1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * extent_map.c
5  *
6  * Block/Cluster mapping functions
7  *
8  * Copyright (C) 2004 Oracle.  All rights reserved.
9  *
10  * This program is free software; you can redistribute it and/or
11  * modify it under the terms of the GNU General Public
12  * License, version 2,  as published by the Free Software Foundation.
13  *
14  * This program is distributed in the hope that it will be useful,
15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
17  * General Public License for more details.
18  *
19  * You should have received a copy of the GNU General Public
20  * License along with this program; if not, write to the
21  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
22  * Boston, MA 021110-1307, USA.
23  */
24 
25 #include <linux/fs.h>
26 #include <linux/init.h>
27 #include <linux/slab.h>
28 #include <linux/types.h>
29 #include <linux/fiemap.h>
30 
31 #include <cluster/masklog.h>
32 
33 #include "ocfs2.h"
34 
35 #include "alloc.h"
36 #include "dlmglue.h"
37 #include "extent_map.h"
38 #include "inode.h"
39 #include "super.h"
40 #include "symlink.h"
41 #include "ocfs2_trace.h"
42 
43 #include "buffer_head_io.h"
44 
45 /*
46  * The extent caching implementation is intentionally trivial.
47  *
48  * We only cache a small number of extents stored directly on the
49  * inode, so linear order operations are acceptable. If we ever want
50  * to increase the size of the extent map, then these algorithms must
51  * get smarter.
52  */
53 
ocfs2_extent_map_init(struct inode * inode)54 void ocfs2_extent_map_init(struct inode *inode)
55 {
56 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
57 
58 	oi->ip_extent_map.em_num_items = 0;
59 	INIT_LIST_HEAD(&oi->ip_extent_map.em_list);
60 }
61 
__ocfs2_extent_map_lookup(struct ocfs2_extent_map * em,unsigned int cpos,struct ocfs2_extent_map_item ** ret_emi)62 static void __ocfs2_extent_map_lookup(struct ocfs2_extent_map *em,
63 				      unsigned int cpos,
64 				      struct ocfs2_extent_map_item **ret_emi)
65 {
66 	unsigned int range;
67 	struct ocfs2_extent_map_item *emi;
68 
69 	*ret_emi = NULL;
70 
71 	list_for_each_entry(emi, &em->em_list, ei_list) {
72 		range = emi->ei_cpos + emi->ei_clusters;
73 
74 		if (cpos >= emi->ei_cpos && cpos < range) {
75 			list_move(&emi->ei_list, &em->em_list);
76 
77 			*ret_emi = emi;
78 			break;
79 		}
80 	}
81 }
82 
ocfs2_extent_map_lookup(struct inode * inode,unsigned int cpos,unsigned int * phys,unsigned int * len,unsigned int * flags)83 static int ocfs2_extent_map_lookup(struct inode *inode, unsigned int cpos,
84 				   unsigned int *phys, unsigned int *len,
85 				   unsigned int *flags)
86 {
87 	unsigned int coff;
88 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
89 	struct ocfs2_extent_map_item *emi;
90 
91 	spin_lock(&oi->ip_lock);
92 
93 	__ocfs2_extent_map_lookup(&oi->ip_extent_map, cpos, &emi);
94 	if (emi) {
95 		coff = cpos - emi->ei_cpos;
96 		*phys = emi->ei_phys + coff;
97 		if (len)
98 			*len = emi->ei_clusters - coff;
99 		if (flags)
100 			*flags = emi->ei_flags;
101 	}
102 
103 	spin_unlock(&oi->ip_lock);
104 
105 	if (emi == NULL)
106 		return -ENOENT;
107 
108 	return 0;
109 }
110 
111 /*
112  * Forget about all clusters equal to or greater than cpos.
113  */
ocfs2_extent_map_trunc(struct inode * inode,unsigned int cpos)114 void ocfs2_extent_map_trunc(struct inode *inode, unsigned int cpos)
115 {
116 	struct ocfs2_extent_map_item *emi, *n;
117 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
118 	struct ocfs2_extent_map *em = &oi->ip_extent_map;
119 	LIST_HEAD(tmp_list);
120 	unsigned int range;
121 
122 	spin_lock(&oi->ip_lock);
123 	list_for_each_entry_safe(emi, n, &em->em_list, ei_list) {
124 		if (emi->ei_cpos >= cpos) {
125 			/* Full truncate of this record. */
126 			list_move(&emi->ei_list, &tmp_list);
127 			BUG_ON(em->em_num_items == 0);
128 			em->em_num_items--;
129 			continue;
130 		}
131 
132 		range = emi->ei_cpos + emi->ei_clusters;
133 		if (range > cpos) {
134 			/* Partial truncate */
135 			emi->ei_clusters = cpos - emi->ei_cpos;
136 		}
137 	}
138 	spin_unlock(&oi->ip_lock);
139 
140 	list_for_each_entry_safe(emi, n, &tmp_list, ei_list) {
141 		list_del(&emi->ei_list);
142 		kfree(emi);
143 	}
144 }
145 
146 /*
147  * Is any part of emi2 contained within emi1
148  */
ocfs2_ei_is_contained(struct ocfs2_extent_map_item * emi1,struct ocfs2_extent_map_item * emi2)149 static int ocfs2_ei_is_contained(struct ocfs2_extent_map_item *emi1,
150 				 struct ocfs2_extent_map_item *emi2)
151 {
152 	unsigned int range1, range2;
153 
154 	/*
155 	 * Check if logical start of emi2 is inside emi1
156 	 */
157 	range1 = emi1->ei_cpos + emi1->ei_clusters;
158 	if (emi2->ei_cpos >= emi1->ei_cpos && emi2->ei_cpos < range1)
159 		return 1;
160 
161 	/*
162 	 * Check if logical end of emi2 is inside emi1
163 	 */
164 	range2 = emi2->ei_cpos + emi2->ei_clusters;
165 	if (range2 > emi1->ei_cpos && range2 <= range1)
166 		return 1;
167 
168 	return 0;
169 }
170 
ocfs2_copy_emi_fields(struct ocfs2_extent_map_item * dest,struct ocfs2_extent_map_item * src)171 static void ocfs2_copy_emi_fields(struct ocfs2_extent_map_item *dest,
172 				  struct ocfs2_extent_map_item *src)
173 {
174 	dest->ei_cpos = src->ei_cpos;
175 	dest->ei_phys = src->ei_phys;
176 	dest->ei_clusters = src->ei_clusters;
177 	dest->ei_flags = src->ei_flags;
178 }
179 
180 /*
181  * Try to merge emi with ins. Returns 1 if merge succeeds, zero
182  * otherwise.
183  */
ocfs2_try_to_merge_extent_map(struct ocfs2_extent_map_item * emi,struct ocfs2_extent_map_item * ins)184 static int ocfs2_try_to_merge_extent_map(struct ocfs2_extent_map_item *emi,
185 					 struct ocfs2_extent_map_item *ins)
186 {
187 	/*
188 	 * Handle contiguousness
189 	 */
190 	if (ins->ei_phys == (emi->ei_phys + emi->ei_clusters) &&
191 	    ins->ei_cpos == (emi->ei_cpos + emi->ei_clusters) &&
192 	    ins->ei_flags == emi->ei_flags) {
193 		emi->ei_clusters += ins->ei_clusters;
194 		return 1;
195 	} else if ((ins->ei_phys + ins->ei_clusters) == emi->ei_phys &&
196 		   (ins->ei_cpos + ins->ei_clusters) == emi->ei_cpos &&
197 		   ins->ei_flags == emi->ei_flags) {
198 		emi->ei_phys = ins->ei_phys;
199 		emi->ei_cpos = ins->ei_cpos;
200 		emi->ei_clusters += ins->ei_clusters;
201 		return 1;
202 	}
203 
204 	/*
205 	 * Overlapping extents - this shouldn't happen unless we've
206 	 * split an extent to change it's flags. That is exceedingly
207 	 * rare, so there's no sense in trying to optimize it yet.
208 	 */
209 	if (ocfs2_ei_is_contained(emi, ins) ||
210 	    ocfs2_ei_is_contained(ins, emi)) {
211 		ocfs2_copy_emi_fields(emi, ins);
212 		return 1;
213 	}
214 
215 	/* No merge was possible. */
216 	return 0;
217 }
218 
219 /*
220  * In order to reduce complexity on the caller, this insert function
221  * is intentionally liberal in what it will accept.
222  *
223  * The only rule is that the truncate call *must* be used whenever
224  * records have been deleted. This avoids inserting overlapping
225  * records with different physical mappings.
226  */
ocfs2_extent_map_insert_rec(struct inode * inode,struct ocfs2_extent_rec * rec)227 void ocfs2_extent_map_insert_rec(struct inode *inode,
228 				 struct ocfs2_extent_rec *rec)
229 {
230 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
231 	struct ocfs2_extent_map *em = &oi->ip_extent_map;
232 	struct ocfs2_extent_map_item *emi, *new_emi = NULL;
233 	struct ocfs2_extent_map_item ins;
234 
235 	ins.ei_cpos = le32_to_cpu(rec->e_cpos);
236 	ins.ei_phys = ocfs2_blocks_to_clusters(inode->i_sb,
237 					       le64_to_cpu(rec->e_blkno));
238 	ins.ei_clusters = le16_to_cpu(rec->e_leaf_clusters);
239 	ins.ei_flags = rec->e_flags;
240 
241 search:
242 	spin_lock(&oi->ip_lock);
243 
244 	list_for_each_entry(emi, &em->em_list, ei_list) {
245 		if (ocfs2_try_to_merge_extent_map(emi, &ins)) {
246 			list_move(&emi->ei_list, &em->em_list);
247 			spin_unlock(&oi->ip_lock);
248 			goto out;
249 		}
250 	}
251 
252 	/*
253 	 * No item could be merged.
254 	 *
255 	 * Either allocate and add a new item, or overwrite the last recently
256 	 * inserted.
257 	 */
258 
259 	if (em->em_num_items < OCFS2_MAX_EXTENT_MAP_ITEMS) {
260 		if (new_emi == NULL) {
261 			spin_unlock(&oi->ip_lock);
262 
263 			new_emi = kmalloc(sizeof(*new_emi), GFP_NOFS);
264 			if (new_emi == NULL)
265 				goto out;
266 
267 			goto search;
268 		}
269 
270 		ocfs2_copy_emi_fields(new_emi, &ins);
271 		list_add(&new_emi->ei_list, &em->em_list);
272 		em->em_num_items++;
273 		new_emi = NULL;
274 	} else {
275 		BUG_ON(list_empty(&em->em_list) || em->em_num_items == 0);
276 		emi = list_entry(em->em_list.prev,
277 				 struct ocfs2_extent_map_item, ei_list);
278 		list_move(&emi->ei_list, &em->em_list);
279 		ocfs2_copy_emi_fields(emi, &ins);
280 	}
281 
282 	spin_unlock(&oi->ip_lock);
283 
284 out:
285 	if (new_emi)
286 		kfree(new_emi);
287 }
288 
ocfs2_last_eb_is_empty(struct inode * inode,struct ocfs2_dinode * di)289 static int ocfs2_last_eb_is_empty(struct inode *inode,
290 				  struct ocfs2_dinode *di)
291 {
292 	int ret, next_free;
293 	u64 last_eb_blk = le64_to_cpu(di->i_last_eb_blk);
294 	struct buffer_head *eb_bh = NULL;
295 	struct ocfs2_extent_block *eb;
296 	struct ocfs2_extent_list *el;
297 
298 	ret = ocfs2_read_extent_block(INODE_CACHE(inode), last_eb_blk, &eb_bh);
299 	if (ret) {
300 		mlog_errno(ret);
301 		goto out;
302 	}
303 
304 	eb = (struct ocfs2_extent_block *) eb_bh->b_data;
305 	el = &eb->h_list;
306 
307 	if (el->l_tree_depth) {
308 		ocfs2_error(inode->i_sb,
309 			    "Inode %lu has non zero tree depth in "
310 			    "leaf block %llu\n", inode->i_ino,
311 			    (unsigned long long)eb_bh->b_blocknr);
312 		ret = -EROFS;
313 		goto out;
314 	}
315 
316 	next_free = le16_to_cpu(el->l_next_free_rec);
317 
318 	if (next_free == 0 ||
319 	    (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0])))
320 		ret = 1;
321 
322 out:
323 	brelse(eb_bh);
324 	return ret;
325 }
326 
327 /*
328  * Return the 1st index within el which contains an extent start
329  * larger than v_cluster.
330  */
ocfs2_search_for_hole_index(struct ocfs2_extent_list * el,u32 v_cluster)331 static int ocfs2_search_for_hole_index(struct ocfs2_extent_list *el,
332 				       u32 v_cluster)
333 {
334 	int i;
335 	struct ocfs2_extent_rec *rec;
336 
337 	for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
338 		rec = &el->l_recs[i];
339 
340 		if (v_cluster < le32_to_cpu(rec->e_cpos))
341 			break;
342 	}
343 
344 	return i;
345 }
346 
347 /*
348  * Figure out the size of a hole which starts at v_cluster within the given
349  * extent list.
350  *
351  * If there is no more allocation past v_cluster, we return the maximum
352  * cluster size minus v_cluster.
353  *
354  * If we have in-inode extents, then el points to the dinode list and
355  * eb_bh is NULL. Otherwise, eb_bh should point to the extent block
356  * containing el.
357  */
ocfs2_figure_hole_clusters(struct ocfs2_caching_info * ci,struct ocfs2_extent_list * el,struct buffer_head * eb_bh,u32 v_cluster,u32 * num_clusters)358 int ocfs2_figure_hole_clusters(struct ocfs2_caching_info *ci,
359 			       struct ocfs2_extent_list *el,
360 			       struct buffer_head *eb_bh,
361 			       u32 v_cluster,
362 			       u32 *num_clusters)
363 {
364 	int ret, i;
365 	struct buffer_head *next_eb_bh = NULL;
366 	struct ocfs2_extent_block *eb, *next_eb;
367 
368 	i = ocfs2_search_for_hole_index(el, v_cluster);
369 
370 	if (i == le16_to_cpu(el->l_next_free_rec) && eb_bh) {
371 		eb = (struct ocfs2_extent_block *)eb_bh->b_data;
372 
373 		/*
374 		 * Check the next leaf for any extents.
375 		 */
376 
377 		if (le64_to_cpu(eb->h_next_leaf_blk) == 0ULL)
378 			goto no_more_extents;
379 
380 		ret = ocfs2_read_extent_block(ci,
381 					      le64_to_cpu(eb->h_next_leaf_blk),
382 					      &next_eb_bh);
383 		if (ret) {
384 			mlog_errno(ret);
385 			goto out;
386 		}
387 
388 		next_eb = (struct ocfs2_extent_block *)next_eb_bh->b_data;
389 		el = &next_eb->h_list;
390 		i = ocfs2_search_for_hole_index(el, v_cluster);
391 	}
392 
393 no_more_extents:
394 	if (i == le16_to_cpu(el->l_next_free_rec)) {
395 		/*
396 		 * We're at the end of our existing allocation. Just
397 		 * return the maximum number of clusters we could
398 		 * possibly allocate.
399 		 */
400 		*num_clusters = UINT_MAX - v_cluster;
401 	} else {
402 		*num_clusters = le32_to_cpu(el->l_recs[i].e_cpos) - v_cluster;
403 	}
404 
405 	ret = 0;
406 out:
407 	brelse(next_eb_bh);
408 	return ret;
409 }
410 
ocfs2_get_clusters_nocache(struct inode * inode,struct buffer_head * di_bh,u32 v_cluster,unsigned int * hole_len,struct ocfs2_extent_rec * ret_rec,unsigned int * is_last)411 static int ocfs2_get_clusters_nocache(struct inode *inode,
412 				      struct buffer_head *di_bh,
413 				      u32 v_cluster, unsigned int *hole_len,
414 				      struct ocfs2_extent_rec *ret_rec,
415 				      unsigned int *is_last)
416 {
417 	int i, ret, tree_height, len;
418 	struct ocfs2_dinode *di;
419 	struct ocfs2_extent_block *uninitialized_var(eb);
420 	struct ocfs2_extent_list *el;
421 	struct ocfs2_extent_rec *rec;
422 	struct buffer_head *eb_bh = NULL;
423 
424 	memset(ret_rec, 0, sizeof(*ret_rec));
425 	if (is_last)
426 		*is_last = 0;
427 
428 	di = (struct ocfs2_dinode *) di_bh->b_data;
429 	el = &di->id2.i_list;
430 	tree_height = le16_to_cpu(el->l_tree_depth);
431 
432 	if (tree_height > 0) {
433 		ret = ocfs2_find_leaf(INODE_CACHE(inode), el, v_cluster,
434 				      &eb_bh);
435 		if (ret) {
436 			mlog_errno(ret);
437 			goto out;
438 		}
439 
440 		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
441 		el = &eb->h_list;
442 
443 		if (el->l_tree_depth) {
444 			ocfs2_error(inode->i_sb,
445 				    "Inode %lu has non zero tree depth in "
446 				    "leaf block %llu\n", inode->i_ino,
447 				    (unsigned long long)eb_bh->b_blocknr);
448 			ret = -EROFS;
449 			goto out;
450 		}
451 	}
452 
453 	i = ocfs2_search_extent_list(el, v_cluster);
454 	if (i == -1) {
455 		/*
456 		 * Holes can be larger than the maximum size of an
457 		 * extent, so we return their lengths in a separate
458 		 * field.
459 		 */
460 		if (hole_len) {
461 			ret = ocfs2_figure_hole_clusters(INODE_CACHE(inode),
462 							 el, eb_bh,
463 							 v_cluster, &len);
464 			if (ret) {
465 				mlog_errno(ret);
466 				goto out;
467 			}
468 
469 			*hole_len = len;
470 		}
471 		goto out_hole;
472 	}
473 
474 	rec = &el->l_recs[i];
475 
476 	BUG_ON(v_cluster < le32_to_cpu(rec->e_cpos));
477 
478 	if (!rec->e_blkno) {
479 		ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
480 			    "record (%u, %u, 0)", inode->i_ino,
481 			    le32_to_cpu(rec->e_cpos),
482 			    ocfs2_rec_clusters(el, rec));
483 		ret = -EROFS;
484 		goto out;
485 	}
486 
487 	*ret_rec = *rec;
488 
489 	/*
490 	 * Checking for last extent is potentially expensive - we
491 	 * might have to look at the next leaf over to see if it's
492 	 * empty.
493 	 *
494 	 * The first two checks are to see whether the caller even
495 	 * cares for this information, and if the extent is at least
496 	 * the last in it's list.
497 	 *
498 	 * If those hold true, then the extent is last if any of the
499 	 * additional conditions hold true:
500 	 *  - Extent list is in-inode
501 	 *  - Extent list is right-most
502 	 *  - Extent list is 2nd to rightmost, with empty right-most
503 	 */
504 	if (is_last) {
505 		if (i == (le16_to_cpu(el->l_next_free_rec) - 1)) {
506 			if (tree_height == 0)
507 				*is_last = 1;
508 			else if (eb->h_blkno == di->i_last_eb_blk)
509 				*is_last = 1;
510 			else if (eb->h_next_leaf_blk == di->i_last_eb_blk) {
511 				ret = ocfs2_last_eb_is_empty(inode, di);
512 				if (ret < 0) {
513 					mlog_errno(ret);
514 					goto out;
515 				}
516 				if (ret == 1)
517 					*is_last = 1;
518 			}
519 		}
520 	}
521 
522 out_hole:
523 	ret = 0;
524 out:
525 	brelse(eb_bh);
526 	return ret;
527 }
528 
ocfs2_relative_extent_offsets(struct super_block * sb,u32 v_cluster,struct ocfs2_extent_rec * rec,u32 * p_cluster,u32 * num_clusters)529 static void ocfs2_relative_extent_offsets(struct super_block *sb,
530 					  u32 v_cluster,
531 					  struct ocfs2_extent_rec *rec,
532 					  u32 *p_cluster, u32 *num_clusters)
533 
534 {
535 	u32 coff = v_cluster - le32_to_cpu(rec->e_cpos);
536 
537 	*p_cluster = ocfs2_blocks_to_clusters(sb, le64_to_cpu(rec->e_blkno));
538 	*p_cluster = *p_cluster + coff;
539 
540 	if (num_clusters)
541 		*num_clusters = le16_to_cpu(rec->e_leaf_clusters) - coff;
542 }
543 
ocfs2_xattr_get_clusters(struct inode * inode,u32 v_cluster,u32 * p_cluster,u32 * num_clusters,struct ocfs2_extent_list * el,unsigned int * extent_flags)544 int ocfs2_xattr_get_clusters(struct inode *inode, u32 v_cluster,
545 			     u32 *p_cluster, u32 *num_clusters,
546 			     struct ocfs2_extent_list *el,
547 			     unsigned int *extent_flags)
548 {
549 	int ret = 0, i;
550 	struct buffer_head *eb_bh = NULL;
551 	struct ocfs2_extent_block *eb;
552 	struct ocfs2_extent_rec *rec;
553 	u32 coff;
554 
555 	if (el->l_tree_depth) {
556 		ret = ocfs2_find_leaf(INODE_CACHE(inode), el, v_cluster,
557 				      &eb_bh);
558 		if (ret) {
559 			mlog_errno(ret);
560 			goto out;
561 		}
562 
563 		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
564 		el = &eb->h_list;
565 
566 		if (el->l_tree_depth) {
567 			ocfs2_error(inode->i_sb,
568 				    "Inode %lu has non zero tree depth in "
569 				    "xattr leaf block %llu\n", inode->i_ino,
570 				    (unsigned long long)eb_bh->b_blocknr);
571 			ret = -EROFS;
572 			goto out;
573 		}
574 	}
575 
576 	i = ocfs2_search_extent_list(el, v_cluster);
577 	if (i == -1) {
578 		ret = -EROFS;
579 		mlog_errno(ret);
580 		goto out;
581 	} else {
582 		rec = &el->l_recs[i];
583 		BUG_ON(v_cluster < le32_to_cpu(rec->e_cpos));
584 
585 		if (!rec->e_blkno) {
586 			ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
587 				    "record (%u, %u, 0) in xattr", inode->i_ino,
588 				    le32_to_cpu(rec->e_cpos),
589 				    ocfs2_rec_clusters(el, rec));
590 			ret = -EROFS;
591 			goto out;
592 		}
593 		coff = v_cluster - le32_to_cpu(rec->e_cpos);
594 		*p_cluster = ocfs2_blocks_to_clusters(inode->i_sb,
595 						    le64_to_cpu(rec->e_blkno));
596 		*p_cluster = *p_cluster + coff;
597 		if (num_clusters)
598 			*num_clusters = ocfs2_rec_clusters(el, rec) - coff;
599 
600 		if (extent_flags)
601 			*extent_flags = rec->e_flags;
602 	}
603 out:
604 	if (eb_bh)
605 		brelse(eb_bh);
606 	return ret;
607 }
608 
ocfs2_get_clusters(struct inode * inode,u32 v_cluster,u32 * p_cluster,u32 * num_clusters,unsigned int * extent_flags)609 int ocfs2_get_clusters(struct inode *inode, u32 v_cluster,
610 		       u32 *p_cluster, u32 *num_clusters,
611 		       unsigned int *extent_flags)
612 {
613 	int ret;
614 	unsigned int uninitialized_var(hole_len), flags = 0;
615 	struct buffer_head *di_bh = NULL;
616 	struct ocfs2_extent_rec rec;
617 
618 	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
619 		ret = -ERANGE;
620 		mlog_errno(ret);
621 		goto out;
622 	}
623 
624 	ret = ocfs2_extent_map_lookup(inode, v_cluster, p_cluster,
625 				      num_clusters, extent_flags);
626 	if (ret == 0)
627 		goto out;
628 
629 	ret = ocfs2_read_inode_block(inode, &di_bh);
630 	if (ret) {
631 		mlog_errno(ret);
632 		goto out;
633 	}
634 
635 	ret = ocfs2_get_clusters_nocache(inode, di_bh, v_cluster, &hole_len,
636 					 &rec, NULL);
637 	if (ret) {
638 		mlog_errno(ret);
639 		goto out;
640 	}
641 
642 	if (rec.e_blkno == 0ULL) {
643 		/*
644 		 * A hole was found. Return some canned values that
645 		 * callers can key on. If asked for, num_clusters will
646 		 * be populated with the size of the hole.
647 		 */
648 		*p_cluster = 0;
649 		if (num_clusters) {
650 			*num_clusters = hole_len;
651 		}
652 	} else {
653 		ocfs2_relative_extent_offsets(inode->i_sb, v_cluster, &rec,
654 					      p_cluster, num_clusters);
655 		flags = rec.e_flags;
656 
657 		ocfs2_extent_map_insert_rec(inode, &rec);
658 	}
659 
660 	if (extent_flags)
661 		*extent_flags = flags;
662 
663 out:
664 	brelse(di_bh);
665 	return ret;
666 }
667 
668 /*
669  * This expects alloc_sem to be held. The allocation cannot change at
670  * all while the map is in the process of being updated.
671  */
ocfs2_extent_map_get_blocks(struct inode * inode,u64 v_blkno,u64 * p_blkno,u64 * ret_count,unsigned int * extent_flags)672 int ocfs2_extent_map_get_blocks(struct inode *inode, u64 v_blkno, u64 *p_blkno,
673 				u64 *ret_count, unsigned int *extent_flags)
674 {
675 	int ret;
676 	int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
677 	u32 cpos, num_clusters, p_cluster;
678 	u64 boff = 0;
679 
680 	cpos = ocfs2_blocks_to_clusters(inode->i_sb, v_blkno);
681 
682 	ret = ocfs2_get_clusters(inode, cpos, &p_cluster, &num_clusters,
683 				 extent_flags);
684 	if (ret) {
685 		mlog_errno(ret);
686 		goto out;
687 	}
688 
689 	/*
690 	 * p_cluster == 0 indicates a hole.
691 	 */
692 	if (p_cluster) {
693 		boff = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
694 		boff += (v_blkno & (u64)(bpc - 1));
695 	}
696 
697 	*p_blkno = boff;
698 
699 	if (ret_count) {
700 		*ret_count = ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
701 		*ret_count -= v_blkno & (u64)(bpc - 1);
702 	}
703 
704 out:
705 	return ret;
706 }
707 
708 /*
709  * The ocfs2_fiemap_inline() may be a little bit misleading, since
710  * it not only handles the fiemap for inlined files, but also deals
711  * with the fast symlink, cause they have no difference for extent
712  * mapping per se.
713  */
ocfs2_fiemap_inline(struct inode * inode,struct buffer_head * di_bh,struct fiemap_extent_info * fieinfo,u64 map_start)714 static int ocfs2_fiemap_inline(struct inode *inode, struct buffer_head *di_bh,
715 			       struct fiemap_extent_info *fieinfo,
716 			       u64 map_start)
717 {
718 	int ret;
719 	unsigned int id_count;
720 	struct ocfs2_dinode *di;
721 	u64 phys;
722 	u32 flags = FIEMAP_EXTENT_DATA_INLINE|FIEMAP_EXTENT_LAST;
723 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
724 
725 	di = (struct ocfs2_dinode *)di_bh->b_data;
726 	if (ocfs2_inode_is_fast_symlink(inode))
727 		id_count = ocfs2_fast_symlink_chars(inode->i_sb);
728 	else
729 		id_count = le16_to_cpu(di->id2.i_data.id_count);
730 
731 	if (map_start < id_count) {
732 		phys = oi->ip_blkno << inode->i_sb->s_blocksize_bits;
733 		if (ocfs2_inode_is_fast_symlink(inode))
734 			phys += offsetof(struct ocfs2_dinode, id2.i_symlink);
735 		else
736 			phys += offsetof(struct ocfs2_dinode,
737 					 id2.i_data.id_data);
738 
739 		ret = fiemap_fill_next_extent(fieinfo, 0, phys, id_count,
740 					      flags);
741 		if (ret < 0)
742 			return ret;
743 	}
744 
745 	return 0;
746 }
747 
748 #define OCFS2_FIEMAP_FLAGS	(FIEMAP_FLAG_SYNC)
749 
ocfs2_fiemap(struct inode * inode,struct fiemap_extent_info * fieinfo,u64 map_start,u64 map_len)750 int ocfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
751 		 u64 map_start, u64 map_len)
752 {
753 	int ret, is_last;
754 	u32 mapping_end, cpos;
755 	unsigned int hole_size;
756 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
757 	u64 len_bytes, phys_bytes, virt_bytes;
758 	struct buffer_head *di_bh = NULL;
759 	struct ocfs2_extent_rec rec;
760 
761 	ret = fiemap_check_flags(fieinfo, OCFS2_FIEMAP_FLAGS);
762 	if (ret)
763 		return ret;
764 
765 	ret = ocfs2_inode_lock(inode, &di_bh, 0);
766 	if (ret) {
767 		mlog_errno(ret);
768 		goto out;
769 	}
770 
771 	down_read(&OCFS2_I(inode)->ip_alloc_sem);
772 
773 	/*
774 	 * Handle inline-data and fast symlink separately.
775 	 */
776 	if ((OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) ||
777 	    ocfs2_inode_is_fast_symlink(inode)) {
778 		ret = ocfs2_fiemap_inline(inode, di_bh, fieinfo, map_start);
779 		goto out_unlock;
780 	}
781 
782 	cpos = map_start >> osb->s_clustersize_bits;
783 	mapping_end = ocfs2_clusters_for_bytes(inode->i_sb,
784 					       map_start + map_len);
785 	is_last = 0;
786 	while (cpos < mapping_end && !is_last) {
787 		u32 fe_flags;
788 
789 		ret = ocfs2_get_clusters_nocache(inode, di_bh, cpos,
790 						 &hole_size, &rec, &is_last);
791 		if (ret) {
792 			mlog_errno(ret);
793 			goto out_unlock;
794 		}
795 
796 		if (rec.e_blkno == 0ULL) {
797 			cpos += hole_size;
798 			continue;
799 		}
800 
801 		fe_flags = 0;
802 		if (rec.e_flags & OCFS2_EXT_UNWRITTEN)
803 			fe_flags |= FIEMAP_EXTENT_UNWRITTEN;
804 		if (rec.e_flags & OCFS2_EXT_REFCOUNTED)
805 			fe_flags |= FIEMAP_EXTENT_SHARED;
806 		if (is_last)
807 			fe_flags |= FIEMAP_EXTENT_LAST;
808 		len_bytes = (u64)le16_to_cpu(rec.e_leaf_clusters) << osb->s_clustersize_bits;
809 		phys_bytes = le64_to_cpu(rec.e_blkno) << osb->sb->s_blocksize_bits;
810 		virt_bytes = (u64)le32_to_cpu(rec.e_cpos) << osb->s_clustersize_bits;
811 
812 		ret = fiemap_fill_next_extent(fieinfo, virt_bytes, phys_bytes,
813 					      len_bytes, fe_flags);
814 		if (ret)
815 			break;
816 
817 		cpos = le32_to_cpu(rec.e_cpos)+ le16_to_cpu(rec.e_leaf_clusters);
818 	}
819 
820 	if (ret > 0)
821 		ret = 0;
822 
823 out_unlock:
824 	brelse(di_bh);
825 
826 	up_read(&OCFS2_I(inode)->ip_alloc_sem);
827 
828 	ocfs2_inode_unlock(inode, 0);
829 out:
830 
831 	return ret;
832 }
833 
ocfs2_seek_data_hole_offset(struct file * file,loff_t * offset,int origin)834 int ocfs2_seek_data_hole_offset(struct file *file, loff_t *offset, int origin)
835 {
836 	struct inode *inode = file->f_mapping->host;
837 	int ret;
838 	unsigned int is_last = 0, is_data = 0;
839 	u16 cs_bits = OCFS2_SB(inode->i_sb)->s_clustersize_bits;
840 	u32 cpos, cend, clen, hole_size;
841 	u64 extoff, extlen;
842 	struct buffer_head *di_bh = NULL;
843 	struct ocfs2_extent_rec rec;
844 
845 	BUG_ON(origin != SEEK_DATA && origin != SEEK_HOLE);
846 
847 	ret = ocfs2_inode_lock(inode, &di_bh, 0);
848 	if (ret) {
849 		mlog_errno(ret);
850 		goto out;
851 	}
852 
853 	down_read(&OCFS2_I(inode)->ip_alloc_sem);
854 
855 	if (*offset >= inode->i_size) {
856 		ret = -ENXIO;
857 		goto out_unlock;
858 	}
859 
860 	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
861 		if (origin == SEEK_HOLE)
862 			*offset = inode->i_size;
863 		goto out_unlock;
864 	}
865 
866 	clen = 0;
867 	cpos = *offset >> cs_bits;
868 	cend = ocfs2_clusters_for_bytes(inode->i_sb, inode->i_size);
869 
870 	while (cpos < cend && !is_last) {
871 		ret = ocfs2_get_clusters_nocache(inode, di_bh, cpos, &hole_size,
872 						 &rec, &is_last);
873 		if (ret) {
874 			mlog_errno(ret);
875 			goto out_unlock;
876 		}
877 
878 		extoff = cpos;
879 		extoff <<= cs_bits;
880 
881 		if (rec.e_blkno == 0ULL) {
882 			clen = hole_size;
883 			is_data = 0;
884 		} else {
885 			clen = le16_to_cpu(rec.e_leaf_clusters) -
886 				(cpos - le32_to_cpu(rec.e_cpos));
887 			is_data = (rec.e_flags & OCFS2_EXT_UNWRITTEN) ?  0 : 1;
888 		}
889 
890 		if ((!is_data && origin == SEEK_HOLE) ||
891 		    (is_data && origin == SEEK_DATA)) {
892 			if (extoff > *offset)
893 				*offset = extoff;
894 			goto out_unlock;
895 		}
896 
897 		if (!is_last)
898 			cpos += clen;
899 	}
900 
901 	if (origin == SEEK_HOLE) {
902 		extoff = cpos;
903 		extoff <<= cs_bits;
904 		extlen = clen;
905 		extlen <<=  cs_bits;
906 
907 		if ((extoff + extlen) > inode->i_size)
908 			extlen = inode->i_size - extoff;
909 		extoff += extlen;
910 		if (extoff > *offset)
911 			*offset = extoff;
912 		goto out_unlock;
913 	}
914 
915 	ret = -ENXIO;
916 
917 out_unlock:
918 
919 	brelse(di_bh);
920 
921 	up_read(&OCFS2_I(inode)->ip_alloc_sem);
922 
923 	ocfs2_inode_unlock(inode, 0);
924 out:
925 	if (ret && ret != -ENXIO)
926 		ret = -ENXIO;
927 	return ret;
928 }
929 
ocfs2_read_virt_blocks(struct inode * inode,u64 v_block,int nr,struct buffer_head * bhs[],int flags,int (* validate)(struct super_block * sb,struct buffer_head * bh))930 int ocfs2_read_virt_blocks(struct inode *inode, u64 v_block, int nr,
931 			   struct buffer_head *bhs[], int flags,
932 			   int (*validate)(struct super_block *sb,
933 					   struct buffer_head *bh))
934 {
935 	int rc = 0;
936 	u64 p_block, p_count;
937 	int i, count, done = 0;
938 
939 	trace_ocfs2_read_virt_blocks(
940 	     inode, (unsigned long long)v_block, nr, bhs, flags,
941 	     validate);
942 
943 	if (((v_block + nr - 1) << inode->i_sb->s_blocksize_bits) >=
944 	    i_size_read(inode)) {
945 		BUG_ON(!(flags & OCFS2_BH_READAHEAD));
946 		goto out;
947 	}
948 
949 	while (done < nr) {
950 		down_read(&OCFS2_I(inode)->ip_alloc_sem);
951 		rc = ocfs2_extent_map_get_blocks(inode, v_block + done,
952 						 &p_block, &p_count, NULL);
953 		up_read(&OCFS2_I(inode)->ip_alloc_sem);
954 		if (rc) {
955 			mlog_errno(rc);
956 			break;
957 		}
958 
959 		if (!p_block) {
960 			rc = -EIO;
961 			mlog(ML_ERROR,
962 			     "Inode #%llu contains a hole at offset %llu\n",
963 			     (unsigned long long)OCFS2_I(inode)->ip_blkno,
964 			     (unsigned long long)(v_block + done) <<
965 			     inode->i_sb->s_blocksize_bits);
966 			break;
967 		}
968 
969 		count = nr - done;
970 		if (p_count < count)
971 			count = p_count;
972 
973 		/*
974 		 * If the caller passed us bhs, they should have come
975 		 * from a previous readahead call to this function.  Thus,
976 		 * they should have the right b_blocknr.
977 		 */
978 		for (i = 0; i < count; i++) {
979 			if (!bhs[done + i])
980 				continue;
981 			BUG_ON(bhs[done + i]->b_blocknr != (p_block + i));
982 		}
983 
984 		rc = ocfs2_read_blocks(INODE_CACHE(inode), p_block, count,
985 				       bhs + done, flags, validate);
986 		if (rc) {
987 			mlog_errno(rc);
988 			break;
989 		}
990 		done += count;
991 	}
992 
993 out:
994 	return rc;
995 }
996 
997 
998