fs/gfs2/bmap.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
   4  *
   5  * This copyrighted material is made available to anyone wishing to use,
   6  * modify, copy, or redistribute it subject to the terms and conditions
   7  * of the GNU General Public License version 2.
   8  */
   9
  10 #include <linux/spinlock.h>
  11 #include <linux/completion.h>
  12 #include <linux/buffer_head.h>
  13 #include <linux/gfs2_ondisk.h>
  14 #include <linux/crc32.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 "super.h"
  25 #include "trans.h"
  26 #include "dir.h"
  27 #include "util.h"
  28 #include "trace_gfs2.h"
  29
  30 /* This doesn't need to be that large as max 64 bit pointers in a 4k
  31  * block is 512, so __u16 is fine for that. It saves stack space to
  32  * keep it small.
  33  */
  34 struct metapath {
  35         struct buffer_head *mp_bh[GFS2_MAX_META_HEIGHT];
  36         __u16 mp_list[GFS2_MAX_META_HEIGHT];
  37 };
  38
  39 typedef int (*block_call_t) (struct gfs2_inode *ip, struct buffer_head *dibh,
  40                              struct buffer_head *bh, __be64 *top,
  41                              __be64 *bottom, unsigned int height,
  42                              void *data);
  43
  44 struct strip_mine {
  45         int sm_first;
  46         unsigned int sm_height;
  47 };
  48
  49 /**
  50  * gfs2_unstuffer_page - unstuff a stuffed inode into a block cached by a page
  51  * @ip: the inode
  52  * @dibh: the dinode buffer
  53  * @block: the block number that was allocated
  54  * @page: The (optional) page. This is looked up if @page is NULL
  55  *
  56  * Returns: errno
  57  */
  58
  59 static int gfs2_unstuffer_page(struct gfs2_inode *ip, struct buffer_head *dibh,
  60                                u64 block, struct page *page)
  61 {
  62         struct inode *inode = &ip->i_inode;
  63         struct buffer_head *bh;
  64         int release = 0;
  65
  66         if (!page || page->index) {
  67                 page = grab_cache_page(inode->i_mapping, 0);
  68                 if (!page)
  69                         return -ENOMEM;
  70                 release = 1;
  71         }
  72
  73         if (!PageUptodate(page)) {
  74                 void *kaddr = kmap(page);
  75                 u64 dsize = i_size_read(inode);
  76
  77                 if (dsize > (dibh->b_size - sizeof(struct gfs2_dinode)))
  78                         dsize = dibh->b_size - sizeof(struct gfs2_dinode);
  79
  80                 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
  81                 memset(kaddr + dsize, 0, PAGE_CACHE_SIZE - dsize);
  82                 kunmap(page);
  83
  84                 SetPageUptodate(page);
  85         }
  86
  87         if (!page_has_buffers(page))
  88                 create_empty_buffers(page, 1 << inode->i_blkbits,
  89                                      (1 << BH_Uptodate));
  90
  91         bh = page_buffers(page);
  92
  93         if (!buffer_mapped(bh))
  94                 map_bh(bh, inode->i_sb, block);
  95
  96         set_buffer_uptodate(bh);
  97         if (!gfs2_is_jdata(ip))
  98                 mark_buffer_dirty(bh);
  99         if (!gfs2_is_writeback(ip))
 100                 gfs2_trans_add_bh(ip->i_gl, bh, 0);
 101
 102         if (release) {
 103                 unlock_page(page);
 104                 page_cache_release(page);
 105         }
 106
 107         return 0;
 108 }
 109
 110 /**
 111  * gfs2_unstuff_dinode - Unstuff a dinode when the data has grown too big
 112  * @ip: The GFS2 inode to unstuff
 113  * @page: The (optional) page. This is looked up if the @page is NULL
 114  *
 115  * This routine unstuffs a dinode and returns it to a "normal" state such
 116  * that the height can be grown in the traditional way.
 117  *
 118  * Returns: errno
 119  */
 120
 121 int gfs2_unstuff_dinode(struct gfs2_inode *ip, struct page *page)
 122 {
 123         struct buffer_head *bh, *dibh;
 124         struct gfs2_dinode *di;
 125         u64 block = 0;
 126         int isdir = gfs2_is_dir(ip);
 127         int error;
 128
 129         down_write(&ip->i_rw_mutex);
 130
 131         error = gfs2_meta_inode_buffer(ip, &dibh);
 132         if (error)
 133                 goto out;
 134
 135         if (i_size_read(&ip->i_inode)) {
 136                 /* Get a free block, fill it with the stuffed data,
 137                    and write it out to disk */
 138
 139                 unsigned int n = 1;
 140                 error = gfs2_alloc_block(ip, &block, &n);
 141                 if (error)
 142                         goto out_brelse;
 143                 if (isdir) {
 144                         gfs2_trans_add_unrevoke(GFS2_SB(&ip->i_inode), block, 1);
 145                         error = gfs2_dir_get_new_buffer(ip, block, &bh);
 146                         if (error)
 147                                 goto out_brelse;
 148                         gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_meta_header),
 149                                               dibh, sizeof(struct gfs2_dinode));
 150                         brelse(bh);
 151                 } else {
 152                         error = gfs2_unstuffer_page(ip, dibh, block, page);
 153                         if (error)
 154                                 goto out_brelse;
 155                 }
 156         }
 157
 158         /*  Set up the pointer to the new block  */
 159
 160         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 161         di = (struct gfs2_dinode *)dibh->b_data;
 162         gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
 163
 164         if (i_size_read(&ip->i_inode)) {
 165                 *(__be64 *)(di + 1) = cpu_to_be64(block);
 166                 gfs2_add_inode_blocks(&ip->i_inode, 1);
 167                 di->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
 168         }
 169
 170         ip->i_height = 1;
 171         di->di_height = cpu_to_be16(1);
 172
 173 out_brelse:
 174         brelse(dibh);
 175 out:
 176         up_write(&ip->i_rw_mutex);
 177         return error;
 178 }
 179
 180
 181 /**
 182  * find_metapath - Find path through the metadata tree
 183  * @sdp: The superblock
 184  * @mp: The metapath to return the result in
 185  * @block: The disk block to look up
 186  * @height: The pre-calculated height of the metadata tree
 187  *
 188  *   This routine returns a struct metapath structure that defines a path
 189  *   through the metadata of inode "ip" to get to block "block".
 190  *
 191  *   Example:
 192  *   Given:  "ip" is a height 3 file, "offset" is 101342453, and this is a
 193  *   filesystem with a blocksize of 4096.
 194  *
 195  *   find_metapath() would return a struct metapath structure set to:
 196  *   mp_offset = 101342453, mp_height = 3, mp_list[0] = 0, mp_list[1] = 48,
 197  *   and mp_list[2] = 165.
 198  *
 199  *   That means that in order to get to the block containing the byte at
 200  *   offset 101342453, we would load the indirect block pointed to by pointer
 201  *   0 in the dinode.  We would then load the indirect block pointed to by
 202  *   pointer 48 in that indirect block.  We would then load the data block
 203  *   pointed to by pointer 165 in that indirect block.
 204  *
 205  *             ----------------------------------------
 206  *             | Dinode |                             |
 207  *             |        |                            4|
 208  *             |        |0 1 2 3 4 5                 9|
 209  *             |        |                            6|
 210  *             ----------------------------------------
 211  *                       |
 212  *                       |
 213  *                       V
 214  *             ----------------------------------------
 215  *             | Indirect Block                       |
 216  *             |                                     5|
 217  *             |            4 4 4 4 4 5 5            1|
 218  *             |0           5 6 7 8 9 0 1            2|
 219  *             ----------------------------------------
 220  *                                |
 221  *                                |
 222  *                                V
 223  *             ----------------------------------------
 224  *             | Indirect Block                       |
 225  *             |                         1 1 1 1 1   5|
 226  *             |                         6 6 6 6 6   1|
 227  *             |0                        3 4 5 6 7   2|
 228  *             ----------------------------------------
 229  *                                           |
 230  *                                           |
 231  *                                           V
 232  *             ----------------------------------------
 233  *             | Data block containing offset         |
 234  *             |            101342453                 |
 235  *             |                                      |
 236  *             |                                      |
 237  *             ----------------------------------------
 238  *
 239  */
 240
 241 static void find_metapath(const struct gfs2_sbd *sdp, u64 block,
 242                           struct metapath *mp, unsigned int height)
 243 {
 244         unsigned int i;
 245
 246         for (i = height; i--;)
 247                 mp->mp_list[i] = do_div(block, sdp->sd_inptrs);
 248
 249 }
 250
 251 static inline unsigned int metapath_branch_start(const struct metapath *mp)
 252 {
 253         if (mp->mp_list[0] == 0)
 254                 return 2;
 255         return 1;
 256 }
 257
 258 /**
 259  * metapointer - Return pointer to start of metadata in a buffer
 260  * @height: The metadata height (0 = dinode)
 261  * @mp: The metapath
 262  *
 263  * Return a pointer to the block number of the next height of the metadata
 264  * tree given a buffer containing the pointer to the current height of the
 265  * metadata tree.
 266  */
 267
 268 static inline __be64 *metapointer(unsigned int height, const struct metapath *mp)
 269 {
 270         struct buffer_head *bh = mp->mp_bh[height];
 271         unsigned int head_size = (height > 0) ?
 272                 sizeof(struct gfs2_meta_header) : sizeof(struct gfs2_dinode);
 273         return ((__be64 *)(bh->b_data + head_size)) + mp->mp_list[height];
 274 }
 275
 276 /**
 277  * lookup_metapath - Walk the metadata tree to a specific point
 278  * @ip: The inode
 279  * @mp: The metapath
 280  *
 281  * Assumes that the inode's buffer has already been looked up and
 282  * hooked onto mp->mp_bh[0] and that the metapath has been initialised
 283  * by find_metapath().
 284  *
 285  * If this function encounters part of the tree which has not been
 286  * allocated, it returns the current height of the tree at the point
 287  * at which it found the unallocated block. Blocks which are found are
 288  * added to the mp->mp_bh[] list.
 289  *
 290  * Returns: error or height of metadata tree
 291  */
 292
 293 static int lookup_metapath(struct gfs2_inode *ip, struct metapath *mp)
 294 {
 295         unsigned int end_of_metadata = ip->i_height - 1;
 296         unsigned int x;
 297         __be64 *ptr;
 298         u64 dblock;
 299         int ret;
 300
 301         for (x = 0; x < end_of_metadata; x++) {
 302                 ptr = metapointer(x, mp);
 303                 dblock = be64_to_cpu(*ptr);
 304                 if (!dblock)
 305                         return x + 1;
 306
 307                 ret = gfs2_meta_indirect_buffer(ip, x+1, dblock, 0, &mp->mp_bh[x+1]);
 308                 if (ret)
 309                         return ret;
 310         }
 311
 312         return ip->i_height;
 313 }
 314
 315 static inline void release_metapath(struct metapath *mp)
 316 {
 317         int i;
 318
 319         for (i = 0; i < GFS2_MAX_META_HEIGHT; i++) {
 320                 if (mp->mp_bh[i] == NULL)
 321                         break;
 322                 brelse(mp->mp_bh[i]);
 323         }
 324 }
 325
 326 /**
 327  * gfs2_extent_length - Returns length of an extent of blocks
 328  * @start: Start of the buffer
 329  * @len: Length of the buffer in bytes
 330  * @ptr: Current position in the buffer
 331  * @limit: Max extent length to return (0 = unlimited)
 332  * @eob: Set to 1 if we hit "end of block"
 333  *
 334  * If the first block is zero (unallocated) it will return the number of
 335  * unallocated blocks in the extent, otherwise it will return the number
 336  * of contiguous blocks in the extent.
 337  *
 338  * Returns: The length of the extent (minimum of one block)
 339  */
 340
 341 static inline unsigned int gfs2_extent_length(void *start, unsigned int len, __be64 *ptr, unsigned limit, int *eob)
 342 {
 343         const __be64 *end = (start + len);
 344         const __be64 *first = ptr;
 345         u64 d = be64_to_cpu(*ptr);
 346
 347         *eob = 0;
 348         do {
 349                 ptr++;
 350                 if (ptr >= end)
 351                         break;
 352                 if (limit && --limit == 0)
 353                         break;
 354                 if (d)
 355                         d++;
 356         } while(be64_to_cpu(*ptr) == d);
 357         if (ptr >= end)
 358                 *eob = 1;
 359         return (ptr - first);
 360 }
 361
 362 static inline void bmap_lock(struct gfs2_inode *ip, int create)
 363 {
 364         if (create)
 365                 down_write(&ip->i_rw_mutex);
 366         else
 367                 down_read(&ip->i_rw_mutex);
 368 }
 369
 370 static inline void bmap_unlock(struct gfs2_inode *ip, int create)
 371 {
 372         if (create)
 373                 up_write(&ip->i_rw_mutex);
 374         else
 375                 up_read(&ip->i_rw_mutex);
 376 }
 377
 378 static inline __be64 *gfs2_indirect_init(struct metapath *mp,
 379                                          struct gfs2_glock *gl, unsigned int i,
 380                                          unsigned offset, u64 bn)
 381 {
 382         __be64 *ptr = (__be64 *)(mp->mp_bh[i - 1]->b_data +
 383                        ((i > 1) ? sizeof(struct gfs2_meta_header) :
 384                                  sizeof(struct gfs2_dinode)));
 385         BUG_ON(i < 1);
 386         BUG_ON(mp->mp_bh[i] != NULL);
 387         mp->mp_bh[i] = gfs2_meta_new(gl, bn);
 388         gfs2_trans_add_bh(gl, mp->mp_bh[i], 1);
 389         gfs2_metatype_set(mp->mp_bh[i], GFS2_METATYPE_IN, GFS2_FORMAT_IN);
 390         gfs2_buffer_clear_tail(mp->mp_bh[i], sizeof(struct gfs2_meta_header));
 391         ptr += offset;
 392         *ptr = cpu_to_be64(bn);
 393         return ptr;
 394 }
 395
 396 enum alloc_state {
 397         ALLOC_DATA = 0,
 398         ALLOC_GROW_DEPTH = 1,
 399         ALLOC_GROW_HEIGHT = 2,
 400         /* ALLOC_UNSTUFF = 3,   TBD and rather complicated */
 401 };
 402
 403 /**
 404  * gfs2_bmap_alloc - Build a metadata tree of the requested height
 405  * @inode: The GFS2 inode
 406  * @lblock: The logical starting block of the extent
 407  * @bh_map: This is used to return the mapping details
 408  * @mp: The metapath
 409  * @sheight: The starting height (i.e. whats already mapped)
 410  * @height: The height to build to
 411  * @maxlen: The max number of data blocks to alloc
 412  *
 413  * In this routine we may have to alloc:
 414  *   i) Indirect blocks to grow the metadata tree height
 415  *  ii) Indirect blocks to fill in lower part of the metadata tree
 416  * iii) Data blocks
 417  *
 418  * The function is in two parts. The first part works out the total
 419  * number of blocks which we need. The second part does the actual
 420  * allocation asking for an extent at a time (if enough contiguous free
 421  * blocks are available, there will only be one request per bmap call)
 422  * and uses the state machine to initialise the blocks in order.
 423  *
 424  * Returns: errno on error
 425  */
 426
 427 static int gfs2_bmap_alloc(struct inode *inode, const sector_t lblock,
 428                            struct buffer_head *bh_map, struct metapath *mp,
 429                            const unsigned int sheight,
 430                            const unsigned int height,
 431                            const unsigned int maxlen)
 432 {
 433         struct gfs2_inode *ip = GFS2_I(inode);
 434         struct gfs2_sbd *sdp = GFS2_SB(inode);
 435         struct buffer_head *dibh = mp->mp_bh[0];
 436         u64 bn, dblock = 0;
 437         unsigned n, i, blks, alloced = 0, iblks = 0, branch_start = 0;
 438         unsigned dblks = 0;
 439         unsigned ptrs_per_blk;
 440         const unsigned end_of_metadata = height - 1;
 441         int eob = 0;
 442         enum alloc_state state;
 443         __be64 *ptr;
 444         __be64 zero_bn = 0;
 445
 446         BUG_ON(sheight < 1);
 447         BUG_ON(dibh == NULL);
 448
 449         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 450
 451         if (height == sheight) {
 452                 struct buffer_head *bh;
 453                 /* Bottom indirect block exists, find unalloced extent size */
 454                 ptr = metapointer(end_of_metadata, mp);
 455                 bh = mp->mp_bh[end_of_metadata];
 456                 dblks = gfs2_extent_length(bh->b_data, bh->b_size, ptr, maxlen,
 457                                            &eob);
 458                 BUG_ON(dblks < 1);
 459                 state = ALLOC_DATA;
 460         } else {
 461                 /* Need to allocate indirect blocks */
 462                 ptrs_per_blk = height > 1 ? sdp->sd_inptrs : sdp->sd_diptrs;
 463                 dblks = min(maxlen, ptrs_per_blk - mp->mp_list[end_of_metadata]);
 464                 if (height == ip->i_height) {
 465                         /* Writing into existing tree, extend tree down */
 466                         iblks = height - sheight;
 467                         state = ALLOC_GROW_DEPTH;
 468                 } else {
 469                         /* Building up tree height */
 470                         state = ALLOC_GROW_HEIGHT;
 471                         iblks = height - ip->i_height;
 472                         branch_start = metapath_branch_start(mp);
 473                         iblks += (height - branch_start);
 474                 }
 475         }
 476
 477         /* start of the second part of the function (state machine) */
 478
 479         blks = dblks + iblks;
 480         i = sheight;
 481         do {
 482                 int error;
 483                 n = blks - alloced;
 484                 error = gfs2_alloc_block(ip, &bn, &n);
 485                 if (error)
 486                         return error;
 487                 alloced += n;
 488                 if (state != ALLOC_DATA || gfs2_is_jdata(ip))
 489                         gfs2_trans_add_unrevoke(sdp, bn, n);
 490                 switch (state) {
 491                 /* Growing height of tree */
 492                 case ALLOC_GROW_HEIGHT:
 493                         if (i == 1) {
 494                                 ptr = (__be64 *)(dibh->b_data +
 495                                                  sizeof(struct gfs2_dinode));
 496                                 zero_bn = *ptr;
 497                         }
 498                         for (; i - 1 < height - ip->i_height && n > 0; i++, n--)
 499                                 gfs2_indirect_init(mp, ip->i_gl, i, 0, bn++);
 500                         if (i - 1 == height - ip->i_height) {
 501                                 i--;
 502                                 gfs2_buffer_copy_tail(mp->mp_bh[i],
 503                                                 sizeof(struct gfs2_meta_header),
 504                                                 dibh, sizeof(struct gfs2_dinode));
 505                                 gfs2_buffer_clear_tail(dibh,
 506                                                 sizeof(struct gfs2_dinode) +
 507                                                 sizeof(__be64));
 508                                 ptr = (__be64 *)(mp->mp_bh[i]->b_data +
 509                                         sizeof(struct gfs2_meta_header));
 510                                 *ptr = zero_bn;
 511                                 state = ALLOC_GROW_DEPTH;
 512                                 for(i = branch_start; i < height; i++) {
 513                                         if (mp->mp_bh[i] == NULL)
 514                                                 break;
 515                                         brelse(mp->mp_bh[i]);
 516                                         mp->mp_bh[i] = NULL;
 517                                 }
 518                                 i = branch_start;
 519                         }
 520                         if (n == 0)
 521                                 break;
 522                 /* Branching from existing tree */
 523                 case ALLOC_GROW_DEPTH:
 524                         if (i > 1 && i < height)
 525                                 gfs2_trans_add_bh(ip->i_gl, mp->mp_bh[i-1], 1);
 526                         for (; i < height && n > 0; i++, n--)
 527                                 gfs2_indirect_init(mp, ip->i_gl, i,
 528                                                    mp->mp_list[i-1], bn++);
 529                         if (i == height)
 530                                 state = ALLOC_DATA;
 531                         if (n == 0)
 532                                 break;
 533                 /* Tree complete, adding data blocks */
 534                 case ALLOC_DATA:
 535                         BUG_ON(n > dblks);
 536                         BUG_ON(mp->mp_bh[end_of_metadata] == NULL);
 537                         gfs2_trans_add_bh(ip->i_gl, mp->mp_bh[end_of_metadata], 1);
 538                         dblks = n;
 539                         ptr = metapointer(end_of_metadata, mp);
 540                         dblock = bn;
 541                         while (n-- > 0)
 542                                 *ptr++ = cpu_to_be64(bn++);
 543                         break;
 544                 }
 545         } while ((state != ALLOC_DATA) || !dblock);
 546
 547         ip->i_height = height;
 548         gfs2_add_inode_blocks(&ip->i_inode, alloced);
 549         gfs2_dinode_out(ip, mp->mp_bh[0]->b_data);
 550         map_bh(bh_map, inode->i_sb, dblock);
 551         bh_map->b_size = dblks << inode->i_blkbits;
 552         set_buffer_new(bh_map);
 553         return 0;
 554 }
 555
 556 /**
 557  * gfs2_block_map - Map a block from an inode to a disk block
 558  * @inode: The inode
 559  * @lblock: The logical block number
 560  * @bh_map: The bh to be mapped
 561  * @create: True if its ok to alloc blocks to satify the request
 562  *
 563  * Sets buffer_mapped() if successful, sets buffer_boundary() if a
 564  * read of metadata will be required before the next block can be
 565  * mapped. Sets buffer_new() if new blocks were allocated.
 566  *
 567  * Returns: errno
 568  */
 569
 570 int gfs2_block_map(struct inode *inode, sector_t lblock,
 571                    struct buffer_head *bh_map, int create)
 572 {
 573         struct gfs2_inode *ip = GFS2_I(inode);
 574         struct gfs2_sbd *sdp = GFS2_SB(inode);
 575         unsigned int bsize = sdp->sd_sb.sb_bsize;
 576         const unsigned int maxlen = bh_map->b_size >> inode->i_blkbits;
 577         const u64 *arr = sdp->sd_heightsize;
 578         __be64 *ptr;
 579         u64 size;
 580         struct metapath mp;
 581         int ret;
 582         int eob;
 583         unsigned int len;
 584         struct buffer_head *bh;
 585         u8 height;
 586
 587         BUG_ON(maxlen == 0);
 588
 589         memset(mp.mp_bh, 0, sizeof(mp.mp_bh));
 590         bmap_lock(ip, create);
 591         clear_buffer_mapped(bh_map);
 592         clear_buffer_new(bh_map);
 593         clear_buffer_boundary(bh_map);
 594         trace_gfs2_bmap(ip, bh_map, lblock, create, 1);
 595         if (gfs2_is_dir(ip)) {
 596                 bsize = sdp->sd_jbsize;
 597                 arr = sdp->sd_jheightsize;
 598         }
 599
 600         ret = gfs2_meta_inode_buffer(ip, &mp.mp_bh[0]);
 601         if (ret)
 602                 goto out;
 603
 604         height = ip->i_height;
 605         size = (lblock + 1) * bsize;
 606         while (size > arr[height])
 607                 height++;
 608         find_metapath(sdp, lblock, &mp, height);
 609         ret = 1;
 610         if (height > ip->i_height || gfs2_is_stuffed(ip))
 611                 goto do_alloc;
 612         ret = lookup_metapath(ip, &mp);
 613         if (ret < 0)
 614                 goto out;
 615         if (ret != ip->i_height)
 616                 goto do_alloc;
 617         ptr = metapointer(ip->i_height - 1, &mp);
 618         if (*ptr == 0)
 619                 goto do_alloc;
 620         map_bh(bh_map, inode->i_sb, be64_to_cpu(*ptr));
 621         bh = mp.mp_bh[ip->i_height - 1];
 622         len = gfs2_extent_length(bh->b_data, bh->b_size, ptr, maxlen, &eob);
 623         bh_map->b_size = (len << inode->i_blkbits);
 624         if (eob)
 625                 set_buffer_boundary(bh_map);
 626         ret = 0;
 627 out:
 628         release_metapath(&mp);
 629         trace_gfs2_bmap(ip, bh_map, lblock, create, ret);
 630         bmap_unlock(ip, create);
 631         return ret;
 632
 633 do_alloc:
 634         /* All allocations are done here, firstly check create flag */
 635         if (!create) {
 636                 BUG_ON(gfs2_is_stuffed(ip));
 637                 ret = 0;
 638                 goto out;
 639         }
 640
 641         /* At this point ret is the tree depth of already allocated blocks */
 642         ret = gfs2_bmap_alloc(inode, lblock, bh_map, &mp, ret, height, maxlen);
 643         goto out;
 644 }
 645
 646 /*
 647  * Deprecated: do not use in new code
 648  */
 649 int gfs2_extent_map(struct inode *inode, u64 lblock, int *new, u64 *dblock, unsigned *extlen)
 650 {
 651         struct buffer_head bh = { .b_state = 0, .b_blocknr = 0 };
 652         int ret;
 653         int create = *new;
 654
 655         BUG_ON(!extlen);
 656         BUG_ON(!dblock);
 657         BUG_ON(!new);
 658
 659         bh.b_size = 1 << (inode->i_blkbits + (create ? 0 : 5));
 660         ret = gfs2_block_map(inode, lblock, &bh, create);
 661         *extlen = bh.b_size >> inode->i_blkbits;
 662         *dblock = bh.b_blocknr;
 663         if (buffer_new(&bh))
 664                 *new = 1;
 665         else
 666                 *new = 0;
 667         return ret;
 668 }
 669
 670 /**
 671  * recursive_scan - recursively scan through the end of a file
 672  * @ip: the inode
 673  * @dibh: the dinode buffer
 674  * @mp: the path through the metadata to the point to start
 675  * @height: the height the recursion is at
 676  * @block: the indirect block to look at
 677  * @first: 1 if this is the first block
 678  * @bc: the call to make for each piece of metadata
 679  * @data: data opaque to this function to pass to @bc
 680  *
 681  * When this is first called @height and @block should be zero and
 682  * @first should be 1.
 683  *
 684  * Returns: errno
 685  */
 686
 687 static int recursive_scan(struct gfs2_inode *ip, struct buffer_head *dibh,
 688                           struct metapath *mp, unsigned int height,
 689                           u64 block, int first, block_call_t bc,
 690                           void *data)
 691 {
 692         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 693         struct buffer_head *bh = NULL;
 694         __be64 *top, *bottom;
 695         u64 bn;
 696         int error;
 697         int mh_size = sizeof(struct gfs2_meta_header);
 698
 699         if (!height) {
 700                 error = gfs2_meta_inode_buffer(ip, &bh);
 701                 if (error)
 702                         return error;
 703                 dibh = bh;
 704
 705                 top = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + mp->mp_list[0];
 706                 bottom = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + sdp->sd_diptrs;
 707         } else {
 708                 error = gfs2_meta_indirect_buffer(ip, height, block, 0, &bh);
 709                 if (error)
 710                         return error;
 711
 712                 top = (__be64 *)(bh->b_data + mh_size) +
 713                                   (first ? mp->mp_list[height] : 0);
 714
 715                 bottom = (__be64 *)(bh->b_data + mh_size) + sdp->sd_inptrs;
 716         }
 717
 718         error = bc(ip, dibh, bh, top, bottom, height, data);
 719         if (error)
 720                 goto out;
 721
 722         if (height < ip->i_height - 1)
 723                 for (; top < bottom; top++, first = 0) {
 724                         if (!*top)
 725                                 continue;
 726
 727                         bn = be64_to_cpu(*top);
 728
 729                         error = recursive_scan(ip, dibh, mp, height + 1, bn,
 730                                                first, bc, data);
 731                         if (error)
 732                                 break;
 733                 }
 734
 735 out:
 736         brelse(bh);
 737         return error;
 738 }
 739
 740 /**
 741  * do_strip - Look for a layer a particular layer of the file and strip it off
 742  * @ip: the inode
 743  * @dibh: the dinode buffer
 744  * @bh: A buffer of pointers
 745  * @top: The first pointer in the buffer
 746  * @bottom: One more than the last pointer
 747  * @height: the height this buffer is at
 748  * @data: a pointer to a struct strip_mine
 749  *
 750  * Returns: errno
 751  */
 752
 753 static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh,
 754                     struct buffer_head *bh, __be64 *top, __be64 *bottom,
 755                     unsigned int height, void *data)
 756 {
 757         struct strip_mine *sm = data;
 758         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 759         struct gfs2_rgrp_list rlist;
 760         u64 bn, bstart;
 761         u32 blen, btotal;
 762         __be64 *p;
 763         unsigned int rg_blocks = 0;
 764         int metadata;
 765         unsigned int revokes = 0;
 766         int x;
 767         int error = 0;
 768
 769         if (!*top)
 770                 sm->sm_first = 0;
 771
 772         if (height != sm->sm_height)
 773                 return 0;
 774
 775         if (sm->sm_first) {
 776                 top++;
 777                 sm->sm_first = 0;
 778         }
 779
 780         metadata = (height != ip->i_height - 1);
 781         if (metadata)
 782                 revokes = (height) ? sdp->sd_inptrs : sdp->sd_diptrs;
 783         else if (ip->i_depth)
 784                 revokes = sdp->sd_inptrs;
 785
 786         if (ip != GFS2_I(sdp->sd_rindex))
 787                 error = gfs2_rindex_hold(sdp, &ip->i_alloc->al_ri_gh);
 788         else if (!sdp->sd_rgrps)
 789                 error = gfs2_ri_update(ip);
 790
 791         if (error)
 792                 return error;
 793
 794         memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
 795         bstart = 0;
 796         blen = 0;
 797
 798         for (p = top; p < bottom; p++) {
 799                 if (!*p)
 800                         continue;
 801
 802                 bn = be64_to_cpu(*p);
 803
 804                 if (bstart + blen == bn)
 805                         blen++;
 806                 else {
 807                         if (bstart)
 808                                 gfs2_rlist_add(sdp, &rlist, bstart);
 809
 810                         bstart = bn;
 811                         blen = 1;
 812                 }
 813         }
 814
 815         if (bstart)
 816                 gfs2_rlist_add(sdp, &rlist, bstart);
 817         else
 818                 goto out; /* Nothing to do */
 819
 820         gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE);
 821
 822         for (x = 0; x < rlist.rl_rgrps; x++) {
 823                 struct gfs2_rgrpd *rgd;
 824                 rgd = rlist.rl_ghs[x].gh_gl->gl_object;
 825                 rg_blocks += rgd->rd_length;
 826         }
 827
 828         error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
 829         if (error)
 830                 goto out_rlist;
 831
 832         error = gfs2_trans_begin(sdp, rg_blocks + RES_DINODE +
 833                                  RES_INDIRECT + RES_STATFS + RES_QUOTA,
 834                                  revokes);
 835         if (error)
 836                 goto out_rg_gunlock;
 837
 838         down_write(&ip->i_rw_mutex);
 839
 840         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 841         gfs2_trans_add_bh(ip->i_gl, bh, 1);
 842
 843         bstart = 0;
 844         blen = 0;
 845         btotal = 0;
 846
 847         for (p = top; p < bottom; p++) {
 848                 if (!*p)
 849                         continue;
 850
 851                 bn = be64_to_cpu(*p);
 852
 853                 if (bstart + blen == bn)
 854                         blen++;
 855                 else {
 856                         if (bstart) {
 857                                 if (metadata)
 858                                         __gfs2_free_meta(ip, bstart, blen);
 859                                 else
 860                                         __gfs2_free_data(ip, bstart, blen);
 861
 862                                 btotal += blen;
 863                         }
 864
 865                         bstart = bn;
 866                         blen = 1;
 867                 }
 868
 869                 *p = 0;
 870                 gfs2_add_inode_blocks(&ip->i_inode, -1);
 871         }
 872         if (bstart) {
 873                 if (metadata)
 874                         __gfs2_free_meta(ip, bstart, blen);
 875                 else
 876                         __gfs2_free_data(ip, bstart, blen);
 877
 878                 btotal += blen;
 879         }
 880
 881         gfs2_statfs_change(sdp, 0, +btotal, 0);
 882         gfs2_quota_change(ip, -(s64)btotal, ip->i_inode.i_uid,
 883                           ip->i_inode.i_gid);
 884
 885         ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
 886
 887         gfs2_dinode_out(ip, dibh->b_data);
 888
 889         up_write(&ip->i_rw_mutex);
 890
 891         gfs2_trans_end(sdp);
 892
 893 out_rg_gunlock:
 894         gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
 895 out_rlist:
 896         gfs2_rlist_free(&rlist);
 897 out:
 898         if (ip != GFS2_I(sdp->sd_rindex))
 899                 gfs2_glock_dq_uninit(&ip->i_alloc->al_ri_gh);
 900         return error;
 901 }
 902
 903 /**
 904  * gfs2_block_truncate_page - Deal with zeroing out data for truncate
 905  *
 906  * This is partly borrowed from ext3.
 907  */
 908 static int gfs2_block_truncate_page(struct address_space *mapping, loff_t from)
 909 {
 910         struct inode *inode = mapping->host;
 911         struct gfs2_inode *ip = GFS2_I(inode);
 912         unsigned long index = from >> PAGE_CACHE_SHIFT;
 913         unsigned offset = from & (PAGE_CACHE_SIZE-1);
 914         unsigned blocksize, iblock, length, pos;
 915         struct buffer_head *bh;
 916         struct page *page;
 917         int err;
 918
 919         page = grab_cache_page(mapping, index);
 920         if (!page)
 921                 return 0;
 922
 923         blocksize = inode->i_sb->s_blocksize;
 924         length = blocksize - (offset & (blocksize - 1));
 925         iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
 926
 927         if (!page_has_buffers(page))
 928                 create_empty_buffers(page, blocksize, 0);
 929
 930         /* Find the buffer that contains "offset" */
 931         bh = page_buffers(page);
 932         pos = blocksize;
 933         while (offset >= pos) {
 934                 bh = bh->b_this_page;
 935                 iblock++;
 936                 pos += blocksize;
 937         }
 938
 939         err = 0;
 940
 941         if (!buffer_mapped(bh)) {
 942                 gfs2_block_map(inode, iblock, bh, 0);
 943                 /* unmapped? It's a hole - nothing to do */
 944                 if (!buffer_mapped(bh))
 945                         goto unlock;
 946         }
 947
 948         /* Ok, it's mapped. Make sure it's up-to-date */
 949         if (PageUptodate(page))
 950                 set_buffer_uptodate(bh);
 951
 952         if (!buffer_uptodate(bh)) {
 953                 err = -EIO;
 954                 ll_rw_block(READ, 1, &bh);
 955                 wait_on_buffer(bh);
 956                 /* Uhhuh. Read error. Complain and punt. */
 957                 if (!buffer_uptodate(bh))
 958                         goto unlock;
 959                 err = 0;
 960         }
 961
 962         if (!gfs2_is_writeback(ip))
 963                 gfs2_trans_add_bh(ip->i_gl, bh, 0);
 964
 965         zero_user(page, offset, length);
 966         mark_buffer_dirty(bh);
 967 unlock:
 968         unlock_page(page);
 969         page_cache_release(page);
 970         return err;
 971 }
 972
 973 static int trunc_start(struct inode *inode, u64 oldsize, u64 newsize)
 974 {
 975         struct gfs2_inode *ip = GFS2_I(inode);
 976         struct gfs2_sbd *sdp = GFS2_SB(inode);
 977         struct address_space *mapping = inode->i_mapping;
 978         struct buffer_head *dibh;
 979         int journaled = gfs2_is_jdata(ip);
 980         int error;
 981
 982         error = gfs2_trans_begin(sdp,
 983                                  RES_DINODE + (journaled ? RES_JDATA : 0), 0);
 984         if (error)
 985                 return error;
 986
 987         error = gfs2_meta_inode_buffer(ip, &dibh);
 988         if (error)
 989                 goto out;
 990
 991         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 992
 993         if (gfs2_is_stuffed(ip)) {
 994                 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + newsize);
 995         } else {
 996                 if (newsize & (u64)(sdp->sd_sb.sb_bsize - 1)) {
 997                         error = gfs2_block_truncate_page(mapping, newsize);
 998                         if (error)
 999                                 goto out_brelse;
1000                 }
1001                 ip->i_diskflags |= GFS2_DIF_TRUNC_IN_PROG;
1002         }
1003
1004         i_size_write(inode, newsize);
1005         ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1006         gfs2_dinode_out(ip, dibh->b_data);
1007
1008         truncate_pagecache(inode, oldsize, newsize);
1009 out_brelse:
1010         brelse(dibh);
1011 out:
1012         gfs2_trans_end(sdp);
1013         return error;
1014 }
1015
1016 static int trunc_dealloc(struct gfs2_inode *ip, u64 size)
1017 {
1018         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1019         unsigned int height = ip->i_height;
1020         u64 lblock;
1021         struct metapath mp;
1022         int error;
1023
1024         if (!size)
1025                 lblock = 0;
1026         else
1027                 lblock = (size - 1) >> sdp->sd_sb.sb_bsize_shift;
1028
1029         find_metapath(sdp, lblock, &mp, ip->i_height);
1030         if (!gfs2_alloc_get(ip))
1031                 return -ENOMEM;
1032
1033         error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1034         if (error)
1035                 goto out;
1036
1037         while (height--) {
1038                 struct strip_mine sm;
1039                 sm.sm_first = !!size;
1040                 sm.sm_height = height;
1041
1042                 error = recursive_scan(ip, NULL, &mp, 0, 0, 1, do_strip, &sm);
1043                 if (error)
1044                         break;
1045         }
1046
1047         gfs2_quota_unhold(ip);
1048
1049 out:
1050         gfs2_alloc_put(ip);
1051         return error;
1052 }
1053
1054 static int trunc_end(struct gfs2_inode *ip)
1055 {
1056         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1057         struct buffer_head *dibh;
1058         int error;
1059
1060         error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1061         if (error)
1062                 return error;
1063
1064         down_write(&ip->i_rw_mutex);
1065
1066         error = gfs2_meta_inode_buffer(ip, &dibh);
1067         if (error)
1068                 goto out;
1069
1070         if (!i_size_read(&ip->i_inode)) {
1071                 ip->i_height = 0;
1072                 ip->i_goal = ip->i_no_addr;
1073                 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
1074         }
1075         ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1076         ip->i_diskflags &= ~GFS2_DIF_TRUNC_IN_PROG;
1077
1078         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1079         gfs2_dinode_out(ip, dibh->b_data);
1080         brelse(dibh);
1081
1082 out:
1083         up_write(&ip->i_rw_mutex);
1084         gfs2_trans_end(sdp);
1085         return error;
1086 }
1087
1088 /**
1089  * do_shrink - make a file smaller
1090  * @inode: the inode
1091  * @oldsize: the current inode size
1092  * @newsize: the size to make the file
1093  *
1094  * Called with an exclusive lock on @inode. The @size must
1095  * be equal to or smaller than the current inode size.
1096  *
1097  * Returns: errno
1098  */
1099
1100 static int do_shrink(struct inode *inode, u64 oldsize, u64 newsize)
1101 {
1102         struct gfs2_inode *ip = GFS2_I(inode);
1103         int error;
1104
1105         error = trunc_start(inode, oldsize, newsize);
1106         if (error < 0)
1107                 return error;
1108         if (gfs2_is_stuffed(ip))
1109                 return 0;
1110
1111         error = trunc_dealloc(ip, newsize);
1112         if (error == 0)
1113                 error = trunc_end(ip);
1114
1115         return error;
1116 }
1117
1118 void gfs2_trim_blocks(struct inode *inode)
1119 {
1120         u64 size = inode->i_size;
1121         int ret;
1122
1123         ret = do_shrink(inode, size, size);
1124         WARN_ON(ret != 0);
1125 }
1126
1127 /**
1128  * do_grow - Touch and update inode size
1129  * @inode: The inode
1130  * @size: The new size
1131  *
1132  * This function updates the timestamps on the inode and
1133  * may also increase the size of the inode. This function
1134  * must not be called with @size any smaller than the current
1135  * inode size.
1136  *
1137  * Although it is not strictly required to unstuff files here,
1138  * earlier versions of GFS2 have a bug in the stuffed file reading
1139  * code which will result in a buffer overrun if the size is larger
1140  * than the max stuffed file size. In order to prevent this from
1141  * occurring, such files are unstuffed, but in other cases we can
1142  * just update the inode size directly.
1143  *
1144  * Returns: 0 on success, or -ve on error
1145  */
1146
1147 static int do_grow(struct inode *inode, u64 size)
1148 {
1149         struct gfs2_inode *ip = GFS2_I(inode);
1150         struct gfs2_sbd *sdp = GFS2_SB(inode);
1151         struct buffer_head *dibh;
1152         struct gfs2_alloc *al = NULL;
1153         int error;
1154
1155         if (gfs2_is_stuffed(ip) &&
1156             (size > (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)))) {
1157                 al = gfs2_alloc_get(ip);
1158                 if (al == NULL)
1159                         return -ENOMEM;
1160
1161                 error = gfs2_quota_lock_check(ip);
1162                 if (error)
1163                         goto do_grow_alloc_put;
1164
1165                 al->al_requested = 1;
1166                 error = gfs2_inplace_reserve(ip);
1167                 if (error)
1168                         goto do_grow_qunlock;
1169         }
1170
1171         error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT, 0);
1172         if (error)
1173                 goto do_grow_release;
1174
1175         if (al) {
1176                 error = gfs2_unstuff_dinode(ip, NULL);
1177                 if (error)
1178                         goto do_end_trans;
1179         }
1180
1181         error = gfs2_meta_inode_buffer(ip, &dibh);
1182         if (error)
1183                 goto do_end_trans;
1184
1185         i_size_write(inode, size);
1186         ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1187         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1188         gfs2_dinode_out(ip, dibh->b_data);
1189         brelse(dibh);
1190
1191 do_end_trans:
1192         gfs2_trans_end(sdp);
1193 do_grow_release:
1194         if (al) {
1195                 gfs2_inplace_release(ip);
1196 do_grow_qunlock:
1197                 gfs2_quota_unlock(ip);
1198 do_grow_alloc_put:
1199                 gfs2_alloc_put(ip);
1200         }
1201         return error;
1202 }
1203
1204 /**
1205  * gfs2_setattr_size - make a file a given size
1206  * @inode: the inode
1207  * @newsize: the size to make the file
1208  *
1209  * The file size can grow, shrink, or stay the same size. This
1210  * is called holding i_mutex and an exclusive glock on the inode
1211  * in question.
1212  *
1213  * Returns: errno
1214  */
1215
1216 int gfs2_setattr_size(struct inode *inode, u64 newsize)
1217 {
1218         int ret;
1219         u64 oldsize;
1220
1221         BUG_ON(!S_ISREG(inode->i_mode));
1222
1223         ret = inode_newsize_ok(inode, newsize);
1224         if (ret)
1225                 return ret;
1226
1227         oldsize = inode->i_size;
1228         if (newsize >= oldsize)
1229                 return do_grow(inode, newsize);
1230
1231         return do_shrink(inode, oldsize, newsize);
1232 }
1233
1234 int gfs2_truncatei_resume(struct gfs2_inode *ip)
1235 {
1236         int error;
1237         error = trunc_dealloc(ip, i_size_read(&ip->i_inode));
1238         if (!error)
1239                 error = trunc_end(ip);
1240         return error;
1241 }
1242
1243 int gfs2_file_dealloc(struct gfs2_inode *ip)
1244 {
1245         return trunc_dealloc(ip, 0);
1246 }
1247
1248 /**
1249  * gfs2_write_alloc_required - figure out if a write will require an allocation
1250  * @ip: the file being written to
1251  * @offset: the offset to write to
1252  * @len: the number of bytes being written
1253  *
1254  * Returns: 1 if an alloc is required, 0 otherwise
1255  */
1256
1257 int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset,
1258                               unsigned int len)
1259 {
1260         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1261         struct buffer_head bh;
1262         unsigned int shift;
1263         u64 lblock, lblock_stop, size;
1264         u64 end_of_file;
1265
1266         if (!len)
1267                 return 0;
1268
1269         if (gfs2_is_stuffed(ip)) {
1270                 if (offset + len >
1271                     sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
1272                         return 1;
1273                 return 0;
1274         }
1275
1276         shift = sdp->sd_sb.sb_bsize_shift;
1277         BUG_ON(gfs2_is_dir(ip));
1278         end_of_file = (i_size_read(&ip->i_inode) + sdp->sd_sb.sb_bsize - 1) >> shift;
1279         lblock = offset >> shift;
1280         lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift;
1281         if (lblock_stop > end_of_file)
1282                 return 1;
1283
1284         size = (lblock_stop - lblock) << shift;
1285         do {
1286                 bh.b_state = 0;
1287                 bh.b_size = size;
1288                 gfs2_block_map(&ip->i_inode, lblock, &bh, 0);
1289                 if (!buffer_mapped(&bh))
1290                         return 1;
1291                 size -= bh.b_size;
1292                 lblock += (bh.b_size >> ip->i_inode.i_blkbits);
1293         } while(size > 0);
1294
1295         return 0;
1296 }
1297