SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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 #include "symlink.h"
  41
  42 #include "buffer_head_io.h"
  43
  44 /*
  45  * The extent caching implementation is intentionally trivial.
  46  *
  47  * We only cache a small number of extents stored directly on the
  48  * inode, so linear order operations are acceptable. If we ever want
  49  * to increase the size of the extent map, then these algorithms must
  50  * get smarter.
  51  */
  52
  53 void ocfs2_extent_map_init(struct inode *inode)
  54 {
  55         struct ocfs2_inode_info *oi = OCFS2_I(inode);
  56
  57         oi->ip_extent_map.em_num_items = 0;
  58         INIT_LIST_HEAD(&oi->ip_extent_map.em_list);
  59 }
  60
  61 static void __ocfs2_extent_map_lookup(struct ocfs2_extent_map *em,
  62                                       unsigned int cpos,
  63                                       struct ocfs2_extent_map_item **ret_emi)
  64 {
  65         unsigned int range;
  66         struct ocfs2_extent_map_item *emi;
  67
  68         *ret_emi = NULL;
  69
  70         list_for_each_entry(emi, &em->em_list, ei_list) {
  71                 range = emi->ei_cpos + emi->ei_clusters;
  72
  73                 if (cpos >= emi->ei_cpos && cpos < range) {
  74                         list_move(&emi->ei_list, &em->em_list);
  75
  76                         *ret_emi = emi;
  77                         break;
  78                 }
  79         }
  80 }
  81
  82 static int ocfs2_extent_map_lookup(struct inode *inode, unsigned int cpos,
  83                                    unsigned int *phys, unsigned int *len,
  84                                    unsigned int *flags)
  85 {
  86         unsigned int coff;
  87         struct ocfs2_inode_info *oi = OCFS2_I(inode);
  88         struct ocfs2_extent_map_item *emi;
  89
  90         spin_lock(&oi->ip_lock);
  91
  92         __ocfs2_extent_map_lookup(&oi->ip_extent_map, cpos, &emi);
  93         if (emi) {
  94                 coff = cpos - emi->ei_cpos;
  95                 *phys = emi->ei_phys + coff;
  96                 if (len)
  97                         *len = emi->ei_clusters - coff;
  98                 if (flags)
  99                         *flags = emi->ei_flags;
 100         }
 101
 102         spin_unlock(&oi->ip_lock);
 103
 104         if (emi == NULL)
 105                 return -ENOENT;
 106
 107         return 0;
 108 }
 109
 110 /*
 111  * Forget about all clusters equal to or greater than cpos.
 112  */
 113 void ocfs2_extent_map_trunc(struct inode *inode, unsigned int cpos)
 114 {
 115         struct ocfs2_extent_map_item *emi, *n;
 116         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 117         struct ocfs2_extent_map *em = &oi->ip_extent_map;
 118         LIST_HEAD(tmp_list);
 119         unsigned int range;
 120
 121         spin_lock(&oi->ip_lock);
 122         list_for_each_entry_safe(emi, n, &em->em_list, ei_list) {
 123                 if (emi->ei_cpos >= cpos) {
 124                         /* Full truncate of this record. */
 125                         list_move(&emi->ei_list, &tmp_list);
 126                         BUG_ON(em->em_num_items == 0);
 127                         em->em_num_items--;
 128                         continue;
 129                 }
 130
 131                 range = emi->ei_cpos + emi->ei_clusters;
 132                 if (range > cpos) {
 133                         /* Partial truncate */
 134                         emi->ei_clusters = cpos - emi->ei_cpos;
 135                 }
 136         }
 137         spin_unlock(&oi->ip_lock);
 138
 139         list_for_each_entry_safe(emi, n, &tmp_list, ei_list) {
 140                 list_del(&emi->ei_list);
 141                 kfree(emi);
 142         }
 143 }
 144
 145 /*
 146  * Is any part of emi2 contained within emi1
 147  */
 148 static int ocfs2_ei_is_contained(struct ocfs2_extent_map_item *emi1,
 149                                  struct ocfs2_extent_map_item *emi2)
 150 {
 151         unsigned int range1, range2;
 152
 153         /*
 154          * Check if logical start of emi2 is inside emi1
 155          */
 156         range1 = emi1->ei_cpos + emi1->ei_clusters;
 157         if (emi2->ei_cpos >= emi1->ei_cpos && emi2->ei_cpos < range1)
 158                 return 1;
 159
 160         /*
 161          * Check if logical end of emi2 is inside emi1
 162          */
 163         range2 = emi2->ei_cpos + emi2->ei_clusters;
 164         if (range2 > emi1->ei_cpos && range2 <= range1)
 165                 return 1;
 166
 167         return 0;
 168 }
 169
 170 static void ocfs2_copy_emi_fields(struct ocfs2_extent_map_item *dest,
 171                                   struct ocfs2_extent_map_item *src)
 172 {
 173         dest->ei_cpos = src->ei_cpos;
 174         dest->ei_phys = src->ei_phys;
 175         dest->ei_clusters = src->ei_clusters;
 176         dest->ei_flags = src->ei_flags;
 177 }
 178
 179 /*
 180  * Try to merge emi with ins. Returns 1 if merge succeeds, zero
 181  * otherwise.
 182  */
 183 static int ocfs2_try_to_merge_extent_map(struct ocfs2_extent_map_item *emi,
 184                                          struct ocfs2_extent_map_item *ins)
 185 {
 186         /*
 187          * Handle contiguousness
 188          */
 189         if (ins->ei_phys == (emi->ei_phys + emi->ei_clusters) &&
 190             ins->ei_cpos == (emi->ei_cpos + emi->ei_clusters) &&
 191             ins->ei_flags == emi->ei_flags) {
 192                 emi->ei_clusters += ins->ei_clusters;
 193                 return 1;
 194         } else if ((ins->ei_phys + ins->ei_clusters) == emi->ei_phys &&
 195                    (ins->ei_cpos + ins->ei_clusters) == emi->ei_cpos &&
 196                    ins->ei_flags == emi->ei_flags) {
 197                 emi->ei_phys = ins->ei_phys;
 198                 emi->ei_cpos = ins->ei_cpos;
 199                 emi->ei_clusters += ins->ei_clusters;
 200                 return 1;
 201         }
 202
 203         /*
 204          * Overlapping extents - this shouldn't happen unless we've
 205          * split an extent to change it's flags. That is exceedingly
 206          * rare, so there's no sense in trying to optimize it yet.
 207          */
 208         if (ocfs2_ei_is_contained(emi, ins) ||
 209             ocfs2_ei_is_contained(ins, emi)) {
 210                 ocfs2_copy_emi_fields(emi, ins);
 211                 return 1;
 212         }
 213
 214         /* No merge was possible. */
 215         return 0;
 216 }
 217
 218 /*
 219  * In order to reduce complexity on the caller, this insert function
 220  * is intentionally liberal in what it will accept.
 221  *
 222  * The only rule is that the truncate call *must* be used whenever
 223  * records have been deleted. This avoids inserting overlapping
 224  * records with different physical mappings.
 225  */
 226 void ocfs2_extent_map_insert_rec(struct inode *inode,
 227                                  struct ocfs2_extent_rec *rec)
 228 {
 229         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 230         struct ocfs2_extent_map *em = &oi->ip_extent_map;
 231         struct ocfs2_extent_map_item *emi, *new_emi = NULL;
 232         struct ocfs2_extent_map_item ins;
 233
 234         ins.ei_cpos = le32_to_cpu(rec->e_cpos);
 235         ins.ei_phys = ocfs2_blocks_to_clusters(inode->i_sb,
 236                                                le64_to_cpu(rec->e_blkno));
 237         ins.ei_clusters = le16_to_cpu(rec->e_leaf_clusters);
 238         ins.ei_flags = rec->e_flags;
 239
 240 search:
 241         spin_lock(&oi->ip_lock);
 242
 243         list_for_each_entry(emi, &em->em_list, ei_list) {
 244                 if (ocfs2_try_to_merge_extent_map(emi, &ins)) {
 245                         list_move(&emi->ei_list, &em->em_list);
 246                         spin_unlock(&oi->ip_lock);
 247                         goto out;
 248                 }
 249         }
 250
 251         /*
 252          * No item could be merged.
 253          *
 254          * Either allocate and add a new item, or overwrite the last recently
 255          * inserted.
 256          */
 257
 258         if (em->em_num_items < OCFS2_MAX_EXTENT_MAP_ITEMS) {
 259                 if (new_emi == NULL) {
 260                         spin_unlock(&oi->ip_lock);
 261
 262                         new_emi = kmalloc(sizeof(*new_emi), GFP_NOFS);
 263                         if (new_emi == NULL)
 264                                 goto out;
 265
 266                         goto search;
 267                 }
 268
 269                 ocfs2_copy_emi_fields(new_emi, &ins);
 270                 list_add(&new_emi->ei_list, &em->em_list);
 271                 em->em_num_items++;
 272                 new_emi = NULL;
 273         } else {
 274                 BUG_ON(list_empty(&em->em_list) || em->em_num_items == 0);
 275                 emi = list_entry(em->em_list.prev,
 276                                  struct ocfs2_extent_map_item, ei_list);
 277                 list_move(&emi->ei_list, &em->em_list);
 278                 ocfs2_copy_emi_fields(emi, &ins);
 279         }
 280
 281         spin_unlock(&oi->ip_lock);
 282
 283 out:
 284         if (new_emi)
 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         mapping_end -= cpos;
 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;
 794                 }
 795
 796                 if (rec.e_blkno == 0ULL) {
 797                         cpos += hole_size;
 798                         continue;
 799                 }
 800
 801                 fe_flags = 0;
 802                 if (rec.e_flags & OCFS2_EXT_UNWRITTEN)
 803                         fe_flags |= FIEMAP_EXTENT_UNWRITTEN;
 804                 if (rec.e_flags & OCFS2_EXT_REFCOUNTED)
 805                         fe_flags |= FIEMAP_EXTENT_SHARED;
 806                 if (is_last)
 807                         fe_flags |= FIEMAP_EXTENT_LAST;
 808                 len_bytes = (u64)le16_to_cpu(rec.e_leaf_clusters) << osb->s_clustersize_bits;
 809                 phys_bytes = le64_to_cpu(rec.e_blkno) << osb->sb->s_blocksize_bits;
 810                 virt_bytes = (u64)le32_to_cpu(rec.e_cpos) << osb->s_clustersize_bits;
 811
 812                 ret = fiemap_fill_next_extent(fieinfo, virt_bytes, phys_bytes,
 813                                               len_bytes, fe_flags);
 814                 if (ret)
 815                         break;
 816
 817                 cpos = le32_to_cpu(rec.e_cpos)+ le16_to_cpu(rec.e_leaf_clusters);
 818         }
 819
 820         if (ret > 0)
 821                 ret = 0;
 822
 823 out_unlock:
 824         brelse(di_bh);
 825
 826         up_read(&OCFS2_I(inode)->ip_alloc_sem);
 827
 828         ocfs2_inode_unlock(inode, 0);
 829 out:
 830
 831         return ret;
 832 }
 833
 834 int ocfs2_read_virt_blocks(struct inode *inode, u64 v_block, int nr,
 835                            struct buffer_head *bhs[], int flags,
 836                            int (*validate)(struct super_block *sb,
 837                                            struct buffer_head *bh))
 838 {
 839         int rc = 0;
 840         u64 p_block, p_count;
 841         int i, count, done = 0;
 842
 843         mlog_entry("(inode = %p, v_block = %llu, nr = %d, bhs = %p, "
 844                    "flags = %x, validate = %p)\n",
 845                    inode, (unsigned long long)v_block, nr, bhs, flags,
 846                    validate);
 847
 848         if (((v_block + nr - 1) << inode->i_sb->s_blocksize_bits) >=
 849             i_size_read(inode)) {
 850                 BUG_ON(!(flags & OCFS2_BH_READAHEAD));
 851                 goto out;
 852         }
 853
 854         while (done < nr) {
 855                 down_read(&OCFS2_I(inode)->ip_alloc_sem);
 856                 rc = ocfs2_extent_map_get_blocks(inode, v_block + done,
 857                                                  &p_block, &p_count, NULL);
 858                 up_read(&OCFS2_I(inode)->ip_alloc_sem);
 859                 if (rc) {
 860                         mlog_errno(rc);
 861                         break;
 862                 }
 863
 864                 if (!p_block) {
 865                         rc = -EIO;
 866                         mlog(ML_ERROR,
 867                              "Inode #%llu contains a hole at offset %llu\n",
 868                              (unsigned long long)OCFS2_I(inode)->ip_blkno,
 869                              (unsigned long long)(v_block + done) <<
 870                              inode->i_sb->s_blocksize_bits);
 871                         break;
 872                 }
 873
 874                 count = nr - done;
 875                 if (p_count < count)
 876                         count = p_count;
 877
 878                 /*
 879                  * If the caller passed us bhs, they should have come
 880                  * from a previous readahead call to this function.  Thus,
 881                  * they should have the right b_blocknr.
 882                  */
 883                 for (i = 0; i < count; i++) {
 884                         if (!bhs[done + i])
 885                                 continue;
 886                         BUG_ON(bhs[done + i]->b_blocknr != (p_block + i));
 887                 }
 888
 889                 rc = ocfs2_read_blocks(INODE_CACHE(inode), p_block, count,
 890                                        bhs + done, flags, validate);
 891                 if (rc) {
 892                         mlog_errno(rc);
 893                         break;
 894                 }
 895                 done += count;
 896         }
 897
 898 out:
 899         mlog_exit(rc);
 900         return rc;
 901 }
 902
 903