fs/ocfs2/alloc.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * alloc.c
   5  *
   6  * Extent allocs and frees
   7  *
   8  * Copyright (C) 2002, 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 as published by the Free Software Foundation; either
  13  * version 2 of the License, or (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public
  21  * License along with this program; if not, write to the
  22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  23  * Boston, MA 021110-1307, USA.
  24  */
  25
  26 #include <linux/fs.h>
  27 #include <linux/types.h>
  28 #include <linux/slab.h>
  29 #include <linux/highmem.h>
  30
  31 #define MLOG_MASK_PREFIX ML_DISK_ALLOC
  32 #include <cluster/masklog.h>
  33
  34 #include "ocfs2.h"
  35
  36 #include "alloc.h"
  37 #include "dlmglue.h"
  38 #include "extent_map.h"
  39 #include "inode.h"
  40 #include "journal.h"
  41 #include "localalloc.h"
  42 #include "suballoc.h"
  43 #include "sysfile.h"
  44 #include "file.h"
  45 #include "super.h"
  46 #include "uptodate.h"
  47
  48 #include "buffer_head_io.h"
  49
  50 static int ocfs2_extent_contig(struct inode *inode,
  51                                struct ocfs2_extent_rec *ext,
  52                                u64 blkno);
  53
  54 static int ocfs2_create_new_meta_bhs(struct ocfs2_super *osb,
  55                                      struct ocfs2_journal_handle *handle,
  56                                      struct inode *inode,
  57                                      int wanted,
  58                                      struct ocfs2_alloc_context *meta_ac,
  59                                      struct buffer_head *bhs[]);
  60
  61 static int ocfs2_add_branch(struct ocfs2_super *osb,
  62                             struct ocfs2_journal_handle *handle,
  63                             struct inode *inode,
  64                             struct buffer_head *fe_bh,
  65                             struct buffer_head *eb_bh,
  66                             struct buffer_head *last_eb_bh,
  67                             struct ocfs2_alloc_context *meta_ac);
  68
  69 static int ocfs2_shift_tree_depth(struct ocfs2_super *osb,
  70                                   struct ocfs2_journal_handle *handle,
  71                                   struct inode *inode,
  72                                   struct buffer_head *fe_bh,
  73                                   struct ocfs2_alloc_context *meta_ac,
  74                                   struct buffer_head **ret_new_eb_bh);
  75
  76 static int ocfs2_do_insert_extent(struct ocfs2_super *osb,
  77                                   struct ocfs2_journal_handle *handle,
  78                                   struct inode *inode,
  79                                   struct buffer_head *fe_bh,
  80                                   u64 blkno,
  81                                   u32 new_clusters);
  82
  83 static int ocfs2_find_branch_target(struct ocfs2_super *osb,
  84                                     struct inode *inode,
  85                                     struct buffer_head *fe_bh,
  86                                     struct buffer_head **target_bh);
  87
  88 static int ocfs2_find_new_last_ext_blk(struct ocfs2_super *osb,
  89                                        struct inode *inode,
  90                                        struct ocfs2_dinode *fe,
  91                                        unsigned int new_i_clusters,
  92                                        struct buffer_head *old_last_eb,
  93                                        struct buffer_head **new_last_eb);
  94
  95 static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc);
  96
  97 static int ocfs2_extent_contig(struct inode *inode,
  98                                struct ocfs2_extent_rec *ext,
  99                                u64 blkno)
 100 {
 101         return blkno == (le64_to_cpu(ext->e_blkno) +
 102                          ocfs2_clusters_to_blocks(inode->i_sb,
 103                                                   le32_to_cpu(ext->e_clusters)));
 104 }
 105
 106 /*
 107  * How many free extents have we got before we need more meta data?
 108  */
 109 int ocfs2_num_free_extents(struct ocfs2_super *osb,
 110                            struct inode *inode,
 111                            struct ocfs2_dinode *fe)
 112 {
 113         int retval;
 114         struct ocfs2_extent_list *el;
 115         struct ocfs2_extent_block *eb;
 116         struct buffer_head *eb_bh = NULL;
 117
 118         mlog_entry_void();
 119
 120         if (!OCFS2_IS_VALID_DINODE(fe)) {
 121                 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
 122                 retval = -EIO;
 123                 goto bail;
 124         }
 125
 126         if (fe->i_last_eb_blk) {
 127                 retval = ocfs2_read_block(osb, le64_to_cpu(fe->i_last_eb_blk),
 128                                           &eb_bh, OCFS2_BH_CACHED, inode);
 129                 if (retval < 0) {
 130                         mlog_errno(retval);
 131                         goto bail;
 132                 }
 133                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
 134                 el = &eb->h_list;
 135         } else
 136                 el = &fe->id2.i_list;
 137
 138         BUG_ON(el->l_tree_depth != 0);
 139
 140         retval = le16_to_cpu(el->l_count) - le16_to_cpu(el->l_next_free_rec);
 141 bail:
 142         if (eb_bh)
 143                 brelse(eb_bh);
 144
 145         mlog_exit(retval);
 146         return retval;
 147 }
 148
 149 /* expects array to already be allocated
 150  *
 151  * sets h_signature, h_blkno, h_suballoc_bit, h_suballoc_slot, and
 152  * l_count for you
 153  */
 154 static int ocfs2_create_new_meta_bhs(struct ocfs2_super *osb,
 155                                      struct ocfs2_journal_handle *handle,
 156                                      struct inode *inode,
 157                                      int wanted,
 158                                      struct ocfs2_alloc_context *meta_ac,
 159                                      struct buffer_head *bhs[])
 160 {
 161         int count, status, i;
 162         u16 suballoc_bit_start;
 163         u32 num_got;
 164         u64 first_blkno;
 165         struct ocfs2_extent_block *eb;
 166
 167         mlog_entry_void();
 168
 169         count = 0;
 170         while (count < wanted) {
 171                 status = ocfs2_claim_metadata(osb,
 172                                               handle,
 173                                               meta_ac,
 174                                               wanted - count,
 175                                               &suballoc_bit_start,
 176                                               &num_got,
 177                                               &first_blkno);
 178                 if (status < 0) {
 179                         mlog_errno(status);
 180                         goto bail;
 181                 }
 182
 183                 for(i = count;  i < (num_got + count); i++) {
 184                         bhs[i] = sb_getblk(osb->sb, first_blkno);
 185                         if (bhs[i] == NULL) {
 186                                 status = -EIO;
 187                                 mlog_errno(status);
 188                                 goto bail;
 189                         }
 190                         ocfs2_set_new_buffer_uptodate(inode, bhs[i]);
 191
 192                         status = ocfs2_journal_access(handle, inode, bhs[i],
 193                                                       OCFS2_JOURNAL_ACCESS_CREATE);
 194                         if (status < 0) {
 195                                 mlog_errno(status);
 196                                 goto bail;
 197                         }
 198
 199                         memset(bhs[i]->b_data, 0, osb->sb->s_blocksize);
 200                         eb = (struct ocfs2_extent_block *) bhs[i]->b_data;
 201                         /* Ok, setup the minimal stuff here. */
 202                         strcpy(eb->h_signature, OCFS2_EXTENT_BLOCK_SIGNATURE);
 203                         eb->h_blkno = cpu_to_le64(first_blkno);
 204                         eb->h_fs_generation = cpu_to_le32(osb->fs_generation);
 205
 206 #ifndef OCFS2_USE_ALL_METADATA_SUBALLOCATORS
 207                         /* we always use slot zero's suballocator */
 208                         eb->h_suballoc_slot = 0;
 209 #else
 210                         eb->h_suballoc_slot = cpu_to_le16(osb->slot_num);
 211 #endif
 212                         eb->h_suballoc_bit = cpu_to_le16(suballoc_bit_start);
 213                         eb->h_list.l_count =
 214                                 cpu_to_le16(ocfs2_extent_recs_per_eb(osb->sb));
 215
 216                         suballoc_bit_start++;
 217                         first_blkno++;
 218
 219                         /* We'll also be dirtied by the caller, so
 220                          * this isn't absolutely necessary. */
 221                         status = ocfs2_journal_dirty(handle, bhs[i]);
 222                         if (status < 0) {
 223                                 mlog_errno(status);
 224                                 goto bail;
 225                         }
 226                 }
 227
 228                 count += num_got;
 229         }
 230
 231         status = 0;
 232 bail:
 233         if (status < 0) {
 234                 for(i = 0; i < wanted; i++) {
 235                         if (bhs[i])
 236                                 brelse(bhs[i]);
 237                         bhs[i] = NULL;
 238                 }
 239         }
 240         mlog_exit(status);
 241         return status;
 242 }
 243
 244 /*
 245  * Add an entire tree branch to our inode. eb_bh is the extent block
 246  * to start at, if we don't want to start the branch at the dinode
 247  * structure.
 248  *
 249  * last_eb_bh is required as we have to update it's next_leaf pointer
 250  * for the new last extent block.
 251  *
 252  * the new branch will be 'empty' in the sense that every block will
 253  * contain a single record with e_clusters == 0.
 254  */
 255 static int ocfs2_add_branch(struct ocfs2_super *osb,
 256                             struct ocfs2_journal_handle *handle,
 257                             struct inode *inode,
 258                             struct buffer_head *fe_bh,
 259                             struct buffer_head *eb_bh,
 260                             struct buffer_head *last_eb_bh,
 261                             struct ocfs2_alloc_context *meta_ac)
 262 {
 263         int status, new_blocks, i;
 264         u64 next_blkno, new_last_eb_blk;
 265         struct buffer_head *bh;
 266         struct buffer_head **new_eb_bhs = NULL;
 267         struct ocfs2_dinode *fe;
 268         struct ocfs2_extent_block *eb;
 269         struct ocfs2_extent_list  *eb_el;
 270         struct ocfs2_extent_list  *el;
 271
 272         mlog_entry_void();
 273
 274         BUG_ON(!last_eb_bh);
 275
 276         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 277
 278         if (eb_bh) {
 279                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
 280                 el = &eb->h_list;
 281         } else
 282                 el = &fe->id2.i_list;
 283
 284         /* we never add a branch to a leaf. */
 285         BUG_ON(!el->l_tree_depth);
 286
 287         new_blocks = le16_to_cpu(el->l_tree_depth);
 288
 289         /* allocate the number of new eb blocks we need */
 290         new_eb_bhs = kcalloc(new_blocks, sizeof(struct buffer_head *),
 291                              GFP_KERNEL);
 292         if (!new_eb_bhs) {
 293                 status = -ENOMEM;
 294                 mlog_errno(status);
 295                 goto bail;
 296         }
 297
 298         status = ocfs2_create_new_meta_bhs(osb, handle, inode, new_blocks,
 299                                            meta_ac, new_eb_bhs);
 300         if (status < 0) {
 301                 mlog_errno(status);
 302                 goto bail;
 303         }
 304
 305         /* Note: new_eb_bhs[new_blocks - 1] is the guy which will be
 306          * linked with the rest of the tree.
 307          * conversly, new_eb_bhs[0] is the new bottommost leaf.
 308          *
 309          * when we leave the loop, new_last_eb_blk will point to the
 310          * newest leaf, and next_blkno will point to the topmost extent
 311          * block. */
 312         next_blkno = new_last_eb_blk = 0;
 313         for(i = 0; i < new_blocks; i++) {
 314                 bh = new_eb_bhs[i];
 315                 eb = (struct ocfs2_extent_block *) bh->b_data;
 316                 if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
 317                         OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
 318                         status = -EIO;
 319                         goto bail;
 320                 }
 321                 eb_el = &eb->h_list;
 322
 323                 status = ocfs2_journal_access(handle, inode, bh,
 324                                               OCFS2_JOURNAL_ACCESS_CREATE);
 325                 if (status < 0) {
 326                         mlog_errno(status);
 327                         goto bail;
 328                 }
 329
 330                 eb->h_next_leaf_blk = 0;
 331                 eb_el->l_tree_depth = cpu_to_le16(i);
 332                 eb_el->l_next_free_rec = cpu_to_le16(1);
 333                 eb_el->l_recs[0].e_cpos = fe->i_clusters;
 334                 eb_el->l_recs[0].e_blkno = cpu_to_le64(next_blkno);
 335                 eb_el->l_recs[0].e_clusters = cpu_to_le32(0);
 336                 if (!eb_el->l_tree_depth)
 337                         new_last_eb_blk = le64_to_cpu(eb->h_blkno);
 338
 339                 status = ocfs2_journal_dirty(handle, bh);
 340                 if (status < 0) {
 341                         mlog_errno(status);
 342                         goto bail;
 343                 }
 344
 345                 next_blkno = le64_to_cpu(eb->h_blkno);
 346         }
 347
 348         /* This is a bit hairy. We want to update up to three blocks
 349          * here without leaving any of them in an inconsistent state
 350          * in case of error. We don't have to worry about
 351          * journal_dirty erroring as it won't unless we've aborted the
 352          * handle (in which case we would never be here) so reserving
 353          * the write with journal_access is all we need to do. */
 354         status = ocfs2_journal_access(handle, inode, last_eb_bh,
 355                                       OCFS2_JOURNAL_ACCESS_WRITE);
 356         if (status < 0) {
 357                 mlog_errno(status);
 358                 goto bail;
 359         }
 360         status = ocfs2_journal_access(handle, inode, fe_bh,
 361                                       OCFS2_JOURNAL_ACCESS_WRITE);
 362         if (status < 0) {
 363                 mlog_errno(status);
 364                 goto bail;
 365         }
 366         if (eb_bh) {
 367                 status = ocfs2_journal_access(handle, inode, eb_bh,
 368                                               OCFS2_JOURNAL_ACCESS_WRITE);
 369                 if (status < 0) {
 370                         mlog_errno(status);
 371                         goto bail;
 372                 }
 373         }
 374
 375         /* Link the new branch into the rest of the tree (el will
 376          * either be on the fe, or the extent block passed in. */
 377         i = le16_to_cpu(el->l_next_free_rec);
 378         el->l_recs[i].e_blkno = cpu_to_le64(next_blkno);
 379         el->l_recs[i].e_cpos = fe->i_clusters;
 380         el->l_recs[i].e_clusters = 0;
 381         le16_add_cpu(&el->l_next_free_rec, 1);
 382
 383         /* fe needs a new last extent block pointer, as does the
 384          * next_leaf on the previously last-extent-block. */
 385         fe->i_last_eb_blk = cpu_to_le64(new_last_eb_blk);
 386
 387         eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
 388         eb->h_next_leaf_blk = cpu_to_le64(new_last_eb_blk);
 389
 390         status = ocfs2_journal_dirty(handle, last_eb_bh);
 391         if (status < 0)
 392                 mlog_errno(status);
 393         status = ocfs2_journal_dirty(handle, fe_bh);
 394         if (status < 0)
 395                 mlog_errno(status);
 396         if (eb_bh) {
 397                 status = ocfs2_journal_dirty(handle, eb_bh);
 398                 if (status < 0)
 399                         mlog_errno(status);
 400         }
 401
 402         status = 0;
 403 bail:
 404         if (new_eb_bhs) {
 405                 for (i = 0; i < new_blocks; i++)
 406                         if (new_eb_bhs[i])
 407                                 brelse(new_eb_bhs[i]);
 408                 kfree(new_eb_bhs);
 409         }
 410
 411         mlog_exit(status);
 412         return status;
 413 }
 414
 415 /*
 416  * adds another level to the allocation tree.
 417  * returns back the new extent block so you can add a branch to it
 418  * after this call.
 419  */
 420 static int ocfs2_shift_tree_depth(struct ocfs2_super *osb,
 421                                   struct ocfs2_journal_handle *handle,
 422                                   struct inode *inode,
 423                                   struct buffer_head *fe_bh,
 424                                   struct ocfs2_alloc_context *meta_ac,
 425                                   struct buffer_head **ret_new_eb_bh)
 426 {
 427         int status, i;
 428         struct buffer_head *new_eb_bh = NULL;
 429         struct ocfs2_dinode *fe;
 430         struct ocfs2_extent_block *eb;
 431         struct ocfs2_extent_list  *fe_el;
 432         struct ocfs2_extent_list  *eb_el;
 433
 434         mlog_entry_void();
 435
 436         status = ocfs2_create_new_meta_bhs(osb, handle, inode, 1, meta_ac,
 437                                            &new_eb_bh);
 438         if (status < 0) {
 439                 mlog_errno(status);
 440                 goto bail;
 441         }
 442
 443         eb = (struct ocfs2_extent_block *) new_eb_bh->b_data;
 444         if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
 445                 OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
 446                 status = -EIO;
 447                 goto bail;
 448         }
 449
 450         eb_el = &eb->h_list;
 451         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 452         fe_el = &fe->id2.i_list;
 453
 454         status = ocfs2_journal_access(handle, inode, new_eb_bh,
 455                                       OCFS2_JOURNAL_ACCESS_CREATE);
 456         if (status < 0) {
 457                 mlog_errno(status);
 458                 goto bail;
 459         }
 460
 461         /* copy the fe data into the new extent block */
 462         eb_el->l_tree_depth = fe_el->l_tree_depth;
 463         eb_el->l_next_free_rec = fe_el->l_next_free_rec;
 464         for(i = 0; i < le16_to_cpu(fe_el->l_next_free_rec); i++) {
 465                 eb_el->l_recs[i].e_cpos = fe_el->l_recs[i].e_cpos;
 466                 eb_el->l_recs[i].e_clusters = fe_el->l_recs[i].e_clusters;
 467                 eb_el->l_recs[i].e_blkno = fe_el->l_recs[i].e_blkno;
 468         }
 469
 470         status = ocfs2_journal_dirty(handle, new_eb_bh);
 471         if (status < 0) {
 472                 mlog_errno(status);
 473                 goto bail;
 474         }
 475
 476         status = ocfs2_journal_access(handle, inode, fe_bh,
 477                                       OCFS2_JOURNAL_ACCESS_WRITE);
 478         if (status < 0) {
 479                 mlog_errno(status);
 480                 goto bail;
 481         }
 482
 483         /* update fe now */
 484         le16_add_cpu(&fe_el->l_tree_depth, 1);
 485         fe_el->l_recs[0].e_cpos = 0;
 486         fe_el->l_recs[0].e_blkno = eb->h_blkno;
 487         fe_el->l_recs[0].e_clusters = fe->i_clusters;
 488         for(i = 1; i < le16_to_cpu(fe_el->l_next_free_rec); i++) {
 489                 fe_el->l_recs[i].e_cpos = 0;
 490                 fe_el->l_recs[i].e_clusters = 0;
 491                 fe_el->l_recs[i].e_blkno = 0;
 492         }
 493         fe_el->l_next_free_rec = cpu_to_le16(1);
 494
 495         /* If this is our 1st tree depth shift, then last_eb_blk
 496          * becomes the allocated extent block */
 497         if (fe_el->l_tree_depth == cpu_to_le16(1))
 498                 fe->i_last_eb_blk = eb->h_blkno;
 499
 500         status = ocfs2_journal_dirty(handle, fe_bh);
 501         if (status < 0) {
 502                 mlog_errno(status);
 503                 goto bail;
 504         }
 505
 506         *ret_new_eb_bh = new_eb_bh;
 507         new_eb_bh = NULL;
 508         status = 0;
 509 bail:
 510         if (new_eb_bh)
 511                 brelse(new_eb_bh);
 512
 513         mlog_exit(status);
 514         return status;
 515 }
 516
 517 /*
 518  * Expects the tree to already have room in the rightmost leaf for the
 519  * extent.  Updates all the extent blocks (and the dinode) on the way
 520  * down.
 521  */
 522 static int ocfs2_do_insert_extent(struct ocfs2_super *osb,
 523                                   struct ocfs2_journal_handle *handle,
 524                                   struct inode *inode,
 525                                   struct buffer_head *fe_bh,
 526                                   u64 start_blk,
 527                                   u32 new_clusters)
 528 {
 529         int status, i, num_bhs = 0;
 530         u64 next_blkno;
 531         u16 next_free;
 532         struct buffer_head **eb_bhs = NULL;
 533         struct ocfs2_dinode *fe;
 534         struct ocfs2_extent_block *eb;
 535         struct ocfs2_extent_list  *el;
 536
 537         mlog_entry_void();
 538
 539         status = ocfs2_journal_access(handle, inode, fe_bh,
 540                                       OCFS2_JOURNAL_ACCESS_WRITE);
 541         if (status < 0) {
 542                 mlog_errno(status);
 543                 goto bail;
 544         }
 545
 546         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 547         el = &fe->id2.i_list;
 548         if (el->l_tree_depth) {
 549                 /* This is another operation where we want to be
 550                  * careful about our tree updates. An error here means
 551                  * none of the previous changes we made should roll
 552                  * forward. As a result, we have to record the buffers
 553                  * for this part of the tree in an array and reserve a
 554                  * journal write to them before making any changes. */
 555                 num_bhs = le16_to_cpu(fe->id2.i_list.l_tree_depth);
 556                 eb_bhs = kcalloc(num_bhs, sizeof(struct buffer_head *),
 557                                  GFP_KERNEL);
 558                 if (!eb_bhs) {
 559                         status = -ENOMEM;
 560                         mlog_errno(status);
 561                         goto bail;
 562                 }
 563
 564                 i = 0;
 565                 while(el->l_tree_depth) {
 566                         next_free = le16_to_cpu(el->l_next_free_rec);
 567                         if (next_free == 0) {
 568                                 ocfs2_error(inode->i_sb,
 569                                             "Dinode %"MLFu64" has a bad "
 570                                             "extent list",
 571                                             OCFS2_I(inode)->ip_blkno);
 572                                 status = -EIO;
 573                                 goto bail;
 574                         }
 575                         next_blkno = le64_to_cpu(el->l_recs[next_free - 1].e_blkno);
 576
 577                         BUG_ON(i >= num_bhs);
 578                         status = ocfs2_read_block(osb, next_blkno, &eb_bhs[i],
 579                                                   OCFS2_BH_CACHED, inode);
 580                         if (status < 0) {
 581                                 mlog_errno(status);
 582                                 goto bail;
 583                         }
 584                         eb = (struct ocfs2_extent_block *) eb_bhs[i]->b_data;
 585                         if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
 586                                 OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb,
 587                                                                  eb);
 588                                 status = -EIO;
 589                                 goto bail;
 590                         }
 591
 592                         status = ocfs2_journal_access(handle, inode, eb_bhs[i],
 593                                                       OCFS2_JOURNAL_ACCESS_WRITE);
 594                         if (status < 0) {
 595                                 mlog_errno(status);
 596                                 goto bail;
 597                         }
 598
 599                         el = &eb->h_list;
 600                         i++;
 601                         /* When we leave this loop, eb_bhs[num_bhs - 1] will
 602                          * hold the bottom-most leaf extent block. */
 603                 }
 604                 BUG_ON(el->l_tree_depth);
 605
 606                 el = &fe->id2.i_list;
 607                 /* If we have tree depth, then the fe update is
 608                  * trivial, and we want to switch el out for the
 609                  * bottom-most leaf in order to update it with the
 610                  * actual extent data below. */
 611                 next_free = le16_to_cpu(el->l_next_free_rec);
 612                 if (next_free == 0) {
 613                         ocfs2_error(inode->i_sb,
 614                                     "Dinode %"MLFu64" has a bad "
 615                                     "extent list",
 616                                     OCFS2_I(inode)->ip_blkno);
 617                         status = -EIO;
 618                         goto bail;
 619                 }
 620                 le32_add_cpu(&el->l_recs[next_free - 1].e_clusters,
 621                              new_clusters);
 622                 /* (num_bhs - 1) to avoid the leaf */
 623                 for(i = 0; i < (num_bhs - 1); i++) {
 624                         eb = (struct ocfs2_extent_block *) eb_bhs[i]->b_data;
 625                         el = &eb->h_list;
 626
 627                         /* finally, make our actual change to the
 628                          * intermediate extent blocks. */
 629                         next_free = le16_to_cpu(el->l_next_free_rec);
 630                         le32_add_cpu(&el->l_recs[next_free - 1].e_clusters,
 631                                      new_clusters);
 632
 633                         status = ocfs2_journal_dirty(handle, eb_bhs[i]);
 634                         if (status < 0)
 635                                 mlog_errno(status);
 636                 }
 637                 BUG_ON(i != (num_bhs - 1));
 638                 /* note that the leaf block wasn't touched in
 639                  * the loop above */
 640                 eb = (struct ocfs2_extent_block *) eb_bhs[num_bhs - 1]->b_data;
 641                 el = &eb->h_list;
 642                 BUG_ON(el->l_tree_depth);
 643         }
 644
 645         /* yay, we can finally add the actual extent now! */
 646         i = le16_to_cpu(el->l_next_free_rec) - 1;
 647         if (le16_to_cpu(el->l_next_free_rec) &&
 648             ocfs2_extent_contig(inode, &el->l_recs[i], start_blk)) {
 649                 le32_add_cpu(&el->l_recs[i].e_clusters, new_clusters);
 650         } else if (le16_to_cpu(el->l_next_free_rec) &&
 651                    (le32_to_cpu(el->l_recs[i].e_clusters) == 0)) {
 652                 /* having an empty extent at eof is legal. */
 653                 if (el->l_recs[i].e_cpos != fe->i_clusters) {
 654                         ocfs2_error(inode->i_sb,
 655                                     "Dinode %"MLFu64" trailing extent is bad: "
 656                                     "cpos (%u) != number of clusters (%u)",
 657                                     le32_to_cpu(el->l_recs[i].e_cpos),
 658                                     le32_to_cpu(fe->i_clusters));
 659                         status = -EIO;
 660                         goto bail;
 661                 }
 662                 el->l_recs[i].e_blkno = cpu_to_le64(start_blk);
 663                 el->l_recs[i].e_clusters = cpu_to_le32(new_clusters);
 664         } else {
 665                 /* No contiguous record, or no empty record at eof, so
 666                  * we add a new one. */
 667
 668                 BUG_ON(le16_to_cpu(el->l_next_free_rec) >=
 669                        le16_to_cpu(el->l_count));
 670                 i = le16_to_cpu(el->l_next_free_rec);
 671
 672                 el->l_recs[i].e_blkno = cpu_to_le64(start_blk);
 673                 el->l_recs[i].e_clusters = cpu_to_le32(new_clusters);
 674                 el->l_recs[i].e_cpos = fe->i_clusters;
 675                 le16_add_cpu(&el->l_next_free_rec, 1);
 676         }
 677
 678         /*
 679          * extent_map errors are not fatal, so they are ignored outside
 680          * of flushing the thing.
 681          */
 682         status = ocfs2_extent_map_append(inode, &el->l_recs[i],
 683                                          new_clusters);
 684         if (status) {
 685                 mlog_errno(status);
 686                 ocfs2_extent_map_drop(inode, le32_to_cpu(fe->i_clusters));
 687         }
 688
 689         status = ocfs2_journal_dirty(handle, fe_bh);
 690         if (status < 0)
 691                 mlog_errno(status);
 692         if (fe->id2.i_list.l_tree_depth) {
 693                 status = ocfs2_journal_dirty(handle, eb_bhs[num_bhs - 1]);
 694                 if (status < 0)
 695                         mlog_errno(status);
 696         }
 697
 698         status = 0;
 699 bail:
 700         if (eb_bhs) {
 701                 for (i = 0; i < num_bhs; i++)
 702                         if (eb_bhs[i])
 703                                 brelse(eb_bhs[i]);
 704                 kfree(eb_bhs);
 705         }
 706
 707         mlog_exit(status);
 708         return status;
 709 }
 710
 711 /*
 712  * Should only be called when there is no space left in any of the
 713  * leaf nodes. What we want to do is find the lowest tree depth
 714  * non-leaf extent block with room for new records. There are three
 715  * valid results of this search:
 716  *
 717  * 1) a lowest extent block is found, then we pass it back in
 718  *    *lowest_eb_bh and return '0'
 719  *
 720  * 2) the search fails to find anything, but the dinode has room. We
 721  *    pass NULL back in *lowest_eb_bh, but still return '0'
 722  *
 723  * 3) the search fails to find anything AND the dinode is full, in
 724  *    which case we return > 0
 725  *
 726  * return status < 0 indicates an error.
 727  */
 728 static int ocfs2_find_branch_target(struct ocfs2_super *osb,
 729                                     struct inode *inode,
 730                                     struct buffer_head *fe_bh,
 731                                     struct buffer_head **target_bh)
 732 {
 733         int status = 0, i;
 734         u64 blkno;
 735         struct ocfs2_dinode *fe;
 736         struct ocfs2_extent_block *eb;
 737         struct ocfs2_extent_list  *el;
 738         struct buffer_head *bh = NULL;
 739         struct buffer_head *lowest_bh = NULL;
 740
 741         mlog_entry_void();
 742
 743         *target_bh = NULL;
 744
 745         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 746         el = &fe->id2.i_list;
 747
 748         while(le16_to_cpu(el->l_tree_depth) > 1) {
 749                 if (le16_to_cpu(el->l_next_free_rec) == 0) {
 750                         ocfs2_error(inode->i_sb, "Dinode %"MLFu64" has empty "
 751                                     "extent list (next_free_rec == 0)",
 752                                     OCFS2_I(inode)->ip_blkno);
 753                         status = -EIO;
 754                         goto bail;
 755                 }
 756                 i = le16_to_cpu(el->l_next_free_rec) - 1;
 757                 blkno = le64_to_cpu(el->l_recs[i].e_blkno);
 758                 if (!blkno) {
 759                         ocfs2_error(inode->i_sb, "Dinode %"MLFu64" has extent "
 760                                     "list where extent # %d has no physical "
 761                                     "block start",
 762                                     OCFS2_I(inode)->ip_blkno, i);
 763                         status = -EIO;
 764                         goto bail;
 765                 }
 766
 767                 if (bh) {
 768                         brelse(bh);
 769                         bh = NULL;
 770                 }
 771
 772                 status = ocfs2_read_block(osb, blkno, &bh, OCFS2_BH_CACHED,
 773                                           inode);
 774                 if (status < 0) {
 775                         mlog_errno(status);
 776                         goto bail;
 777                 }
 778
 779                 eb = (struct ocfs2_extent_block *) bh->b_data;
 780                 if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
 781                         OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
 782                         status = -EIO;
 783                         goto bail;
 784                 }
 785                 el = &eb->h_list;
 786
 787                 if (le16_to_cpu(el->l_next_free_rec) <
 788                     le16_to_cpu(el->l_count)) {
 789                         if (lowest_bh)
 790                                 brelse(lowest_bh);
 791                         lowest_bh = bh;
 792                         get_bh(lowest_bh);
 793                 }
 794         }
 795
 796         /* If we didn't find one and the fe doesn't have any room,
 797          * then return '1' */
 798         if (!lowest_bh
 799             && (fe->id2.i_list.l_next_free_rec == fe->id2.i_list.l_count))
 800                 status = 1;
 801
 802         *target_bh = lowest_bh;
 803 bail:
 804         if (bh)
 805                 brelse(bh);
 806
 807         mlog_exit(status);
 808         return status;
 809 }
 810
 811 /* the caller needs to update fe->i_clusters */
 812 int ocfs2_insert_extent(struct ocfs2_super *osb,
 813                         struct ocfs2_journal_handle *handle,
 814                         struct inode *inode,
 815                         struct buffer_head *fe_bh,
 816                         u64 start_blk,
 817                         u32 new_clusters,
 818                         struct ocfs2_alloc_context *meta_ac)
 819 {
 820         int status, i, shift;
 821         struct buffer_head *last_eb_bh = NULL;
 822         struct buffer_head *bh = NULL;
 823         struct ocfs2_dinode *fe;
 824         struct ocfs2_extent_block *eb;
 825         struct ocfs2_extent_list  *el;
 826
 827         mlog_entry_void();
 828
 829         mlog(0, "add %u clusters starting at block %"MLFu64" to "
 830                 "inode %"MLFu64"\n",
 831              new_clusters, start_blk, OCFS2_I(inode)->ip_blkno);
 832
 833         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 834         el = &fe->id2.i_list;
 835
 836         if (el->l_tree_depth) {
 837                 /* jump to end of tree */
 838                 status = ocfs2_read_block(osb, le64_to_cpu(fe->i_last_eb_blk),
 839                                           &last_eb_bh, OCFS2_BH_CACHED, inode);
 840                 if (status < 0) {
 841                         mlog_exit(status);
 842                         goto bail;
 843                 }
 844                 eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
 845                 el = &eb->h_list;
 846         }
 847
 848         /* Can we allocate without adding/shifting tree bits? */
 849         i = le16_to_cpu(el->l_next_free_rec) - 1;
 850         if (le16_to_cpu(el->l_next_free_rec) == 0
 851             || (le16_to_cpu(el->l_next_free_rec) < le16_to_cpu(el->l_count))
 852             || le32_to_cpu(el->l_recs[i].e_clusters) == 0
 853             || ocfs2_extent_contig(inode, &el->l_recs[i], start_blk))
 854                 goto out_add;
 855
 856         mlog(0, "ocfs2_allocate_extent: couldn't do a simple add, traversing "
 857              "tree now.\n");
 858
 859         shift = ocfs2_find_branch_target(osb, inode, fe_bh, &bh);
 860         if (shift < 0) {
 861                 status = shift;
 862                 mlog_errno(status);
 863                 goto bail;
 864         }
 865
 866         /* We traveled all the way to the bottom of the allocation tree
 867          * and didn't find room for any more extents - we need to add
 868          * another tree level */
 869         if (shift) {
 870                 /* if we hit a leaf, we'd better be empty :) */
 871                 BUG_ON(le16_to_cpu(el->l_next_free_rec) !=
 872                        le16_to_cpu(el->l_count));
 873                 BUG_ON(bh);
 874                 mlog(0, "ocfs2_allocate_extent: need to shift tree depth "
 875                      "(current = %u)\n",
 876                      le16_to_cpu(fe->id2.i_list.l_tree_depth));
 877
 878                 /* ocfs2_shift_tree_depth will return us a buffer with
 879                  * the new extent block (so we can pass that to
 880                  * ocfs2_add_branch). */
 881                 status = ocfs2_shift_tree_depth(osb, handle, inode, fe_bh,
 882                                                 meta_ac, &bh);
 883                 if (status < 0) {
 884                         mlog_errno(status);
 885                         goto bail;
 886                 }
 887                 /* Special case: we have room now if we shifted from
 888                  * tree_depth 0 */
 889                 if (fe->id2.i_list.l_tree_depth == cpu_to_le16(1))
 890                         goto out_add;
 891         }
 892
 893         /* call ocfs2_add_branch to add the final part of the tree with
 894          * the new data. */
 895         mlog(0, "ocfs2_allocate_extent: add branch. bh = %p\n", bh);
 896         status = ocfs2_add_branch(osb, handle, inode, fe_bh, bh, last_eb_bh,
 897                                   meta_ac);
 898         if (status < 0) {
 899                 mlog_errno(status);
 900                 goto bail;
 901         }
 902
 903 out_add:
 904         /* Finally, we can add clusters. */
 905         status = ocfs2_do_insert_extent(osb, handle, inode, fe_bh,
 906                                         start_blk, new_clusters);
 907         if (status < 0)
 908                 mlog_errno(status);
 909
 910 bail:
 911         if (bh)
 912                 brelse(bh);
 913
 914         if (last_eb_bh)
 915                 brelse(last_eb_bh);
 916
 917         mlog_exit(status);
 918         return status;
 919 }
 920
 921 static inline int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb)
 922 {
 923         struct buffer_head *tl_bh = osb->osb_tl_bh;
 924         struct ocfs2_dinode *di;
 925         struct ocfs2_truncate_log *tl;
 926
 927         di = (struct ocfs2_dinode *) tl_bh->b_data;
 928         tl = &di->id2.i_dealloc;
 929
 930         mlog_bug_on_msg(le16_to_cpu(tl->tl_used) > le16_to_cpu(tl->tl_count),
 931                         "slot %d, invalid truncate log parameters: used = "
 932                         "%u, count = %u\n", osb->slot_num,
 933                         le16_to_cpu(tl->tl_used), le16_to_cpu(tl->tl_count));
 934         return le16_to_cpu(tl->tl_used) == le16_to_cpu(tl->tl_count);
 935 }
 936
 937 static int ocfs2_truncate_log_can_coalesce(struct ocfs2_truncate_log *tl,
 938                                            unsigned int new_start)
 939 {
 940         unsigned int tail_index;
 941         unsigned int current_tail;
 942
 943         /* No records, nothing to coalesce */
 944         if (!le16_to_cpu(tl->tl_used))
 945                 return 0;
 946
 947         tail_index = le16_to_cpu(tl->tl_used) - 1;
 948         current_tail = le32_to_cpu(tl->tl_recs[tail_index].t_start);
 949         current_tail += le32_to_cpu(tl->tl_recs[tail_index].t_clusters);
 950
 951         return current_tail == new_start;
 952 }
 953
 954 static int ocfs2_truncate_log_append(struct ocfs2_super *osb,
 955                                      struct ocfs2_journal_handle *handle,
 956                                      u64 start_blk,
 957                                      unsigned int num_clusters)
 958 {
 959         int status, index;
 960         unsigned int start_cluster, tl_count;
 961         struct inode *tl_inode = osb->osb_tl_inode;
 962         struct buffer_head *tl_bh = osb->osb_tl_bh;
 963         struct ocfs2_dinode *di;
 964         struct ocfs2_truncate_log *tl;
 965
 966         mlog_entry("start_blk = %"MLFu64", num_clusters = %u\n", start_blk,
 967                    num_clusters);
 968
 969         BUG_ON(mutex_trylock(&tl_inode->i_mutex));
 970
 971         start_cluster = ocfs2_blocks_to_clusters(osb->sb, start_blk);
 972
 973         di = (struct ocfs2_dinode *) tl_bh->b_data;
 974         tl = &di->id2.i_dealloc;
 975         if (!OCFS2_IS_VALID_DINODE(di)) {
 976                 OCFS2_RO_ON_INVALID_DINODE(osb->sb, di);
 977                 status = -EIO;
 978                 goto bail;
 979         }
 980
 981         tl_count = le16_to_cpu(tl->tl_count);
 982         mlog_bug_on_msg(tl_count > ocfs2_truncate_recs_per_inode(osb->sb) ||
 983                         tl_count == 0,
 984                         "Truncate record count on #%"MLFu64" invalid ("
 985                         "wanted %u, actual %u\n", OCFS2_I(tl_inode)->ip_blkno,
 986                         ocfs2_truncate_recs_per_inode(osb->sb),
 987                         le16_to_cpu(tl->tl_count));
 988
 989         /* Caller should have known to flush before calling us. */
 990         index = le16_to_cpu(tl->tl_used);
 991         if (index >= tl_count) {
 992                 status = -ENOSPC;
 993                 mlog_errno(status);
 994                 goto bail;
 995         }
 996
 997         status = ocfs2_journal_access(handle, tl_inode, tl_bh,
 998                                       OCFS2_JOURNAL_ACCESS_WRITE);
 999         if (status < 0) {
1000                 mlog_errno(status);
1001                 goto bail;
1002         }
1003
1004         mlog(0, "Log truncate of %u clusters starting at cluster %u to "
1005              "%"MLFu64" (index = %d)\n", num_clusters, start_cluster,
1006              OCFS2_I(tl_inode)->ip_blkno, index);
1007
1008         if (ocfs2_truncate_log_can_coalesce(tl, start_cluster)) {
1009                 /*
1010                  * Move index back to the record we are coalescing with.
1011                  * ocfs2_truncate_log_can_coalesce() guarantees nonzero
1012                  */
1013                 index--;
1014
1015                 num_clusters += le32_to_cpu(tl->tl_recs[index].t_clusters);
1016                 mlog(0, "Coalesce with index %u (start = %u, clusters = %u)\n",
1017                      index, le32_to_cpu(tl->tl_recs[index].t_start),
1018                      num_clusters);
1019         } else {
1020                 tl->tl_recs[index].t_start = cpu_to_le32(start_cluster);
1021                 tl->tl_used = cpu_to_le16(index + 1);
1022         }
1023         tl->tl_recs[index].t_clusters = cpu_to_le32(num_clusters);
1024
1025         status = ocfs2_journal_dirty(handle, tl_bh);
1026         if (status < 0) {
1027                 mlog_errno(status);
1028                 goto bail;
1029         }
1030
1031 bail:
1032         mlog_exit(status);
1033         return status;
1034 }
1035
1036 static int ocfs2_replay_truncate_records(struct ocfs2_super *osb,
1037                                          struct ocfs2_journal_handle *handle,
1038                                          struct inode *data_alloc_inode,
1039                                          struct buffer_head *data_alloc_bh)
1040 {
1041         int status = 0;
1042         int i;
1043         unsigned int num_clusters;
1044         u64 start_blk;
1045         struct ocfs2_truncate_rec rec;
1046         struct ocfs2_dinode *di;
1047         struct ocfs2_truncate_log *tl;
1048         struct inode *tl_inode = osb->osb_tl_inode;
1049         struct buffer_head *tl_bh = osb->osb_tl_bh;
1050
1051         mlog_entry_void();
1052
1053         di = (struct ocfs2_dinode *) tl_bh->b_data;
1054         tl = &di->id2.i_dealloc;
1055         i = le16_to_cpu(tl->tl_used) - 1;
1056         while (i >= 0) {
1057                 /* Caller has given us at least enough credits to
1058                  * update the truncate log dinode */
1059                 status = ocfs2_journal_access(handle, tl_inode, tl_bh,
1060                                               OCFS2_JOURNAL_ACCESS_WRITE);
1061                 if (status < 0) {
1062                         mlog_errno(status);
1063                         goto bail;
1064                 }
1065
1066                 tl->tl_used = cpu_to_le16(i);
1067
1068                 status = ocfs2_journal_dirty(handle, tl_bh);
1069                 if (status < 0) {
1070                         mlog_errno(status);
1071                         goto bail;
1072                 }
1073
1074                 /* TODO: Perhaps we can calculate the bulk of the
1075                  * credits up front rather than extending like
1076                  * this. */
1077                 status = ocfs2_extend_trans(handle,
1078                                             OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC);
1079                 if (status < 0) {
1080                         mlog_errno(status);
1081                         goto bail;
1082                 }
1083
1084                 rec = tl->tl_recs[i];
1085                 start_blk = ocfs2_clusters_to_blocks(data_alloc_inode->i_sb,
1086                                                     le32_to_cpu(rec.t_start));
1087                 num_clusters = le32_to_cpu(rec.t_clusters);
1088
1089                 /* if start_blk is not set, we ignore the record as
1090                  * invalid. */
1091                 if (start_blk) {
1092                         mlog(0, "free record %d, start = %u, clusters = %u\n",
1093                              i, le32_to_cpu(rec.t_start), num_clusters);
1094
1095                         status = ocfs2_free_clusters(handle, data_alloc_inode,
1096                                                      data_alloc_bh, start_blk,
1097                                                      num_clusters);
1098                         if (status < 0) {
1099                                 mlog_errno(status);
1100                                 goto bail;
1101                         }
1102                 }
1103                 i--;
1104         }
1105
1106 bail:
1107         mlog_exit(status);
1108         return status;
1109 }
1110
1111 /* Expects you to already be holding tl_inode->i_mutex */
1112 static int __ocfs2_flush_truncate_log(struct ocfs2_super *osb)
1113 {
1114         int status;
1115         unsigned int num_to_flush;
1116         struct ocfs2_journal_handle *handle = NULL;
1117         struct inode *tl_inode = osb->osb_tl_inode;
1118         struct inode *data_alloc_inode = NULL;
1119         struct buffer_head *tl_bh = osb->osb_tl_bh;
1120         struct buffer_head *data_alloc_bh = NULL;
1121         struct ocfs2_dinode *di;
1122         struct ocfs2_truncate_log *tl;
1123
1124         mlog_entry_void();
1125
1126         BUG_ON(mutex_trylock(&tl_inode->i_mutex));
1127
1128         di = (struct ocfs2_dinode *) tl_bh->b_data;
1129         tl = &di->id2.i_dealloc;
1130         if (!OCFS2_IS_VALID_DINODE(di)) {
1131                 OCFS2_RO_ON_INVALID_DINODE(osb->sb, di);
1132                 status = -EIO;
1133                 goto bail;
1134         }
1135
1136         num_to_flush = le16_to_cpu(tl->tl_used);
1137         mlog(0, "Flush %u records from truncate log #%"MLFu64"\n",
1138              num_to_flush, OCFS2_I(tl_inode)->ip_blkno);
1139         if (!num_to_flush) {
1140                 status = 0;
1141                 goto bail;
1142         }
1143
1144         handle = ocfs2_alloc_handle(osb);
1145         if (!handle) {
1146                 status = -ENOMEM;
1147                 mlog_errno(status);
1148                 goto bail;
1149         }
1150
1151         data_alloc_inode = ocfs2_get_system_file_inode(osb,
1152                                                        GLOBAL_BITMAP_SYSTEM_INODE,
1153                                                        OCFS2_INVALID_SLOT);
1154         if (!data_alloc_inode) {
1155                 status = -EINVAL;
1156                 mlog(ML_ERROR, "Could not get bitmap inode!\n");
1157                 goto bail;
1158         }
1159
1160         ocfs2_handle_add_inode(handle, data_alloc_inode);
1161         status = ocfs2_meta_lock(data_alloc_inode, handle, &data_alloc_bh, 1);
1162         if (status < 0) {
1163                 mlog_errno(status);
1164                 goto bail;
1165         }
1166
1167         handle = ocfs2_start_trans(osb, handle, OCFS2_TRUNCATE_LOG_UPDATE);
1168         if (IS_ERR(handle)) {
1169                 status = PTR_ERR(handle);
1170                 handle = NULL;
1171                 mlog_errno(status);
1172                 goto bail;
1173         }
1174
1175         status = ocfs2_replay_truncate_records(osb, handle, data_alloc_inode,
1176                                                data_alloc_bh);
1177         if (status < 0) {
1178                 mlog_errno(status);
1179                 goto bail;
1180         }
1181
1182 bail:
1183         if (handle)
1184                 ocfs2_commit_trans(handle);
1185
1186         if (data_alloc_inode)
1187                 iput(data_alloc_inode);
1188
1189         if (data_alloc_bh)
1190                 brelse(data_alloc_bh);
1191
1192         mlog_exit(status);
1193         return status;
1194 }
1195
1196 int ocfs2_flush_truncate_log(struct ocfs2_super *osb)
1197 {
1198         int status;
1199         struct inode *tl_inode = osb->osb_tl_inode;
1200
1201         mutex_lock(&tl_inode->i_mutex);
1202         status = __ocfs2_flush_truncate_log(osb);
1203         mutex_unlock(&tl_inode->i_mutex);
1204
1205         return status;
1206 }
1207
1208 static void ocfs2_truncate_log_worker(void *data)
1209 {
1210         int status;
1211         struct ocfs2_super *osb = data;
1212
1213         mlog_entry_void();
1214
1215         status = ocfs2_flush_truncate_log(osb);
1216         if (status < 0)
1217                 mlog_errno(status);
1218
1219         mlog_exit(status);
1220 }
1221
1222 #define OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL (2 * HZ)
1223 void ocfs2_schedule_truncate_log_flush(struct ocfs2_super *osb,
1224                                        int cancel)
1225 {
1226         if (osb->osb_tl_inode) {
1227                 /* We want to push off log flushes while truncates are
1228                  * still running. */
1229                 if (cancel)
1230                         cancel_delayed_work(&osb->osb_truncate_log_wq);
1231
1232                 queue_delayed_work(ocfs2_wq, &osb->osb_truncate_log_wq,
1233                                    OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL);
1234         }
1235 }
1236
1237 static int ocfs2_get_truncate_log_info(struct ocfs2_super *osb,
1238                                        int slot_num,
1239                                        struct inode **tl_inode,
1240                                        struct buffer_head **tl_bh)
1241 {
1242         int status;
1243         struct inode *inode = NULL;
1244         struct buffer_head *bh = NULL;
1245
1246         inode = ocfs2_get_system_file_inode(osb,
1247                                            TRUNCATE_LOG_SYSTEM_INODE,
1248                                            slot_num);
1249         if (!inode) {
1250                 status = -EINVAL;
1251                 mlog(ML_ERROR, "Could not get load truncate log inode!\n");
1252                 goto bail;
1253         }
1254
1255         status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh,
1256                                   OCFS2_BH_CACHED, inode);
1257         if (status < 0) {
1258                 iput(inode);
1259                 mlog_errno(status);
1260                 goto bail;
1261         }
1262
1263         *tl_inode = inode;
1264         *tl_bh    = bh;
1265 bail:
1266         mlog_exit(status);
1267         return status;
1268 }
1269
1270 /* called during the 1st stage of node recovery. we stamp a clean
1271  * truncate log and pass back a copy for processing later. if the
1272  * truncate log does not require processing, a *tl_copy is set to
1273  * NULL. */
1274 int ocfs2_begin_truncate_log_recovery(struct ocfs2_super *osb,
1275                                       int slot_num,
1276                                       struct ocfs2_dinode **tl_copy)
1277 {
1278         int status;
1279         struct inode *tl_inode = NULL;
1280         struct buffer_head *tl_bh = NULL;
1281         struct ocfs2_dinode *di;
1282         struct ocfs2_truncate_log *tl;
1283
1284         *tl_copy = NULL;
1285
1286         mlog(0, "recover truncate log from slot %d\n", slot_num);
1287
1288         status = ocfs2_get_truncate_log_info(osb, slot_num, &tl_inode, &tl_bh);
1289         if (status < 0) {
1290                 mlog_errno(status);
1291                 goto bail;
1292         }
1293
1294         di = (struct ocfs2_dinode *) tl_bh->b_data;
1295         tl = &di->id2.i_dealloc;
1296         if (!OCFS2_IS_VALID_DINODE(di)) {
1297                 OCFS2_RO_ON_INVALID_DINODE(tl_inode->i_sb, di);
1298                 status = -EIO;
1299                 goto bail;
1300         }
1301
1302         if (le16_to_cpu(tl->tl_used)) {
1303                 mlog(0, "We'll have %u logs to recover\n",
1304                      le16_to_cpu(tl->tl_used));
1305
1306                 *tl_copy = kmalloc(tl_bh->b_size, GFP_KERNEL);
1307                 if (!(*tl_copy)) {
1308                         status = -ENOMEM;
1309                         mlog_errno(status);
1310                         goto bail;
1311                 }
1312
1313                 /* Assuming the write-out below goes well, this copy
1314                  * will be passed back to recovery for processing. */
1315                 memcpy(*tl_copy, tl_bh->b_data, tl_bh->b_size);
1316
1317                 /* All we need to do to clear the truncate log is set
1318                  * tl_used. */
1319                 tl->tl_used = 0;
1320
1321                 status = ocfs2_write_block(osb, tl_bh, tl_inode);
1322                 if (status < 0) {
1323                         mlog_errno(status);
1324                         goto bail;
1325                 }
1326         }
1327
1328 bail:
1329         if (tl_inode)
1330                 iput(tl_inode);
1331         if (tl_bh)
1332                 brelse(tl_bh);
1333
1334         if (status < 0 && (*tl_copy)) {
1335                 kfree(*tl_copy);
1336                 *tl_copy = NULL;
1337         }
1338
1339         mlog_exit(status);
1340         return status;
1341 }
1342
1343 int ocfs2_complete_truncate_log_recovery(struct ocfs2_super *osb,
1344                                          struct ocfs2_dinode *tl_copy)
1345 {
1346         int status = 0;
1347         int i;
1348         unsigned int clusters, num_recs, start_cluster;
1349         u64 start_blk;
1350         struct ocfs2_journal_handle *handle;
1351         struct inode *tl_inode = osb->osb_tl_inode;
1352         struct ocfs2_truncate_log *tl;
1353
1354         mlog_entry_void();
1355
1356         if (OCFS2_I(tl_inode)->ip_blkno == le64_to_cpu(tl_copy->i_blkno)) {
1357                 mlog(ML_ERROR, "Asked to recover my own truncate log!\n");
1358                 return -EINVAL;
1359         }
1360
1361         tl = &tl_copy->id2.i_dealloc;
1362         num_recs = le16_to_cpu(tl->tl_used);
1363         mlog(0, "cleanup %u records from %"MLFu64"\n", num_recs,
1364              tl_copy->i_blkno);
1365
1366         mutex_lock(&tl_inode->i_mutex);
1367         for(i = 0; i < num_recs; i++) {
1368                 if (ocfs2_truncate_log_needs_flush(osb)) {
1369                         status = __ocfs2_flush_truncate_log(osb);
1370                         if (status < 0) {
1371                                 mlog_errno(status);
1372                                 goto bail_up;
1373                         }
1374                 }
1375
1376                 handle = ocfs2_start_trans(osb, NULL,
1377                                            OCFS2_TRUNCATE_LOG_UPDATE);
1378                 if (IS_ERR(handle)) {
1379                         status = PTR_ERR(handle);
1380                         mlog_errno(status);
1381                         goto bail_up;
1382                 }
1383
1384                 clusters = le32_to_cpu(tl->tl_recs[i].t_clusters);
1385                 start_cluster = le32_to_cpu(tl->tl_recs[i].t_start);
1386                 start_blk = ocfs2_clusters_to_blocks(osb->sb, start_cluster);
1387
1388                 status = ocfs2_truncate_log_append(osb, handle,
1389                                                    start_blk, clusters);
1390                 ocfs2_commit_trans(handle);
1391                 if (status < 0) {
1392                         mlog_errno(status);
1393                         goto bail_up;
1394                 }
1395         }
1396
1397 bail_up:
1398         mutex_unlock(&tl_inode->i_mutex);
1399
1400         mlog_exit(status);
1401         return status;
1402 }
1403
1404 void ocfs2_truncate_log_shutdown(struct ocfs2_super *osb)
1405 {
1406         int status;
1407         struct inode *tl_inode = osb->osb_tl_inode;
1408
1409         mlog_entry_void();
1410
1411         if (tl_inode) {
1412                 cancel_delayed_work(&osb->osb_truncate_log_wq);
1413                 flush_workqueue(ocfs2_wq);
1414
1415                 status = ocfs2_flush_truncate_log(osb);
1416                 if (status < 0)
1417                         mlog_errno(status);
1418
1419                 brelse(osb->osb_tl_bh);
1420                 iput(osb->osb_tl_inode);
1421         }
1422
1423         mlog_exit_void();
1424 }
1425
1426 int ocfs2_truncate_log_init(struct ocfs2_super *osb)
1427 {
1428         int status;
1429         struct inode *tl_inode = NULL;
1430         struct buffer_head *tl_bh = NULL;
1431
1432         mlog_entry_void();
1433
1434         status = ocfs2_get_truncate_log_info(osb,
1435                                              osb->slot_num,
1436                                              &tl_inode,
1437                                              &tl_bh);
1438         if (status < 0)
1439                 mlog_errno(status);
1440
1441         /* ocfs2_truncate_log_shutdown keys on the existence of
1442          * osb->osb_tl_inode so we don't set any of the osb variables
1443          * until we're sure all is well. */
1444         INIT_WORK(&osb->osb_truncate_log_wq, ocfs2_truncate_log_worker, osb);
1445         osb->osb_tl_bh    = tl_bh;
1446         osb->osb_tl_inode = tl_inode;
1447
1448         mlog_exit(status);
1449         return status;
1450 }
1451
1452 /* This function will figure out whether the currently last extent
1453  * block will be deleted, and if it will, what the new last extent
1454  * block will be so we can update his h_next_leaf_blk field, as well
1455  * as the dinodes i_last_eb_blk */
1456 static int ocfs2_find_new_last_ext_blk(struct ocfs2_super *osb,
1457                                        struct inode *inode,
1458                                        struct ocfs2_dinode *fe,
1459                                        u32 new_i_clusters,
1460                                        struct buffer_head *old_last_eb,
1461                                        struct buffer_head **new_last_eb)
1462 {
1463         int i, status = 0;
1464         u64 block = 0;
1465         struct ocfs2_extent_block *eb;
1466         struct ocfs2_extent_list *el;
1467         struct buffer_head *bh = NULL;
1468
1469         *new_last_eb = NULL;
1470
1471         if (!OCFS2_IS_VALID_DINODE(fe)) {
1472                 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
1473                 status = -EIO;
1474                 goto bail;
1475         }
1476
1477         /* we have no tree, so of course, no last_eb. */
1478         if (!fe->id2.i_list.l_tree_depth)
1479                 goto bail;
1480
1481         /* trunc to zero special case - this makes tree_depth = 0
1482          * regardless of what it is.  */
1483         if (!new_i_clusters)
1484                 goto bail;
1485
1486         eb = (struct ocfs2_extent_block *) old_last_eb->b_data;
1487         el = &(eb->h_list);
1488         BUG_ON(!el->l_next_free_rec);
1489
1490         /* Make sure that this guy will actually be empty after we
1491          * clear away the data. */
1492         if (le32_to_cpu(el->l_recs[0].e_cpos) < new_i_clusters)
1493                 goto bail;
1494
1495         /* Ok, at this point, we know that last_eb will definitely
1496          * change, so lets traverse the tree and find the second to
1497          * last extent block. */
1498         el = &(fe->id2.i_list);
1499         /* go down the tree, */
1500         do {
1501                 for(i = (le16_to_cpu(el->l_next_free_rec) - 1); i >= 0; i--) {
1502                         if (le32_to_cpu(el->l_recs[i].e_cpos) <
1503                             new_i_clusters) {
1504                                 block = le64_to_cpu(el->l_recs[i].e_blkno);
1505                                 break;
1506                         }
1507                 }
1508                 BUG_ON(i < 0);
1509
1510                 if (bh) {
1511                         brelse(bh);
1512                         bh = NULL;
1513                 }
1514
1515                 status = ocfs2_read_block(osb, block, &bh, OCFS2_BH_CACHED,
1516                                          inode);
1517                 if (status < 0) {
1518                         mlog_errno(status);
1519                         goto bail;
1520                 }
1521                 eb = (struct ocfs2_extent_block *) bh->b_data;
1522                 el = &eb->h_list;
1523                 if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
1524                         OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
1525                         status = -EIO;
1526                         goto bail;
1527                 }
1528         } while (el->l_tree_depth);
1529
1530         *new_last_eb = bh;
1531         get_bh(*new_last_eb);
1532         mlog(0, "returning block %"MLFu64"\n", le64_to_cpu(eb->h_blkno));
1533 bail:
1534         if (bh)
1535                 brelse(bh);
1536
1537         return status;
1538 }
1539
1540 static int ocfs2_do_truncate(struct ocfs2_super *osb,
1541                              unsigned int clusters_to_del,
1542                              struct inode *inode,
1543                              struct buffer_head *fe_bh,
1544                              struct buffer_head *old_last_eb_bh,
1545                              struct ocfs2_journal_handle *handle,
1546                              struct ocfs2_truncate_context *tc)
1547 {
1548         int status, i, depth;
1549         struct ocfs2_dinode *fe;
1550         struct ocfs2_extent_block *eb;
1551         struct ocfs2_extent_block *last_eb = NULL;
1552         struct ocfs2_extent_list *el;
1553         struct buffer_head *eb_bh = NULL;
1554         struct buffer_head *last_eb_bh = NULL;
1555         u64 next_eb = 0;
1556         u64 delete_blk = 0;
1557
1558         fe = (struct ocfs2_dinode *) fe_bh->b_data;
1559
1560         status = ocfs2_find_new_last_ext_blk(osb,
1561                                              inode,
1562                                              fe,
1563                                              le32_to_cpu(fe->i_clusters) -
1564                                                         clusters_to_del,
1565                                              old_last_eb_bh,
1566                                              &last_eb_bh);
1567         if (status < 0) {
1568                 mlog_errno(status);
1569                 goto bail;
1570         }
1571         if (last_eb_bh)
1572                 last_eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
1573
1574         status = ocfs2_journal_access(handle, inode, fe_bh,
1575                                       OCFS2_JOURNAL_ACCESS_WRITE);
1576         if (status < 0) {
1577                 mlog_errno(status);
1578                 goto bail;
1579         }
1580         el = &(fe->id2.i_list);
1581
1582         spin_lock(&OCFS2_I(inode)->ip_lock);
1583         OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters) -
1584                                       clusters_to_del;
1585         spin_unlock(&OCFS2_I(inode)->ip_lock);
1586         le32_add_cpu(&fe->i_clusters, -clusters_to_del);
1587         fe->i_mtime = cpu_to_le64(CURRENT_TIME.tv_sec);
1588         fe->i_mtime_nsec = cpu_to_le32(CURRENT_TIME.tv_nsec);
1589
1590         i = le16_to_cpu(el->l_next_free_rec) - 1;
1591
1592         BUG_ON(le32_to_cpu(el->l_recs[i].e_clusters) < clusters_to_del);
1593         le32_add_cpu(&el->l_recs[i].e_clusters, -clusters_to_del);
1594         /* tree depth zero, we can just delete the clusters, otherwise
1595          * we need to record the offset of the next level extent block
1596          * as we may overwrite it. */
1597         if (!el->l_tree_depth)
1598                 delete_blk = le64_to_cpu(el->l_recs[i].e_blkno)
1599                         + ocfs2_clusters_to_blocks(osb->sb,
1600                                         le32_to_cpu(el->l_recs[i].e_clusters));
1601         else
1602                 next_eb = le64_to_cpu(el->l_recs[i].e_blkno);
1603
1604         if (!el->l_recs[i].e_clusters) {
1605                 /* if we deleted the whole extent record, then clear
1606                  * out the other fields and update the extent
1607                  * list. For depth > 0 trees, we've already recorded
1608                  * the extent block in 'next_eb' */
1609                 el->l_recs[i].e_cpos = 0;
1610                 el->l_recs[i].e_blkno = 0;
1611                 BUG_ON(!el->l_next_free_rec);
1612                 le16_add_cpu(&el->l_next_free_rec, -1);
1613         }
1614
1615         depth = le16_to_cpu(el->l_tree_depth);
1616         if (!fe->i_clusters) {
1617                 /* trunc to zero is a special case. */
1618                 el->l_tree_depth = 0;
1619                 fe->i_last_eb_blk = 0;
1620         } else if (last_eb)
1621                 fe->i_last_eb_blk = last_eb->h_blkno;
1622
1623         status = ocfs2_journal_dirty(handle, fe_bh);
1624         if (status < 0) {
1625                 mlog_errno(status);
1626                 goto bail;
1627         }
1628
1629         if (last_eb) {
1630                 /* If there will be a new last extent block, then by
1631                  * definition, there cannot be any leaves to the right of
1632                  * him. */
1633                 status = ocfs2_journal_access(handle, inode, last_eb_bh,
1634                                               OCFS2_JOURNAL_ACCESS_WRITE);
1635                 if (status < 0) {
1636                         mlog_errno(status);
1637                         goto bail;
1638                 }
1639                 last_eb->h_next_leaf_blk = 0;
1640                 status = ocfs2_journal_dirty(handle, last_eb_bh);
1641                 if (status < 0) {
1642                         mlog_errno(status);
1643                         goto bail;
1644                 }
1645         }
1646
1647         /* if our tree depth > 0, update all the tree blocks below us. */
1648         while (depth) {
1649                 mlog(0, "traveling tree (depth = %d, next_eb = %"MLFu64")\n",
1650                      depth,  next_eb);
1651                 status = ocfs2_read_block(osb, next_eb, &eb_bh,
1652                                           OCFS2_BH_CACHED, inode);
1653                 if (status < 0) {
1654                         mlog_errno(status);
1655                         goto bail;
1656                 }
1657                 eb = (struct ocfs2_extent_block *)eb_bh->b_data;
1658                 if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
1659                         OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
1660                         status = -EIO;
1661                         goto bail;
1662                 }
1663                 el = &(eb->h_list);
1664
1665                 status = ocfs2_journal_access(handle, inode, eb_bh,
1666                                               OCFS2_JOURNAL_ACCESS_WRITE);
1667                 if (status < 0) {
1668                         mlog_errno(status);
1669                         goto bail;
1670                 }
1671
1672                 BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0);
1673                 BUG_ON(depth != (le16_to_cpu(el->l_tree_depth) + 1));
1674
1675                 i = le16_to_cpu(el->l_next_free_rec) - 1;
1676
1677                 mlog(0, "extent block %"MLFu64", before: record %d: "
1678                      "(%u, %u, %"MLFu64"), next = %u\n",
1679                      le64_to_cpu(eb->h_blkno), i,
1680                      le32_to_cpu(el->l_recs[i].e_cpos),
1681                      le32_to_cpu(el->l_recs[i].e_clusters),
1682                      le64_to_cpu(el->l_recs[i].e_blkno),
1683                      le16_to_cpu(el->l_next_free_rec));
1684
1685                 BUG_ON(le32_to_cpu(el->l_recs[i].e_clusters) < clusters_to_del);
1686                 le32_add_cpu(&el->l_recs[i].e_clusters, -clusters_to_del);
1687
1688                 next_eb = le64_to_cpu(el->l_recs[i].e_blkno);
1689                 /* bottom-most block requires us to delete data.*/
1690                 if (!el->l_tree_depth)
1691                         delete_blk = le64_to_cpu(el->l_recs[i].e_blkno)
1692                                 + ocfs2_clusters_to_blocks(osb->sb,
1693                                         le32_to_cpu(el->l_recs[i].e_clusters));
1694                 if (!el->l_recs[i].e_clusters) {
1695                         el->l_recs[i].e_cpos = 0;
1696                         el->l_recs[i].e_blkno = 0;
1697                         BUG_ON(!el->l_next_free_rec);
1698                         le16_add_cpu(&el->l_next_free_rec, -1);
1699                 }
1700                 mlog(0, "extent block %"MLFu64", after: record %d: "
1701                      "(%u, %u, %"MLFu64"), next = %u\n",
1702                      le64_to_cpu(eb->h_blkno), i,
1703                      le32_to_cpu(el->l_recs[i].e_cpos),
1704                      le32_to_cpu(el->l_recs[i].e_clusters),
1705                      le64_to_cpu(el->l_recs[i].e_blkno),
1706                      le16_to_cpu(el->l_next_free_rec));
1707
1708                 status = ocfs2_journal_dirty(handle, eb_bh);
1709                 if (status < 0) {
1710                         mlog_errno(status);
1711                         goto bail;
1712                 }
1713
1714                 if (!el->l_next_free_rec) {
1715                         mlog(0, "deleting this extent block.\n");
1716
1717                         ocfs2_remove_from_cache(inode, eb_bh);
1718
1719                         BUG_ON(eb->h_suballoc_slot);
1720                         BUG_ON(el->l_recs[0].e_clusters);
1721                         BUG_ON(el->l_recs[0].e_cpos);
1722                         BUG_ON(el->l_recs[0].e_blkno);
1723                         status = ocfs2_free_extent_block(handle,
1724                                                          tc->tc_ext_alloc_inode,
1725                                                          tc->tc_ext_alloc_bh,
1726                                                          eb);
1727                         if (status < 0) {
1728                                 mlog_errno(status);
1729                                 goto bail;
1730                         }
1731                 }
1732                 brelse(eb_bh);
1733                 eb_bh = NULL;
1734                 depth--;
1735         }
1736
1737         BUG_ON(!delete_blk);
1738         status = ocfs2_truncate_log_append(osb, handle, delete_blk,
1739                                            clusters_to_del);
1740         if (status < 0) {
1741                 mlog_errno(status);
1742                 goto bail;
1743         }
1744         status = 0;
1745 bail:
1746         if (!status)
1747                 ocfs2_extent_map_trunc(inode, le32_to_cpu(fe->i_clusters));
1748         else
1749                 ocfs2_extent_map_drop(inode, 0);
1750         mlog_exit(status);
1751         return status;
1752 }
1753
1754 /*
1755  * It is expected, that by the time you call this function,
1756  * inode->i_size and fe->i_size have been adjusted.
1757  *
1758  * WARNING: This will kfree the truncate context
1759  */
1760 int ocfs2_commit_truncate(struct ocfs2_super *osb,
1761                           struct inode *inode,
1762                           struct buffer_head *fe_bh,
1763                           struct ocfs2_truncate_context *tc)
1764 {
1765         int status, i, credits, tl_sem = 0;
1766         u32 clusters_to_del, target_i_clusters;
1767         u64 last_eb = 0;
1768         struct ocfs2_dinode *fe;
1769         struct ocfs2_extent_block *eb;
1770         struct ocfs2_extent_list *el;
1771         struct buffer_head *last_eb_bh;
1772         struct ocfs2_journal_handle *handle = NULL;
1773         struct inode *tl_inode = osb->osb_tl_inode;
1774
1775         mlog_entry_void();
1776
1777         down_write(&OCFS2_I(inode)->ip_alloc_sem);
1778
1779         target_i_clusters = ocfs2_clusters_for_bytes(osb->sb,
1780                                                      i_size_read(inode));
1781
1782         last_eb_bh = tc->tc_last_eb_bh;
1783         tc->tc_last_eb_bh = NULL;
1784
1785         fe = (struct ocfs2_dinode *) fe_bh->b_data;
1786
1787         if (fe->id2.i_list.l_tree_depth) {
1788                 eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
1789                 el = &eb->h_list;
1790         } else
1791                 el = &fe->id2.i_list;
1792         last_eb = le64_to_cpu(fe->i_last_eb_blk);
1793 start:
1794         mlog(0, "ocfs2_commit_truncate: fe->i_clusters = %u, "
1795              "last_eb = %"MLFu64", fe->i_last_eb_blk = %"MLFu64", "
1796              "fe->id2.i_list.l_tree_depth = %u last_eb_bh = %p\n",
1797              le32_to_cpu(fe->i_clusters), last_eb,
1798              le64_to_cpu(fe->i_last_eb_blk),
1799              le16_to_cpu(fe->id2.i_list.l_tree_depth), last_eb_bh);
1800
1801         if (last_eb != le64_to_cpu(fe->i_last_eb_blk)) {
1802                 mlog(0, "last_eb changed!\n");
1803                 BUG_ON(!fe->id2.i_list.l_tree_depth);
1804                 last_eb = le64_to_cpu(fe->i_last_eb_blk);
1805                 /* i_last_eb_blk may have changed, read it if
1806                  * necessary. We don't have to worry about the
1807                  * truncate to zero case here (where there becomes no
1808                  * last_eb) because we never loop back after our work
1809                  * is done. */
1810                 if (last_eb_bh) {
1811                         brelse(last_eb_bh);
1812                         last_eb_bh = NULL;
1813                 }
1814
1815                 status = ocfs2_read_block(osb, last_eb,
1816                                           &last_eb_bh, OCFS2_BH_CACHED,
1817                                           inode);
1818                 if (status < 0) {
1819                         mlog_errno(status);
1820                         goto bail;
1821                 }
1822                 eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
1823                 if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
1824                         OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
1825                         status = -EIO;
1826                         goto bail;
1827                 }
1828                 el = &(eb->h_list);
1829         }
1830
1831         /* by now, el will point to the extent list on the bottom most
1832          * portion of this tree. */
1833         i = le16_to_cpu(el->l_next_free_rec) - 1;
1834         if (le32_to_cpu(el->l_recs[i].e_cpos) >= target_i_clusters)
1835                 clusters_to_del = le32_to_cpu(el->l_recs[i].e_clusters);
1836         else
1837                 clusters_to_del = (le32_to_cpu(el->l_recs[i].e_clusters) +
1838                                    le32_to_cpu(el->l_recs[i].e_cpos)) -
1839                                   target_i_clusters;
1840
1841         mlog(0, "clusters_to_del = %u in this pass\n", clusters_to_del);
1842
1843         mutex_lock(&tl_inode->i_mutex);
1844         tl_sem = 1;
1845         /* ocfs2_truncate_log_needs_flush guarantees us at least one
1846          * record is free for use. If there isn't any, we flush to get
1847          * an empty truncate log.  */
1848         if (ocfs2_truncate_log_needs_flush(osb)) {
1849                 status = __ocfs2_flush_truncate_log(osb);
1850                 if (status < 0) {
1851                         mlog_errno(status);
1852                         goto bail;
1853                 }
1854         }
1855
1856         credits = ocfs2_calc_tree_trunc_credits(osb->sb, clusters_to_del,
1857                                                 fe, el);
1858         handle = ocfs2_start_trans(osb, NULL, credits);
1859         if (IS_ERR(handle)) {
1860                 status = PTR_ERR(handle);
1861                 handle = NULL;
1862                 mlog_errno(status);
1863                 goto bail;
1864         }
1865
1866         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1867         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
1868         if (status < 0)
1869                 mlog_errno(status);
1870
1871         status = ocfs2_do_truncate(osb, clusters_to_del, inode, fe_bh,
1872                                    last_eb_bh, handle, tc);
1873         if (status < 0) {
1874                 mlog_errno(status);
1875                 goto bail;
1876         }
1877
1878         mutex_unlock(&tl_inode->i_mutex);
1879         tl_sem = 0;
1880
1881         ocfs2_commit_trans(handle);
1882         handle = NULL;
1883
1884         BUG_ON(le32_to_cpu(fe->i_clusters) < target_i_clusters);
1885         if (le32_to_cpu(fe->i_clusters) > target_i_clusters)
1886                 goto start;
1887 bail:
1888         up_write(&OCFS2_I(inode)->ip_alloc_sem);
1889
1890         ocfs2_schedule_truncate_log_flush(osb, 1);
1891
1892         if (tl_sem)
1893                 mutex_unlock(&tl_inode->i_mutex);
1894
1895         if (handle)
1896                 ocfs2_commit_trans(handle);
1897
1898         if (last_eb_bh)
1899                 brelse(last_eb_bh);
1900
1901         /* This will drop the ext_alloc cluster lock for us */
1902         ocfs2_free_truncate_context(tc);
1903
1904         mlog_exit(status);
1905         return status;
1906 }
1907
1908
1909 /*
1910  * Expects the inode to already be locked. This will figure out which
1911  * inodes need to be locked and will put them on the returned truncate
1912  * context.
1913  */
1914 int ocfs2_prepare_truncate(struct ocfs2_super *osb,
1915                            struct inode *inode,
1916                            struct buffer_head *fe_bh,
1917                            struct ocfs2_truncate_context **tc)
1918 {
1919         int status, metadata_delete;
1920         unsigned int new_i_clusters;
1921         struct ocfs2_dinode *fe;
1922         struct ocfs2_extent_block *eb;
1923         struct ocfs2_extent_list *el;
1924         struct buffer_head *last_eb_bh = NULL;
1925         struct inode *ext_alloc_inode = NULL;
1926         struct buffer_head *ext_alloc_bh = NULL;
1927
1928         mlog_entry_void();
1929
1930         *tc = NULL;
1931
1932         new_i_clusters = ocfs2_clusters_for_bytes(osb->sb,
1933                                                   i_size_read(inode));
1934         fe = (struct ocfs2_dinode *) fe_bh->b_data;
1935
1936         mlog(0, "fe->i_clusters = %u, new_i_clusters = %u, fe->i_size ="
1937              "%"MLFu64"\n", fe->i_clusters, new_i_clusters, fe->i_size);
1938
1939         if (le32_to_cpu(fe->i_clusters) <= new_i_clusters) {
1940                 ocfs2_error(inode->i_sb, "Dinode %"MLFu64" has cluster count "
1941                             "%u and size %"MLFu64" whereas struct inode has "
1942                             "cluster count %u and size %llu which caused an "
1943                             "invalid truncate to %u clusters.",
1944                             le64_to_cpu(fe->i_blkno),
1945                             le32_to_cpu(fe->i_clusters),
1946                             le64_to_cpu(fe->i_size),
1947                             OCFS2_I(inode)->ip_clusters, i_size_read(inode),
1948                             new_i_clusters);
1949                 mlog_meta_lvb(ML_ERROR, &OCFS2_I(inode)->ip_meta_lockres);
1950                 status = -EIO;
1951                 goto bail;
1952         }
1953
1954         *tc = kcalloc(1, sizeof(struct ocfs2_truncate_context), GFP_KERNEL);
1955         if (!(*tc)) {
1956                 status = -ENOMEM;
1957                 mlog_errno(status);
1958                 goto bail;
1959         }
1960
1961         metadata_delete = 0;
1962         if (fe->id2.i_list.l_tree_depth) {
1963                 /* If we have a tree, then the truncate may result in
1964                  * metadata deletes. Figure this out from the
1965                  * rightmost leaf block.*/
1966                 status = ocfs2_read_block(osb, le64_to_cpu(fe->i_last_eb_blk),
1967                                           &last_eb_bh, OCFS2_BH_CACHED, inode);
1968                 if (status < 0) {
1969                         mlog_errno(status);
1970                         goto bail;
1971                 }
1972                 eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
1973                 if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
1974                         OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
1975
1976                         brelse(last_eb_bh);
1977                         status = -EIO;
1978                         goto bail;
1979                 }
1980                 el = &(eb->h_list);
1981                 if (le32_to_cpu(el->l_recs[0].e_cpos) >= new_i_clusters)
1982                         metadata_delete = 1;
1983         }
1984
1985         (*tc)->tc_last_eb_bh = last_eb_bh;
1986
1987         if (metadata_delete) {
1988                 mlog(0, "Will have to delete metadata for this trunc. "
1989                      "locking allocator.\n");
1990                 ext_alloc_inode = ocfs2_get_system_file_inode(osb, EXTENT_ALLOC_SYSTEM_INODE, 0);
1991                 if (!ext_alloc_inode) {
1992                         status = -ENOMEM;
1993                         mlog_errno(status);
1994                         goto bail;
1995                 }
1996
1997                 mutex_lock(&ext_alloc_inode->i_mutex);
1998                 (*tc)->tc_ext_alloc_inode = ext_alloc_inode;
1999
2000                 status = ocfs2_meta_lock(ext_alloc_inode,
2001                                          NULL,
2002                                          &ext_alloc_bh,
2003                                          1);
2004                 if (status < 0) {
2005                         mlog_errno(status);
2006                         goto bail;
2007                 }
2008                 (*tc)->tc_ext_alloc_bh = ext_alloc_bh;
2009                 (*tc)->tc_ext_alloc_locked = 1;
2010         }
2011
2012         status = 0;
2013 bail:
2014         if (status < 0) {
2015                 if (*tc)
2016                         ocfs2_free_truncate_context(*tc);
2017                 *tc = NULL;
2018         }
2019         mlog_exit_void();
2020         return status;
2021 }
2022
2023 static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc)
2024 {
2025         if (tc->tc_ext_alloc_inode) {
2026                 if (tc->tc_ext_alloc_locked)
2027                         ocfs2_meta_unlock(tc->tc_ext_alloc_inode, 1);
2028
2029                 mutex_unlock(&tc->tc_ext_alloc_inode->i_mutex);
2030                 iput(tc->tc_ext_alloc_inode);
2031         }
2032
2033         if (tc->tc_ext_alloc_bh)
2034                 brelse(tc->tc_ext_alloc_bh);
2035
2036         if (tc->tc_last_eb_bh)
2037                 brelse(tc->tc_last_eb_bh);
2038
2039         kfree(tc);
2040 }