fs/ocfs2/file.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * file.c
   5  *
   6  * File open, close, extend, truncate
   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/capability.h>
  27 #include <linux/fs.h>
  28 #include <linux/types.h>
  29 #include <linux/slab.h>
  30 #include <linux/highmem.h>
  31 #include <linux/pagemap.h>
  32 #include <linux/uio.h>
  33 #include <linux/sched.h>
  34 #include <linux/splice.h>
  35 #include <linux/mount.h>
  36 #include <linux/writeback.h>
  37 #include <linux/falloc.h>
  38
  39 #define MLOG_MASK_PREFIX ML_INODE
  40 #include <cluster/masklog.h>
  41
  42 #include "ocfs2.h"
  43
  44 #include "alloc.h"
  45 #include "aops.h"
  46 #include "dir.h"
  47 #include "dlmglue.h"
  48 #include "extent_map.h"
  49 #include "file.h"
  50 #include "sysfile.h"
  51 #include "inode.h"
  52 #include "ioctl.h"
  53 #include "journal.h"
  54 #include "mmap.h"
  55 #include "suballoc.h"
  56 #include "super.h"
  57
  58 #include "buffer_head_io.h"
  59
  60 static int ocfs2_sync_inode(struct inode *inode)
  61 {
  62         filemap_fdatawrite(inode->i_mapping);
  63         return sync_mapping_buffers(inode->i_mapping);
  64 }
  65
  66 static int ocfs2_file_open(struct inode *inode, struct file *file)
  67 {
  68         int status;
  69         int mode = file->f_flags;
  70         struct ocfs2_inode_info *oi = OCFS2_I(inode);
  71
  72         mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
  73                    file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name);
  74
  75         spin_lock(&oi->ip_lock);
  76
  77         /* Check that the inode hasn't been wiped from disk by another
  78          * node. If it hasn't then we're safe as long as we hold the
  79          * spin lock until our increment of open count. */
  80         if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
  81                 spin_unlock(&oi->ip_lock);
  82
  83                 status = -ENOENT;
  84                 goto leave;
  85         }
  86
  87         if (mode & O_DIRECT)
  88                 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
  89
  90         oi->ip_open_count++;
  91         spin_unlock(&oi->ip_lock);
  92         status = 0;
  93 leave:
  94         mlog_exit(status);
  95         return status;
  96 }
  97
  98 static int ocfs2_file_release(struct inode *inode, struct file *file)
  99 {
 100         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 101
 102         mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
 103                        file->f_path.dentry->d_name.len,
 104                        file->f_path.dentry->d_name.name);
 105
 106         spin_lock(&oi->ip_lock);
 107         if (!--oi->ip_open_count)
 108                 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
 109         spin_unlock(&oi->ip_lock);
 110
 111         mlog_exit(0);
 112
 113         return 0;
 114 }
 115
 116 static int ocfs2_sync_file(struct file *file,
 117                            struct dentry *dentry,
 118                            int datasync)
 119 {
 120         int err = 0;
 121         journal_t *journal;
 122         struct inode *inode = dentry->d_inode;
 123         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 124
 125         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync,
 126                    dentry->d_name.len, dentry->d_name.name);
 127
 128         err = ocfs2_sync_inode(dentry->d_inode);
 129         if (err)
 130                 goto bail;
 131
 132         journal = osb->journal->j_journal;
 133         err = journal_force_commit(journal);
 134
 135 bail:
 136         mlog_exit(err);
 137
 138         return (err < 0) ? -EIO : 0;
 139 }
 140
 141 int ocfs2_should_update_atime(struct inode *inode,
 142                               struct vfsmount *vfsmnt)
 143 {
 144         struct timespec now;
 145         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 146
 147         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 148                 return 0;
 149
 150         if ((inode->i_flags & S_NOATIME) ||
 151             ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
 152                 return 0;
 153
 154         /*
 155          * We can be called with no vfsmnt structure - NFSD will
 156          * sometimes do this.
 157          *
 158          * Note that our action here is different than touch_atime() -
 159          * if we can't tell whether this is a noatime mount, then we
 160          * don't know whether to trust the value of s_atime_quantum.
 161          */
 162         if (vfsmnt == NULL)
 163                 return 0;
 164
 165         if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
 166             ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
 167                 return 0;
 168
 169         if (vfsmnt->mnt_flags & MNT_RELATIME) {
 170                 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
 171                     (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
 172                         return 1;
 173
 174                 return 0;
 175         }
 176
 177         now = CURRENT_TIME;
 178         if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
 179                 return 0;
 180         else
 181                 return 1;
 182 }
 183
 184 int ocfs2_update_inode_atime(struct inode *inode,
 185                              struct buffer_head *bh)
 186 {
 187         int ret;
 188         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 189         handle_t *handle;
 190         struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
 191
 192         mlog_entry_void();
 193
 194         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 195         if (handle == NULL) {
 196                 ret = -ENOMEM;
 197                 mlog_errno(ret);
 198                 goto out;
 199         }
 200
 201         ret = ocfs2_journal_access(handle, inode, bh,
 202                                    OCFS2_JOURNAL_ACCESS_WRITE);
 203         if (ret) {
 204                 mlog_errno(ret);
 205                 goto out_commit;
 206         }
 207
 208         /*
 209          * Don't use ocfs2_mark_inode_dirty() here as we don't always
 210          * have i_mutex to guard against concurrent changes to other
 211          * inode fields.
 212          */
 213         inode->i_atime = CURRENT_TIME;
 214         di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
 215         di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
 216
 217         ret = ocfs2_journal_dirty(handle, bh);
 218         if (ret < 0)
 219                 mlog_errno(ret);
 220
 221 out_commit:
 222         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 223 out:
 224         mlog_exit(ret);
 225         return ret;
 226 }
 227
 228 static int ocfs2_set_inode_size(handle_t *handle,
 229                                 struct inode *inode,
 230                                 struct buffer_head *fe_bh,
 231                                 u64 new_i_size)
 232 {
 233         int status;
 234
 235         mlog_entry_void();
 236         i_size_write(inode, new_i_size);
 237         inode->i_blocks = ocfs2_inode_sector_count(inode);
 238         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 239
 240         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 241         if (status < 0) {
 242                 mlog_errno(status);
 243                 goto bail;
 244         }
 245
 246 bail:
 247         mlog_exit(status);
 248         return status;
 249 }
 250
 251 static int ocfs2_simple_size_update(struct inode *inode,
 252                                     struct buffer_head *di_bh,
 253                                     u64 new_i_size)
 254 {
 255         int ret;
 256         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 257         handle_t *handle = NULL;
 258
 259         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 260         if (handle == NULL) {
 261                 ret = -ENOMEM;
 262                 mlog_errno(ret);
 263                 goto out;
 264         }
 265
 266         ret = ocfs2_set_inode_size(handle, inode, di_bh,
 267                                    new_i_size);
 268         if (ret < 0)
 269                 mlog_errno(ret);
 270
 271         ocfs2_commit_trans(osb, handle);
 272 out:
 273         return ret;
 274 }
 275
 276 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 277                                      struct inode *inode,
 278                                      struct buffer_head *fe_bh,
 279                                      u64 new_i_size)
 280 {
 281         int status;
 282         handle_t *handle;
 283         struct ocfs2_dinode *di;
 284         u64 cluster_bytes;
 285
 286         mlog_entry_void();
 287
 288         /* TODO: This needs to actually orphan the inode in this
 289          * transaction. */
 290
 291         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 292         if (IS_ERR(handle)) {
 293                 status = PTR_ERR(handle);
 294                 mlog_errno(status);
 295                 goto out;
 296         }
 297
 298         status = ocfs2_journal_access(handle, inode, fe_bh,
 299                                       OCFS2_JOURNAL_ACCESS_WRITE);
 300         if (status < 0) {
 301                 mlog_errno(status);
 302                 goto out_commit;
 303         }
 304
 305         /*
 306          * Do this before setting i_size.
 307          */
 308         cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
 309         status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
 310                                                cluster_bytes);
 311         if (status) {
 312                 mlog_errno(status);
 313                 goto out_commit;
 314         }
 315
 316         i_size_write(inode, new_i_size);
 317         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 318
 319         di = (struct ocfs2_dinode *) fe_bh->b_data;
 320         di->i_size = cpu_to_le64(new_i_size);
 321         di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
 322         di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 323
 324         status = ocfs2_journal_dirty(handle, fe_bh);
 325         if (status < 0)
 326                 mlog_errno(status);
 327
 328 out_commit:
 329         ocfs2_commit_trans(osb, handle);
 330 out:
 331
 332         mlog_exit(status);
 333         return status;
 334 }
 335
 336 static int ocfs2_truncate_file(struct inode *inode,
 337                                struct buffer_head *di_bh,
 338                                u64 new_i_size)
 339 {
 340         int status = 0;
 341         struct ocfs2_dinode *fe = NULL;
 342         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 343         struct ocfs2_truncate_context *tc = NULL;
 344
 345         mlog_entry("(inode = %llu, new_i_size = %llu\n",
 346                    (unsigned long long)OCFS2_I(inode)->ip_blkno,
 347                    (unsigned long long)new_i_size);
 348
 349         fe = (struct ocfs2_dinode *) di_bh->b_data;
 350         if (!OCFS2_IS_VALID_DINODE(fe)) {
 351                 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
 352                 status = -EIO;
 353                 goto bail;
 354         }
 355
 356         mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
 357                         "Inode %llu, inode i_size = %lld != di "
 358                         "i_size = %llu, i_flags = 0x%x\n",
 359                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 360                         i_size_read(inode),
 361                         (unsigned long long)le64_to_cpu(fe->i_size),
 362                         le32_to_cpu(fe->i_flags));
 363
 364         if (new_i_size > le64_to_cpu(fe->i_size)) {
 365                 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
 366                      (unsigned long long)le64_to_cpu(fe->i_size),
 367                      (unsigned long long)new_i_size);
 368                 status = -EINVAL;
 369                 mlog_errno(status);
 370                 goto bail;
 371         }
 372
 373         mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
 374              (unsigned long long)le64_to_cpu(fe->i_blkno),
 375              (unsigned long long)le64_to_cpu(fe->i_size),
 376              (unsigned long long)new_i_size);
 377
 378         /* lets handle the simple truncate cases before doing any more
 379          * cluster locking. */
 380         if (new_i_size == le64_to_cpu(fe->i_size))
 381                 goto bail;
 382
 383         down_write(&OCFS2_I(inode)->ip_alloc_sem);
 384
 385         /* This forces other nodes to sync and drop their pages. Do
 386          * this even if we have a truncate without allocation change -
 387          * ocfs2 cluster sizes can be much greater than page size, so
 388          * we have to truncate them anyway.  */
 389         status = ocfs2_data_lock(inode, 1);
 390         if (status < 0) {
 391                 up_write(&OCFS2_I(inode)->ip_alloc_sem);
 392
 393                 mlog_errno(status);
 394                 goto bail;
 395         }
 396
 397         unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
 398         truncate_inode_pages(inode->i_mapping, new_i_size);
 399
 400         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 401                 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
 402                                                i_size_read(inode), 0);
 403                 if (status)
 404                         mlog_errno(status);
 405
 406                 goto bail_unlock_data;
 407         }
 408
 409         /* alright, we're going to need to do a full blown alloc size
 410          * change. Orphan the inode so that recovery can complete the
 411          * truncate if necessary. This does the task of marking
 412          * i_size. */
 413         status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
 414         if (status < 0) {
 415                 mlog_errno(status);
 416                 goto bail_unlock_data;
 417         }
 418
 419         status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
 420         if (status < 0) {
 421                 mlog_errno(status);
 422                 goto bail_unlock_data;
 423         }
 424
 425         status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
 426         if (status < 0) {
 427                 mlog_errno(status);
 428                 goto bail_unlock_data;
 429         }
 430
 431         /* TODO: orphan dir cleanup here. */
 432 bail_unlock_data:
 433         ocfs2_data_unlock(inode, 1);
 434
 435         up_write(&OCFS2_I(inode)->ip_alloc_sem);
 436
 437 bail:
 438
 439         mlog_exit(status);
 440         return status;
 441 }
 442
 443 /*
 444  * extend allocation only here.
 445  * we'll update all the disk stuff, and oip->alloc_size
 446  *
 447  * expect stuff to be locked, a transaction started and enough data /
 448  * metadata reservations in the contexts.
 449  *
 450  * Will return -EAGAIN, and a reason if a restart is needed.
 451  * If passed in, *reason will always be set, even in error.
 452  */
 453 int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
 454                                struct inode *inode,
 455                                u32 *logical_offset,
 456                                u32 clusters_to_add,
 457                                int mark_unwritten,
 458                                struct buffer_head *fe_bh,
 459                                handle_t *handle,
 460                                struct ocfs2_alloc_context *data_ac,
 461                                struct ocfs2_alloc_context *meta_ac,
 462                                enum ocfs2_alloc_restarted *reason_ret)
 463 {
 464         int status = 0;
 465         int free_extents;
 466         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
 467         enum ocfs2_alloc_restarted reason = RESTART_NONE;
 468         u32 bit_off, num_bits;
 469         u64 block;
 470         u8 flags = 0;
 471
 472         BUG_ON(!clusters_to_add);
 473
 474         if (mark_unwritten)
 475                 flags = OCFS2_EXT_UNWRITTEN;
 476
 477         free_extents = ocfs2_num_free_extents(osb, inode, fe);
 478         if (free_extents < 0) {
 479                 status = free_extents;
 480                 mlog_errno(status);
 481                 goto leave;
 482         }
 483
 484         /* there are two cases which could cause us to EAGAIN in the
 485          * we-need-more-metadata case:
 486          * 1) we haven't reserved *any*
 487          * 2) we are so fragmented, we've needed to add metadata too
 488          *    many times. */
 489         if (!free_extents && !meta_ac) {
 490                 mlog(0, "we haven't reserved any metadata!\n");
 491                 status = -EAGAIN;
 492                 reason = RESTART_META;
 493                 goto leave;
 494         } else if ((!free_extents)
 495                    && (ocfs2_alloc_context_bits_left(meta_ac)
 496                        < ocfs2_extend_meta_needed(fe))) {
 497                 mlog(0, "filesystem is really fragmented...\n");
 498                 status = -EAGAIN;
 499                 reason = RESTART_META;
 500                 goto leave;
 501         }
 502
 503         status = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
 504                                         clusters_to_add, &bit_off, &num_bits);
 505         if (status < 0) {
 506                 if (status != -ENOSPC)
 507                         mlog_errno(status);
 508                 goto leave;
 509         }
 510
 511         BUG_ON(num_bits > clusters_to_add);
 512
 513         /* reserve our write early -- insert_extent may update the inode */
 514         status = ocfs2_journal_access(handle, inode, fe_bh,
 515                                       OCFS2_JOURNAL_ACCESS_WRITE);
 516         if (status < 0) {
 517                 mlog_errno(status);
 518                 goto leave;
 519         }
 520
 521         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
 522         mlog(0, "Allocating %u clusters at block %u for inode %llu\n",
 523              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
 524         status = ocfs2_insert_extent(osb, handle, inode, fe_bh,
 525                                      *logical_offset, block, num_bits,
 526                                      flags, meta_ac);
 527         if (status < 0) {
 528                 mlog_errno(status);
 529                 goto leave;
 530         }
 531
 532         status = ocfs2_journal_dirty(handle, fe_bh);
 533         if (status < 0) {
 534                 mlog_errno(status);
 535                 goto leave;
 536         }
 537
 538         clusters_to_add -= num_bits;
 539         *logical_offset += num_bits;
 540
 541         if (clusters_to_add) {
 542                 mlog(0, "need to alloc once more, clusters = %u, wanted = "
 543                      "%u\n", fe->i_clusters, clusters_to_add);
 544                 status = -EAGAIN;
 545                 reason = RESTART_TRANS;
 546         }
 547
 548 leave:
 549         mlog_exit(status);
 550         if (reason_ret)
 551                 *reason_ret = reason;
 552         return status;
 553 }
 554
 555 /*
 556  * For a given allocation, determine which allocators will need to be
 557  * accessed, and lock them, reserving the appropriate number of bits.
 558  *
 559  * Sparse file systems call this from ocfs2_write_begin_nolock()
 560  * and ocfs2_allocate_unwritten_extents().
 561  *
 562  * File systems which don't support holes call this from
 563  * ocfs2_extend_allocation().
 564  */
 565 int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_dinode *di,
 566                           u32 clusters_to_add, u32 extents_to_split,
 567                           struct ocfs2_alloc_context **data_ac,
 568                           struct ocfs2_alloc_context **meta_ac)
 569 {
 570         int ret = 0, num_free_extents;
 571         unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
 572         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 573
 574         *meta_ac = NULL;
 575         if (data_ac)
 576                 *data_ac = NULL;
 577
 578         BUG_ON(clusters_to_add != 0 && data_ac == NULL);
 579
 580         mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
 581              "clusters_to_add = %u, extents_to_split = %u\n",
 582              (unsigned long long)OCFS2_I(inode)->ip_blkno, i_size_read(inode),
 583              le32_to_cpu(di->i_clusters), clusters_to_add, extents_to_split);
 584
 585         num_free_extents = ocfs2_num_free_extents(osb, inode, di);
 586         if (num_free_extents < 0) {
 587                 ret = num_free_extents;
 588                 mlog_errno(ret);
 589                 goto out;
 590         }
 591
 592         /*
 593          * Sparse allocation file systems need to be more conservative
 594          * with reserving room for expansion - the actual allocation
 595          * happens while we've got a journal handle open so re-taking
 596          * a cluster lock (because we ran out of room for another
 597          * extent) will violate ordering rules.
 598          *
 599          * Most of the time we'll only be seeing this 1 cluster at a time
 600          * anyway.
 601          *
 602          * Always lock for any unwritten extents - we might want to
 603          * add blocks during a split.
 604          */
 605         if (!num_free_extents ||
 606             (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
 607                 ret = ocfs2_reserve_new_metadata(osb, di, meta_ac);
 608                 if (ret < 0) {
 609                         if (ret != -ENOSPC)
 610                                 mlog_errno(ret);
 611                         goto out;
 612                 }
 613         }
 614
 615         if (clusters_to_add == 0)
 616                 goto out;
 617
 618         ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
 619         if (ret < 0) {
 620                 if (ret != -ENOSPC)
 621                         mlog_errno(ret);
 622                 goto out;
 623         }
 624
 625 out:
 626         if (ret) {
 627                 if (*meta_ac) {
 628                         ocfs2_free_alloc_context(*meta_ac);
 629                         *meta_ac = NULL;
 630                 }
 631
 632                 /*
 633                  * We cannot have an error and a non null *data_ac.
 634                  */
 635         }
 636
 637         return ret;
 638 }
 639
 640 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
 641                                      u32 clusters_to_add, int mark_unwritten)
 642 {
 643         int status = 0;
 644         int restart_func = 0;
 645         int credits;
 646         u32 prev_clusters;
 647         struct buffer_head *bh = NULL;
 648         struct ocfs2_dinode *fe = NULL;
 649         handle_t *handle = NULL;
 650         struct ocfs2_alloc_context *data_ac = NULL;
 651         struct ocfs2_alloc_context *meta_ac = NULL;
 652         enum ocfs2_alloc_restarted why;
 653         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 654
 655         mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
 656
 657         /*
 658          * This function only exists for file systems which don't
 659          * support holes.
 660          */
 661         BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
 662
 663         status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh,
 664                                   OCFS2_BH_CACHED, inode);
 665         if (status < 0) {
 666                 mlog_errno(status);
 667                 goto leave;
 668         }
 669
 670         fe = (struct ocfs2_dinode *) bh->b_data;
 671         if (!OCFS2_IS_VALID_DINODE(fe)) {
 672                 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
 673                 status = -EIO;
 674                 goto leave;
 675         }
 676
 677 restart_all:
 678         BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
 679
 680         status = ocfs2_lock_allocators(inode, fe, clusters_to_add, 0, &data_ac,
 681                                        &meta_ac);
 682         if (status) {
 683                 mlog_errno(status);
 684                 goto leave;
 685         }
 686
 687         credits = ocfs2_calc_extend_credits(osb->sb, fe, clusters_to_add);
 688         handle = ocfs2_start_trans(osb, credits);
 689         if (IS_ERR(handle)) {
 690                 status = PTR_ERR(handle);
 691                 handle = NULL;
 692                 mlog_errno(status);
 693                 goto leave;
 694         }
 695
 696 restarted_transaction:
 697         /* reserve a write to the file entry early on - that we if we
 698          * run out of credits in the allocation path, we can still
 699          * update i_size. */
 700         status = ocfs2_journal_access(handle, inode, bh,
 701                                       OCFS2_JOURNAL_ACCESS_WRITE);
 702         if (status < 0) {
 703                 mlog_errno(status);
 704                 goto leave;
 705         }
 706
 707         prev_clusters = OCFS2_I(inode)->ip_clusters;
 708
 709         status = ocfs2_do_extend_allocation(osb,
 710                                             inode,
 711                                             &logical_start,
 712                                             clusters_to_add,
 713                                             mark_unwritten,
 714                                             bh,
 715                                             handle,
 716                                             data_ac,
 717                                             meta_ac,
 718                                             &why);
 719         if ((status < 0) && (status != -EAGAIN)) {
 720                 if (status != -ENOSPC)
 721                         mlog_errno(status);
 722                 goto leave;
 723         }
 724
 725         status = ocfs2_journal_dirty(handle, bh);
 726         if (status < 0) {
 727                 mlog_errno(status);
 728                 goto leave;
 729         }
 730
 731         spin_lock(&OCFS2_I(inode)->ip_lock);
 732         clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
 733         spin_unlock(&OCFS2_I(inode)->ip_lock);
 734
 735         if (why != RESTART_NONE && clusters_to_add) {
 736                 if (why == RESTART_META) {
 737                         mlog(0, "restarting function.\n");
 738                         restart_func = 1;
 739                 } else {
 740                         BUG_ON(why != RESTART_TRANS);
 741
 742                         mlog(0, "restarting transaction.\n");
 743                         /* TODO: This can be more intelligent. */
 744                         credits = ocfs2_calc_extend_credits(osb->sb,
 745                                                             fe,
 746                                                             clusters_to_add);
 747                         status = ocfs2_extend_trans(handle, credits);
 748                         if (status < 0) {
 749                                 /* handle still has to be committed at
 750                                  * this point. */
 751                                 status = -ENOMEM;
 752                                 mlog_errno(status);
 753                                 goto leave;
 754                         }
 755                         goto restarted_transaction;
 756                 }
 757         }
 758
 759         mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
 760              le32_to_cpu(fe->i_clusters),
 761              (unsigned long long)le64_to_cpu(fe->i_size));
 762         mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
 763              OCFS2_I(inode)->ip_clusters, i_size_read(inode));
 764
 765 leave:
 766         if (handle) {
 767                 ocfs2_commit_trans(osb, handle);
 768                 handle = NULL;
 769         }
 770         if (data_ac) {
 771                 ocfs2_free_alloc_context(data_ac);
 772                 data_ac = NULL;
 773         }
 774         if (meta_ac) {
 775                 ocfs2_free_alloc_context(meta_ac);
 776                 meta_ac = NULL;
 777         }
 778         if ((!status) && restart_func) {
 779                 restart_func = 0;
 780                 goto restart_all;
 781         }
 782         if (bh) {
 783                 brelse(bh);
 784                 bh = NULL;
 785         }
 786
 787         mlog_exit(status);
 788         return status;
 789 }
 790
 791 /* Some parts of this taken from generic_cont_expand, which turned out
 792  * to be too fragile to do exactly what we need without us having to
 793  * worry about recursive locking in ->prepare_write() and
 794  * ->commit_write(). */
 795 static int ocfs2_write_zero_page(struct inode *inode,
 796                                  u64 size)
 797 {
 798         struct address_space *mapping = inode->i_mapping;
 799         struct page *page;
 800         unsigned long index;
 801         unsigned int offset;
 802         handle_t *handle = NULL;
 803         int ret;
 804
 805         offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */
 806         /* ugh.  in prepare/commit_write, if from==to==start of block, we
 807         ** skip the prepare.  make sure we never send an offset for the start
 808         ** of a block
 809         */
 810         if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) {
 811                 offset++;
 812         }
 813         index = size >> PAGE_CACHE_SHIFT;
 814
 815         page = grab_cache_page(mapping, index);
 816         if (!page) {
 817                 ret = -ENOMEM;
 818                 mlog_errno(ret);
 819                 goto out;
 820         }
 821
 822         ret = ocfs2_prepare_write_nolock(inode, page, offset, offset);
 823         if (ret < 0) {
 824                 mlog_errno(ret);
 825                 goto out_unlock;
 826         }
 827
 828         if (ocfs2_should_order_data(inode)) {
 829                 handle = ocfs2_start_walk_page_trans(inode, page, offset,
 830                                                      offset);
 831                 if (IS_ERR(handle)) {
 832                         ret = PTR_ERR(handle);
 833                         handle = NULL;
 834                         goto out_unlock;
 835                 }
 836         }
 837
 838         /* must not update i_size! */
 839         ret = block_commit_write(page, offset, offset);
 840         if (ret < 0)
 841                 mlog_errno(ret);
 842         else
 843                 ret = 0;
 844
 845         if (handle)
 846                 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 847 out_unlock:
 848         unlock_page(page);
 849         page_cache_release(page);
 850 out:
 851         return ret;
 852 }
 853
 854 static int ocfs2_zero_extend(struct inode *inode,
 855                              u64 zero_to_size)
 856 {
 857         int ret = 0;
 858         u64 start_off;
 859         struct super_block *sb = inode->i_sb;
 860
 861         start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
 862         while (start_off < zero_to_size) {
 863                 ret = ocfs2_write_zero_page(inode, start_off);
 864                 if (ret < 0) {
 865                         mlog_errno(ret);
 866                         goto out;
 867                 }
 868
 869                 start_off += sb->s_blocksize;
 870
 871                 /*
 872                  * Very large extends have the potential to lock up
 873                  * the cpu for extended periods of time.
 874                  */
 875                 cond_resched();
 876         }
 877
 878 out:
 879         return ret;
 880 }
 881
 882 int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size, u64 zero_to)
 883 {
 884         int ret;
 885         u32 clusters_to_add;
 886         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 887
 888         clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
 889         if (clusters_to_add < oi->ip_clusters)
 890                 clusters_to_add = 0;
 891         else
 892                 clusters_to_add -= oi->ip_clusters;
 893
 894         if (clusters_to_add) {
 895                 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
 896                                                 clusters_to_add, 0);
 897                 if (ret) {
 898                         mlog_errno(ret);
 899                         goto out;
 900                 }
 901         }
 902
 903         /*
 904          * Call this even if we don't add any clusters to the tree. We
 905          * still need to zero the area between the old i_size and the
 906          * new i_size.
 907          */
 908         ret = ocfs2_zero_extend(inode, zero_to);
 909         if (ret < 0)
 910                 mlog_errno(ret);
 911
 912 out:
 913         return ret;
 914 }
 915
 916 static int ocfs2_extend_file(struct inode *inode,
 917                              struct buffer_head *di_bh,
 918                              u64 new_i_size)
 919 {
 920         int ret = 0, data_locked = 0;
 921         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 922
 923         BUG_ON(!di_bh);
 924
 925         /* setattr sometimes calls us like this. */
 926         if (new_i_size == 0)
 927                 goto out;
 928
 929         if (i_size_read(inode) == new_i_size)
 930                 goto out;
 931         BUG_ON(new_i_size < i_size_read(inode));
 932
 933         /*
 934          * Fall through for converting inline data, even if the fs
 935          * supports sparse files.
 936          *
 937          * The check for inline data here is legal - nobody can add
 938          * the feature since we have i_mutex. We must check it again
 939          * after acquiring ip_alloc_sem though, as paths like mmap
 940          * might have raced us to converting the inode to extents.
 941          */
 942         if (!(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
 943             && ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
 944                 goto out_update_size;
 945
 946         /*
 947          * protect the pages that ocfs2_zero_extend is going to be
 948          * pulling into the page cache.. we do this before the
 949          * metadata extend so that we don't get into the situation
 950          * where we've extended the metadata but can't get the data
 951          * lock to zero.
 952          */
 953         ret = ocfs2_data_lock(inode, 1);
 954         if (ret < 0) {
 955                 mlog_errno(ret);
 956                 goto out;
 957         }
 958         data_locked = 1;
 959
 960         /*
 961          * The alloc sem blocks people in read/write from reading our
 962          * allocation until we're done changing it. We depend on
 963          * i_mutex to block other extend/truncate calls while we're
 964          * here.
 965          */
 966         down_write(&oi->ip_alloc_sem);
 967
 968         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 969                 /*
 970                  * We can optimize small extends by keeping the inodes
 971                  * inline data.
 972                  */
 973                 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
 974                         up_write(&oi->ip_alloc_sem);
 975                         goto out_update_size;
 976                 }
 977
 978                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
 979                 if (ret) {
 980                         up_write(&oi->ip_alloc_sem);
 981
 982                         mlog_errno(ret);
 983                         goto out_unlock;
 984                 }
 985         }
 986
 987         if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
 988                 ret = ocfs2_extend_no_holes(inode, new_i_size, new_i_size);
 989
 990         up_write(&oi->ip_alloc_sem);
 991
 992         if (ret < 0) {
 993                 mlog_errno(ret);
 994                 goto out_unlock;
 995         }
 996
 997 out_update_size:
 998         ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
 999         if (ret < 0)
1000                 mlog_errno(ret);
1001
1002 out_unlock:
1003         if (data_locked)
1004                 ocfs2_data_unlock(inode, 1);
1005
1006 out:
1007         return ret;
1008 }
1009
1010 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1011 {
1012         int status = 0, size_change;
1013         struct inode *inode = dentry->d_inode;
1014         struct super_block *sb = inode->i_sb;
1015         struct ocfs2_super *osb = OCFS2_SB(sb);
1016         struct buffer_head *bh = NULL;
1017         handle_t *handle = NULL;
1018
1019         mlog_entry("(0x%p, '%.*s')\n", dentry,
1020                    dentry->d_name.len, dentry->d_name.name);
1021
1022         if (attr->ia_valid & ATTR_MODE)
1023                 mlog(0, "mode change: %d\n", attr->ia_mode);
1024         if (attr->ia_valid & ATTR_UID)
1025                 mlog(0, "uid change: %d\n", attr->ia_uid);
1026         if (attr->ia_valid & ATTR_GID)
1027                 mlog(0, "gid change: %d\n", attr->ia_gid);
1028         if (attr->ia_valid & ATTR_SIZE)
1029                 mlog(0, "size change...\n");
1030         if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
1031                 mlog(0, "time change...\n");
1032
1033 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1034                            | ATTR_GID | ATTR_UID | ATTR_MODE)
1035         if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
1036                 mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
1037                 return 0;
1038         }
1039
1040         status = inode_change_ok(inode, attr);
1041         if (status)
1042                 return status;
1043
1044         size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1045         if (size_change) {
1046                 status = ocfs2_rw_lock(inode, 1);
1047                 if (status < 0) {
1048                         mlog_errno(status);
1049                         goto bail;
1050                 }
1051         }
1052
1053         status = ocfs2_meta_lock(inode, &bh, 1);
1054         if (status < 0) {
1055                 if (status != -ENOENT)
1056                         mlog_errno(status);
1057                 goto bail_unlock_rw;
1058         }
1059
1060         if (size_change && attr->ia_size != i_size_read(inode)) {
1061                 if (attr->ia_size > sb->s_maxbytes) {
1062                         status = -EFBIG;
1063                         goto bail_unlock;
1064                 }
1065
1066                 if (i_size_read(inode) > attr->ia_size)
1067                         status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1068                 else
1069                         status = ocfs2_extend_file(inode, bh, attr->ia_size);
1070                 if (status < 0) {
1071                         if (status != -ENOSPC)
1072                                 mlog_errno(status);
1073                         status = -ENOSPC;
1074                         goto bail_unlock;
1075                 }
1076         }
1077
1078         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1079         if (IS_ERR(handle)) {
1080                 status = PTR_ERR(handle);
1081                 mlog_errno(status);
1082                 goto bail_unlock;
1083         }
1084
1085         /*
1086          * This will intentionally not wind up calling vmtruncate(),
1087          * since all the work for a size change has been done above.
1088          * Otherwise, we could get into problems with truncate as
1089          * ip_alloc_sem is used there to protect against i_size
1090          * changes.
1091          */
1092         status = inode_setattr(inode, attr);
1093         if (status < 0) {
1094                 mlog_errno(status);
1095                 goto bail_commit;
1096         }
1097
1098         status = ocfs2_mark_inode_dirty(handle, inode, bh);
1099         if (status < 0)
1100                 mlog_errno(status);
1101
1102 bail_commit:
1103         ocfs2_commit_trans(osb, handle);
1104 bail_unlock:
1105         ocfs2_meta_unlock(inode, 1);
1106 bail_unlock_rw:
1107         if (size_change)
1108                 ocfs2_rw_unlock(inode, 1);
1109 bail:
1110         if (bh)
1111                 brelse(bh);
1112
1113         mlog_exit(status);
1114         return status;
1115 }
1116
1117 int ocfs2_getattr(struct vfsmount *mnt,
1118                   struct dentry *dentry,
1119                   struct kstat *stat)
1120 {
1121         struct inode *inode = dentry->d_inode;
1122         struct super_block *sb = dentry->d_inode->i_sb;
1123         struct ocfs2_super *osb = sb->s_fs_info;
1124         int err;
1125
1126         mlog_entry_void();
1127
1128         err = ocfs2_inode_revalidate(dentry);
1129         if (err) {
1130                 if (err != -ENOENT)
1131                         mlog_errno(err);
1132                 goto bail;
1133         }
1134
1135         generic_fillattr(inode, stat);
1136
1137         /* We set the blksize from the cluster size for performance */
1138         stat->blksize = osb->s_clustersize;
1139
1140 bail:
1141         mlog_exit(err);
1142
1143         return err;
1144 }
1145
1146 int ocfs2_permission(struct inode *inode, int mask, struct nameidata *nd)
1147 {
1148         int ret;
1149
1150         mlog_entry_void();
1151
1152         ret = ocfs2_meta_lock(inode, NULL, 0);
1153         if (ret) {
1154                 if (ret != -ENOENT)
1155                         mlog_errno(ret);
1156                 goto out;
1157         }
1158
1159         ret = generic_permission(inode, mask, NULL);
1160
1161         ocfs2_meta_unlock(inode, 0);
1162 out:
1163         mlog_exit(ret);
1164         return ret;
1165 }
1166
1167 static int __ocfs2_write_remove_suid(struct inode *inode,
1168                                      struct buffer_head *bh)
1169 {
1170         int ret;
1171         handle_t *handle;
1172         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1173         struct ocfs2_dinode *di;
1174
1175         mlog_entry("(Inode %llu, mode 0%o)\n",
1176                    (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode);
1177
1178         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1179         if (handle == NULL) {
1180                 ret = -ENOMEM;
1181                 mlog_errno(ret);
1182                 goto out;
1183         }
1184
1185         ret = ocfs2_journal_access(handle, inode, bh,
1186                                    OCFS2_JOURNAL_ACCESS_WRITE);
1187         if (ret < 0) {
1188                 mlog_errno(ret);
1189                 goto out_trans;
1190         }
1191
1192         inode->i_mode &= ~S_ISUID;
1193         if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1194                 inode->i_mode &= ~S_ISGID;
1195
1196         di = (struct ocfs2_dinode *) bh->b_data;
1197         di->i_mode = cpu_to_le16(inode->i_mode);
1198
1199         ret = ocfs2_journal_dirty(handle, bh);
1200         if (ret < 0)
1201                 mlog_errno(ret);
1202
1203 out_trans:
1204         ocfs2_commit_trans(osb, handle);
1205 out:
1206         mlog_exit(ret);
1207         return ret;
1208 }
1209
1210 /*
1211  * Will look for holes and unwritten extents in the range starting at
1212  * pos for count bytes (inclusive).
1213  */
1214 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1215                                        size_t count)
1216 {
1217         int ret = 0;
1218         unsigned int extent_flags;
1219         u32 cpos, clusters, extent_len, phys_cpos;
1220         struct super_block *sb = inode->i_sb;
1221
1222         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1223         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1224
1225         while (clusters) {
1226                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1227                                          &extent_flags);
1228                 if (ret < 0) {
1229                         mlog_errno(ret);
1230                         goto out;
1231                 }
1232
1233                 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1234                         ret = 1;
1235                         break;
1236                 }
1237
1238                 if (extent_len > clusters)
1239                         extent_len = clusters;
1240
1241                 clusters -= extent_len;
1242                 cpos += extent_len;
1243         }
1244 out:
1245         return ret;
1246 }
1247
1248 static int ocfs2_write_remove_suid(struct inode *inode)
1249 {
1250         int ret;
1251         struct buffer_head *bh = NULL;
1252         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1253
1254         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1255                                oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode);
1256         if (ret < 0) {
1257                 mlog_errno(ret);
1258                 goto out;
1259         }
1260
1261         ret =  __ocfs2_write_remove_suid(inode, bh);
1262 out:
1263         brelse(bh);
1264         return ret;
1265 }
1266
1267 /*
1268  * Allocate enough extents to cover the region starting at byte offset
1269  * start for len bytes. Existing extents are skipped, any extents
1270  * added are marked as "unwritten".
1271  */
1272 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1273                                             u64 start, u64 len)
1274 {
1275         int ret;
1276         u32 cpos, phys_cpos, clusters, alloc_size;
1277         u64 end = start + len;
1278         struct buffer_head *di_bh = NULL;
1279
1280         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1281                 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1282                                        OCFS2_I(inode)->ip_blkno, &di_bh,
1283                                        OCFS2_BH_CACHED, inode);
1284                 if (ret) {
1285                         mlog_errno(ret);
1286                         goto out;
1287                 }
1288
1289                 /*
1290                  * Nothing to do if the requested reservation range
1291                  * fits within the inode.
1292                  */
1293                 if (ocfs2_size_fits_inline_data(di_bh, end))
1294                         goto out;
1295
1296                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1297                 if (ret) {
1298                         mlog_errno(ret);
1299                         goto out;
1300                 }
1301         }
1302
1303         /*
1304          * We consider both start and len to be inclusive.
1305          */
1306         cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1307         clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1308         clusters -= cpos;
1309
1310         while (clusters) {
1311                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1312                                          &alloc_size, NULL);
1313                 if (ret) {
1314                         mlog_errno(ret);
1315                         goto out;
1316                 }
1317
1318                 /*
1319                  * Hole or existing extent len can be arbitrary, so
1320                  * cap it to our own allocation request.
1321                  */
1322                 if (alloc_size > clusters)
1323                         alloc_size = clusters;
1324
1325                 if (phys_cpos) {
1326                         /*
1327                          * We already have an allocation at this
1328                          * region so we can safely skip it.
1329                          */
1330                         goto next;
1331                 }
1332
1333                 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1334                 if (ret) {
1335                         if (ret != -ENOSPC)
1336                                 mlog_errno(ret);
1337                         goto out;
1338                 }
1339
1340 next:
1341                 cpos += alloc_size;
1342                 clusters -= alloc_size;
1343         }
1344
1345         ret = 0;
1346 out:
1347
1348         brelse(di_bh);
1349         return ret;
1350 }
1351
1352 static int __ocfs2_remove_inode_range(struct inode *inode,
1353                                       struct buffer_head *di_bh,
1354                                       u32 cpos, u32 phys_cpos, u32 len,
1355                                       struct ocfs2_cached_dealloc_ctxt *dealloc)
1356 {
1357         int ret;
1358         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
1359         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1360         struct inode *tl_inode = osb->osb_tl_inode;
1361         handle_t *handle;
1362         struct ocfs2_alloc_context *meta_ac = NULL;
1363         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1364
1365         ret = ocfs2_lock_allocators(inode, di, 0, 1, NULL, &meta_ac);
1366         if (ret) {
1367                 mlog_errno(ret);
1368                 return ret;
1369         }
1370
1371         mutex_lock(&tl_inode->i_mutex);
1372
1373         if (ocfs2_truncate_log_needs_flush(osb)) {
1374                 ret = __ocfs2_flush_truncate_log(osb);
1375                 if (ret < 0) {
1376                         mlog_errno(ret);
1377                         goto out;
1378                 }
1379         }
1380
1381         handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
1382         if (handle == NULL) {
1383                 ret = -ENOMEM;
1384                 mlog_errno(ret);
1385                 goto out;
1386         }
1387
1388         ret = ocfs2_journal_access(handle, inode, di_bh,
1389                                    OCFS2_JOURNAL_ACCESS_WRITE);
1390         if (ret) {
1391                 mlog_errno(ret);
1392                 goto out;
1393         }
1394
1395         ret = ocfs2_remove_extent(inode, di_bh, cpos, len, handle, meta_ac,
1396                                   dealloc);
1397         if (ret) {
1398                 mlog_errno(ret);
1399                 goto out_commit;
1400         }
1401
1402         OCFS2_I(inode)->ip_clusters -= len;
1403         di->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1404
1405         ret = ocfs2_journal_dirty(handle, di_bh);
1406         if (ret) {
1407                 mlog_errno(ret);
1408                 goto out_commit;
1409         }
1410
1411         ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
1412         if (ret)
1413                 mlog_errno(ret);
1414
1415 out_commit:
1416         ocfs2_commit_trans(osb, handle);
1417 out:
1418         mutex_unlock(&tl_inode->i_mutex);
1419
1420         if (meta_ac)
1421                 ocfs2_free_alloc_context(meta_ac);
1422
1423         return ret;
1424 }
1425
1426 /*
1427  * Truncate a byte range, avoiding pages within partial clusters. This
1428  * preserves those pages for the zeroing code to write to.
1429  */
1430 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1431                                          u64 byte_len)
1432 {
1433         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1434         loff_t start, end;
1435         struct address_space *mapping = inode->i_mapping;
1436
1437         start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1438         end = byte_start + byte_len;
1439         end = end & ~(osb->s_clustersize - 1);
1440
1441         if (start < end) {
1442                 unmap_mapping_range(mapping, start, end - start, 0);
1443                 truncate_inode_pages_range(mapping, start, end - 1);
1444         }
1445 }
1446
1447 static int ocfs2_zero_partial_clusters(struct inode *inode,
1448                                        u64 start, u64 len)
1449 {
1450         int ret = 0;
1451         u64 tmpend, end = start + len;
1452         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1453         unsigned int csize = osb->s_clustersize;
1454         handle_t *handle;
1455
1456         /*
1457          * The "start" and "end" values are NOT necessarily part of
1458          * the range whose allocation is being deleted. Rather, this
1459          * is what the user passed in with the request. We must zero
1460          * partial clusters here. There's no need to worry about
1461          * physical allocation - the zeroing code knows to skip holes.
1462          */
1463         mlog(0, "byte start: %llu, end: %llu\n",
1464              (unsigned long long)start, (unsigned long long)end);
1465
1466         /*
1467          * If both edges are on a cluster boundary then there's no
1468          * zeroing required as the region is part of the allocation to
1469          * be truncated.
1470          */
1471         if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1472                 goto out;
1473
1474         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1475         if (handle == NULL) {
1476                 ret = -ENOMEM;
1477                 mlog_errno(ret);
1478                 goto out;
1479         }
1480
1481         /*
1482          * We want to get the byte offset of the end of the 1st cluster.
1483          */
1484         tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1485         if (tmpend > end)
1486                 tmpend = end;
1487
1488         mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1489              (unsigned long long)start, (unsigned long long)tmpend);
1490
1491         ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1492         if (ret)
1493                 mlog_errno(ret);
1494
1495         if (tmpend < end) {
1496                 /*
1497                  * This may make start and end equal, but the zeroing
1498                  * code will skip any work in that case so there's no
1499                  * need to catch it up here.
1500                  */
1501                 start = end & ~(osb->s_clustersize - 1);
1502
1503                 mlog(0, "2nd range: start: %llu, end: %llu\n",
1504                      (unsigned long long)start, (unsigned long long)end);
1505
1506                 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1507                 if (ret)
1508                         mlog_errno(ret);
1509         }
1510
1511         ocfs2_commit_trans(osb, handle);
1512 out:
1513         return ret;
1514 }
1515
1516 static int ocfs2_remove_inode_range(struct inode *inode,
1517                                     struct buffer_head *di_bh, u64 byte_start,
1518                                     u64 byte_len)
1519 {
1520         int ret = 0;
1521         u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size;
1522         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1523         struct ocfs2_cached_dealloc_ctxt dealloc;
1524
1525         ocfs2_init_dealloc_ctxt(&dealloc);
1526
1527         if (byte_len == 0)
1528                 return 0;
1529
1530         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1531                 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1532                                             byte_start + byte_len, 1);
1533                 if (ret)
1534                         mlog_errno(ret);
1535                 return ret;
1536         }
1537
1538         trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1539         trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits;
1540         if (trunc_len >= trunc_start)
1541                 trunc_len -= trunc_start;
1542         else
1543                 trunc_len = 0;
1544
1545         mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n",
1546              (unsigned long long)OCFS2_I(inode)->ip_blkno,
1547              (unsigned long long)byte_start,
1548              (unsigned long long)byte_len, trunc_start, trunc_len);
1549
1550         ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1551         if (ret) {
1552                 mlog_errno(ret);
1553                 goto out;
1554         }
1555
1556         cpos = trunc_start;
1557         while (trunc_len) {
1558                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1559                                          &alloc_size, NULL);
1560                 if (ret) {
1561                         mlog_errno(ret);
1562                         goto out;
1563                 }
1564
1565                 if (alloc_size > trunc_len)
1566                         alloc_size = trunc_len;
1567
1568                 /* Only do work for non-holes */
1569                 if (phys_cpos != 0) {
1570                         ret = __ocfs2_remove_inode_range(inode, di_bh, cpos,
1571                                                          phys_cpos, alloc_size,
1572                                                          &dealloc);
1573                         if (ret) {
1574                                 mlog_errno(ret);
1575                                 goto out;
1576                         }
1577                 }
1578
1579                 cpos += alloc_size;
1580                 trunc_len -= alloc_size;
1581         }
1582
1583         ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1584
1585 out:
1586         ocfs2_schedule_truncate_log_flush(osb, 1);
1587         ocfs2_run_deallocs(osb, &dealloc);
1588
1589         return ret;
1590 }
1591
1592 /*
1593  * Parts of this function taken from xfs_change_file_space()
1594  */
1595 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1596                                      loff_t f_pos, unsigned int cmd,
1597                                      struct ocfs2_space_resv *sr,
1598                                      int change_size)
1599 {
1600         int ret;
1601         s64 llen;
1602         loff_t size;
1603         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1604         struct buffer_head *di_bh = NULL;
1605         handle_t *handle;
1606         unsigned long long max_off = inode->i_sb->s_maxbytes;
1607
1608         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1609                 return -EROFS;
1610
1611         mutex_lock(&inode->i_mutex);
1612
1613         /*
1614          * This prevents concurrent writes on other nodes
1615          */
1616         ret = ocfs2_rw_lock(inode, 1);
1617         if (ret) {
1618                 mlog_errno(ret);
1619                 goto out;
1620         }
1621
1622         ret = ocfs2_meta_lock(inode, &di_bh, 1);
1623         if (ret) {
1624                 mlog_errno(ret);
1625                 goto out_rw_unlock;
1626         }
1627
1628         if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1629                 ret = -EPERM;
1630                 goto out_meta_unlock;
1631         }
1632
1633         switch (sr->l_whence) {
1634         case 0: /*SEEK_SET*/
1635                 break;
1636         case 1: /*SEEK_CUR*/
1637                 sr->l_start += f_pos;
1638                 break;
1639         case 2: /*SEEK_END*/
1640                 sr->l_start += i_size_read(inode);
1641                 break;
1642         default:
1643                 ret = -EINVAL;
1644                 goto out_meta_unlock;
1645         }
1646         sr->l_whence = 0;
1647
1648         llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1649
1650         if (sr->l_start < 0
1651             || sr->l_start > max_off
1652             || (sr->l_start + llen) < 0
1653             || (sr->l_start + llen) > max_off) {
1654                 ret = -EINVAL;
1655                 goto out_meta_unlock;
1656         }
1657         size = sr->l_start + sr->l_len;
1658
1659         if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1660                 if (sr->l_len <= 0) {
1661                         ret = -EINVAL;
1662                         goto out_meta_unlock;
1663                 }
1664         }
1665
1666         if (file && should_remove_suid(file->f_path.dentry)) {
1667                 ret = __ocfs2_write_remove_suid(inode, di_bh);
1668                 if (ret) {
1669                         mlog_errno(ret);
1670                         goto out_meta_unlock;
1671                 }
1672         }
1673
1674         down_write(&OCFS2_I(inode)->ip_alloc_sem);
1675         switch (cmd) {
1676         case OCFS2_IOC_RESVSP:
1677         case OCFS2_IOC_RESVSP64:
1678                 /*
1679                  * This takes unsigned offsets, but the signed ones we
1680                  * pass have been checked against overflow above.
1681                  */
1682                 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1683                                                        sr->l_len);
1684                 break;
1685         case OCFS2_IOC_UNRESVSP:
1686         case OCFS2_IOC_UNRESVSP64:
1687                 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1688                                                sr->l_len);
1689                 break;
1690         default:
1691                 ret = -EINVAL;
1692         }
1693         up_write(&OCFS2_I(inode)->ip_alloc_sem);
1694         if (ret) {
1695                 mlog_errno(ret);
1696                 goto out_meta_unlock;
1697         }
1698
1699         /*
1700          * We update c/mtime for these changes
1701          */
1702         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1703         if (IS_ERR(handle)) {
1704                 ret = PTR_ERR(handle);
1705                 mlog_errno(ret);
1706                 goto out_meta_unlock;
1707         }
1708
1709         if (change_size && i_size_read(inode) < size)
1710                 i_size_write(inode, size);
1711
1712         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1713         ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1714         if (ret < 0)
1715                 mlog_errno(ret);
1716
1717         ocfs2_commit_trans(osb, handle);
1718
1719 out_meta_unlock:
1720         brelse(di_bh);
1721         ocfs2_meta_unlock(inode, 1);
1722 out_rw_unlock:
1723         ocfs2_rw_unlock(inode, 1);
1724
1725         mutex_unlock(&inode->i_mutex);
1726 out:
1727         return ret;
1728 }
1729
1730 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1731                             struct ocfs2_space_resv *sr)
1732 {
1733         struct inode *inode = file->f_path.dentry->d_inode;
1734         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);;
1735
1736         if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1737             !ocfs2_writes_unwritten_extents(osb))
1738                 return -ENOTTY;
1739         else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1740                  !ocfs2_sparse_alloc(osb))
1741                 return -ENOTTY;
1742
1743         if (!S_ISREG(inode->i_mode))
1744                 return -EINVAL;
1745
1746         if (!(file->f_mode & FMODE_WRITE))
1747                 return -EBADF;
1748
1749         return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1750 }
1751
1752 static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset,
1753                             loff_t len)
1754 {
1755         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1756         struct ocfs2_space_resv sr;
1757         int change_size = 1;
1758
1759         if (!ocfs2_writes_unwritten_extents(osb))
1760                 return -EOPNOTSUPP;
1761
1762         if (S_ISDIR(inode->i_mode))
1763                 return -ENODEV;
1764
1765         if (mode & FALLOC_FL_KEEP_SIZE)
1766                 change_size = 0;
1767
1768         sr.l_whence = 0;
1769         sr.l_start = (s64)offset;
1770         sr.l_len = (s64)len;
1771
1772         return __ocfs2_change_file_space(NULL, inode, offset,
1773                                          OCFS2_IOC_RESVSP64, &sr, change_size);
1774 }
1775
1776 static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
1777                                          loff_t *ppos,
1778                                          size_t count,
1779                                          int appending,
1780                                          int *direct_io)
1781 {
1782         int ret = 0, meta_level = 0;
1783         struct inode *inode = dentry->d_inode;
1784         loff_t saved_pos, end;
1785
1786         /*
1787          * We start with a read level meta lock and only jump to an ex
1788          * if we need to make modifications here.
1789          */
1790         for(;;) {
1791                 ret = ocfs2_meta_lock(inode, NULL, meta_level);
1792                 if (ret < 0) {
1793                         meta_level = -1;
1794                         mlog_errno(ret);
1795                         goto out;
1796                 }
1797
1798                 /* Clear suid / sgid if necessary. We do this here
1799                  * instead of later in the write path because
1800                  * remove_suid() calls ->setattr without any hint that
1801                  * we may have already done our cluster locking. Since
1802                  * ocfs2_setattr() *must* take cluster locks to
1803                  * proceeed, this will lead us to recursively lock the
1804                  * inode. There's also the dinode i_size state which
1805                  * can be lost via setattr during extending writes (we
1806                  * set inode->i_size at the end of a write. */
1807                 if (should_remove_suid(dentry)) {
1808                         if (meta_level == 0) {
1809                                 ocfs2_meta_unlock(inode, meta_level);
1810                                 meta_level = 1;
1811                                 continue;
1812                         }
1813
1814                         ret = ocfs2_write_remove_suid(inode);
1815                         if (ret < 0) {
1816                                 mlog_errno(ret);
1817                                 goto out_unlock;
1818                         }
1819                 }
1820
1821                 /* work on a copy of ppos until we're sure that we won't have
1822                  * to recalculate it due to relocking. */
1823                 if (appending) {
1824                         saved_pos = i_size_read(inode);
1825                         mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
1826                 } else {
1827                         saved_pos = *ppos;
1828                 }
1829
1830                 end = saved_pos + count;
1831
1832                 /*
1833                  * Skip the O_DIRECT checks if we don't need
1834                  * them.
1835                  */
1836                 if (!direct_io || !(*direct_io))
1837                         break;
1838
1839                 /*
1840                  * There's no sane way to do direct writes to an inode
1841                  * with inline data.
1842                  */
1843                 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1844                         *direct_io = 0;
1845                         break;
1846                 }
1847
1848                 /*
1849                  * Allowing concurrent direct writes means
1850                  * i_size changes wouldn't be synchronized, so
1851                  * one node could wind up truncating another
1852                  * nodes writes.
1853                  */
1854                 if (end > i_size_read(inode)) {
1855                         *direct_io = 0;
1856                         break;
1857                 }
1858
1859                 /*
1860                  * We don't fill holes during direct io, so
1861                  * check for them here. If any are found, the
1862                  * caller will have to retake some cluster
1863                  * locks and initiate the io as buffered.
1864                  */
1865                 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
1866                 if (ret == 1) {
1867                         *direct_io = 0;
1868                         ret = 0;
1869                 } else if (ret < 0)
1870                         mlog_errno(ret);
1871                 break;
1872         }
1873
1874         if (appending)
1875                 *ppos = saved_pos;
1876
1877 out_unlock:
1878         ocfs2_meta_unlock(inode, meta_level);
1879
1880 out:
1881         return ret;
1882 }
1883
1884 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
1885                                     const struct iovec *iov,
1886                                     unsigned long nr_segs,
1887                                     loff_t pos)
1888 {
1889         int ret, direct_io, appending, rw_level, have_alloc_sem  = 0;
1890         int can_do_direct;
1891         ssize_t written = 0;
1892         size_t ocount;          /* original count */
1893         size_t count;           /* after file limit checks */
1894         loff_t old_size, *ppos = &iocb->ki_pos;
1895         u32 old_clusters;
1896         struct file *file = iocb->ki_filp;
1897         struct inode *inode = file->f_path.dentry->d_inode;
1898         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1899
1900         mlog_entry("(0x%p, %u, '%.*s')\n", file,
1901                    (unsigned int)nr_segs,
1902                    file->f_path.dentry->d_name.len,
1903                    file->f_path.dentry->d_name.name);
1904
1905         if (iocb->ki_left == 0)
1906                 return 0;
1907
1908         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1909
1910         appending = file->f_flags & O_APPEND ? 1 : 0;
1911         direct_io = file->f_flags & O_DIRECT ? 1 : 0;
1912
1913         mutex_lock(&inode->i_mutex);
1914
1915 relock:
1916         /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
1917         if (direct_io) {
1918                 down_read(&inode->i_alloc_sem);
1919                 have_alloc_sem = 1;
1920         }
1921
1922         /* concurrent O_DIRECT writes are allowed */
1923         rw_level = !direct_io;
1924         ret = ocfs2_rw_lock(inode, rw_level);
1925         if (ret < 0) {
1926                 mlog_errno(ret);
1927                 goto out_sems;
1928         }
1929
1930         can_do_direct = direct_io;
1931         ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos,
1932                                             iocb->ki_left, appending,
1933                                             &can_do_direct);
1934         if (ret < 0) {
1935                 mlog_errno(ret);
1936                 goto out;
1937         }
1938
1939         /*
1940          * We can't complete the direct I/O as requested, fall back to
1941          * buffered I/O.
1942          */
1943         if (direct_io && !can_do_direct) {
1944                 ocfs2_rw_unlock(inode, rw_level);
1945                 up_read(&inode->i_alloc_sem);
1946
1947                 have_alloc_sem = 0;
1948                 rw_level = -1;
1949
1950                 direct_io = 0;
1951                 goto relock;
1952         }
1953
1954         /*
1955          * To later detect whether a journal commit for sync writes is
1956          * necessary, we sample i_size, and cluster count here.
1957          */
1958         old_size = i_size_read(inode);
1959         old_clusters = OCFS2_I(inode)->ip_clusters;
1960
1961         /* communicate with ocfs2_dio_end_io */
1962         ocfs2_iocb_set_rw_locked(iocb, rw_level);
1963
1964         if (direct_io) {
1965                 ret = generic_segment_checks(iov, &nr_segs, &ocount,
1966                                              VERIFY_READ);
1967                 if (ret)
1968                         goto out_dio;
1969
1970                 ret = generic_write_checks(file, ppos, &count,
1971                                            S_ISBLK(inode->i_mode));
1972                 if (ret)
1973                         goto out_dio;
1974
1975                 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
1976                                                     ppos, count, ocount);
1977                 if (written < 0) {
1978                         ret = written;
1979                         goto out_dio;
1980                 }
1981         } else {
1982                 written = generic_file_aio_write_nolock(iocb, iov, nr_segs,
1983                                                         *ppos);
1984         }
1985
1986 out_dio:
1987         /* buffered aio wouldn't have proper lock coverage today */
1988         BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
1989
1990         if ((file->f_flags & O_SYNC && !direct_io) || IS_SYNC(inode)) {
1991                 /*
1992                  * The generic write paths have handled getting data
1993                  * to disk, but since we don't make use of the dirty
1994                  * inode list, a manual journal commit is necessary
1995                  * here.
1996                  */
1997                 if (old_size != i_size_read(inode) ||
1998                     old_clusters != OCFS2_I(inode)->ip_clusters) {
1999                         ret = journal_force_commit(osb->journal->j_journal);
2000                         if (ret < 0)
2001                                 written = ret;
2002                 }
2003         }
2004
2005         /*
2006          * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2007          * function pointer which is called when o_direct io completes so that
2008          * it can unlock our rw lock.  (it's the clustered equivalent of
2009          * i_alloc_sem; protects truncate from racing with pending ios).
2010          * Unfortunately there are error cases which call end_io and others
2011          * that don't.  so we don't have to unlock the rw_lock if either an
2012          * async dio is going to do it in the future or an end_io after an
2013          * error has already done it.
2014          */
2015         if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2016                 rw_level = -1;
2017                 have_alloc_sem = 0;
2018         }
2019
2020 out:
2021         if (rw_level != -1)
2022                 ocfs2_rw_unlock(inode, rw_level);
2023
2024 out_sems:
2025         if (have_alloc_sem)
2026                 up_read(&inode->i_alloc_sem);
2027
2028         mutex_unlock(&inode->i_mutex);
2029
2030         mlog_exit(ret);
2031         return written ? written : ret;
2032 }
2033
2034 static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2035                                        struct file *out,
2036                                        loff_t *ppos,
2037                                        size_t len,
2038                                        unsigned int flags)
2039 {
2040         int ret;
2041         struct inode *inode = out->f_path.dentry->d_inode;
2042
2043         mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe,
2044                    (unsigned int)len,
2045                    out->f_path.dentry->d_name.len,
2046                    out->f_path.dentry->d_name.name);
2047
2048         inode_double_lock(inode, pipe->inode);
2049
2050         ret = ocfs2_rw_lock(inode, 1);
2051         if (ret < 0) {
2052                 mlog_errno(ret);
2053                 goto out;
2054         }
2055
2056         ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0,
2057                                             NULL);
2058         if (ret < 0) {
2059                 mlog_errno(ret);
2060                 goto out_unlock;
2061         }
2062
2063         ret = generic_file_splice_write_nolock(pipe, out, ppos, len, flags);
2064
2065 out_unlock:
2066         ocfs2_rw_unlock(inode, 1);
2067 out:
2068         inode_double_unlock(inode, pipe->inode);
2069
2070         mlog_exit(ret);
2071         return ret;
2072 }
2073
2074 static ssize_t ocfs2_file_splice_read(struct file *in,
2075                                       loff_t *ppos,
2076                                       struct pipe_inode_info *pipe,
2077                                       size_t len,
2078                                       unsigned int flags)
2079 {
2080         int ret = 0;
2081         struct inode *inode = in->f_path.dentry->d_inode;
2082
2083         mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe,
2084                    (unsigned int)len,
2085                    in->f_path.dentry->d_name.len,
2086                    in->f_path.dentry->d_name.name);
2087
2088         /*
2089          * See the comment in ocfs2_file_aio_read()
2090          */
2091         ret = ocfs2_meta_lock(inode, NULL, 0);
2092         if (ret < 0) {
2093                 mlog_errno(ret);
2094                 goto bail;
2095         }
2096         ocfs2_meta_unlock(inode, 0);
2097
2098         ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2099
2100 bail:
2101         mlog_exit(ret);
2102         return ret;
2103 }
2104
2105 static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
2106                                    const struct iovec *iov,
2107                                    unsigned long nr_segs,
2108                                    loff_t pos)
2109 {
2110         int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
2111         struct file *filp = iocb->ki_filp;
2112         struct inode *inode = filp->f_path.dentry->d_inode;
2113
2114         mlog_entry("(0x%p, %u, '%.*s')\n", filp,
2115                    (unsigned int)nr_segs,
2116                    filp->f_path.dentry->d_name.len,
2117                    filp->f_path.dentry->d_name.name);
2118
2119         if (!inode) {
2120                 ret = -EINVAL;
2121                 mlog_errno(ret);
2122                 goto bail;
2123         }
2124
2125         /*
2126          * buffered reads protect themselves in ->readpage().  O_DIRECT reads
2127          * need locks to protect pending reads from racing with truncate.
2128          */
2129         if (filp->f_flags & O_DIRECT) {
2130                 down_read(&inode->i_alloc_sem);
2131                 have_alloc_sem = 1;
2132
2133                 ret = ocfs2_rw_lock(inode, 0);
2134                 if (ret < 0) {
2135                         mlog_errno(ret);
2136                         goto bail;
2137                 }
2138                 rw_level = 0;
2139                 /* communicate with ocfs2_dio_end_io */
2140                 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2141         }
2142
2143         /*
2144          * We're fine letting folks race truncates and extending
2145          * writes with read across the cluster, just like they can
2146          * locally. Hence no rw_lock during read.
2147          *
2148          * Take and drop the meta data lock to update inode fields
2149          * like i_size. This allows the checks down below
2150          * generic_file_aio_read() a chance of actually working.
2151          */
2152         ret = ocfs2_meta_lock_atime(inode, filp->f_vfsmnt, &lock_level);
2153         if (ret < 0) {
2154                 mlog_errno(ret);
2155                 goto bail;
2156         }
2157         ocfs2_meta_unlock(inode, lock_level);
2158
2159         ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
2160         if (ret == -EINVAL)
2161                 mlog(ML_ERROR, "generic_file_aio_read returned -EINVAL\n");
2162
2163         /* buffered aio wouldn't have proper lock coverage today */
2164         BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2165
2166         /* see ocfs2_file_aio_write */
2167         if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2168                 rw_level = -1;
2169                 have_alloc_sem = 0;
2170         }
2171
2172 bail:
2173         if (have_alloc_sem)
2174                 up_read(&inode->i_alloc_sem);
2175         if (rw_level != -1)
2176                 ocfs2_rw_unlock(inode, rw_level);
2177         mlog_exit(ret);
2178
2179         return ret;
2180 }
2181
2182 const struct inode_operations ocfs2_file_iops = {
2183         .setattr        = ocfs2_setattr,
2184         .getattr        = ocfs2_getattr,
2185         .permission     = ocfs2_permission,
2186         .fallocate      = ocfs2_fallocate,
2187 };
2188
2189 const struct inode_operations ocfs2_special_file_iops = {
2190         .setattr        = ocfs2_setattr,
2191         .getattr        = ocfs2_getattr,
2192         .permission     = ocfs2_permission,
2193 };
2194
2195 const struct file_operations ocfs2_fops = {
2196         .read           = do_sync_read,
2197         .write          = do_sync_write,
2198         .mmap           = ocfs2_mmap,
2199         .fsync          = ocfs2_sync_file,
2200         .release        = ocfs2_file_release,
2201         .open           = ocfs2_file_open,
2202         .aio_read       = ocfs2_file_aio_read,
2203         .aio_write      = ocfs2_file_aio_write,
2204         .ioctl          = ocfs2_ioctl,
2205 #ifdef CONFIG_COMPAT
2206         .compat_ioctl   = ocfs2_compat_ioctl,
2207 #endif
2208         .splice_read    = ocfs2_file_splice_read,
2209         .splice_write   = ocfs2_file_splice_write,
2210 };
2211
2212 const struct file_operations ocfs2_dops = {
2213         .read           = generic_read_dir,
2214         .readdir        = ocfs2_readdir,
2215         .fsync          = ocfs2_sync_file,
2216         .ioctl          = ocfs2_ioctl,
2217 #ifdef CONFIG_COMPAT
2218         .compat_ioctl   = ocfs2_compat_ioctl,
2219 #endif
2220 };