linux/fs/ocfs2/aops.c
<<
>>
Prefs
   1/* -*- mode: c; c-basic-offset: 8; -*-
   2 * vim: noexpandtab sw=8 ts=8 sts=0:
   3 *
   4 * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   5 *
   6 * This program is free software; you can redistribute it and/or
   7 * modify it under the terms of the GNU General Public
   8 * License as published by the Free Software Foundation; either
   9 * version 2 of the License, or (at your option) any later version.
  10 *
  11 * This program is distributed in the hope that it will be useful,
  12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14 * General Public License for more details.
  15 *
  16 * You should have received a copy of the GNU General Public
  17 * License along with this program; if not, write to the
  18 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  19 * Boston, MA 021110-1307, USA.
  20 */
  21
  22#include <linux/fs.h>
  23#include <linux/slab.h>
  24#include <linux/highmem.h>
  25#include <linux/pagemap.h>
  26#include <asm/byteorder.h>
  27#include <linux/swap.h>
  28#include <linux/pipe_fs_i.h>
  29#include <linux/mpage.h>
  30#include <linux/quotaops.h>
  31
  32#include <cluster/masklog.h>
  33
  34#include "ocfs2.h"
  35
  36#include "alloc.h"
  37#include "aops.h"
  38#include "dlmglue.h"
  39#include "extent_map.h"
  40#include "file.h"
  41#include "inode.h"
  42#include "journal.h"
  43#include "suballoc.h"
  44#include "super.h"
  45#include "symlink.h"
  46#include "refcounttree.h"
  47#include "ocfs2_trace.h"
  48
  49#include "buffer_head_io.h"
  50
  51static int ocfs2_symlink_get_block(struct inode *inode, sector_t iblock,
  52                                   struct buffer_head *bh_result, int create)
  53{
  54        int err = -EIO;
  55        int status;
  56        struct ocfs2_dinode *fe = NULL;
  57        struct buffer_head *bh = NULL;
  58        struct buffer_head *buffer_cache_bh = NULL;
  59        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
  60        void *kaddr;
  61
  62        trace_ocfs2_symlink_get_block(
  63                        (unsigned long long)OCFS2_I(inode)->ip_blkno,
  64                        (unsigned long long)iblock, bh_result, create);
  65
  66        BUG_ON(ocfs2_inode_is_fast_symlink(inode));
  67
  68        if ((iblock << inode->i_sb->s_blocksize_bits) > PATH_MAX + 1) {
  69                mlog(ML_ERROR, "block offset > PATH_MAX: %llu",
  70                     (unsigned long long)iblock);
  71                goto bail;
  72        }
  73
  74        status = ocfs2_read_inode_block(inode, &bh);
  75        if (status < 0) {
  76                mlog_errno(status);
  77                goto bail;
  78        }
  79        fe = (struct ocfs2_dinode *) bh->b_data;
  80
  81        if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb,
  82                                                    le32_to_cpu(fe->i_clusters))) {
  83                mlog(ML_ERROR, "block offset is outside the allocated size: "
  84                     "%llu\n", (unsigned long long)iblock);
  85                goto bail;
  86        }
  87
  88        /* We don't use the page cache to create symlink data, so if
  89         * need be, copy it over from the buffer cache. */
  90        if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) {
  91                u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) +
  92                            iblock;
  93                buffer_cache_bh = sb_getblk(osb->sb, blkno);
  94                if (!buffer_cache_bh) {
  95                        mlog(ML_ERROR, "couldn't getblock for symlink!\n");
  96                        goto bail;
  97                }
  98
  99                /* we haven't locked out transactions, so a commit
 100                 * could've happened. Since we've got a reference on
 101                 * the bh, even if it commits while we're doing the
 102                 * copy, the data is still good. */
 103                if (buffer_jbd(buffer_cache_bh)
 104                    && ocfs2_inode_is_new(inode)) {
 105                        kaddr = kmap_atomic(bh_result->b_page);
 106                        if (!kaddr) {
 107                                mlog(ML_ERROR, "couldn't kmap!\n");
 108                                goto bail;
 109                        }
 110                        memcpy(kaddr + (bh_result->b_size * iblock),
 111                               buffer_cache_bh->b_data,
 112                               bh_result->b_size);
 113                        kunmap_atomic(kaddr);
 114                        set_buffer_uptodate(bh_result);
 115                }
 116                brelse(buffer_cache_bh);
 117        }
 118
 119        map_bh(bh_result, inode->i_sb,
 120               le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + iblock);
 121
 122        err = 0;
 123
 124bail:
 125        brelse(bh);
 126
 127        return err;
 128}
 129
 130int ocfs2_get_block(struct inode *inode, sector_t iblock,
 131                    struct buffer_head *bh_result, int create)
 132{
 133        int err = 0;
 134        unsigned int ext_flags;
 135        u64 max_blocks = bh_result->b_size >> inode->i_blkbits;
 136        u64 p_blkno, count, past_eof;
 137        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 138
 139        trace_ocfs2_get_block((unsigned long long)OCFS2_I(inode)->ip_blkno,
 140                              (unsigned long long)iblock, bh_result, create);
 141
 142        if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
 143                mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n",
 144                     inode, inode->i_ino);
 145
 146        if (S_ISLNK(inode->i_mode)) {
 147                /* this always does I/O for some reason. */
 148                err = ocfs2_symlink_get_block(inode, iblock, bh_result, create);
 149                goto bail;
 150        }
 151
 152        err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, &count,
 153                                          &ext_flags);
 154        if (err) {
 155                mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, "
 156                     "%llu, NULL)\n", err, inode, (unsigned long long)iblock,
 157                     (unsigned long long)p_blkno);
 158                goto bail;
 159        }
 160
 161        if (max_blocks < count)
 162                count = max_blocks;
 163
 164        /*
 165         * ocfs2 never allocates in this function - the only time we
 166         * need to use BH_New is when we're extending i_size on a file
 167         * system which doesn't support holes, in which case BH_New
 168         * allows __block_write_begin() to zero.
 169         *
 170         * If we see this on a sparse file system, then a truncate has
 171         * raced us and removed the cluster. In this case, we clear
 172         * the buffers dirty and uptodate bits and let the buffer code
 173         * ignore it as a hole.
 174         */
 175        if (create && p_blkno == 0 && ocfs2_sparse_alloc(osb)) {
 176                clear_buffer_dirty(bh_result);
 177                clear_buffer_uptodate(bh_result);
 178                goto bail;
 179        }
 180
 181        /* Treat the unwritten extent as a hole for zeroing purposes. */
 182        if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
 183                map_bh(bh_result, inode->i_sb, p_blkno);
 184
 185        bh_result->b_size = count << inode->i_blkbits;
 186
 187        if (!ocfs2_sparse_alloc(osb)) {
 188                if (p_blkno == 0) {
 189                        err = -EIO;
 190                        mlog(ML_ERROR,
 191                             "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
 192                             (unsigned long long)iblock,
 193                             (unsigned long long)p_blkno,
 194                             (unsigned long long)OCFS2_I(inode)->ip_blkno);
 195                        mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters);
 196                        dump_stack();
 197                        goto bail;
 198                }
 199        }
 200
 201        past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
 202
 203        trace_ocfs2_get_block_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
 204                                  (unsigned long long)past_eof);
 205        if (create && (iblock >= past_eof))
 206                set_buffer_new(bh_result);
 207
 208bail:
 209        if (err < 0)
 210                err = -EIO;
 211
 212        return err;
 213}
 214
 215int ocfs2_read_inline_data(struct inode *inode, struct page *page,
 216                           struct buffer_head *di_bh)
 217{
 218        void *kaddr;
 219        loff_t size;
 220        struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 221
 222        if (!(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)) {
 223                ocfs2_error(inode->i_sb, "Inode %llu lost inline data flag",
 224                            (unsigned long long)OCFS2_I(inode)->ip_blkno);
 225                return -EROFS;
 226        }
 227
 228        size = i_size_read(inode);
 229
 230        if (size > PAGE_CACHE_SIZE ||
 231            size > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) {
 232                ocfs2_error(inode->i_sb,
 233                            "Inode %llu has with inline data has bad size: %Lu",
 234                            (unsigned long long)OCFS2_I(inode)->ip_blkno,
 235                            (unsigned long long)size);
 236                return -EROFS;
 237        }
 238
 239        kaddr = kmap_atomic(page);
 240        if (size)
 241                memcpy(kaddr, di->id2.i_data.id_data, size);
 242        /* Clear the remaining part of the page */
 243        memset(kaddr + size, 0, PAGE_CACHE_SIZE - size);
 244        flush_dcache_page(page);
 245        kunmap_atomic(kaddr);
 246
 247        SetPageUptodate(page);
 248
 249        return 0;
 250}
 251
 252static int ocfs2_readpage_inline(struct inode *inode, struct page *page)
 253{
 254        int ret;
 255        struct buffer_head *di_bh = NULL;
 256
 257        BUG_ON(!PageLocked(page));
 258        BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
 259
 260        ret = ocfs2_read_inode_block(inode, &di_bh);
 261        if (ret) {
 262                mlog_errno(ret);
 263                goto out;
 264        }
 265
 266        ret = ocfs2_read_inline_data(inode, page, di_bh);
 267out:
 268        unlock_page(page);
 269
 270        brelse(di_bh);
 271        return ret;
 272}
 273
 274static int ocfs2_readpage(struct file *file, struct page *page)
 275{
 276        struct inode *inode = page->mapping->host;
 277        struct ocfs2_inode_info *oi = OCFS2_I(inode);
 278        loff_t start = (loff_t)page->index << PAGE_CACHE_SHIFT;
 279        int ret, unlock = 1;
 280
 281        trace_ocfs2_readpage((unsigned long long)oi->ip_blkno,
 282                             (page ? page->index : 0));
 283
 284        ret = ocfs2_inode_lock_with_page(inode, NULL, 0, page);
 285        if (ret != 0) {
 286                if (ret == AOP_TRUNCATED_PAGE)
 287                        unlock = 0;
 288                mlog_errno(ret);
 289                goto out;
 290        }
 291
 292        if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
 293                /*
 294                 * Unlock the page and cycle ip_alloc_sem so that we don't
 295                 * busyloop waiting for ip_alloc_sem to unlock
 296                 */
 297                ret = AOP_TRUNCATED_PAGE;
 298                unlock_page(page);
 299                unlock = 0;
 300                down_read(&oi->ip_alloc_sem);
 301                up_read(&oi->ip_alloc_sem);
 302                goto out_inode_unlock;
 303        }
 304
 305        /*
 306         * i_size might have just been updated as we grabed the meta lock.  We
 307         * might now be discovering a truncate that hit on another node.
 308         * block_read_full_page->get_block freaks out if it is asked to read
 309         * beyond the end of a file, so we check here.  Callers
 310         * (generic_file_read, vm_ops->fault) are clever enough to check i_size
 311         * and notice that the page they just read isn't needed.
 312         *
 313         * XXX sys_readahead() seems to get that wrong?
 314         */
 315        if (start >= i_size_read(inode)) {
 316                zero_user(page, 0, PAGE_SIZE);
 317                SetPageUptodate(page);
 318                ret = 0;
 319                goto out_alloc;
 320        }
 321
 322        if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
 323                ret = ocfs2_readpage_inline(inode, page);
 324        else
 325                ret = block_read_full_page(page, ocfs2_get_block);
 326        unlock = 0;
 327
 328out_alloc:
 329        up_read(&OCFS2_I(inode)->ip_alloc_sem);
 330out_inode_unlock:
 331        ocfs2_inode_unlock(inode, 0);
 332out:
 333        if (unlock)
 334                unlock_page(page);
 335        return ret;
 336}
 337
 338/*
 339 * This is used only for read-ahead. Failures or difficult to handle
 340 * situations are safe to ignore.
 341 *
 342 * Right now, we don't bother with BH_Boundary - in-inode extent lists
 343 * are quite large (243 extents on 4k blocks), so most inodes don't
 344 * grow out to a tree. If need be, detecting boundary extents could
 345 * trivially be added in a future version of ocfs2_get_block().
 346 */
 347static int ocfs2_readpages(struct file *filp, struct address_space *mapping,
 348                           struct list_head *pages, unsigned nr_pages)
 349{
 350        int ret, err = -EIO;
 351        struct inode *inode = mapping->host;
 352        struct ocfs2_inode_info *oi = OCFS2_I(inode);
 353        loff_t start;
 354        struct page *last;
 355
 356        /*
 357         * Use the nonblocking flag for the dlm code to avoid page
 358         * lock inversion, but don't bother with retrying.
 359         */
 360        ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK);
 361        if (ret)
 362                return err;
 363
 364        if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
 365                ocfs2_inode_unlock(inode, 0);
 366                return err;
 367        }
 368
 369        /*
 370         * Don't bother with inline-data. There isn't anything
 371         * to read-ahead in that case anyway...
 372         */
 373        if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
 374                goto out_unlock;
 375
 376        /*
 377         * Check whether a remote node truncated this file - we just
 378         * drop out in that case as it's not worth handling here.
 379         */
 380        last = list_entry(pages->prev, struct page, lru);
 381        start = (loff_t)last->index << PAGE_CACHE_SHIFT;
 382        if (start >= i_size_read(inode))
 383                goto out_unlock;
 384
 385        err = mpage_readpages(mapping, pages, nr_pages, ocfs2_get_block);
 386
 387out_unlock:
 388        up_read(&oi->ip_alloc_sem);
 389        ocfs2_inode_unlock(inode, 0);
 390
 391        return err;
 392}
 393
 394/* Note: Because we don't support holes, our allocation has
 395 * already happened (allocation writes zeros to the file data)
 396 * so we don't have to worry about ordered writes in
 397 * ocfs2_writepage.
 398 *
 399 * ->writepage is called during the process of invalidating the page cache
 400 * during blocked lock processing.  It can't block on any cluster locks
 401 * to during block mapping.  It's relying on the fact that the block
 402 * mapping can't have disappeared under the dirty pages that it is
 403 * being asked to write back.
 404 */
 405static int ocfs2_writepage(struct page *page, struct writeback_control *wbc)
 406{
 407        trace_ocfs2_writepage(
 408                (unsigned long long)OCFS2_I(page->mapping->host)->ip_blkno,
 409                page->index);
 410
 411        return block_write_full_page(page, ocfs2_get_block, wbc);
 412}
 413
 414/* Taken from ext3. We don't necessarily need the full blown
 415 * functionality yet, but IMHO it's better to cut and paste the whole
 416 * thing so we can avoid introducing our own bugs (and easily pick up
 417 * their fixes when they happen) --Mark */
 418int walk_page_buffers(  handle_t *handle,
 419                        struct buffer_head *head,
 420                        unsigned from,
 421                        unsigned to,
 422                        int *partial,
 423                        int (*fn)(      handle_t *handle,
 424                                        struct buffer_head *bh))
 425{
 426        struct buffer_head *bh;
 427        unsigned block_start, block_end;
 428        unsigned blocksize = head->b_size;
 429        int err, ret = 0;
 430        struct buffer_head *next;
 431
 432        for (   bh = head, block_start = 0;
 433                ret == 0 && (bh != head || !block_start);
 434                block_start = block_end, bh = next)
 435        {
 436                next = bh->b_this_page;
 437                block_end = block_start + blocksize;
 438                if (block_end <= from || block_start >= to) {
 439                        if (partial && !buffer_uptodate(bh))
 440                                *partial = 1;
 441                        continue;
 442                }
 443                err = (*fn)(handle, bh);
 444                if (!ret)
 445                        ret = err;
 446        }
 447        return ret;
 448}
 449
 450static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block)
 451{
 452        sector_t status;
 453        u64 p_blkno = 0;
 454        int err = 0;
 455        struct inode *inode = mapping->host;
 456
 457        trace_ocfs2_bmap((unsigned long long)OCFS2_I(inode)->ip_blkno,
 458                         (unsigned long long)block);
 459
 460        /* We don't need to lock journal system files, since they aren't
 461         * accessed concurrently from multiple nodes.
 462         */
 463        if (!INODE_JOURNAL(inode)) {
 464                err = ocfs2_inode_lock(inode, NULL, 0);
 465                if (err) {
 466                        if (err != -ENOENT)
 467                                mlog_errno(err);
 468                        goto bail;
 469                }
 470                down_read(&OCFS2_I(inode)->ip_alloc_sem);
 471        }
 472
 473        if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
 474                err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
 475                                                  NULL);
 476
 477        if (!INODE_JOURNAL(inode)) {
 478                up_read(&OCFS2_I(inode)->ip_alloc_sem);
 479                ocfs2_inode_unlock(inode, 0);
 480        }
 481
 482        if (err) {
 483                mlog(ML_ERROR, "get_blocks() failed, block = %llu\n",
 484                     (unsigned long long)block);
 485                mlog_errno(err);
 486                goto bail;
 487        }
 488
 489bail:
 490        status = err ? 0 : p_blkno;
 491
 492        return status;
 493}
 494
 495/*
 496 * TODO: Make this into a generic get_blocks function.
 497 *
 498 * From do_direct_io in direct-io.c:
 499 *  "So what we do is to permit the ->get_blocks function to populate
 500 *   bh.b_size with the size of IO which is permitted at this offset and
 501 *   this i_blkbits."
 502 *
 503 * This function is called directly from get_more_blocks in direct-io.c.
 504 *
 505 * called like this: dio->get_blocks(dio->inode, fs_startblk,
 506 *                                      fs_count, map_bh, dio->rw == WRITE);
 507 *
 508 * Note that we never bother to allocate blocks here, and thus ignore the
 509 * create argument.
 510 */
 511static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock,
 512                                     struct buffer_head *bh_result, int create)
 513{
 514        int ret;
 515        u64 p_blkno, inode_blocks, contig_blocks;
 516        unsigned int ext_flags;
 517        unsigned char blocksize_bits = inode->i_sb->s_blocksize_bits;
 518        unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits;
 519
 520        /* This function won't even be called if the request isn't all
 521         * nicely aligned and of the right size, so there's no need
 522         * for us to check any of that. */
 523
 524        inode_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
 525
 526        /* This figures out the size of the next contiguous block, and
 527         * our logical offset */
 528        ret = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno,
 529                                          &contig_blocks, &ext_flags);
 530        if (ret) {
 531                mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n",
 532                     (unsigned long long)iblock);
 533                ret = -EIO;
 534                goto bail;
 535        }
 536
 537        /* We should already CoW the refcounted extent in case of create. */
 538        BUG_ON(create && (ext_flags & OCFS2_EXT_REFCOUNTED));
 539
 540        /*
 541         * get_more_blocks() expects us to describe a hole by clearing
 542         * the mapped bit on bh_result().
 543         *
 544         * Consider an unwritten extent as a hole.
 545         */
 546        if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
 547                map_bh(bh_result, inode->i_sb, p_blkno);
 548        else
 549                clear_buffer_mapped(bh_result);
 550
 551        /* make sure we don't map more than max_blocks blocks here as
 552           that's all the kernel will handle at this point. */
 553        if (max_blocks < contig_blocks)
 554                contig_blocks = max_blocks;
 555        bh_result->b_size = contig_blocks << blocksize_bits;
 556bail:
 557        return ret;
 558}
 559
 560/*
 561 * ocfs2_dio_end_io is called by the dio core when a dio is finished.  We're
 562 * particularly interested in the aio/dio case.  We use the rw_lock DLM lock
 563 * to protect io on one node from truncation on another.
 564 */
 565static void ocfs2_dio_end_io(struct kiocb *iocb,
 566                             loff_t offset,
 567                             ssize_t bytes,
 568                             void *private,
 569                             int ret,
 570                             bool is_async)
 571{
 572        struct inode *inode = file_inode(iocb->ki_filp);
 573        int level;
 574        wait_queue_head_t *wq = ocfs2_ioend_wq(inode);
 575
 576        /* this io's submitter should not have unlocked this before we could */
 577        BUG_ON(!ocfs2_iocb_is_rw_locked(iocb));
 578
 579        if (ocfs2_iocb_is_sem_locked(iocb))
 580                ocfs2_iocb_clear_sem_locked(iocb);
 581
 582        if (ocfs2_iocb_is_unaligned_aio(iocb)) {
 583                ocfs2_iocb_clear_unaligned_aio(iocb);
 584
 585                if (atomic_dec_and_test(&OCFS2_I(inode)->ip_unaligned_aio) &&
 586                    waitqueue_active(wq)) {
 587                        wake_up_all(wq);
 588                }
 589        }
 590
 591        ocfs2_iocb_clear_rw_locked(iocb);
 592
 593        level = ocfs2_iocb_rw_locked_level(iocb);
 594        ocfs2_rw_unlock(inode, level);
 595
 596        inode_dio_done(inode);
 597        if (is_async)
 598                aio_complete(iocb, ret, 0);
 599}
 600
 601/*
 602 * ocfs2_invalidatepage() and ocfs2_releasepage() are shamelessly stolen
 603 * from ext3.  PageChecked() bits have been removed as OCFS2 does not
 604 * do journalled data.
 605 */
 606static void ocfs2_invalidatepage(struct page *page, unsigned int offset,
 607                                 unsigned int length)
 608{
 609        journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal;
 610
 611        jbd2_journal_invalidatepage(journal, page, offset, length);
 612}
 613
 614static int ocfs2_releasepage(struct page *page, gfp_t wait)
 615{
 616        journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal;
 617
 618        if (!page_has_buffers(page))
 619                return 0;
 620        return jbd2_journal_try_to_free_buffers(journal, page, wait);
 621}
 622
 623static ssize_t ocfs2_direct_IO(int rw,
 624                               struct kiocb *iocb,
 625                               const struct iovec *iov,
 626                               loff_t offset,
 627                               unsigned long nr_segs)
 628{
 629        struct file *file = iocb->ki_filp;
 630        struct inode *inode = file_inode(file)->i_mapping->host;
 631
 632        /*
 633         * Fallback to buffered I/O if we see an inode without
 634         * extents.
 635         */
 636        if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
 637                return 0;
 638
 639        /* Fallback to buffered I/O if we are appending. */
 640        if (i_size_read(inode) <= offset)
 641                return 0;
 642
 643        return __blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev,
 644                                    iov, offset, nr_segs,
 645                                    ocfs2_direct_IO_get_blocks,
 646                                    ocfs2_dio_end_io, NULL, 0);
 647}
 648
 649static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb,
 650                                            u32 cpos,
 651                                            unsigned int *start,
 652                                            unsigned int *end)
 653{
 654        unsigned int cluster_start = 0, cluster_end = PAGE_CACHE_SIZE;
 655
 656        if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) {
 657                unsigned int cpp;
 658
 659                cpp = 1 << (PAGE_CACHE_SHIFT - osb->s_clustersize_bits);
 660
 661                cluster_start = cpos % cpp;
 662                cluster_start = cluster_start << osb->s_clustersize_bits;
 663
 664                cluster_end = cluster_start + osb->s_clustersize;
 665        }
 666
 667        BUG_ON(cluster_start > PAGE_SIZE);
 668        BUG_ON(cluster_end > PAGE_SIZE);
 669
 670        if (start)
 671                *start = cluster_start;
 672        if (end)
 673                *end = cluster_end;
 674}
 675
 676/*
 677 * 'from' and 'to' are the region in the page to avoid zeroing.
 678 *
 679 * If pagesize > clustersize, this function will avoid zeroing outside
 680 * of the cluster boundary.
 681 *
 682 * from == to == 0 is code for "zero the entire cluster region"
 683 */
 684static void ocfs2_clear_page_regions(struct page *page,
 685                                     struct ocfs2_super *osb, u32 cpos,
 686                                     unsigned from, unsigned to)
 687{
 688        void *kaddr;
 689        unsigned int cluster_start, cluster_end;
 690
 691        ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end);
 692
 693        kaddr = kmap_atomic(page);
 694
 695        if (from || to) {
 696                if (from > cluster_start)
 697                        memset(kaddr + cluster_start, 0, from - cluster_start);
 698                if (to < cluster_end)
 699                        memset(kaddr + to, 0, cluster_end - to);
 700        } else {
 701                memset(kaddr + cluster_start, 0, cluster_end - cluster_start);
 702        }
 703
 704        kunmap_atomic(kaddr);
 705}
 706
 707/*
 708 * Nonsparse file systems fully allocate before we get to the write
 709 * code. This prevents ocfs2_write() from tagging the write as an
 710 * allocating one, which means ocfs2_map_page_blocks() might try to
 711 * read-in the blocks at the tail of our file. Avoid reading them by
 712 * testing i_size against each block offset.
 713 */
 714static int ocfs2_should_read_blk(struct inode *inode, struct page *page,
 715                                 unsigned int block_start)
 716{
 717        u64 offset = page_offset(page) + block_start;
 718
 719        if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
 720                return 1;
 721
 722        if (i_size_read(inode) > offset)
 723                return 1;
 724
 725        return 0;
 726}
 727
 728/*
 729 * Some of this taken from __block_write_begin(). We already have our
 730 * mapping by now though, and the entire write will be allocating or
 731 * it won't, so not much need to use BH_New.
 732 *
 733 * This will also skip zeroing, which is handled externally.
 734 */
 735int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
 736                          struct inode *inode, unsigned int from,
 737                          unsigned int to, int new)
 738{
 739        int ret = 0;
 740        struct buffer_head *head, *bh, *wait[2], **wait_bh = wait;
 741        unsigned int block_end, block_start;
 742        unsigned int bsize = 1 << inode->i_blkbits;
 743
 744        if (!page_has_buffers(page))
 745                create_empty_buffers(page, bsize, 0);
 746
 747        head = page_buffers(page);
 748        for (bh = head, block_start = 0; bh != head || !block_start;
 749             bh = bh->b_this_page, block_start += bsize) {
 750                block_end = block_start + bsize;
 751
 752                clear_buffer_new(bh);
 753
 754                /*
 755                 * Ignore blocks outside of our i/o range -
 756                 * they may belong to unallocated clusters.
 757                 */
 758                if (block_start >= to || block_end <= from) {
 759                        if (PageUptodate(page))
 760                                set_buffer_uptodate(bh);
 761                        continue;
 762                }
 763
 764                /*
 765                 * For an allocating write with cluster size >= page
 766                 * size, we always write the entire page.
 767                 */
 768                if (new)
 769                        set_buffer_new(bh);
 770
 771                if (!buffer_mapped(bh)) {
 772                        map_bh(bh, inode->i_sb, *p_blkno);
 773                        unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
 774                }
 775
 776                if (PageUptodate(page)) {
 777                        if (!buffer_uptodate(bh))
 778                                set_buffer_uptodate(bh);
 779                } else if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
 780                           !buffer_new(bh) &&
 781                           ocfs2_should_read_blk(inode, page, block_start) &&
 782                           (block_start < from || block_end > to)) {
 783                        ll_rw_block(READ, 1, &bh);
 784                        *wait_bh++=bh;
 785                }
 786
 787                *p_blkno = *p_blkno + 1;
 788        }
 789
 790        /*
 791         * If we issued read requests - let them complete.
 792         */
 793        while(wait_bh > wait) {
 794                wait_on_buffer(*--wait_bh);
 795                if (!buffer_uptodate(*wait_bh))
 796                        ret = -EIO;
 797        }
 798
 799        if (ret == 0 || !new)
 800                return ret;
 801
 802        /*
 803         * If we get -EIO above, zero out any newly allocated blocks
 804         * to avoid exposing stale data.
 805         */
 806        bh = head;
 807        block_start = 0;
 808        do {
 809                block_end = block_start + bsize;
 810                if (block_end <= from)
 811                        goto next_bh;
 812                if (block_start >= to)
 813                        break;
 814
 815                zero_user(page, block_start, bh->b_size);
 816                set_buffer_uptodate(bh);
 817                mark_buffer_dirty(bh);
 818
 819next_bh:
 820                block_start = block_end;
 821                bh = bh->b_this_page;
 822        } while (bh != head);
 823
 824        return ret;
 825}
 826
 827#if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
 828#define OCFS2_MAX_CTXT_PAGES    1
 829#else
 830#define OCFS2_MAX_CTXT_PAGES    (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE)
 831#endif
 832
 833#define OCFS2_MAX_CLUSTERS_PER_PAGE     (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE)
 834
 835/*
 836 * Describe the state of a single cluster to be written to.
 837 */
 838struct ocfs2_write_cluster_desc {
 839        u32             c_cpos;
 840        u32             c_phys;
 841        /*
 842         * Give this a unique field because c_phys eventually gets
 843         * filled.
 844         */
 845        unsigned        c_new;
 846        unsigned        c_unwritten;
 847        unsigned        c_needs_zero;
 848};
 849
 850struct ocfs2_write_ctxt {
 851        /* Logical cluster position / len of write */
 852        u32                             w_cpos;
 853        u32                             w_clen;
 854
 855        /* First cluster allocated in a nonsparse extend */
 856        u32                             w_first_new_cpos;
 857
 858        struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE];
 859
 860        /*
 861         * This is true if page_size > cluster_size.
 862         *
 863         * It triggers a set of special cases during write which might
 864         * have to deal with allocating writes to partial pages.
 865         */
 866        unsigned int                    w_large_pages;
 867
 868        /*
 869         * Pages involved in this write.
 870         *
 871         * w_target_page is the page being written to by the user.
 872         *
 873         * w_pages is an array of pages which always contains
 874         * w_target_page, and in the case of an allocating write with
 875         * page_size < cluster size, it will contain zero'd and mapped
 876         * pages adjacent to w_target_page which need to be written
 877         * out in so that future reads from that region will get
 878         * zero's.
 879         */
 880        unsigned int                    w_num_pages;
 881        struct page                     *w_pages[OCFS2_MAX_CTXT_PAGES];
 882        struct page                     *w_target_page;
 883
 884        /*
 885         * w_target_locked is used for page_mkwrite path indicating no unlocking
 886         * against w_target_page in ocfs2_write_end_nolock.
 887         */
 888        unsigned int                    w_target_locked:1;
 889
 890        /*
 891         * ocfs2_write_end() uses this to know what the real range to
 892         * write in the target should be.
 893         */
 894        unsigned int                    w_target_from;
 895        unsigned int                    w_target_to;
 896
 897        /*
 898         * We could use journal_current_handle() but this is cleaner,
 899         * IMHO -Mark
 900         */
 901        handle_t                        *w_handle;
 902
 903        struct buffer_head              *w_di_bh;
 904
 905        struct ocfs2_cached_dealloc_ctxt w_dealloc;
 906};
 907
 908void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages)
 909{
 910        int i;
 911
 912        for(i = 0; i < num_pages; i++) {
 913                if (pages[i]) {
 914                        unlock_page(pages[i]);
 915                        mark_page_accessed(pages[i]);
 916                        page_cache_release(pages[i]);
 917                }
 918        }
 919}
 920
 921static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt *wc)
 922{
 923        int i;
 924
 925        /*
 926         * w_target_locked is only set to true in the page_mkwrite() case.
 927         * The intent is to allow us to lock the target page from write_begin()
 928         * to write_end(). The caller must hold a ref on w_target_page.
 929         */
 930        if (wc->w_target_locked) {
 931                BUG_ON(!wc->w_target_page);
 932                for (i = 0; i < wc->w_num_pages; i++) {
 933                        if (wc->w_target_page == wc->w_pages[i]) {
 934                                wc->w_pages[i] = NULL;
 935                                break;
 936                        }
 937                }
 938                mark_page_accessed(wc->w_target_page);
 939                page_cache_release(wc->w_target_page);
 940        }
 941        ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages);
 942
 943        brelse(wc->w_di_bh);
 944        kfree(wc);
 945}
 946
 947static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp,
 948                                  struct ocfs2_super *osb, loff_t pos,
 949                                  unsigned len, struct buffer_head *di_bh)
 950{
 951        u32 cend;
 952        struct ocfs2_write_ctxt *wc;
 953
 954        wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS);
 955        if (!wc)
 956                return -ENOMEM;
 957
 958        wc->w_cpos = pos >> osb->s_clustersize_bits;
 959        wc->w_first_new_cpos = UINT_MAX;
 960        cend = (pos + len - 1) >> osb->s_clustersize_bits;
 961        wc->w_clen = cend - wc->w_cpos + 1;
 962        get_bh(di_bh);
 963        wc->w_di_bh = di_bh;
 964
 965        if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits))
 966                wc->w_large_pages = 1;
 967        else
 968                wc->w_large_pages = 0;
 969
 970        ocfs2_init_dealloc_ctxt(&wc->w_dealloc);
 971
 972        *wcp = wc;
 973
 974        return 0;
 975}
 976
 977/*
 978 * If a page has any new buffers, zero them out here, and mark them uptodate
 979 * and dirty so they'll be written out (in order to prevent uninitialised
 980 * block data from leaking). And clear the new bit.
 981 */
 982static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to)
 983{
 984        unsigned int block_start, block_end;
 985        struct buffer_head *head, *bh;
 986
 987        BUG_ON(!PageLocked(page));
 988        if (!page_has_buffers(page))
 989                return;
 990
 991        bh = head = page_buffers(page);
 992        block_start = 0;
 993        do {
 994                block_end = block_start + bh->b_size;
 995
 996                if (buffer_new(bh)) {
 997                        if (block_end > from && block_start < to) {
 998                                if (!PageUptodate(page)) {
 999                                        unsigned start, end;
1000
1001                                        start = max(from, block_start);
1002                                        end = min(to, block_end);
1003
1004                                        zero_user_segment(page, start, end);
1005                                        set_buffer_uptodate(bh);
1006                                }
1007
1008                                clear_buffer_new(bh);
1009                                mark_buffer_dirty(bh);
1010                        }
1011                }
1012
1013                block_start = block_end;
1014                bh = bh->b_this_page;
1015        } while (bh != head);
1016}
1017
1018/*
1019 * Only called when we have a failure during allocating write to write
1020 * zero's to the newly allocated region.
1021 */
1022static void ocfs2_write_failure(struct inode *inode,
1023                                struct ocfs2_write_ctxt *wc,
1024                                loff_t user_pos, unsigned user_len)
1025{
1026        int i;
1027        unsigned from = user_pos & (PAGE_CACHE_SIZE - 1),
1028                to = user_pos + user_len;
1029        struct page *tmppage;
1030
1031        ocfs2_zero_new_buffers(wc->w_target_page, from, to);
1032
1033        for(i = 0; i < wc->w_num_pages; i++) {
1034                tmppage = wc->w_pages[i];
1035
1036                if (page_has_buffers(tmppage)) {
1037                        if (ocfs2_should_order_data(inode))
1038                                ocfs2_jbd2_file_inode(wc->w_handle, inode);
1039
1040                        block_commit_write(tmppage, from, to);
1041                }
1042        }
1043}
1044
1045static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno,
1046                                        struct ocfs2_write_ctxt *wc,
1047                                        struct page *page, u32 cpos,
1048                                        loff_t user_pos, unsigned user_len,
1049                                        int new)
1050{
1051        int ret;
1052        unsigned int map_from = 0, map_to = 0;
1053        unsigned int cluster_start, cluster_end;
1054        unsigned int user_data_from = 0, user_data_to = 0;
1055
1056        ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos,
1057                                        &cluster_start, &cluster_end);
1058
1059        /* treat the write as new if the a hole/lseek spanned across
1060         * the page boundary.
1061         */
1062        new = new | ((i_size_read(inode) <= page_offset(page)) &&
1063                        (page_offset(page) <= user_pos));
1064
1065        if (page == wc->w_target_page) {
1066                map_from = user_pos & (PAGE_CACHE_SIZE - 1);
1067                map_to = map_from + user_len;
1068
1069                if (new)
1070                        ret = ocfs2_map_page_blocks(page, p_blkno, inode,
1071                                                    cluster_start, cluster_end,
1072                                                    new);
1073                else
1074                        ret = ocfs2_map_page_blocks(page, p_blkno, inode,
1075                                                    map_from, map_to, new);
1076                if (ret) {
1077                        mlog_errno(ret);
1078                        goto out;
1079                }
1080
1081                user_data_from = map_from;
1082                user_data_to = map_to;
1083                if (new) {
1084                        map_from = cluster_start;
1085                        map_to = cluster_end;
1086                }
1087        } else {
1088                /*
1089                 * If we haven't allocated the new page yet, we
1090                 * shouldn't be writing it out without copying user
1091                 * data. This is likely a math error from the caller.
1092                 */
1093                BUG_ON(!new);
1094
1095                map_from = cluster_start;
1096                map_to = cluster_end;
1097
1098                ret = ocfs2_map_page_blocks(page, p_blkno, inode,
1099                                            cluster_start, cluster_end, new);
1100                if (ret) {
1101                        mlog_errno(ret);
1102                        goto out;
1103                }
1104        }
1105
1106        /*
1107         * Parts of newly allocated pages need to be zero'd.
1108         *
1109         * Above, we have also rewritten 'to' and 'from' - as far as
1110         * the rest of the function is concerned, the entire cluster
1111         * range inside of a page needs to be written.
1112         *
1113         * We can skip this if the page is up to date - it's already
1114         * been zero'd from being read in as a hole.
1115         */
1116        if (new && !PageUptodate(page))
1117                ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb),
1118                                         cpos, user_data_from, user_data_to);
1119
1120        flush_dcache_page(page);
1121
1122out:
1123        return ret;
1124}
1125
1126/*
1127 * This function will only grab one clusters worth of pages.
1128 */
1129static int ocfs2_grab_pages_for_write(struct address_space *mapping,
1130                                      struct ocfs2_write_ctxt *wc,
1131                                      u32 cpos, loff_t user_pos,
1132                                      unsigned user_len, int new,
1133                                      struct page *mmap_page)
1134{
1135        int ret = 0, i;
1136        unsigned long start, target_index, end_index, index;
1137        struct inode *inode = mapping->host;
1138        loff_t last_byte;
1139
1140        target_index = user_pos >> PAGE_CACHE_SHIFT;
1141
1142        /*
1143         * Figure out how many pages we'll be manipulating here. For
1144         * non allocating write, we just change the one
1145         * page. Otherwise, we'll need a whole clusters worth.  If we're
1146         * writing past i_size, we only need enough pages to cover the
1147         * last page of the write.
1148         */
1149        if (new) {
1150                wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb);
1151                start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos);
1152                /*
1153                 * We need the index *past* the last page we could possibly
1154                 * touch.  This is the page past the end of the write or
1155                 * i_size, whichever is greater.
1156                 */
1157                last_byte = max(user_pos + user_len, i_size_read(inode));
1158                BUG_ON(last_byte < 1);
1159                end_index = ((last_byte - 1) >> PAGE_CACHE_SHIFT) + 1;
1160                if ((start + wc->w_num_pages) > end_index)
1161                        wc->w_num_pages = end_index - start;
1162        } else {
1163                wc->w_num_pages = 1;
1164                start = target_index;
1165        }
1166
1167        for(i = 0; i < wc->w_num_pages; i++) {
1168                index = start + i;
1169
1170                if (index == target_index && mmap_page) {
1171                        /*
1172                         * ocfs2_pagemkwrite() is a little different
1173                         * and wants us to directly use the page
1174                         * passed in.
1175                         */
1176                        lock_page(mmap_page);
1177
1178                        /* Exit and let the caller retry */
1179                        if (mmap_page->mapping != mapping) {
1180                                WARN_ON(mmap_page->mapping);
1181                                unlock_page(mmap_page);
1182                                ret = -EAGAIN;
1183                                goto out;
1184                        }
1185
1186                        page_cache_get(mmap_page);
1187                        wc->w_pages[i] = mmap_page;
1188                        wc->w_target_locked = true;
1189                } else {
1190                        wc->w_pages[i] = find_or_create_page(mapping, index,
1191                                                             GFP_NOFS);
1192                        if (!wc->w_pages[i]) {
1193                                ret = -ENOMEM;
1194                                mlog_errno(ret);
1195                                goto out;
1196                        }
1197                }
1198                wait_for_stable_page(wc->w_pages[i]);
1199
1200                if (index == target_index)
1201                        wc->w_target_page = wc->w_pages[i];
1202        }
1203out:
1204        if (ret)
1205                wc->w_target_locked = false;
1206        return ret;
1207}
1208
1209/*
1210 * Prepare a single cluster for write one cluster into the file.
1211 */
1212static int ocfs2_write_cluster(struct address_space *mapping,
1213                               u32 phys, unsigned int unwritten,
1214                               unsigned int should_zero,
1215                               struct ocfs2_alloc_context *data_ac,
1216                               struct ocfs2_alloc_context *meta_ac,
1217                               struct ocfs2_write_ctxt *wc, u32 cpos,
1218                               loff_t user_pos, unsigned user_len)
1219{
1220        int ret, i, new;
1221        u64 v_blkno, p_blkno;
1222        struct inode *inode = mapping->host;
1223        struct ocfs2_extent_tree et;
1224
1225        new = phys == 0 ? 1 : 0;
1226        if (new) {
1227                u32 tmp_pos;
1228
1229                /*
1230                 * This is safe to call with the page locks - it won't take
1231                 * any additional semaphores or cluster locks.
1232                 */
1233                tmp_pos = cpos;
1234                ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode,
1235                                           &tmp_pos, 1, 0, wc->w_di_bh,
1236                                           wc->w_handle, data_ac,
1237                                           meta_ac, NULL);
1238                /*
1239                 * This shouldn't happen because we must have already
1240                 * calculated the correct meta data allocation required. The
1241                 * internal tree allocation code should know how to increase
1242                 * transaction credits itself.
1243                 *
1244                 * If need be, we could handle -EAGAIN for a
1245                 * RESTART_TRANS here.
1246                 */
1247                mlog_bug_on_msg(ret == -EAGAIN,
1248                                "Inode %llu: EAGAIN return during allocation.\n",
1249                                (unsigned long long)OCFS2_I(inode)->ip_blkno);
1250                if (ret < 0) {
1251                        mlog_errno(ret);
1252                        goto out;
1253                }
1254        } else if (unwritten) {
1255                ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
1256                                              wc->w_di_bh);
1257                ret = ocfs2_mark_extent_written(inode, &et,
1258                                                wc->w_handle, cpos, 1, phys,
1259                                                meta_ac, &wc->w_dealloc);
1260                if (ret < 0) {
1261                        mlog_errno(ret);
1262                        goto out;
1263                }
1264        }
1265
1266        if (should_zero)
1267                v_blkno = ocfs2_clusters_to_blocks(inode->i_sb, cpos);
1268        else
1269                v_blkno = user_pos >> inode->i_sb->s_blocksize_bits;
1270
1271        /*
1272         * The only reason this should fail is due to an inability to
1273         * find the extent added.
1274         */
1275        ret = ocfs2_extent_map_get_blocks(inode, v_blkno, &p_blkno, NULL,
1276                                          NULL);
1277        if (ret < 0) {
1278                ocfs2_error(inode->i_sb, "Corrupting extend for inode %llu, "
1279                            "at logical block %llu",
1280                            (unsigned long long)OCFS2_I(inode)->ip_blkno,
1281                            (unsigned long long)v_blkno);
1282                goto out;
1283        }
1284
1285        BUG_ON(p_blkno == 0);
1286
1287        for(i = 0; i < wc->w_num_pages; i++) {
1288                int tmpret;
1289
1290                tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc,
1291                                                      wc->w_pages[i], cpos,
1292                                                      user_pos, user_len,
1293                                                      should_zero);
1294                if (tmpret) {
1295                        mlog_errno(tmpret);
1296                        if (ret == 0)
1297                                ret = tmpret;
1298                }
1299        }
1300
1301        /*
1302         * We only have cleanup to do in case of allocating write.
1303         */
1304        if (ret && new)
1305                ocfs2_write_failure(inode, wc, user_pos, user_len);
1306
1307out:
1308
1309        return ret;
1310}
1311
1312static int ocfs2_write_cluster_by_desc(struct address_space *mapping,
1313                                       struct ocfs2_alloc_context *data_ac,
1314                                       struct ocfs2_alloc_context *meta_ac,
1315                                       struct ocfs2_write_ctxt *wc,
1316                                       loff_t pos, unsigned len)
1317{
1318        int ret, i;
1319        loff_t cluster_off;
1320        unsigned int local_len = len;
1321        struct ocfs2_write_cluster_desc *desc;
1322        struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb);
1323
1324        for (i = 0; i < wc->w_clen; i++) {
1325                desc = &wc->w_desc[i];
1326
1327                /*
1328                 * We have to make sure that the total write passed in
1329                 * doesn't extend past a single cluster.
1330                 */
1331                local_len = len;
1332                cluster_off = pos & (osb->s_clustersize - 1);
1333                if ((cluster_off + local_len) > osb->s_clustersize)
1334                        local_len = osb->s_clustersize - cluster_off;
1335
1336                ret = ocfs2_write_cluster(mapping, desc->c_phys,
1337                                          desc->c_unwritten,
1338                                          desc->c_needs_zero,
1339                                          data_ac, meta_ac,
1340                                          wc, desc->c_cpos, pos, local_len);
1341                if (ret) {
1342                        mlog_errno(ret);
1343                        goto out;
1344                }
1345
1346                len -= local_len;
1347                pos += local_len;
1348        }
1349
1350        ret = 0;
1351out:
1352        return ret;
1353}
1354
1355/*
1356 * ocfs2_write_end() wants to know which parts of the target page it
1357 * should complete the write on. It's easiest to compute them ahead of
1358 * time when a more complete view of the write is available.
1359 */
1360static void ocfs2_set_target_boundaries(struct ocfs2_super *osb,
1361                                        struct ocfs2_write_ctxt *wc,
1362                                        loff_t pos, unsigned len, int alloc)
1363{
1364        struct ocfs2_write_cluster_desc *desc;
1365
1366        wc->w_target_from = pos & (PAGE_CACHE_SIZE - 1);
1367        wc->w_target_to = wc->w_target_from + len;
1368
1369        if (alloc == 0)
1370                return;
1371
1372        /*
1373         * Allocating write - we may have different boundaries based
1374         * on page size and cluster size.
1375         *
1376         * NOTE: We can no longer compute one value from the other as
1377         * the actual write length and user provided length may be
1378         * different.
1379         */
1380
1381        if (wc->w_large_pages) {
1382                /*
1383                 * We only care about the 1st and last cluster within
1384                 * our range and whether they should be zero'd or not. Either
1385                 * value may be extended out to the start/end of a
1386                 * newly allocated cluster.
1387                 */
1388                desc = &wc->w_desc[0];
1389                if (desc->c_needs_zero)
1390                        ocfs2_figure_cluster_boundaries(osb,
1391                                                        desc->c_cpos,
1392                                                        &wc->w_target_from,
1393                                                        NULL);
1394
1395                desc = &wc->w_desc[wc->w_clen - 1];
1396                if (desc->c_needs_zero)
1397                        ocfs2_figure_cluster_boundaries(osb,
1398                                                        desc->c_cpos,
1399                                                        NULL,
1400                                                        &wc->w_target_to);
1401        } else {
1402                wc->w_target_from = 0;
1403                wc->w_target_to = PAGE_CACHE_SIZE;
1404        }
1405}
1406
1407/*
1408 * Populate each single-cluster write descriptor in the write context
1409 * with information about the i/o to be done.
1410 *
1411 * Returns the number of clusters that will have to be allocated, as
1412 * well as a worst case estimate of the number of extent records that
1413 * would have to be created during a write to an unwritten region.
1414 */
1415static int ocfs2_populate_write_desc(struct inode *inode,
1416                                     struct ocfs2_write_ctxt *wc,
1417                                     unsigned int *clusters_to_alloc,
1418                                     unsigned int *extents_to_split)
1419{
1420        int ret;
1421        struct ocfs2_write_cluster_desc *desc;
1422        unsigned int num_clusters = 0;
1423        unsigned int ext_flags = 0;
1424        u32 phys = 0;
1425        int i;
1426
1427        *clusters_to_alloc = 0;
1428        *extents_to_split = 0;
1429
1430        for (i = 0; i < wc->w_clen; i++) {
1431                desc = &wc->w_desc[i];
1432                desc->c_cpos = wc->w_cpos + i;
1433
1434                if (num_clusters == 0) {
1435                        /*
1436                         * Need to look up the next extent record.
1437                         */
1438                        ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys,
1439                                                 &num_clusters, &ext_flags);
1440                        if (ret) {
1441                                mlog_errno(ret);
1442                                goto out;
1443                        }
1444
1445                        /* We should already CoW the refcountd extent. */
1446                        BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1447
1448                        /*
1449                         * Assume worst case - that we're writing in
1450                         * the middle of the extent.
1451                         *
1452                         * We can assume that the write proceeds from
1453                         * left to right, in which case the extent
1454                         * insert code is smart enough to coalesce the
1455                         * next splits into the previous records created.
1456                         */
1457                        if (ext_flags & OCFS2_EXT_UNWRITTEN)
1458                                *extents_to_split = *extents_to_split + 2;
1459                } else if (phys) {
1460                        /*
1461                         * Only increment phys if it doesn't describe
1462                         * a hole.
1463                         */
1464                        phys++;
1465                }
1466
1467                /*
1468                 * If w_first_new_cpos is < UINT_MAX, we have a non-sparse
1469                 * file that got extended.  w_first_new_cpos tells us
1470                 * where the newly allocated clusters are so we can
1471                 * zero them.
1472                 */
1473                if (desc->c_cpos >= wc->w_first_new_cpos) {
1474                        BUG_ON(phys == 0);
1475                        desc->c_needs_zero = 1;
1476                }
1477
1478                desc->c_phys = phys;
1479                if (phys == 0) {
1480                        desc->c_new = 1;
1481                        desc->c_needs_zero = 1;
1482                        *clusters_to_alloc = *clusters_to_alloc + 1;
1483                }
1484
1485                if (ext_flags & OCFS2_EXT_UNWRITTEN) {
1486                        desc->c_unwritten = 1;
1487                        desc->c_needs_zero = 1;
1488                }
1489
1490                num_clusters--;
1491        }
1492
1493        ret = 0;
1494out:
1495        return ret;
1496}
1497
1498static int ocfs2_write_begin_inline(struct address_space *mapping,
1499                                    struct inode *inode,
1500                                    struct ocfs2_write_ctxt *wc)
1501{
1502        int ret;
1503        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1504        struct page *page;
1505        handle_t *handle;
1506        struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1507
1508        page = find_or_create_page(mapping, 0, GFP_NOFS);
1509        if (!page) {
1510                ret = -ENOMEM;
1511                mlog_errno(ret);
1512                goto out;
1513        }
1514        /*
1515         * If we don't set w_num_pages then this page won't get unlocked
1516         * and freed on cleanup of the write context.
1517         */
1518        wc->w_pages[0] = wc->w_target_page = page;
1519        wc->w_num_pages = 1;
1520
1521        handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1522        if (IS_ERR(handle)) {
1523                ret = PTR_ERR(handle);
1524                mlog_errno(ret);
1525                goto out;
1526        }
1527
1528        ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
1529                                      OCFS2_JOURNAL_ACCESS_WRITE);
1530        if (ret) {
1531                ocfs2_commit_trans(osb, handle);
1532
1533                mlog_errno(ret);
1534                goto out;
1535        }
1536
1537        if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1538                ocfs2_set_inode_data_inline(inode, di);
1539
1540        if (!PageUptodate(page)) {
1541                ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh);
1542                if (ret) {
1543                        ocfs2_commit_trans(osb, handle);
1544
1545                        goto out;
1546                }
1547        }
1548
1549        wc->w_handle = handle;
1550out:
1551        return ret;
1552}
1553
1554int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size)
1555{
1556        struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1557
1558        if (new_size <= le16_to_cpu(di->id2.i_data.id_count))
1559                return 1;
1560        return 0;
1561}
1562
1563static int ocfs2_try_to_write_inline_data(struct address_space *mapping,
1564                                          struct inode *inode, loff_t pos,
1565                                          unsigned len, struct page *mmap_page,
1566                                          struct ocfs2_write_ctxt *wc)
1567{
1568        int ret, written = 0;
1569        loff_t end = pos + len;
1570        struct ocfs2_inode_info *oi = OCFS2_I(inode);
1571        struct ocfs2_dinode *di = NULL;
1572
1573        trace_ocfs2_try_to_write_inline_data((unsigned long long)oi->ip_blkno,
1574                                             len, (unsigned long long)pos,
1575                                             oi->ip_dyn_features);
1576
1577        /*
1578         * Handle inodes which already have inline data 1st.
1579         */
1580        if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1581                if (mmap_page == NULL &&
1582                    ocfs2_size_fits_inline_data(wc->w_di_bh, end))
1583                        goto do_inline_write;
1584
1585                /*
1586                 * The write won't fit - we have to give this inode an
1587                 * inline extent list now.
1588                 */
1589                ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh);
1590                if (ret)
1591                        mlog_errno(ret);
1592                goto out;
1593        }
1594
1595        /*
1596         * Check whether the inode can accept inline data.
1597         */
1598        if (oi->ip_clusters != 0 || i_size_read(inode) != 0)
1599                return 0;
1600
1601        /*
1602         * Check whether the write can fit.
1603         */
1604        di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1605        if (mmap_page ||
1606            end > ocfs2_max_inline_data_with_xattr(inode->i_sb, di))
1607                return 0;
1608
1609do_inline_write:
1610        ret = ocfs2_write_begin_inline(mapping, inode, wc);
1611        if (ret) {
1612                mlog_errno(ret);
1613                goto out;
1614        }
1615
1616        /*
1617         * This signals to the caller that the data can be written
1618         * inline.
1619         */
1620        written = 1;
1621out:
1622        return written ? written : ret;
1623}
1624
1625/*
1626 * This function only does anything for file systems which can't
1627 * handle sparse files.
1628 *
1629 * What we want to do here is fill in any hole between the current end
1630 * of allocation and the end of our write. That way the rest of the
1631 * write path can treat it as an non-allocating write, which has no
1632 * special case code for sparse/nonsparse files.
1633 */
1634static int ocfs2_expand_nonsparse_inode(struct inode *inode,
1635                                        struct buffer_head *di_bh,
1636                                        loff_t pos, unsigned len,
1637                                        struct ocfs2_write_ctxt *wc)
1638{
1639        int ret;
1640        loff_t newsize = pos + len;
1641
1642        BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
1643
1644        if (newsize <= i_size_read(inode))
1645                return 0;
1646
1647        ret = ocfs2_extend_no_holes(inode, di_bh, newsize, pos);
1648        if (ret)
1649                mlog_errno(ret);
1650
1651        wc->w_first_new_cpos =
1652                ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode));
1653
1654        return ret;
1655}
1656
1657static int ocfs2_zero_tail(struct inode *inode, struct buffer_head *di_bh,
1658                           loff_t pos)
1659{
1660        int ret = 0;
1661
1662        BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
1663        if (pos > i_size_read(inode))
1664                ret = ocfs2_zero_extend(inode, di_bh, pos);
1665
1666        return ret;
1667}
1668
1669/*
1670 * Try to flush truncate logs if we can free enough clusters from it.
1671 * As for return value, "< 0" means error, "0" no space and "1" means
1672 * we have freed enough spaces and let the caller try to allocate again.
1673 */
1674static int ocfs2_try_to_free_truncate_log(struct ocfs2_super *osb,
1675                                          unsigned int needed)
1676{
1677        tid_t target;
1678        int ret = 0;
1679        unsigned int truncated_clusters;
1680
1681        mutex_lock(&osb->osb_tl_inode->i_mutex);
1682        truncated_clusters = osb->truncated_clusters;
1683        mutex_unlock(&osb->osb_tl_inode->i_mutex);
1684
1685        /*
1686         * Check whether we can succeed in allocating if we free
1687         * the truncate log.
1688         */
1689        if (truncated_clusters < needed)
1690                goto out;
1691
1692        ret = ocfs2_flush_truncate_log(osb);
1693        if (ret) {
1694                mlog_errno(ret);
1695                goto out;
1696        }
1697
1698        if (jbd2_journal_start_commit(osb->journal->j_journal, &target)) {
1699                jbd2_log_wait_commit(osb->journal->j_journal, target);
1700                ret = 1;
1701        }
1702out:
1703        return ret;
1704}
1705
1706int ocfs2_write_begin_nolock(struct file *filp,
1707                             struct address_space *mapping,
1708                             loff_t pos, unsigned len, unsigned flags,
1709                             struct page **pagep, void **fsdata,
1710                             struct buffer_head *di_bh, struct page *mmap_page)
1711{
1712        int ret, cluster_of_pages, credits = OCFS2_INODE_UPDATE_CREDITS;
1713        unsigned int clusters_to_alloc, extents_to_split, clusters_need = 0;
1714        struct ocfs2_write_ctxt *wc;
1715        struct inode *inode = mapping->host;
1716        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1717        struct ocfs2_dinode *di;
1718        struct ocfs2_alloc_context *data_ac = NULL;
1719        struct ocfs2_alloc_context *meta_ac = NULL;
1720        handle_t *handle;
1721        struct ocfs2_extent_tree et;
1722        int try_free = 1, ret1;
1723
1724try_again:
1725        ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, di_bh);
1726        if (ret) {
1727                mlog_errno(ret);
1728                return ret;
1729        }
1730
1731        if (ocfs2_supports_inline_data(osb)) {
1732                ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len,
1733                                                     mmap_page, wc);
1734                if (ret == 1) {
1735                        ret = 0;
1736                        goto success;
1737                }
1738                if (ret < 0) {
1739                        mlog_errno(ret);
1740                        goto out;
1741                }
1742        }
1743
1744        if (ocfs2_sparse_alloc(osb))
1745                ret = ocfs2_zero_tail(inode, di_bh, pos);
1746        else
1747                ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos, len,
1748                                                   wc);
1749        if (ret) {
1750                mlog_errno(ret);
1751                goto out;
1752        }
1753
1754        ret = ocfs2_check_range_for_refcount(inode, pos, len);
1755        if (ret < 0) {
1756                mlog_errno(ret);
1757                goto out;
1758        } else if (ret == 1) {
1759                clusters_need = wc->w_clen;
1760                ret = ocfs2_refcount_cow(inode, di_bh,
1761                                         wc->w_cpos, wc->w_clen, UINT_MAX);
1762                if (ret) {
1763                        mlog_errno(ret);
1764                        goto out;
1765                }
1766        }
1767
1768        ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc,
1769                                        &extents_to_split);
1770        if (ret) {
1771                mlog_errno(ret);
1772                goto out;
1773        }
1774        clusters_need += clusters_to_alloc;
1775
1776        di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1777
1778        trace_ocfs2_write_begin_nolock(
1779                        (unsigned long long)OCFS2_I(inode)->ip_blkno,
1780                        (long long)i_size_read(inode),
1781                        le32_to_cpu(di->i_clusters),
1782                        pos, len, flags, mmap_page,
1783                        clusters_to_alloc, extents_to_split);
1784
1785        /*
1786         * We set w_target_from, w_target_to here so that
1787         * ocfs2_write_end() knows which range in the target page to
1788         * write out. An allocation requires that we write the entire
1789         * cluster range.
1790         */
1791        if (clusters_to_alloc || extents_to_split) {
1792                /*
1793                 * XXX: We are stretching the limits of
1794                 * ocfs2_lock_allocators(). It greatly over-estimates
1795                 * the work to be done.
1796                 */
1797                ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
1798                                              wc->w_di_bh);
1799                ret = ocfs2_lock_allocators(inode, &et,
1800                                            clusters_to_alloc, extents_to_split,
1801                                            &data_ac, &meta_ac);
1802                if (ret) {
1803                        mlog_errno(ret);
1804                        goto out;
1805                }
1806
1807                if (data_ac)
1808                        data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv;
1809
1810                credits = ocfs2_calc_extend_credits(inode->i_sb,
1811                                                    &di->id2.i_list,
1812                                                    clusters_to_alloc);
1813
1814        }
1815
1816        /*
1817         * We have to zero sparse allocated clusters, unwritten extent clusters,
1818         * and non-sparse clusters we just extended.  For non-sparse writes,
1819         * we know zeros will only be needed in the first and/or last cluster.
1820         */
1821        if (clusters_to_alloc || extents_to_split ||
1822            (wc->w_clen && (wc->w_desc[0].c_needs_zero ||
1823                            wc->w_desc[wc->w_clen - 1].c_needs_zero)))
1824                cluster_of_pages = 1;
1825        else
1826                cluster_of_pages = 0;
1827
1828        ocfs2_set_target_boundaries(osb, wc, pos, len, cluster_of_pages);
1829
1830        handle = ocfs2_start_trans(osb, credits);
1831        if (IS_ERR(handle)) {
1832                ret = PTR_ERR(handle);
1833                mlog_errno(ret);
1834                goto out;
1835        }
1836
1837        wc->w_handle = handle;
1838
1839        if (clusters_to_alloc) {
1840                ret = dquot_alloc_space_nodirty(inode,
1841                        ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
1842                if (ret)
1843                        goto out_commit;
1844        }
1845        /*
1846         * We don't want this to fail in ocfs2_write_end(), so do it
1847         * here.
1848         */
1849        ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
1850                                      OCFS2_JOURNAL_ACCESS_WRITE);
1851        if (ret) {
1852                mlog_errno(ret);
1853                goto out_quota;
1854        }
1855
1856        /*
1857         * Fill our page array first. That way we've grabbed enough so
1858         * that we can zero and flush if we error after adding the
1859         * extent.
1860         */
1861        ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos, len,
1862                                         cluster_of_pages, mmap_page);
1863        if (ret && ret != -EAGAIN) {
1864                mlog_errno(ret);
1865                goto out_quota;
1866        }
1867
1868        /*
1869         * ocfs2_grab_pages_for_write() returns -EAGAIN if it could not lock
1870         * the target page. In this case, we exit with no error and no target
1871         * page. This will trigger the caller, page_mkwrite(), to re-try
1872         * the operation.
1873         */
1874        if (ret == -EAGAIN) {
1875                BUG_ON(wc->w_target_page);
1876                ret = 0;
1877                goto out_quota;
1878        }
1879
1880        ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos,
1881                                          len);
1882        if (ret) {
1883                mlog_errno(ret);
1884                goto out_quota;
1885        }
1886
1887        if (data_ac)
1888                ocfs2_free_alloc_context(data_ac);
1889        if (meta_ac)
1890                ocfs2_free_alloc_context(meta_ac);
1891
1892success:
1893        *pagep = wc->w_target_page;
1894        *fsdata = wc;
1895        return 0;
1896out_quota:
1897        if (clusters_to_alloc)
1898                dquot_free_space(inode,
1899                          ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
1900out_commit:
1901        ocfs2_commit_trans(osb, handle);
1902
1903out:
1904        ocfs2_free_write_ctxt(wc);
1905
1906        if (data_ac)
1907                ocfs2_free_alloc_context(data_ac);
1908        if (meta_ac)
1909                ocfs2_free_alloc_context(meta_ac);
1910
1911        if (ret == -ENOSPC && try_free) {
1912                /*
1913                 * Try to free some truncate log so that we can have enough
1914                 * clusters to allocate.
1915                 */
1916                try_free = 0;
1917
1918                ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need);
1919                if (ret1 == 1)
1920                        goto try_again;
1921
1922                if (ret1 < 0)
1923                        mlog_errno(ret1);
1924        }
1925
1926        return ret;
1927}
1928
1929static int ocfs2_write_begin(struct file *file, struct address_space *mapping,
1930                             loff_t pos, unsigned len, unsigned flags,
1931                             struct page **pagep, void **fsdata)
1932{
1933        int ret;
1934        struct buffer_head *di_bh = NULL;
1935        struct inode *inode = mapping->host;
1936
1937        ret = ocfs2_inode_lock(inode, &di_bh, 1);
1938        if (ret) {
1939                mlog_errno(ret);
1940                return ret;
1941        }
1942
1943        /*
1944         * Take alloc sem here to prevent concurrent lookups. That way
1945         * the mapping, zeroing and tree manipulation within
1946         * ocfs2_write() will be safe against ->readpage(). This
1947         * should also serve to lock out allocation from a shared
1948         * writeable region.
1949         */
1950        down_write(&OCFS2_I(inode)->ip_alloc_sem);
1951
1952        ret = ocfs2_write_begin_nolock(file, mapping, pos, len, flags, pagep,
1953                                       fsdata, di_bh, NULL);
1954        if (ret) {
1955                mlog_errno(ret);
1956                goto out_fail;
1957        }
1958
1959        brelse(di_bh);
1960
1961        return 0;
1962
1963out_fail:
1964        up_write(&OCFS2_I(inode)->ip_alloc_sem);
1965
1966        brelse(di_bh);
1967        ocfs2_inode_unlock(inode, 1);
1968
1969        return ret;
1970}
1971
1972static void ocfs2_write_end_inline(struct inode *inode, loff_t pos,
1973                                   unsigned len, unsigned *copied,
1974                                   struct ocfs2_dinode *di,
1975                                   struct ocfs2_write_ctxt *wc)
1976{
1977        void *kaddr;
1978
1979        if (unlikely(*copied < len)) {
1980                if (!PageUptodate(wc->w_target_page)) {
1981                        *copied = 0;
1982                        return;
1983                }
1984        }
1985
1986        kaddr = kmap_atomic(wc->w_target_page);
1987        memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied);
1988        kunmap_atomic(kaddr);
1989
1990        trace_ocfs2_write_end_inline(
1991             (unsigned long long)OCFS2_I(inode)->ip_blkno,
1992             (unsigned long long)pos, *copied,
1993             le16_to_cpu(di->id2.i_data.id_count),
1994             le16_to_cpu(di->i_dyn_features));
1995}
1996
1997int ocfs2_write_end_nolock(struct address_space *mapping,
1998                           loff_t pos, unsigned len, unsigned copied,
1999                           struct page *page, void *fsdata)
2000{
2001        int i;
2002        unsigned from, to, start = pos & (PAGE_CACHE_SIZE - 1);
2003        struct inode *inode = mapping->host;
2004        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2005        struct ocfs2_write_ctxt *wc = fsdata;
2006        struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
2007        handle_t *handle = wc->w_handle;
2008        struct page *tmppage;
2009
2010        if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2011                ocfs2_write_end_inline(inode, pos, len, &copied, di, wc);
2012                goto out_write_size;
2013        }
2014
2015        if (unlikely(copied < len)) {
2016                if (!PageUptodate(wc->w_target_page))
2017                        copied = 0;
2018
2019                ocfs2_zero_new_buffers(wc->w_target_page, start+copied,
2020                                       start+len);
2021        }
2022        flush_dcache_page(wc->w_target_page);
2023
2024        for(i = 0; i < wc->w_num_pages; i++) {
2025                tmppage = wc->w_pages[i];
2026
2027                if (tmppage == wc->w_target_page) {
2028                        from = wc->w_target_from;
2029                        to = wc->w_target_to;
2030
2031                        BUG_ON(from > PAGE_CACHE_SIZE ||
2032                               to > PAGE_CACHE_SIZE ||
2033                               to < from);
2034                } else {
2035                        /*
2036                         * Pages adjacent to the target (if any) imply
2037                         * a hole-filling write in which case we want
2038                         * to flush their entire range.
2039                         */
2040                        from = 0;
2041                        to = PAGE_CACHE_SIZE;
2042                }
2043
2044                if (page_has_buffers(tmppage)) {
2045                        if (ocfs2_should_order_data(inode))
2046                                ocfs2_jbd2_file_inode(wc->w_handle, inode);
2047                        block_commit_write(tmppage, from, to);
2048                }
2049        }
2050
2051out_write_size:
2052        pos += copied;
2053        if (pos > inode->i_size) {
2054                i_size_write(inode, pos);
2055                mark_inode_dirty(inode);
2056        }
2057        inode->i_blocks = ocfs2_inode_sector_count(inode);
2058        di->i_size = cpu_to_le64((u64)i_size_read(inode));
2059        inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2060        di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
2061        di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
2062        ocfs2_journal_dirty(handle, wc->w_di_bh);
2063
2064        ocfs2_commit_trans(osb, handle);
2065
2066        ocfs2_run_deallocs(osb, &wc->w_dealloc);
2067
2068        ocfs2_free_write_ctxt(wc);
2069
2070        return copied;
2071}
2072
2073static int ocfs2_write_end(struct file *file, struct address_space *mapping,
2074                           loff_t pos, unsigned len, unsigned copied,
2075                           struct page *page, void *fsdata)
2076{
2077        int ret;
2078        struct inode *inode = mapping->host;
2079
2080        ret = ocfs2_write_end_nolock(mapping, pos, len, copied, page, fsdata);
2081
2082        up_write(&OCFS2_I(inode)->ip_alloc_sem);
2083        ocfs2_inode_unlock(inode, 1);
2084
2085        return ret;
2086}
2087
2088const struct address_space_operations ocfs2_aops = {
2089        .readpage               = ocfs2_readpage,
2090        .readpages              = ocfs2_readpages,
2091        .writepage              = ocfs2_writepage,
2092        .write_begin            = ocfs2_write_begin,
2093        .write_end              = ocfs2_write_end,
2094        .bmap                   = ocfs2_bmap,
2095        .direct_IO              = ocfs2_direct_IO,
2096        .invalidatepage         = ocfs2_invalidatepage,
2097        .releasepage            = ocfs2_releasepage,
2098        .migratepage            = buffer_migrate_page,
2099        .is_partially_uptodate  = block_is_partially_uptodate,
2100        .error_remove_page      = generic_error_remove_page,
2101};
2102
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.