Merge remote-tracking branches 'asoc/topic/mc13783', 'asoc/topic/msm8916', 'asoc...
[linux-2.6-block.git] / fs / xfs / libxfs / xfs_iext_tree.c
1 /*
2  * Copyright (c) 2017 Christoph Hellwig.
3  *
4  * This program is free software; you can redistribute it and/or modify it
5  * under the terms and conditions of the GNU General Public License,
6  * version 2, as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope it will be useful, but WITHOUT
9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
11  * more details.
12  */
13
14 #include <linux/cache.h>
15 #include <linux/kernel.h>
16 #include <linux/slab.h>
17 #include "xfs.h"
18 #include "xfs_format.h"
19 #include "xfs_bit.h"
20 #include "xfs_log_format.h"
21 #include "xfs_inode.h"
22 #include "xfs_inode_fork.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_mount.h"
25 #include "xfs_trace.h"
26
27 /*
28  * In-core extent record layout:
29  *
30  * +-------+----------------------------+
31  * | 00:53 | all 54 bits of startoff    |
32  * | 54:63 | low 10 bits of startblock  |
33  * +-------+----------------------------+
34  * | 00:20 | all 21 bits of length      |
35  * |    21 | unwritten extent bit       |
36  * | 22:63 | high 42 bits of startblock |
37  * +-------+----------------------------+
38  */
39 #define XFS_IEXT_STARTOFF_MASK          xfs_mask64lo(BMBT_STARTOFF_BITLEN)
40 #define XFS_IEXT_LENGTH_MASK            xfs_mask64lo(BMBT_BLOCKCOUNT_BITLEN)
41 #define XFS_IEXT_STARTBLOCK_MASK        xfs_mask64lo(BMBT_STARTBLOCK_BITLEN)
42
43 struct xfs_iext_rec {
44         uint64_t                        lo;
45         uint64_t                        hi;
46 };
47
48 /*
49  * Given that the length can't be a zero, only an empty hi value indicates an
50  * unused record.
51  */
52 static bool xfs_iext_rec_is_empty(struct xfs_iext_rec *rec)
53 {
54         return rec->hi == 0;
55 }
56
57 static inline void xfs_iext_rec_clear(struct xfs_iext_rec *rec)
58 {
59         rec->lo = 0;
60         rec->hi = 0;
61 }
62
63 static void
64 xfs_iext_set(
65         struct xfs_iext_rec     *rec,
66         struct xfs_bmbt_irec    *irec)
67 {
68         ASSERT((irec->br_startoff & ~XFS_IEXT_STARTOFF_MASK) == 0);
69         ASSERT((irec->br_blockcount & ~XFS_IEXT_LENGTH_MASK) == 0);
70         ASSERT((irec->br_startblock & ~XFS_IEXT_STARTBLOCK_MASK) == 0);
71
72         rec->lo = irec->br_startoff & XFS_IEXT_STARTOFF_MASK;
73         rec->hi = irec->br_blockcount & XFS_IEXT_LENGTH_MASK;
74
75         rec->lo |= (irec->br_startblock << 54);
76         rec->hi |= ((irec->br_startblock & ~xfs_mask64lo(10)) << (22 - 10));
77
78         if (irec->br_state == XFS_EXT_UNWRITTEN)
79                 rec->hi |= (1 << 21);
80 }
81
82 static void
83 xfs_iext_get(
84         struct xfs_bmbt_irec    *irec,
85         struct xfs_iext_rec     *rec)
86 {
87         irec->br_startoff = rec->lo & XFS_IEXT_STARTOFF_MASK;
88         irec->br_blockcount = rec->hi & XFS_IEXT_LENGTH_MASK;
89
90         irec->br_startblock = rec->lo >> 54;
91         irec->br_startblock |= (rec->hi & xfs_mask64hi(42)) >> (22 - 10);
92
93         if (rec->hi & (1 << 21))
94                 irec->br_state = XFS_EXT_UNWRITTEN;
95         else
96                 irec->br_state = XFS_EXT_NORM;
97 }
98
99 enum {
100         NODE_SIZE       = 256,
101         KEYS_PER_NODE   = NODE_SIZE / (sizeof(uint64_t) + sizeof(void *)),
102         RECS_PER_LEAF   = (NODE_SIZE - (2 * sizeof(struct xfs_iext_leaf *))) /
103                                 sizeof(struct xfs_iext_rec),
104 };
105
106 /*
107  * In-core extent btree block layout:
108  *
109  * There are two types of blocks in the btree: leaf and inner (non-leaf) blocks.
110  *
111  * The leaf blocks are made up by %KEYS_PER_NODE extent records, which each
112  * contain the startoffset, blockcount, startblock and unwritten extent flag.
113  * See above for the exact format, followed by pointers to the previous and next
114  * leaf blocks (if there are any).
115  *
116  * The inner (non-leaf) blocks first contain KEYS_PER_NODE lookup keys, followed
117  * by an equal number of pointers to the btree blocks at the next lower level.
118  *
119  *              +-------+-------+-------+-------+-------+----------+----------+
120  * Leaf:        | rec 1 | rec 2 | rec 3 | rec 4 | rec N | prev-ptr | next-ptr |
121  *              +-------+-------+-------+-------+-------+----------+----------+
122  *
123  *              +-------+-------+-------+-------+-------+-------+------+-------+
124  * Inner:       | key 1 | key 2 | key 3 | key N | ptr 1 | ptr 2 | ptr3 | ptr N |
125  *              +-------+-------+-------+-------+-------+-------+------+-------+
126  */
127 struct xfs_iext_node {
128         uint64_t                keys[KEYS_PER_NODE];
129 #define XFS_IEXT_KEY_INVALID    (1ULL << 63)
130         void                    *ptrs[KEYS_PER_NODE];
131 };
132
133 struct xfs_iext_leaf {
134         struct xfs_iext_rec     recs[RECS_PER_LEAF];
135         struct xfs_iext_leaf    *prev;
136         struct xfs_iext_leaf    *next;
137 };
138
139 inline xfs_extnum_t xfs_iext_count(struct xfs_ifork *ifp)
140 {
141         return ifp->if_bytes / sizeof(struct xfs_iext_rec);
142 }
143
144 static inline int xfs_iext_max_recs(struct xfs_ifork *ifp)
145 {
146         if (ifp->if_height == 1)
147                 return xfs_iext_count(ifp);
148         return RECS_PER_LEAF;
149 }
150
151 static inline struct xfs_iext_rec *cur_rec(struct xfs_iext_cursor *cur)
152 {
153         return &cur->leaf->recs[cur->pos];
154 }
155
156 static inline bool xfs_iext_valid(struct xfs_ifork *ifp,
157                 struct xfs_iext_cursor *cur)
158 {
159         if (!cur->leaf)
160                 return false;
161         if (cur->pos < 0 || cur->pos >= xfs_iext_max_recs(ifp))
162                 return false;
163         if (xfs_iext_rec_is_empty(cur_rec(cur)))
164                 return false;
165         return true;
166 }
167
168 static void *
169 xfs_iext_find_first_leaf(
170         struct xfs_ifork        *ifp)
171 {
172         struct xfs_iext_node    *node = ifp->if_u1.if_root;
173         int                     height;
174
175         if (!ifp->if_height)
176                 return NULL;
177
178         for (height = ifp->if_height; height > 1; height--) {
179                 node = node->ptrs[0];
180                 ASSERT(node);
181         }
182
183         return node;
184 }
185
186 static void *
187 xfs_iext_find_last_leaf(
188         struct xfs_ifork        *ifp)
189 {
190         struct xfs_iext_node    *node = ifp->if_u1.if_root;
191         int                     height, i;
192
193         if (!ifp->if_height)
194                 return NULL;
195
196         for (height = ifp->if_height; height > 1; height--) {
197                 for (i = 1; i < KEYS_PER_NODE; i++)
198                         if (!node->ptrs[i])
199                                 break;
200                 node = node->ptrs[i - 1];
201                 ASSERT(node);
202         }
203
204         return node;
205 }
206
207 void
208 xfs_iext_first(
209         struct xfs_ifork        *ifp,
210         struct xfs_iext_cursor  *cur)
211 {
212         cur->pos = 0;
213         cur->leaf = xfs_iext_find_first_leaf(ifp);
214 }
215
216 void
217 xfs_iext_last(
218         struct xfs_ifork        *ifp,
219         struct xfs_iext_cursor  *cur)
220 {
221         int                     i;
222
223         cur->leaf = xfs_iext_find_last_leaf(ifp);
224         if (!cur->leaf) {
225                 cur->pos = 0;
226                 return;
227         }
228
229         for (i = 1; i < xfs_iext_max_recs(ifp); i++) {
230                 if (xfs_iext_rec_is_empty(&cur->leaf->recs[i]))
231                         break;
232         }
233         cur->pos = i - 1;
234 }
235
236 void
237 xfs_iext_next(
238         struct xfs_ifork        *ifp,
239         struct xfs_iext_cursor  *cur)
240 {
241         if (!cur->leaf) {
242                 ASSERT(cur->pos <= 0 || cur->pos >= RECS_PER_LEAF);
243                 xfs_iext_first(ifp, cur);
244                 return;
245         }
246
247         ASSERT(cur->pos >= 0);
248         ASSERT(cur->pos < xfs_iext_max_recs(ifp));
249
250         cur->pos++;
251         if (ifp->if_height > 1 && !xfs_iext_valid(ifp, cur) &&
252             cur->leaf->next) {
253                 cur->leaf = cur->leaf->next;
254                 cur->pos = 0;
255         }
256 }
257
258 void
259 xfs_iext_prev(
260         struct xfs_ifork        *ifp,
261         struct xfs_iext_cursor  *cur)
262 {
263         if (!cur->leaf) {
264                 ASSERT(cur->pos <= 0 || cur->pos >= RECS_PER_LEAF);
265                 xfs_iext_last(ifp, cur);
266                 return;
267         }
268
269         ASSERT(cur->pos >= 0);
270         ASSERT(cur->pos <= RECS_PER_LEAF);
271
272 recurse:
273         do {
274                 cur->pos--;
275                 if (xfs_iext_valid(ifp, cur))
276                         return;
277         } while (cur->pos > 0);
278
279         if (ifp->if_height > 1 && cur->leaf->prev) {
280                 cur->leaf = cur->leaf->prev;
281                 cur->pos = RECS_PER_LEAF;
282                 goto recurse;
283         }
284 }
285
286 static inline int
287 xfs_iext_key_cmp(
288         struct xfs_iext_node    *node,
289         int                     n,
290         xfs_fileoff_t           offset)
291 {
292         if (node->keys[n] > offset)
293                 return 1;
294         if (node->keys[n] < offset)
295                 return -1;
296         return 0;
297 }
298
299 static inline int
300 xfs_iext_rec_cmp(
301         struct xfs_iext_rec     *rec,
302         xfs_fileoff_t           offset)
303 {
304         uint64_t                rec_offset = rec->lo & XFS_IEXT_STARTOFF_MASK;
305         uint32_t                rec_len = rec->hi & XFS_IEXT_LENGTH_MASK;
306
307         if (rec_offset > offset)
308                 return 1;
309         if (rec_offset + rec_len <= offset)
310                 return -1;
311         return 0;
312 }
313
314 static void *
315 xfs_iext_find_level(
316         struct xfs_ifork        *ifp,
317         xfs_fileoff_t           offset,
318         int                     level)
319 {
320         struct xfs_iext_node    *node = ifp->if_u1.if_root;
321         int                     height, i;
322
323         if (!ifp->if_height)
324                 return NULL;
325
326         for (height = ifp->if_height; height > level; height--) {
327                 for (i = 1; i < KEYS_PER_NODE; i++)
328                         if (xfs_iext_key_cmp(node, i, offset) > 0)
329                                 break;
330
331                 node = node->ptrs[i - 1];
332                 if (!node)
333                         break;
334         }
335
336         return node;
337 }
338
339 static int
340 xfs_iext_node_pos(
341         struct xfs_iext_node    *node,
342         xfs_fileoff_t           offset)
343 {
344         int                     i;
345
346         for (i = 1; i < KEYS_PER_NODE; i++) {
347                 if (xfs_iext_key_cmp(node, i, offset) > 0)
348                         break;
349         }
350
351         return i - 1;
352 }
353
354 static int
355 xfs_iext_node_insert_pos(
356         struct xfs_iext_node    *node,
357         xfs_fileoff_t           offset)
358 {
359         int                     i;
360
361         for (i = 0; i < KEYS_PER_NODE; i++) {
362                 if (xfs_iext_key_cmp(node, i, offset) > 0)
363                         return i;
364         }
365
366         return KEYS_PER_NODE;
367 }
368
369 static int
370 xfs_iext_node_nr_entries(
371         struct xfs_iext_node    *node,
372         int                     start)
373 {
374         int                     i;
375
376         for (i = start; i < KEYS_PER_NODE; i++) {
377                 if (node->keys[i] == XFS_IEXT_KEY_INVALID)
378                         break;
379         }
380
381         return i;
382 }
383
384 static int
385 xfs_iext_leaf_nr_entries(
386         struct xfs_ifork        *ifp,
387         struct xfs_iext_leaf    *leaf,
388         int                     start)
389 {
390         int                     i;
391
392         for (i = start; i < xfs_iext_max_recs(ifp); i++) {
393                 if (xfs_iext_rec_is_empty(&leaf->recs[i]))
394                         break;
395         }
396
397         return i;
398 }
399
400 static inline uint64_t
401 xfs_iext_leaf_key(
402         struct xfs_iext_leaf    *leaf,
403         int                     n)
404 {
405         return leaf->recs[n].lo & XFS_IEXT_STARTOFF_MASK;
406 }
407
408 static void
409 xfs_iext_grow(
410         struct xfs_ifork        *ifp)
411 {
412         struct xfs_iext_node    *node = kmem_zalloc(NODE_SIZE, KM_NOFS);
413         int                     i;
414
415         if (ifp->if_height == 1) {
416                 struct xfs_iext_leaf *prev = ifp->if_u1.if_root;
417
418                 node->keys[0] = xfs_iext_leaf_key(prev, 0);
419                 node->ptrs[0] = prev;
420         } else  {
421                 struct xfs_iext_node *prev = ifp->if_u1.if_root;
422
423                 ASSERT(ifp->if_height > 1);
424
425                 node->keys[0] = prev->keys[0];
426                 node->ptrs[0] = prev;
427         }
428
429         for (i = 1; i < KEYS_PER_NODE; i++)
430                 node->keys[i] = XFS_IEXT_KEY_INVALID;
431
432         ifp->if_u1.if_root = node;
433         ifp->if_height++;
434 }
435
436 static void
437 xfs_iext_update_node(
438         struct xfs_ifork        *ifp,
439         xfs_fileoff_t           old_offset,
440         xfs_fileoff_t           new_offset,
441         int                     level,
442         void                    *ptr)
443 {
444         struct xfs_iext_node    *node = ifp->if_u1.if_root;
445         int                     height, i;
446
447         for (height = ifp->if_height; height > level; height--) {
448                 for (i = 0; i < KEYS_PER_NODE; i++) {
449                         if (i > 0 && xfs_iext_key_cmp(node, i, old_offset) > 0)
450                                 break;
451                         if (node->keys[i] == old_offset)
452                                 node->keys[i] = new_offset;
453                 }
454                 node = node->ptrs[i - 1];
455                 ASSERT(node);
456         }
457
458         ASSERT(node == ptr);
459 }
460
461 static struct xfs_iext_node *
462 xfs_iext_split_node(
463         struct xfs_iext_node    **nodep,
464         int                     *pos,
465         int                     *nr_entries)
466 {
467         struct xfs_iext_node    *node = *nodep;
468         struct xfs_iext_node    *new = kmem_zalloc(NODE_SIZE, KM_NOFS);
469         const int               nr_move = KEYS_PER_NODE / 2;
470         int                     nr_keep = nr_move + (KEYS_PER_NODE & 1);
471         int                     i = 0;
472
473         /* for sequential append operations just spill over into the new node */
474         if (*pos == KEYS_PER_NODE) {
475                 *nodep = new;
476                 *pos = 0;
477                 *nr_entries = 0;
478                 goto done;
479         }
480
481
482         for (i = 0; i < nr_move; i++) {
483                 new->keys[i] = node->keys[nr_keep + i];
484                 new->ptrs[i] = node->ptrs[nr_keep + i];
485
486                 node->keys[nr_keep + i] = XFS_IEXT_KEY_INVALID;
487                 node->ptrs[nr_keep + i] = NULL;
488         }
489
490         if (*pos >= nr_keep) {
491                 *nodep = new;
492                 *pos -= nr_keep;
493                 *nr_entries = nr_move;
494         } else {
495                 *nr_entries = nr_keep;
496         }
497 done:
498         for (; i < KEYS_PER_NODE; i++)
499                 new->keys[i] = XFS_IEXT_KEY_INVALID;
500         return new;
501 }
502
503 static void
504 xfs_iext_insert_node(
505         struct xfs_ifork        *ifp,
506         uint64_t                offset,
507         void                    *ptr,
508         int                     level)
509 {
510         struct xfs_iext_node    *node, *new;
511         int                     i, pos, nr_entries;
512
513 again:
514         if (ifp->if_height < level)
515                 xfs_iext_grow(ifp);
516
517         new = NULL;
518         node = xfs_iext_find_level(ifp, offset, level);
519         pos = xfs_iext_node_insert_pos(node, offset);
520         nr_entries = xfs_iext_node_nr_entries(node, pos);
521
522         ASSERT(pos >= nr_entries || xfs_iext_key_cmp(node, pos, offset) != 0);
523         ASSERT(nr_entries <= KEYS_PER_NODE);
524
525         if (nr_entries == KEYS_PER_NODE)
526                 new = xfs_iext_split_node(&node, &pos, &nr_entries);
527
528         /*
529          * Update the pointers in higher levels if the first entry changes
530          * in an existing node.
531          */
532         if (node != new && pos == 0 && nr_entries > 0)
533                 xfs_iext_update_node(ifp, node->keys[0], offset, level, node);
534
535         for (i = nr_entries; i > pos; i--) {
536                 node->keys[i] = node->keys[i - 1];
537                 node->ptrs[i] = node->ptrs[i - 1];
538         }
539         node->keys[pos] = offset;
540         node->ptrs[pos] = ptr;
541
542         if (new) {
543                 offset = new->keys[0];
544                 ptr = new;
545                 level++;
546                 goto again;
547         }
548 }
549
550 static struct xfs_iext_leaf *
551 xfs_iext_split_leaf(
552         struct xfs_iext_cursor  *cur,
553         int                     *nr_entries)
554 {
555         struct xfs_iext_leaf    *leaf = cur->leaf;
556         struct xfs_iext_leaf    *new = kmem_zalloc(NODE_SIZE, KM_NOFS);
557         const int               nr_move = RECS_PER_LEAF / 2;
558         int                     nr_keep = nr_move + (RECS_PER_LEAF & 1);
559         int                     i;
560
561         /* for sequential append operations just spill over into the new node */
562         if (cur->pos == RECS_PER_LEAF) {
563                 cur->leaf = new;
564                 cur->pos = 0;
565                 *nr_entries = 0;
566                 goto done;
567         }
568
569         for (i = 0; i < nr_move; i++) {
570                 new->recs[i] = leaf->recs[nr_keep + i];
571                 xfs_iext_rec_clear(&leaf->recs[nr_keep + i]);
572         }
573
574         if (cur->pos >= nr_keep) {
575                 cur->leaf = new;
576                 cur->pos -= nr_keep;
577                 *nr_entries = nr_move;
578         } else {
579                 *nr_entries = nr_keep;
580         }
581 done:
582         if (leaf->next)
583                 leaf->next->prev = new;
584         new->next = leaf->next;
585         new->prev = leaf;
586         leaf->next = new;
587         return new;
588 }
589
590 static void
591 xfs_iext_alloc_root(
592         struct xfs_ifork        *ifp,
593         struct xfs_iext_cursor  *cur)
594 {
595         ASSERT(ifp->if_bytes == 0);
596
597         ifp->if_u1.if_root = kmem_zalloc(sizeof(struct xfs_iext_rec), KM_NOFS);
598         ifp->if_height = 1;
599
600         /* now that we have a node step into it */
601         cur->leaf = ifp->if_u1.if_root;
602         cur->pos = 0;
603 }
604
605 static void
606 xfs_iext_realloc_root(
607         struct xfs_ifork        *ifp,
608         struct xfs_iext_cursor  *cur)
609 {
610         size_t new_size = ifp->if_bytes + sizeof(struct xfs_iext_rec);
611         void *new;
612
613         /* account for the prev/next pointers */
614         if (new_size / sizeof(struct xfs_iext_rec) == RECS_PER_LEAF)
615                 new_size = NODE_SIZE;
616
617         new = kmem_realloc(ifp->if_u1.if_root, new_size, KM_NOFS);
618         memset(new + ifp->if_bytes, 0, new_size - ifp->if_bytes);
619         ifp->if_u1.if_root = new;
620         cur->leaf = new;
621 }
622
623 void
624 xfs_iext_insert(
625         struct xfs_inode        *ip,
626         struct xfs_iext_cursor  *cur,
627         struct xfs_bmbt_irec    *irec,
628         int                     state)
629 {
630         struct xfs_ifork        *ifp = xfs_iext_state_to_fork(ip, state);
631         xfs_fileoff_t           offset = irec->br_startoff;
632         struct xfs_iext_leaf    *new = NULL;
633         int                     nr_entries, i;
634
635         if (ifp->if_height == 0)
636                 xfs_iext_alloc_root(ifp, cur);
637         else if (ifp->if_height == 1)
638                 xfs_iext_realloc_root(ifp, cur);
639
640         nr_entries = xfs_iext_leaf_nr_entries(ifp, cur->leaf, cur->pos);
641         ASSERT(nr_entries <= RECS_PER_LEAF);
642         ASSERT(cur->pos >= nr_entries ||
643                xfs_iext_rec_cmp(cur_rec(cur), irec->br_startoff) != 0);
644
645         if (nr_entries == RECS_PER_LEAF)
646                 new = xfs_iext_split_leaf(cur, &nr_entries);
647
648         /*
649          * Update the pointers in higher levels if the first entry changes
650          * in an existing node.
651          */
652         if (cur->leaf != new && cur->pos == 0 && nr_entries > 0) {
653                 xfs_iext_update_node(ifp, xfs_iext_leaf_key(cur->leaf, 0),
654                                 offset, 1, cur->leaf);
655         }
656
657         for (i = nr_entries; i > cur->pos; i--)
658                 cur->leaf->recs[i] = cur->leaf->recs[i - 1];
659         xfs_iext_set(cur_rec(cur), irec);
660         ifp->if_bytes += sizeof(struct xfs_iext_rec);
661
662         trace_xfs_iext_insert(ip, cur, state, _RET_IP_);
663
664         if (new)
665                 xfs_iext_insert_node(ifp, xfs_iext_leaf_key(new, 0), new, 2);
666 }
667
668 static struct xfs_iext_node *
669 xfs_iext_rebalance_node(
670         struct xfs_iext_node    *parent,
671         int                     *pos,
672         struct xfs_iext_node    *node,
673         int                     nr_entries)
674 {
675         /*
676          * If the neighbouring nodes are completely full, or have different
677          * parents, we might never be able to merge our node, and will only
678          * delete it once the number of entries hits zero.
679          */
680         if (nr_entries == 0)
681                 return node;
682
683         if (*pos > 0) {
684                 struct xfs_iext_node *prev = parent->ptrs[*pos - 1];
685                 int nr_prev = xfs_iext_node_nr_entries(prev, 0), i;
686
687                 if (nr_prev + nr_entries <= KEYS_PER_NODE) {
688                         for (i = 0; i < nr_entries; i++) {
689                                 prev->keys[nr_prev + i] = node->keys[i];
690                                 prev->ptrs[nr_prev + i] = node->ptrs[i];
691                         }
692                         return node;
693                 }
694         }
695
696         if (*pos + 1 < xfs_iext_node_nr_entries(parent, *pos)) {
697                 struct xfs_iext_node *next = parent->ptrs[*pos + 1];
698                 int nr_next = xfs_iext_node_nr_entries(next, 0), i;
699
700                 if (nr_entries + nr_next <= KEYS_PER_NODE) {
701                         /*
702                          * Merge the next node into this node so that we don't
703                          * have to do an additional update of the keys in the
704                          * higher levels.
705                          */
706                         for (i = 0; i < nr_next; i++) {
707                                 node->keys[nr_entries + i] = next->keys[i];
708                                 node->ptrs[nr_entries + i] = next->ptrs[i];
709                         }
710
711                         ++*pos;
712                         return next;
713                 }
714         }
715
716         return NULL;
717 }
718
719 static void
720 xfs_iext_remove_node(
721         struct xfs_ifork        *ifp,
722         xfs_fileoff_t           offset,
723         void                    *victim)
724 {
725         struct xfs_iext_node    *node, *parent;
726         int                     level = 2, pos, nr_entries, i;
727
728         ASSERT(level <= ifp->if_height);
729         node = xfs_iext_find_level(ifp, offset, level);
730         pos = xfs_iext_node_pos(node, offset);
731 again:
732         ASSERT(node->ptrs[pos]);
733         ASSERT(node->ptrs[pos] == victim);
734         kmem_free(victim);
735
736         nr_entries = xfs_iext_node_nr_entries(node, pos) - 1;
737         offset = node->keys[0];
738         for (i = pos; i < nr_entries; i++) {
739                 node->keys[i] = node->keys[i + 1];
740                 node->ptrs[i] = node->ptrs[i + 1];
741         }
742         node->keys[nr_entries] = XFS_IEXT_KEY_INVALID;
743         node->ptrs[nr_entries] = NULL;
744
745         if (pos == 0 && nr_entries > 0) {
746                 xfs_iext_update_node(ifp, offset, node->keys[0], level, node);
747                 offset = node->keys[0];
748         }
749
750         if (nr_entries >= KEYS_PER_NODE / 2)
751                 return;
752
753         if (level < ifp->if_height) {
754                 /*
755                  * If we aren't at the root yet try to find a neighbour node to
756                  * merge with (or delete the node if it is empty), and then
757                  * recurse up to the next level.
758                  */
759                 level++;
760                 parent = xfs_iext_find_level(ifp, offset, level);
761                 pos = xfs_iext_node_pos(parent, offset);
762
763                 ASSERT(pos != KEYS_PER_NODE);
764                 ASSERT(parent->ptrs[pos] == node);
765
766                 node = xfs_iext_rebalance_node(parent, &pos, node, nr_entries);
767                 if (node) {
768                         victim = node;
769                         node = parent;
770                         goto again;
771                 }
772         } else if (nr_entries == 1) {
773                 /*
774                  * If we are at the root and only one entry is left we can just
775                  * free this node and update the root pointer.
776                  */
777                 ASSERT(node == ifp->if_u1.if_root);
778                 ifp->if_u1.if_root = node->ptrs[0];
779                 ifp->if_height--;
780                 kmem_free(node);
781         }
782 }
783
784 static void
785 xfs_iext_rebalance_leaf(
786         struct xfs_ifork        *ifp,
787         struct xfs_iext_cursor  *cur,
788         struct xfs_iext_leaf    *leaf,
789         xfs_fileoff_t           offset,
790         int                     nr_entries)
791 {
792         /*
793          * If the neighbouring nodes are completely full we might never be able
794          * to merge our node, and will only delete it once the number of
795          * entries hits zero.
796          */
797         if (nr_entries == 0)
798                 goto remove_node;
799
800         if (leaf->prev) {
801                 int nr_prev = xfs_iext_leaf_nr_entries(ifp, leaf->prev, 0), i;
802
803                 if (nr_prev + nr_entries <= RECS_PER_LEAF) {
804                         for (i = 0; i < nr_entries; i++)
805                                 leaf->prev->recs[nr_prev + i] = leaf->recs[i];
806
807                         if (cur->leaf == leaf) {
808                                 cur->leaf = leaf->prev;
809                                 cur->pos += nr_prev;
810                         }
811                         goto remove_node;
812                 }
813         }
814
815         if (leaf->next) {
816                 int nr_next = xfs_iext_leaf_nr_entries(ifp, leaf->next, 0), i;
817
818                 if (nr_entries + nr_next <= RECS_PER_LEAF) {
819                         /*
820                          * Merge the next node into this node so that we don't
821                          * have to do an additional update of the keys in the
822                          * higher levels.
823                          */
824                         for (i = 0; i < nr_next; i++) {
825                                 leaf->recs[nr_entries + i] =
826                                         leaf->next->recs[i];
827                         }
828
829                         if (cur->leaf == leaf->next) {
830                                 cur->leaf = leaf;
831                                 cur->pos += nr_entries;
832                         }
833
834                         offset = xfs_iext_leaf_key(leaf->next, 0);
835                         leaf = leaf->next;
836                         goto remove_node;
837                 }
838         }
839
840         return;
841 remove_node:
842         if (leaf->prev)
843                 leaf->prev->next = leaf->next;
844         if (leaf->next)
845                 leaf->next->prev = leaf->prev;
846         xfs_iext_remove_node(ifp, offset, leaf);
847 }
848
849 static void
850 xfs_iext_free_last_leaf(
851         struct xfs_ifork        *ifp)
852 {
853         ifp->if_height--;
854         kmem_free(ifp->if_u1.if_root);
855         ifp->if_u1.if_root = NULL;
856 }
857
858 void
859 xfs_iext_remove(
860         struct xfs_inode        *ip,
861         struct xfs_iext_cursor  *cur,
862         int                     state)
863 {
864         struct xfs_ifork        *ifp = xfs_iext_state_to_fork(ip, state);
865         struct xfs_iext_leaf    *leaf = cur->leaf;
866         xfs_fileoff_t           offset = xfs_iext_leaf_key(leaf, 0);
867         int                     i, nr_entries;
868
869         trace_xfs_iext_remove(ip, cur, state, _RET_IP_);
870
871         ASSERT(ifp->if_height > 0);
872         ASSERT(ifp->if_u1.if_root != NULL);
873         ASSERT(xfs_iext_valid(ifp, cur));
874
875         nr_entries = xfs_iext_leaf_nr_entries(ifp, leaf, cur->pos) - 1;
876         for (i = cur->pos; i < nr_entries; i++)
877                 leaf->recs[i] = leaf->recs[i + 1];
878         xfs_iext_rec_clear(&leaf->recs[nr_entries]);
879         ifp->if_bytes -= sizeof(struct xfs_iext_rec);
880
881         if (cur->pos == 0 && nr_entries > 0) {
882                 xfs_iext_update_node(ifp, offset, xfs_iext_leaf_key(leaf, 0), 1,
883                                 leaf);
884                 offset = xfs_iext_leaf_key(leaf, 0);
885         } else if (cur->pos == nr_entries) {
886                 if (ifp->if_height > 1 && leaf->next)
887                         cur->leaf = leaf->next;
888                 else
889                         cur->leaf = NULL;
890                 cur->pos = 0;
891         }
892
893         if (nr_entries >= RECS_PER_LEAF / 2)
894                 return;
895
896         if (ifp->if_height > 1)
897                 xfs_iext_rebalance_leaf(ifp, cur, leaf, offset, nr_entries);
898         else if (nr_entries == 0)
899                 xfs_iext_free_last_leaf(ifp);
900 }
901
902 /*
903  * Lookup the extent covering bno.
904  *
905  * If there is an extent covering bno return the extent index, and store the
906  * expanded extent structure in *gotp, and the extent cursor in *cur.
907  * If there is no extent covering bno, but there is an extent after it (e.g.
908  * it lies in a hole) return that extent in *gotp and its cursor in *cur
909  * instead.
910  * If bno is beyond the last extent return false, and return an invalid
911  * cursor value.
912  */
913 bool
914 xfs_iext_lookup_extent(
915         struct xfs_inode        *ip,
916         struct xfs_ifork        *ifp,
917         xfs_fileoff_t           offset,
918         struct xfs_iext_cursor  *cur,
919         struct xfs_bmbt_irec    *gotp)
920 {
921         XFS_STATS_INC(ip->i_mount, xs_look_exlist);
922
923         cur->leaf = xfs_iext_find_level(ifp, offset, 1);
924         if (!cur->leaf) {
925                 cur->pos = 0;
926                 return false;
927         }
928
929         for (cur->pos = 0; cur->pos < xfs_iext_max_recs(ifp); cur->pos++) {
930                 struct xfs_iext_rec *rec = cur_rec(cur);
931
932                 if (xfs_iext_rec_is_empty(rec))
933                         break;
934                 if (xfs_iext_rec_cmp(rec, offset) >= 0)
935                         goto found;
936         }
937
938         /* Try looking in the next node for an entry > offset */
939         if (ifp->if_height == 1 || !cur->leaf->next)
940                 return false;
941         cur->leaf = cur->leaf->next;
942         cur->pos = 0;
943         if (!xfs_iext_valid(ifp, cur))
944                 return false;
945 found:
946         xfs_iext_get(gotp, cur_rec(cur));
947         return true;
948 }
949
950 /*
951  * Returns the last extent before end, and if this extent doesn't cover
952  * end, update end to the end of the extent.
953  */
954 bool
955 xfs_iext_lookup_extent_before(
956         struct xfs_inode        *ip,
957         struct xfs_ifork        *ifp,
958         xfs_fileoff_t           *end,
959         struct xfs_iext_cursor  *cur,
960         struct xfs_bmbt_irec    *gotp)
961 {
962         /* could be optimized to not even look up the next on a match.. */
963         if (xfs_iext_lookup_extent(ip, ifp, *end - 1, cur, gotp) &&
964             gotp->br_startoff <= *end - 1)
965                 return true;
966         if (!xfs_iext_prev_extent(ifp, cur, gotp))
967                 return false;
968         *end = gotp->br_startoff + gotp->br_blockcount;
969         return true;
970 }
971
972 void
973 xfs_iext_update_extent(
974         struct xfs_inode        *ip,
975         int                     state,
976         struct xfs_iext_cursor  *cur,
977         struct xfs_bmbt_irec    *new)
978 {
979         struct xfs_ifork        *ifp = xfs_iext_state_to_fork(ip, state);
980
981         if (cur->pos == 0) {
982                 struct xfs_bmbt_irec    old;
983
984                 xfs_iext_get(&old, cur_rec(cur));
985                 if (new->br_startoff != old.br_startoff) {
986                         xfs_iext_update_node(ifp, old.br_startoff,
987                                         new->br_startoff, 1, cur->leaf);
988                 }
989         }
990
991         trace_xfs_bmap_pre_update(ip, cur, state, _RET_IP_);
992         xfs_iext_set(cur_rec(cur), new);
993         trace_xfs_bmap_post_update(ip, cur, state, _RET_IP_);
994 }
995
996 /*
997  * Return true if the cursor points at an extent and return the extent structure
998  * in gotp.  Else return false.
999  */
1000 bool
1001 xfs_iext_get_extent(
1002         struct xfs_ifork        *ifp,
1003         struct xfs_iext_cursor  *cur,
1004         struct xfs_bmbt_irec    *gotp)
1005 {
1006         if (!xfs_iext_valid(ifp, cur))
1007                 return false;
1008         xfs_iext_get(gotp, cur_rec(cur));
1009         return true;
1010 }
1011
1012 /*
1013  * This is a recursive function, because of that we need to be extremely
1014  * careful with stack usage.
1015  */
1016 static void
1017 xfs_iext_destroy_node(
1018         struct xfs_iext_node    *node,
1019         int                     level)
1020 {
1021         int                     i;
1022
1023         if (level > 1) {
1024                 for (i = 0; i < KEYS_PER_NODE; i++) {
1025                         if (node->keys[i] == XFS_IEXT_KEY_INVALID)
1026                                 break;
1027                         xfs_iext_destroy_node(node->ptrs[i], level - 1);
1028                 }
1029         }
1030
1031         kmem_free(node);
1032 }
1033
1034 void
1035 xfs_iext_destroy(
1036         struct xfs_ifork        *ifp)
1037 {
1038         xfs_iext_destroy_node(ifp->if_u1.if_root, ifp->if_height);
1039
1040         ifp->if_bytes = 0;
1041         ifp->if_height = 0;
1042         ifp->if_u1.if_root = NULL;
1043 }