root/fs/gfs2/bmap.c

/* [<][>][^][v][top][bottom][index][help] */

DEFINITIONS

This source file includes following definitions.
  1. gfs2_unstuffer_page
  2. gfs2_unstuff_dinode
  3. find_metapath
  4. metapath_branch_start
  5. metaptr1
  6. metapointer
  7. metaend
  8. clone_metapath
  9. gfs2_metapath_ra
  10. __fillup_metapath
  11. lookup_metapath
  12. fillup_metapath
  13. metapath_to_block
  14. release_metapath
  15. gfs2_extent_length
  16. gfs2_hole_walker
  17. gfs2_hole_size
  18. gfs2_indirect_init
  19. gfs2_iomap_alloc
  20. gfs2_alloc_size
  21. gfs2_iomap_get
  22. gfs2_lblk_to_dblk
  23. gfs2_write_lock
  24. gfs2_write_unlock
  25. gfs2_iomap_page_prepare
  26. gfs2_iomap_page_done
  27. gfs2_iomap_begin_write
  28. gfs2_iomap_need_write_lock
  29. gfs2_iomap_begin
  30. gfs2_iomap_end
  31. gfs2_block_map
  32. gfs2_extent_map
  33. gfs2_block_zero_range
  34. gfs2_journaled_truncate
  35. trunc_start
  36. gfs2_iomap_get_alloc
  37. sweep_bh_for_rgrps
  38. mp_eq_to_hgt
  39. find_nonnull_ptr
  40. metapointer_range
  41. walk_done
  42. punch_hole
  43. trunc_end
  44. do_shrink
  45. gfs2_trim_blocks
  46. do_grow
  47. gfs2_setattr_size
  48. gfs2_truncatei_resume
  49. gfs2_file_dealloc
  50. gfs2_free_journal_extents
  51. gfs2_add_jextent
  52. gfs2_map_journal_extents
  53. gfs2_write_alloc_required
  54. stuffed_zero_range
  55. gfs2_journaled_truncate_range
  56. __gfs2_punch_hole

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   4  * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
   5  */
   6 
   7 #include <linux/spinlock.h>
   8 #include <linux/completion.h>
   9 #include <linux/buffer_head.h>
  10 #include <linux/blkdev.h>
  11 #include <linux/gfs2_ondisk.h>
  12 #include <linux/crc32.h>
  13 #include <linux/iomap.h>
  14 #include <linux/ktime.h>
  15 
  16 #include "gfs2.h"
  17 #include "incore.h"
  18 #include "bmap.h"
  19 #include "glock.h"
  20 #include "inode.h"
  21 #include "meta_io.h"
  22 #include "quota.h"
  23 #include "rgrp.h"
  24 #include "log.h"
  25 #include "super.h"
  26 #include "trans.h"
  27 #include "dir.h"
  28 #include "util.h"
  29 #include "aops.h"
  30 #include "trace_gfs2.h"
  31 
  32 /* This doesn't need to be that large as max 64 bit pointers in a 4k
  33  * block is 512, so __u16 is fine for that. It saves stack space to
  34  * keep it small.
  35  */
  36 struct metapath {
  37         struct buffer_head *mp_bh[GFS2_MAX_META_HEIGHT];
  38         __u16 mp_list[GFS2_MAX_META_HEIGHT];
  39         int mp_fheight; /* find_metapath height */
  40         int mp_aheight; /* actual height (lookup height) */
  41 };
  42 
  43 static int punch_hole(struct gfs2_inode *ip, u64 offset, u64 length);
  44 
  45 /**
  46  * gfs2_unstuffer_page - unstuff a stuffed inode into a block cached by a page
  47  * @ip: the inode
  48  * @dibh: the dinode buffer
  49  * @block: the block number that was allocated
  50  * @page: The (optional) page. This is looked up if @page is NULL
  51  *
  52  * Returns: errno
  53  */
  54 
  55 static int gfs2_unstuffer_page(struct gfs2_inode *ip, struct buffer_head *dibh,
  56                                u64 block, struct page *page)
  57 {
  58         struct inode *inode = &ip->i_inode;
  59         struct buffer_head *bh;
  60         int release = 0;
  61 
  62         if (!page || page->index) {
  63                 page = find_or_create_page(inode->i_mapping, 0, GFP_NOFS);
  64                 if (!page)
  65                         return -ENOMEM;
  66                 release = 1;
  67         }
  68 
  69         if (!PageUptodate(page)) {
  70                 void *kaddr = kmap(page);
  71                 u64 dsize = i_size_read(inode);
  72  
  73                 if (dsize > gfs2_max_stuffed_size(ip))
  74                         dsize = gfs2_max_stuffed_size(ip);
  75 
  76                 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
  77                 memset(kaddr + dsize, 0, PAGE_SIZE - dsize);
  78                 kunmap(page);
  79 
  80                 SetPageUptodate(page);
  81         }
  82 
  83         if (!page_has_buffers(page))
  84                 create_empty_buffers(page, BIT(inode->i_blkbits),
  85                                      BIT(BH_Uptodate));
  86 
  87         bh = page_buffers(page);
  88 
  89         if (!buffer_mapped(bh))
  90                 map_bh(bh, inode->i_sb, block);
  91 
  92         set_buffer_uptodate(bh);
  93         if (gfs2_is_jdata(ip))
  94                 gfs2_trans_add_data(ip->i_gl, bh);
  95         else {
  96                 mark_buffer_dirty(bh);
  97                 gfs2_ordered_add_inode(ip);
  98         }
  99 
 100         if (release) {
 101                 unlock_page(page);
 102                 put_page(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  * @page: The (optional) page. This is looked up if the @page is NULL
 112  *
 113  * This routine unstuffs a dinode and returns it to a "normal" state such
 114  * that the height can be grown in the traditional way.
 115  *
 116  * Returns: errno
 117  */
 118 
 119 int gfs2_unstuff_dinode(struct gfs2_inode *ip, struct page *page)
 120 {
 121         struct buffer_head *bh, *dibh;
 122         struct gfs2_dinode *di;
 123         u64 block = 0;
 124         int isdir = gfs2_is_dir(ip);
 125         int error;
 126 
 127         down_write(&ip->i_rw_mutex);
 128 
 129         error = gfs2_meta_inode_buffer(ip, &dibh);
 130         if (error)
 131                 goto out;
 132 
 133         if (i_size_read(&ip->i_inode)) {
 134                 /* Get a free block, fill it with the stuffed data,
 135                    and write it out to disk */
 136 
 137                 unsigned int n = 1;
 138                 error = gfs2_alloc_blocks(ip, &block, &n, 0, NULL);
 139                 if (error)
 140                         goto out_brelse;
 141                 if (isdir) {
 142                         gfs2_trans_remove_revoke(GFS2_SB(&ip->i_inode), block, 1);
 143                         error = gfs2_dir_get_new_buffer(ip, block, &bh);
 144                         if (error)
 145                                 goto out_brelse;
 146                         gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_meta_header),
 147                                               dibh, sizeof(struct gfs2_dinode));
 148                         brelse(bh);
 149                 } else {
 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_meta(ip->i_gl, dibh);
 159         di = (struct gfs2_dinode *)dibh->b_data;
 160         gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
 161 
 162         if (i_size_read(&ip->i_inode)) {
 163                 *(__be64 *)(di + 1) = cpu_to_be64(block);
 164                 gfs2_add_inode_blocks(&ip->i_inode, 1);
 165                 di->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
 166         }
 167 
 168         ip->i_height = 1;
 169         di->di_height = cpu_to_be16(1);
 170 
 171 out_brelse:
 172         brelse(dibh);
 173 out:
 174         up_write(&ip->i_rw_mutex);
 175         return error;
 176 }
 177 
 178 
 179 /**
 180  * find_metapath - Find path through the metadata tree
 181  * @sdp: The superblock
 182  * @block: The disk block to look up
 183  * @mp: The metapath to return the result in
 184  * @height: The pre-calculated height of the metadata tree
 185  *
 186  *   This routine returns a struct metapath structure that defines a path
 187  *   through the metadata of inode "ip" to get to block "block".
 188  *
 189  *   Example:
 190  *   Given:  "ip" is a height 3 file, "offset" is 101342453, and this is a
 191  *   filesystem with a blocksize of 4096.
 192  *
 193  *   find_metapath() would return a struct metapath structure set to:
 194  *   mp_fheight = 3, mp_list[0] = 0, mp_list[1] = 48, and mp_list[2] = 165.
 195  *
 196  *   That means that in order to get to the block containing the byte at
 197  *   offset 101342453, we would load the indirect block pointed to by pointer
 198  *   0 in the dinode.  We would then load the indirect block pointed to by
 199  *   pointer 48 in that indirect block.  We would then load the data block
 200  *   pointed to by pointer 165 in that indirect block.
 201  *
 202  *             ----------------------------------------
 203  *             | Dinode |                             |
 204  *             |        |                            4|
 205  *             |        |0 1 2 3 4 5                 9|
 206  *             |        |                            6|
 207  *             ----------------------------------------
 208  *                       |
 209  *                       |
 210  *                       V
 211  *             ----------------------------------------
 212  *             | Indirect Block                       |
 213  *             |                                     5|
 214  *             |            4 4 4 4 4 5 5            1|
 215  *             |0           5 6 7 8 9 0 1            2|
 216  *             ----------------------------------------
 217  *                                |
 218  *                                |
 219  *                                V
 220  *             ----------------------------------------
 221  *             | Indirect Block                       |
 222  *             |                         1 1 1 1 1   5|
 223  *             |                         6 6 6 6 6   1|
 224  *             |0                        3 4 5 6 7   2|
 225  *             ----------------------------------------
 226  *                                           |
 227  *                                           |
 228  *                                           V
 229  *             ----------------------------------------
 230  *             | Data block containing offset         |
 231  *             |            101342453                 |
 232  *             |                                      |
 233  *             |                                      |
 234  *             ----------------------------------------
 235  *
 236  */
 237 
 238 static void find_metapath(const struct gfs2_sbd *sdp, u64 block,
 239                           struct metapath *mp, unsigned int height)
 240 {
 241         unsigned int i;
 242 
 243         mp->mp_fheight = height;
 244         for (i = height; i--;)
 245                 mp->mp_list[i] = do_div(block, sdp->sd_inptrs);
 246 }
 247 
 248 static inline unsigned int metapath_branch_start(const struct metapath *mp)
 249 {
 250         if (mp->mp_list[0] == 0)
 251                 return 2;
 252         return 1;
 253 }
 254 
 255 /**
 256  * metaptr1 - Return the first possible metadata pointer in a metapath buffer
 257  * @height: The metadata height (0 = dinode)
 258  * @mp: The metapath
 259  */
 260 static inline __be64 *metaptr1(unsigned int height, const struct metapath *mp)
 261 {
 262         struct buffer_head *bh = mp->mp_bh[height];
 263         if (height == 0)
 264                 return ((__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)));
 265         return ((__be64 *)(bh->b_data + sizeof(struct gfs2_meta_header)));
 266 }
 267 
 268 /**
 269  * metapointer - Return pointer to start of metadata in a buffer
 270  * @height: The metadata height (0 = dinode)
 271  * @mp: The metapath
 272  *
 273  * Return a pointer to the block number of the next height of the metadata
 274  * tree given a buffer containing the pointer to the current height of the
 275  * metadata tree.
 276  */
 277 
 278 static inline __be64 *metapointer(unsigned int height, const struct metapath *mp)
 279 {
 280         __be64 *p = metaptr1(height, mp);
 281         return p + mp->mp_list[height];
 282 }
 283 
 284 static inline const __be64 *metaend(unsigned int height, const struct metapath *mp)
 285 {
 286         const struct buffer_head *bh = mp->mp_bh[height];
 287         return (const __be64 *)(bh->b_data + bh->b_size);
 288 }
 289 
 290 static void clone_metapath(struct metapath *clone, struct metapath *mp)
 291 {
 292         unsigned int hgt;
 293 
 294         *clone = *mp;
 295         for (hgt = 0; hgt < mp->mp_aheight; hgt++)
 296                 get_bh(clone->mp_bh[hgt]);
 297 }
 298 
 299 static void gfs2_metapath_ra(struct gfs2_glock *gl, __be64 *start, __be64 *end)
 300 {
 301         const __be64 *t;
 302 
 303         for (t = start; t < end; t++) {
 304                 struct buffer_head *rabh;
 305 
 306                 if (!*t)
 307                         continue;
 308 
 309                 rabh = gfs2_getbuf(gl, be64_to_cpu(*t), CREATE);
 310                 if (trylock_buffer(rabh)) {
 311                         if (!buffer_uptodate(rabh)) {
 312                                 rabh->b_end_io = end_buffer_read_sync;
 313                                 submit_bh(REQ_OP_READ,
 314                                           REQ_RAHEAD | REQ_META | REQ_PRIO,
 315                                           rabh);
 316                                 continue;
 317                         }
 318                         unlock_buffer(rabh);
 319                 }
 320                 brelse(rabh);
 321         }
 322 }
 323 
 324 static int __fillup_metapath(struct gfs2_inode *ip, struct metapath *mp,
 325                              unsigned int x, unsigned int h)
 326 {
 327         for (; x < h; x++) {
 328                 __be64 *ptr = metapointer(x, mp);
 329                 u64 dblock = be64_to_cpu(*ptr);
 330                 int ret;
 331 
 332                 if (!dblock)
 333                         break;
 334                 ret = gfs2_meta_indirect_buffer(ip, x + 1, dblock, &mp->mp_bh[x + 1]);
 335                 if (ret)
 336                         return ret;
 337         }
 338         mp->mp_aheight = x + 1;
 339         return 0;
 340 }
 341 
 342 /**
 343  * lookup_metapath - Walk the metadata tree to a specific point
 344  * @ip: The inode
 345  * @mp: The metapath
 346  *
 347  * Assumes that the inode's buffer has already been looked up and
 348  * hooked onto mp->mp_bh[0] and that the metapath has been initialised
 349  * by find_metapath().
 350  *
 351  * If this function encounters part of the tree which has not been
 352  * allocated, it returns the current height of the tree at the point
 353  * at which it found the unallocated block. Blocks which are found are
 354  * added to the mp->mp_bh[] list.
 355  *
 356  * Returns: error
 357  */
 358 
 359 static int lookup_metapath(struct gfs2_inode *ip, struct metapath *mp)
 360 {
 361         return __fillup_metapath(ip, mp, 0, ip->i_height - 1);
 362 }
 363 
 364 /**
 365  * fillup_metapath - fill up buffers for the metadata path to a specific height
 366  * @ip: The inode
 367  * @mp: The metapath
 368  * @h: The height to which it should be mapped
 369  *
 370  * Similar to lookup_metapath, but does lookups for a range of heights
 371  *
 372  * Returns: error or the number of buffers filled
 373  */
 374 
 375 static int fillup_metapath(struct gfs2_inode *ip, struct metapath *mp, int h)
 376 {
 377         unsigned int x = 0;
 378         int ret;
 379 
 380         if (h) {
 381                 /* find the first buffer we need to look up. */
 382                 for (x = h - 1; x > 0; x--) {
 383                         if (mp->mp_bh[x])
 384                                 break;
 385                 }
 386         }
 387         ret = __fillup_metapath(ip, mp, x, h);
 388         if (ret)
 389                 return ret;
 390         return mp->mp_aheight - x - 1;
 391 }
 392 
 393 static sector_t metapath_to_block(struct gfs2_sbd *sdp, struct metapath *mp)
 394 {
 395         sector_t factor = 1, block = 0;
 396         int hgt;
 397 
 398         for (hgt = mp->mp_fheight - 1; hgt >= 0; hgt--) {
 399                 if (hgt < mp->mp_aheight)
 400                         block += mp->mp_list[hgt] * factor;
 401                 factor *= sdp->sd_inptrs;
 402         }
 403         return block;
 404 }
 405 
 406 static void release_metapath(struct metapath *mp)
 407 {
 408         int i;
 409 
 410         for (i = 0; i < GFS2_MAX_META_HEIGHT; i++) {
 411                 if (mp->mp_bh[i] == NULL)
 412                         break;
 413                 brelse(mp->mp_bh[i]);
 414                 mp->mp_bh[i] = NULL;
 415         }
 416 }
 417 
 418 /**
 419  * gfs2_extent_length - Returns length of an extent of blocks
 420  * @bh: The metadata block
 421  * @ptr: Current position in @bh
 422  * @limit: Max extent length to return
 423  * @eob: Set to 1 if we hit "end of block"
 424  *
 425  * Returns: The length of the extent (minimum of one block)
 426  */
 427 
 428 static inline unsigned int gfs2_extent_length(struct buffer_head *bh, __be64 *ptr, size_t limit, int *eob)
 429 {
 430         const __be64 *end = (__be64 *)(bh->b_data + bh->b_size);
 431         const __be64 *first = ptr;
 432         u64 d = be64_to_cpu(*ptr);
 433 
 434         *eob = 0;
 435         do {
 436                 ptr++;
 437                 if (ptr >= end)
 438                         break;
 439                 d++;
 440         } while(be64_to_cpu(*ptr) == d);
 441         if (ptr >= end)
 442                 *eob = 1;
 443         return ptr - first;
 444 }
 445 
 446 enum walker_status { WALK_STOP, WALK_FOLLOW, WALK_CONTINUE };
 447 
 448 /*
 449  * gfs2_metadata_walker - walk an indirect block
 450  * @mp: Metapath to indirect block
 451  * @ptrs: Number of pointers to look at
 452  *
 453  * When returning WALK_FOLLOW, the walker must update @mp to point at the right
 454  * indirect block to follow.
 455  */
 456 typedef enum walker_status (*gfs2_metadata_walker)(struct metapath *mp,
 457                                                    unsigned int ptrs);
 458 
 459 /*
 460  * gfs2_walk_metadata - walk a tree of indirect blocks
 461  * @inode: The inode
 462  * @mp: Starting point of walk
 463  * @max_len: Maximum number of blocks to walk
 464  * @walker: Called during the walk
 465  *
 466  * Returns 1 if the walk was stopped by @walker, 0 if we went past @max_len or
 467  * past the end of metadata, and a negative error code otherwise.
 468  */
 469 
 470 static int gfs2_walk_metadata(struct inode *inode, struct metapath *mp,
 471                 u64 max_len, gfs2_metadata_walker walker)
 472 {
 473         struct gfs2_inode *ip = GFS2_I(inode);
 474         struct gfs2_sbd *sdp = GFS2_SB(inode);
 475         u64 factor = 1;
 476         unsigned int hgt;
 477         int ret;
 478 
 479         /*
 480          * The walk starts in the lowest allocated indirect block, which may be
 481          * before the position indicated by @mp.  Adjust @max_len accordingly
 482          * to avoid a short walk.
 483          */
 484         for (hgt = mp->mp_fheight - 1; hgt >= mp->mp_aheight; hgt--) {
 485                 max_len += mp->mp_list[hgt] * factor;
 486                 mp->mp_list[hgt] = 0;
 487                 factor *= sdp->sd_inptrs;
 488         }
 489 
 490         for (;;) {
 491                 u16 start = mp->mp_list[hgt];
 492                 enum walker_status status;
 493                 unsigned int ptrs;
 494                 u64 len;
 495 
 496                 /* Walk indirect block. */
 497                 ptrs = (hgt >= 1 ? sdp->sd_inptrs : sdp->sd_diptrs) - start;
 498                 len = ptrs * factor;
 499                 if (len > max_len)
 500                         ptrs = DIV_ROUND_UP_ULL(max_len, factor);
 501                 status = walker(mp, ptrs);
 502                 switch (status) {
 503                 case WALK_STOP:
 504                         return 1;
 505                 case WALK_FOLLOW:
 506                         BUG_ON(mp->mp_aheight == mp->mp_fheight);
 507                         ptrs = mp->mp_list[hgt] - start;
 508                         len = ptrs * factor;
 509                         break;
 510                 case WALK_CONTINUE:
 511                         break;
 512                 }
 513                 if (len >= max_len)
 514                         break;
 515                 max_len -= len;
 516                 if (status == WALK_FOLLOW)
 517                         goto fill_up_metapath;
 518 
 519 lower_metapath:
 520                 /* Decrease height of metapath. */
 521                 brelse(mp->mp_bh[hgt]);
 522                 mp->mp_bh[hgt] = NULL;
 523                 mp->mp_list[hgt] = 0;
 524                 if (!hgt)
 525                         break;
 526                 hgt--;
 527                 factor *= sdp->sd_inptrs;
 528 
 529                 /* Advance in metadata tree. */
 530                 (mp->mp_list[hgt])++;
 531                 if (hgt) {
 532                         if (mp->mp_list[hgt] >= sdp->sd_inptrs)
 533                                 goto lower_metapath;
 534                 } else {
 535                         if (mp->mp_list[hgt] >= sdp->sd_diptrs)
 536                                 break;
 537                 }
 538 
 539 fill_up_metapath:
 540                 /* Increase height of metapath. */
 541                 ret = fillup_metapath(ip, mp, ip->i_height - 1);
 542                 if (ret < 0)
 543                         return ret;
 544                 hgt += ret;
 545                 for (; ret; ret--)
 546                         do_div(factor, sdp->sd_inptrs);
 547                 mp->mp_aheight = hgt + 1;
 548         }
 549         return 0;
 550 }
 551 
 552 static enum walker_status gfs2_hole_walker(struct metapath *mp,
 553                                            unsigned int ptrs)
 554 {
 555         const __be64 *start, *ptr, *end;
 556         unsigned int hgt;
 557 
 558         hgt = mp->mp_aheight - 1;
 559         start = metapointer(hgt, mp);
 560         end = start + ptrs;
 561 
 562         for (ptr = start; ptr < end; ptr++) {
 563                 if (*ptr) {
 564                         mp->mp_list[hgt] += ptr - start;
 565                         if (mp->mp_aheight == mp->mp_fheight)
 566                                 return WALK_STOP;
 567                         return WALK_FOLLOW;
 568                 }
 569         }
 570         return WALK_CONTINUE;
 571 }
 572 
 573 /**
 574  * gfs2_hole_size - figure out the size of a hole
 575  * @inode: The inode
 576  * @lblock: The logical starting block number
 577  * @len: How far to look (in blocks)
 578  * @mp: The metapath at lblock
 579  * @iomap: The iomap to store the hole size in
 580  *
 581  * This function modifies @mp.
 582  *
 583  * Returns: errno on error
 584  */
 585 static int gfs2_hole_size(struct inode *inode, sector_t lblock, u64 len,
 586                           struct metapath *mp, struct iomap *iomap)
 587 {
 588         struct metapath clone;
 589         u64 hole_size;
 590         int ret;
 591 
 592         clone_metapath(&clone, mp);
 593         ret = gfs2_walk_metadata(inode, &clone, len, gfs2_hole_walker);
 594         if (ret < 0)
 595                 goto out;
 596 
 597         if (ret == 1)
 598                 hole_size = metapath_to_block(GFS2_SB(inode), &clone) - lblock;
 599         else
 600                 hole_size = len;
 601         iomap->length = hole_size << inode->i_blkbits;
 602         ret = 0;
 603 
 604 out:
 605         release_metapath(&clone);
 606         return ret;
 607 }
 608 
 609 static inline __be64 *gfs2_indirect_init(struct metapath *mp,
 610                                          struct gfs2_glock *gl, unsigned int i,
 611                                          unsigned offset, u64 bn)
 612 {
 613         __be64 *ptr = (__be64 *)(mp->mp_bh[i - 1]->b_data +
 614                        ((i > 1) ? sizeof(struct gfs2_meta_header) :
 615                                  sizeof(struct gfs2_dinode)));
 616         BUG_ON(i < 1);
 617         BUG_ON(mp->mp_bh[i] != NULL);
 618         mp->mp_bh[i] = gfs2_meta_new(gl, bn);
 619         gfs2_trans_add_meta(gl, mp->mp_bh[i]);
 620         gfs2_metatype_set(mp->mp_bh[i], GFS2_METATYPE_IN, GFS2_FORMAT_IN);
 621         gfs2_buffer_clear_tail(mp->mp_bh[i], sizeof(struct gfs2_meta_header));
 622         ptr += offset;
 623         *ptr = cpu_to_be64(bn);
 624         return ptr;
 625 }
 626 
 627 enum alloc_state {
 628         ALLOC_DATA = 0,
 629         ALLOC_GROW_DEPTH = 1,
 630         ALLOC_GROW_HEIGHT = 2,
 631         /* ALLOC_UNSTUFF = 3,   TBD and rather complicated */
 632 };
 633 
 634 /**
 635  * gfs2_iomap_alloc - Build a metadata tree of the requested height
 636  * @inode: The GFS2 inode
 637  * @iomap: The iomap structure
 638  * @mp: The metapath, with proper height information calculated
 639  *
 640  * In this routine we may have to alloc:
 641  *   i) Indirect blocks to grow the metadata tree height
 642  *  ii) Indirect blocks to fill in lower part of the metadata tree
 643  * iii) Data blocks
 644  *
 645  * This function is called after gfs2_iomap_get, which works out the
 646  * total number of blocks which we need via gfs2_alloc_size.
 647  *
 648  * We then do the actual allocation asking for an extent at a time (if
 649  * enough contiguous free blocks are available, there will only be one
 650  * allocation request per call) and uses the state machine to initialise
 651  * the blocks in order.
 652  *
 653  * Right now, this function will allocate at most one indirect block
 654  * worth of data -- with a default block size of 4K, that's slightly
 655  * less than 2M.  If this limitation is ever removed to allow huge
 656  * allocations, we would probably still want to limit the iomap size we
 657  * return to avoid stalling other tasks during huge writes; the next
 658  * iomap iteration would then find the blocks already allocated.
 659  *
 660  * Returns: errno on error
 661  */
 662 
 663 static int gfs2_iomap_alloc(struct inode *inode, struct iomap *iomap,
 664                             struct metapath *mp)
 665 {
 666         struct gfs2_inode *ip = GFS2_I(inode);
 667         struct gfs2_sbd *sdp = GFS2_SB(inode);
 668         struct buffer_head *dibh = mp->mp_bh[0];
 669         u64 bn;
 670         unsigned n, i, blks, alloced = 0, iblks = 0, branch_start = 0;
 671         size_t dblks = iomap->length >> inode->i_blkbits;
 672         const unsigned end_of_metadata = mp->mp_fheight - 1;
 673         int ret;
 674         enum alloc_state state;
 675         __be64 *ptr;
 676         __be64 zero_bn = 0;
 677 
 678         BUG_ON(mp->mp_aheight < 1);
 679         BUG_ON(dibh == NULL);
 680         BUG_ON(dblks < 1);
 681 
 682         gfs2_trans_add_meta(ip->i_gl, dibh);
 683 
 684         down_write(&ip->i_rw_mutex);
 685 
 686         if (mp->mp_fheight == mp->mp_aheight) {
 687                 /* Bottom indirect block exists */
 688                 state = ALLOC_DATA;
 689         } else {
 690                 /* Need to allocate indirect blocks */
 691                 if (mp->mp_fheight == ip->i_height) {
 692                         /* Writing into existing tree, extend tree down */
 693                         iblks = mp->mp_fheight - mp->mp_aheight;
 694                         state = ALLOC_GROW_DEPTH;
 695                 } else {
 696                         /* Building up tree height */
 697                         state = ALLOC_GROW_HEIGHT;
 698                         iblks = mp->mp_fheight - ip->i_height;
 699                         branch_start = metapath_branch_start(mp);
 700                         iblks += (mp->mp_fheight - branch_start);
 701                 }
 702         }
 703 
 704         /* start of the second part of the function (state machine) */
 705 
 706         blks = dblks + iblks;
 707         i = mp->mp_aheight;
 708         do {
 709                 n = blks - alloced;
 710                 ret = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL);
 711                 if (ret)
 712                         goto out;
 713                 alloced += n;
 714                 if (state != ALLOC_DATA || gfs2_is_jdata(ip))
 715                         gfs2_trans_remove_revoke(sdp, bn, n);
 716                 switch (state) {
 717                 /* Growing height of tree */
 718                 case ALLOC_GROW_HEIGHT:
 719                         if (i == 1) {
 720                                 ptr = (__be64 *)(dibh->b_data +
 721                                                  sizeof(struct gfs2_dinode));
 722                                 zero_bn = *ptr;
 723                         }
 724                         for (; i - 1 < mp->mp_fheight - ip->i_height && n > 0;
 725                              i++, n--)
 726                                 gfs2_indirect_init(mp, ip->i_gl, i, 0, bn++);
 727                         if (i - 1 == mp->mp_fheight - ip->i_height) {
 728                                 i--;
 729                                 gfs2_buffer_copy_tail(mp->mp_bh[i],
 730                                                 sizeof(struct gfs2_meta_header),
 731                                                 dibh, sizeof(struct gfs2_dinode));
 732                                 gfs2_buffer_clear_tail(dibh,
 733                                                 sizeof(struct gfs2_dinode) +
 734                                                 sizeof(__be64));
 735                                 ptr = (__be64 *)(mp->mp_bh[i]->b_data +
 736                                         sizeof(struct gfs2_meta_header));
 737                                 *ptr = zero_bn;
 738                                 state = ALLOC_GROW_DEPTH;
 739                                 for(i = branch_start; i < mp->mp_fheight; i++) {
 740                                         if (mp->mp_bh[i] == NULL)
 741                                                 break;
 742                                         brelse(mp->mp_bh[i]);
 743                                         mp->mp_bh[i] = NULL;
 744                                 }
 745                                 i = branch_start;
 746                         }
 747                         if (n == 0)
 748                                 break;
 749                 /* fall through - To branching from existing tree */
 750                 case ALLOC_GROW_DEPTH:
 751                         if (i > 1 && i < mp->mp_fheight)
 752                                 gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[i-1]);
 753                         for (; i < mp->mp_fheight && n > 0; i++, n--)
 754                                 gfs2_indirect_init(mp, ip->i_gl, i,
 755                                                    mp->mp_list[i-1], bn++);
 756                         if (i == mp->mp_fheight)
 757                                 state = ALLOC_DATA;
 758                         if (n == 0)
 759                                 break;
 760                 /* fall through - To tree complete, adding data blocks */
 761                 case ALLOC_DATA:
 762                         BUG_ON(n > dblks);
 763                         BUG_ON(mp->mp_bh[end_of_metadata] == NULL);
 764                         gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[end_of_metadata]);
 765                         dblks = n;
 766                         ptr = metapointer(end_of_metadata, mp);
 767                         iomap->addr = bn << inode->i_blkbits;
 768                         iomap->flags |= IOMAP_F_MERGED | IOMAP_F_NEW;
 769                         while (n-- > 0)
 770                                 *ptr++ = cpu_to_be64(bn++);
 771                         break;
 772                 }
 773         } while (iomap->addr == IOMAP_NULL_ADDR);
 774 
 775         iomap->type = IOMAP_MAPPED;
 776         iomap->length = (u64)dblks << inode->i_blkbits;
 777         ip->i_height = mp->mp_fheight;
 778         gfs2_add_inode_blocks(&ip->i_inode, alloced);
 779         gfs2_dinode_out(ip, dibh->b_data);
 780 out:
 781         up_write(&ip->i_rw_mutex);
 782         return ret;
 783 }
 784 
 785 #define IOMAP_F_GFS2_BOUNDARY IOMAP_F_PRIVATE
 786 
 787 /**
 788  * gfs2_alloc_size - Compute the maximum allocation size
 789  * @inode: The inode
 790  * @mp: The metapath
 791  * @size: Requested size in blocks
 792  *
 793  * Compute the maximum size of the next allocation at @mp.
 794  *
 795  * Returns: size in blocks
 796  */
 797 static u64 gfs2_alloc_size(struct inode *inode, struct metapath *mp, u64 size)
 798 {
 799         struct gfs2_inode *ip = GFS2_I(inode);
 800         struct gfs2_sbd *sdp = GFS2_SB(inode);
 801         const __be64 *first, *ptr, *end;
 802 
 803         /*
 804          * For writes to stuffed files, this function is called twice via
 805          * gfs2_iomap_get, before and after unstuffing. The size we return the
 806          * first time needs to be large enough to get the reservation and
 807          * allocation sizes right.  The size we return the second time must
 808          * be exact or else gfs2_iomap_alloc won't do the right thing.
 809          */
 810 
 811         if (gfs2_is_stuffed(ip) || mp->mp_fheight != mp->mp_aheight) {
 812                 unsigned int maxsize = mp->mp_fheight > 1 ?
 813                         sdp->sd_inptrs : sdp->sd_diptrs;
 814                 maxsize -= mp->mp_list[mp->mp_fheight - 1];
 815                 if (size > maxsize)
 816                         size = maxsize;
 817                 return size;
 818         }
 819 
 820         first = metapointer(ip->i_height - 1, mp);
 821         end = metaend(ip->i_height - 1, mp);
 822         if (end - first > size)
 823                 end = first + size;
 824         for (ptr = first; ptr < end; ptr++) {
 825                 if (*ptr)
 826                         break;
 827         }
 828         return ptr - first;
 829 }
 830 
 831 /**
 832  * gfs2_iomap_get - Map blocks from an inode to disk blocks
 833  * @inode: The inode
 834  * @pos: Starting position in bytes
 835  * @length: Length to map, in bytes
 836  * @flags: iomap flags
 837  * @iomap: The iomap structure
 838  * @mp: The metapath
 839  *
 840  * Returns: errno
 841  */
 842 static int gfs2_iomap_get(struct inode *inode, loff_t pos, loff_t length,
 843                           unsigned flags, struct iomap *iomap,
 844                           struct metapath *mp)
 845 {
 846         struct gfs2_inode *ip = GFS2_I(inode);
 847         struct gfs2_sbd *sdp = GFS2_SB(inode);
 848         loff_t size = i_size_read(inode);
 849         __be64 *ptr;
 850         sector_t lblock;
 851         sector_t lblock_stop;
 852         int ret;
 853         int eob;
 854         u64 len;
 855         struct buffer_head *dibh = NULL, *bh;
 856         u8 height;
 857 
 858         if (!length)
 859                 return -EINVAL;
 860 
 861         down_read(&ip->i_rw_mutex);
 862 
 863         ret = gfs2_meta_inode_buffer(ip, &dibh);
 864         if (ret)
 865                 goto unlock;
 866         mp->mp_bh[0] = dibh;
 867 
 868         if (gfs2_is_stuffed(ip)) {
 869                 if (flags & IOMAP_WRITE) {
 870                         loff_t max_size = gfs2_max_stuffed_size(ip);
 871 
 872                         if (pos + length > max_size)
 873                                 goto unstuff;
 874                         iomap->length = max_size;
 875                 } else {
 876                         if (pos >= size) {
 877                                 if (flags & IOMAP_REPORT) {
 878                                         ret = -ENOENT;
 879                                         goto unlock;
 880                                 } else {
 881                                         iomap->offset = pos;
 882                                         iomap->length = length;
 883                                         goto hole_found;
 884                                 }
 885                         }
 886                         iomap->length = size;
 887                 }
 888                 iomap->addr = (ip->i_no_addr << inode->i_blkbits) +
 889                               sizeof(struct gfs2_dinode);
 890                 iomap->type = IOMAP_INLINE;
 891                 iomap->inline_data = dibh->b_data + sizeof(struct gfs2_dinode);
 892                 goto out;
 893         }
 894 
 895 unstuff:
 896         lblock = pos >> inode->i_blkbits;
 897         iomap->offset = lblock << inode->i_blkbits;
 898         lblock_stop = (pos + length - 1) >> inode->i_blkbits;
 899         len = lblock_stop - lblock + 1;
 900         iomap->length = len << inode->i_blkbits;
 901 
 902         height = ip->i_height;
 903         while ((lblock + 1) * sdp->sd_sb.sb_bsize > sdp->sd_heightsize[height])
 904                 height++;
 905         find_metapath(sdp, lblock, mp, height);
 906         if (height > ip->i_height || gfs2_is_stuffed(ip))
 907                 goto do_alloc;
 908 
 909         ret = lookup_metapath(ip, mp);
 910         if (ret)
 911                 goto unlock;
 912 
 913         if (mp->mp_aheight != ip->i_height)
 914                 goto do_alloc;
 915 
 916         ptr = metapointer(ip->i_height - 1, mp);
 917         if (*ptr == 0)
 918                 goto do_alloc;
 919 
 920         bh = mp->mp_bh[ip->i_height - 1];
 921         len = gfs2_extent_length(bh, ptr, len, &eob);
 922 
 923         iomap->addr = be64_to_cpu(*ptr) << inode->i_blkbits;
 924         iomap->length = len << inode->i_blkbits;
 925         iomap->type = IOMAP_MAPPED;
 926         iomap->flags |= IOMAP_F_MERGED;
 927         if (eob)
 928                 iomap->flags |= IOMAP_F_GFS2_BOUNDARY;
 929 
 930 out:
 931         iomap->bdev = inode->i_sb->s_bdev;
 932 unlock:
 933         up_read(&ip->i_rw_mutex);
 934         return ret;
 935 
 936 do_alloc:
 937         if (flags & IOMAP_REPORT) {
 938                 if (pos >= size)
 939                         ret = -ENOENT;
 940                 else if (height == ip->i_height)
 941                         ret = gfs2_hole_size(inode, lblock, len, mp, iomap);
 942                 else
 943                         iomap->length = size - pos;
 944         } else if (flags & IOMAP_WRITE) {
 945                 u64 alloc_size;
 946 
 947                 if (flags & IOMAP_DIRECT)
 948                         goto out;  /* (see gfs2_file_direct_write) */
 949 
 950                 len = gfs2_alloc_size(inode, mp, len);
 951                 alloc_size = len << inode->i_blkbits;
 952                 if (alloc_size < iomap->length)
 953                         iomap->length = alloc_size;
 954         } else {
 955                 if (pos < size && height == ip->i_height)
 956                         ret = gfs2_hole_size(inode, lblock, len, mp, iomap);
 957         }
 958 hole_found:
 959         iomap->addr = IOMAP_NULL_ADDR;
 960         iomap->type = IOMAP_HOLE;
 961         goto out;
 962 }
 963 
 964 /**
 965  * gfs2_lblk_to_dblk - convert logical block to disk block
 966  * @inode: the inode of the file we're mapping
 967  * @lblock: the block relative to the start of the file
 968  * @dblock: the returned dblock, if no error
 969  *
 970  * This function maps a single block from a file logical block (relative to
 971  * the start of the file) to a file system absolute block using iomap.
 972  *
 973  * Returns: the absolute file system block, or an error
 974  */
 975 int gfs2_lblk_to_dblk(struct inode *inode, u32 lblock, u64 *dblock)
 976 {
 977         struct iomap iomap = { };
 978         struct metapath mp = { .mp_aheight = 1, };
 979         loff_t pos = (loff_t)lblock << inode->i_blkbits;
 980         int ret;
 981 
 982         ret = gfs2_iomap_get(inode, pos, i_blocksize(inode), 0, &iomap, &mp);
 983         release_metapath(&mp);
 984         if (ret == 0)
 985                 *dblock = iomap.addr >> inode->i_blkbits;
 986 
 987         return ret;
 988 }
 989 
 990 static int gfs2_write_lock(struct inode *inode)
 991 {
 992         struct gfs2_inode *ip = GFS2_I(inode);
 993         struct gfs2_sbd *sdp = GFS2_SB(inode);
 994         int error;
 995 
 996         gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
 997         error = gfs2_glock_nq(&ip->i_gh);
 998         if (error)
 999                 goto out_uninit;
1000         if (&ip->i_inode == sdp->sd_rindex) {
1001                 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
1002 
1003                 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
1004                                            GL_NOCACHE, &m_ip->i_gh);
1005                 if (error)
1006                         goto out_unlock;
1007         }
1008         return 0;
1009 
1010 out_unlock:
1011         gfs2_glock_dq(&ip->i_gh);
1012 out_uninit:
1013         gfs2_holder_uninit(&ip->i_gh);
1014         return error;
1015 }
1016 
1017 static void gfs2_write_unlock(struct inode *inode)
1018 {
1019         struct gfs2_inode *ip = GFS2_I(inode);
1020         struct gfs2_sbd *sdp = GFS2_SB(inode);
1021 
1022         if (&ip->i_inode == sdp->sd_rindex) {
1023                 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
1024 
1025                 gfs2_glock_dq_uninit(&m_ip->i_gh);
1026         }
1027         gfs2_glock_dq_uninit(&ip->i_gh);
1028 }
1029 
1030 static int gfs2_iomap_page_prepare(struct inode *inode, loff_t pos,
1031                                    unsigned len, struct iomap *iomap)
1032 {
1033         unsigned int blockmask = i_blocksize(inode) - 1;
1034         struct gfs2_sbd *sdp = GFS2_SB(inode);
1035         unsigned int blocks;
1036 
1037         blocks = ((pos & blockmask) + len + blockmask) >> inode->i_blkbits;
1038         return gfs2_trans_begin(sdp, RES_DINODE + blocks, 0);
1039 }
1040 
1041 static void gfs2_iomap_page_done(struct inode *inode, loff_t pos,
1042                                  unsigned copied, struct page *page,
1043                                  struct iomap *iomap)
1044 {
1045         struct gfs2_trans *tr = current->journal_info;
1046         struct gfs2_inode *ip = GFS2_I(inode);
1047         struct gfs2_sbd *sdp = GFS2_SB(inode);
1048 
1049         if (page && !gfs2_is_stuffed(ip))
1050                 gfs2_page_add_databufs(ip, page, offset_in_page(pos), copied);
1051 
1052         if (tr->tr_num_buf_new)
1053                 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1054 
1055         gfs2_trans_end(sdp);
1056 }
1057 
1058 static const struct iomap_page_ops gfs2_iomap_page_ops = {
1059         .page_prepare = gfs2_iomap_page_prepare,
1060         .page_done = gfs2_iomap_page_done,
1061 };
1062 
1063 static int gfs2_iomap_begin_write(struct inode *inode, loff_t pos,
1064                                   loff_t length, unsigned flags,
1065                                   struct iomap *iomap,
1066                                   struct metapath *mp)
1067 {
1068         struct gfs2_inode *ip = GFS2_I(inode);
1069         struct gfs2_sbd *sdp = GFS2_SB(inode);
1070         bool unstuff;
1071         int ret;
1072 
1073         unstuff = gfs2_is_stuffed(ip) &&
1074                   pos + length > gfs2_max_stuffed_size(ip);
1075 
1076         if (unstuff || iomap->type == IOMAP_HOLE) {
1077                 unsigned int data_blocks, ind_blocks;
1078                 struct gfs2_alloc_parms ap = {};
1079                 unsigned int rblocks;
1080                 struct gfs2_trans *tr;
1081 
1082                 gfs2_write_calc_reserv(ip, iomap->length, &data_blocks,
1083                                        &ind_blocks);
1084                 ap.target = data_blocks + ind_blocks;
1085                 ret = gfs2_quota_lock_check(ip, &ap);
1086                 if (ret)
1087                         return ret;
1088 
1089                 ret = gfs2_inplace_reserve(ip, &ap);
1090                 if (ret)
1091                         goto out_qunlock;
1092 
1093                 rblocks = RES_DINODE + ind_blocks;
1094                 if (gfs2_is_jdata(ip))
1095                         rblocks += data_blocks;
1096                 if (ind_blocks || data_blocks)
1097                         rblocks += RES_STATFS + RES_QUOTA;
1098                 if (inode == sdp->sd_rindex)
1099                         rblocks += 2 * RES_STATFS;
1100                 rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
1101 
1102                 ret = gfs2_trans_begin(sdp, rblocks,
1103                                        iomap->length >> inode->i_blkbits);
1104                 if (ret)
1105                         goto out_trans_fail;
1106 
1107                 if (unstuff) {
1108                         ret = gfs2_unstuff_dinode(ip, NULL);
1109                         if (ret)
1110                                 goto out_trans_end;
1111                         release_metapath(mp);
1112                         ret = gfs2_iomap_get(inode, iomap->offset,
1113                                              iomap->length, flags, iomap, mp);
1114                         if (ret)
1115                                 goto out_trans_end;
1116                 }
1117 
1118                 if (iomap->type == IOMAP_HOLE) {
1119                         ret = gfs2_iomap_alloc(inode, iomap, mp);
1120                         if (ret) {
1121                                 gfs2_trans_end(sdp);
1122                                 gfs2_inplace_release(ip);
1123                                 punch_hole(ip, iomap->offset, iomap->length);
1124                                 goto out_qunlock;
1125                         }
1126                 }
1127 
1128                 tr = current->journal_info;
1129                 if (tr->tr_num_buf_new)
1130                         __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1131 
1132                 gfs2_trans_end(sdp);
1133         }
1134 
1135         if (gfs2_is_stuffed(ip) || gfs2_is_jdata(ip))
1136                 iomap->page_ops = &gfs2_iomap_page_ops;
1137         return 0;
1138 
1139 out_trans_end:
1140         gfs2_trans_end(sdp);
1141 out_trans_fail:
1142         gfs2_inplace_release(ip);
1143 out_qunlock:
1144         gfs2_quota_unlock(ip);
1145         return ret;
1146 }
1147 
1148 static inline bool gfs2_iomap_need_write_lock(unsigned flags)
1149 {
1150         return (flags & IOMAP_WRITE) && !(flags & IOMAP_DIRECT);
1151 }
1152 
1153 static int gfs2_iomap_begin(struct inode *inode, loff_t pos, loff_t length,
1154                             unsigned flags, struct iomap *iomap)
1155 {
1156         struct gfs2_inode *ip = GFS2_I(inode);
1157         struct metapath mp = { .mp_aheight = 1, };
1158         int ret;
1159 
1160         iomap->flags |= IOMAP_F_BUFFER_HEAD;
1161 
1162         trace_gfs2_iomap_start(ip, pos, length, flags);
1163         if (gfs2_iomap_need_write_lock(flags)) {
1164                 ret = gfs2_write_lock(inode);
1165                 if (ret)
1166                         goto out;
1167         }
1168 
1169         ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp);
1170         if (ret)
1171                 goto out_unlock;
1172 
1173         switch(flags & (IOMAP_WRITE | IOMAP_ZERO)) {
1174         case IOMAP_WRITE:
1175                 if (flags & IOMAP_DIRECT) {
1176                         /*
1177                          * Silently fall back to buffered I/O for stuffed files
1178                          * or if we've got a hole (see gfs2_file_direct_write).
1179                          */
1180                         if (iomap->type != IOMAP_MAPPED)
1181                                 ret = -ENOTBLK;
1182                         goto out_unlock;
1183                 }
1184                 break;
1185         case IOMAP_ZERO:
1186                 if (iomap->type == IOMAP_HOLE)
1187                         goto out_unlock;
1188                 break;
1189         default:
1190                 goto out_unlock;
1191         }
1192 
1193         ret = gfs2_iomap_begin_write(inode, pos, length, flags, iomap, &mp);
1194 
1195 out_unlock:
1196         if (ret && gfs2_iomap_need_write_lock(flags))
1197                 gfs2_write_unlock(inode);
1198         release_metapath(&mp);
1199 out:
1200         trace_gfs2_iomap_end(ip, iomap, ret);
1201         return ret;
1202 }
1203 
1204 static int gfs2_iomap_end(struct inode *inode, loff_t pos, loff_t length,
1205                           ssize_t written, unsigned flags, struct iomap *iomap)
1206 {
1207         struct gfs2_inode *ip = GFS2_I(inode);
1208         struct gfs2_sbd *sdp = GFS2_SB(inode);
1209 
1210         switch (flags & (IOMAP_WRITE | IOMAP_ZERO)) {
1211         case IOMAP_WRITE:
1212                 if (flags & IOMAP_DIRECT)
1213                         return 0;
1214                 break;
1215         case IOMAP_ZERO:
1216                  if (iomap->type == IOMAP_HOLE)
1217                          return 0;
1218                  break;
1219         default:
1220                  return 0;
1221         }
1222 
1223         if (!gfs2_is_stuffed(ip))
1224                 gfs2_ordered_add_inode(ip);
1225 
1226         if (inode == sdp->sd_rindex)
1227                 adjust_fs_space(inode);
1228 
1229         gfs2_inplace_release(ip);
1230 
1231         if (length != written && (iomap->flags & IOMAP_F_NEW)) {
1232                 /* Deallocate blocks that were just allocated. */
1233                 loff_t blockmask = i_blocksize(inode) - 1;
1234                 loff_t end = (pos + length) & ~blockmask;
1235 
1236                 pos = (pos + written + blockmask) & ~blockmask;
1237                 if (pos < end) {
1238                         truncate_pagecache_range(inode, pos, end - 1);
1239                         punch_hole(ip, pos, end - pos);
1240                 }
1241         }
1242 
1243         if (ip->i_qadata && ip->i_qadata->qa_qd_num)
1244                 gfs2_quota_unlock(ip);
1245 
1246         if (unlikely(!written))
1247                 goto out_unlock;
1248 
1249         if (iomap->flags & IOMAP_F_SIZE_CHANGED)
1250                 mark_inode_dirty(inode);
1251         set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
1252 
1253 out_unlock:
1254         if (gfs2_iomap_need_write_lock(flags))
1255                 gfs2_write_unlock(inode);
1256         return 0;
1257 }
1258 
1259 const struct iomap_ops gfs2_iomap_ops = {
1260         .iomap_begin = gfs2_iomap_begin,
1261         .iomap_end = gfs2_iomap_end,
1262 };
1263 
1264 /**
1265  * gfs2_block_map - Map one or more blocks of an inode to a disk block
1266  * @inode: The inode
1267  * @lblock: The logical block number
1268  * @bh_map: The bh to be mapped
1269  * @create: True if its ok to alloc blocks to satify the request
1270  *
1271  * The size of the requested mapping is defined in bh_map->b_size.
1272  *
1273  * Clears buffer_mapped(bh_map) and leaves bh_map->b_size unchanged
1274  * when @lblock is not mapped.  Sets buffer_mapped(bh_map) and
1275  * bh_map->b_size to indicate the size of the mapping when @lblock and
1276  * successive blocks are mapped, up to the requested size.
1277  *
1278  * Sets buffer_boundary() if a read of metadata will be required
1279  * before the next block can be mapped. Sets buffer_new() if new
1280  * blocks were allocated.
1281  *
1282  * Returns: errno
1283  */
1284 
1285 int gfs2_block_map(struct inode *inode, sector_t lblock,
1286                    struct buffer_head *bh_map, int create)
1287 {
1288         struct gfs2_inode *ip = GFS2_I(inode);
1289         loff_t pos = (loff_t)lblock << inode->i_blkbits;
1290         loff_t length = bh_map->b_size;
1291         struct metapath mp = { .mp_aheight = 1, };
1292         struct iomap iomap = { };
1293         int ret;
1294 
1295         clear_buffer_mapped(bh_map);
1296         clear_buffer_new(bh_map);
1297         clear_buffer_boundary(bh_map);
1298         trace_gfs2_bmap(ip, bh_map, lblock, create, 1);
1299 
1300         if (create) {
1301                 ret = gfs2_iomap_get(inode, pos, length, IOMAP_WRITE, &iomap, &mp);
1302                 if (!ret && iomap.type == IOMAP_HOLE)
1303                         ret = gfs2_iomap_alloc(inode, &iomap, &mp);
1304                 release_metapath(&mp);
1305         } else {
1306                 ret = gfs2_iomap_get(inode, pos, length, 0, &iomap, &mp);
1307                 release_metapath(&mp);
1308         }
1309         if (ret)
1310                 goto out;
1311 
1312         if (iomap.length > bh_map->b_size) {
1313                 iomap.length = bh_map->b_size;
1314                 iomap.flags &= ~IOMAP_F_GFS2_BOUNDARY;
1315         }
1316         if (iomap.addr != IOMAP_NULL_ADDR)
1317                 map_bh(bh_map, inode->i_sb, iomap.addr >> inode->i_blkbits);
1318         bh_map->b_size = iomap.length;
1319         if (iomap.flags & IOMAP_F_GFS2_BOUNDARY)
1320                 set_buffer_boundary(bh_map);
1321         if (iomap.flags & IOMAP_F_NEW)
1322                 set_buffer_new(bh_map);
1323 
1324 out:
1325         trace_gfs2_bmap(ip, bh_map, lblock, create, ret);
1326         return ret;
1327 }
1328 
1329 /*
1330  * Deprecated: do not use in new code
1331  */
1332 int gfs2_extent_map(struct inode *inode, u64 lblock, int *new, u64 *dblock, unsigned *extlen)
1333 {
1334         struct buffer_head bh = { .b_state = 0, .b_blocknr = 0 };
1335         int ret;
1336         int create = *new;
1337 
1338         BUG_ON(!extlen);
1339         BUG_ON(!dblock);
1340         BUG_ON(!new);
1341 
1342         bh.b_size = BIT(inode->i_blkbits + (create ? 0 : 5));
1343         ret = gfs2_block_map(inode, lblock, &bh, create);
1344         *extlen = bh.b_size >> inode->i_blkbits;
1345         *dblock = bh.b_blocknr;
1346         if (buffer_new(&bh))
1347                 *new = 1;
1348         else
1349                 *new = 0;
1350         return ret;
1351 }
1352 
1353 static int gfs2_block_zero_range(struct inode *inode, loff_t from,
1354                                  unsigned int length)
1355 {
1356         return iomap_zero_range(inode, from, length, NULL, &gfs2_iomap_ops);
1357 }
1358 
1359 #define GFS2_JTRUNC_REVOKES 8192
1360 
1361 /**
1362  * gfs2_journaled_truncate - Wrapper for truncate_pagecache for jdata files
1363  * @inode: The inode being truncated
1364  * @oldsize: The original (larger) size
1365  * @newsize: The new smaller size
1366  *
1367  * With jdata files, we have to journal a revoke for each block which is
1368  * truncated. As a result, we need to split this into separate transactions
1369  * if the number of pages being truncated gets too large.
1370  */
1371 
1372 static int gfs2_journaled_truncate(struct inode *inode, u64 oldsize, u64 newsize)
1373 {
1374         struct gfs2_sbd *sdp = GFS2_SB(inode);
1375         u64 max_chunk = GFS2_JTRUNC_REVOKES * sdp->sd_vfs->s_blocksize;
1376         u64 chunk;
1377         int error;
1378 
1379         while (oldsize != newsize) {
1380                 struct gfs2_trans *tr;
1381                 unsigned int offs;
1382 
1383                 chunk = oldsize - newsize;
1384                 if (chunk > max_chunk)
1385                         chunk = max_chunk;
1386 
1387                 offs = oldsize & ~PAGE_MASK;
1388                 if (offs && chunk > PAGE_SIZE)
1389                         chunk = offs + ((chunk - offs) & PAGE_MASK);
1390 
1391                 truncate_pagecache(inode, oldsize - chunk);
1392                 oldsize -= chunk;
1393 
1394                 tr = current->journal_info;
1395                 if (!test_bit(TR_TOUCHED, &tr->tr_flags))
1396                         continue;
1397 
1398                 gfs2_trans_end(sdp);
1399                 error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES);
1400                 if (error)
1401                         return error;
1402         }
1403 
1404         return 0;
1405 }
1406 
1407 static int trunc_start(struct inode *inode, u64 newsize)
1408 {
1409         struct gfs2_inode *ip = GFS2_I(inode);
1410         struct gfs2_sbd *sdp = GFS2_SB(inode);
1411         struct buffer_head *dibh = NULL;
1412         int journaled = gfs2_is_jdata(ip);
1413         u64 oldsize = inode->i_size;
1414         int error;
1415 
1416         if (journaled)
1417                 error = gfs2_trans_begin(sdp, RES_DINODE + RES_JDATA, GFS2_JTRUNC_REVOKES);
1418         else
1419                 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1420         if (error)
1421                 return error;
1422 
1423         error = gfs2_meta_inode_buffer(ip, &dibh);
1424         if (error)
1425                 goto out;
1426 
1427         gfs2_trans_add_meta(ip->i_gl, dibh);
1428 
1429         if (gfs2_is_stuffed(ip)) {
1430                 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + newsize);
1431         } else {
1432                 unsigned int blocksize = i_blocksize(inode);
1433                 unsigned int offs = newsize & (blocksize - 1);
1434                 if (offs) {
1435                         error = gfs2_block_zero_range(inode, newsize,
1436                                                       blocksize - offs);
1437                         if (error)
1438                                 goto out;
1439                 }
1440                 ip->i_diskflags |= GFS2_DIF_TRUNC_IN_PROG;
1441         }
1442 
1443         i_size_write(inode, newsize);
1444         ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
1445         gfs2_dinode_out(ip, dibh->b_data);
1446 
1447         if (journaled)
1448                 error = gfs2_journaled_truncate(inode, oldsize, newsize);
1449         else
1450                 truncate_pagecache(inode, newsize);
1451 
1452 out:
1453         brelse(dibh);
1454         if (current->journal_info)
1455                 gfs2_trans_end(sdp);
1456         return error;
1457 }
1458 
1459 int gfs2_iomap_get_alloc(struct inode *inode, loff_t pos, loff_t length,
1460                          struct iomap *iomap)
1461 {
1462         struct metapath mp = { .mp_aheight = 1, };
1463         int ret;
1464 
1465         ret = gfs2_iomap_get(inode, pos, length, IOMAP_WRITE, iomap, &mp);
1466         if (!ret && iomap->type == IOMAP_HOLE)
1467                 ret = gfs2_iomap_alloc(inode, iomap, &mp);
1468         release_metapath(&mp);
1469         return ret;
1470 }
1471 
1472 /**
1473  * sweep_bh_for_rgrps - find an rgrp in a meta buffer and free blocks therein
1474  * @ip: inode
1475  * @rg_gh: holder of resource group glock
1476  * @bh: buffer head to sweep
1477  * @start: starting point in bh
1478  * @end: end point in bh
1479  * @meta: true if bh points to metadata (rather than data)
1480  * @btotal: place to keep count of total blocks freed
1481  *
1482  * We sweep a metadata buffer (provided by the metapath) for blocks we need to
1483  * free, and free them all. However, we do it one rgrp at a time. If this
1484  * block has references to multiple rgrps, we break it into individual
1485  * transactions. This allows other processes to use the rgrps while we're
1486  * focused on a single one, for better concurrency / performance.
1487  * At every transaction boundary, we rewrite the inode into the journal.
1488  * That way the bitmaps are kept consistent with the inode and we can recover
1489  * if we're interrupted by power-outages.
1490  *
1491  * Returns: 0, or return code if an error occurred.
1492  *          *btotal has the total number of blocks freed
1493  */
1494 static int sweep_bh_for_rgrps(struct gfs2_inode *ip, struct gfs2_holder *rd_gh,
1495                               struct buffer_head *bh, __be64 *start, __be64 *end,
1496                               bool meta, u32 *btotal)
1497 {
1498         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1499         struct gfs2_rgrpd *rgd;
1500         struct gfs2_trans *tr;
1501         __be64 *p;
1502         int blks_outside_rgrp;
1503         u64 bn, bstart, isize_blks;
1504         s64 blen; /* needs to be s64 or gfs2_add_inode_blocks breaks */
1505         int ret = 0;
1506         bool buf_in_tr = false; /* buffer was added to transaction */
1507 
1508 more_rgrps:
1509         rgd = NULL;
1510         if (gfs2_holder_initialized(rd_gh)) {
1511                 rgd = gfs2_glock2rgrp(rd_gh->gh_gl);
1512                 gfs2_assert_withdraw(sdp,
1513                              gfs2_glock_is_locked_by_me(rd_gh->gh_gl));
1514         }
1515         blks_outside_rgrp = 0;
1516         bstart = 0;
1517         blen = 0;
1518 
1519         for (p = start; p < end; p++) {
1520                 if (!*p)
1521                         continue;
1522                 bn = be64_to_cpu(*p);
1523 
1524                 if (rgd) {
1525                         if (!rgrp_contains_block(rgd, bn)) {
1526                                 blks_outside_rgrp++;
1527                                 continue;
1528                         }
1529                 } else {
1530                         rgd = gfs2_blk2rgrpd(sdp, bn, true);
1531                         if (unlikely(!rgd)) {
1532                                 ret = -EIO;
1533                                 goto out;
1534                         }
1535                         ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
1536                                                  0, rd_gh);
1537                         if (ret)
1538                                 goto out;
1539 
1540                         /* Must be done with the rgrp glock held: */
1541                         if (gfs2_rs_active(&ip->i_res) &&
1542                             rgd == ip->i_res.rs_rbm.rgd)
1543                                 gfs2_rs_deltree(&ip->i_res);
1544                 }
1545 
1546                 /* The size of our transactions will be unknown until we
1547                    actually process all the metadata blocks that relate to
1548                    the rgrp. So we estimate. We know it can't be more than
1549                    the dinode's i_blocks and we don't want to exceed the
1550                    journal flush threshold, sd_log_thresh2. */
1551                 if (current->journal_info == NULL) {
1552                         unsigned int jblocks_rqsted, revokes;
1553 
1554                         jblocks_rqsted = rgd->rd_length + RES_DINODE +
1555                                 RES_INDIRECT;
1556                         isize_blks = gfs2_get_inode_blocks(&ip->i_inode);
1557                         if (isize_blks > atomic_read(&sdp->sd_log_thresh2))
1558                                 jblocks_rqsted +=
1559                                         atomic_read(&sdp->sd_log_thresh2);
1560                         else
1561                                 jblocks_rqsted += isize_blks;
1562                         revokes = jblocks_rqsted;
1563                         if (meta)
1564                                 revokes += end - start;
1565                         else if (ip->i_depth)
1566                                 revokes += sdp->sd_inptrs;
1567                         ret = gfs2_trans_begin(sdp, jblocks_rqsted, revokes);
1568                         if (ret)
1569                                 goto out_unlock;
1570                         down_write(&ip->i_rw_mutex);
1571                 }
1572                 /* check if we will exceed the transaction blocks requested */
1573                 tr = current->journal_info;
1574                 if (tr->tr_num_buf_new + RES_STATFS +
1575                     RES_QUOTA >= atomic_read(&sdp->sd_log_thresh2)) {
1576                         /* We set blks_outside_rgrp to ensure the loop will
1577                            be repeated for the same rgrp, but with a new
1578                            transaction. */
1579                         blks_outside_rgrp++;
1580                         /* This next part is tricky. If the buffer was added
1581                            to the transaction, we've already set some block
1582                            pointers to 0, so we better follow through and free
1583                            them, or we will introduce corruption (so break).
1584                            This may be impossible, or at least rare, but I
1585                            decided to cover the case regardless.
1586 
1587                            If the buffer was not added to the transaction
1588                            (this call), doing so would exceed our transaction
1589                            size, so we need to end the transaction and start a
1590                            new one (so goto). */
1591 
1592                         if (buf_in_tr)
1593                                 break;
1594                         goto out_unlock;
1595                 }
1596 
1597                 gfs2_trans_add_meta(ip->i_gl, bh);
1598                 buf_in_tr = true;
1599                 *p = 0;
1600                 if (bstart + blen == bn) {
1601                         blen++;
1602                         continue;
1603                 }
1604                 if (bstart) {
1605                         __gfs2_free_blocks(ip, rgd, bstart, (u32)blen, meta);
1606                         (*btotal) += blen;
1607                         gfs2_add_inode_blocks(&ip->i_inode, -blen);
1608                 }
1609                 bstart = bn;
1610                 blen = 1;
1611         }
1612         if (bstart) {
1613                 __gfs2_free_blocks(ip, rgd, bstart, (u32)blen, meta);
1614                 (*btotal) += blen;
1615                 gfs2_add_inode_blocks(&ip->i_inode, -blen);
1616         }
1617 out_unlock:
1618         if (!ret && blks_outside_rgrp) { /* If buffer still has non-zero blocks
1619                                             outside the rgrp we just processed,
1620                                             do it all over again. */
1621                 if (current->journal_info) {
1622                         struct buffer_head *dibh;
1623 
1624                         ret = gfs2_meta_inode_buffer(ip, &dibh);
1625                         if (ret)
1626                                 goto out;
1627 
1628                         /* Every transaction boundary, we rewrite the dinode
1629                            to keep its di_blocks current in case of failure. */
1630                         ip->i_inode.i_mtime = ip->i_inode.i_ctime =
1631                                 current_time(&ip->i_inode);
1632                         gfs2_trans_add_meta(ip->i_gl, dibh);
1633                         gfs2_dinode_out(ip, dibh->b_data);
1634                         brelse(dibh);
1635                         up_write(&ip->i_rw_mutex);
1636                         gfs2_trans_end(sdp);
1637                         buf_in_tr = false;
1638                 }
1639                 gfs2_glock_dq_uninit(rd_gh);
1640                 cond_resched();
1641                 goto more_rgrps;
1642         }
1643 out:
1644         return ret;
1645 }
1646 
1647 static bool mp_eq_to_hgt(struct metapath *mp, __u16 *list, unsigned int h)
1648 {
1649         if (memcmp(mp->mp_list, list, h * sizeof(mp->mp_list[0])))
1650                 return false;
1651         return true;
1652 }
1653 
1654 /**
1655  * find_nonnull_ptr - find a non-null pointer given a metapath and height
1656  * @mp: starting metapath
1657  * @h: desired height to search
1658  *
1659  * Assumes the metapath is valid (with buffers) out to height h.
1660  * Returns: true if a non-null pointer was found in the metapath buffer
1661  *          false if all remaining pointers are NULL in the buffer
1662  */
1663 static bool find_nonnull_ptr(struct gfs2_sbd *sdp, struct metapath *mp,
1664                              unsigned int h,
1665                              __u16 *end_list, unsigned int end_aligned)
1666 {
1667         struct buffer_head *bh = mp->mp_bh[h];
1668         __be64 *first, *ptr, *end;
1669 
1670         first = metaptr1(h, mp);
1671         ptr = first + mp->mp_list[h];
1672         end = (__be64 *)(bh->b_data + bh->b_size);
1673         if (end_list && mp_eq_to_hgt(mp, end_list, h)) {
1674                 bool keep_end = h < end_aligned;
1675                 end = first + end_list[h] + keep_end;
1676         }
1677 
1678         while (ptr < end) {
1679                 if (*ptr) { /* if we have a non-null pointer */
1680                         mp->mp_list[h] = ptr - first;
1681                         h++;
1682                         if (h < GFS2_MAX_META_HEIGHT)
1683                                 mp->mp_list[h] = 0;
1684                         return true;
1685                 }
1686                 ptr++;
1687         }
1688         return false;
1689 }
1690 
1691 enum dealloc_states {
1692         DEALLOC_MP_FULL = 0,    /* Strip a metapath with all buffers read in */
1693         DEALLOC_MP_LOWER = 1,   /* lower the metapath strip height */
1694         DEALLOC_FILL_MP = 2,  /* Fill in the metapath to the given height. */
1695         DEALLOC_DONE = 3,       /* process complete */
1696 };
1697 
1698 static inline void
1699 metapointer_range(struct metapath *mp, int height,
1700                   __u16 *start_list, unsigned int start_aligned,
1701                   __u16 *end_list, unsigned int end_aligned,
1702                   __be64 **start, __be64 **end)
1703 {
1704         struct buffer_head *bh = mp->mp_bh[height];
1705         __be64 *first;
1706 
1707         first = metaptr1(height, mp);
1708         *start = first;
1709         if (mp_eq_to_hgt(mp, start_list, height)) {
1710                 bool keep_start = height < start_aligned;
1711                 *start = first + start_list[height] + keep_start;
1712         }
1713         *end = (__be64 *)(bh->b_data + bh->b_size);
1714         if (end_list && mp_eq_to_hgt(mp, end_list, height)) {
1715                 bool keep_end = height < end_aligned;
1716                 *end = first + end_list[height] + keep_end;
1717         }
1718 }
1719 
1720 static inline bool walk_done(struct gfs2_sbd *sdp,
1721                              struct metapath *mp, int height,
1722                              __u16 *end_list, unsigned int end_aligned)
1723 {
1724         __u16 end;
1725 
1726         if (end_list) {
1727                 bool keep_end = height < end_aligned;
1728                 if (!mp_eq_to_hgt(mp, end_list, height))
1729                         return false;
1730                 end = end_list[height] + keep_end;
1731         } else
1732                 end = (height > 0) ? sdp->sd_inptrs : sdp->sd_diptrs;
1733         return mp->mp_list[height] >= end;
1734 }
1735 
1736 /**
1737  * punch_hole - deallocate blocks in a file
1738  * @ip: inode to truncate
1739  * @offset: the start of the hole
1740  * @length: the size of the hole (or 0 for truncate)
1741  *
1742  * Punch a hole into a file or truncate a file at a given position.  This
1743  * function operates in whole blocks (@offset and @length are rounded
1744  * accordingly); partially filled blocks must be cleared otherwise.
1745  *
1746  * This function works from the bottom up, and from the right to the left. In
1747  * other words, it strips off the highest layer (data) before stripping any of
1748  * the metadata. Doing it this way is best in case the operation is interrupted
1749  * by power failure, etc.  The dinode is rewritten in every transaction to
1750  * guarantee integrity.
1751  */
1752 static int punch_hole(struct gfs2_inode *ip, u64 offset, u64 length)
1753 {
1754         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1755         u64 maxsize = sdp->sd_heightsize[ip->i_height];
1756         struct metapath mp = {};
1757         struct buffer_head *dibh, *bh;
1758         struct gfs2_holder rd_gh;
1759         unsigned int bsize_shift = sdp->sd_sb.sb_bsize_shift;
1760         u64 lblock = (offset + (1 << bsize_shift) - 1) >> bsize_shift;
1761         __u16 start_list[GFS2_MAX_META_HEIGHT];
1762         __u16 __end_list[GFS2_MAX_META_HEIGHT], *end_list = NULL;
1763         unsigned int start_aligned, uninitialized_var(end_aligned);
1764         unsigned int strip_h = ip->i_height - 1;
1765         u32 btotal = 0;
1766         int ret, state;
1767         int mp_h; /* metapath buffers are read in to this height */
1768         u64 prev_bnr = 0;
1769         __be64 *start, *end;
1770 
1771         if (offset >= maxsize) {
1772                 /*
1773                  * The starting point lies beyond the allocated meta-data;
1774                  * there are no blocks do deallocate.
1775                  */
1776                 return 0;
1777         }
1778 
1779         /*
1780          * The start position of the hole is defined by lblock, start_list, and
1781          * start_aligned.  The end position of the hole is defined by lend,
1782          * end_list, and end_aligned.
1783          *
1784          * start_aligned and end_aligned define down to which height the start
1785          * and end positions are aligned to the metadata tree (i.e., the
1786          * position is a multiple of the metadata granularity at the height
1787          * above).  This determines at which heights additional meta pointers
1788          * needs to be preserved for the remaining data.
1789          */
1790 
1791         if (length) {
1792                 u64 end_offset = offset + length;
1793                 u64 lend;
1794 
1795                 /*
1796                  * Clip the end at the maximum file size for the given height:
1797                  * that's how far the metadata goes; files bigger than that
1798                  * will have additional layers of indirection.
1799                  */
1800                 if (end_offset > maxsize)
1801                         end_offset = maxsize;
1802                 lend = end_offset >> bsize_shift;
1803 
1804                 if (lblock >= lend)
1805                         return 0;
1806 
1807                 find_metapath(sdp, lend, &mp, ip->i_height);
1808                 end_list = __end_list;
1809                 memcpy(end_list, mp.mp_list, sizeof(mp.mp_list));
1810 
1811                 for (mp_h = ip->i_height - 1; mp_h > 0; mp_h--) {
1812                         if (end_list[mp_h])
1813                                 break;
1814                 }
1815                 end_aligned = mp_h;
1816         }
1817 
1818         find_metapath(sdp, lblock, &mp, ip->i_height);
1819         memcpy(start_list, mp.mp_list, sizeof(start_list));
1820 
1821         for (mp_h = ip->i_height - 1; mp_h > 0; mp_h--) {
1822                 if (start_list[mp_h])
1823                         break;
1824         }
1825         start_aligned = mp_h;
1826 
1827         ret = gfs2_meta_inode_buffer(ip, &dibh);
1828         if (ret)
1829                 return ret;
1830 
1831         mp.mp_bh[0] = dibh;
1832         ret = lookup_metapath(ip, &mp);
1833         if (ret)
1834                 goto out_metapath;
1835 
1836         /* issue read-ahead on metadata */
1837         for (mp_h = 0; mp_h < mp.mp_aheight - 1; mp_h++) {
1838                 metapointer_range(&mp, mp_h, start_list, start_aligned,
1839                                   end_list, end_aligned, &start, &end);
1840                 gfs2_metapath_ra(ip->i_gl, start, end);
1841         }
1842 
1843         if (mp.mp_aheight == ip->i_height)
1844                 state = DEALLOC_MP_FULL; /* We have a complete metapath */
1845         else
1846                 state = DEALLOC_FILL_MP; /* deal with partial metapath */
1847 
1848         ret = gfs2_rindex_update(sdp);
1849         if (ret)
1850                 goto out_metapath;
1851 
1852         ret = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1853         if (ret)
1854                 goto out_metapath;
1855         gfs2_holder_mark_uninitialized(&rd_gh);
1856 
1857         mp_h = strip_h;
1858 
1859         while (state != DEALLOC_DONE) {
1860                 switch (state) {
1861                 /* Truncate a full metapath at the given strip height.
1862                  * Note that strip_h == mp_h in order to be in this state. */
1863                 case DEALLOC_MP_FULL:
1864                         bh = mp.mp_bh[mp_h];
1865                         gfs2_assert_withdraw(sdp, bh);
1866                         if (gfs2_assert_withdraw(sdp,
1867                                                  prev_bnr != bh->b_blocknr)) {
1868                                 fs_emerg(sdp, "inode %llu, block:%llu, i_h:%u,"
1869                                          "s_h:%u, mp_h:%u\n",
1870                                        (unsigned long long)ip->i_no_addr,
1871                                        prev_bnr, ip->i_height, strip_h, mp_h);
1872                         }
1873                         prev_bnr = bh->b_blocknr;
1874 
1875                         if (gfs2_metatype_check(sdp, bh,
1876                                                 (mp_h ? GFS2_METATYPE_IN :
1877                                                         GFS2_METATYPE_DI))) {
1878                                 ret = -EIO;
1879                                 goto out;
1880                         }
1881 
1882                         /*
1883                          * Below, passing end_aligned as 0 gives us the
1884                          * metapointer range excluding the end point: the end
1885                          * point is the first metapath we must not deallocate!
1886                          */
1887 
1888                         metapointer_range(&mp, mp_h, start_list, start_aligned,
1889                                           end_list, 0 /* end_aligned */,
1890                                           &start, &end);
1891                         ret = sweep_bh_for_rgrps(ip, &rd_gh, mp.mp_bh[mp_h],
1892                                                  start, end,
1893                                                  mp_h != ip->i_height - 1,
1894                                                  &btotal);
1895 
1896                         /* If we hit an error or just swept dinode buffer,
1897                            just exit. */
1898                         if (ret || !mp_h) {
1899                                 state = DEALLOC_DONE;
1900                                 break;
1901                         }
1902                         state = DEALLOC_MP_LOWER;
1903                         break;
1904 
1905                 /* lower the metapath strip height */
1906                 case DEALLOC_MP_LOWER:
1907                         /* We're done with the current buffer, so release it,
1908                            unless it's the dinode buffer. Then back up to the
1909                            previous pointer. */
1910                         if (mp_h) {
1911                                 brelse(mp.mp_bh[mp_h]);
1912                                 mp.mp_bh[mp_h] = NULL;
1913                         }
1914                         /* If we can't get any lower in height, we've stripped
1915                            off all we can. Next step is to back up and start
1916                            stripping the previous level of metadata. */
1917                         if (mp_h == 0) {
1918                                 strip_h--;
1919                                 memcpy(mp.mp_list, start_list, sizeof(start_list));
1920                                 mp_h = strip_h;
1921                                 state = DEALLOC_FILL_MP;
1922                                 break;
1923                         }
1924                         mp.mp_list[mp_h] = 0;
1925                         mp_h--; /* search one metadata height down */
1926                         mp.mp_list[mp_h]++;
1927                         if (walk_done(sdp, &mp, mp_h, end_list, end_aligned))
1928                                 break;
1929                         /* Here we've found a part of the metapath that is not
1930                          * allocated. We need to search at that height for the
1931                          * next non-null pointer. */
1932                         if (find_nonnull_ptr(sdp, &mp, mp_h, end_list, end_aligned)) {
1933                                 state = DEALLOC_FILL_MP;
1934                                 mp_h++;
1935                         }
1936                         /* No more non-null pointers at this height. Back up
1937                            to the previous height and try again. */
1938                         break; /* loop around in the same state */
1939 
1940                 /* Fill the metapath with buffers to the given height. */
1941                 case DEALLOC_FILL_MP:
1942                         /* Fill the buffers out to the current height. */
1943                         ret = fillup_metapath(ip, &mp, mp_h);
1944                         if (ret < 0)
1945                                 goto out;
1946 
1947                         /* On the first pass, issue read-ahead on metadata. */
1948                         if (mp.mp_aheight > 1 && strip_h == ip->i_height - 1) {
1949                                 unsigned int height = mp.mp_aheight - 1;
1950 
1951                                 /* No read-ahead for data blocks. */
1952                                 if (mp.mp_aheight - 1 == strip_h)
1953                                         height--;
1954 
1955                                 for (; height >= mp.mp_aheight - ret; height--) {
1956                                         metapointer_range(&mp, height,
1957                                                           start_list, start_aligned,
1958                                                           end_list, end_aligned,
1959                                                           &start, &end);
1960                                         gfs2_metapath_ra(ip->i_gl, start, end);
1961                                 }
1962                         }
1963 
1964                         /* If buffers found for the entire strip height */
1965                         if (mp.mp_aheight - 1 == strip_h) {
1966                                 state = DEALLOC_MP_FULL;
1967                                 break;
1968                         }
1969                         if (mp.mp_aheight < ip->i_height) /* We have a partial height */
1970                                 mp_h = mp.mp_aheight - 1;
1971 
1972                         /* If we find a non-null block pointer, crawl a bit
1973                            higher up in the metapath and try again, otherwise
1974                            we need to look lower for a new starting point. */
1975                         if (find_nonnull_ptr(sdp, &mp, mp_h, end_list, end_aligned))
1976                                 mp_h++;
1977                         else
1978                                 state = DEALLOC_MP_LOWER;
1979                         break;
1980                 }
1981         }
1982 
1983         if (btotal) {
1984                 if (current->journal_info == NULL) {
1985                         ret = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS +
1986                                                RES_QUOTA, 0);
1987                         if (ret)
1988                                 goto out;
1989                         down_write(&ip->i_rw_mutex);
1990                 }
1991                 gfs2_statfs_change(sdp, 0, +btotal, 0);
1992                 gfs2_quota_change(ip, -(s64)btotal, ip->i_inode.i_uid,
1993                                   ip->i_inode.i_gid);
1994                 ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
1995                 gfs2_trans_add_meta(ip->i_gl, dibh);
1996                 gfs2_dinode_out(ip, dibh->b_data);
1997                 up_write(&ip->i_rw_mutex);
1998                 gfs2_trans_end(sdp);
1999         }
2000 
2001 out:
2002         if (gfs2_holder_initialized(&rd_gh))
2003                 gfs2_glock_dq_uninit(&rd_gh);
2004         if (current->journal_info) {
2005                 up_write(&ip->i_rw_mutex);
2006                 gfs2_trans_end(sdp);
2007                 cond_resched();
2008         }
2009         gfs2_quota_unhold(ip);
2010 out_metapath:
2011         release_metapath(&mp);
2012         return ret;
2013 }
2014 
2015 static int trunc_end(struct gfs2_inode *ip)
2016 {
2017         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
2018         struct buffer_head *dibh;
2019         int error;
2020 
2021         error = gfs2_trans_begin(sdp, RES_DINODE, 0);
2022         if (error)
2023                 return error;
2024 
2025         down_write(&ip->i_rw_mutex);
2026 
2027         error = gfs2_meta_inode_buffer(ip, &dibh);
2028         if (error)
2029                 goto out;
2030 
2031         if (!i_size_read(&ip->i_inode)) {
2032                 ip->i_height = 0;
2033                 ip->i_goal = ip->i_no_addr;
2034                 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
2035                 gfs2_ordered_del_inode(ip);
2036         }
2037         ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
2038         ip->i_diskflags &= ~GFS2_DIF_TRUNC_IN_PROG;
2039 
2040         gfs2_trans_add_meta(ip->i_gl, dibh);
2041         gfs2_dinode_out(ip, dibh->b_data);
2042         brelse(dibh);
2043 
2044 out:
2045         up_write(&ip->i_rw_mutex);
2046         gfs2_trans_end(sdp);
2047         return error;
2048 }
2049 
2050 /**
2051  * do_shrink - make a file smaller
2052  * @inode: the inode
2053  * @newsize: the size to make the file
2054  *
2055  * Called with an exclusive lock on @inode. The @size must
2056  * be equal to or smaller than the current inode size.
2057  *
2058  * Returns: errno
2059  */
2060 
2061 static int do_shrink(struct inode *inode, u64 newsize)
2062 {
2063         struct gfs2_inode *ip = GFS2_I(inode);
2064         int error;
2065 
2066         error = trunc_start(inode, newsize);
2067         if (error < 0)
2068                 return error;
2069         if (gfs2_is_stuffed(ip))
2070                 return 0;
2071 
2072         error = punch_hole(ip, newsize, 0);
2073         if (error == 0)
2074                 error = trunc_end(ip);
2075 
2076         return error;
2077 }
2078 
2079 void gfs2_trim_blocks(struct inode *inode)
2080 {
2081         int ret;
2082 
2083         ret = do_shrink(inode, inode->i_size);
2084         WARN_ON(ret != 0);
2085 }
2086 
2087 /**
2088  * do_grow - Touch and update inode size
2089  * @inode: The inode
2090  * @size: The new size
2091  *
2092  * This function updates the timestamps on the inode and
2093  * may also increase the size of the inode. This function
2094  * must not be called with @size any smaller than the current
2095  * inode size.
2096  *
2097  * Although it is not strictly required to unstuff files here,
2098  * earlier versions of GFS2 have a bug in the stuffed file reading
2099  * code which will result in a buffer overrun if the size is larger
2100  * than the max stuffed file size. In order to prevent this from
2101  * occurring, such files are unstuffed, but in other cases we can
2102  * just update the inode size directly.
2103  *
2104  * Returns: 0 on success, or -ve on error
2105  */
2106 
2107 static int do_grow(struct inode *inode, u64 size)
2108 {
2109         struct gfs2_inode *ip = GFS2_I(inode);
2110         struct gfs2_sbd *sdp = GFS2_SB(inode);
2111         struct gfs2_alloc_parms ap = { .target = 1, };
2112         struct buffer_head *dibh;
2113         int error;
2114         int unstuff = 0;
2115 
2116         if (gfs2_is_stuffed(ip) && size > gfs2_max_stuffed_size(ip)) {
2117                 error = gfs2_quota_lock_check(ip, &ap);
2118                 if (error)
2119                         return error;
2120 
2121                 error = gfs2_inplace_reserve(ip, &ap);
2122                 if (error)
2123                         goto do_grow_qunlock;
2124                 unstuff = 1;
2125         }
2126 
2127         error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT +
2128                                  (unstuff &&
2129                                   gfs2_is_jdata(ip) ? RES_JDATA : 0) +
2130                                  (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF ?
2131                                   0 : RES_QUOTA), 0);
2132         if (error)
2133                 goto do_grow_release;
2134 
2135         if (unstuff) {
2136                 error = gfs2_unstuff_dinode(ip, NULL);
2137                 if (error)
2138                         goto do_end_trans;
2139         }
2140 
2141         error = gfs2_meta_inode_buffer(ip, &dibh);
2142         if (error)
2143                 goto do_end_trans;
2144 
2145         truncate_setsize(inode, size);
2146         ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
2147         gfs2_trans_add_meta(ip->i_gl, dibh);
2148         gfs2_dinode_out(ip, dibh->b_data);
2149         brelse(dibh);
2150 
2151 do_end_trans:
2152         gfs2_trans_end(sdp);
2153 do_grow_release:
2154         if (unstuff) {
2155                 gfs2_inplace_release(ip);
2156 do_grow_qunlock:
2157                 gfs2_quota_unlock(ip);
2158         }
2159         return error;
2160 }
2161 
2162 /**
2163  * gfs2_setattr_size - make a file a given size
2164  * @inode: the inode
2165  * @newsize: the size to make the file
2166  *
2167  * The file size can grow, shrink, or stay the same size. This
2168  * is called holding i_rwsem and an exclusive glock on the inode
2169  * in question.
2170  *
2171  * Returns: errno
2172  */
2173 
2174 int gfs2_setattr_size(struct inode *inode, u64 newsize)
2175 {
2176         struct gfs2_inode *ip = GFS2_I(inode);
2177         int ret;
2178 
2179         BUG_ON(!S_ISREG(inode->i_mode));
2180 
2181         ret = inode_newsize_ok(inode, newsize);
2182         if (ret)
2183                 return ret;
2184 
2185         inode_dio_wait(inode);
2186 
2187         ret = gfs2_rsqa_alloc(ip);
2188         if (ret)
2189                 goto out;
2190 
2191         if (newsize >= inode->i_size) {
2192                 ret = do_grow(inode, newsize);
2193                 goto out;
2194         }
2195 
2196         ret = do_shrink(inode, newsize);
2197 out:
2198         gfs2_rsqa_delete(ip, NULL);
2199         return ret;
2200 }
2201 
2202 int gfs2_truncatei_resume(struct gfs2_inode *ip)
2203 {
2204         int error;
2205         error = punch_hole(ip, i_size_read(&ip->i_inode), 0);
2206         if (!error)
2207                 error = trunc_end(ip);
2208         return error;
2209 }
2210 
2211 int gfs2_file_dealloc(struct gfs2_inode *ip)
2212 {
2213         return punch_hole(ip, 0, 0);
2214 }
2215 
2216 /**
2217  * gfs2_free_journal_extents - Free cached journal bmap info
2218  * @jd: The journal
2219  *
2220  */
2221 
2222 void gfs2_free_journal_extents(struct gfs2_jdesc *jd)
2223 {
2224         struct gfs2_journal_extent *jext;
2225 
2226         while(!list_empty(&jd->extent_list)) {
2227                 jext = list_entry(jd->extent_list.next, struct gfs2_journal_extent, list);
2228                 list_del(&jext->list);
2229                 kfree(jext);
2230         }
2231 }
2232 
2233 /**
2234  * gfs2_add_jextent - Add or merge a new extent to extent cache
2235  * @jd: The journal descriptor
2236  * @lblock: The logical block at start of new extent
2237  * @dblock: The physical block at start of new extent
2238  * @blocks: Size of extent in fs blocks
2239  *
2240  * Returns: 0 on success or -ENOMEM
2241  */
2242 
2243 static int gfs2_add_jextent(struct gfs2_jdesc *jd, u64 lblock, u64 dblock, u64 blocks)
2244 {
2245         struct gfs2_journal_extent *jext;
2246 
2247         if (!list_empty(&jd->extent_list)) {
2248                 jext = list_entry(jd->extent_list.prev, struct gfs2_journal_extent, list);
2249                 if ((jext->dblock + jext->blocks) == dblock) {
2250                         jext->blocks += blocks;
2251                         return 0;
2252                 }
2253         }
2254 
2255         jext = kzalloc(sizeof(struct gfs2_journal_extent), GFP_NOFS);
2256         if (jext == NULL)
2257                 return -ENOMEM;
2258         jext->dblock = dblock;
2259         jext->lblock = lblock;
2260         jext->blocks = blocks;
2261         list_add_tail(&jext->list, &jd->extent_list);
2262         jd->nr_extents++;
2263         return 0;
2264 }
2265 
2266 /**
2267  * gfs2_map_journal_extents - Cache journal bmap info
2268  * @sdp: The super block
2269  * @jd: The journal to map
2270  *
2271  * Create a reusable "extent" mapping from all logical
2272  * blocks to all physical blocks for the given journal.  This will save
2273  * us time when writing journal blocks.  Most journals will have only one
2274  * extent that maps all their logical blocks.  That's because gfs2.mkfs
2275  * arranges the journal blocks sequentially to maximize performance.
2276  * So the extent would map the first block for the entire file length.
2277  * However, gfs2_jadd can happen while file activity is happening, so
2278  * those journals may not be sequential.  Less likely is the case where
2279  * the users created their own journals by mounting the metafs and
2280  * laying it out.  But it's still possible.  These journals might have
2281  * several extents.
2282  *
2283  * Returns: 0 on success, or error on failure
2284  */
2285 
2286 int gfs2_map_journal_extents(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
2287 {
2288         u64 lblock = 0;
2289         u64 lblock_stop;
2290         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
2291         struct buffer_head bh;
2292         unsigned int shift = sdp->sd_sb.sb_bsize_shift;
2293         u64 size;
2294         int rc;
2295         ktime_t start, end;
2296 
2297         start = ktime_get();
2298         lblock_stop = i_size_read(jd->jd_inode) >> shift;
2299         size = (lblock_stop - lblock) << shift;
2300         jd->nr_extents = 0;
2301         WARN_ON(!list_empty(&jd->extent_list));
2302 
2303         do {
2304                 bh.b_state = 0;
2305                 bh.b_blocknr = 0;
2306                 bh.b_size = size;
2307                 rc = gfs2_block_map(jd->jd_inode, lblock, &bh, 0);
2308                 if (rc || !buffer_mapped(&bh))
2309                         goto fail;
2310                 rc = gfs2_add_jextent(jd, lblock, bh.b_blocknr, bh.b_size >> shift);
2311                 if (rc)
2312                         goto fail;
2313                 size -= bh.b_size;
2314                 lblock += (bh.b_size >> ip->i_inode.i_blkbits);
2315         } while(size > 0);
2316 
2317         end = ktime_get();
2318         fs_info(sdp, "journal %d mapped with %u extents in %lldms\n", jd->jd_jid,
2319                 jd->nr_extents, ktime_ms_delta(end, start));
2320         return 0;
2321 
2322 fail:
2323         fs_warn(sdp, "error %d mapping journal %u at offset %llu (extent %u)\n",
2324                 rc, jd->jd_jid,
2325                 (unsigned long long)(i_size_read(jd->jd_inode) - size),
2326                 jd->nr_extents);
2327         fs_warn(sdp, "bmap=%d lblock=%llu block=%llu, state=0x%08lx, size=%llu\n",
2328                 rc, (unsigned long long)lblock, (unsigned long long)bh.b_blocknr,
2329                 bh.b_state, (unsigned long long)bh.b_size);
2330         gfs2_free_journal_extents(jd);
2331         return rc;
2332 }
2333 
2334 /**
2335  * gfs2_write_alloc_required - figure out if a write will require an allocation
2336  * @ip: the file being written to
2337  * @offset: the offset to write to
2338  * @len: the number of bytes being written
2339  *
2340  * Returns: 1 if an alloc is required, 0 otherwise
2341  */
2342 
2343 int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset,
2344                               unsigned int len)
2345 {
2346         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
2347         struct buffer_head bh;
2348         unsigned int shift;
2349         u64 lblock, lblock_stop, size;
2350         u64 end_of_file;
2351 
2352         if (!len)
2353                 return 0;
2354 
2355         if (gfs2_is_stuffed(ip)) {
2356                 if (offset + len > gfs2_max_stuffed_size(ip))
2357                         return 1;
2358                 return 0;
2359         }
2360 
2361         shift = sdp->sd_sb.sb_bsize_shift;
2362         BUG_ON(gfs2_is_dir(ip));
2363         end_of_file = (i_size_read(&ip->i_inode) + sdp->sd_sb.sb_bsize - 1) >> shift;
2364         lblock = offset >> shift;
2365         lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift;
2366         if (lblock_stop > end_of_file && ip != GFS2_I(sdp->sd_rindex))
2367                 return 1;
2368 
2369         size = (lblock_stop - lblock) << shift;
2370         do {
2371                 bh.b_state = 0;
2372                 bh.b_size = size;
2373                 gfs2_block_map(&ip->i_inode, lblock, &bh, 0);
2374                 if (!buffer_mapped(&bh))
2375                         return 1;
2376                 size -= bh.b_size;
2377                 lblock += (bh.b_size >> ip->i_inode.i_blkbits);
2378         } while(size > 0);
2379 
2380         return 0;
2381 }
2382 
2383 static int stuffed_zero_range(struct inode *inode, loff_t offset, loff_t length)
2384 {
2385         struct gfs2_inode *ip = GFS2_I(inode);
2386         struct buffer_head *dibh;
2387         int error;
2388 
2389         if (offset >= inode->i_size)
2390                 return 0;
2391         if (offset + length > inode->i_size)
2392                 length = inode->i_size - offset;
2393 
2394         error = gfs2_meta_inode_buffer(ip, &dibh);
2395         if (error)
2396                 return error;
2397         gfs2_trans_add_meta(ip->i_gl, dibh);
2398         memset(dibh->b_data + sizeof(struct gfs2_dinode) + offset, 0,
2399                length);
2400         brelse(dibh);
2401         return 0;
2402 }
2403 
2404 static int gfs2_journaled_truncate_range(struct inode *inode, loff_t offset,
2405                                          loff_t length)
2406 {
2407         struct gfs2_sbd *sdp = GFS2_SB(inode);
2408         loff_t max_chunk = GFS2_JTRUNC_REVOKES * sdp->sd_vfs->s_blocksize;
2409         int error;
2410 
2411         while (length) {
2412                 struct gfs2_trans *tr;
2413                 loff_t chunk;
2414                 unsigned int offs;
2415 
2416                 chunk = length;
2417                 if (chunk > max_chunk)
2418                         chunk = max_chunk;
2419 
2420                 offs = offset & ~PAGE_MASK;
2421                 if (offs && chunk > PAGE_SIZE)
2422                         chunk = offs + ((chunk - offs) & PAGE_MASK);
2423 
2424                 truncate_pagecache_range(inode, offset, chunk);
2425                 offset += chunk;
2426                 length -= chunk;
2427 
2428                 tr = current->journal_info;
2429                 if (!test_bit(TR_TOUCHED, &tr->tr_flags))
2430                         continue;
2431 
2432                 gfs2_trans_end(sdp);
2433                 error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES);
2434                 if (error)
2435                         return error;
2436         }
2437         return 0;
2438 }
2439 
2440 int __gfs2_punch_hole(struct file *file, loff_t offset, loff_t length)
2441 {
2442         struct inode *inode = file_inode(file);
2443         struct gfs2_inode *ip = GFS2_I(inode);
2444         struct gfs2_sbd *sdp = GFS2_SB(inode);
2445         int error;
2446 
2447         if (gfs2_is_jdata(ip))
2448                 error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_JDATA,
2449                                          GFS2_JTRUNC_REVOKES);
2450         else
2451                 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
2452         if (error)
2453                 return error;
2454 
2455         if (gfs2_is_stuffed(ip)) {
2456                 error = stuffed_zero_range(inode, offset, length);
2457                 if (error)
2458                         goto out;
2459         } else {
2460                 unsigned int start_off, end_len, blocksize;
2461 
2462                 blocksize = i_blocksize(inode);
2463                 start_off = offset & (blocksize - 1);
2464                 end_len = (offset + length) & (blocksize - 1);
2465                 if (start_off) {
2466                         unsigned int len = length;
2467                         if (length > blocksize - start_off)
2468                                 len = blocksize - start_off;
2469                         error = gfs2_block_zero_range(inode, offset, len);
2470                         if (error)
2471                                 goto out;
2472                         if (start_off + length < blocksize)
2473                                 end_len = 0;
2474                 }
2475                 if (end_len) {
2476                         error = gfs2_block_zero_range(inode,
2477                                 offset + length - end_len, end_len);
2478                         if (error)
2479                                 goto out;
2480                 }
2481         }
2482 
2483         if (gfs2_is_jdata(ip)) {
2484                 BUG_ON(!current->journal_info);
2485                 gfs2_journaled_truncate_range(inode, offset, length);
2486         } else
2487                 truncate_pagecache_range(inode, offset, offset + length - 1);
2488 
2489         file_update_time(file);
2490         mark_inode_dirty(inode);
2491 
2492         if (current->journal_info)
2493                 gfs2_trans_end(sdp);
2494 
2495         if (!gfs2_is_stuffed(ip))
2496                 error = punch_hole(ip, offset, length);
2497 
2498 out:
2499         if (current->journal_info)
2500                 gfs2_trans_end(sdp);
2501         return error;
2502 }

/* [<][>][^][v][top][bottom][index][help] */