fs/gfs2/bmap.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2006 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 version 2.
   8  */
   9
  10 #include <linux/slab.h>
  11 #include <linux/spinlock.h>
  12 #include <linux/completion.h>
  13 #include <linux/buffer_head.h>
  14 #include <linux/gfs2_ondisk.h>
  15 #include <linux/crc32.h>
  16 #include <linux/lm_interface.h>
  17
  18 #include "gfs2.h"
  19 #include "incore.h"
  20 #include "bmap.h"
  21 #include "glock.h"
  22 #include "inode.h"
  23 #include "meta_io.h"
  24 #include "quota.h"
  25 #include "rgrp.h"
  26 #include "trans.h"
  27 #include "dir.h"
  28 #include "util.h"
  29 #include "ops_address.h"
  30
  31 /* This doesn't need to be that large as max 64 bit pointers in a 4k
  32  * block is 512, so __u16 is fine for that. It saves stack space to
  33  * keep it small.
  34  */
  35 struct metapath {
  36         __u16 mp_list[GFS2_MAX_META_HEIGHT];
  37 };
  38
  39 typedef int (*block_call_t) (struct gfs2_inode *ip, struct buffer_head *dibh,
  40                              struct buffer_head *bh, __be64 *top,
  41                              __be64 *bottom, unsigned int height,
  42                              void *data);
  43
  44 struct strip_mine {
  45         int sm_first;
  46         unsigned int sm_height;
  47 };
  48
  49 /**
  50  * gfs2_unstuffer_page - unstuff a stuffed inode into a block cached by a page
  51  * @ip: the inode
  52  * @dibh: the dinode buffer
  53  * @block: the block number that was allocated
  54  * @private: any locked page held by the caller process
  55  *
  56  * Returns: errno
  57  */
  58
  59 static int gfs2_unstuffer_page(struct gfs2_inode *ip, struct buffer_head *dibh,
  60                                u64 block, struct page *page)
  61 {
  62         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
  63         struct inode *inode = &ip->i_inode;
  64         struct buffer_head *bh;
  65         int release = 0;
  66
  67         if (!page || page->index) {
  68                 page = grab_cache_page(inode->i_mapping, 0);
  69                 if (!page)
  70                         return -ENOMEM;
  71                 release = 1;
  72         }
  73
  74         if (!PageUptodate(page)) {
  75                 void *kaddr = kmap(page);
  76
  77                 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
  78                        ip->i_di.di_size);
  79                 memset(kaddr + ip->i_di.di_size, 0,
  80                        PAGE_CACHE_SIZE - ip->i_di.di_size);
  81                 kunmap(page);
  82
  83                 SetPageUptodate(page);
  84         }
  85
  86         if (!page_has_buffers(page))
  87                 create_empty_buffers(page, 1 << inode->i_blkbits,
  88                                      (1 << BH_Uptodate));
  89
  90         bh = page_buffers(page);
  91
  92         if (!buffer_mapped(bh))
  93                 map_bh(bh, inode->i_sb, block);
  94
  95         set_buffer_uptodate(bh);
  96         if (!gfs2_is_jdata(ip))
  97                 mark_buffer_dirty(bh);
  98         if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip))
  99                 gfs2_trans_add_bh(ip->i_gl, bh, 0);
 100
 101         if (release) {
 102                 unlock_page(page);
 103                 page_cache_release(page);
 104         }
 105
 106         return 0;
 107 }
 108
 109 /**
 110  * gfs2_unstuff_dinode - Unstuff a dinode when the data has grown too big
 111  * @ip: The GFS2 inode to unstuff
 112  * @unstuffer: the routine that handles unstuffing a non-zero length file
 113  * @private: private data for the unstuffer
 114  *
 115  * This routine unstuffs a dinode and returns it to a "normal" state such
 116  * that the height can be grown in the traditional way.
 117  *
 118  * Returns: errno
 119  */
 120
 121 int gfs2_unstuff_dinode(struct gfs2_inode *ip, struct page *page)
 122 {
 123         struct buffer_head *bh, *dibh;
 124         struct gfs2_dinode *di;
 125         u64 block = 0;
 126         int isdir = gfs2_is_dir(ip);
 127         int error;
 128
 129         down_write(&ip->i_rw_mutex);
 130
 131         error = gfs2_meta_inode_buffer(ip, &dibh);
 132         if (error)
 133                 goto out;
 134
 135         if (ip->i_di.di_size) {
 136                 /* Get a free block, fill it with the stuffed data,
 137                    and write it out to disk */
 138
 139                 if (isdir) {
 140                         block = gfs2_alloc_meta(ip);
 141
 142                         error = gfs2_dir_get_new_buffer(ip, block, &bh);
 143                         if (error)
 144                                 goto out_brelse;
 145                         gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_meta_header),
 146                                               dibh, sizeof(struct gfs2_dinode));
 147                         brelse(bh);
 148                 } else {
 149                         block = gfs2_alloc_data(ip);
 150
 151                         error = gfs2_unstuffer_page(ip, dibh, block, page);
 152                         if (error)
 153                                 goto out_brelse;
 154                 }
 155         }
 156
 157         /*  Set up the pointer to the new block  */
 158
 159         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 160         di = (struct gfs2_dinode *)dibh->b_data;
 161         gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
 162
 163         if (ip->i_di.di_size) {
 164                 *(__be64 *)(di + 1) = cpu_to_be64(block);
 165                 ip->i_di.di_blocks++;
 166                 gfs2_set_inode_blocks(&ip->i_inode);
 167                 di->di_blocks = cpu_to_be64(ip->i_di.di_blocks);
 168         }
 169
 170         ip->i_di.di_height = 1;
 171         di->di_height = cpu_to_be16(1);
 172
 173 out_brelse:
 174         brelse(dibh);
 175 out:
 176         up_write(&ip->i_rw_mutex);
 177         return error;
 178 }
 179
 180 /**
 181  * calc_tree_height - Calculate the height of a metadata tree
 182  * @ip: The GFS2 inode
 183  * @size: The proposed size of the file
 184  *
 185  * Work out how tall a metadata tree needs to be in order to accommodate a
 186  * file of a particular size. If size is less than the current size of
 187  * the inode, then the current size of the inode is used instead of the
 188  * supplied one.
 189  *
 190  * Returns: the height the tree should be
 191  */
 192
 193 static unsigned int calc_tree_height(struct gfs2_inode *ip, u64 size)
 194 {
 195         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 196         u64 *arr;
 197         unsigned int max, height;
 198
 199         if (ip->i_di.di_size > size)
 200                 size = ip->i_di.di_size;
 201
 202         if (gfs2_is_dir(ip)) {
 203                 arr = sdp->sd_jheightsize;
 204                 max = sdp->sd_max_jheight;
 205         } else {
 206                 arr = sdp->sd_heightsize;
 207                 max = sdp->sd_max_height;
 208         }
 209
 210         for (height = 0; height < max; height++)
 211                 if (arr[height] >= size)
 212                         break;
 213
 214         return height;
 215 }
 216
 217 /**
 218  * build_height - Build a metadata tree of the requested height
 219  * @ip: The GFS2 inode
 220  * @height: The height to build to
 221  *
 222  *
 223  * Returns: errno
 224  */
 225
 226 static int build_height(struct inode *inode, unsigned height)
 227 {
 228         struct gfs2_inode *ip = GFS2_I(inode);
 229         unsigned new_height = height - ip->i_di.di_height;
 230         struct buffer_head *dibh;
 231         struct buffer_head *blocks[GFS2_MAX_META_HEIGHT];
 232         struct gfs2_dinode *di;
 233         int error;
 234         __be64 *bp;
 235         u64 bn;
 236         unsigned n;
 237
 238         if (height <= ip->i_di.di_height)
 239                 return 0;
 240
 241         error = gfs2_meta_inode_buffer(ip, &dibh);
 242         if (error)
 243                 return error;
 244
 245         for(n = 0; n < new_height; n++) {
 246                 bn = gfs2_alloc_meta(ip);
 247                 blocks[n] = gfs2_meta_new(ip->i_gl, bn);
 248                 gfs2_trans_add_bh(ip->i_gl, blocks[n], 1);
 249         }
 250
 251         n = 0;
 252         bn = blocks[0]->b_blocknr;
 253         if (new_height > 1) {
 254                 for(; n < new_height-1; n++) {
 255                         gfs2_metatype_set(blocks[n], GFS2_METATYPE_IN,
 256                                           GFS2_FORMAT_IN);
 257                         gfs2_buffer_clear_tail(blocks[n],
 258                                                sizeof(struct gfs2_meta_header));
 259                         bp = (__be64 *)(blocks[n]->b_data +
 260                                      sizeof(struct gfs2_meta_header));
 261                         *bp = cpu_to_be64(blocks[n+1]->b_blocknr);
 262                         brelse(blocks[n]);
 263                         blocks[n] = NULL;
 264                 }
 265         }
 266         gfs2_metatype_set(blocks[n], GFS2_METATYPE_IN, GFS2_FORMAT_IN);
 267         gfs2_buffer_copy_tail(blocks[n], sizeof(struct gfs2_meta_header),
 268                               dibh, sizeof(struct gfs2_dinode));
 269         brelse(blocks[n]);
 270         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 271         di = (struct gfs2_dinode *)dibh->b_data;
 272         gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
 273         *(__be64 *)(di + 1) = cpu_to_be64(bn);
 274         ip->i_di.di_height += new_height;
 275         ip->i_di.di_blocks += new_height;
 276         gfs2_set_inode_blocks(&ip->i_inode);
 277         di->di_height = cpu_to_be16(ip->i_di.di_height);
 278         di->di_blocks = cpu_to_be64(ip->i_di.di_blocks);
 279         brelse(dibh);
 280         return error;
 281 }
 282
 283 /**
 284  * find_metapath - Find path through the metadata tree
 285  * @ip: The inode pointer
 286  * @mp: The metapath to return the result in
 287  * @block: The disk block to look up
 288  *
 289  *   This routine returns a struct metapath structure that defines a path
 290  *   through the metadata of inode "ip" to get to block "block".
 291  *
 292  *   Example:
 293  *   Given:  "ip" is a height 3 file, "offset" is 101342453, and this is a
 294  *   filesystem with a blocksize of 4096.
 295  *
 296  *   find_metapath() would return a struct metapath structure set to:
 297  *   mp_offset = 101342453, mp_height = 3, mp_list[0] = 0, mp_list[1] = 48,
 298  *   and mp_list[2] = 165.
 299  *
 300  *   That means that in order to get to the block containing the byte at
 301  *   offset 101342453, we would load the indirect block pointed to by pointer
 302  *   0 in the dinode.  We would then load the indirect block pointed to by
 303  *   pointer 48 in that indirect block.  We would then load the data block
 304  *   pointed to by pointer 165 in that indirect block.
 305  *
 306  *             ----------------------------------------
 307  *             | Dinode |                             |
 308  *             |        |                            4|
 309  *             |        |0 1 2 3 4 5                 9|
 310  *             |        |                            6|
 311  *             ----------------------------------------
 312  *                       |
 313  *                       |
 314  *                       V
 315  *             ----------------------------------------
 316  *             | Indirect Block                       |
 317  *             |                                     5|
 318  *             |            4 4 4 4 4 5 5            1|
 319  *             |0           5 6 7 8 9 0 1            2|
 320  *             ----------------------------------------
 321  *                                |
 322  *                                |
 323  *                                V
 324  *             ----------------------------------------
 325  *             | Indirect Block                       |
 326  *             |                         1 1 1 1 1   5|
 327  *             |                         6 6 6 6 6   1|
 328  *             |0                        3 4 5 6 7   2|
 329  *             ----------------------------------------
 330  *                                           |
 331  *                                           |
 332  *                                           V
 333  *             ----------------------------------------
 334  *             | Data block containing offset         |
 335  *             |            101342453                 |
 336  *             |                                      |
 337  *             |                                      |
 338  *             ----------------------------------------
 339  *
 340  */
 341
 342 static void find_metapath(struct gfs2_inode *ip, u64 block,
 343                           struct metapath *mp)
 344 {
 345         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 346         u64 b = block;
 347         unsigned int i;
 348
 349         for (i = ip->i_di.di_height; i--;)
 350                 mp->mp_list[i] = do_div(b, sdp->sd_inptrs);
 351
 352 }
 353
 354 /**
 355  * metapointer - Return pointer to start of metadata in a buffer
 356  * @bh: The buffer
 357  * @height: The metadata height (0 = dinode)
 358  * @mp: The metapath
 359  *
 360  * Return a pointer to the block number of the next height of the metadata
 361  * tree given a buffer containing the pointer to the current height of the
 362  * metadata tree.
 363  */
 364
 365 static inline __be64 *metapointer(struct buffer_head *bh, int *boundary,
 366                                unsigned int height, const struct metapath *mp)
 367 {
 368         unsigned int head_size = (height > 0) ?
 369                 sizeof(struct gfs2_meta_header) : sizeof(struct gfs2_dinode);
 370         __be64 *ptr;
 371         *boundary = 0;
 372         ptr = ((__be64 *)(bh->b_data + head_size)) + mp->mp_list[height];
 373         if (ptr + 1 == (__be64 *)(bh->b_data + bh->b_size))
 374                 *boundary = 1;
 375         return ptr;
 376 }
 377
 378 /**
 379  * lookup_block - Get the next metadata block in metadata tree
 380  * @ip: The GFS2 inode
 381  * @bh: Buffer containing the pointers to metadata blocks
 382  * @height: The height of the tree (0 = dinode)
 383  * @mp: The metapath
 384  * @create: Non-zero if we may create a new meatdata block
 385  * @new: Used to indicate if we did create a new metadata block
 386  * @block: the returned disk block number
 387  *
 388  * Given a metatree, complete to a particular height, checks to see if the next
 389  * height of the tree exists. If not the next height of the tree is created.
 390  * The block number of the next height of the metadata tree is returned.
 391  *
 392  */
 393
 394 static int lookup_block(struct gfs2_inode *ip, struct buffer_head *bh,
 395                         unsigned int height, struct metapath *mp, int create,
 396                         int *new, u64 *block)
 397 {
 398         int boundary;
 399         __be64 *ptr = metapointer(bh, &boundary, height, mp);
 400
 401         if (*ptr) {
 402                 *block = be64_to_cpu(*ptr);
 403                 return boundary;
 404         }
 405
 406         *block = 0;
 407
 408         if (!create)
 409                 return 0;
 410
 411         if (height == ip->i_di.di_height - 1 && !gfs2_is_dir(ip))
 412                 *block = gfs2_alloc_data(ip);
 413         else
 414                 *block = gfs2_alloc_meta(ip);
 415
 416         gfs2_trans_add_bh(ip->i_gl, bh, 1);
 417
 418         *ptr = cpu_to_be64(*block);
 419         ip->i_di.di_blocks++;
 420         gfs2_set_inode_blocks(&ip->i_inode);
 421
 422         *new = 1;
 423         return 0;
 424 }
 425
 426 static inline void bmap_lock(struct inode *inode, int create)
 427 {
 428         struct gfs2_inode *ip = GFS2_I(inode);
 429         if (create)
 430                 down_write(&ip->i_rw_mutex);
 431         else
 432                 down_read(&ip->i_rw_mutex);
 433 }
 434
 435 static inline void bmap_unlock(struct inode *inode, int create)
 436 {
 437         struct gfs2_inode *ip = GFS2_I(inode);
 438         if (create)
 439                 up_write(&ip->i_rw_mutex);
 440         else
 441                 up_read(&ip->i_rw_mutex);
 442 }
 443
 444 /**
 445  * gfs2_block_map - Map a block from an inode to a disk block
 446  * @inode: The inode
 447  * @lblock: The logical block number
 448  * @bh_map: The bh to be mapped
 449  *
 450  * Find the block number on the current device which corresponds to an
 451  * inode's block. If the block had to be created, "new" will be set.
 452  *
 453  * Returns: errno
 454  */
 455
 456 int gfs2_block_map(struct inode *inode, u64 lblock, int create,
 457                    struct buffer_head *bh_map)
 458 {
 459         struct gfs2_inode *ip = GFS2_I(inode);
 460         struct gfs2_sbd *sdp = GFS2_SB(inode);
 461         struct buffer_head *bh;
 462         unsigned int bsize;
 463         unsigned int height;
 464         unsigned int end_of_metadata;
 465         unsigned int x;
 466         int error = 0;
 467         int new = 0;
 468         u64 dblock = 0;
 469         int boundary;
 470         unsigned int maxlen = bh_map->b_size >> inode->i_blkbits;
 471         struct metapath mp;
 472         u64 size;
 473
 474         BUG_ON(maxlen == 0);
 475
 476         if (gfs2_assert_warn(sdp, !gfs2_is_stuffed(ip)))
 477                 return 0;
 478
 479         bmap_lock(inode, create);
 480         clear_buffer_mapped(bh_map);
 481         clear_buffer_new(bh_map);
 482         clear_buffer_boundary(bh_map);
 483         bsize = gfs2_is_dir(ip) ? sdp->sd_jbsize : sdp->sd_sb.sb_bsize;
 484         size = (lblock + 1) * bsize;
 485
 486         if (size > ip->i_di.di_size) {
 487                 height = calc_tree_height(ip, size);
 488                 if (ip->i_di.di_height < height) {
 489                         if (!create)
 490                                 goto out_ok;
 491
 492                         error = build_height(inode, height);
 493                         if (error)
 494                                 goto out_fail;
 495                 }
 496         }
 497
 498         find_metapath(ip, lblock, &mp);
 499         end_of_metadata = ip->i_di.di_height - 1;
 500         error = gfs2_meta_inode_buffer(ip, &bh);
 501         if (error)
 502                 goto out_fail;
 503
 504         for (x = 0; x < end_of_metadata; x++) {
 505                 lookup_block(ip, bh, x, &mp, create, &new, &dblock);
 506                 brelse(bh);
 507                 if (!dblock)
 508                         goto out_ok;
 509
 510                 error = gfs2_meta_indirect_buffer(ip, x+1, dblock, new, &bh);
 511                 if (error)
 512                         goto out_fail;
 513         }
 514
 515         boundary = lookup_block(ip, bh, end_of_metadata, &mp, create, &new, &dblock);
 516         if (dblock) {
 517                 map_bh(bh_map, inode->i_sb, dblock);
 518                 if (boundary)
 519                         set_buffer_boundary(bh_map);
 520                 if (new) {
 521                         struct buffer_head *dibh;
 522                         error = gfs2_meta_inode_buffer(ip, &dibh);
 523                         if (!error) {
 524                                 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 525                                 gfs2_dinode_out(ip, dibh->b_data);
 526                                 brelse(dibh);
 527                         }
 528                         set_buffer_new(bh_map);
 529                         goto out_brelse;
 530                 }
 531                 while(--maxlen && !buffer_boundary(bh_map)) {
 532                         u64 eblock;
 533
 534                         mp.mp_list[end_of_metadata]++;
 535                         boundary = lookup_block(ip, bh, end_of_metadata, &mp, 0, &new, &eblock);
 536                         if (eblock != ++dblock)
 537                                 break;
 538                         bh_map->b_size += (1 << inode->i_blkbits);
 539                         if (boundary)
 540                                 set_buffer_boundary(bh_map);
 541                 }
 542         }
 543 out_brelse:
 544         brelse(bh);
 545 out_ok:
 546         error = 0;
 547 out_fail:
 548         bmap_unlock(inode, create);
 549         return error;
 550 }
 551
 552 int gfs2_extent_map(struct inode *inode, u64 lblock, int *new, u64 *dblock, unsigned *extlen)
 553 {
 554         struct buffer_head bh = { .b_state = 0, .b_blocknr = 0 };
 555         int ret;
 556         int create = *new;
 557
 558         BUG_ON(!extlen);
 559         BUG_ON(!dblock);
 560         BUG_ON(!new);
 561
 562         bh.b_size = 1 << (inode->i_blkbits + 5);
 563         ret = gfs2_block_map(inode, lblock, create, &bh);
 564         *extlen = bh.b_size >> inode->i_blkbits;
 565         *dblock = bh.b_blocknr;
 566         if (buffer_new(&bh))
 567                 *new = 1;
 568         else
 569                 *new = 0;
 570         return ret;
 571 }
 572
 573 /**
 574  * recursive_scan - recursively scan through the end of a file
 575  * @ip: the inode
 576  * @dibh: the dinode buffer
 577  * @mp: the path through the metadata to the point to start
 578  * @height: the height the recursion is at
 579  * @block: the indirect block to look at
 580  * @first: 1 if this is the first block
 581  * @bc: the call to make for each piece of metadata
 582  * @data: data opaque to this function to pass to @bc
 583  *
 584  * When this is first called @height and @block should be zero and
 585  * @first should be 1.
 586  *
 587  * Returns: errno
 588  */
 589
 590 static int recursive_scan(struct gfs2_inode *ip, struct buffer_head *dibh,
 591                           struct metapath *mp, unsigned int height,
 592                           u64 block, int first, block_call_t bc,
 593                           void *data)
 594 {
 595         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 596         struct buffer_head *bh = NULL;
 597         __be64 *top, *bottom;
 598         u64 bn;
 599         int error;
 600         int mh_size = sizeof(struct gfs2_meta_header);
 601
 602         if (!height) {
 603                 error = gfs2_meta_inode_buffer(ip, &bh);
 604                 if (error)
 605                         return error;
 606                 dibh = bh;
 607
 608                 top = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + mp->mp_list[0];
 609                 bottom = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + sdp->sd_diptrs;
 610         } else {
 611                 error = gfs2_meta_indirect_buffer(ip, height, block, 0, &bh);
 612                 if (error)
 613                         return error;
 614
 615                 top = (__be64 *)(bh->b_data + mh_size) +
 616                                   (first ? mp->mp_list[height] : 0);
 617
 618                 bottom = (__be64 *)(bh->b_data + mh_size) + sdp->sd_inptrs;
 619         }
 620
 621         error = bc(ip, dibh, bh, top, bottom, height, data);
 622         if (error)
 623                 goto out;
 624
 625         if (height < ip->i_di.di_height - 1)
 626                 for (; top < bottom; top++, first = 0) {
 627                         if (!*top)
 628                                 continue;
 629
 630                         bn = be64_to_cpu(*top);
 631
 632                         error = recursive_scan(ip, dibh, mp, height + 1, bn,
 633                                                first, bc, data);
 634                         if (error)
 635                                 break;
 636                 }
 637
 638 out:
 639         brelse(bh);
 640         return error;
 641 }
 642
 643 /**
 644  * do_strip - Look for a layer a particular layer of the file and strip it off
 645  * @ip: the inode
 646  * @dibh: the dinode buffer
 647  * @bh: A buffer of pointers
 648  * @top: The first pointer in the buffer
 649  * @bottom: One more than the last pointer
 650  * @height: the height this buffer is at
 651  * @data: a pointer to a struct strip_mine
 652  *
 653  * Returns: errno
 654  */
 655
 656 static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh,
 657                     struct buffer_head *bh, __be64 *top, __be64 *bottom,
 658                     unsigned int height, void *data)
 659 {
 660         struct strip_mine *sm = data;
 661         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 662         struct gfs2_rgrp_list rlist;
 663         u64 bn, bstart;
 664         u32 blen;
 665         __be64 *p;
 666         unsigned int rg_blocks = 0;
 667         int metadata;
 668         unsigned int revokes = 0;
 669         int x;
 670         int error;
 671
 672         if (!*top)
 673                 sm->sm_first = 0;
 674
 675         if (height != sm->sm_height)
 676                 return 0;
 677
 678         if (sm->sm_first) {
 679                 top++;
 680                 sm->sm_first = 0;
 681         }
 682
 683         metadata = (height != ip->i_di.di_height - 1);
 684         if (metadata)
 685                 revokes = (height) ? sdp->sd_inptrs : sdp->sd_diptrs;
 686
 687         error = gfs2_rindex_hold(sdp, &ip->i_alloc.al_ri_gh);
 688         if (error)
 689                 return error;
 690
 691         memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
 692         bstart = 0;
 693         blen = 0;
 694
 695         for (p = top; p < bottom; p++) {
 696                 if (!*p)
 697                         continue;
 698
 699                 bn = be64_to_cpu(*p);
 700
 701                 if (bstart + blen == bn)
 702                         blen++;
 703                 else {
 704                         if (bstart)
 705                                 gfs2_rlist_add(sdp, &rlist, bstart);
 706
 707                         bstart = bn;
 708                         blen = 1;
 709                 }
 710         }
 711
 712         if (bstart)
 713                 gfs2_rlist_add(sdp, &rlist, bstart);
 714         else
 715                 goto out; /* Nothing to do */
 716
 717         gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE, 0);
 718
 719         for (x = 0; x < rlist.rl_rgrps; x++) {
 720                 struct gfs2_rgrpd *rgd;
 721                 rgd = rlist.rl_ghs[x].gh_gl->gl_object;
 722                 rg_blocks += rgd->rd_length;
 723         }
 724
 725         error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
 726         if (error)
 727                 goto out_rlist;
 728
 729         error = gfs2_trans_begin(sdp, rg_blocks + RES_DINODE +
 730                                  RES_INDIRECT + RES_STATFS + RES_QUOTA,
 731                                  revokes);
 732         if (error)
 733                 goto out_rg_gunlock;
 734
 735         down_write(&ip->i_rw_mutex);
 736
 737         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 738         gfs2_trans_add_bh(ip->i_gl, bh, 1);
 739
 740         bstart = 0;
 741         blen = 0;
 742
 743         for (p = top; p < bottom; p++) {
 744                 if (!*p)
 745                         continue;
 746
 747                 bn = be64_to_cpu(*p);
 748
 749                 if (bstart + blen == bn)
 750                         blen++;
 751                 else {
 752                         if (bstart) {
 753                                 if (metadata)
 754                                         gfs2_free_meta(ip, bstart, blen);
 755                                 else
 756                                         gfs2_free_data(ip, bstart, blen);
 757                         }
 758
 759                         bstart = bn;
 760                         blen = 1;
 761                 }
 762
 763                 *p = 0;
 764                 if (!ip->i_di.di_blocks)
 765                         gfs2_consist_inode(ip);
 766                 ip->i_di.di_blocks--;
 767                 gfs2_set_inode_blocks(&ip->i_inode);
 768         }
 769         if (bstart) {
 770                 if (metadata)
 771                         gfs2_free_meta(ip, bstart, blen);
 772                 else
 773                         gfs2_free_data(ip, bstart, blen);
 774         }
 775
 776         ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
 777
 778         gfs2_dinode_out(ip, dibh->b_data);
 779
 780         up_write(&ip->i_rw_mutex);
 781
 782         gfs2_trans_end(sdp);
 783
 784 out_rg_gunlock:
 785         gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
 786 out_rlist:
 787         gfs2_rlist_free(&rlist);
 788 out:
 789         gfs2_glock_dq_uninit(&ip->i_alloc.al_ri_gh);
 790         return error;
 791 }
 792
 793 /**
 794  * do_grow - Make a file look bigger than it is
 795  * @ip: the inode
 796  * @size: the size to set the file to
 797  *
 798  * Called with an exclusive lock on @ip.
 799  *
 800  * Returns: errno
 801  */
 802
 803 static int do_grow(struct gfs2_inode *ip, u64 size)
 804 {
 805         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 806         struct gfs2_alloc *al;
 807         struct buffer_head *dibh;
 808         unsigned int h;
 809         int error;
 810
 811         al = gfs2_alloc_get(ip);
 812
 813         error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
 814         if (error)
 815                 goto out;
 816
 817         error = gfs2_quota_check(ip, ip->i_inode.i_uid, ip->i_inode.i_gid);
 818         if (error)
 819                 goto out_gunlock_q;
 820
 821         al->al_requested = sdp->sd_max_height + RES_DATA;
 822
 823         error = gfs2_inplace_reserve(ip);
 824         if (error)
 825                 goto out_gunlock_q;
 826
 827         error = gfs2_trans_begin(sdp,
 828                         sdp->sd_max_height + al->al_rgd->rd_length +
 829                         RES_JDATA + RES_DINODE + RES_STATFS + RES_QUOTA, 0);
 830         if (error)
 831                 goto out_ipres;
 832
 833         if (size > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
 834                 if (gfs2_is_stuffed(ip)) {
 835                         error = gfs2_unstuff_dinode(ip, NULL);
 836                         if (error)
 837                                 goto out_end_trans;
 838                 }
 839
 840                 h = calc_tree_height(ip, size);
 841                 if (ip->i_di.di_height < h) {
 842                         down_write(&ip->i_rw_mutex);
 843                         error = build_height(&ip->i_inode, h);
 844                         up_write(&ip->i_rw_mutex);
 845                         if (error)
 846                                 goto out_end_trans;
 847                 }
 848         }
 849
 850         ip->i_di.di_size = size;
 851         ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
 852
 853         error = gfs2_meta_inode_buffer(ip, &dibh);
 854         if (error)
 855                 goto out_end_trans;
 856
 857         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 858         gfs2_dinode_out(ip, dibh->b_data);
 859         brelse(dibh);
 860
 861 out_end_trans:
 862         gfs2_trans_end(sdp);
 863 out_ipres:
 864         gfs2_inplace_release(ip);
 865 out_gunlock_q:
 866         gfs2_quota_unlock(ip);
 867 out:
 868         gfs2_alloc_put(ip);
 869         return error;
 870 }
 871
 872
 873 /**
 874  * gfs2_block_truncate_page - Deal with zeroing out data for truncate
 875  *
 876  * This is partly borrowed from ext3.
 877  */
 878 static int gfs2_block_truncate_page(struct address_space *mapping)
 879 {
 880         struct inode *inode = mapping->host;
 881         struct gfs2_inode *ip = GFS2_I(inode);
 882         struct gfs2_sbd *sdp = GFS2_SB(inode);
 883         loff_t from = inode->i_size;
 884         unsigned long index = from >> PAGE_CACHE_SHIFT;
 885         unsigned offset = from & (PAGE_CACHE_SIZE-1);
 886         unsigned blocksize, iblock, length, pos;
 887         struct buffer_head *bh;
 888         struct page *page;
 889         int err;
 890
 891         page = grab_cache_page(mapping, index);
 892         if (!page)
 893                 return 0;
 894
 895         blocksize = inode->i_sb->s_blocksize;
 896         length = blocksize - (offset & (blocksize - 1));
 897         iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
 898
 899         if (!page_has_buffers(page))
 900                 create_empty_buffers(page, blocksize, 0);
 901
 902         /* Find the buffer that contains "offset" */
 903         bh = page_buffers(page);
 904         pos = blocksize;
 905         while (offset >= pos) {
 906                 bh = bh->b_this_page;
 907                 iblock++;
 908                 pos += blocksize;
 909         }
 910
 911         err = 0;
 912
 913         if (!buffer_mapped(bh)) {
 914                 gfs2_get_block(inode, iblock, bh, 0);
 915                 /* unmapped? It's a hole - nothing to do */
 916                 if (!buffer_mapped(bh))
 917                         goto unlock;
 918         }
 919
 920         /* Ok, it's mapped. Make sure it's up-to-date */
 921         if (PageUptodate(page))
 922                 set_buffer_uptodate(bh);
 923
 924         if (!buffer_uptodate(bh)) {
 925                 err = -EIO;
 926                 ll_rw_block(READ, 1, &bh);
 927                 wait_on_buffer(bh);
 928                 /* Uhhuh. Read error. Complain and punt. */
 929                 if (!buffer_uptodate(bh))
 930                         goto unlock;
 931                 err = 0;
 932         }
 933
 934         if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip))
 935                 gfs2_trans_add_bh(ip->i_gl, bh, 0);
 936
 937         zero_user_page(page, offset, length, KM_USER0);
 938
 939 unlock:
 940         unlock_page(page);
 941         page_cache_release(page);
 942         return err;
 943 }
 944
 945 static int trunc_start(struct gfs2_inode *ip, u64 size)
 946 {
 947         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 948         struct buffer_head *dibh;
 949         int journaled = gfs2_is_jdata(ip);
 950         int error;
 951
 952         error = gfs2_trans_begin(sdp,
 953                                  RES_DINODE + (journaled ? RES_JDATA : 0), 0);
 954         if (error)
 955                 return error;
 956
 957         error = gfs2_meta_inode_buffer(ip, &dibh);
 958         if (error)
 959                 goto out;
 960
 961         if (gfs2_is_stuffed(ip)) {
 962                 ip->i_di.di_size = size;
 963                 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
 964                 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 965                 gfs2_dinode_out(ip, dibh->b_data);
 966                 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + size);
 967                 error = 1;
 968
 969         } else {
 970                 if (size & (u64)(sdp->sd_sb.sb_bsize - 1))
 971                         error = gfs2_block_truncate_page(ip->i_inode.i_mapping);
 972
 973                 if (!error) {
 974                         ip->i_di.di_size = size;
 975                         ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
 976                         ip->i_di.di_flags |= GFS2_DIF_TRUNC_IN_PROG;
 977                         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 978                         gfs2_dinode_out(ip, dibh->b_data);
 979                 }
 980         }
 981
 982         brelse(dibh);
 983
 984 out:
 985         gfs2_trans_end(sdp);
 986         return error;
 987 }
 988
 989 static int trunc_dealloc(struct gfs2_inode *ip, u64 size)
 990 {
 991         unsigned int height = ip->i_di.di_height;
 992         u64 lblock;
 993         struct metapath mp;
 994         int error;
 995
 996         if (!size)
 997                 lblock = 0;
 998         else
 999                 lblock = (size - 1) >> GFS2_SB(&ip->i_inode)->sd_sb.sb_bsize_shift;
1000
1001         find_metapath(ip, lblock, &mp);
1002         gfs2_alloc_get(ip);
1003
1004         error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1005         if (error)
1006                 goto out;
1007
1008         while (height--) {
1009                 struct strip_mine sm;
1010                 sm.sm_first = !!size;
1011                 sm.sm_height = height;
1012
1013                 error = recursive_scan(ip, NULL, &mp, 0, 0, 1, do_strip, &sm);
1014                 if (error)
1015                         break;
1016         }
1017
1018         gfs2_quota_unhold(ip);
1019
1020 out:
1021         gfs2_alloc_put(ip);
1022         return error;
1023 }
1024
1025 static int trunc_end(struct gfs2_inode *ip)
1026 {
1027         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1028         struct buffer_head *dibh;
1029         int error;
1030
1031         error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1032         if (error)
1033                 return error;
1034
1035         down_write(&ip->i_rw_mutex);
1036
1037         error = gfs2_meta_inode_buffer(ip, &dibh);
1038         if (error)
1039                 goto out;
1040
1041         if (!ip->i_di.di_size) {
1042                 ip->i_di.di_height = 0;
1043                 ip->i_di.di_goal_meta =
1044                         ip->i_di.di_goal_data =
1045                         ip->i_no_addr;
1046                 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
1047         }
1048         ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1049         ip->i_di.di_flags &= ~GFS2_DIF_TRUNC_IN_PROG;
1050
1051         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1052         gfs2_dinode_out(ip, dibh->b_data);
1053         brelse(dibh);
1054
1055 out:
1056         up_write(&ip->i_rw_mutex);
1057         gfs2_trans_end(sdp);
1058         return error;
1059 }
1060
1061 /**
1062  * do_shrink - make a file smaller
1063  * @ip: the inode
1064  * @size: the size to make the file
1065  * @truncator: function to truncate the last partial block
1066  *
1067  * Called with an exclusive lock on @ip.
1068  *
1069  * Returns: errno
1070  */
1071
1072 static int do_shrink(struct gfs2_inode *ip, u64 size)
1073 {
1074         int error;
1075
1076         error = trunc_start(ip, size);
1077         if (error < 0)
1078                 return error;
1079         if (error > 0)
1080                 return 0;
1081
1082         error = trunc_dealloc(ip, size);
1083         if (!error)
1084                 error = trunc_end(ip);
1085
1086         return error;
1087 }
1088
1089 static int do_touch(struct gfs2_inode *ip, u64 size)
1090 {
1091         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1092         struct buffer_head *dibh;
1093         int error;
1094
1095         error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1096         if (error)
1097                 return error;
1098
1099         down_write(&ip->i_rw_mutex);
1100
1101         error = gfs2_meta_inode_buffer(ip, &dibh);
1102         if (error)
1103                 goto do_touch_out;
1104
1105         ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1106         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1107         gfs2_dinode_out(ip, dibh->b_data);
1108         brelse(dibh);
1109
1110 do_touch_out:
1111         up_write(&ip->i_rw_mutex);
1112         gfs2_trans_end(sdp);
1113         return error;
1114 }
1115
1116 /**
1117  * gfs2_truncatei - make a file a given size
1118  * @ip: the inode
1119  * @size: the size to make the file
1120  * @truncator: function to truncate the last partial block
1121  *
1122  * The file size can grow, shrink, or stay the same size.
1123  *
1124  * Returns: errno
1125  */
1126
1127 int gfs2_truncatei(struct gfs2_inode *ip, u64 size)
1128 {
1129         int error;
1130
1131         if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), S_ISREG(ip->i_inode.i_mode)))
1132                 return -EINVAL;
1133
1134         if (size > ip->i_di.di_size)
1135                 error = do_grow(ip, size);
1136         else if (size < ip->i_di.di_size)
1137                 error = do_shrink(ip, size);
1138         else
1139                 /* update time stamps */
1140                 error = do_touch(ip, size);
1141
1142         return error;
1143 }
1144
1145 int gfs2_truncatei_resume(struct gfs2_inode *ip)
1146 {
1147         int error;
1148         error = trunc_dealloc(ip, ip->i_di.di_size);
1149         if (!error)
1150                 error = trunc_end(ip);
1151         return error;
1152 }
1153
1154 int gfs2_file_dealloc(struct gfs2_inode *ip)
1155 {
1156         return trunc_dealloc(ip, 0);
1157 }
1158
1159 /**
1160  * gfs2_write_calc_reserv - calculate number of blocks needed to write to a file
1161  * @ip: the file
1162  * @len: the number of bytes to be written to the file
1163  * @data_blocks: returns the number of data blocks required
1164  * @ind_blocks: returns the number of indirect blocks required
1165  *
1166  */
1167
1168 void gfs2_write_calc_reserv(struct gfs2_inode *ip, unsigned int len,
1169                             unsigned int *data_blocks, unsigned int *ind_blocks)
1170 {
1171         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1172         unsigned int tmp;
1173
1174         if (gfs2_is_dir(ip)) {
1175                 *data_blocks = DIV_ROUND_UP(len, sdp->sd_jbsize) + 2;
1176                 *ind_blocks = 3 * (sdp->sd_max_jheight - 1);
1177         } else {
1178                 *data_blocks = (len >> sdp->sd_sb.sb_bsize_shift) + 3;
1179                 *ind_blocks = 3 * (sdp->sd_max_height - 1);
1180         }
1181
1182         for (tmp = *data_blocks; tmp > sdp->sd_diptrs;) {
1183                 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
1184                 *ind_blocks += tmp;
1185         }
1186 }
1187
1188 /**
1189  * gfs2_write_alloc_required - figure out if a write will require an allocation
1190  * @ip: the file being written to
1191  * @offset: the offset to write to
1192  * @len: the number of bytes being written
1193  * @alloc_required: set to 1 if an alloc is required, 0 otherwise
1194  *
1195  * Returns: errno
1196  */
1197
1198 int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset,
1199                               unsigned int len, int *alloc_required)
1200 {
1201         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1202         u64 lblock, lblock_stop, dblock;
1203         u32 extlen;
1204         int new = 0;
1205         int error = 0;
1206
1207         *alloc_required = 0;
1208
1209         if (!len)
1210                 return 0;
1211
1212         if (gfs2_is_stuffed(ip)) {
1213                 if (offset + len >
1214                     sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
1215                         *alloc_required = 1;
1216                 return 0;
1217         }
1218
1219         if (gfs2_is_dir(ip)) {
1220                 unsigned int bsize = sdp->sd_jbsize;
1221                 lblock = offset;
1222                 do_div(lblock, bsize);
1223                 lblock_stop = offset + len + bsize - 1;
1224                 do_div(lblock_stop, bsize);
1225         } else {
1226                 unsigned int shift = sdp->sd_sb.sb_bsize_shift;
1227                 lblock = offset >> shift;
1228                 lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift;
1229         }
1230
1231         for (; lblock < lblock_stop; lblock += extlen) {
1232                 error = gfs2_extent_map(&ip->i_inode, lblock, &new, &dblock, &extlen);
1233                 if (error)
1234                         return error;
1235
1236                 if (!dblock) {
1237                         *alloc_required = 1;
1238                         return 0;
1239                 }
1240         }
1241
1242         return 0;
1243 }
1244