extent_map.c

   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
  40void 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
  48static 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
  69static 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 */
 100void 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 */
 135static 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
 157static 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 */
 170static 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 */
 213void 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
 227search:
 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
 270out:
 271	kfree(new_emi);
 272}
 273
 274static 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
 307out:
 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 */
 316static 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 */
 343int 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
 378no_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;
 391out:
 392	brelse(next_eb_bh);
 393	return ret;
 394}
 395
 396static 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	if (le16_to_cpu(el->l_next_free_rec) > le16_to_cpu(el->l_count)) {
 439		ocfs2_error(inode->i_sb,
 440			    "Inode %lu has an invalid extent (next_free_rec %u, count %u)\n",
 441			    inode->i_ino,
 442			    le16_to_cpu(el->l_next_free_rec),
 443			    le16_to_cpu(el->l_count));
 444		ret = -EROFS;
 445		goto out;
 446	}
 447
 448	i = ocfs2_search_extent_list(el, v_cluster);
 449	if (i == -1) {
 450		/*
 451		 * Holes can be larger than the maximum size of an
 452		 * extent, so we return their lengths in a separate
 453		 * field.
 454		 */
 455		if (hole_len) {
 456			ret = ocfs2_figure_hole_clusters(INODE_CACHE(inode),
 457							 el, eb_bh,
 458							 v_cluster, &len);
 459			if (ret) {
 460				mlog_errno(ret);
 461				goto out;
 462			}
 463
 464			*hole_len = len;
 465		}
 466		goto out_hole;
 467	}
 468
 469	rec = &el->l_recs[i];
 470
 471	BUG_ON(v_cluster < le32_to_cpu(rec->e_cpos));
 472
 473	if (!rec->e_blkno) {
 474		ocfs2_error(inode->i_sb,
 475			    "Inode %lu has bad extent record (%u, %u, 0)\n",
 476			    inode->i_ino,
 477			    le32_to_cpu(rec->e_cpos),
 478			    ocfs2_rec_clusters(el, rec));
 479		ret = -EROFS;
 480		goto out;
 481	}
 482
 483	*ret_rec = *rec;
 484
 485	/*
 486	 * Checking for last extent is potentially expensive - we
 487	 * might have to look at the next leaf over to see if it's
 488	 * empty.
 489	 *
 490	 * The first two checks are to see whether the caller even
 491	 * cares for this information, and if the extent is at least
 492	 * the last in it's list.
 493	 *
 494	 * If those hold true, then the extent is last if any of the
 495	 * additional conditions hold true:
 496	 *  - Extent list is in-inode
 497	 *  - Extent list is right-most
 498	 *  - Extent list is 2nd to rightmost, with empty right-most
 499	 */
 500	if (is_last) {
 501		if (i == (le16_to_cpu(el->l_next_free_rec) - 1)) {
 502			if (tree_height == 0)
 503				*is_last = 1;
 504			else if (eb->h_blkno == di->i_last_eb_blk)
 505				*is_last = 1;
 506			else if (eb->h_next_leaf_blk == di->i_last_eb_blk) {
 507				ret = ocfs2_last_eb_is_empty(inode, di);
 508				if (ret < 0) {
 509					mlog_errno(ret);
 510					goto out;
 511				}
 512				if (ret == 1)
 513					*is_last = 1;
 514			}
 515		}
 516	}
 517
 518out_hole:
 519	ret = 0;
 520out:
 521	brelse(eb_bh);
 522	return ret;
 523}
 524
 525static void ocfs2_relative_extent_offsets(struct super_block *sb,
 526					  u32 v_cluster,
 527					  struct ocfs2_extent_rec *rec,
 528					  u32 *p_cluster, u32 *num_clusters)
 529
 530{
 531	u32 coff = v_cluster - le32_to_cpu(rec->e_cpos);
 532
 533	*p_cluster = ocfs2_blocks_to_clusters(sb, le64_to_cpu(rec->e_blkno));
 534	*p_cluster = *p_cluster + coff;
 535
 536	if (num_clusters)
 537		*num_clusters = le16_to_cpu(rec->e_leaf_clusters) - coff;
 538}
 539
 540int ocfs2_xattr_get_clusters(struct inode *inode, u32 v_cluster,
 541			     u32 *p_cluster, u32 *num_clusters,
 542			     struct ocfs2_extent_list *el,
 543			     unsigned int *extent_flags)
 544{
 545	int ret = 0, i;
 546	struct buffer_head *eb_bh = NULL;
 547	struct ocfs2_extent_block *eb;
 548	struct ocfs2_extent_rec *rec;
 549	u32 coff;
 550
 551	if (el->l_tree_depth) {
 552		ret = ocfs2_find_leaf(INODE_CACHE(inode), el, v_cluster,
 553				      &eb_bh);
 554		if (ret) {
 555			mlog_errno(ret);
 556			goto out;
 557		}
 558
 559		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
 560		el = &eb->h_list;
 561
 562		if (el->l_tree_depth) {
 563			ocfs2_error(inode->i_sb,
 564				    "Inode %lu has non zero tree depth in xattr leaf block %llu\n",
 565				    inode->i_ino,
 566				    (unsigned long long)eb_bh->b_blocknr);
 567			ret = -EROFS;
 568			goto out;
 569		}
 570	}
 571
 572	i = ocfs2_search_extent_list(el, v_cluster);
 573	if (i == -1) {
 574		ret = -EROFS;
 575		mlog_errno(ret);
 576		goto out;
 577	} else {
 578		rec = &el->l_recs[i];
 579		BUG_ON(v_cluster < le32_to_cpu(rec->e_cpos));
 580
 581		if (!rec->e_blkno) {
 582			ocfs2_error(inode->i_sb,
 583				    "Inode %lu has bad extent record (%u, %u, 0) in xattr\n",
 584				    inode->i_ino,
 585				    le32_to_cpu(rec->e_cpos),
 586				    ocfs2_rec_clusters(el, rec));
 587			ret = -EROFS;
 588			goto out;
 589		}
 590		coff = v_cluster - le32_to_cpu(rec->e_cpos);
 591		*p_cluster = ocfs2_blocks_to_clusters(inode->i_sb,
 592						    le64_to_cpu(rec->e_blkno));
 593		*p_cluster = *p_cluster + coff;
 594		if (num_clusters)
 595			*num_clusters = ocfs2_rec_clusters(el, rec) - coff;
 596
 597		if (extent_flags)
 598			*extent_flags = rec->e_flags;
 599	}
 600out:
 601	brelse(eb_bh);
 602	return ret;
 603}
 604
 605int ocfs2_get_clusters(struct inode *inode, u32 v_cluster,
 606		       u32 *p_cluster, u32 *num_clusters,
 607		       unsigned int *extent_flags)
 608{
 609	int ret;
 610	unsigned int hole_len, flags = 0;
 611	struct buffer_head *di_bh = NULL;
 612	struct ocfs2_extent_rec rec;
 613
 614	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 615		ret = -ERANGE;
 616		mlog_errno(ret);
 617		goto out;
 618	}
 619
 620	ret = ocfs2_extent_map_lookup(inode, v_cluster, p_cluster,
 621				      num_clusters, extent_flags);
 622	if (ret == 0)
 623		goto out;
 624
 625	ret = ocfs2_read_inode_block(inode, &di_bh);
 626	if (ret) {
 627		mlog_errno(ret);
 628		goto out;
 629	}
 630
 631	ret = ocfs2_get_clusters_nocache(inode, di_bh, v_cluster, &hole_len,
 632					 &rec, NULL);
 633	if (ret) {
 634		mlog_errno(ret);
 635		goto out;
 636	}
 637
 638	if (rec.e_blkno == 0ULL) {
 639		/*
 640		 * A hole was found. Return some canned values that
 641		 * callers can key on. If asked for, num_clusters will
 642		 * be populated with the size of the hole.
 643		 */
 644		*p_cluster = 0;
 645		if (num_clusters) {
 646			*num_clusters = hole_len;
 647		}
 648	} else {
 649		ocfs2_relative_extent_offsets(inode->i_sb, v_cluster, &rec,
 650					      p_cluster, num_clusters);
 651		flags = rec.e_flags;
 652
 653		ocfs2_extent_map_insert_rec(inode, &rec);
 654	}
 655
 656	if (extent_flags)
 657		*extent_flags = flags;
 658
 659out:
 660	brelse(di_bh);
 661	return ret;
 662}
 663
 664/*
 665 * This expects alloc_sem to be held. The allocation cannot change at
 666 * all while the map is in the process of being updated.
 667 */
 668int ocfs2_extent_map_get_blocks(struct inode *inode, u64 v_blkno, u64 *p_blkno,
 669				u64 *ret_count, unsigned int *extent_flags)
 670{
 671	int ret;
 672	int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
 673	u32 cpos, num_clusters, p_cluster;
 674	u64 boff = 0;
 675
 676	cpos = ocfs2_blocks_to_clusters(inode->i_sb, v_blkno);
 677
 678	ret = ocfs2_get_clusters(inode, cpos, &p_cluster, &num_clusters,
 679				 extent_flags);
 680	if (ret) {
 681		mlog_errno(ret);
 682		goto out;
 683	}
 684
 685	/*
 686	 * p_cluster == 0 indicates a hole.
 687	 */
 688	if (p_cluster) {
 689		boff = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
 690		boff += (v_blkno & (u64)(bpc - 1));
 691	}
 692
 693	*p_blkno = boff;
 694
 695	if (ret_count) {
 696		*ret_count = ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
 697		*ret_count -= v_blkno & (u64)(bpc - 1);
 698	}
 699
 700out:
 701	return ret;
 702}
 703
 704/*
 705 * The ocfs2_fiemap_inline() may be a little bit misleading, since
 706 * it not only handles the fiemap for inlined files, but also deals
 707 * with the fast symlink, cause they have no difference for extent
 708 * mapping per se.
 709 *
 710 * Must be called with ip_alloc_sem semaphore held.
 711 */
 712static int ocfs2_fiemap_inline(struct inode *inode, struct buffer_head *di_bh,
 713			       struct fiemap_extent_info *fieinfo,
 714			       u64 map_start)
 715{
 716	int ret;
 717	unsigned int id_count;
 718	struct ocfs2_dinode *di;
 719	u64 phys;
 720	u32 flags = FIEMAP_EXTENT_DATA_INLINE|FIEMAP_EXTENT_LAST;
 721	struct ocfs2_inode_info *oi = OCFS2_I(inode);
 722	lockdep_assert_held_read(&oi->ip_alloc_sem);
 723
 724	di = (struct ocfs2_dinode *)di_bh->b_data;
 725	if (ocfs2_inode_is_fast_symlink(inode))
 726		id_count = ocfs2_fast_symlink_chars(inode->i_sb);
 727	else
 728		id_count = le16_to_cpu(di->id2.i_data.id_count);
 729
 730	if (map_start < id_count) {
 731		phys = oi->ip_blkno << inode->i_sb->s_blocksize_bits;
 732		if (ocfs2_inode_is_fast_symlink(inode))
 733			phys += offsetof(struct ocfs2_dinode, id2.i_symlink);
 734		else
 735			phys += offsetof(struct ocfs2_dinode,
 736					 id2.i_data.id_data);
 737
 738		/* Release the ip_alloc_sem to prevent deadlock on page fault */
 739		up_read(&OCFS2_I(inode)->ip_alloc_sem);
 740		ret = fiemap_fill_next_extent(fieinfo, 0, phys, id_count,
 741					      flags);
 742		down_read(&OCFS2_I(inode)->ip_alloc_sem);
 743		if (ret < 0)
 744			return ret;
 745	}
 746
 747	return 0;
 748}
 749
 750int 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_prep(inode, fieinfo, map_start, &map_len, 0);
 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		/* Release the ip_alloc_sem to prevent deadlock on page fault */
 812		up_read(&OCFS2_I(inode)->ip_alloc_sem);
 813		ret = fiemap_fill_next_extent(fieinfo, virt_bytes, phys_bytes,
 814					      len_bytes, fe_flags);
 815		down_read(&OCFS2_I(inode)->ip_alloc_sem);
 816		if (ret)
 817			break;
 818
 819		cpos = le32_to_cpu(rec.e_cpos)+ le16_to_cpu(rec.e_leaf_clusters);
 820	}
 821
 822	if (ret > 0)
 823		ret = 0;
 824
 825out_unlock:
 826	brelse(di_bh);
 827
 828	up_read(&OCFS2_I(inode)->ip_alloc_sem);
 829
 830	ocfs2_inode_unlock(inode, 0);
 831out:
 832
 833	return ret;
 834}
 835
 836/* Is IO overwriting allocated blocks? */
 837int ocfs2_overwrite_io(struct inode *inode, struct buffer_head *di_bh,
 838		       u64 map_start, u64 map_len)
 839{
 840	int ret = 0, is_last;
 841	u32 mapping_end, cpos;
 842	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 843	struct ocfs2_extent_rec rec;
 844
 845	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 846		if (ocfs2_size_fits_inline_data(di_bh, map_start + map_len))
 847			return ret;
 848		else
 849			return -EAGAIN;
 850	}
 851
 852	cpos = map_start >> osb->s_clustersize_bits;
 853	mapping_end = ocfs2_clusters_for_bytes(inode->i_sb,
 854					       map_start + map_len);
 855	is_last = 0;
 856	while (cpos < mapping_end && !is_last) {
 857		ret = ocfs2_get_clusters_nocache(inode, di_bh, cpos,
 858						 NULL, &rec, &is_last);
 859		if (ret) {
 860			mlog_errno(ret);
 861			goto out;
 862		}
 863
 864		if (rec.e_blkno == 0ULL)
 865			break;
 866
 867		if (rec.e_flags & OCFS2_EXT_REFCOUNTED)
 868			break;
 869
 870		cpos = le32_to_cpu(rec.e_cpos) +
 871			le16_to_cpu(rec.e_leaf_clusters);
 872	}
 873
 874	if (cpos < mapping_end)
 875		ret = -EAGAIN;
 876out:
 877	return ret;
 878}
 879
 880int ocfs2_seek_data_hole_offset(struct file *file, loff_t *offset, int whence)
 881{
 882	struct inode *inode = file->f_mapping->host;
 883	int ret;
 884	unsigned int is_last = 0, is_data = 0;
 885	u16 cs_bits = OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 886	u32 cpos, cend, clen, hole_size;
 887	u64 extoff, extlen;
 888	struct buffer_head *di_bh = NULL;
 889	struct ocfs2_extent_rec rec;
 890
 891	BUG_ON(whence != SEEK_DATA && whence != SEEK_HOLE);
 892
 893	ret = ocfs2_inode_lock(inode, &di_bh, 0);
 894	if (ret) {
 895		mlog_errno(ret);
 896		goto out;
 897	}
 898
 899	down_read(&OCFS2_I(inode)->ip_alloc_sem);
 900
 901	if (*offset >= i_size_read(inode)) {
 902		ret = -ENXIO;
 903		goto out_unlock;
 904	}
 905
 906	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 907		if (whence == SEEK_HOLE)
 908			*offset = i_size_read(inode);
 909		goto out_unlock;
 910	}
 911
 912	clen = 0;
 913	cpos = *offset >> cs_bits;
 914	cend = ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode));
 915
 916	while (cpos < cend && !is_last) {
 917		ret = ocfs2_get_clusters_nocache(inode, di_bh, cpos, &hole_size,
 918						 &rec, &is_last);
 919		if (ret) {
 920			mlog_errno(ret);
 921			goto out_unlock;
 922		}
 923
 924		extoff = cpos;
 925		extoff <<= cs_bits;
 926
 927		if (rec.e_blkno == 0ULL) {
 928			clen = hole_size;
 929			is_data = 0;
 930		} else {
 931			clen = le16_to_cpu(rec.e_leaf_clusters) -
 932				(cpos - le32_to_cpu(rec.e_cpos));
 933			is_data = (rec.e_flags & OCFS2_EXT_UNWRITTEN) ?  0 : 1;
 934		}
 935
 936		if ((!is_data && whence == SEEK_HOLE) ||
 937		    (is_data && whence == SEEK_DATA)) {
 938			if (extoff > *offset)
 939				*offset = extoff;
 940			goto out_unlock;
 941		}
 942
 943		if (!is_last)
 944			cpos += clen;
 945	}
 946
 947	if (whence == SEEK_HOLE) {
 948		extoff = cpos;
 949		extoff <<= cs_bits;
 950		extlen = clen;
 951		extlen <<=  cs_bits;
 952
 953		if ((extoff + extlen) > i_size_read(inode))
 954			extlen = i_size_read(inode) - extoff;
 955		extoff += extlen;
 956		if (extoff > *offset)
 957			*offset = extoff;
 958		goto out_unlock;
 959	}
 960
 961	ret = -ENXIO;
 962
 963out_unlock:
 964
 965	brelse(di_bh);
 966
 967	up_read(&OCFS2_I(inode)->ip_alloc_sem);
 968
 969	ocfs2_inode_unlock(inode, 0);
 970out:
 971	return ret;
 972}
 973
 974int ocfs2_read_virt_blocks(struct inode *inode, u64 v_block, int nr,
 975			   struct buffer_head *bhs[], int flags,
 976			   int (*validate)(struct super_block *sb,
 977					   struct buffer_head *bh))
 978{
 979	int rc = 0;
 980	u64 p_block, p_count;
 981	int i, count, done = 0;
 982
 983	trace_ocfs2_read_virt_blocks(
 984	     inode, (unsigned long long)v_block, nr, bhs, flags,
 985	     validate);
 986
 987	if (((v_block + nr - 1) << inode->i_sb->s_blocksize_bits) >=
 988	    i_size_read(inode)) {
 989		BUG_ON(!(flags & OCFS2_BH_READAHEAD));
 990		goto out;
 991	}
 992
 993	while (done < nr) {
 994		if (!down_read_trylock(&OCFS2_I(inode)->ip_alloc_sem)) {
 995			rc = -EAGAIN;
 996			mlog(ML_ERROR,
 997				 "Inode #%llu ip_alloc_sem is temporarily unavailable\n",
 998				 (unsigned long long)OCFS2_I(inode)->ip_blkno);
 999			break;
1000		}
1001		rc = ocfs2_extent_map_get_blocks(inode, v_block + done,
1002						 &p_block, &p_count, NULL);
1003		up_read(&OCFS2_I(inode)->ip_alloc_sem);
1004		if (rc) {
1005			mlog_errno(rc);
1006			break;
1007		}
1008
1009		if (!p_block) {
1010			rc = -EIO;
1011			mlog(ML_ERROR,
1012			     "Inode #%llu contains a hole at offset %llu\n",
1013			     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1014			     (unsigned long long)(v_block + done) <<
1015			     inode->i_sb->s_blocksize_bits);
1016			break;
1017		}
1018
1019		count = nr - done;
1020		if (p_count < count)
1021			count = p_count;
1022
1023		/*
1024		 * If the caller passed us bhs, they should have come
1025		 * from a previous readahead call to this function.  Thus,
1026		 * they should have the right b_blocknr.
1027		 */
1028		for (i = 0; i < count; i++) {
1029			if (!bhs[done + i])
1030				continue;
1031			BUG_ON(bhs[done + i]->b_blocknr != (p_block + i));
1032		}
1033
1034		rc = ocfs2_read_blocks(INODE_CACHE(inode), p_block, count,
1035				       bhs + done, flags, validate);
1036		if (rc) {
1037			mlog_errno(rc);
1038			break;
1039		}
1040		done += count;
1041	}
1042
1043out:
1044	return rc;
1045}
1046
1047