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/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
  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
  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
  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  */
 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  */
 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
 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  */
 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  */
 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         kfree(new_emi);
 286 }
 287
 288 static int ocfs2_last_eb_is_empty(struct inode *inode,
 289                                   struct ocfs2_dinode *di)
 290 {
 291         int ret, next_free;
 292         u64 last_eb_blk = le64_to_cpu(di->i_last_eb_blk);
 293         struct buffer_head *eb_bh = NULL;
 294         struct ocfs2_extent_block *eb;
 295         struct ocfs2_extent_list *el;
 296
 297         ret = ocfs2_read_extent_block(INODE_CACHE(inode), last_eb_blk, &eb_bh);
 298         if (ret) {
 299                 mlog_errno(ret);
 300                 goto out;
 301         }
 302
 303         eb = (struct ocfs2_extent_block *) eb_bh->b_data;
 304         el = &eb->h_list;
 305
 306         if (el->l_tree_depth) {
 307                 ocfs2_error(inode->i_sb,
 308                             "Inode %lu has non zero tree depth in "
 309                             "leaf block %llu\n", inode->i_ino,
 310                             (unsigned long long)eb_bh->b_blocknr);
 311                 ret = -EROFS;
 312                 goto out;
 313         }
 314
 315         next_free = le16_to_cpu(el->l_next_free_rec);
 316
 317         if (next_free == 0 ||
 318             (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0])))
 319                 ret = 1;
 320
 321 out:
 322         brelse(eb_bh);
 323         return ret;
 324 }
 325
 326 /*
 327  * Return the 1st index within el which contains an extent start
 328  * larger than v_cluster.
 329  */
 330 static int ocfs2_search_for_hole_index(struct ocfs2_extent_list *el,
 331                                        u32 v_cluster)
 332 {
 333         int i;
 334         struct ocfs2_extent_rec *rec;
 335
 336         for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
 337                 rec = &el->l_recs[i];
 338
 339                 if (v_cluster < le32_to_cpu(rec->e_cpos))
 340                         break;
 341         }
 342
 343         return i;
 344 }
 345
 346 /*
 347  * Figure out the size of a hole which starts at v_cluster within the given
 348  * extent list.
 349  *
 350  * If there is no more allocation past v_cluster, we return the maximum
 351  * cluster size minus v_cluster.
 352  *
 353  * If we have in-inode extents, then el points to the dinode list and
 354  * eb_bh is NULL. Otherwise, eb_bh should point to the extent block
 355  * containing el.
 356  */
 357 int ocfs2_figure_hole_clusters(struct ocfs2_caching_info *ci,
 358                                struct ocfs2_extent_list *el,
 359                                struct buffer_head *eb_bh,
 360                                u32 v_cluster,
 361                                u32 *num_clusters)
 362 {
 363         int ret, i;
 364         struct buffer_head *next_eb_bh = NULL;
 365         struct ocfs2_extent_block *eb, *next_eb;
 366
 367         i = ocfs2_search_for_hole_index(el, v_cluster);
 368
 369         if (i == le16_to_cpu(el->l_next_free_rec) && eb_bh) {
 370                 eb = (struct ocfs2_extent_block *)eb_bh->b_data;
 371
 372                 /*
 373                  * Check the next leaf for any extents.
 374                  */
 375
 376                 if (le64_to_cpu(eb->h_next_leaf_blk) == 0ULL)
 377                         goto no_more_extents;
 378
 379                 ret = ocfs2_read_extent_block(ci,
 380                                               le64_to_cpu(eb->h_next_leaf_blk),
 381                                               &next_eb_bh);
 382                 if (ret) {
 383                         mlog_errno(ret);
 384                         goto out;
 385                 }
 386
 387                 next_eb = (struct ocfs2_extent_block *)next_eb_bh->b_data;
 388                 el = &next_eb->h_list;
 389                 i = ocfs2_search_for_hole_index(el, v_cluster);
 390         }
 391
 392 no_more_extents:
 393         if (i == le16_to_cpu(el->l_next_free_rec)) {
 394                 /*
 395                  * We're at the end of our existing allocation. Just
 396                  * return the maximum number of clusters we could
 397                  * possibly allocate.
 398                  */
 399                 *num_clusters = UINT_MAX - v_cluster;
 400         } else {
 401                 *num_clusters = le32_to_cpu(el->l_recs[i].e_cpos) - v_cluster;
 402         }
 403
 404         ret = 0;
 405 out:
 406         brelse(next_eb_bh);
 407         return ret;
 408 }
 409
 410 static int ocfs2_get_clusters_nocache(struct inode *inode,
 411                                       struct buffer_head *di_bh,
 412                                       u32 v_cluster, unsigned int *hole_len,
 413                                       struct ocfs2_extent_rec *ret_rec,
 414                                       unsigned int *is_last)
 415 {
 416         int i, ret, tree_height, len;
 417         struct ocfs2_dinode *di;
 418         struct ocfs2_extent_block *uninitialized_var(eb);
 419         struct ocfs2_extent_list *el;
 420         struct ocfs2_extent_rec *rec;
 421         struct buffer_head *eb_bh = NULL;
 422
 423         memset(ret_rec, 0, sizeof(*ret_rec));
 424         if (is_last)
 425                 *is_last = 0;
 426
 427         di = (struct ocfs2_dinode *) di_bh->b_data;
 428         el = &di->id2.i_list;
 429         tree_height = le16_to_cpu(el->l_tree_depth);
 430
 431         if (tree_height > 0) {
 432                 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, v_cluster,
 433                                       &eb_bh);
 434                 if (ret) {
 435                         mlog_errno(ret);
 436                         goto out;
 437                 }
 438
 439                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
 440                 el = &eb->h_list;
 441
 442                 if (el->l_tree_depth) {
 443                         ocfs2_error(inode->i_sb,
 444                                     "Inode %lu has non zero tree depth in "
 445                                     "leaf block %llu\n", inode->i_ino,
 446                                     (unsigned long long)eb_bh->b_blocknr);
 447                         ret = -EROFS;
 448                         goto out;
 449                 }
 450         }
 451
 452         i = ocfs2_search_extent_list(el, v_cluster);
 453         if (i == -1) {
 454                 /*
 455                  * Holes can be larger than the maximum size of an
 456                  * extent, so we return their lengths in a separate
 457                  * field.
 458                  */
 459                 if (hole_len) {
 460                         ret = ocfs2_figure_hole_clusters(INODE_CACHE(inode),
 461                                                          el, eb_bh,
 462                                                          v_cluster, &len);
 463                         if (ret) {
 464                                 mlog_errno(ret);
 465                                 goto out;
 466                         }
 467
 468                         *hole_len = len;
 469                 }
 470                 goto out_hole;
 471         }
 472
 473         rec = &el->l_recs[i];
 474
 475         BUG_ON(v_cluster < le32_to_cpu(rec->e_cpos));
 476
 477         if (!rec->e_blkno) {
 478                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
 479                             "record (%u, %u, 0)", inode->i_ino,
 480                             le32_to_cpu(rec->e_cpos),
 481                             ocfs2_rec_clusters(el, rec));
 482                 ret = -EROFS;
 483                 goto out;
 484         }
 485
 486         *ret_rec = *rec;
 487
 488         /*
 489          * Checking for last extent is potentially expensive - we
 490          * might have to look at the next leaf over to see if it's
 491          * empty.
 492          *
 493          * The first two checks are to see whether the caller even
 494          * cares for this information, and if the extent is at least
 495          * the last in it's list.
 496          *
 497          * If those hold true, then the extent is last if any of the
 498          * additional conditions hold true:
 499          *  - Extent list is in-inode
 500          *  - Extent list is right-most
 501          *  - Extent list is 2nd to rightmost, with empty right-most
 502          */
 503         if (is_last) {
 504                 if (i == (le16_to_cpu(el->l_next_free_rec) - 1)) {
 505                         if (tree_height == 0)
 506                                 *is_last = 1;
 507                         else if (eb->h_blkno == di->i_last_eb_blk)
 508                                 *is_last = 1;
 509                         else if (eb->h_next_leaf_blk == di->i_last_eb_blk) {
 510                                 ret = ocfs2_last_eb_is_empty(inode, di);
 511                                 if (ret < 0) {
 512                                         mlog_errno(ret);
 513                                         goto out;
 514                                 }
 515                                 if (ret == 1)
 516                                         *is_last = 1;
 517                         }
 518                 }
 519         }
 520
 521 out_hole:
 522         ret = 0;
 523 out:
 524         brelse(eb_bh);
 525         return ret;
 526 }
 527
 528 static void ocfs2_relative_extent_offsets(struct super_block *sb,
 529                                           u32 v_cluster,
 530                                           struct ocfs2_extent_rec *rec,
 531                                           u32 *p_cluster, u32 *num_clusters)
 532
 533 {
 534         u32 coff = v_cluster - le32_to_cpu(rec->e_cpos);
 535
 536         *p_cluster = ocfs2_blocks_to_clusters(sb, le64_to_cpu(rec->e_blkno));
 537         *p_cluster = *p_cluster + coff;
 538
 539         if (num_clusters)
 540                 *num_clusters = le16_to_cpu(rec->e_leaf_clusters) - coff;
 541 }
 542
 543 int ocfs2_xattr_get_clusters(struct inode *inode, u32 v_cluster,
 544                              u32 *p_cluster, u32 *num_clusters,
 545                              struct ocfs2_extent_list *el,
 546                              unsigned int *extent_flags)
 547 {
 548         int ret = 0, i;
 549         struct buffer_head *eb_bh = NULL;
 550         struct ocfs2_extent_block *eb;
 551         struct ocfs2_extent_rec *rec;
 552         u32 coff;
 553
 554         if (el->l_tree_depth) {
 555                 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, v_cluster,
 556                                       &eb_bh);
 557                 if (ret) {
 558                         mlog_errno(ret);
 559                         goto out;
 560                 }
 561
 562                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
 563                 el = &eb->h_list;
 564
 565                 if (el->l_tree_depth) {
 566                         ocfs2_error(inode->i_sb,
 567                                     "Inode %lu has non zero tree depth in "
 568                                     "xattr leaf block %llu\n", inode->i_ino,
 569                                     (unsigned long long)eb_bh->b_blocknr);
 570                         ret = -EROFS;
 571                         goto out;
 572                 }
 573         }
 574
 575         i = ocfs2_search_extent_list(el, v_cluster);
 576         if (i == -1) {
 577                 ret = -EROFS;
 578                 mlog_errno(ret);
 579                 goto out;
 580         } else {
 581                 rec = &el->l_recs[i];
 582                 BUG_ON(v_cluster < le32_to_cpu(rec->e_cpos));
 583
 584                 if (!rec->e_blkno) {
 585                         ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
 586                                     "record (%u, %u, 0) in xattr", inode->i_ino,
 587                                     le32_to_cpu(rec->e_cpos),
 588                                     ocfs2_rec_clusters(el, rec));
 589                         ret = -EROFS;
 590                         goto out;
 591                 }
 592                 coff = v_cluster - le32_to_cpu(rec->e_cpos);
 593                 *p_cluster = ocfs2_blocks_to_clusters(inode->i_sb,
 594                                                     le64_to_cpu(rec->e_blkno));
 595                 *p_cluster = *p_cluster + coff;
 596                 if (num_clusters)
 597                         *num_clusters = ocfs2_rec_clusters(el, rec) - coff;
 598
 599                 if (extent_flags)
 600                         *extent_flags = rec->e_flags;
 601         }
 602 out:
 603         if (eb_bh)
 604                 brelse(eb_bh);
 605         return ret;
 606 }
 607
 608 int ocfs2_get_clusters(struct inode *inode, u32 v_cluster,
 609                        u32 *p_cluster, u32 *num_clusters,
 610                        unsigned int *extent_flags)
 611 {
 612         int ret;
 613         unsigned int uninitialized_var(hole_len), flags = 0;
 614         struct buffer_head *di_bh = NULL;
 615         struct ocfs2_extent_rec rec;
 616
 617         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 618                 ret = -ERANGE;
 619                 mlog_errno(ret);
 620                 goto out;
 621         }
 622
 623         ret = ocfs2_extent_map_lookup(inode, v_cluster, p_cluster,
 624                                       num_clusters, extent_flags);
 625         if (ret == 0)
 626                 goto out;
 627
 628         ret = ocfs2_read_inode_block(inode, &di_bh);
 629         if (ret) {
 630                 mlog_errno(ret);
 631                 goto out;
 632         }
 633
 634         ret = ocfs2_get_clusters_nocache(inode, di_bh, v_cluster, &hole_len,
 635                                          &rec, NULL);
 636         if (ret) {
 637                 mlog_errno(ret);
 638                 goto out;
 639         }
 640
 641         if (rec.e_blkno == 0ULL) {
 642                 /*
 643                  * A hole was found. Return some canned values that
 644                  * callers can key on. If asked for, num_clusters will
 645                  * be populated with the size of the hole.
 646                  */
 647                 *p_cluster = 0;
 648                 if (num_clusters) {
 649                         *num_clusters = hole_len;
 650                 }
 651         } else {
 652                 ocfs2_relative_extent_offsets(inode->i_sb, v_cluster, &rec,
 653                                               p_cluster, num_clusters);
 654                 flags = rec.e_flags;
 655
 656                 ocfs2_extent_map_insert_rec(inode, &rec);
 657         }
 658
 659         if (extent_flags)
 660                 *extent_flags = flags;
 661
 662 out:
 663         brelse(di_bh);
 664         return ret;
 665 }
 666
 667 /*
 668  * This expects alloc_sem to be held. The allocation cannot change at
 669  * all while the map is in the process of being updated.
 670  */
 671 int ocfs2_extent_map_get_blocks(struct inode *inode, u64 v_blkno, u64 *p_blkno,
 672                                 u64 *ret_count, unsigned int *extent_flags)
 673 {
 674         int ret;
 675         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
 676         u32 cpos, num_clusters, p_cluster;
 677         u64 boff = 0;
 678
 679         cpos = ocfs2_blocks_to_clusters(inode->i_sb, v_blkno);
 680
 681         ret = ocfs2_get_clusters(inode, cpos, &p_cluster, &num_clusters,
 682                                  extent_flags);
 683         if (ret) {
 684                 mlog_errno(ret);
 685                 goto out;
 686         }
 687
 688         /*
 689          * p_cluster == 0 indicates a hole.
 690          */
 691         if (p_cluster) {
 692                 boff = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
 693                 boff += (v_blkno & (u64)(bpc - 1));
 694         }
 695
 696         *p_blkno = boff;
 697
 698         if (ret_count) {
 699                 *ret_count = ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
 700                 *ret_count -= v_blkno & (u64)(bpc - 1);
 701         }
 702
 703 out:
 704         return ret;
 705 }
 706
 707 /*
 708  * The ocfs2_fiemap_inline() may be a little bit misleading, since
 709  * it not only handles the fiemap for inlined files, but also deals
 710  * with the fast symlink, cause they have no difference for extent
 711  * mapping per se.
 712  */
 713 static int ocfs2_fiemap_inline(struct inode *inode, struct buffer_head *di_bh,
 714                                struct fiemap_extent_info *fieinfo,
 715                                u64 map_start)
 716 {
 717         int ret;
 718         unsigned int id_count;
 719         struct ocfs2_dinode *di;
 720         u64 phys;
 721         u32 flags = FIEMAP_EXTENT_DATA_INLINE|FIEMAP_EXTENT_LAST;
 722         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 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                 ret = fiemap_fill_next_extent(fieinfo, 0, phys, id_count,
 739                                               flags);
 740                 if (ret < 0)
 741                         return ret;
 742         }
 743
 744         return 0;
 745 }
 746
 747 #define OCFS2_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
 748
 749 int ocfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 750                  u64 map_start, u64 map_len)
 751 {
 752         int ret, is_last;
 753         u32 mapping_end, cpos;
 754         unsigned int hole_size;
 755         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 756         u64 len_bytes, phys_bytes, virt_bytes;
 757         struct buffer_head *di_bh = NULL;
 758         struct ocfs2_extent_rec rec;
 759
 760         ret = fiemap_check_flags(fieinfo, OCFS2_FIEMAP_FLAGS);
 761         if (ret)
 762                 return ret;
 763
 764         ret = ocfs2_inode_lock(inode, &di_bh, 0);
 765         if (ret) {
 766                 mlog_errno(ret);
 767                 goto out;
 768         }
 769
 770         down_read(&OCFS2_I(inode)->ip_alloc_sem);
 771
 772         /*
 773          * Handle inline-data and fast symlink separately.
 774          */
 775         if ((OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) ||
 776             ocfs2_inode_is_fast_symlink(inode)) {
 777                 ret = ocfs2_fiemap_inline(inode, di_bh, fieinfo, map_start);
 778                 goto out_unlock;
 779         }
 780
 781         cpos = map_start >> osb->s_clustersize_bits;
 782         mapping_end = ocfs2_clusters_for_bytes(inode->i_sb,
 783                                                map_start + map_len);
 784         is_last = 0;
 785         while (cpos < mapping_end && !is_last) {
 786                 u32 fe_flags;
 787
 788                 ret = ocfs2_get_clusters_nocache(inode, di_bh, cpos,
 789                                                  &hole_size, &rec, &is_last);
 790                 if (ret) {
 791                         mlog_errno(ret);
 792                         goto out_unlock;
 793                 }
 794
 795                 if (rec.e_blkno == 0ULL) {
 796                         cpos += hole_size;
 797                         continue;
 798                 }
 799
 800                 fe_flags = 0;
 801                 if (rec.e_flags & OCFS2_EXT_UNWRITTEN)
 802                         fe_flags |= FIEMAP_EXTENT_UNWRITTEN;
 803                 if (rec.e_flags & OCFS2_EXT_REFCOUNTED)
 804                         fe_flags |= FIEMAP_EXTENT_SHARED;
 805                 if (is_last)
 806                         fe_flags |= FIEMAP_EXTENT_LAST;
 807                 len_bytes = (u64)le16_to_cpu(rec.e_leaf_clusters) << osb->s_clustersize_bits;
 808                 phys_bytes = le64_to_cpu(rec.e_blkno) << osb->sb->s_blocksize_bits;
 809                 virt_bytes = (u64)le32_to_cpu(rec.e_cpos) << osb->s_clustersize_bits;
 810
 811                 ret = fiemap_fill_next_extent(fieinfo, virt_bytes, phys_bytes,
 812                                               len_bytes, fe_flags);
 813                 if (ret)
 814                         break;
 815
 816                 cpos = le32_to_cpu(rec.e_cpos)+ le16_to_cpu(rec.e_leaf_clusters);
 817         }
 818
 819         if (ret > 0)
 820                 ret = 0;
 821
 822 out_unlock:
 823         brelse(di_bh);
 824
 825         up_read(&OCFS2_I(inode)->ip_alloc_sem);
 826
 827         ocfs2_inode_unlock(inode, 0);
 828 out:
 829
 830         return ret;
 831 }
 832
 833 int ocfs2_seek_data_hole_offset(struct file *file, loff_t *offset, int whence)
 834 {
 835         struct inode *inode = file->f_mapping->host;
 836         int ret;
 837         unsigned int is_last = 0, is_data = 0;
 838         u16 cs_bits = OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 839         u32 cpos, cend, clen, hole_size;
 840         u64 extoff, extlen;
 841         struct buffer_head *di_bh = NULL;
 842         struct ocfs2_extent_rec rec;
 843
 844         BUG_ON(whence != SEEK_DATA && whence != SEEK_HOLE);
 845
 846         ret = ocfs2_inode_lock(inode, &di_bh, 0);
 847         if (ret) {
 848                 mlog_errno(ret);
 849                 goto out;
 850         }
 851
 852         down_read(&OCFS2_I(inode)->ip_alloc_sem);
 853
 854         if (*offset >= i_size_read(inode)) {
 855                 ret = -ENXIO;
 856                 goto out_unlock;
 857         }
 858
 859         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 860                 if (whence == SEEK_HOLE)
 861                         *offset = i_size_read(inode);
 862                 goto out_unlock;
 863         }
 864
 865         clen = 0;
 866         cpos = *offset >> cs_bits;
 867         cend = ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode));
 868
 869         while (cpos < cend && !is_last) {
 870                 ret = ocfs2_get_clusters_nocache(inode, di_bh, cpos, &hole_size,
 871                                                  &rec, &is_last);
 872                 if (ret) {
 873                         mlog_errno(ret);
 874                         goto out_unlock;
 875                 }
 876
 877                 extoff = cpos;
 878                 extoff <<= cs_bits;
 879
 880                 if (rec.e_blkno == 0ULL) {
 881                         clen = hole_size;
 882                         is_data = 0;
 883                 } else {
 884                         clen = le16_to_cpu(rec.e_leaf_clusters) -
 885                                 (cpos - le32_to_cpu(rec.e_cpos));
 886                         is_data = (rec.e_flags & OCFS2_EXT_UNWRITTEN) ?  0 : 1;
 887                 }
 888
 889                 if ((!is_data && whence == SEEK_HOLE) ||
 890                     (is_data && whence == SEEK_DATA)) {
 891                         if (extoff > *offset)
 892                                 *offset = extoff;
 893                         goto out_unlock;
 894                 }
 895
 896                 if (!is_last)
 897                         cpos += clen;
 898         }
 899
 900         if (whence == SEEK_HOLE) {
 901                 extoff = cpos;
 902                 extoff <<= cs_bits;
 903                 extlen = clen;
 904                 extlen <<=  cs_bits;
 905
 906                 if ((extoff + extlen) > i_size_read(inode))
 907                         extlen = i_size_read(inode) - extoff;
 908                 extoff += extlen;
 909                 if (extoff > *offset)
 910                         *offset = extoff;
 911                 goto out_unlock;
 912         }
 913
 914         ret = -ENXIO;
 915
 916 out_unlock:
 917
 918         brelse(di_bh);
 919
 920         up_read(&OCFS2_I(inode)->ip_alloc_sem);
 921
 922         ocfs2_inode_unlock(inode, 0);
 923 out:
 924         return ret;
 925 }
 926
 927 int ocfs2_read_virt_blocks(struct inode *inode, u64 v_block, int nr,
 928                            struct buffer_head *bhs[], int flags,
 929                            int (*validate)(struct super_block *sb,
 930                                            struct buffer_head *bh))
 931 {
 932         int rc = 0;
 933         u64 p_block, p_count;
 934         int i, count, done = 0;
 935
 936         trace_ocfs2_read_virt_blocks(
 937              inode, (unsigned long long)v_block, nr, bhs, flags,
 938              validate);
 939
 940         if (((v_block + nr - 1) << inode->i_sb->s_blocksize_bits) >=
 941             i_size_read(inode)) {
 942                 BUG_ON(!(flags & OCFS2_BH_READAHEAD));
 943                 goto out;
 944         }
 945
 946         while (done < nr) {
 947                 down_read(&OCFS2_I(inode)->ip_alloc_sem);
 948                 rc = ocfs2_extent_map_get_blocks(inode, v_block + done,
 949                                                  &p_block, &p_count, NULL);
 950                 up_read(&OCFS2_I(inode)->ip_alloc_sem);
 951                 if (rc) {
 952                         mlog_errno(rc);
 953                         break;
 954                 }
 955
 956                 if (!p_block) {
 957                         rc = -EIO;
 958                         mlog(ML_ERROR,
 959                              "Inode #%llu contains a hole at offset %llu\n",
 960                              (unsigned long long)OCFS2_I(inode)->ip_blkno,
 961                              (unsigned long long)(v_block + done) <<
 962                              inode->i_sb->s_blocksize_bits);
 963                         break;
 964                 }
 965
 966                 count = nr - done;
 967                 if (p_count < count)
 968                         count = p_count;
 969
 970                 /*
 971                  * If the caller passed us bhs, they should have come
 972                  * from a previous readahead call to this function.  Thus,
 973                  * they should have the right b_blocknr.
 974                  */
 975                 for (i = 0; i < count; i++) {
 976                         if (!bhs[done + i])
 977                                 continue;
 978                         BUG_ON(bhs[done + i]->b_blocknr != (p_block + i));
 979                 }
 980
 981                 rc = ocfs2_read_blocks(INODE_CACHE(inode), p_block, count,
 982                                        bhs + done, flags, validate);
 983                 if (rc) {
 984                         mlog_errno(rc);
 985                         break;
 986                 }
 987                 done += count;
 988         }
 989
 990 out:
 991         return rc;
 992 }
 993
 994