Merge branch 'work.audit' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[linux-2.6-microblaze.git] / fs / xfs / libxfs / xfs_iext_tree.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2017 Christoph Hellwig.
4  */
5
6 #include "xfs.h"
7 #include "xfs_shared.h"
8 #include "xfs_format.h"
9 #include "xfs_bit.h"
10 #include "xfs_log_format.h"
11 #include "xfs_inode.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_trace.h"
15
16 /*
17  * In-core extent record layout:
18  *
19  * +-------+----------------------------+
20  * | 00:53 | all 54 bits of startoff    |
21  * | 54:63 | low 10 bits of startblock  |
22  * +-------+----------------------------+
23  * | 00:20 | all 21 bits of length      |
24  * |    21 | unwritten extent bit       |
25  * | 22:63 | high 42 bits of startblock |
26  * +-------+----------------------------+
27  */
28 #define XFS_IEXT_STARTOFF_MASK          xfs_mask64lo(BMBT_STARTOFF_BITLEN)
29 #define XFS_IEXT_LENGTH_MASK            xfs_mask64lo(BMBT_BLOCKCOUNT_BITLEN)
30 #define XFS_IEXT_STARTBLOCK_MASK        xfs_mask64lo(BMBT_STARTBLOCK_BITLEN)
31
32 struct xfs_iext_rec {
33         uint64_t                        lo;
34         uint64_t                        hi;
35 };
36
37 /*
38  * Given that the length can't be a zero, only an empty hi value indicates an
39  * unused record.
40  */
41 static bool xfs_iext_rec_is_empty(struct xfs_iext_rec *rec)
42 {
43         return rec->hi == 0;
44 }
45
46 static inline void xfs_iext_rec_clear(struct xfs_iext_rec *rec)
47 {
48         rec->lo = 0;
49         rec->hi = 0;
50 }
51
52 static void
53 xfs_iext_set(
54         struct xfs_iext_rec     *rec,
55         struct xfs_bmbt_irec    *irec)
56 {
57         ASSERT((irec->br_startoff & ~XFS_IEXT_STARTOFF_MASK) == 0);
58         ASSERT((irec->br_blockcount & ~XFS_IEXT_LENGTH_MASK) == 0);
59         ASSERT((irec->br_startblock & ~XFS_IEXT_STARTBLOCK_MASK) == 0);
60
61         rec->lo = irec->br_startoff & XFS_IEXT_STARTOFF_MASK;
62         rec->hi = irec->br_blockcount & XFS_IEXT_LENGTH_MASK;
63
64         rec->lo |= (irec->br_startblock << 54);
65         rec->hi |= ((irec->br_startblock & ~xfs_mask64lo(10)) << (22 - 10));
66
67         if (irec->br_state == XFS_EXT_UNWRITTEN)
68                 rec->hi |= (1 << 21);
69 }
70
71 static void
72 xfs_iext_get(
73         struct xfs_bmbt_irec    *irec,
74         struct xfs_iext_rec     *rec)
75 {
76         irec->br_startoff = rec->lo & XFS_IEXT_STARTOFF_MASK;
77         irec->br_blockcount = rec->hi & XFS_IEXT_LENGTH_MASK;
78
79         irec->br_startblock = rec->lo >> 54;
80         irec->br_startblock |= (rec->hi & xfs_mask64hi(42)) >> (22 - 10);
81
82         if (rec->hi & (1 << 21))
83                 irec->br_state = XFS_EXT_UNWRITTEN;
84         else
85                 irec->br_state = XFS_EXT_NORM;
86 }
87
88 enum {
89         NODE_SIZE       = 256,
90         KEYS_PER_NODE   = NODE_SIZE / (sizeof(uint64_t) + sizeof(void *)),
91         RECS_PER_LEAF   = (NODE_SIZE - (2 * sizeof(struct xfs_iext_leaf *))) /
92                                 sizeof(struct xfs_iext_rec),
93 };
94
95 /*
96  * In-core extent btree block layout:
97  *
98  * There are two types of blocks in the btree: leaf and inner (non-leaf) blocks.
99  *
100  * The leaf blocks are made up by %KEYS_PER_NODE extent records, which each
101  * contain the startoffset, blockcount, startblock and unwritten extent flag.
102  * See above for the exact format, followed by pointers to the previous and next
103  * leaf blocks (if there are any).
104  *
105  * The inner (non-leaf) blocks first contain KEYS_PER_NODE lookup keys, followed
106  * by an equal number of pointers to the btree blocks at the next lower level.
107  *
108  *              +-------+-------+-------+-------+-------+----------+----------+
109  * Leaf:        | rec 1 | rec 2 | rec 3 | rec 4 | rec N | prev-ptr | next-ptr |
110  *              +-------+-------+-------+-------+-------+----------+----------+
111  *
112  *              +-------+-------+-------+-------+-------+-------+------+-------+
113  * Inner:       | key 1 | key 2 | key 3 | key N | ptr 1 | ptr 2 | ptr3 | ptr N |
114  *              +-------+-------+-------+-------+-------+-------+------+-------+
115  */
116 struct xfs_iext_node {
117         uint64_t                keys[KEYS_PER_NODE];
118 #define XFS_IEXT_KEY_INVALID    (1ULL << 63)
119         void                    *ptrs[KEYS_PER_NODE];
120 };
121
122 struct xfs_iext_leaf {
123         struct xfs_iext_rec     recs[RECS_PER_LEAF];
124         struct xfs_iext_leaf    *prev;
125         struct xfs_iext_leaf    *next;
126 };
127
128 inline xfs_extnum_t xfs_iext_count(struct xfs_ifork *ifp)
129 {
130         return ifp->if_bytes / sizeof(struct xfs_iext_rec);
131 }
132
133 static inline int xfs_iext_max_recs(struct xfs_ifork *ifp)
134 {
135         if (ifp->if_height == 1)
136                 return xfs_iext_count(ifp);
137         return RECS_PER_LEAF;
138 }
139
140 static inline struct xfs_iext_rec *cur_rec(struct xfs_iext_cursor *cur)
141 {
142         return &cur->leaf->recs[cur->pos];
143 }
144
145 static inline bool xfs_iext_valid(struct xfs_ifork *ifp,
146                 struct xfs_iext_cursor *cur)
147 {
148         if (!cur->leaf)
149                 return false;
150         if (cur->pos < 0 || cur->pos >= xfs_iext_max_recs(ifp))
151                 return false;
152         if (xfs_iext_rec_is_empty(cur_rec(cur)))
153                 return false;
154         return true;
155 }
156
157 static void *
158 xfs_iext_find_first_leaf(
159         struct xfs_ifork        *ifp)
160 {
161         struct xfs_iext_node    *node = ifp->if_u1.if_root;
162         int                     height;
163
164         if (!ifp->if_height)
165                 return NULL;
166
167         for (height = ifp->if_height; height > 1; height--) {
168                 node = node->ptrs[0];
169                 ASSERT(node);
170         }
171
172         return node;
173 }
174
175 static void *
176 xfs_iext_find_last_leaf(
177         struct xfs_ifork        *ifp)
178 {
179         struct xfs_iext_node    *node = ifp->if_u1.if_root;
180         int                     height, i;
181
182         if (!ifp->if_height)
183                 return NULL;
184
185         for (height = ifp->if_height; height > 1; height--) {
186                 for (i = 1; i < KEYS_PER_NODE; i++)
187                         if (!node->ptrs[i])
188                                 break;
189                 node = node->ptrs[i - 1];
190                 ASSERT(node);
191         }
192
193         return node;
194 }
195
196 void
197 xfs_iext_first(
198         struct xfs_ifork        *ifp,
199         struct xfs_iext_cursor  *cur)
200 {
201         cur->pos = 0;
202         cur->leaf = xfs_iext_find_first_leaf(ifp);
203 }
204
205 void
206 xfs_iext_last(
207         struct xfs_ifork        *ifp,
208         struct xfs_iext_cursor  *cur)
209 {
210         int                     i;
211
212         cur->leaf = xfs_iext_find_last_leaf(ifp);
213         if (!cur->leaf) {
214                 cur->pos = 0;
215                 return;
216         }
217
218         for (i = 1; i < xfs_iext_max_recs(ifp); i++) {
219                 if (xfs_iext_rec_is_empty(&cur->leaf->recs[i]))
220                         break;
221         }
222         cur->pos = i - 1;
223 }
224
225 void
226 xfs_iext_next(
227         struct xfs_ifork        *ifp,
228         struct xfs_iext_cursor  *cur)
229 {
230         if (!cur->leaf) {
231                 ASSERT(cur->pos <= 0 || cur->pos >= RECS_PER_LEAF);
232                 xfs_iext_first(ifp, cur);
233                 return;
234         }
235
236         ASSERT(cur->pos >= 0);
237         ASSERT(cur->pos < xfs_iext_max_recs(ifp));
238
239         cur->pos++;
240         if (ifp->if_height > 1 && !xfs_iext_valid(ifp, cur) &&
241             cur->leaf->next) {
242                 cur->leaf = cur->leaf->next;
243                 cur->pos = 0;
244         }
245 }
246
247 void
248 xfs_iext_prev(
249         struct xfs_ifork        *ifp,
250         struct xfs_iext_cursor  *cur)
251 {
252         if (!cur->leaf) {
253                 ASSERT(cur->pos <= 0 || cur->pos >= RECS_PER_LEAF);
254                 xfs_iext_last(ifp, cur);
255                 return;
256         }
257
258         ASSERT(cur->pos >= 0);
259         ASSERT(cur->pos <= RECS_PER_LEAF);
260
261 recurse:
262         do {
263                 cur->pos--;
264                 if (xfs_iext_valid(ifp, cur))
265                         return;
266         } while (cur->pos > 0);
267
268         if (ifp->if_height > 1 && cur->leaf->prev) {
269                 cur->leaf = cur->leaf->prev;
270                 cur->pos = RECS_PER_LEAF;
271                 goto recurse;
272         }
273 }
274
275 static inline int
276 xfs_iext_key_cmp(
277         struct xfs_iext_node    *node,
278         int                     n,
279         xfs_fileoff_t           offset)
280 {
281         if (node->keys[n] > offset)
282                 return 1;
283         if (node->keys[n] < offset)
284                 return -1;
285         return 0;
286 }
287
288 static inline int
289 xfs_iext_rec_cmp(
290         struct xfs_iext_rec     *rec,
291         xfs_fileoff_t           offset)
292 {
293         uint64_t                rec_offset = rec->lo & XFS_IEXT_STARTOFF_MASK;
294         uint32_t                rec_len = rec->hi & XFS_IEXT_LENGTH_MASK;
295
296         if (rec_offset > offset)
297                 return 1;
298         if (rec_offset + rec_len <= offset)
299                 return -1;
300         return 0;
301 }
302
303 static void *
304 xfs_iext_find_level(
305         struct xfs_ifork        *ifp,
306         xfs_fileoff_t           offset,
307         int                     level)
308 {
309         struct xfs_iext_node    *node = ifp->if_u1.if_root;
310         int                     height, i;
311
312         if (!ifp->if_height)
313                 return NULL;
314
315         for (height = ifp->if_height; height > level; height--) {
316                 for (i = 1; i < KEYS_PER_NODE; i++)
317                         if (xfs_iext_key_cmp(node, i, offset) > 0)
318                                 break;
319
320                 node = node->ptrs[i - 1];
321                 if (!node)
322                         break;
323         }
324
325         return node;
326 }
327
328 static int
329 xfs_iext_node_pos(
330         struct xfs_iext_node    *node,
331         xfs_fileoff_t           offset)
332 {
333         int                     i;
334
335         for (i = 1; i < KEYS_PER_NODE; i++) {
336                 if (xfs_iext_key_cmp(node, i, offset) > 0)
337                         break;
338         }
339
340         return i - 1;
341 }
342
343 static int
344 xfs_iext_node_insert_pos(
345         struct xfs_iext_node    *node,
346         xfs_fileoff_t           offset)
347 {
348         int                     i;
349
350         for (i = 0; i < KEYS_PER_NODE; i++) {
351                 if (xfs_iext_key_cmp(node, i, offset) > 0)
352                         return i;
353         }
354
355         return KEYS_PER_NODE;
356 }
357
358 static int
359 xfs_iext_node_nr_entries(
360         struct xfs_iext_node    *node,
361         int                     start)
362 {
363         int                     i;
364
365         for (i = start; i < KEYS_PER_NODE; i++) {
366                 if (node->keys[i] == XFS_IEXT_KEY_INVALID)
367                         break;
368         }
369
370         return i;
371 }
372
373 static int
374 xfs_iext_leaf_nr_entries(
375         struct xfs_ifork        *ifp,
376         struct xfs_iext_leaf    *leaf,
377         int                     start)
378 {
379         int                     i;
380
381         for (i = start; i < xfs_iext_max_recs(ifp); i++) {
382                 if (xfs_iext_rec_is_empty(&leaf->recs[i]))
383                         break;
384         }
385
386         return i;
387 }
388
389 static inline uint64_t
390 xfs_iext_leaf_key(
391         struct xfs_iext_leaf    *leaf,
392         int                     n)
393 {
394         return leaf->recs[n].lo & XFS_IEXT_STARTOFF_MASK;
395 }
396
397 static void
398 xfs_iext_grow(
399         struct xfs_ifork        *ifp)
400 {
401         struct xfs_iext_node    *node = kmem_zalloc(NODE_SIZE, KM_NOFS);
402         int                     i;
403
404         if (ifp->if_height == 1) {
405                 struct xfs_iext_leaf *prev = ifp->if_u1.if_root;
406
407                 node->keys[0] = xfs_iext_leaf_key(prev, 0);
408                 node->ptrs[0] = prev;
409         } else  {
410                 struct xfs_iext_node *prev = ifp->if_u1.if_root;
411
412                 ASSERT(ifp->if_height > 1);
413
414                 node->keys[0] = prev->keys[0];
415                 node->ptrs[0] = prev;
416         }
417
418         for (i = 1; i < KEYS_PER_NODE; i++)
419                 node->keys[i] = XFS_IEXT_KEY_INVALID;
420
421         ifp->if_u1.if_root = node;
422         ifp->if_height++;
423 }
424
425 static void
426 xfs_iext_update_node(
427         struct xfs_ifork        *ifp,
428         xfs_fileoff_t           old_offset,
429         xfs_fileoff_t           new_offset,
430         int                     level,
431         void                    *ptr)
432 {
433         struct xfs_iext_node    *node = ifp->if_u1.if_root;
434         int                     height, i;
435
436         for (height = ifp->if_height; height > level; height--) {
437                 for (i = 0; i < KEYS_PER_NODE; i++) {
438                         if (i > 0 && xfs_iext_key_cmp(node, i, old_offset) > 0)
439                                 break;
440                         if (node->keys[i] == old_offset)
441                                 node->keys[i] = new_offset;
442                 }
443                 node = node->ptrs[i - 1];
444                 ASSERT(node);
445         }
446
447         ASSERT(node == ptr);
448 }
449
450 static struct xfs_iext_node *
451 xfs_iext_split_node(
452         struct xfs_iext_node    **nodep,
453         int                     *pos,
454         int                     *nr_entries)
455 {
456         struct xfs_iext_node    *node = *nodep;
457         struct xfs_iext_node    *new = kmem_zalloc(NODE_SIZE, KM_NOFS);
458         const int               nr_move = KEYS_PER_NODE / 2;
459         int                     nr_keep = nr_move + (KEYS_PER_NODE & 1);
460         int                     i = 0;
461
462         /* for sequential append operations just spill over into the new node */
463         if (*pos == KEYS_PER_NODE) {
464                 *nodep = new;
465                 *pos = 0;
466                 *nr_entries = 0;
467                 goto done;
468         }
469
470
471         for (i = 0; i < nr_move; i++) {
472                 new->keys[i] = node->keys[nr_keep + i];
473                 new->ptrs[i] = node->ptrs[nr_keep + i];
474
475                 node->keys[nr_keep + i] = XFS_IEXT_KEY_INVALID;
476                 node->ptrs[nr_keep + i] = NULL;
477         }
478
479         if (*pos >= nr_keep) {
480                 *nodep = new;
481                 *pos -= nr_keep;
482                 *nr_entries = nr_move;
483         } else {
484                 *nr_entries = nr_keep;
485         }
486 done:
487         for (; i < KEYS_PER_NODE; i++)
488                 new->keys[i] = XFS_IEXT_KEY_INVALID;
489         return new;
490 }
491
492 static void
493 xfs_iext_insert_node(
494         struct xfs_ifork        *ifp,
495         uint64_t                offset,
496         void                    *ptr,
497         int                     level)
498 {
499         struct xfs_iext_node    *node, *new;
500         int                     i, pos, nr_entries;
501
502 again:
503         if (ifp->if_height < level)
504                 xfs_iext_grow(ifp);
505
506         new = NULL;
507         node = xfs_iext_find_level(ifp, offset, level);
508         pos = xfs_iext_node_insert_pos(node, offset);
509         nr_entries = xfs_iext_node_nr_entries(node, pos);
510
511         ASSERT(pos >= nr_entries || xfs_iext_key_cmp(node, pos, offset) != 0);
512         ASSERT(nr_entries <= KEYS_PER_NODE);
513
514         if (nr_entries == KEYS_PER_NODE)
515                 new = xfs_iext_split_node(&node, &pos, &nr_entries);
516
517         /*
518          * Update the pointers in higher levels if the first entry changes
519          * in an existing node.
520          */
521         if (node != new && pos == 0 && nr_entries > 0)
522                 xfs_iext_update_node(ifp, node->keys[0], offset, level, node);
523
524         for (i = nr_entries; i > pos; i--) {
525                 node->keys[i] = node->keys[i - 1];
526                 node->ptrs[i] = node->ptrs[i - 1];
527         }
528         node->keys[pos] = offset;
529         node->ptrs[pos] = ptr;
530
531         if (new) {
532                 offset = new->keys[0];
533                 ptr = new;
534                 level++;
535                 goto again;
536         }
537 }
538
539 static struct xfs_iext_leaf *
540 xfs_iext_split_leaf(
541         struct xfs_iext_cursor  *cur,
542         int                     *nr_entries)
543 {
544         struct xfs_iext_leaf    *leaf = cur->leaf;
545         struct xfs_iext_leaf    *new = kmem_zalloc(NODE_SIZE, KM_NOFS);
546         const int               nr_move = RECS_PER_LEAF / 2;
547         int                     nr_keep = nr_move + (RECS_PER_LEAF & 1);
548         int                     i;
549
550         /* for sequential append operations just spill over into the new node */
551         if (cur->pos == RECS_PER_LEAF) {
552                 cur->leaf = new;
553                 cur->pos = 0;
554                 *nr_entries = 0;
555                 goto done;
556         }
557
558         for (i = 0; i < nr_move; i++) {
559                 new->recs[i] = leaf->recs[nr_keep + i];
560                 xfs_iext_rec_clear(&leaf->recs[nr_keep + i]);
561         }
562
563         if (cur->pos >= nr_keep) {
564                 cur->leaf = new;
565                 cur->pos -= nr_keep;
566                 *nr_entries = nr_move;
567         } else {
568                 *nr_entries = nr_keep;
569         }
570 done:
571         if (leaf->next)
572                 leaf->next->prev = new;
573         new->next = leaf->next;
574         new->prev = leaf;
575         leaf->next = new;
576         return new;
577 }
578
579 static void
580 xfs_iext_alloc_root(
581         struct xfs_ifork        *ifp,
582         struct xfs_iext_cursor  *cur)
583 {
584         ASSERT(ifp->if_bytes == 0);
585
586         ifp->if_u1.if_root = kmem_zalloc(sizeof(struct xfs_iext_rec), KM_NOFS);
587         ifp->if_height = 1;
588
589         /* now that we have a node step into it */
590         cur->leaf = ifp->if_u1.if_root;
591         cur->pos = 0;
592 }
593
594 static void
595 xfs_iext_realloc_root(
596         struct xfs_ifork        *ifp,
597         struct xfs_iext_cursor  *cur)
598 {
599         int64_t new_size = ifp->if_bytes + sizeof(struct xfs_iext_rec);
600         void *new;
601
602         /* account for the prev/next pointers */
603         if (new_size / sizeof(struct xfs_iext_rec) == RECS_PER_LEAF)
604                 new_size = NODE_SIZE;
605
606         new = krealloc(ifp->if_u1.if_root, new_size, GFP_NOFS | __GFP_NOFAIL);
607         memset(new + ifp->if_bytes, 0, new_size - ifp->if_bytes);
608         ifp->if_u1.if_root = new;
609         cur->leaf = new;
610 }
611
612 /*
613  * Increment the sequence counter on extent tree changes. If we are on a COW
614  * fork, this allows the writeback code to skip looking for a COW extent if the
615  * COW fork hasn't changed. We use WRITE_ONCE here to ensure the update to the
616  * sequence counter is seen before the modifications to the extent tree itself
617  * take effect.
618  */
619 static inline void xfs_iext_inc_seq(struct xfs_ifork *ifp)
620 {
621         WRITE_ONCE(ifp->if_seq, READ_ONCE(ifp->if_seq) + 1);
622 }
623
624 void
625 xfs_iext_insert(
626         struct xfs_inode        *ip,
627         struct xfs_iext_cursor  *cur,
628         struct xfs_bmbt_irec    *irec,
629         int                     state)
630 {
631         struct xfs_ifork        *ifp = xfs_iext_state_to_fork(ip, state);
632         xfs_fileoff_t           offset = irec->br_startoff;
633         struct xfs_iext_leaf    *new = NULL;
634         int                     nr_entries, i;
635
636         xfs_iext_inc_seq(ifp);
637
638         if (ifp->if_height == 0)
639                 xfs_iext_alloc_root(ifp, cur);
640         else if (ifp->if_height == 1)
641                 xfs_iext_realloc_root(ifp, cur);
642
643         nr_entries = xfs_iext_leaf_nr_entries(ifp, cur->leaf, cur->pos);
644         ASSERT(nr_entries <= RECS_PER_LEAF);
645         ASSERT(cur->pos >= nr_entries ||
646                xfs_iext_rec_cmp(cur_rec(cur), irec->br_startoff) != 0);
647
648         if (nr_entries == RECS_PER_LEAF)
649                 new = xfs_iext_split_leaf(cur, &nr_entries);
650
651         /*
652          * Update the pointers in higher levels if the first entry changes
653          * in an existing node.
654          */
655         if (cur->leaf != new && cur->pos == 0 && nr_entries > 0) {
656                 xfs_iext_update_node(ifp, xfs_iext_leaf_key(cur->leaf, 0),
657                                 offset, 1, cur->leaf);
658         }
659
660         for (i = nr_entries; i > cur->pos; i--)
661                 cur->leaf->recs[i] = cur->leaf->recs[i - 1];
662         xfs_iext_set(cur_rec(cur), irec);
663         ifp->if_bytes += sizeof(struct xfs_iext_rec);
664
665         trace_xfs_iext_insert(ip, cur, state, _RET_IP_);
666
667         if (new)
668                 xfs_iext_insert_node(ifp, xfs_iext_leaf_key(new, 0), new, 2);
669 }
670
671 static struct xfs_iext_node *
672 xfs_iext_rebalance_node(
673         struct xfs_iext_node    *parent,
674         int                     *pos,
675         struct xfs_iext_node    *node,
676         int                     nr_entries)
677 {
678         /*
679          * If the neighbouring nodes are completely full, or have different
680          * parents, we might never be able to merge our node, and will only
681          * delete it once the number of entries hits zero.
682          */
683         if (nr_entries == 0)
684                 return node;
685
686         if (*pos > 0) {
687                 struct xfs_iext_node *prev = parent->ptrs[*pos - 1];
688                 int nr_prev = xfs_iext_node_nr_entries(prev, 0), i;
689
690                 if (nr_prev + nr_entries <= KEYS_PER_NODE) {
691                         for (i = 0; i < nr_entries; i++) {
692                                 prev->keys[nr_prev + i] = node->keys[i];
693                                 prev->ptrs[nr_prev + i] = node->ptrs[i];
694                         }
695                         return node;
696                 }
697         }
698
699         if (*pos + 1 < xfs_iext_node_nr_entries(parent, *pos)) {
700                 struct xfs_iext_node *next = parent->ptrs[*pos + 1];
701                 int nr_next = xfs_iext_node_nr_entries(next, 0), i;
702
703                 if (nr_entries + nr_next <= KEYS_PER_NODE) {
704                         /*
705                          * Merge the next node into this node so that we don't
706                          * have to do an additional update of the keys in the
707                          * higher levels.
708                          */
709                         for (i = 0; i < nr_next; i++) {
710                                 node->keys[nr_entries + i] = next->keys[i];
711                                 node->ptrs[nr_entries + i] = next->ptrs[i];
712                         }
713
714                         ++*pos;
715                         return next;
716                 }
717         }
718
719         return NULL;
720 }
721
722 static void
723 xfs_iext_remove_node(
724         struct xfs_ifork        *ifp,
725         xfs_fileoff_t           offset,
726         void                    *victim)
727 {
728         struct xfs_iext_node    *node, *parent;
729         int                     level = 2, pos, nr_entries, i;
730
731         ASSERT(level <= ifp->if_height);
732         node = xfs_iext_find_level(ifp, offset, level);
733         pos = xfs_iext_node_pos(node, offset);
734 again:
735         ASSERT(node->ptrs[pos]);
736         ASSERT(node->ptrs[pos] == victim);
737         kmem_free(victim);
738
739         nr_entries = xfs_iext_node_nr_entries(node, pos) - 1;
740         offset = node->keys[0];
741         for (i = pos; i < nr_entries; i++) {
742                 node->keys[i] = node->keys[i + 1];
743                 node->ptrs[i] = node->ptrs[i + 1];
744         }
745         node->keys[nr_entries] = XFS_IEXT_KEY_INVALID;
746         node->ptrs[nr_entries] = NULL;
747
748         if (pos == 0 && nr_entries > 0) {
749                 xfs_iext_update_node(ifp, offset, node->keys[0], level, node);
750                 offset = node->keys[0];
751         }
752
753         if (nr_entries >= KEYS_PER_NODE / 2)
754                 return;
755
756         if (level < ifp->if_height) {
757                 /*
758                  * If we aren't at the root yet try to find a neighbour node to
759                  * merge with (or delete the node if it is empty), and then
760                  * recurse up to the next level.
761                  */
762                 level++;
763                 parent = xfs_iext_find_level(ifp, offset, level);
764                 pos = xfs_iext_node_pos(parent, offset);
765
766                 ASSERT(pos != KEYS_PER_NODE);
767                 ASSERT(parent->ptrs[pos] == node);
768
769                 node = xfs_iext_rebalance_node(parent, &pos, node, nr_entries);
770                 if (node) {
771                         victim = node;
772                         node = parent;
773                         goto again;
774                 }
775         } else if (nr_entries == 1) {
776                 /*
777                  * If we are at the root and only one entry is left we can just
778                  * free this node and update the root pointer.
779                  */
780                 ASSERT(node == ifp->if_u1.if_root);
781                 ifp->if_u1.if_root = node->ptrs[0];
782                 ifp->if_height--;
783                 kmem_free(node);
784         }
785 }
786
787 static void
788 xfs_iext_rebalance_leaf(
789         struct xfs_ifork        *ifp,
790         struct xfs_iext_cursor  *cur,
791         struct xfs_iext_leaf    *leaf,
792         xfs_fileoff_t           offset,
793         int                     nr_entries)
794 {
795         /*
796          * If the neighbouring nodes are completely full we might never be able
797          * to merge our node, and will only delete it once the number of
798          * entries hits zero.
799          */
800         if (nr_entries == 0)
801                 goto remove_node;
802
803         if (leaf->prev) {
804                 int nr_prev = xfs_iext_leaf_nr_entries(ifp, leaf->prev, 0), i;
805
806                 if (nr_prev + nr_entries <= RECS_PER_LEAF) {
807                         for (i = 0; i < nr_entries; i++)
808                                 leaf->prev->recs[nr_prev + i] = leaf->recs[i];
809
810                         if (cur->leaf == leaf) {
811                                 cur->leaf = leaf->prev;
812                                 cur->pos += nr_prev;
813                         }
814                         goto remove_node;
815                 }
816         }
817
818         if (leaf->next) {
819                 int nr_next = xfs_iext_leaf_nr_entries(ifp, leaf->next, 0), i;
820
821                 if (nr_entries + nr_next <= RECS_PER_LEAF) {
822                         /*
823                          * Merge the next node into this node so that we don't
824                          * have to do an additional update of the keys in the
825                          * higher levels.
826                          */
827                         for (i = 0; i < nr_next; i++) {
828                                 leaf->recs[nr_entries + i] =
829                                         leaf->next->recs[i];
830                         }
831
832                         if (cur->leaf == leaf->next) {
833                                 cur->leaf = leaf;
834                                 cur->pos += nr_entries;
835                         }
836
837                         offset = xfs_iext_leaf_key(leaf->next, 0);
838                         leaf = leaf->next;
839                         goto remove_node;
840                 }
841         }
842
843         return;
844 remove_node:
845         if (leaf->prev)
846                 leaf->prev->next = leaf->next;
847         if (leaf->next)
848                 leaf->next->prev = leaf->prev;
849         xfs_iext_remove_node(ifp, offset, leaf);
850 }
851
852 static void
853 xfs_iext_free_last_leaf(
854         struct xfs_ifork        *ifp)
855 {
856         ifp->if_height--;
857         kmem_free(ifp->if_u1.if_root);
858         ifp->if_u1.if_root = NULL;
859 }
860
861 void
862 xfs_iext_remove(
863         struct xfs_inode        *ip,
864         struct xfs_iext_cursor  *cur,
865         int                     state)
866 {
867         struct xfs_ifork        *ifp = xfs_iext_state_to_fork(ip, state);
868         struct xfs_iext_leaf    *leaf = cur->leaf;
869         xfs_fileoff_t           offset = xfs_iext_leaf_key(leaf, 0);
870         int                     i, nr_entries;
871
872         trace_xfs_iext_remove(ip, cur, state, _RET_IP_);
873
874         ASSERT(ifp->if_height > 0);
875         ASSERT(ifp->if_u1.if_root != NULL);
876         ASSERT(xfs_iext_valid(ifp, cur));
877
878         xfs_iext_inc_seq(ifp);
879
880         nr_entries = xfs_iext_leaf_nr_entries(ifp, leaf, cur->pos) - 1;
881         for (i = cur->pos; i < nr_entries; i++)
882                 leaf->recs[i] = leaf->recs[i + 1];
883         xfs_iext_rec_clear(&leaf->recs[nr_entries]);
884         ifp->if_bytes -= sizeof(struct xfs_iext_rec);
885
886         if (cur->pos == 0 && nr_entries > 0) {
887                 xfs_iext_update_node(ifp, offset, xfs_iext_leaf_key(leaf, 0), 1,
888                                 leaf);
889                 offset = xfs_iext_leaf_key(leaf, 0);
890         } else if (cur->pos == nr_entries) {
891                 if (ifp->if_height > 1 && leaf->next)
892                         cur->leaf = leaf->next;
893                 else
894                         cur->leaf = NULL;
895                 cur->pos = 0;
896         }
897
898         if (nr_entries >= RECS_PER_LEAF / 2)
899                 return;
900
901         if (ifp->if_height > 1)
902                 xfs_iext_rebalance_leaf(ifp, cur, leaf, offset, nr_entries);
903         else if (nr_entries == 0)
904                 xfs_iext_free_last_leaf(ifp);
905 }
906
907 /*
908  * Lookup the extent covering bno.
909  *
910  * If there is an extent covering bno return the extent index, and store the
911  * expanded extent structure in *gotp, and the extent cursor in *cur.
912  * If there is no extent covering bno, but there is an extent after it (e.g.
913  * it lies in a hole) return that extent in *gotp and its cursor in *cur
914  * instead.
915  * If bno is beyond the last extent return false, and return an invalid
916  * cursor value.
917  */
918 bool
919 xfs_iext_lookup_extent(
920         struct xfs_inode        *ip,
921         struct xfs_ifork        *ifp,
922         xfs_fileoff_t           offset,
923         struct xfs_iext_cursor  *cur,
924         struct xfs_bmbt_irec    *gotp)
925 {
926         XFS_STATS_INC(ip->i_mount, xs_look_exlist);
927
928         cur->leaf = xfs_iext_find_level(ifp, offset, 1);
929         if (!cur->leaf) {
930                 cur->pos = 0;
931                 return false;
932         }
933
934         for (cur->pos = 0; cur->pos < xfs_iext_max_recs(ifp); cur->pos++) {
935                 struct xfs_iext_rec *rec = cur_rec(cur);
936
937                 if (xfs_iext_rec_is_empty(rec))
938                         break;
939                 if (xfs_iext_rec_cmp(rec, offset) >= 0)
940                         goto found;
941         }
942
943         /* Try looking in the next node for an entry > offset */
944         if (ifp->if_height == 1 || !cur->leaf->next)
945                 return false;
946         cur->leaf = cur->leaf->next;
947         cur->pos = 0;
948         if (!xfs_iext_valid(ifp, cur))
949                 return false;
950 found:
951         xfs_iext_get(gotp, cur_rec(cur));
952         return true;
953 }
954
955 /*
956  * Returns the last extent before end, and if this extent doesn't cover
957  * end, update end to the end of the extent.
958  */
959 bool
960 xfs_iext_lookup_extent_before(
961         struct xfs_inode        *ip,
962         struct xfs_ifork        *ifp,
963         xfs_fileoff_t           *end,
964         struct xfs_iext_cursor  *cur,
965         struct xfs_bmbt_irec    *gotp)
966 {
967         /* could be optimized to not even look up the next on a match.. */
968         if (xfs_iext_lookup_extent(ip, ifp, *end - 1, cur, gotp) &&
969             gotp->br_startoff <= *end - 1)
970                 return true;
971         if (!xfs_iext_prev_extent(ifp, cur, gotp))
972                 return false;
973         *end = gotp->br_startoff + gotp->br_blockcount;
974         return true;
975 }
976
977 void
978 xfs_iext_update_extent(
979         struct xfs_inode        *ip,
980         int                     state,
981         struct xfs_iext_cursor  *cur,
982         struct xfs_bmbt_irec    *new)
983 {
984         struct xfs_ifork        *ifp = xfs_iext_state_to_fork(ip, state);
985
986         xfs_iext_inc_seq(ifp);
987
988         if (cur->pos == 0) {
989                 struct xfs_bmbt_irec    old;
990
991                 xfs_iext_get(&old, cur_rec(cur));
992                 if (new->br_startoff != old.br_startoff) {
993                         xfs_iext_update_node(ifp, old.br_startoff,
994                                         new->br_startoff, 1, cur->leaf);
995                 }
996         }
997
998         trace_xfs_bmap_pre_update(ip, cur, state, _RET_IP_);
999         xfs_iext_set(cur_rec(cur), new);
1000         trace_xfs_bmap_post_update(ip, cur, state, _RET_IP_);
1001 }
1002
1003 /*
1004  * Return true if the cursor points at an extent and return the extent structure
1005  * in gotp.  Else return false.
1006  */
1007 bool
1008 xfs_iext_get_extent(
1009         struct xfs_ifork        *ifp,
1010         struct xfs_iext_cursor  *cur,
1011         struct xfs_bmbt_irec    *gotp)
1012 {
1013         if (!xfs_iext_valid(ifp, cur))
1014                 return false;
1015         xfs_iext_get(gotp, cur_rec(cur));
1016         return true;
1017 }
1018
1019 /*
1020  * This is a recursive function, because of that we need to be extremely
1021  * careful with stack usage.
1022  */
1023 static void
1024 xfs_iext_destroy_node(
1025         struct xfs_iext_node    *node,
1026         int                     level)
1027 {
1028         int                     i;
1029
1030         if (level > 1) {
1031                 for (i = 0; i < KEYS_PER_NODE; i++) {
1032                         if (node->keys[i] == XFS_IEXT_KEY_INVALID)
1033                                 break;
1034                         xfs_iext_destroy_node(node->ptrs[i], level - 1);
1035                 }
1036         }
1037
1038         kmem_free(node);
1039 }
1040
1041 void
1042 xfs_iext_destroy(
1043         struct xfs_ifork        *ifp)
1044 {
1045         xfs_iext_destroy_node(ifp->if_u1.if_root, ifp->if_height);
1046
1047         ifp->if_bytes = 0;
1048         ifp->if_height = 0;
1049         ifp->if_u1.if_root = NULL;
1050 }