fs/ocfs2/extent_map.c

   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/types.h>
  28 #include <linux/fiemap.h>
  29
  30 #define MLOG_MASK_PREFIX ML_EXTENT_MAP
  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
  41 #include "buffer_head_io.h"
  42
  43 /*
  44  * The extent caching implementation is intentionally trivial.
  45  *
  46  * We only cache a small number of extents stored directly on the
  47  * inode, so linear order operations are acceptable. If we ever want
  48  * to increase the size of the extent map, then these algorithms must
  49  * get smarter.
  50  */
  51
  52 void ocfs2_extent_map_init(struct inode *inode)
  53 {
  54         struct ocfs2_inode_info *oi = OCFS2_I(inode);
  55
  56         oi->ip_extent_map.em_num_items = 0;
  57         INIT_LIST_HEAD(&oi->ip_extent_map.em_list);
  58 }
  59
  60 static void __ocfs2_extent_map_lookup(struct ocfs2_extent_map *em,
  61                                       unsigned int cpos,
  62                                       struct ocfs2_extent_map_item **ret_emi)
  63 {
  64         unsigned int range;
  65         struct ocfs2_extent_map_item *emi;
  66
  67         *ret_emi = NULL;
  68
  69         list_for_each_entry(emi, &em->em_list, ei_list) {
  70                 range = emi->ei_cpos + emi->ei_clusters;
  71
  72                 if (cpos >= emi->ei_cpos && cpos < range) {
  73                         list_move(&emi->ei_list, &em->em_list);
  74
  75                         *ret_emi = emi;
  76                         break;
  77                 }
  78         }
  79 }
  80
  81 static int ocfs2_extent_map_lookup(struct inode *inode, unsigned int cpos,
  82                                    unsigned int *phys, unsigned int *len,
  83                                    unsigned int *flags)
  84 {
  85         unsigned int coff;
  86         struct ocfs2_inode_info *oi = OCFS2_I(inode);
  87         struct ocfs2_extent_map_item *emi;
  88
  89         spin_lock(&oi->ip_lock);
  90
  91         __ocfs2_extent_map_lookup(&oi->ip_extent_map, cpos, &emi);
  92         if (emi) {
  93                 coff = cpos - emi->ei_cpos;
  94                 *phys = emi->ei_phys + coff;
  95                 if (len)
  96                         *len = emi->ei_clusters - coff;
  97                 if (flags)
  98                         *flags = emi->ei_flags;
  99         }
 100
 101         spin_unlock(&oi->ip_lock);
 102
 103         if (emi == NULL)
 104                 return -ENOENT;
 105
 106         return 0;
 107 }
 108
 109 /*
 110  * Forget about all clusters equal to or greater than cpos.
 111  */
 112 void ocfs2_extent_map_trunc(struct inode *inode, unsigned int cpos)
 113 {
 114         struct ocfs2_extent_map_item *emi, *n;
 115         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 116         struct ocfs2_extent_map *em = &oi->ip_extent_map;
 117         LIST_HEAD(tmp_list);
 118         unsigned int range;
 119
 120         spin_lock(&oi->ip_lock);
 121         list_for_each_entry_safe(emi, n, &em->em_list, ei_list) {
 122                 if (emi->ei_cpos >= cpos) {
 123                         /* Full truncate of this record. */
 124                         list_move(&emi->ei_list, &tmp_list);
 125                         BUG_ON(em->em_num_items == 0);
 126                         em->em_num_items--;
 127                         continue;
 128                 }
 129
 130                 range = emi->ei_cpos + emi->ei_clusters;
 131                 if (range > cpos) {
 132                         /* Partial truncate */
 133                         emi->ei_clusters = cpos - emi->ei_cpos;
 134                 }
 135         }
 136         spin_unlock(&oi->ip_lock);
 137
 138         list_for_each_entry_safe(emi, n, &tmp_list, ei_list) {
 139                 list_del(&emi->ei_list);
 140                 kfree(emi);
 141         }
 142 }
 143
 144 /*
 145  * Is any part of emi2 contained within emi1
 146  */
 147 static int ocfs2_ei_is_contained(struct ocfs2_extent_map_item *emi1,
 148                                  struct ocfs2_extent_map_item *emi2)
 149 {
 150         unsigned int range1, range2;
 151
 152         /*
 153          * Check if logical start of emi2 is inside emi1
 154          */
 155         range1 = emi1->ei_cpos + emi1->ei_clusters;
 156         if (emi2->ei_cpos >= emi1->ei_cpos && emi2->ei_cpos < range1)
 157                 return 1;
 158
 159         /*
 160          * Check if logical end of emi2 is inside emi1
 161          */
 162         range2 = emi2->ei_cpos + emi2->ei_clusters;
 163         if (range2 > emi1->ei_cpos && range2 <= range1)
 164                 return 1;
 165
 166         return 0;
 167 }
 168
 169 static void ocfs2_copy_emi_fields(struct ocfs2_extent_map_item *dest,
 170                                   struct ocfs2_extent_map_item *src)
 171 {
 172         dest->ei_cpos = src->ei_cpos;
 173         dest->ei_phys = src->ei_phys;
 174         dest->ei_clusters = src->ei_clusters;
 175         dest->ei_flags = src->ei_flags;
 176 }
 177
 178 /*
 179  * Try to merge emi with ins. Returns 1 if merge succeeds, zero
 180  * otherwise.
 181  */
 182 static int ocfs2_try_to_merge_extent_map(struct ocfs2_extent_map_item *emi,
 183                                          struct ocfs2_extent_map_item *ins)
 184 {
 185         /*
 186          * Handle contiguousness
 187          */
 188         if (ins->ei_phys == (emi->ei_phys + emi->ei_clusters) &&
 189             ins->ei_cpos == (emi->ei_cpos + emi->ei_clusters) &&
 190             ins->ei_flags == emi->ei_flags) {
 191                 emi->ei_clusters += ins->ei_clusters;
 192                 return 1;
 193         } else if ((ins->ei_phys + ins->ei_clusters) == emi->ei_phys &&
 194                    (ins->ei_cpos + ins->ei_clusters) == emi->ei_phys &&
 195                    ins->ei_flags == emi->ei_flags) {
 196                 emi->ei_phys = ins->ei_phys;
 197                 emi->ei_cpos = ins->ei_cpos;
 198                 emi->ei_clusters += ins->ei_clusters;
 199                 return 1;
 200         }
 201
 202         /*
 203          * Overlapping extents - this shouldn't happen unless we've
 204          * split an extent to change it's flags. That is exceedingly
 205          * rare, so there's no sense in trying to optimize it yet.
 206          */
 207         if (ocfs2_ei_is_contained(emi, ins) ||
 208             ocfs2_ei_is_contained(ins, emi)) {
 209                 ocfs2_copy_emi_fields(emi, ins);
 210                 return 1;
 211         }
 212
 213         /* No merge was possible. */
 214         return 0;
 215 }
 216
 217 /*
 218  * In order to reduce complexity on the caller, this insert function
 219  * is intentionally liberal in what it will accept.
 220  *
 221  * The only rule is that the truncate call *must* be used whenever
 222  * records have been deleted. This avoids inserting overlapping
 223  * records with different physical mappings.
 224  */
 225 void ocfs2_extent_map_insert_rec(struct inode *inode,
 226                                  struct ocfs2_extent_rec *rec)
 227 {
 228         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 229         struct ocfs2_extent_map *em = &oi->ip_extent_map;
 230         struct ocfs2_extent_map_item *emi, *new_emi = NULL;
 231         struct ocfs2_extent_map_item ins;
 232
 233         ins.ei_cpos = le32_to_cpu(rec->e_cpos);
 234         ins.ei_phys = ocfs2_blocks_to_clusters(inode->i_sb,
 235                                                le64_to_cpu(rec->e_blkno));
 236         ins.ei_clusters = le16_to_cpu(rec->e_leaf_clusters);
 237         ins.ei_flags = rec->e_flags;
 238
 239 search:
 240         spin_lock(&oi->ip_lock);
 241
 242         list_for_each_entry(emi, &em->em_list, ei_list) {
 243                 if (ocfs2_try_to_merge_extent_map(emi, &ins)) {
 244                         list_move(&emi->ei_list, &em->em_list);
 245                         spin_unlock(&oi->ip_lock);
 246                         goto out;
 247                 }
 248         }
 249
 250         /*
 251          * No item could be merged.
 252          *
 253          * Either allocate and add a new item, or overwrite the last recently
 254          * inserted.
 255          */
 256
 257         if (em->em_num_items < OCFS2_MAX_EXTENT_MAP_ITEMS) {
 258                 if (new_emi == NULL) {
 259                         spin_unlock(&oi->ip_lock);
 260
 261                         new_emi = kmalloc(sizeof(*new_emi), GFP_NOFS);
 262                         if (new_emi == NULL)
 263                                 goto out;
 264
 265                         goto search;
 266                 }
 267
 268                 ocfs2_copy_emi_fields(new_emi, &ins);
 269                 list_add(&new_emi->ei_list, &em->em_list);
 270                 em->em_num_items++;
 271                 new_emi = NULL;
 272         } else {
 273                 BUG_ON(list_empty(&em->em_list) || em->em_num_items == 0);
 274                 emi = list_entry(em->em_list.prev,
 275                                  struct ocfs2_extent_map_item, ei_list);
 276                 list_move(&emi->ei_list, &em->em_list);
 277                 ocfs2_copy_emi_fields(emi, &ins);
 278         }
 279
 280         spin_unlock(&oi->ip_lock);
 281
 282 out:
 283         if (new_emi)
 284                 kfree(new_emi);
 285 }
 286
 287 static int ocfs2_last_eb_is_empty(struct inode *inode,
 288                                   struct ocfs2_dinode *di)
 289 {
 290         int ret, next_free;
 291         u64 last_eb_blk = le64_to_cpu(di->i_last_eb_blk);
 292         struct buffer_head *eb_bh = NULL;
 293         struct ocfs2_extent_block *eb;
 294         struct ocfs2_extent_list *el;
 295
 296         ret = ocfs2_read_extent_block(inode, last_eb_blk, &eb_bh);
 297         if (ret) {
 298                 mlog_errno(ret);
 299                 goto out;
 300         }
 301
 302         eb = (struct ocfs2_extent_block *) eb_bh->b_data;
 303         el = &eb->h_list;
 304
 305         if (el->l_tree_depth) {
 306                 ocfs2_error(inode->i_sb,
 307                             "Inode %lu has non zero tree depth in "
 308                             "leaf block %llu\n", inode->i_ino,
 309                             (unsigned long long)eb_bh->b_blocknr);
 310                 ret = -EROFS;
 311                 goto out;
 312         }
 313
 314         next_free = le16_to_cpu(el->l_next_free_rec);
 315
 316         if (next_free == 0 ||
 317             (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0])))
 318                 ret = 1;
 319
 320 out:
 321         brelse(eb_bh);
 322         return ret;
 323 }
 324
 325 /*
 326  * Return the 1st index within el which contains an extent start
 327  * larger than v_cluster.
 328  */
 329 static int ocfs2_search_for_hole_index(struct ocfs2_extent_list *el,
 330                                        u32 v_cluster)
 331 {
 332         int i;
 333         struct ocfs2_extent_rec *rec;
 334
 335         for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
 336                 rec = &el->l_recs[i];
 337
 338                 if (v_cluster < le32_to_cpu(rec->e_cpos))
 339                         break;
 340         }
 341
 342         return i;
 343 }
 344
 345 /*
 346  * Figure out the size of a hole which starts at v_cluster within the given
 347  * extent list.
 348  *
 349  * If there is no more allocation past v_cluster, we return the maximum
 350  * cluster size minus v_cluster.
 351  *
 352  * If we have in-inode extents, then el points to the dinode list and
 353  * eb_bh is NULL. Otherwise, eb_bh should point to the extent block
 354  * containing el.
 355  */
 356 static int ocfs2_figure_hole_clusters(struct inode *inode,
 357                                       struct ocfs2_extent_list *el,
 358                                       struct buffer_head *eb_bh,
 359                                       u32 v_cluster,
 360                                       u32 *num_clusters)
 361 {
 362         int ret, i;
 363         struct buffer_head *next_eb_bh = NULL;
 364         struct ocfs2_extent_block *eb, *next_eb;
 365
 366         i = ocfs2_search_for_hole_index(el, v_cluster);
 367
 368         if (i == le16_to_cpu(el->l_next_free_rec) && eb_bh) {
 369                 eb = (struct ocfs2_extent_block *)eb_bh->b_data;
 370
 371                 /*
 372                  * Check the next leaf for any extents.
 373                  */
 374
 375                 if (le64_to_cpu(eb->h_next_leaf_blk) == 0ULL)
 376                         goto no_more_extents;
 377
 378                 ret = ocfs2_read_extent_block(inode,
 379                                               le64_to_cpu(eb->h_next_leaf_blk),
 380                                               &next_eb_bh);
 381                 if (ret) {
 382                         mlog_errno(ret);
 383                         goto out;
 384                 }
 385
 386                 next_eb = (struct ocfs2_extent_block *)next_eb_bh->b_data;
 387                 el = &next_eb->h_list;
 388                 i = ocfs2_search_for_hole_index(el, v_cluster);
 389         }
 390
 391 no_more_extents:
 392         if (i == le16_to_cpu(el->l_next_free_rec)) {
 393                 /*
 394                  * We're at the end of our existing allocation. Just
 395                  * return the maximum number of clusters we could
 396                  * possibly allocate.
 397                  */
 398                 *num_clusters = UINT_MAX - v_cluster;
 399         } else {
 400                 *num_clusters = le32_to_cpu(el->l_recs[i].e_cpos) - v_cluster;
 401         }
 402
 403         ret = 0;
 404 out:
 405         brelse(next_eb_bh);
 406         return ret;
 407 }
 408
 409 static int ocfs2_get_clusters_nocache(struct inode *inode,
 410                                       struct buffer_head *di_bh,
 411                                       u32 v_cluster, unsigned int *hole_len,
 412                                       struct ocfs2_extent_rec *ret_rec,
 413                                       unsigned int *is_last)
 414 {
 415         int i, ret, tree_height, len;
 416         struct ocfs2_dinode *di;
 417         struct ocfs2_extent_block *uninitialized_var(eb);
 418         struct ocfs2_extent_list *el;
 419         struct ocfs2_extent_rec *rec;
 420         struct buffer_head *eb_bh = NULL;
 421
 422         memset(ret_rec, 0, sizeof(*ret_rec));
 423         if (is_last)
 424                 *is_last = 0;
 425
 426         di = (struct ocfs2_dinode *) di_bh->b_data;
 427         el = &di->id2.i_list;
 428         tree_height = le16_to_cpu(el->l_tree_depth);
 429
 430         if (tree_height > 0) {
 431                 ret = ocfs2_find_leaf(inode, el, v_cluster, &eb_bh);
 432                 if (ret) {
 433                         mlog_errno(ret);
 434                         goto out;
 435                 }
 436
 437                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
 438                 el = &eb->h_list;
 439
 440                 if (el->l_tree_depth) {
 441                         ocfs2_error(inode->i_sb,
 442                                     "Inode %lu has non zero tree depth in "
 443                                     "leaf block %llu\n", inode->i_ino,
 444                                     (unsigned long long)eb_bh->b_blocknr);
 445                         ret = -EROFS;
 446                         goto out;
 447                 }
 448         }
 449
 450         i = ocfs2_search_extent_list(el, v_cluster);
 451         if (i == -1) {
 452                 /*
 453                  * Holes can be larger than the maximum size of an
 454                  * extent, so we return their lengths in a seperate
 455                  * field.
 456                  */
 457                 if (hole_len) {
 458                         ret = ocfs2_figure_hole_clusters(inode, el, eb_bh,
 459                                                          v_cluster, &len);
 460                         if (ret) {
 461                                 mlog_errno(ret);
 462                                 goto out;
 463                         }
 464
 465                         *hole_len = len;
 466                 }
 467                 goto out_hole;
 468         }
 469
 470         rec = &el->l_recs[i];
 471
 472         BUG_ON(v_cluster < le32_to_cpu(rec->e_cpos));
 473
 474         if (!rec->e_blkno) {
 475                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
 476                             "record (%u, %u, 0)", 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
 518 out_hole:
 519         ret = 0;
 520 out:
 521         brelse(eb_bh);
 522         return ret;
 523 }
 524
 525 static 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
 540 int ocfs2_xattr_get_clusters(struct inode *inode, u32 v_cluster,
 541                              u32 *p_cluster, u32 *num_clusters,
 542                              struct ocfs2_extent_list *el)
 543 {
 544         int ret = 0, i;
 545         struct buffer_head *eb_bh = NULL;
 546         struct ocfs2_extent_block *eb;
 547         struct ocfs2_extent_rec *rec;
 548         u32 coff;
 549
 550         if (el->l_tree_depth) {
 551                 ret = ocfs2_find_leaf(inode, el, v_cluster, &eb_bh);
 552                 if (ret) {
 553                         mlog_errno(ret);
 554                         goto out;
 555                 }
 556
 557                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
 558                 el = &eb->h_list;
 559
 560                 if (el->l_tree_depth) {
 561                         ocfs2_error(inode->i_sb,
 562                                     "Inode %lu has non zero tree depth in "
 563                                     "xattr leaf block %llu\n", inode->i_ino,
 564                                     (unsigned long long)eb_bh->b_blocknr);
 565                         ret = -EROFS;
 566                         goto out;
 567                 }
 568         }
 569
 570         i = ocfs2_search_extent_list(el, v_cluster);
 571         if (i == -1) {
 572                 ret = -EROFS;
 573                 mlog_errno(ret);
 574                 goto out;
 575         } else {
 576                 rec = &el->l_recs[i];
 577                 BUG_ON(v_cluster < le32_to_cpu(rec->e_cpos));
 578
 579                 if (!rec->e_blkno) {
 580                         ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
 581                                     "record (%u, %u, 0) in xattr", inode->i_ino,
 582                                     le32_to_cpu(rec->e_cpos),
 583                                     ocfs2_rec_clusters(el, rec));
 584                         ret = -EROFS;
 585                         goto out;
 586                 }
 587                 coff = v_cluster - le32_to_cpu(rec->e_cpos);
 588                 *p_cluster = ocfs2_blocks_to_clusters(inode->i_sb,
 589                                                     le64_to_cpu(rec->e_blkno));
 590                 *p_cluster = *p_cluster + coff;
 591                 if (num_clusters)
 592                         *num_clusters = ocfs2_rec_clusters(el, rec) - coff;
 593         }
 594 out:
 595         if (eb_bh)
 596                 brelse(eb_bh);
 597         return ret;
 598 }
 599
 600 int ocfs2_get_clusters(struct inode *inode, u32 v_cluster,
 601                        u32 *p_cluster, u32 *num_clusters,
 602                        unsigned int *extent_flags)
 603 {
 604         int ret;
 605         unsigned int uninitialized_var(hole_len), flags = 0;
 606         struct buffer_head *di_bh = NULL;
 607         struct ocfs2_extent_rec rec;
 608
 609         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 610                 ret = -ERANGE;
 611                 mlog_errno(ret);
 612                 goto out;
 613         }
 614
 615         ret = ocfs2_extent_map_lookup(inode, v_cluster, p_cluster,
 616                                       num_clusters, extent_flags);
 617         if (ret == 0)
 618                 goto out;
 619
 620         ret = ocfs2_read_inode_block(inode, &di_bh);
 621         if (ret) {
 622                 mlog_errno(ret);
 623                 goto out;
 624         }
 625
 626         ret = ocfs2_get_clusters_nocache(inode, di_bh, v_cluster, &hole_len,
 627                                          &rec, NULL);
 628         if (ret) {
 629                 mlog_errno(ret);
 630                 goto out;
 631         }
 632
 633         if (rec.e_blkno == 0ULL) {
 634                 /*
 635                  * A hole was found. Return some canned values that
 636                  * callers can key on. If asked for, num_clusters will
 637                  * be populated with the size of the hole.
 638                  */
 639                 *p_cluster = 0;
 640                 if (num_clusters) {
 641                         *num_clusters = hole_len;
 642                 }
 643         } else {
 644                 ocfs2_relative_extent_offsets(inode->i_sb, v_cluster, &rec,
 645                                               p_cluster, num_clusters);
 646                 flags = rec.e_flags;
 647
 648                 ocfs2_extent_map_insert_rec(inode, &rec);
 649         }
 650
 651         if (extent_flags)
 652                 *extent_flags = flags;
 653
 654 out:
 655         brelse(di_bh);
 656         return ret;
 657 }
 658
 659 /*
 660  * This expects alloc_sem to be held. The allocation cannot change at
 661  * all while the map is in the process of being updated.
 662  */
 663 int ocfs2_extent_map_get_blocks(struct inode *inode, u64 v_blkno, u64 *p_blkno,
 664                                 u64 *ret_count, unsigned int *extent_flags)
 665 {
 666         int ret;
 667         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
 668         u32 cpos, num_clusters, p_cluster;
 669         u64 boff = 0;
 670
 671         cpos = ocfs2_blocks_to_clusters(inode->i_sb, v_blkno);
 672
 673         ret = ocfs2_get_clusters(inode, cpos, &p_cluster, &num_clusters,
 674                                  extent_flags);
 675         if (ret) {
 676                 mlog_errno(ret);
 677                 goto out;
 678         }
 679
 680         /*
 681          * p_cluster == 0 indicates a hole.
 682          */
 683         if (p_cluster) {
 684                 boff = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
 685                 boff += (v_blkno & (u64)(bpc - 1));
 686         }
 687
 688         *p_blkno = boff;
 689
 690         if (ret_count) {
 691                 *ret_count = ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
 692                 *ret_count -= v_blkno & (u64)(bpc - 1);
 693         }
 694
 695 out:
 696         return ret;
 697 }
 698
 699 static int ocfs2_fiemap_inline(struct inode *inode, struct buffer_head *di_bh,
 700                                struct fiemap_extent_info *fieinfo,
 701                                u64 map_start)
 702 {
 703         int ret;
 704         unsigned int id_count;
 705         struct ocfs2_dinode *di;
 706         u64 phys;
 707         u32 flags = FIEMAP_EXTENT_DATA_INLINE|FIEMAP_EXTENT_LAST;
 708         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 709
 710         di = (struct ocfs2_dinode *)di_bh->b_data;
 711         id_count = le16_to_cpu(di->id2.i_data.id_count);
 712
 713         if (map_start < id_count) {
 714                 phys = oi->ip_blkno << inode->i_sb->s_blocksize_bits;
 715                 phys += offsetof(struct ocfs2_dinode, id2.i_data.id_data);
 716
 717                 ret = fiemap_fill_next_extent(fieinfo, 0, phys, id_count,
 718                                               flags);
 719                 if (ret < 0)
 720                         return ret;
 721         }
 722
 723         return 0;
 724 }
 725
 726 #define OCFS2_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
 727
 728 int ocfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 729                  u64 map_start, u64 map_len)
 730 {
 731         int ret, is_last;
 732         u32 mapping_end, cpos;
 733         unsigned int hole_size;
 734         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 735         u64 len_bytes, phys_bytes, virt_bytes;
 736         struct buffer_head *di_bh = NULL;
 737         struct ocfs2_extent_rec rec;
 738
 739         ret = fiemap_check_flags(fieinfo, OCFS2_FIEMAP_FLAGS);
 740         if (ret)
 741                 return ret;
 742
 743         ret = ocfs2_inode_lock(inode, &di_bh, 0);
 744         if (ret) {
 745                 mlog_errno(ret);
 746                 goto out;
 747         }
 748
 749         down_read(&OCFS2_I(inode)->ip_alloc_sem);
 750
 751         /*
 752          * Handle inline-data separately.
 753          */
 754         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 755                 ret = ocfs2_fiemap_inline(inode, di_bh, fieinfo, map_start);
 756                 goto out_unlock;
 757         }
 758
 759         cpos = map_start >> osb->s_clustersize_bits;
 760         mapping_end = ocfs2_clusters_for_bytes(inode->i_sb,
 761                                                map_start + map_len);
 762         mapping_end -= cpos;
 763         is_last = 0;
 764         while (cpos < mapping_end && !is_last) {
 765                 u32 fe_flags;
 766
 767                 ret = ocfs2_get_clusters_nocache(inode, di_bh, cpos,
 768                                                  &hole_size, &rec, &is_last);
 769                 if (ret) {
 770                         mlog_errno(ret);
 771                         goto out;
 772                 }
 773
 774                 if (rec.e_blkno == 0ULL) {
 775                         cpos += hole_size;
 776                         continue;
 777                 }
 778
 779                 fe_flags = 0;
 780                 if (rec.e_flags & OCFS2_EXT_UNWRITTEN)
 781                         fe_flags |= FIEMAP_EXTENT_UNWRITTEN;
 782                 if (is_last)
 783                         fe_flags |= FIEMAP_EXTENT_LAST;
 784                 len_bytes = (u64)le16_to_cpu(rec.e_leaf_clusters) << osb->s_clustersize_bits;
 785                 phys_bytes = le64_to_cpu(rec.e_blkno) << osb->sb->s_blocksize_bits;
 786                 virt_bytes = (u64)le32_to_cpu(rec.e_cpos) << osb->s_clustersize_bits;
 787
 788                 ret = fiemap_fill_next_extent(fieinfo, virt_bytes, phys_bytes,
 789                                               len_bytes, fe_flags);
 790                 if (ret)
 791                         break;
 792
 793                 cpos = le32_to_cpu(rec.e_cpos)+ le16_to_cpu(rec.e_leaf_clusters);
 794         }
 795
 796         if (ret > 0)
 797                 ret = 0;
 798
 799 out_unlock:
 800         brelse(di_bh);
 801
 802         up_read(&OCFS2_I(inode)->ip_alloc_sem);
 803
 804         ocfs2_inode_unlock(inode, 0);
 805 out:
 806
 807         return ret;
 808 }
 809
 810 int ocfs2_read_virt_blocks(struct inode *inode, u64 v_block, int nr,
 811                            struct buffer_head *bhs[], int flags,
 812                            int (*validate)(struct super_block *sb,
 813                                            struct buffer_head *bh))
 814 {
 815         int rc = 0;
 816         u64 p_block, p_count;
 817         int i, count, done = 0;
 818
 819         mlog_entry("(inode = %p, v_block = %llu, nr = %d, bhs = %p, "
 820                    "flags = %x, validate = %p)\n",
 821                    inode, (unsigned long long)v_block, nr, bhs, flags,
 822                    validate);
 823
 824         if (((v_block + nr - 1) << inode->i_sb->s_blocksize_bits) >=
 825             i_size_read(inode)) {
 826                 BUG_ON(!(flags & OCFS2_BH_READAHEAD));
 827                 goto out;
 828         }
 829
 830         while (done < nr) {
 831                 down_read(&OCFS2_I(inode)->ip_alloc_sem);
 832                 rc = ocfs2_extent_map_get_blocks(inode, v_block + done,
 833                                                  &p_block, &p_count, NULL);
 834                 up_read(&OCFS2_I(inode)->ip_alloc_sem);
 835                 if (rc) {
 836                         mlog_errno(rc);
 837                         break;
 838                 }
 839
 840                 if (!p_block) {
 841                         rc = -EIO;
 842                         mlog(ML_ERROR,
 843                              "Inode #%llu contains a hole at offset %llu\n",
 844                              (unsigned long long)OCFS2_I(inode)->ip_blkno,
 845                              (unsigned long long)(v_block + done) <<
 846                              inode->i_sb->s_blocksize_bits);
 847                         break;
 848                 }
 849
 850                 count = nr - done;
 851                 if (p_count < count)
 852                         count = p_count;
 853
 854                 /*
 855                  * If the caller passed us bhs, they should have come
 856                  * from a previous readahead call to this function.  Thus,
 857                  * they should have the right b_blocknr.
 858                  */
 859                 for (i = 0; i < count; i++) {
 860                         if (!bhs[done + i])
 861                                 continue;
 862                         BUG_ON(bhs[done + i]->b_blocknr != (p_block + i));
 863                 }
 864
 865                 rc = ocfs2_read_blocks(inode, p_block, count, bhs + done,
 866                                        flags, validate);
 867                 if (rc) {
 868                         mlog_errno(rc);
 869                         break;
 870                 }
 871                 done += count;
 872         }
 873
 874 out:
 875         mlog_exit(rc);
 876         return rc;
 877 }
 878
 879