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 inode *inode = &ip->i_inode;
  63         struct buffer_head *bh;
  64         int release = 0;
  65
  66         if (!page || page->index) {
  67                 page = grab_cache_page(inode->i_mapping, 0);
  68                 if (!page)
  69                         return -ENOMEM;
  70                 release = 1;
  71         }
  72
  73         if (!PageUptodate(page)) {
  74                 void *kaddr = kmap(page);
  75
  76                 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
  77                        ip->i_di.di_size);
  78                 memset(kaddr + ip->i_di.di_size, 0,
  79                        PAGE_CACHE_SIZE - ip->i_di.di_size);
  80                 kunmap(page);
  81
  82                 SetPageUptodate(page);
  83         }
  84
  85         if (!page_has_buffers(page))
  86                 create_empty_buffers(page, 1 << inode->i_blkbits,
  87                                      (1 << BH_Uptodate));
  88
  89         bh = page_buffers(page);
  90
  91         if (!buffer_mapped(bh))
  92                 map_bh(bh, inode->i_sb, block);
  93
  94         set_buffer_uptodate(bh);
  95         if (!gfs2_is_jdata(ip))
  96                 mark_buffer_dirty(bh);
  97         if (!gfs2_is_writeback(ip))
  98                 gfs2_trans_add_bh(ip->i_gl, bh, 0);
  99
 100         if (release) {
 101                 unlock_page(page);
 102                 page_cache_release(page);
 103         }
 104
 105         return 0;
 106 }
 107
 108 /**
 109  * gfs2_unstuff_dinode - Unstuff a dinode when the data has grown too big
 110  * @ip: The GFS2 inode to unstuff
 111  * @unstuffer: the routine that handles unstuffing a non-zero length file
 112  * @private: private data for the unstuffer
 113  *
 114  * This routine unstuffs a dinode and returns it to a "normal" state such
 115  * that the height can be grown in the traditional way.
 116  *
 117  * Returns: errno
 118  */
 119
 120 int gfs2_unstuff_dinode(struct gfs2_inode *ip, struct page *page)
 121 {
 122         struct buffer_head *bh, *dibh;
 123         struct gfs2_dinode *di;
 124         u64 block = 0;
 125         int isdir = gfs2_is_dir(ip);
 126         int error;
 127
 128         down_write(&ip->i_rw_mutex);
 129
 130         error = gfs2_meta_inode_buffer(ip, &dibh);
 131         if (error)
 132                 goto out;
 133
 134         if (ip->i_di.di_size) {
 135                 /* Get a free block, fill it with the stuffed data,
 136                    and write it out to disk */
 137
 138                 if (isdir) {
 139                         block = gfs2_alloc_meta(ip);
 140
 141                         error = gfs2_dir_get_new_buffer(ip, block, &bh);
 142                         if (error)
 143                                 goto out_brelse;
 144                         gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_meta_header),
 145                                               dibh, sizeof(struct gfs2_dinode));
 146                         brelse(bh);
 147                 } else {
 148                         block = gfs2_alloc_data(ip);
 149
 150                         error = gfs2_unstuffer_page(ip, dibh, block, page);
 151                         if (error)
 152                                 goto out_brelse;
 153                 }
 154         }
 155
 156         /*  Set up the pointer to the new block  */
 157
 158         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 159         di = (struct gfs2_dinode *)dibh->b_data;
 160         gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
 161
 162         if (ip->i_di.di_size) {
 163                 *(__be64 *)(di + 1) = cpu_to_be64(block);
 164                 ip->i_di.di_blocks++;
 165                 gfs2_set_inode_blocks(&ip->i_inode);
 166                 di->di_blocks = cpu_to_be64(ip->i_di.di_blocks);
 167         }
 168
 169         ip->i_di.di_height = 1;
 170         di->di_height = cpu_to_be16(1);
 171
 172 out_brelse:
 173         brelse(dibh);
 174 out:
 175         up_write(&ip->i_rw_mutex);
 176         return error;
 177 }
 178
 179 /**
 180  * calc_tree_height - Calculate the height of a metadata tree
 181  * @ip: The GFS2 inode
 182  * @size: The proposed size of the file
 183  *
 184  * Work out how tall a metadata tree needs to be in order to accommodate a
 185  * file of a particular size. If size is less than the current size of
 186  * the inode, then the current size of the inode is used instead of the
 187  * supplied one.
 188  *
 189  * Returns: the height the tree should be
 190  */
 191
 192 static unsigned int calc_tree_height(struct gfs2_inode *ip, u64 size)
 193 {
 194         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 195         u64 *arr;
 196         unsigned int max, height;
 197
 198         if (ip->i_di.di_size > size)
 199                 size = ip->i_di.di_size;
 200
 201         if (gfs2_is_dir(ip)) {
 202                 arr = sdp->sd_jheightsize;
 203                 max = sdp->sd_max_jheight;
 204         } else {
 205                 arr = sdp->sd_heightsize;
 206                 max = sdp->sd_max_height;
 207         }
 208
 209         for (height = 0; height < max; height++)
 210                 if (arr[height] >= size)
 211                         break;
 212
 213         return height;
 214 }
 215
 216 /**
 217  * build_height - Build a metadata tree of the requested height
 218  * @ip: The GFS2 inode
 219  * @height: The height to build to
 220  *
 221  *
 222  * Returns: errno
 223  */
 224
 225 static int build_height(struct inode *inode, unsigned height)
 226 {
 227         struct gfs2_inode *ip = GFS2_I(inode);
 228         unsigned new_height = height - ip->i_di.di_height;
 229         struct buffer_head *dibh;
 230         struct buffer_head *blocks[GFS2_MAX_META_HEIGHT];
 231         struct gfs2_dinode *di;
 232         int error;
 233         __be64 *bp;
 234         u64 bn;
 235         unsigned n;
 236
 237         if (height <= ip->i_di.di_height)
 238                 return 0;
 239
 240         error = gfs2_meta_inode_buffer(ip, &dibh);
 241         if (error)
 242                 return error;
 243
 244         for(n = 0; n < new_height; n++) {
 245                 bn = gfs2_alloc_meta(ip);
 246                 blocks[n] = gfs2_meta_new(ip->i_gl, bn);
 247                 gfs2_trans_add_bh(ip->i_gl, blocks[n], 1);
 248         }
 249
 250         n = 0;
 251         bn = blocks[0]->b_blocknr;
 252         if (new_height > 1) {
 253                 for(; n < new_height-1; n++) {
 254                         gfs2_metatype_set(blocks[n], GFS2_METATYPE_IN,
 255                                           GFS2_FORMAT_IN);
 256                         gfs2_buffer_clear_tail(blocks[n],
 257                                                sizeof(struct gfs2_meta_header));
 258                         bp = (__be64 *)(blocks[n]->b_data +
 259                                      sizeof(struct gfs2_meta_header));
 260                         *bp = cpu_to_be64(blocks[n+1]->b_blocknr);
 261                         brelse(blocks[n]);
 262                         blocks[n] = NULL;
 263                 }
 264         }
 265         gfs2_metatype_set(blocks[n], GFS2_METATYPE_IN, GFS2_FORMAT_IN);
 266         gfs2_buffer_copy_tail(blocks[n], sizeof(struct gfs2_meta_header),
 267                               dibh, sizeof(struct gfs2_dinode));
 268         brelse(blocks[n]);
 269         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 270         di = (struct gfs2_dinode *)dibh->b_data;
 271         gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
 272         *(__be64 *)(di + 1) = cpu_to_be64(bn);
 273         ip->i_di.di_height += new_height;
 274         ip->i_di.di_blocks += new_height;
 275         gfs2_set_inode_blocks(&ip->i_inode);
 276         di->di_height = cpu_to_be16(ip->i_di.di_height);
 277         di->di_blocks = cpu_to_be64(ip->i_di.di_blocks);
 278         brelse(dibh);
 279         return error;
 280 }
 281
 282 /**
 283  * find_metapath - Find path through the metadata tree
 284  * @ip: The inode pointer
 285  * @mp: The metapath to return the result in
 286  * @block: The disk block to look up
 287  *
 288  *   This routine returns a struct metapath structure that defines a path
 289  *   through the metadata of inode "ip" to get to block "block".
 290  *
 291  *   Example:
 292  *   Given:  "ip" is a height 3 file, "offset" is 101342453, and this is a
 293  *   filesystem with a blocksize of 4096.
 294  *
 295  *   find_metapath() would return a struct metapath structure set to:
 296  *   mp_offset = 101342453, mp_height = 3, mp_list[0] = 0, mp_list[1] = 48,
 297  *   and mp_list[2] = 165.
 298  *
 299  *   That means that in order to get to the block containing the byte at
 300  *   offset 101342453, we would load the indirect block pointed to by pointer
 301  *   0 in the dinode.  We would then load the indirect block pointed to by
 302  *   pointer 48 in that indirect block.  We would then load the data block
 303  *   pointed to by pointer 165 in that indirect block.
 304  *
 305  *             ----------------------------------------
 306  *             | Dinode |                             |
 307  *             |        |                            4|
 308  *             |        |0 1 2 3 4 5                 9|
 309  *             |        |                            6|
 310  *             ----------------------------------------
 311  *                       |
 312  *                       |
 313  *                       V
 314  *             ----------------------------------------
 315  *             | Indirect Block                       |
 316  *             |                                     5|
 317  *             |            4 4 4 4 4 5 5            1|
 318  *             |0           5 6 7 8 9 0 1            2|
 319  *             ----------------------------------------
 320  *                                |
 321  *                                |
 322  *                                V
 323  *             ----------------------------------------
 324  *             | Indirect Block                       |
 325  *             |                         1 1 1 1 1   5|
 326  *             |                         6 6 6 6 6   1|
 327  *             |0                        3 4 5 6 7   2|
 328  *             ----------------------------------------
 329  *                                           |
 330  *                                           |
 331  *                                           V
 332  *             ----------------------------------------
 333  *             | Data block containing offset         |
 334  *             |            101342453                 |
 335  *             |                                      |
 336  *             |                                      |
 337  *             ----------------------------------------
 338  *
 339  */
 340
 341 static void find_metapath(struct gfs2_inode *ip, u64 block,
 342                           struct metapath *mp)
 343 {
 344         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 345         u64 b = block;
 346         unsigned int i;
 347
 348         for (i = ip->i_di.di_height; i--;)
 349                 mp->mp_list[i] = do_div(b, sdp->sd_inptrs);
 350
 351 }
 352
 353 /**
 354  * metapointer - Return pointer to start of metadata in a buffer
 355  * @bh: The buffer
 356  * @height: The metadata height (0 = dinode)
 357  * @mp: The metapath
 358  *
 359  * Return a pointer to the block number of the next height of the metadata
 360  * tree given a buffer containing the pointer to the current height of the
 361  * metadata tree.
 362  */
 363
 364 static inline __be64 *metapointer(struct buffer_head *bh, int *boundary,
 365                                unsigned int height, const struct metapath *mp)
 366 {
 367         unsigned int head_size = (height > 0) ?
 368                 sizeof(struct gfs2_meta_header) : sizeof(struct gfs2_dinode);
 369         __be64 *ptr;
 370         *boundary = 0;
 371         ptr = ((__be64 *)(bh->b_data + head_size)) + mp->mp_list[height];
 372         if (ptr + 1 == (__be64 *)(bh->b_data + bh->b_size))
 373                 *boundary = 1;
 374         return ptr;
 375 }
 376
 377 /**
 378  * lookup_block - Get the next metadata block in metadata tree
 379  * @ip: The GFS2 inode
 380  * @bh: Buffer containing the pointers to metadata blocks
 381  * @height: The height of the tree (0 = dinode)
 382  * @mp: The metapath
 383  * @create: Non-zero if we may create a new meatdata block
 384  * @new: Used to indicate if we did create a new metadata block
 385  * @block: the returned disk block number
 386  *
 387  * Given a metatree, complete to a particular height, checks to see if the next
 388  * height of the tree exists. If not the next height of the tree is created.
 389  * The block number of the next height of the metadata tree is returned.
 390  *
 391  */
 392
 393 static int lookup_block(struct gfs2_inode *ip, struct buffer_head *bh,
 394                         unsigned int height, struct metapath *mp, int create,
 395                         int *new, u64 *block)
 396 {
 397         int boundary;
 398         __be64 *ptr = metapointer(bh, &boundary, height, mp);
 399
 400         if (*ptr) {
 401                 *block = be64_to_cpu(*ptr);
 402                 return boundary;
 403         }
 404
 405         *block = 0;
 406
 407         if (!create)
 408                 return 0;
 409
 410         if (height == ip->i_di.di_height - 1 && !gfs2_is_dir(ip))
 411                 *block = gfs2_alloc_data(ip);
 412         else
 413                 *block = gfs2_alloc_meta(ip);
 414
 415         gfs2_trans_add_bh(ip->i_gl, bh, 1);
 416
 417         *ptr = cpu_to_be64(*block);
 418         ip->i_di.di_blocks++;
 419         gfs2_set_inode_blocks(&ip->i_inode);
 420
 421         *new = 1;
 422         return 0;
 423 }
 424
 425 static inline void bmap_lock(struct inode *inode, int create)
 426 {
 427         struct gfs2_inode *ip = GFS2_I(inode);
 428         if (create)
 429                 down_write(&ip->i_rw_mutex);
 430         else
 431                 down_read(&ip->i_rw_mutex);
 432 }
 433
 434 static inline void bmap_unlock(struct inode *inode, int create)
 435 {
 436         struct gfs2_inode *ip = GFS2_I(inode);
 437         if (create)
 438                 up_write(&ip->i_rw_mutex);
 439         else
 440                 up_read(&ip->i_rw_mutex);
 441 }
 442
 443 /**
 444  * gfs2_block_map - Map a block from an inode to a disk block
 445  * @inode: The inode
 446  * @lblock: The logical block number
 447  * @bh_map: The bh to be mapped
 448  *
 449  * Find the block number on the current device which corresponds to an
 450  * inode's block. If the block had to be created, "new" will be set.
 451  *
 452  * Returns: errno
 453  */
 454
 455 int gfs2_block_map(struct inode *inode, sector_t lblock,
 456                    struct buffer_head *bh_map, int create)
 457 {
 458         struct gfs2_inode *ip = GFS2_I(inode);
 459         struct gfs2_sbd *sdp = GFS2_SB(inode);
 460         struct buffer_head *bh;
 461         unsigned int bsize;
 462         unsigned int height;
 463         unsigned int end_of_metadata;
 464         unsigned int x;
 465         int error = 0;
 466         int new = 0;
 467         u64 dblock = 0;
 468         int boundary;
 469         unsigned int maxlen = bh_map->b_size >> inode->i_blkbits;
 470         struct metapath mp;
 471         u64 size;
 472         struct buffer_head *dibh = NULL;
 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         dibh = bh;
 504         get_bh(dibh);
 505
 506         for (x = 0; x < end_of_metadata; x++) {
 507                 lookup_block(ip, bh, x, &mp, create, &new, &dblock);
 508                 brelse(bh);
 509                 if (!dblock)
 510                         goto out_ok;
 511
 512                 error = gfs2_meta_indirect_buffer(ip, x+1, dblock, new, &bh);
 513                 if (error)
 514                         goto out_fail;
 515         }
 516
 517         boundary = lookup_block(ip, bh, end_of_metadata, &mp, create, &new, &dblock);
 518         if (dblock) {
 519                 map_bh(bh_map, inode->i_sb, dblock);
 520                 if (boundary)
 521                         set_buffer_boundary(bh_map);
 522                 if (new) {
 523                         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 524                         gfs2_dinode_out(ip, dibh->b_data);
 525                         set_buffer_new(bh_map);
 526                         goto out_brelse;
 527                 }
 528                 while(--maxlen && !buffer_boundary(bh_map)) {
 529                         u64 eblock;
 530
 531                         mp.mp_list[end_of_metadata]++;
 532                         boundary = lookup_block(ip, bh, end_of_metadata, &mp, 0, &new, &eblock);
 533                         if (eblock != ++dblock)
 534                                 break;
 535                         bh_map->b_size += (1 << inode->i_blkbits);
 536                         if (boundary)
 537                                 set_buffer_boundary(bh_map);
 538                 }
 539         }
 540 out_brelse:
 541         brelse(bh);
 542 out_ok:
 543         error = 0;
 544 out_fail:
 545         if (dibh)
 546                 brelse(dibh);
 547         bmap_unlock(inode, create);
 548         return error;
 549 }
 550
 551 int gfs2_extent_map(struct inode *inode, u64 lblock, int *new, u64 *dblock, unsigned *extlen)
 552 {
 553         struct buffer_head bh = { .b_state = 0, .b_blocknr = 0 };
 554         int ret;
 555         int create = *new;
 556
 557         BUG_ON(!extlen);
 558         BUG_ON(!dblock);
 559         BUG_ON(!new);
 560
 561         bh.b_size = 1 << (inode->i_blkbits + 5);
 562         ret = gfs2_block_map(inode, lblock, &bh, create);
 563         *extlen = bh.b_size >> inode->i_blkbits;
 564         *dblock = bh.b_blocknr;
 565         if (buffer_new(&bh))
 566                 *new = 1;
 567         else
 568                 *new = 0;
 569         return ret;
 570 }
 571
 572 /**
 573  * recursive_scan - recursively scan through the end of a file
 574  * @ip: the inode
 575  * @dibh: the dinode buffer
 576  * @mp: the path through the metadata to the point to start
 577  * @height: the height the recursion is at
 578  * @block: the indirect block to look at
 579  * @first: 1 if this is the first block
 580  * @bc: the call to make for each piece of metadata
 581  * @data: data opaque to this function to pass to @bc
 582  *
 583  * When this is first called @height and @block should be zero and
 584  * @first should be 1.
 585  *
 586  * Returns: errno
 587  */
 588
 589 static int recursive_scan(struct gfs2_inode *ip, struct buffer_head *dibh,
 590                           struct metapath *mp, unsigned int height,
 591                           u64 block, int first, block_call_t bc,
 592                           void *data)
 593 {
 594         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 595         struct buffer_head *bh = NULL;
 596         __be64 *top, *bottom;
 597         u64 bn;
 598         int error;
 599         int mh_size = sizeof(struct gfs2_meta_header);
 600
 601         if (!height) {
 602                 error = gfs2_meta_inode_buffer(ip, &bh);
 603                 if (error)
 604                         return error;
 605                 dibh = bh;
 606
 607                 top = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + mp->mp_list[0];
 608                 bottom = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + sdp->sd_diptrs;
 609         } else {
 610                 error = gfs2_meta_indirect_buffer(ip, height, block, 0, &bh);
 611                 if (error)
 612                         return error;
 613
 614                 top = (__be64 *)(bh->b_data + mh_size) +
 615                                   (first ? mp->mp_list[height] : 0);
 616
 617                 bottom = (__be64 *)(bh->b_data + mh_size) + sdp->sd_inptrs;
 618         }
 619
 620         error = bc(ip, dibh, bh, top, bottom, height, data);
 621         if (error)
 622                 goto out;
 623
 624         if (height < ip->i_di.di_height - 1)
 625                 for (; top < bottom; top++, first = 0) {
 626                         if (!*top)
 627                                 continue;
 628
 629                         bn = be64_to_cpu(*top);
 630
 631                         error = recursive_scan(ip, dibh, mp, height + 1, bn,
 632                                                first, bc, data);
 633                         if (error)
 634                                 break;
 635                 }
 636
 637 out:
 638         brelse(bh);
 639         return error;
 640 }
 641
 642 /**
 643  * do_strip - Look for a layer a particular layer of the file and strip it off
 644  * @ip: the inode
 645  * @dibh: the dinode buffer
 646  * @bh: A buffer of pointers
 647  * @top: The first pointer in the buffer
 648  * @bottom: One more than the last pointer
 649  * @height: the height this buffer is at
 650  * @data: a pointer to a struct strip_mine
 651  *
 652  * Returns: errno
 653  */
 654
 655 static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh,
 656                     struct buffer_head *bh, __be64 *top, __be64 *bottom,
 657                     unsigned int height, void *data)
 658 {
 659         struct strip_mine *sm = data;
 660         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 661         struct gfs2_rgrp_list rlist;
 662         u64 bn, bstart;
 663         u32 blen;
 664         __be64 *p;
 665         unsigned int rg_blocks = 0;
 666         int metadata;
 667         unsigned int revokes = 0;
 668         int x;
 669         int error;
 670
 671         if (!*top)
 672                 sm->sm_first = 0;
 673
 674         if (height != sm->sm_height)
 675                 return 0;
 676
 677         if (sm->sm_first) {
 678                 top++;
 679                 sm->sm_first = 0;
 680         }
 681
 682         metadata = (height != ip->i_di.di_height - 1);
 683         if (metadata)
 684                 revokes = (height) ? sdp->sd_inptrs : sdp->sd_diptrs;
 685
 686         error = gfs2_rindex_hold(sdp, &ip->i_alloc->al_ri_gh);
 687         if (error)
 688                 return error;
 689
 690         memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
 691         bstart = 0;
 692         blen = 0;
 693
 694         for (p = top; p < bottom; p++) {
 695                 if (!*p)
 696                         continue;
 697
 698                 bn = be64_to_cpu(*p);
 699
 700                 if (bstart + blen == bn)
 701                         blen++;
 702                 else {
 703                         if (bstart)
 704                                 gfs2_rlist_add(sdp, &rlist, bstart);
 705
 706                         bstart = bn;
 707                         blen = 1;
 708                 }
 709         }
 710
 711         if (bstart)
 712                 gfs2_rlist_add(sdp, &rlist, bstart);
 713         else
 714                 goto out; /* Nothing to do */
 715
 716         gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE, 0);
 717
 718         for (x = 0; x < rlist.rl_rgrps; x++) {
 719                 struct gfs2_rgrpd *rgd;
 720                 rgd = rlist.rl_ghs[x].gh_gl->gl_object;
 721                 rg_blocks += rgd->rd_length;
 722         }
 723
 724         error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
 725         if (error)
 726                 goto out_rlist;
 727
 728         error = gfs2_trans_begin(sdp, rg_blocks + RES_DINODE +
 729                                  RES_INDIRECT + RES_STATFS + RES_QUOTA,
 730                                  revokes);
 731         if (error)
 732                 goto out_rg_gunlock;
 733
 734         down_write(&ip->i_rw_mutex);
 735
 736         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 737         gfs2_trans_add_bh(ip->i_gl, bh, 1);
 738
 739         bstart = 0;
 740         blen = 0;
 741
 742         for (p = top; p < bottom; p++) {
 743                 if (!*p)
 744                         continue;
 745
 746                 bn = be64_to_cpu(*p);
 747
 748                 if (bstart + blen == bn)
 749                         blen++;
 750                 else {
 751                         if (bstart) {
 752                                 if (metadata)
 753                                         gfs2_free_meta(ip, bstart, blen);
 754                                 else
 755                                         gfs2_free_data(ip, bstart, blen);
 756                         }
 757
 758                         bstart = bn;
 759                         blen = 1;
 760                 }
 761
 762                 *p = 0;
 763                 if (!ip->i_di.di_blocks)
 764                         gfs2_consist_inode(ip);
 765                 ip->i_di.di_blocks--;
 766                 gfs2_set_inode_blocks(&ip->i_inode);
 767         }
 768         if (bstart) {
 769                 if (metadata)
 770                         gfs2_free_meta(ip, bstart, blen);
 771                 else
 772                         gfs2_free_data(ip, bstart, blen);
 773         }
 774
 775         ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
 776
 777         gfs2_dinode_out(ip, dibh->b_data);
 778
 779         up_write(&ip->i_rw_mutex);
 780
 781         gfs2_trans_end(sdp);
 782
 783 out_rg_gunlock:
 784         gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
 785 out_rlist:
 786         gfs2_rlist_free(&rlist);
 787 out:
 788         gfs2_glock_dq_uninit(&ip->i_alloc->al_ri_gh);
 789         return error;
 790 }
 791
 792 /**
 793  * do_grow - Make a file look bigger than it is
 794  * @ip: the inode
 795  * @size: the size to set the file to
 796  *
 797  * Called with an exclusive lock on @ip.
 798  *
 799  * Returns: errno
 800  */
 801
 802 static int do_grow(struct gfs2_inode *ip, u64 size)
 803 {
 804         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 805         struct gfs2_alloc *al;
 806         struct buffer_head *dibh;
 807         unsigned int h;
 808         int error;
 809
 810         al = gfs2_alloc_get(ip);
 811
 812         error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
 813         if (error)
 814                 goto out;
 815
 816         error = gfs2_quota_check(ip, ip->i_inode.i_uid, ip->i_inode.i_gid);
 817         if (error)
 818                 goto out_gunlock_q;
 819
 820         al->al_requested = sdp->sd_max_height + RES_DATA;
 821
 822         error = gfs2_inplace_reserve(ip);
 823         if (error)
 824                 goto out_gunlock_q;
 825
 826         error = gfs2_trans_begin(sdp,
 827                         sdp->sd_max_height + al->al_rgd->rd_length +
 828                         RES_JDATA + RES_DINODE + RES_STATFS + RES_QUOTA, 0);
 829         if (error)
 830                 goto out_ipres;
 831
 832         if (size > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
 833                 if (gfs2_is_stuffed(ip)) {
 834                         error = gfs2_unstuff_dinode(ip, NULL);
 835                         if (error)
 836                                 goto out_end_trans;
 837                 }
 838
 839                 h = calc_tree_height(ip, size);
 840                 if (ip->i_di.di_height < h) {
 841                         down_write(&ip->i_rw_mutex);
 842                         error = build_height(&ip->i_inode, h);
 843                         up_write(&ip->i_rw_mutex);
 844                         if (error)
 845                                 goto out_end_trans;
 846                 }
 847         }
 848
 849         ip->i_di.di_size = size;
 850         ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
 851
 852         error = gfs2_meta_inode_buffer(ip, &dibh);
 853         if (error)
 854                 goto out_end_trans;
 855
 856         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 857         gfs2_dinode_out(ip, dibh->b_data);
 858         brelse(dibh);
 859
 860 out_end_trans:
 861         gfs2_trans_end(sdp);
 862 out_ipres:
 863         gfs2_inplace_release(ip);
 864 out_gunlock_q:
 865         gfs2_quota_unlock(ip);
 866 out:
 867         gfs2_alloc_put(ip);
 868         return error;
 869 }
 870
 871
 872 /**
 873  * gfs2_block_truncate_page - Deal with zeroing out data for truncate
 874  *
 875  * This is partly borrowed from ext3.
 876  */
 877 static int gfs2_block_truncate_page(struct address_space *mapping)
 878 {
 879         struct inode *inode = mapping->host;
 880         struct gfs2_inode *ip = GFS2_I(inode);
 881         loff_t from = inode->i_size;
 882         unsigned long index = from >> PAGE_CACHE_SHIFT;
 883         unsigned offset = from & (PAGE_CACHE_SIZE-1);
 884         unsigned blocksize, iblock, length, pos;
 885         struct buffer_head *bh;
 886         struct page *page;
 887         int err;
 888
 889         page = grab_cache_page(mapping, index);
 890         if (!page)
 891                 return 0;
 892
 893         blocksize = inode->i_sb->s_blocksize;
 894         length = blocksize - (offset & (blocksize - 1));
 895         iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
 896
 897         if (!page_has_buffers(page))
 898                 create_empty_buffers(page, blocksize, 0);
 899
 900         /* Find the buffer that contains "offset" */
 901         bh = page_buffers(page);
 902         pos = blocksize;
 903         while (offset >= pos) {
 904                 bh = bh->b_this_page;
 905                 iblock++;
 906                 pos += blocksize;
 907         }
 908
 909         err = 0;
 910
 911         if (!buffer_mapped(bh)) {
 912                 gfs2_block_map(inode, iblock, bh, 0);
 913                 /* unmapped? It's a hole - nothing to do */
 914                 if (!buffer_mapped(bh))
 915                         goto unlock;
 916         }
 917
 918         /* Ok, it's mapped. Make sure it's up-to-date */
 919         if (PageUptodate(page))
 920                 set_buffer_uptodate(bh);
 921
 922         if (!buffer_uptodate(bh)) {
 923                 err = -EIO;
 924                 ll_rw_block(READ, 1, &bh);
 925                 wait_on_buffer(bh);
 926                 /* Uhhuh. Read error. Complain and punt. */
 927                 if (!buffer_uptodate(bh))
 928                         goto unlock;
 929                 err = 0;
 930         }
 931
 932         if (!gfs2_is_writeback(ip))
 933                 gfs2_trans_add_bh(ip->i_gl, bh, 0);
 934
 935         zero_user_page(page, offset, length, KM_USER0);
 936
 937 unlock:
 938         unlock_page(page);
 939         page_cache_release(page);
 940         return err;
 941 }
 942
 943 static int trunc_start(struct gfs2_inode *ip, u64 size)
 944 {
 945         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 946         struct buffer_head *dibh;
 947         int journaled = gfs2_is_jdata(ip);
 948         int error;
 949
 950         error = gfs2_trans_begin(sdp,
 951                                  RES_DINODE + (journaled ? RES_JDATA : 0), 0);
 952         if (error)
 953                 return error;
 954
 955         error = gfs2_meta_inode_buffer(ip, &dibh);
 956         if (error)
 957                 goto out;
 958
 959         if (gfs2_is_stuffed(ip)) {
 960                 ip->i_di.di_size = size;
 961                 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
 962                 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 963                 gfs2_dinode_out(ip, dibh->b_data);
 964                 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + size);
 965                 error = 1;
 966
 967         } else {
 968                 if (size & (u64)(sdp->sd_sb.sb_bsize - 1))
 969                         error = gfs2_block_truncate_page(ip->i_inode.i_mapping);
 970
 971                 if (!error) {
 972                         ip->i_di.di_size = size;
 973                         ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
 974                         ip->i_di.di_flags |= GFS2_DIF_TRUNC_IN_PROG;
 975                         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 976                         gfs2_dinode_out(ip, dibh->b_data);
 977                 }
 978         }
 979
 980         brelse(dibh);
 981
 982 out:
 983         gfs2_trans_end(sdp);
 984         return error;
 985 }
 986
 987 static int trunc_dealloc(struct gfs2_inode *ip, u64 size)
 988 {
 989         unsigned int height = ip->i_di.di_height;
 990         u64 lblock;
 991         struct metapath mp;
 992         int error;
 993
 994         if (!size)
 995                 lblock = 0;
 996         else
 997                 lblock = (size - 1) >> GFS2_SB(&ip->i_inode)->sd_sb.sb_bsize_shift;
 998
 999         find_metapath(ip, lblock, &mp);
1000         gfs2_alloc_get(ip);
1001
1002         error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1003         if (error)
1004                 goto out;
1005
1006         while (height--) {
1007                 struct strip_mine sm;
1008                 sm.sm_first = !!size;
1009                 sm.sm_height = height;
1010
1011                 error = recursive_scan(ip, NULL, &mp, 0, 0, 1, do_strip, &sm);
1012                 if (error)
1013                         break;
1014         }
1015
1016         gfs2_quota_unhold(ip);
1017
1018 out:
1019         gfs2_alloc_put(ip);
1020         return error;
1021 }
1022
1023 static int trunc_end(struct gfs2_inode *ip)
1024 {
1025         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1026         struct buffer_head *dibh;
1027         int error;
1028
1029         error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1030         if (error)
1031                 return error;
1032
1033         down_write(&ip->i_rw_mutex);
1034
1035         error = gfs2_meta_inode_buffer(ip, &dibh);
1036         if (error)
1037                 goto out;
1038
1039         if (!ip->i_di.di_size) {
1040                 ip->i_di.di_height = 0;
1041                 ip->i_di.di_goal_meta =
1042                         ip->i_di.di_goal_data =
1043                         ip->i_no_addr;
1044                 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
1045         }
1046         ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1047         ip->i_di.di_flags &= ~GFS2_DIF_TRUNC_IN_PROG;
1048
1049         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1050         gfs2_dinode_out(ip, dibh->b_data);
1051         brelse(dibh);
1052
1053 out:
1054         up_write(&ip->i_rw_mutex);
1055         gfs2_trans_end(sdp);
1056         return error;
1057 }
1058
1059 /**
1060  * do_shrink - make a file smaller
1061  * @ip: the inode
1062  * @size: the size to make the file
1063  * @truncator: function to truncate the last partial block
1064  *
1065  * Called with an exclusive lock on @ip.
1066  *
1067  * Returns: errno
1068  */
1069
1070 static int do_shrink(struct gfs2_inode *ip, u64 size)
1071 {
1072         int error;
1073
1074         error = trunc_start(ip, size);
1075         if (error < 0)
1076                 return error;
1077         if (error > 0)
1078                 return 0;
1079
1080         error = trunc_dealloc(ip, size);
1081         if (!error)
1082                 error = trunc_end(ip);
1083
1084         return error;
1085 }
1086
1087 static int do_touch(struct gfs2_inode *ip, u64 size)
1088 {
1089         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1090         struct buffer_head *dibh;
1091         int error;
1092
1093         error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1094         if (error)
1095                 return error;
1096
1097         down_write(&ip->i_rw_mutex);
1098
1099         error = gfs2_meta_inode_buffer(ip, &dibh);
1100         if (error)
1101                 goto do_touch_out;
1102
1103         ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1104         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1105         gfs2_dinode_out(ip, dibh->b_data);
1106         brelse(dibh);
1107
1108 do_touch_out:
1109         up_write(&ip->i_rw_mutex);
1110         gfs2_trans_end(sdp);
1111         return error;
1112 }
1113
1114 /**
1115  * gfs2_truncatei - make a file a given size
1116  * @ip: the inode
1117  * @size: the size to make the file
1118  * @truncator: function to truncate the last partial block
1119  *
1120  * The file size can grow, shrink, or stay the same size.
1121  *
1122  * Returns: errno
1123  */
1124
1125 int gfs2_truncatei(struct gfs2_inode *ip, u64 size)
1126 {
1127         int error;
1128
1129         if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), S_ISREG(ip->i_inode.i_mode)))
1130                 return -EINVAL;
1131
1132         if (size > ip->i_di.di_size)
1133                 error = do_grow(ip, size);
1134         else if (size < ip->i_di.di_size)
1135                 error = do_shrink(ip, size);
1136         else
1137                 /* update time stamps */
1138                 error = do_touch(ip, size);
1139
1140         return error;
1141 }
1142
1143 int gfs2_truncatei_resume(struct gfs2_inode *ip)
1144 {
1145         int error;
1146         error = trunc_dealloc(ip, ip->i_di.di_size);
1147         if (!error)
1148                 error = trunc_end(ip);
1149         return error;
1150 }
1151
1152 int gfs2_file_dealloc(struct gfs2_inode *ip)
1153 {
1154         return trunc_dealloc(ip, 0);
1155 }
1156
1157 /**
1158  * gfs2_write_calc_reserv - calculate number of blocks needed to write to a file
1159  * @ip: the file
1160  * @len: the number of bytes to be written to the file
1161  * @data_blocks: returns the number of data blocks required
1162  * @ind_blocks: returns the number of indirect blocks required
1163  *
1164  */
1165
1166 void gfs2_write_calc_reserv(struct gfs2_inode *ip, unsigned int len,
1167                             unsigned int *data_blocks, unsigned int *ind_blocks)
1168 {
1169         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1170         unsigned int tmp;
1171
1172         if (gfs2_is_dir(ip)) {
1173                 *data_blocks = DIV_ROUND_UP(len, sdp->sd_jbsize) + 2;
1174                 *ind_blocks = 3 * (sdp->sd_max_jheight - 1);
1175         } else {
1176                 *data_blocks = (len >> sdp->sd_sb.sb_bsize_shift) + 3;
1177                 *ind_blocks = 3 * (sdp->sd_max_height - 1);
1178         }
1179
1180         for (tmp = *data_blocks; tmp > sdp->sd_diptrs;) {
1181                 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
1182                 *ind_blocks += tmp;
1183         }
1184 }
1185
1186 /**
1187  * gfs2_write_alloc_required - figure out if a write will require an allocation
1188  * @ip: the file being written to
1189  * @offset: the offset to write to
1190  * @len: the number of bytes being written
1191  * @alloc_required: set to 1 if an alloc is required, 0 otherwise
1192  *
1193  * Returns: errno
1194  */
1195
1196 int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset,
1197                               unsigned int len, int *alloc_required)
1198 {
1199         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1200         u64 lblock, lblock_stop, dblock;
1201         u32 extlen;
1202         int new = 0;
1203         int error = 0;
1204
1205         *alloc_required = 0;
1206
1207         if (!len)
1208                 return 0;
1209
1210         if (gfs2_is_stuffed(ip)) {
1211                 if (offset + len >
1212                     sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
1213                         *alloc_required = 1;
1214                 return 0;
1215         }
1216
1217         if (gfs2_is_dir(ip)) {
1218                 unsigned int bsize = sdp->sd_jbsize;
1219                 lblock = offset;
1220                 do_div(lblock, bsize);
1221                 lblock_stop = offset + len + bsize - 1;
1222                 do_div(lblock_stop, bsize);
1223         } else {
1224                 unsigned int shift = sdp->sd_sb.sb_bsize_shift;
1225                 u64 end_of_file = (ip->i_di.di_size + sdp->sd_sb.sb_bsize - 1) >> shift;
1226                 lblock = offset >> shift;
1227                 lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift;
1228                 if (lblock_stop > end_of_file) {
1229                         *alloc_required = 1;
1230                         return 0;
1231                 }
1232         }
1233
1234         for (; lblock < lblock_stop; lblock += extlen) {
1235                 error = gfs2_extent_map(&ip->i_inode, lblock, &new, &dblock, &extlen);
1236                 if (error)
1237                         return error;
1238
1239                 if (!dblock) {
1240                         *alloc_required = 1;
1241                         return 0;
1242                 }
1243         }
1244
1245         return 0;
1246 }
1247