xref: /linux/fs/ocfs2/extent_map.c (revision f22f9aaf6c3d92ebd5ad9e67acc03afebaaeb289)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * extent_map.c
4  *
5  * Block/Cluster mapping functions
6  *
7  * Copyright (C) 2004 Oracle.  All rights reserved.
8  */
9 
10 #include <linux/fs.h>
11 #include <linux/init.h>
12 #include <linux/slab.h>
13 #include <linux/types.h>
14 #include <linux/fiemap.h>
15 
16 #include <cluster/masklog.h>
17 
18 #include "ocfs2.h"
19 
20 #include "alloc.h"
21 #include "dlmglue.h"
22 #include "extent_map.h"
23 #include "inode.h"
24 #include "super.h"
25 #include "symlink.h"
26 #include "aops.h"
27 #include "ocfs2_trace.h"
28 
29 #include "buffer_head_io.h"
30 
31 /*
32  * The extent caching implementation is intentionally trivial.
33  *
34  * We only cache a small number of extents stored directly on the
35  * inode, so linear order operations are acceptable. If we ever want
36  * to increase the size of the extent map, then these algorithms must
37  * get smarter.
38  */
39 
40 void ocfs2_extent_map_init(struct inode *inode)
41 {
42 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
43 
44 	oi->ip_extent_map.em_num_items = 0;
45 	INIT_LIST_HEAD(&oi->ip_extent_map.em_list);
46 }
47 
48 static void __ocfs2_extent_map_lookup(struct ocfs2_extent_map *em,
49 				      unsigned int cpos,
50 				      struct ocfs2_extent_map_item **ret_emi)
51 {
52 	unsigned int range;
53 	struct ocfs2_extent_map_item *emi;
54 
55 	*ret_emi = NULL;
56 
57 	list_for_each_entry(emi, &em->em_list, ei_list) {
58 		range = emi->ei_cpos + emi->ei_clusters;
59 
60 		if (cpos >= emi->ei_cpos && cpos < range) {
61 			list_move(&emi->ei_list, &em->em_list);
62 
63 			*ret_emi = emi;
64 			break;
65 		}
66 	}
67 }
68 
69 static int ocfs2_extent_map_lookup(struct inode *inode, unsigned int cpos,
70 				   unsigned int *phys, unsigned int *len,
71 				   unsigned int *flags)
72 {
73 	unsigned int coff;
74 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
75 	struct ocfs2_extent_map_item *emi;
76 
77 	spin_lock(&oi->ip_lock);
78 
79 	__ocfs2_extent_map_lookup(&oi->ip_extent_map, cpos, &emi);
80 	if (emi) {
81 		coff = cpos - emi->ei_cpos;
82 		*phys = emi->ei_phys + coff;
83 		if (len)
84 			*len = emi->ei_clusters - coff;
85 		if (flags)
86 			*flags = emi->ei_flags;
87 	}
88 
89 	spin_unlock(&oi->ip_lock);
90 
91 	if (emi == NULL)
92 		return -ENOENT;
93 
94 	return 0;
95 }
96 
97 /*
98  * Forget about all clusters equal to or greater than cpos.
99  */
100 void ocfs2_extent_map_trunc(struct inode *inode, unsigned int cpos)
101 {
102 	struct ocfs2_extent_map_item *emi, *n;
103 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
104 	struct ocfs2_extent_map *em = &oi->ip_extent_map;
105 	LIST_HEAD(tmp_list);
106 	unsigned int range;
107 
108 	spin_lock(&oi->ip_lock);
109 	list_for_each_entry_safe(emi, n, &em->em_list, ei_list) {
110 		if (emi->ei_cpos >= cpos) {
111 			/* Full truncate of this record. */
112 			list_move(&emi->ei_list, &tmp_list);
113 			BUG_ON(em->em_num_items == 0);
114 			em->em_num_items--;
115 			continue;
116 		}
117 
118 		range = emi->ei_cpos + emi->ei_clusters;
119 		if (range > cpos) {
120 			/* Partial truncate */
121 			emi->ei_clusters = cpos - emi->ei_cpos;
122 		}
123 	}
124 	spin_unlock(&oi->ip_lock);
125 
126 	list_for_each_entry_safe(emi, n, &tmp_list, ei_list) {
127 		list_del(&emi->ei_list);
128 		kfree(emi);
129 	}
130 }
131 
132 /*
133  * Is any part of emi2 contained within emi1
134  */
135 static int ocfs2_ei_is_contained(struct ocfs2_extent_map_item *emi1,
136 				 struct ocfs2_extent_map_item *emi2)
137 {
138 	unsigned int range1, range2;
139 
140 	/*
141 	 * Check if logical start of emi2 is inside emi1
142 	 */
143 	range1 = emi1->ei_cpos + emi1->ei_clusters;
144 	if (emi2->ei_cpos >= emi1->ei_cpos && emi2->ei_cpos < range1)
145 		return 1;
146 
147 	/*
148 	 * Check if logical end of emi2 is inside emi1
149 	 */
150 	range2 = emi2->ei_cpos + emi2->ei_clusters;
151 	if (range2 > emi1->ei_cpos && range2 <= range1)
152 		return 1;
153 
154 	return 0;
155 }
156 
157 static void ocfs2_copy_emi_fields(struct ocfs2_extent_map_item *dest,
158 				  struct ocfs2_extent_map_item *src)
159 {
160 	dest->ei_cpos = src->ei_cpos;
161 	dest->ei_phys = src->ei_phys;
162 	dest->ei_clusters = src->ei_clusters;
163 	dest->ei_flags = src->ei_flags;
164 }
165 
166 /*
167  * Try to merge emi with ins. Returns 1 if merge succeeds, zero
168  * otherwise.
169  */
170 static int ocfs2_try_to_merge_extent_map(struct ocfs2_extent_map_item *emi,
171 					 struct ocfs2_extent_map_item *ins)
172 {
173 	/*
174 	 * Handle contiguousness
175 	 */
176 	if (ins->ei_phys == (emi->ei_phys + emi->ei_clusters) &&
177 	    ins->ei_cpos == (emi->ei_cpos + emi->ei_clusters) &&
178 	    ins->ei_flags == emi->ei_flags) {
179 		emi->ei_clusters += ins->ei_clusters;
180 		return 1;
181 	} else if ((ins->ei_phys + ins->ei_clusters) == emi->ei_phys &&
182 		   (ins->ei_cpos + ins->ei_clusters) == emi->ei_cpos &&
183 		   ins->ei_flags == emi->ei_flags) {
184 		emi->ei_phys = ins->ei_phys;
185 		emi->ei_cpos = ins->ei_cpos;
186 		emi->ei_clusters += ins->ei_clusters;
187 		return 1;
188 	}
189 
190 	/*
191 	 * Overlapping extents - this shouldn't happen unless we've
192 	 * split an extent to change it's flags. That is exceedingly
193 	 * rare, so there's no sense in trying to optimize it yet.
194 	 */
195 	if (ocfs2_ei_is_contained(emi, ins) ||
196 	    ocfs2_ei_is_contained(ins, emi)) {
197 		ocfs2_copy_emi_fields(emi, ins);
198 		return 1;
199 	}
200 
201 	/* No merge was possible. */
202 	return 0;
203 }
204 
205 /*
206  * In order to reduce complexity on the caller, this insert function
207  * is intentionally liberal in what it will accept.
208  *
209  * The only rule is that the truncate call *must* be used whenever
210  * records have been deleted. This avoids inserting overlapping
211  * records with different physical mappings.
212  */
213 void ocfs2_extent_map_insert_rec(struct inode *inode,
214 				 struct ocfs2_extent_rec *rec)
215 {
216 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
217 	struct ocfs2_extent_map *em = &oi->ip_extent_map;
218 	struct ocfs2_extent_map_item *emi, *new_emi = NULL;
219 	struct ocfs2_extent_map_item ins;
220 
221 	ins.ei_cpos = le32_to_cpu(rec->e_cpos);
222 	ins.ei_phys = ocfs2_blocks_to_clusters(inode->i_sb,
223 					       le64_to_cpu(rec->e_blkno));
224 	ins.ei_clusters = le16_to_cpu(rec->e_leaf_clusters);
225 	ins.ei_flags = rec->e_flags;
226 
227 search:
228 	spin_lock(&oi->ip_lock);
229 
230 	list_for_each_entry(emi, &em->em_list, ei_list) {
231 		if (ocfs2_try_to_merge_extent_map(emi, &ins)) {
232 			list_move(&emi->ei_list, &em->em_list);
233 			spin_unlock(&oi->ip_lock);
234 			goto out;
235 		}
236 	}
237 
238 	/*
239 	 * No item could be merged.
240 	 *
241 	 * Either allocate and add a new item, or overwrite the last recently
242 	 * inserted.
243 	 */
244 
245 	if (em->em_num_items < OCFS2_MAX_EXTENT_MAP_ITEMS) {
246 		if (new_emi == NULL) {
247 			spin_unlock(&oi->ip_lock);
248 
249 			new_emi = kmalloc(sizeof(*new_emi), GFP_NOFS);
250 			if (new_emi == NULL)
251 				goto out;
252 
253 			goto search;
254 		}
255 
256 		ocfs2_copy_emi_fields(new_emi, &ins);
257 		list_add(&new_emi->ei_list, &em->em_list);
258 		em->em_num_items++;
259 		new_emi = NULL;
260 	} else {
261 		BUG_ON(list_empty(&em->em_list) || em->em_num_items == 0);
262 		emi = list_entry(em->em_list.prev,
263 				 struct ocfs2_extent_map_item, ei_list);
264 		list_move(&emi->ei_list, &em->em_list);
265 		ocfs2_copy_emi_fields(emi, &ins);
266 	}
267 
268 	spin_unlock(&oi->ip_lock);
269 
270 out:
271 	kfree(new_emi);
272 }
273 
274 static int ocfs2_last_eb_is_empty(struct inode *inode,
275 				  struct ocfs2_dinode *di)
276 {
277 	int ret, next_free;
278 	u64 last_eb_blk = le64_to_cpu(di->i_last_eb_blk);
279 	struct buffer_head *eb_bh = NULL;
280 	struct ocfs2_extent_block *eb;
281 	struct ocfs2_extent_list *el;
282 
283 	ret = ocfs2_read_extent_block(INODE_CACHE(inode), last_eb_blk, &eb_bh);
284 	if (ret) {
285 		mlog_errno(ret);
286 		goto out;
287 	}
288 
289 	eb = (struct ocfs2_extent_block *) eb_bh->b_data;
290 	el = &eb->h_list;
291 
292 	if (el->l_tree_depth) {
293 		ocfs2_error(inode->i_sb,
294 			    "Inode %lu has non zero tree depth in leaf block %llu\n",
295 			    inode->i_ino,
296 			    (unsigned long long)eb_bh->b_blocknr);
297 		ret = -EROFS;
298 		goto out;
299 	}
300 
301 	next_free = le16_to_cpu(el->l_next_free_rec);
302 
303 	if (next_free == 0 ||
304 	    (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0])))
305 		ret = 1;
306 
307 out:
308 	brelse(eb_bh);
309 	return ret;
310 }
311 
312 /*
313  * Return the 1st index within el which contains an extent start
314  * larger than v_cluster.
315  */
316 static int ocfs2_search_for_hole_index(struct ocfs2_extent_list *el,
317 				       u32 v_cluster)
318 {
319 	int i;
320 	struct ocfs2_extent_rec *rec;
321 
322 	for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
323 		rec = &el->l_recs[i];
324 
325 		if (v_cluster < le32_to_cpu(rec->e_cpos))
326 			break;
327 	}
328 
329 	return i;
330 }
331 
332 /*
333  * Figure out the size of a hole which starts at v_cluster within the given
334  * extent list.
335  *
336  * If there is no more allocation past v_cluster, we return the maximum
337  * cluster size minus v_cluster.
338  *
339  * If we have in-inode extents, then el points to the dinode list and
340  * eb_bh is NULL. Otherwise, eb_bh should point to the extent block
341  * containing el.
342  */
343 int ocfs2_figure_hole_clusters(struct ocfs2_caching_info *ci,
344 			       struct ocfs2_extent_list *el,
345 			       struct buffer_head *eb_bh,
346 			       u32 v_cluster,
347 			       u32 *num_clusters)
348 {
349 	int ret, i;
350 	struct buffer_head *next_eb_bh = NULL;
351 	struct ocfs2_extent_block *eb, *next_eb;
352 
353 	i = ocfs2_search_for_hole_index(el, v_cluster);
354 
355 	if (i == le16_to_cpu(el->l_next_free_rec) && eb_bh) {
356 		eb = (struct ocfs2_extent_block *)eb_bh->b_data;
357 
358 		/*
359 		 * Check the next leaf for any extents.
360 		 */
361 
362 		if (le64_to_cpu(eb->h_next_leaf_blk) == 0ULL)
363 			goto no_more_extents;
364 
365 		ret = ocfs2_read_extent_block(ci,
366 					      le64_to_cpu(eb->h_next_leaf_blk),
367 					      &next_eb_bh);
368 		if (ret) {
369 			mlog_errno(ret);
370 			goto out;
371 		}
372 
373 		next_eb = (struct ocfs2_extent_block *)next_eb_bh->b_data;
374 		el = &next_eb->h_list;
375 		i = ocfs2_search_for_hole_index(el, v_cluster);
376 	}
377 
378 no_more_extents:
379 	if (i == le16_to_cpu(el->l_next_free_rec)) {
380 		/*
381 		 * We're at the end of our existing allocation. Just
382 		 * return the maximum number of clusters we could
383 		 * possibly allocate.
384 		 */
385 		*num_clusters = UINT_MAX - v_cluster;
386 	} else {
387 		*num_clusters = le32_to_cpu(el->l_recs[i].e_cpos) - v_cluster;
388 	}
389 
390 	ret = 0;
391 out:
392 	brelse(next_eb_bh);
393 	return ret;
394 }
395 
396 static int ocfs2_get_clusters_nocache(struct inode *inode,
397 				      struct buffer_head *di_bh,
398 				      u32 v_cluster, unsigned int *hole_len,
399 				      struct ocfs2_extent_rec *ret_rec,
400 				      unsigned int *is_last)
401 {
402 	int i, ret, tree_height, len;
403 	struct ocfs2_dinode *di;
404 	struct ocfs2_extent_block *eb;
405 	struct ocfs2_extent_list *el;
406 	struct ocfs2_extent_rec *rec;
407 	struct buffer_head *eb_bh = NULL;
408 
409 	memset(ret_rec, 0, sizeof(*ret_rec));
410 	if (is_last)
411 		*is_last = 0;
412 
413 	di = (struct ocfs2_dinode *) di_bh->b_data;
414 	el = &di->id2.i_list;
415 	tree_height = le16_to_cpu(el->l_tree_depth);
416 
417 	if (tree_height > 0) {
418 		ret = ocfs2_find_leaf(INODE_CACHE(inode), el, v_cluster,
419 				      &eb_bh);
420 		if (ret) {
421 			mlog_errno(ret);
422 			goto out;
423 		}
424 
425 		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
426 		el = &eb->h_list;
427 
428 		if (el->l_tree_depth) {
429 			ocfs2_error(inode->i_sb,
430 				    "Inode %lu has non zero tree depth in leaf block %llu\n",
431 				    inode->i_ino,
432 				    (unsigned long long)eb_bh->b_blocknr);
433 			ret = -EROFS;
434 			goto out;
435 		}
436 	}
437 
438 	i = ocfs2_search_extent_list(el, v_cluster);
439 	if (i == -1) {
440 		/*
441 		 * Holes can be larger than the maximum size of an
442 		 * extent, so we return their lengths in a separate
443 		 * field.
444 		 */
445 		if (hole_len) {
446 			ret = ocfs2_figure_hole_clusters(INODE_CACHE(inode),
447 							 el, eb_bh,
448 							 v_cluster, &len);
449 			if (ret) {
450 				mlog_errno(ret);
451 				goto out;
452 			}
453 
454 			*hole_len = len;
455 		}
456 		goto out_hole;
457 	}
458 
459 	rec = &el->l_recs[i];
460 
461 	BUG_ON(v_cluster < le32_to_cpu(rec->e_cpos));
462 
463 	if (!rec->e_blkno) {
464 		ocfs2_error(inode->i_sb,
465 			    "Inode %lu has bad extent record (%u, %u, 0)\n",
466 			    inode->i_ino,
467 			    le32_to_cpu(rec->e_cpos),
468 			    ocfs2_rec_clusters(el, rec));
469 		ret = -EROFS;
470 		goto out;
471 	}
472 
473 	*ret_rec = *rec;
474 
475 	/*
476 	 * Checking for last extent is potentially expensive - we
477 	 * might have to look at the next leaf over to see if it's
478 	 * empty.
479 	 *
480 	 * The first two checks are to see whether the caller even
481 	 * cares for this information, and if the extent is at least
482 	 * the last in it's list.
483 	 *
484 	 * If those hold true, then the extent is last if any of the
485 	 * additional conditions hold true:
486 	 *  - Extent list is in-inode
487 	 *  - Extent list is right-most
488 	 *  - Extent list is 2nd to rightmost, with empty right-most
489 	 */
490 	if (is_last) {
491 		if (i == (le16_to_cpu(el->l_next_free_rec) - 1)) {
492 			if (tree_height == 0)
493 				*is_last = 1;
494 			else if (eb->h_blkno == di->i_last_eb_blk)
495 				*is_last = 1;
496 			else if (eb->h_next_leaf_blk == di->i_last_eb_blk) {
497 				ret = ocfs2_last_eb_is_empty(inode, di);
498 				if (ret < 0) {
499 					mlog_errno(ret);
500 					goto out;
501 				}
502 				if (ret == 1)
503 					*is_last = 1;
504 			}
505 		}
506 	}
507 
508 out_hole:
509 	ret = 0;
510 out:
511 	brelse(eb_bh);
512 	return ret;
513 }
514 
515 static void ocfs2_relative_extent_offsets(struct super_block *sb,
516 					  u32 v_cluster,
517 					  struct ocfs2_extent_rec *rec,
518 					  u32 *p_cluster, u32 *num_clusters)
519 
520 {
521 	u32 coff = v_cluster - le32_to_cpu(rec->e_cpos);
522 
523 	*p_cluster = ocfs2_blocks_to_clusters(sb, le64_to_cpu(rec->e_blkno));
524 	*p_cluster = *p_cluster + coff;
525 
526 	if (num_clusters)
527 		*num_clusters = le16_to_cpu(rec->e_leaf_clusters) - coff;
528 }
529 
530 int ocfs2_xattr_get_clusters(struct inode *inode, u32 v_cluster,
531 			     u32 *p_cluster, u32 *num_clusters,
532 			     struct ocfs2_extent_list *el,
533 			     unsigned int *extent_flags)
534 {
535 	int ret = 0, i;
536 	struct buffer_head *eb_bh = NULL;
537 	struct ocfs2_extent_block *eb;
538 	struct ocfs2_extent_rec *rec;
539 	u32 coff;
540 
541 	if (el->l_tree_depth) {
542 		ret = ocfs2_find_leaf(INODE_CACHE(inode), el, v_cluster,
543 				      &eb_bh);
544 		if (ret) {
545 			mlog_errno(ret);
546 			goto out;
547 		}
548 
549 		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
550 		el = &eb->h_list;
551 
552 		if (el->l_tree_depth) {
553 			ocfs2_error(inode->i_sb,
554 				    "Inode %lu has non zero tree depth in xattr leaf block %llu\n",
555 				    inode->i_ino,
556 				    (unsigned long long)eb_bh->b_blocknr);
557 			ret = -EROFS;
558 			goto out;
559 		}
560 	}
561 
562 	i = ocfs2_search_extent_list(el, v_cluster);
563 	if (i == -1) {
564 		ret = -EROFS;
565 		mlog_errno(ret);
566 		goto out;
567 	} else {
568 		rec = &el->l_recs[i];
569 		BUG_ON(v_cluster < le32_to_cpu(rec->e_cpos));
570 
571 		if (!rec->e_blkno) {
572 			ocfs2_error(inode->i_sb,
573 				    "Inode %lu has bad extent record (%u, %u, 0) in xattr\n",
574 				    inode->i_ino,
575 				    le32_to_cpu(rec->e_cpos),
576 				    ocfs2_rec_clusters(el, rec));
577 			ret = -EROFS;
578 			goto out;
579 		}
580 		coff = v_cluster - le32_to_cpu(rec->e_cpos);
581 		*p_cluster = ocfs2_blocks_to_clusters(inode->i_sb,
582 						    le64_to_cpu(rec->e_blkno));
583 		*p_cluster = *p_cluster + coff;
584 		if (num_clusters)
585 			*num_clusters = ocfs2_rec_clusters(el, rec) - coff;
586 
587 		if (extent_flags)
588 			*extent_flags = rec->e_flags;
589 	}
590 out:
591 	brelse(eb_bh);
592 	return ret;
593 }
594 
595 int ocfs2_get_clusters(struct inode *inode, u32 v_cluster,
596 		       u32 *p_cluster, u32 *num_clusters,
597 		       unsigned int *extent_flags)
598 {
599 	int ret;
600 	unsigned int hole_len, flags = 0;
601 	struct buffer_head *di_bh = NULL;
602 	struct ocfs2_extent_rec rec;
603 
604 	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
605 		ret = -ERANGE;
606 		mlog_errno(ret);
607 		goto out;
608 	}
609 
610 	ret = ocfs2_extent_map_lookup(inode, v_cluster, p_cluster,
611 				      num_clusters, extent_flags);
612 	if (ret == 0)
613 		goto out;
614 
615 	ret = ocfs2_read_inode_block(inode, &di_bh);
616 	if (ret) {
617 		mlog_errno(ret);
618 		goto out;
619 	}
620 
621 	ret = ocfs2_get_clusters_nocache(inode, di_bh, v_cluster, &hole_len,
622 					 &rec, NULL);
623 	if (ret) {
624 		mlog_errno(ret);
625 		goto out;
626 	}
627 
628 	if (rec.e_blkno == 0ULL) {
629 		/*
630 		 * A hole was found. Return some canned values that
631 		 * callers can key on. If asked for, num_clusters will
632 		 * be populated with the size of the hole.
633 		 */
634 		*p_cluster = 0;
635 		if (num_clusters) {
636 			*num_clusters = hole_len;
637 		}
638 	} else {
639 		ocfs2_relative_extent_offsets(inode->i_sb, v_cluster, &rec,
640 					      p_cluster, num_clusters);
641 		flags = rec.e_flags;
642 
643 		ocfs2_extent_map_insert_rec(inode, &rec);
644 	}
645 
646 	if (extent_flags)
647 		*extent_flags = flags;
648 
649 out:
650 	brelse(di_bh);
651 	return ret;
652 }
653 
654 /*
655  * This expects alloc_sem to be held. The allocation cannot change at
656  * all while the map is in the process of being updated.
657  */
658 int ocfs2_extent_map_get_blocks(struct inode *inode, u64 v_blkno, u64 *p_blkno,
659 				u64 *ret_count, unsigned int *extent_flags)
660 {
661 	int ret;
662 	int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
663 	u32 cpos, num_clusters, p_cluster;
664 	u64 boff = 0;
665 
666 	cpos = ocfs2_blocks_to_clusters(inode->i_sb, v_blkno);
667 
668 	ret = ocfs2_get_clusters(inode, cpos, &p_cluster, &num_clusters,
669 				 extent_flags);
670 	if (ret) {
671 		mlog_errno(ret);
672 		goto out;
673 	}
674 
675 	/*
676 	 * p_cluster == 0 indicates a hole.
677 	 */
678 	if (p_cluster) {
679 		boff = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
680 		boff += (v_blkno & (u64)(bpc - 1));
681 	}
682 
683 	*p_blkno = boff;
684 
685 	if (ret_count) {
686 		*ret_count = ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
687 		*ret_count -= v_blkno & (u64)(bpc - 1);
688 	}
689 
690 out:
691 	return ret;
692 }
693 
694 /*
695  * The ocfs2_fiemap_inline() may be a little bit misleading, since
696  * it not only handles the fiemap for inlined files, but also deals
697  * with the fast symlink, cause they have no difference for extent
698  * mapping per se.
699  */
700 static int ocfs2_fiemap_inline(struct inode *inode, struct buffer_head *di_bh,
701 			       struct fiemap_extent_info *fieinfo,
702 			       u64 map_start)
703 {
704 	int ret;
705 	unsigned int id_count;
706 	struct ocfs2_dinode *di;
707 	u64 phys;
708 	u32 flags = FIEMAP_EXTENT_DATA_INLINE|FIEMAP_EXTENT_LAST;
709 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
710 
711 	di = (struct ocfs2_dinode *)di_bh->b_data;
712 	if (ocfs2_inode_is_fast_symlink(inode))
713 		id_count = ocfs2_fast_symlink_chars(inode->i_sb);
714 	else
715 		id_count = le16_to_cpu(di->id2.i_data.id_count);
716 
717 	if (map_start < id_count) {
718 		phys = oi->ip_blkno << inode->i_sb->s_blocksize_bits;
719 		if (ocfs2_inode_is_fast_symlink(inode))
720 			phys += offsetof(struct ocfs2_dinode, id2.i_symlink);
721 		else
722 			phys += offsetof(struct ocfs2_dinode,
723 					 id2.i_data.id_data);
724 
725 		ret = fiemap_fill_next_extent(fieinfo, 0, phys, id_count,
726 					      flags);
727 		if (ret < 0)
728 			return ret;
729 	}
730 
731 	return 0;
732 }
733 
734 int ocfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
735 		 u64 map_start, u64 map_len)
736 {
737 	int ret, is_last;
738 	u32 mapping_end, cpos;
739 	unsigned int hole_size;
740 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
741 	u64 len_bytes, phys_bytes, virt_bytes;
742 	struct buffer_head *di_bh = NULL;
743 	struct ocfs2_extent_rec rec;
744 
745 	ret = fiemap_prep(inode, fieinfo, map_start, &map_len, 0);
746 	if (ret)
747 		return ret;
748 
749 	ret = ocfs2_inode_lock(inode, &di_bh, 0);
750 	if (ret) {
751 		mlog_errno(ret);
752 		goto out;
753 	}
754 
755 	down_read(&OCFS2_I(inode)->ip_alloc_sem);
756 
757 	/*
758 	 * Handle inline-data and fast symlink separately.
759 	 */
760 	if ((OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) ||
761 	    ocfs2_inode_is_fast_symlink(inode)) {
762 		ret = ocfs2_fiemap_inline(inode, di_bh, fieinfo, map_start);
763 		goto out_unlock;
764 	}
765 
766 	cpos = map_start >> osb->s_clustersize_bits;
767 	mapping_end = ocfs2_clusters_for_bytes(inode->i_sb,
768 					       map_start + map_len);
769 	is_last = 0;
770 	while (cpos < mapping_end && !is_last) {
771 		u32 fe_flags;
772 
773 		ret = ocfs2_get_clusters_nocache(inode, di_bh, cpos,
774 						 &hole_size, &rec, &is_last);
775 		if (ret) {
776 			mlog_errno(ret);
777 			goto out_unlock;
778 		}
779 
780 		if (rec.e_blkno == 0ULL) {
781 			cpos += hole_size;
782 			continue;
783 		}
784 
785 		fe_flags = 0;
786 		if (rec.e_flags & OCFS2_EXT_UNWRITTEN)
787 			fe_flags |= FIEMAP_EXTENT_UNWRITTEN;
788 		if (rec.e_flags & OCFS2_EXT_REFCOUNTED)
789 			fe_flags |= FIEMAP_EXTENT_SHARED;
790 		if (is_last)
791 			fe_flags |= FIEMAP_EXTENT_LAST;
792 		len_bytes = (u64)le16_to_cpu(rec.e_leaf_clusters) << osb->s_clustersize_bits;
793 		phys_bytes = le64_to_cpu(rec.e_blkno) << osb->sb->s_blocksize_bits;
794 		virt_bytes = (u64)le32_to_cpu(rec.e_cpos) << osb->s_clustersize_bits;
795 
796 		ret = fiemap_fill_next_extent(fieinfo, virt_bytes, phys_bytes,
797 					      len_bytes, fe_flags);
798 		if (ret)
799 			break;
800 
801 		cpos = le32_to_cpu(rec.e_cpos)+ le16_to_cpu(rec.e_leaf_clusters);
802 	}
803 
804 	if (ret > 0)
805 		ret = 0;
806 
807 out_unlock:
808 	brelse(di_bh);
809 
810 	up_read(&OCFS2_I(inode)->ip_alloc_sem);
811 
812 	ocfs2_inode_unlock(inode, 0);
813 out:
814 
815 	return ret;
816 }
817 
818 /* Is IO overwriting allocated blocks? */
819 int ocfs2_overwrite_io(struct inode *inode, struct buffer_head *di_bh,
820 		       u64 map_start, u64 map_len)
821 {
822 	int ret = 0, is_last;
823 	u32 mapping_end, cpos;
824 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
825 	struct ocfs2_extent_rec rec;
826 
827 	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
828 		if (ocfs2_size_fits_inline_data(di_bh, map_start + map_len))
829 			return ret;
830 		else
831 			return -EAGAIN;
832 	}
833 
834 	cpos = map_start >> osb->s_clustersize_bits;
835 	mapping_end = ocfs2_clusters_for_bytes(inode->i_sb,
836 					       map_start + map_len);
837 	is_last = 0;
838 	while (cpos < mapping_end && !is_last) {
839 		ret = ocfs2_get_clusters_nocache(inode, di_bh, cpos,
840 						 NULL, &rec, &is_last);
841 		if (ret) {
842 			mlog_errno(ret);
843 			goto out;
844 		}
845 
846 		if (rec.e_blkno == 0ULL)
847 			break;
848 
849 		if (rec.e_flags & OCFS2_EXT_REFCOUNTED)
850 			break;
851 
852 		cpos = le32_to_cpu(rec.e_cpos) +
853 			le16_to_cpu(rec.e_leaf_clusters);
854 	}
855 
856 	if (cpos < mapping_end)
857 		ret = -EAGAIN;
858 out:
859 	return ret;
860 }
861 
862 int ocfs2_seek_data_hole_offset(struct file *file, loff_t *offset, int whence)
863 {
864 	struct inode *inode = file->f_mapping->host;
865 	int ret;
866 	unsigned int is_last = 0, is_data = 0;
867 	u16 cs_bits = OCFS2_SB(inode->i_sb)->s_clustersize_bits;
868 	u32 cpos, cend, clen, hole_size;
869 	u64 extoff, extlen;
870 	struct buffer_head *di_bh = NULL;
871 	struct ocfs2_extent_rec rec;
872 
873 	BUG_ON(whence != SEEK_DATA && whence != SEEK_HOLE);
874 
875 	ret = ocfs2_inode_lock(inode, &di_bh, 0);
876 	if (ret) {
877 		mlog_errno(ret);
878 		goto out;
879 	}
880 
881 	down_read(&OCFS2_I(inode)->ip_alloc_sem);
882 
883 	if (*offset >= i_size_read(inode)) {
884 		ret = -ENXIO;
885 		goto out_unlock;
886 	}
887 
888 	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
889 		if (whence == SEEK_HOLE)
890 			*offset = i_size_read(inode);
891 		goto out_unlock;
892 	}
893 
894 	clen = 0;
895 	cpos = *offset >> cs_bits;
896 	cend = ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode));
897 
898 	while (cpos < cend && !is_last) {
899 		ret = ocfs2_get_clusters_nocache(inode, di_bh, cpos, &hole_size,
900 						 &rec, &is_last);
901 		if (ret) {
902 			mlog_errno(ret);
903 			goto out_unlock;
904 		}
905 
906 		extoff = cpos;
907 		extoff <<= cs_bits;
908 
909 		if (rec.e_blkno == 0ULL) {
910 			clen = hole_size;
911 			is_data = 0;
912 		} else {
913 			clen = le16_to_cpu(rec.e_leaf_clusters) -
914 				(cpos - le32_to_cpu(rec.e_cpos));
915 			is_data = (rec.e_flags & OCFS2_EXT_UNWRITTEN) ?  0 : 1;
916 		}
917 
918 		if ((!is_data && whence == SEEK_HOLE) ||
919 		    (is_data && whence == SEEK_DATA)) {
920 			if (extoff > *offset)
921 				*offset = extoff;
922 			goto out_unlock;
923 		}
924 
925 		if (!is_last)
926 			cpos += clen;
927 	}
928 
929 	if (whence == SEEK_HOLE) {
930 		extoff = cpos;
931 		extoff <<= cs_bits;
932 		extlen = clen;
933 		extlen <<=  cs_bits;
934 
935 		if ((extoff + extlen) > i_size_read(inode))
936 			extlen = i_size_read(inode) - extoff;
937 		extoff += extlen;
938 		if (extoff > *offset)
939 			*offset = extoff;
940 		goto out_unlock;
941 	}
942 
943 	ret = -ENXIO;
944 
945 out_unlock:
946 
947 	brelse(di_bh);
948 
949 	up_read(&OCFS2_I(inode)->ip_alloc_sem);
950 
951 	ocfs2_inode_unlock(inode, 0);
952 out:
953 	return ret;
954 }
955 
956 int ocfs2_read_virt_blocks(struct inode *inode, u64 v_block, int nr,
957 			   struct buffer_head *bhs[], int flags,
958 			   int (*validate)(struct super_block *sb,
959 					   struct buffer_head *bh))
960 {
961 	int rc = 0;
962 	u64 p_block, p_count;
963 	int i, count, done = 0;
964 
965 	trace_ocfs2_read_virt_blocks(
966 	     inode, (unsigned long long)v_block, nr, bhs, flags,
967 	     validate);
968 
969 	if (((v_block + nr - 1) << inode->i_sb->s_blocksize_bits) >=
970 	    i_size_read(inode)) {
971 		BUG_ON(!(flags & OCFS2_BH_READAHEAD));
972 		goto out;
973 	}
974 
975 	while (done < nr) {
976 		down_read(&OCFS2_I(inode)->ip_alloc_sem);
977 		rc = ocfs2_extent_map_get_blocks(inode, v_block + done,
978 						 &p_block, &p_count, NULL);
979 		up_read(&OCFS2_I(inode)->ip_alloc_sem);
980 		if (rc) {
981 			mlog_errno(rc);
982 			break;
983 		}
984 
985 		if (!p_block) {
986 			rc = -EIO;
987 			mlog(ML_ERROR,
988 			     "Inode #%llu contains a hole at offset %llu\n",
989 			     (unsigned long long)OCFS2_I(inode)->ip_blkno,
990 			     (unsigned long long)(v_block + done) <<
991 			     inode->i_sb->s_blocksize_bits);
992 			break;
993 		}
994 
995 		count = nr - done;
996 		if (p_count < count)
997 			count = p_count;
998 
999 		/*
1000 		 * If the caller passed us bhs, they should have come
1001 		 * from a previous readahead call to this function.  Thus,
1002 		 * they should have the right b_blocknr.
1003 		 */
1004 		for (i = 0; i < count; i++) {
1005 			if (!bhs[done + i])
1006 				continue;
1007 			BUG_ON(bhs[done + i]->b_blocknr != (p_block + i));
1008 		}
1009 
1010 		rc = ocfs2_read_blocks(INODE_CACHE(inode), p_block, count,
1011 				       bhs + done, flags, validate);
1012 		if (rc) {
1013 			mlog_errno(rc);
1014 			break;
1015 		}
1016 		done += count;
1017 	}
1018 
1019 out:
1020 	return rc;
1021 }
1022 
1023 
1024