SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/gfs2/ops_address.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2005 Red Hat, Inc.  All rights reserved.
   4  *
   5  * This copyrighted material is made available to anyone wishing to use,
   6  * modify, copy, or redistribute it subject to the terms and conditions
   7  * of the GNU General Public License v.2.
   8  */
   9
  10 #include <linux/sched.h>
  11 #include <linux/slab.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/completion.h>
  14 #include <linux/buffer_head.h>
  15 #include <linux/pagemap.h>
  16 #include <linux/mpage.h>
  17 #include <linux/fs.h>
  18 #include <asm/semaphore.h>
  19
  20 #include "gfs2.h"
  21 #include "bmap.h"
  22 #include "glock.h"
  23 #include "inode.h"
  24 #include "log.h"
  25 #include "meta_io.h"
  26 #include "ops_address.h"
  27 #include "page.h"
  28 #include "quota.h"
  29 #include "trans.h"
  30 #include "rgrp.h"
  31 #include "ops_file.h"
  32
  33 /**
  34  * gfs2_get_block - Fills in a buffer head with details about a block
  35  * @inode: The inode
  36  * @lblock: The block number to look up
  37  * @bh_result: The buffer head to return the result in
  38  * @create: Non-zero if we may add block to the file
  39  *
  40  * Returns: errno
  41  */
  42
  43 int gfs2_get_block(struct inode *inode, sector_t lblock,
  44                    struct buffer_head *bh_result, int create)
  45 {
  46         struct gfs2_inode *ip = get_v2ip(inode);
  47         int new = create;
  48         uint64_t dblock;
  49         int error;
  50
  51         error = gfs2_block_map(ip, lblock, &new, &dblock, NULL);
  52         if (error)
  53                 return error;
  54
  55         if (!dblock)
  56                 return 0;
  57
  58         map_bh(bh_result, inode->i_sb, dblock);
  59         if (new)
  60                 set_buffer_new(bh_result);
  61
  62         return 0;
  63 }
  64
  65 /**
  66  * get_block_noalloc - Fills in a buffer head with details about a block
  67  * @inode: The inode
  68  * @lblock: The block number to look up
  69  * @bh_result: The buffer head to return the result in
  70  * @create: Non-zero if we may add block to the file
  71  *
  72  * Returns: errno
  73  */
  74
  75 static int get_block_noalloc(struct inode *inode, sector_t lblock,
  76                              struct buffer_head *bh_result, int create)
  77 {
  78         struct gfs2_inode *ip = get_v2ip(inode);
  79         int new = 0;
  80         uint64_t dblock;
  81         int error;
  82
  83         error = gfs2_block_map(ip, lblock, &new, &dblock, NULL);
  84         if (error)
  85                 return error;
  86
  87         if (dblock)
  88                 map_bh(bh_result, inode->i_sb, dblock);
  89         else if (gfs2_assert_withdraw(ip->i_sbd, !create))
  90                 error = -EIO;
  91
  92         return error;
  93 }
  94
  95 static int get_blocks(struct inode *inode, sector_t lblock,
  96                       unsigned long max_blocks, struct buffer_head *bh_result,
  97                       int create)
  98 {
  99         struct gfs2_inode *ip = get_v2ip(inode);
 100         int new = create;
 101         uint64_t dblock;
 102         uint32_t extlen;
 103         int error;
 104
 105         error = gfs2_block_map(ip, lblock, &new, &dblock, &extlen);
 106         if (error)
 107                 return error;
 108
 109         if (!dblock)
 110                 return 0;
 111
 112         map_bh(bh_result, inode->i_sb, dblock);
 113         if (new)
 114                 set_buffer_new(bh_result);
 115
 116         if (extlen > max_blocks)
 117                 extlen = max_blocks;
 118         bh_result->b_size = extlen << inode->i_blkbits;
 119
 120         return 0;
 121 }
 122
 123 static int get_blocks_noalloc(struct inode *inode, sector_t lblock,
 124                               unsigned long max_blocks,
 125                               struct buffer_head *bh_result, int create)
 126 {
 127         struct gfs2_inode *ip = get_v2ip(inode);
 128         int new = 0;
 129         uint64_t dblock;
 130         uint32_t extlen;
 131         int error;
 132
 133         error = gfs2_block_map(ip, lblock, &new, &dblock, &extlen);
 134         if (error)
 135                 return error;
 136
 137         if (dblock) {
 138                 map_bh(bh_result, inode->i_sb, dblock);
 139                 if (extlen > max_blocks)
 140                         extlen = max_blocks;
 141                 bh_result->b_size = extlen << inode->i_blkbits;
 142         } else if (gfs2_assert_withdraw(ip->i_sbd, !create))
 143                 error = -EIO;
 144
 145         return error;
 146 }
 147
 148 /**
 149  * gfs2_writepage - Write complete page
 150  * @page: Page to write
 151  *
 152  * Returns: errno
 153  *
 154  * Some of this is copied from block_write_full_page() although we still
 155  * call it to do most of the work.
 156  */
 157
 158 static int gfs2_writepage(struct page *page, struct writeback_control *wbc)
 159 {
 160         struct inode *inode = page->mapping->host;
 161         struct gfs2_inode *ip = get_v2ip(page->mapping->host);
 162         struct gfs2_sbd *sdp = ip->i_sbd;
 163         loff_t i_size = i_size_read(inode);
 164         pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
 165         unsigned offset;
 166         int error;
 167         int done_trans = 0;
 168
 169         if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) {
 170                 unlock_page(page);
 171                 return -EIO;
 172         }
 173         if (get_transaction)
 174                 goto out_ignore;
 175
 176         /* Is the page fully outside i_size? (truncate in progress) */
 177         offset = i_size & (PAGE_CACHE_SIZE-1);
 178         if (page->index >= end_index+1 || !offset) {
 179                 page->mapping->a_ops->invalidatepage(page, 0);
 180                 unlock_page(page);
 181                 return 0; /* don't care */
 182         }
 183
 184         if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip)) {
 185                 error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
 186                 if (error)
 187                         goto out_ignore;
 188                 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
 189                 done_trans = 1;
 190         }
 191         error = block_write_full_page(page, get_block_noalloc, wbc);
 192         if (done_trans)
 193                 gfs2_trans_end(sdp);
 194         gfs2_meta_cache_flush(ip);
 195         return error;
 196
 197 out_ignore:
 198         redirty_page_for_writepage(wbc, page);
 199         unlock_page(page);
 200         return 0;
 201 }
 202
 203 /**
 204  * stuffed_readpage - Fill in a Linux page with stuffed file data
 205  * @ip: the inode
 206  * @page: the page
 207  *
 208  * Returns: errno
 209  */
 210
 211 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
 212 {
 213         struct buffer_head *dibh;
 214         void *kaddr;
 215         int error;
 216
 217         error = gfs2_meta_inode_buffer(ip, &dibh);
 218         if (error)
 219                 return error;
 220
 221         kaddr = kmap_atomic(page, KM_USER0);
 222         memcpy((char *)kaddr,
 223                dibh->b_data + sizeof(struct gfs2_dinode),
 224                ip->i_di.di_size);
 225         memset((char *)kaddr + ip->i_di.di_size,
 226                0,
 227                PAGE_CACHE_SIZE - ip->i_di.di_size);
 228         kunmap_atomic(page, KM_USER0);
 229
 230         brelse(dibh);
 231
 232         SetPageUptodate(page);
 233
 234         return 0;
 235 }
 236
 237 static int zero_readpage(struct page *page)
 238 {
 239         void *kaddr;
 240
 241         kaddr = kmap_atomic(page, KM_USER0);
 242         memset(kaddr, 0, PAGE_CACHE_SIZE);
 243         kunmap_atomic(page, KM_USER0);
 244
 245         SetPageUptodate(page);
 246         unlock_page(page);
 247
 248         return 0;
 249 }
 250
 251 /**
 252  * gfs2_readpage - readpage with locking
 253  * @file: The file to read a page for. N.B. This may be NULL if we are
 254  * reading an internal file.
 255  * @page: The page to read
 256  *
 257  * Returns: errno
 258  */
 259
 260 static int gfs2_readpage(struct file *file, struct page *page)
 261 {
 262         struct gfs2_inode *ip = get_v2ip(page->mapping->host);
 263         struct gfs2_sbd *sdp = ip->i_sbd;
 264         struct gfs2_holder gh;
 265         int error;
 266
 267         if (file != &gfs2_internal_file_sentinal) {
 268                 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
 269                 error = gfs2_glock_nq_m_atime(1, &gh);
 270                 if (error)
 271                         goto out_unlock;
 272         }
 273
 274         if (gfs2_is_stuffed(ip)) {
 275                 if (!page->index) {
 276                         error = stuffed_readpage(ip, page);
 277                         unlock_page(page);
 278                 } else
 279                         error = zero_readpage(page);
 280         } else
 281                 error = mpage_readpage(page, gfs2_get_block);
 282
 283         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 284                 error = -EIO;
 285
 286         if (file != &gfs2_internal_file_sentinal) {
 287                 gfs2_glock_dq_m(1, &gh);
 288                 gfs2_holder_uninit(&gh);
 289         }
 290 out:
 291         return error;
 292 out_unlock:
 293         unlock_page(page);
 294         goto out;
 295 }
 296
 297 /**
 298  * gfs2_prepare_write - Prepare to write a page to a file
 299  * @file: The file to write to
 300  * @page: The page which is to be prepared for writing
 301  * @from: From (byte range within page)
 302  * @to: To (byte range within page)
 303  *
 304  * Returns: errno
 305  */
 306
 307 static int gfs2_prepare_write(struct file *file, struct page *page,
 308                               unsigned from, unsigned to)
 309 {
 310         struct gfs2_inode *ip = get_v2ip(page->mapping->host);
 311         struct gfs2_sbd *sdp = ip->i_sbd;
 312         unsigned int data_blocks, ind_blocks, rblocks;
 313         int alloc_required;
 314         int error = 0;
 315         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + from;
 316         loff_t end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
 317         struct gfs2_alloc *al;
 318
 319         gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME, &ip->i_gh);
 320         error = gfs2_glock_nq_m_atime(1, &ip->i_gh);
 321         if (error)
 322                 goto out_uninit;
 323
 324         gfs2_write_calc_reserv(ip, to - from, &data_blocks, &ind_blocks);
 325
 326         error = gfs2_write_alloc_required(ip, pos, from - to, &alloc_required);
 327         if (error)
 328                 goto out_unlock;
 329
 330
 331         if (alloc_required) {
 332                 al = gfs2_alloc_get(ip);
 333
 334                 error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
 335                 if (error)
 336                         goto out_alloc_put;
 337
 338                 error = gfs2_quota_check(ip, ip->i_di.di_uid, ip->i_di.di_gid);
 339                 if (error)
 340                         goto out_qunlock;
 341
 342                 al->al_requested = data_blocks + ind_blocks;
 343                 error = gfs2_inplace_reserve(ip);
 344                 if (error)
 345                         goto out_qunlock;
 346         }
 347
 348         rblocks = RES_DINODE + ind_blocks;
 349         if (gfs2_is_jdata(ip))
 350                 rblocks += data_blocks ? data_blocks : 1;
 351         if (ind_blocks || data_blocks)
 352                 rblocks += RES_STATFS + RES_QUOTA;
 353
 354         error = gfs2_trans_begin(sdp, rblocks, 0);
 355         if (error)
 356                 goto out;
 357
 358         if (gfs2_is_stuffed(ip)) {
 359                 if (end > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
 360                         error = gfs2_unstuff_dinode(ip, gfs2_unstuffer_page,
 361                                                     page);
 362                         if (error == 0)
 363                                 goto prepare_write;
 364                 } else if (!PageUptodate(page))
 365                         error = stuffed_readpage(ip, page);
 366                 goto out;
 367         }
 368
 369 prepare_write:
 370         error = block_prepare_write(page, from, to, gfs2_get_block);
 371
 372 out:
 373         if (error) {
 374                 gfs2_trans_end(sdp);
 375                 if (alloc_required) {
 376                         gfs2_inplace_release(ip);
 377 out_qunlock:
 378                         gfs2_quota_unlock(ip);
 379 out_alloc_put:
 380                         gfs2_alloc_put(ip);
 381                 }
 382 out_unlock:
 383                 gfs2_glock_dq_m(1, &ip->i_gh);
 384 out_uninit:
 385                 gfs2_holder_uninit(&ip->i_gh);
 386         }
 387
 388         return error;
 389 }
 390
 391 /**
 392  * gfs2_commit_write - Commit write to a file
 393  * @file: The file to write to
 394  * @page: The page containing the data
 395  * @from: From (byte range within page)
 396  * @to: To (byte range within page)
 397  *
 398  * Returns: errno
 399  */
 400
 401 static int gfs2_commit_write(struct file *file, struct page *page,
 402                              unsigned from, unsigned to)
 403 {
 404         struct inode *inode = page->mapping->host;
 405         struct gfs2_inode *ip = get_v2ip(inode);
 406         struct gfs2_sbd *sdp = ip->i_sbd;
 407         int error = -EOPNOTSUPP;
 408         struct buffer_head *dibh;
 409         struct gfs2_alloc *al = &ip->i_alloc;;
 410
 411         if (gfs2_assert_withdraw(sdp, gfs2_glock_is_locked_by_me(ip->i_gl)))
 412                 goto fail_nounlock;
 413
 414         error = gfs2_meta_inode_buffer(ip, &dibh);
 415         if (error)
 416                 goto fail_endtrans;
 417
 418         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 419
 420         if (gfs2_is_stuffed(ip)) {
 421                 uint64_t file_size;
 422                 void *kaddr;
 423
 424                 file_size = ((uint64_t)page->index << PAGE_CACHE_SHIFT) + to;
 425
 426                 kaddr = kmap_atomic(page, KM_USER0);
 427                 memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from,
 428                        (char *)kaddr + from, to - from);
 429                 kunmap_atomic(page, KM_USER0);
 430
 431                 SetPageUptodate(page);
 432
 433                 if (inode->i_size < file_size)
 434                         i_size_write(inode, file_size);
 435         } else {
 436                 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED ||
 437                     gfs2_is_jdata(ip))
 438                         gfs2_page_add_databufs(ip, page, from, to);
 439                 error = generic_commit_write(file, page, from, to);
 440                 if (error)
 441                         goto fail;
 442         }
 443
 444         if (ip->i_di.di_size < inode->i_size)
 445                 ip->i_di.di_size = inode->i_size;
 446
 447         gfs2_dinode_out(&ip->i_di, dibh->b_data);
 448         brelse(dibh);
 449         gfs2_trans_end(sdp);
 450         if (al->al_requested) {
 451                 gfs2_inplace_release(ip);
 452                 gfs2_quota_unlock(ip);
 453                 gfs2_alloc_put(ip);
 454         }
 455         gfs2_glock_dq_m(1, &ip->i_gh);
 456         gfs2_holder_uninit(&ip->i_gh);
 457         return 0;
 458
 459 fail:
 460         brelse(dibh);
 461 fail_endtrans:
 462         gfs2_trans_end(sdp);
 463         if (al->al_requested) {
 464                 gfs2_inplace_release(ip);
 465                 gfs2_quota_unlock(ip);
 466                 gfs2_alloc_put(ip);
 467         }
 468         gfs2_glock_dq_m(1, &ip->i_gh);
 469         gfs2_holder_uninit(&ip->i_gh);
 470 fail_nounlock:
 471         ClearPageUptodate(page);
 472         return error;
 473 }
 474
 475 /**
 476  * gfs2_bmap - Block map function
 477  * @mapping: Address space info
 478  * @lblock: The block to map
 479  *
 480  * Returns: The disk address for the block or 0 on hole or error
 481  */
 482
 483 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
 484 {
 485         struct gfs2_inode *ip = get_v2ip(mapping->host);
 486         struct gfs2_holder i_gh;
 487         sector_t dblock = 0;
 488         int error;
 489
 490         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
 491         if (error)
 492                 return 0;
 493
 494         if (!gfs2_is_stuffed(ip))
 495                 dblock = generic_block_bmap(mapping, lblock, gfs2_get_block);
 496
 497         gfs2_glock_dq_uninit(&i_gh);
 498
 499         return dblock;
 500 }
 501
 502 static void discard_buffer(struct gfs2_sbd *sdp, struct buffer_head *bh)
 503 {
 504         struct gfs2_bufdata *bd;
 505
 506         gfs2_log_lock(sdp);
 507         bd = get_v2bd(bh);
 508         if (bd) {
 509                 bd->bd_bh = NULL;
 510                 set_v2bd(bh, NULL);
 511                 gfs2_log_unlock(sdp);
 512                 brelse(bh);
 513         } else
 514                 gfs2_log_unlock(sdp);
 515
 516         lock_buffer(bh);
 517         clear_buffer_dirty(bh);
 518         bh->b_bdev = NULL;
 519         clear_buffer_mapped(bh);
 520         clear_buffer_req(bh);
 521         clear_buffer_new(bh);
 522         clear_buffer_delay(bh);
 523         unlock_buffer(bh);
 524 }
 525
 526 static int gfs2_invalidatepage(struct page *page, unsigned long offset)
 527 {
 528         struct gfs2_sbd *sdp = get_v2sdp(page->mapping->host->i_sb);
 529         struct buffer_head *head, *bh, *next;
 530         unsigned int curr_off = 0;
 531         int ret = 1;
 532
 533         BUG_ON(!PageLocked(page));
 534         if (!page_has_buffers(page))
 535                 return 1;
 536
 537         bh = head = page_buffers(page);
 538         do {
 539                 unsigned int next_off = curr_off + bh->b_size;
 540                 next = bh->b_this_page;
 541
 542                 if (offset <= curr_off)
 543                         discard_buffer(sdp, bh);
 544
 545                 curr_off = next_off;
 546                 bh = next;
 547         } while (bh != head);
 548
 549         if (!offset)
 550                 ret = try_to_release_page(page, 0);
 551
 552         return ret;
 553 }
 554
 555 static ssize_t gfs2_direct_IO_write(struct kiocb *iocb, const struct iovec *iov,
 556                                     loff_t offset, unsigned long nr_segs)
 557 {
 558         struct file *file = iocb->ki_filp;
 559         struct inode *inode = file->f_mapping->host;
 560         struct gfs2_inode *ip = get_v2ip(inode);
 561         struct gfs2_holder gh;
 562         int rv;
 563
 564         /*
 565          * Shared lock, even though its write, since we do no allocation
 566          * on this path. All we need change is atime.
 567          */
 568         gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
 569         rv = gfs2_glock_nq_m_atime(1, &gh);
 570         if (rv)
 571                 goto out;
 572
 573         /*
 574          * Should we return an error here? I can't see that O_DIRECT for
 575          * a journaled file makes any sense. For now we'll silently fall
 576          * back to buffered I/O, likewise we do the same for stuffed
 577          * files since they are (a) small and (b) unaligned.
 578          */
 579         if (gfs2_is_jdata(ip))
 580                 goto out;
 581
 582         if (gfs2_is_stuffed(ip))
 583                 goto out;
 584
 585         rv = __blockdev_direct_IO(WRITE, iocb, inode, inode->i_sb->s_bdev,
 586                                   iov, offset, nr_segs, get_blocks_noalloc,
 587                                   NULL, DIO_OWN_LOCKING);
 588 out:
 589         gfs2_glock_dq_m(1, &gh);
 590         gfs2_holder_uninit(&gh);
 591
 592         return rv;
 593 }
 594
 595 /**
 596  * gfs2_direct_IO
 597  *
 598  * This is called with a shared lock already held for the read path.
 599  * Currently, no locks are held when the write path is called.
 600  */
 601 static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
 602                               const struct iovec *iov, loff_t offset,
 603                               unsigned long nr_segs)
 604 {
 605         struct file *file = iocb->ki_filp;
 606         struct inode *inode = file->f_mapping->host;
 607         struct gfs2_inode *ip = get_v2ip(inode);
 608         struct gfs2_sbd *sdp = ip->i_sbd;
 609
 610         if (rw == WRITE)
 611                 return gfs2_direct_IO_write(iocb, iov, offset, nr_segs);
 612
 613         if (gfs2_assert_warn(sdp, gfs2_glock_is_locked_by_me(ip->i_gl)) ||
 614             gfs2_assert_warn(sdp, !gfs2_is_stuffed(ip)))
 615                 return -EINVAL;
 616
 617         return __blockdev_direct_IO(READ, iocb, inode, inode->i_sb->s_bdev, iov,
 618                                     offset, nr_segs, get_blocks, NULL,
 619                                     DIO_OWN_LOCKING);
 620 }
 621
 622 struct address_space_operations gfs2_file_aops = {
 623         .writepage = gfs2_writepage,
 624         .readpage = gfs2_readpage,
 625         .sync_page = block_sync_page,
 626         .prepare_write = gfs2_prepare_write,
 627         .commit_write = gfs2_commit_write,
 628         .bmap = gfs2_bmap,
 629         .invalidatepage = gfs2_invalidatepage,
 630         .direct_IO = gfs2_direct_IO,
 631 };
 632