fs/gfs2/bmap.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
   4  *
   5  * This copyrighted material is made available to anyone wishing to use,
   6  * modify, copy, or redistribute it subject to the terms and conditions
   7  * of the GNU General Public License version 2.
   8  */
   9
  10 #include <linux/slab.h>
  11 #include <linux/spinlock.h>
  12 #include <linux/completion.h>
  13 #include <linux/buffer_head.h>
  14 #include <linux/gfs2_ondisk.h>
  15 #include <linux/crc32.h>
  16
  17 #include "gfs2.h"
  18 #include "incore.h"
  19 #include "bmap.h"
  20 #include "glock.h"
  21 #include "inode.h"
  22 #include "meta_io.h"
  23 #include "quota.h"
  24 #include "rgrp.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  * @private: any locked page held by the caller process
  55  *
  56  * Returns: errno
  57  */
  58
  59 static int gfs2_unstuffer_page(struct gfs2_inode *ip, struct buffer_head *dibh,
  60                                u64 block, struct page *page)
  61 {
  62         struct inode *inode = &ip->i_inode;
  63         struct buffer_head *bh;
  64         int release = 0;
  65
  66         if (!page || page->index) {
  67                 page = grab_cache_page(inode->i_mapping, 0);
  68                 if (!page)
  69                         return -ENOMEM;
  70                 release = 1;
  71         }
  72
  73         if (!PageUptodate(page)) {
  74                 void *kaddr = kmap(page);
  75
  76                 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
  77                        ip->i_disksize);
  78                 memset(kaddr + ip->i_disksize, 0,
  79                        PAGE_CACHE_SIZE - ip->i_disksize);
  80                 kunmap(page);
  81
  82                 SetPageUptodate(page);
  83         }
  84
  85         if (!page_has_buffers(page))
  86                 create_empty_buffers(page, 1 << inode->i_blkbits,
  87                                      (1 << BH_Uptodate));
  88
  89         bh = page_buffers(page);
  90
  91         if (!buffer_mapped(bh))
  92                 map_bh(bh, inode->i_sb, block);
  93
  94         set_buffer_uptodate(bh);
  95         if (!gfs2_is_jdata(ip))
  96                 mark_buffer_dirty(bh);
  97         if (!gfs2_is_writeback(ip))
  98                 gfs2_trans_add_bh(ip->i_gl, bh, 0);
  99
 100         if (release) {
 101                 unlock_page(page);
 102                 page_cache_release(page);
 103         }
 104
 105         return 0;
 106 }
 107
 108 /**
 109  * gfs2_unstuff_dinode - Unstuff a dinode when the data has grown too big
 110  * @ip: The GFS2 inode to unstuff
 111  * @unstuffer: the routine that handles unstuffing a non-zero length file
 112  * @private: private data for the unstuffer
 113  *
 114  * This routine unstuffs a dinode and returns it to a "normal" state such
 115  * that the height can be grown in the traditional way.
 116  *
 117  * Returns: errno
 118  */
 119
 120 int gfs2_unstuff_dinode(struct gfs2_inode *ip, struct page *page)
 121 {
 122         struct buffer_head *bh, *dibh;
 123         struct gfs2_dinode *di;
 124         u64 block = 0;
 125         int isdir = gfs2_is_dir(ip);
 126         int error;
 127
 128         down_write(&ip->i_rw_mutex);
 129
 130         error = gfs2_meta_inode_buffer(ip, &dibh);
 131         if (error)
 132                 goto out;
 133
 134         if (ip->i_disksize) {
 135                 /* Get a free block, fill it with the stuffed data,
 136                    and write it out to disk */
 137
 138                 unsigned int n = 1;
 139                 error = gfs2_alloc_block(ip, &block, &n);
 140                 if (error)
 141                         goto out_brelse;
 142                 if (isdir) {
 143                         gfs2_trans_add_unrevoke(GFS2_SB(&ip->i_inode), block, 1);
 144                         error = gfs2_dir_get_new_buffer(ip, block, &bh);
 145                         if (error)
 146                                 goto out_brelse;
 147                         gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_meta_header),
 148                                               dibh, sizeof(struct gfs2_dinode));
 149                         brelse(bh);
 150                 } else {
 151                         error = gfs2_unstuffer_page(ip, dibh, block, page);
 152                         if (error)
 153                                 goto out_brelse;
 154                 }
 155         }
 156
 157         /*  Set up the pointer to the new block  */
 158
 159         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 160         di = (struct gfs2_dinode *)dibh->b_data;
 161         gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
 162
 163         if (ip->i_disksize) {
 164                 *(__be64 *)(di + 1) = cpu_to_be64(block);
 165                 gfs2_add_inode_blocks(&ip->i_inode, 1);
 166                 di->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
 167         }
 168
 169         ip->i_height = 1;
 170         di->di_height = cpu_to_be16(1);
 171
 172 out_brelse:
 173         brelse(dibh);
 174 out:
 175         up_write(&ip->i_rw_mutex);
 176         return error;
 177 }
 178
 179
 180 /**
 181  * find_metapath - Find path through the metadata tree
 182  * @sdp: The superblock
 183  * @mp: The metapath to return the result in
 184  * @block: The disk block to look up
 185  * @height: The pre-calculated height of the metadata tree
 186  *
 187  *   This routine returns a struct metapath structure that defines a path
 188  *   through the metadata of inode "ip" to get to block "block".
 189  *
 190  *   Example:
 191  *   Given:  "ip" is a height 3 file, "offset" is 101342453, and this is a
 192  *   filesystem with a blocksize of 4096.
 193  *
 194  *   find_metapath() would return a struct metapath structure set to:
 195  *   mp_offset = 101342453, mp_height = 3, mp_list[0] = 0, mp_list[1] = 48,
 196  *   and mp_list[2] = 165.
 197  *
 198  *   That means that in order to get to the block containing the byte at
 199  *   offset 101342453, we would load the indirect block pointed to by pointer
 200  *   0 in the dinode.  We would then load the indirect block pointed to by
 201  *   pointer 48 in that indirect block.  We would then load the data block
 202  *   pointed to by pointer 165 in that indirect block.
 203  *
 204  *             ----------------------------------------
 205  *             | Dinode |                             |
 206  *             |        |                            4|
 207  *             |        |0 1 2 3 4 5                 9|
 208  *             |        |                            6|
 209  *             ----------------------------------------
 210  *                       |
 211  *                       |
 212  *                       V
 213  *             ----------------------------------------
 214  *             | Indirect Block                       |
 215  *             |                                     5|
 216  *             |            4 4 4 4 4 5 5            1|
 217  *             |0           5 6 7 8 9 0 1            2|
 218  *             ----------------------------------------
 219  *                                |
 220  *                                |
 221  *                                V
 222  *             ----------------------------------------
 223  *             | Indirect Block                       |
 224  *             |                         1 1 1 1 1   5|
 225  *             |                         6 6 6 6 6   1|
 226  *             |0                        3 4 5 6 7   2|
 227  *             ----------------------------------------
 228  *                                           |
 229  *                                           |
 230  *                                           V
 231  *             ----------------------------------------
 232  *             | Data block containing offset         |
 233  *             |            101342453                 |
 234  *             |                                      |
 235  *             |                                      |
 236  *             ----------------------------------------
 237  *
 238  */
 239
 240 static void find_metapath(const struct gfs2_sbd *sdp, u64 block,
 241                           struct metapath *mp, unsigned int height)
 242 {
 243         unsigned int i;
 244
 245         for (i = height; i--;)
 246                 mp->mp_list[i] = do_div(block, sdp->sd_inptrs);
 247
 248 }
 249
 250 static inline unsigned int metapath_branch_start(const struct metapath *mp)
 251 {
 252         if (mp->mp_list[0] == 0)
 253                 return 2;
 254         return 1;
 255 }
 256
 257 /**
 258  * metapointer - Return pointer to start of metadata in a buffer
 259  * @height: The metadata height (0 = dinode)
 260  * @mp: The metapath
 261  *
 262  * Return a pointer to the block number of the next height of the metadata
 263  * tree given a buffer containing the pointer to the current height of the
 264  * metadata tree.
 265  */
 266
 267 static inline __be64 *metapointer(unsigned int height, const struct metapath *mp)
 268 {
 269         struct buffer_head *bh = mp->mp_bh[height];
 270         unsigned int head_size = (height > 0) ?
 271                 sizeof(struct gfs2_meta_header) : sizeof(struct gfs2_dinode);
 272         return ((__be64 *)(bh->b_data + head_size)) + mp->mp_list[height];
 273 }
 274
 275 /**
 276  * lookup_metapath - Walk the metadata tree to a specific point
 277  * @ip: The inode
 278  * @mp: The metapath
 279  *
 280  * Assumes that the inode's buffer has already been looked up and
 281  * hooked onto mp->mp_bh[0] and that the metapath has been initialised
 282  * by find_metapath().
 283  *
 284  * If this function encounters part of the tree which has not been
 285  * allocated, it returns the current height of the tree at the point
 286  * at which it found the unallocated block. Blocks which are found are
 287  * added to the mp->mp_bh[] list.
 288  *
 289  * Returns: error or height of metadata tree
 290  */
 291
 292 static int lookup_metapath(struct gfs2_inode *ip, struct metapath *mp)
 293 {
 294         unsigned int end_of_metadata = ip->i_height - 1;
 295         unsigned int x;
 296         __be64 *ptr;
 297         u64 dblock;
 298         int ret;
 299
 300         for (x = 0; x < end_of_metadata; x++) {
 301                 ptr = metapointer(x, mp);
 302                 dblock = be64_to_cpu(*ptr);
 303                 if (!dblock)
 304                         return x + 1;
 305
 306                 ret = gfs2_meta_indirect_buffer(ip, x+1, dblock, 0, &mp->mp_bh[x+1]);
 307                 if (ret)
 308                         return ret;
 309         }
 310
 311         return ip->i_height;
 312 }
 313
 314 static inline void release_metapath(struct metapath *mp)
 315 {
 316         int i;
 317
 318         for (i = 0; i < GFS2_MAX_META_HEIGHT; i++) {
 319                 if (mp->mp_bh[i] == NULL)
 320                         break;
 321                 brelse(mp->mp_bh[i]);
 322         }
 323 }
 324
 325 /**
 326  * gfs2_extent_length - Returns length of an extent of blocks
 327  * @start: Start of the buffer
 328  * @len: Length of the buffer in bytes
 329  * @ptr: Current position in the buffer
 330  * @limit: Max extent length to return (0 = unlimited)
 331  * @eob: Set to 1 if we hit "end of block"
 332  *
 333  * If the first block is zero (unallocated) it will return the number of
 334  * unallocated blocks in the extent, otherwise it will return the number
 335  * of contiguous blocks in the extent.
 336  *
 337  * Returns: The length of the extent (minimum of one block)
 338  */
 339
 340 static inline unsigned int gfs2_extent_length(void *start, unsigned int len, __be64 *ptr, unsigned limit, int *eob)
 341 {
 342         const __be64 *end = (start + len);
 343         const __be64 *first = ptr;
 344         u64 d = be64_to_cpu(*ptr);
 345
 346         *eob = 0;
 347         do {
 348                 ptr++;
 349                 if (ptr >= end)
 350                         break;
 351                 if (limit && --limit == 0)
 352                         break;
 353                 if (d)
 354                         d++;
 355         } while(be64_to_cpu(*ptr) == d);
 356         if (ptr >= end)
 357                 *eob = 1;
 358         return (ptr - first);
 359 }
 360
 361 static inline void bmap_lock(struct gfs2_inode *ip, int create)
 362 {
 363         if (create)
 364                 down_write(&ip->i_rw_mutex);
 365         else
 366                 down_read(&ip->i_rw_mutex);
 367 }
 368
 369 static inline void bmap_unlock(struct gfs2_inode *ip, int create)
 370 {
 371         if (create)
 372                 up_write(&ip->i_rw_mutex);
 373         else
 374                 up_read(&ip->i_rw_mutex);
 375 }
 376
 377 static inline __be64 *gfs2_indirect_init(struct metapath *mp,
 378                                          struct gfs2_glock *gl, unsigned int i,
 379                                          unsigned offset, u64 bn)
 380 {
 381         __be64 *ptr = (__be64 *)(mp->mp_bh[i - 1]->b_data +
 382                        ((i > 1) ? sizeof(struct gfs2_meta_header) :
 383                                  sizeof(struct gfs2_dinode)));
 384         BUG_ON(i < 1);
 385         BUG_ON(mp->mp_bh[i] != NULL);
 386         mp->mp_bh[i] = gfs2_meta_new(gl, bn);
 387         gfs2_trans_add_bh(gl, mp->mp_bh[i], 1);
 388         gfs2_metatype_set(mp->mp_bh[i], GFS2_METATYPE_IN, GFS2_FORMAT_IN);
 389         gfs2_buffer_clear_tail(mp->mp_bh[i], sizeof(struct gfs2_meta_header));
 390         ptr += offset;
 391         *ptr = cpu_to_be64(bn);
 392         return ptr;
 393 }
 394
 395 enum alloc_state {
 396         ALLOC_DATA = 0,
 397         ALLOC_GROW_DEPTH = 1,
 398         ALLOC_GROW_HEIGHT = 2,
 399         /* ALLOC_UNSTUFF = 3,   TBD and rather complicated */
 400 };
 401
 402 /**
 403  * gfs2_bmap_alloc - Build a metadata tree of the requested height
 404  * @inode: The GFS2 inode
 405  * @lblock: The logical starting block of the extent
 406  * @bh_map: This is used to return the mapping details
 407  * @mp: The metapath
 408  * @sheight: The starting height (i.e. whats already mapped)
 409  * @height: The height to build to
 410  * @maxlen: The max number of data blocks to alloc
 411  *
 412  * In this routine we may have to alloc:
 413  *   i) Indirect blocks to grow the metadata tree height
 414  *  ii) Indirect blocks to fill in lower part of the metadata tree
 415  * iii) Data blocks
 416  *
 417  * The function is in two parts. The first part works out the total
 418  * number of blocks which we need. The second part does the actual
 419  * allocation asking for an extent at a time (if enough contiguous free
 420  * blocks are available, there will only be one request per bmap call)
 421  * and uses the state machine to initialise the blocks in order.
 422  *
 423  * Returns: errno on error
 424  */
 425
 426 static int gfs2_bmap_alloc(struct inode *inode, const sector_t lblock,
 427                            struct buffer_head *bh_map, struct metapath *mp,
 428                            const unsigned int sheight,
 429                            const unsigned int height,
 430                            const unsigned int maxlen)
 431 {
 432         struct gfs2_inode *ip = GFS2_I(inode);
 433         struct gfs2_sbd *sdp = GFS2_SB(inode);
 434         struct buffer_head *dibh = mp->mp_bh[0];
 435         u64 bn, dblock = 0;
 436         unsigned n, i, blks, alloced = 0, iblks = 0, branch_start = 0;
 437         unsigned dblks = 0;
 438         unsigned ptrs_per_blk;
 439         const unsigned end_of_metadata = height - 1;
 440         int eob = 0;
 441         enum alloc_state state;
 442         __be64 *ptr;
 443         __be64 zero_bn = 0;
 444
 445         BUG_ON(sheight < 1);
 446         BUG_ON(dibh == NULL);
 447
 448         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 449
 450         if (height == sheight) {
 451                 struct buffer_head *bh;
 452                 /* Bottom indirect block exists, find unalloced extent size */
 453                 ptr = metapointer(end_of_metadata, mp);
 454                 bh = mp->mp_bh[end_of_metadata];
 455                 dblks = gfs2_extent_length(bh->b_data, bh->b_size, ptr, maxlen,
 456                                            &eob);
 457                 BUG_ON(dblks < 1);
 458                 state = ALLOC_DATA;
 459         } else {
 460                 /* Need to allocate indirect blocks */
 461                 ptrs_per_blk = height > 1 ? sdp->sd_inptrs : sdp->sd_diptrs;
 462                 dblks = min(maxlen, ptrs_per_blk - mp->mp_list[end_of_metadata]);
 463                 if (height == ip->i_height) {
 464                         /* Writing into existing tree, extend tree down */
 465                         iblks = height - sheight;
 466                         state = ALLOC_GROW_DEPTH;
 467                 } else {
 468                         /* Building up tree height */
 469                         state = ALLOC_GROW_HEIGHT;
 470                         iblks = height - ip->i_height;
 471                         branch_start = metapath_branch_start(mp);
 472                         iblks += (height - branch_start);
 473                 }
 474         }
 475
 476         /* start of the second part of the function (state machine) */
 477
 478         blks = dblks + iblks;
 479         i = sheight;
 480         do {
 481                 int error;
 482                 n = blks - alloced;
 483                 error = gfs2_alloc_block(ip, &bn, &n);
 484                 if (error)
 485                         return error;
 486                 alloced += n;
 487                 if (state != ALLOC_DATA || gfs2_is_jdata(ip))
 488                         gfs2_trans_add_unrevoke(sdp, bn, n);
 489                 switch (state) {
 490                 /* Growing height of tree */
 491                 case ALLOC_GROW_HEIGHT:
 492                         if (i == 1) {
 493                                 ptr = (__be64 *)(dibh->b_data +
 494                                                  sizeof(struct gfs2_dinode));
 495                                 zero_bn = *ptr;
 496                         }
 497                         for (; i - 1 < height - ip->i_height && n > 0; i++, n--)
 498                                 gfs2_indirect_init(mp, ip->i_gl, i, 0, bn++);
 499                         if (i - 1 == height - ip->i_height) {
 500                                 i--;
 501                                 gfs2_buffer_copy_tail(mp->mp_bh[i],
 502                                                 sizeof(struct gfs2_meta_header),
 503                                                 dibh, sizeof(struct gfs2_dinode));
 504                                 gfs2_buffer_clear_tail(dibh,
 505                                                 sizeof(struct gfs2_dinode) +
 506                                                 sizeof(__be64));
 507                                 ptr = (__be64 *)(mp->mp_bh[i]->b_data +
 508                                         sizeof(struct gfs2_meta_header));
 509                                 *ptr = zero_bn;
 510                                 state = ALLOC_GROW_DEPTH;
 511                                 for(i = branch_start; i < height; i++) {
 512                                         if (mp->mp_bh[i] == NULL)
 513                                                 break;
 514                                         brelse(mp->mp_bh[i]);
 515                                         mp->mp_bh[i] = NULL;
 516                                 }
 517                                 i = branch_start;
 518                         }
 519                         if (n == 0)
 520                                 break;
 521                 /* Branching from existing tree */
 522                 case ALLOC_GROW_DEPTH:
 523                         if (i > 1 && i < height)
 524                                 gfs2_trans_add_bh(ip->i_gl, mp->mp_bh[i-1], 1);
 525                         for (; i < height && n > 0; i++, n--)
 526                                 gfs2_indirect_init(mp, ip->i_gl, i,
 527                                                    mp->mp_list[i-1], bn++);
 528                         if (i == height)
 529                                 state = ALLOC_DATA;
 530                         if (n == 0)
 531                                 break;
 532                 /* Tree complete, adding data blocks */
 533                 case ALLOC_DATA:
 534                         BUG_ON(n > dblks);
 535                         BUG_ON(mp->mp_bh[end_of_metadata] == NULL);
 536                         gfs2_trans_add_bh(ip->i_gl, mp->mp_bh[end_of_metadata], 1);
 537                         dblks = n;
 538                         ptr = metapointer(end_of_metadata, mp);
 539                         dblock = bn;
 540                         while (n-- > 0)
 541                                 *ptr++ = cpu_to_be64(bn++);
 542                         break;
 543                 }
 544         } while (state != ALLOC_DATA);
 545
 546         ip->i_height = height;
 547         gfs2_add_inode_blocks(&ip->i_inode, alloced);
 548         gfs2_dinode_out(ip, mp->mp_bh[0]->b_data);
 549         map_bh(bh_map, inode->i_sb, dblock);
 550         bh_map->b_size = dblks << inode->i_blkbits;
 551         set_buffer_new(bh_map);
 552         return 0;
 553 }
 554
 555 /**
 556  * gfs2_block_map - Map a block from an inode to a disk block
 557  * @inode: The inode
 558  * @lblock: The logical block number
 559  * @bh_map: The bh to be mapped
 560  * @create: True if its ok to alloc blocks to satify the request
 561  *
 562  * Sets buffer_mapped() if successful, sets buffer_boundary() if a
 563  * read of metadata will be required before the next block can be
 564  * mapped. Sets buffer_new() if new blocks were allocated.
 565  *
 566  * Returns: errno
 567  */
 568
 569 int gfs2_block_map(struct inode *inode, sector_t lblock,
 570                    struct buffer_head *bh_map, int create)
 571 {
 572         struct gfs2_inode *ip = GFS2_I(inode);
 573         struct gfs2_sbd *sdp = GFS2_SB(inode);
 574         unsigned int bsize = sdp->sd_sb.sb_bsize;
 575         const unsigned int maxlen = bh_map->b_size >> inode->i_blkbits;
 576         const u64 *arr = sdp->sd_heightsize;
 577         __be64 *ptr;
 578         u64 size;
 579         struct metapath mp;
 580         int ret;
 581         int eob;
 582         unsigned int len;
 583         struct buffer_head *bh;
 584         u8 height;
 585
 586         BUG_ON(maxlen == 0);
 587
 588         memset(mp.mp_bh, 0, sizeof(mp.mp_bh));
 589         bmap_lock(ip, create);
 590         clear_buffer_mapped(bh_map);
 591         clear_buffer_new(bh_map);
 592         clear_buffer_boundary(bh_map);
 593         trace_gfs2_bmap(ip, bh_map, lblock, create, 1);
 594         if (gfs2_is_dir(ip)) {
 595                 bsize = sdp->sd_jbsize;
 596                 arr = sdp->sd_jheightsize;
 597         }
 598
 599         ret = gfs2_meta_inode_buffer(ip, &mp.mp_bh[0]);
 600         if (ret)
 601                 goto out;
 602
 603         height = ip->i_height;
 604         size = (lblock + 1) * bsize;
 605         while (size > arr[height])
 606                 height++;
 607         find_metapath(sdp, lblock, &mp, height);
 608         ret = 1;
 609         if (height > ip->i_height || gfs2_is_stuffed(ip))
 610                 goto do_alloc;
 611         ret = lookup_metapath(ip, &mp);
 612         if (ret < 0)
 613                 goto out;
 614         if (ret != ip->i_height)
 615                 goto do_alloc;
 616         ptr = metapointer(ip->i_height - 1, &mp);
 617         if (*ptr == 0)
 618                 goto do_alloc;
 619         map_bh(bh_map, inode->i_sb, be64_to_cpu(*ptr));
 620         bh = mp.mp_bh[ip->i_height - 1];
 621         len = gfs2_extent_length(bh->b_data, bh->b_size, ptr, maxlen, &eob);
 622         bh_map->b_size = (len << inode->i_blkbits);
 623         if (eob)
 624                 set_buffer_boundary(bh_map);
 625         ret = 0;
 626 out:
 627         release_metapath(&mp);
 628         trace_gfs2_bmap(ip, bh_map, lblock, create, ret);
 629         bmap_unlock(ip, create);
 630         return ret;
 631
 632 do_alloc:
 633         /* All allocations are done here, firstly check create flag */
 634         if (!create) {
 635                 BUG_ON(gfs2_is_stuffed(ip));
 636                 ret = 0;
 637                 goto out;
 638         }
 639
 640         /* At this point ret is the tree depth of already allocated blocks */
 641         ret = gfs2_bmap_alloc(inode, lblock, bh_map, &mp, ret, height, maxlen);
 642         goto out;
 643 }
 644
 645 /*
 646  * Deprecated: do not use in new code
 647  */
 648 int gfs2_extent_map(struct inode *inode, u64 lblock, int *new, u64 *dblock, unsigned *extlen)
 649 {
 650         struct buffer_head bh = { .b_state = 0, .b_blocknr = 0 };
 651         int ret;
 652         int create = *new;
 653
 654         BUG_ON(!extlen);
 655         BUG_ON(!dblock);
 656         BUG_ON(!new);
 657
 658         bh.b_size = 1 << (inode->i_blkbits + (create ? 0 : 5));
 659         ret = gfs2_block_map(inode, lblock, &bh, create);
 660         *extlen = bh.b_size >> inode->i_blkbits;
 661         *dblock = bh.b_blocknr;
 662         if (buffer_new(&bh))
 663                 *new = 1;
 664         else
 665                 *new = 0;
 666         return ret;
 667 }
 668
 669 /**
 670  * recursive_scan - recursively scan through the end of a file
 671  * @ip: the inode
 672  * @dibh: the dinode buffer
 673  * @mp: the path through the metadata to the point to start
 674  * @height: the height the recursion is at
 675  * @block: the indirect block to look at
 676  * @first: 1 if this is the first block
 677  * @bc: the call to make for each piece of metadata
 678  * @data: data opaque to this function to pass to @bc
 679  *
 680  * When this is first called @height and @block should be zero and
 681  * @first should be 1.
 682  *
 683  * Returns: errno
 684  */
 685
 686 static int recursive_scan(struct gfs2_inode *ip, struct buffer_head *dibh,
 687                           struct metapath *mp, unsigned int height,
 688                           u64 block, int first, block_call_t bc,
 689                           void *data)
 690 {
 691         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 692         struct buffer_head *bh = NULL;
 693         __be64 *top, *bottom;
 694         u64 bn;
 695         int error;
 696         int mh_size = sizeof(struct gfs2_meta_header);
 697
 698         if (!height) {
 699                 error = gfs2_meta_inode_buffer(ip, &bh);
 700                 if (error)
 701                         return error;
 702                 dibh = bh;
 703
 704                 top = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + mp->mp_list[0];
 705                 bottom = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + sdp->sd_diptrs;
 706         } else {
 707                 error = gfs2_meta_indirect_buffer(ip, height, block, 0, &bh);
 708                 if (error)
 709                         return error;
 710
 711                 top = (__be64 *)(bh->b_data + mh_size) +
 712                                   (first ? mp->mp_list[height] : 0);
 713
 714                 bottom = (__be64 *)(bh->b_data + mh_size) + sdp->sd_inptrs;
 715         }
 716
 717         error = bc(ip, dibh, bh, top, bottom, height, data);
 718         if (error)
 719                 goto out;
 720
 721         if (height < ip->i_height - 1)
 722                 for (; top < bottom; top++, first = 0) {
 723                         if (!*top)
 724                                 continue;
 725
 726                         bn = be64_to_cpu(*top);
 727
 728                         error = recursive_scan(ip, dibh, mp, height + 1, bn,
 729                                                first, bc, data);
 730                         if (error)
 731                                 break;
 732                 }
 733
 734 out:
 735         brelse(bh);
 736         return error;
 737 }
 738
 739 /**
 740  * do_strip - Look for a layer a particular layer of the file and strip it off
 741  * @ip: the inode
 742  * @dibh: the dinode buffer
 743  * @bh: A buffer of pointers
 744  * @top: The first pointer in the buffer
 745  * @bottom: One more than the last pointer
 746  * @height: the height this buffer is at
 747  * @data: a pointer to a struct strip_mine
 748  *
 749  * Returns: errno
 750  */
 751
 752 static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh,
 753                     struct buffer_head *bh, __be64 *top, __be64 *bottom,
 754                     unsigned int height, void *data)
 755 {
 756         struct strip_mine *sm = data;
 757         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 758         struct gfs2_rgrp_list rlist;
 759         u64 bn, bstart;
 760         u32 blen;
 761         __be64 *p;
 762         unsigned int rg_blocks = 0;
 763         int metadata;
 764         unsigned int revokes = 0;
 765         int x;
 766         int error;
 767
 768         if (!*top)
 769                 sm->sm_first = 0;
 770
 771         if (height != sm->sm_height)
 772                 return 0;
 773
 774         if (sm->sm_first) {
 775                 top++;
 776                 sm->sm_first = 0;
 777         }
 778
 779         metadata = (height != ip->i_height - 1);
 780         if (metadata)
 781                 revokes = (height) ? sdp->sd_inptrs : sdp->sd_diptrs;
 782
 783         error = gfs2_rindex_hold(sdp, &ip->i_alloc->al_ri_gh);
 784         if (error)
 785                 return error;
 786
 787         memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
 788         bstart = 0;
 789         blen = 0;
 790
 791         for (p = top; p < bottom; p++) {
 792                 if (!*p)
 793                         continue;
 794
 795                 bn = be64_to_cpu(*p);
 796
 797                 if (bstart + blen == bn)
 798                         blen++;
 799                 else {
 800                         if (bstart)
 801                                 gfs2_rlist_add(sdp, &rlist, bstart);
 802
 803                         bstart = bn;
 804                         blen = 1;
 805                 }
 806         }
 807
 808         if (bstart)
 809                 gfs2_rlist_add(sdp, &rlist, bstart);
 810         else
 811                 goto out; /* Nothing to do */
 812
 813         gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE);
 814
 815         for (x = 0; x < rlist.rl_rgrps; x++) {
 816                 struct gfs2_rgrpd *rgd;
 817                 rgd = rlist.rl_ghs[x].gh_gl->gl_object;
 818                 rg_blocks += rgd->rd_length;
 819         }
 820
 821         error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
 822         if (error)
 823                 goto out_rlist;
 824
 825         error = gfs2_trans_begin(sdp, rg_blocks + RES_DINODE +
 826                                  RES_INDIRECT + RES_STATFS + RES_QUOTA,
 827                                  revokes);
 828         if (error)
 829                 goto out_rg_gunlock;
 830
 831         down_write(&ip->i_rw_mutex);
 832
 833         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 834         gfs2_trans_add_bh(ip->i_gl, bh, 1);
 835
 836         bstart = 0;
 837         blen = 0;
 838
 839         for (p = top; p < bottom; p++) {
 840                 if (!*p)
 841                         continue;
 842
 843                 bn = be64_to_cpu(*p);
 844
 845                 if (bstart + blen == bn)
 846                         blen++;
 847                 else {
 848                         if (bstart) {
 849                                 if (metadata)
 850                                         gfs2_free_meta(ip, bstart, blen);
 851                                 else
 852                                         gfs2_free_data(ip, bstart, blen);
 853                         }
 854
 855                         bstart = bn;
 856                         blen = 1;
 857                 }
 858
 859                 *p = 0;
 860                 gfs2_add_inode_blocks(&ip->i_inode, -1);
 861         }
 862         if (bstart) {
 863                 if (metadata)
 864                         gfs2_free_meta(ip, bstart, blen);
 865                 else
 866                         gfs2_free_data(ip, bstart, blen);
 867         }
 868
 869         ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
 870
 871         gfs2_dinode_out(ip, dibh->b_data);
 872
 873         up_write(&ip->i_rw_mutex);
 874
 875         gfs2_trans_end(sdp);
 876
 877 out_rg_gunlock:
 878         gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
 879 out_rlist:
 880         gfs2_rlist_free(&rlist);
 881 out:
 882         gfs2_glock_dq_uninit(&ip->i_alloc->al_ri_gh);
 883         return error;
 884 }
 885
 886 /**
 887  * do_grow - Make a file look bigger than it is
 888  * @ip: the inode
 889  * @size: the size to set the file to
 890  *
 891  * Called with an exclusive lock on @ip.
 892  *
 893  * Returns: errno
 894  */
 895
 896 static int do_grow(struct gfs2_inode *ip, u64 size)
 897 {
 898         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 899         struct gfs2_alloc *al;
 900         struct buffer_head *dibh;
 901         int error;
 902
 903         al = gfs2_alloc_get(ip);
 904         if (!al)
 905                 return -ENOMEM;
 906
 907         error = gfs2_quota_lock_check(ip);
 908         if (error)
 909                 goto out;
 910
 911         al->al_requested = sdp->sd_max_height + RES_DATA;
 912
 913         error = gfs2_inplace_reserve(ip);
 914         if (error)
 915                 goto out_gunlock_q;
 916
 917         error = gfs2_trans_begin(sdp,
 918                         sdp->sd_max_height + al->al_rgd->rd_length +
 919                         RES_JDATA + RES_DINODE + RES_STATFS + RES_QUOTA, 0);
 920         if (error)
 921                 goto out_ipres;
 922
 923         error = gfs2_meta_inode_buffer(ip, &dibh);
 924         if (error)
 925                 goto out_end_trans;
 926
 927         if (size > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
 928                 if (gfs2_is_stuffed(ip)) {
 929                         error = gfs2_unstuff_dinode(ip, NULL);
 930                         if (error)
 931                                 goto out_brelse;
 932                 }
 933         }
 934
 935         ip->i_disksize = size;
 936         ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
 937         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 938         gfs2_dinode_out(ip, dibh->b_data);
 939
 940 out_brelse:
 941         brelse(dibh);
 942 out_end_trans:
 943         gfs2_trans_end(sdp);
 944 out_ipres:
 945         gfs2_inplace_release(ip);
 946 out_gunlock_q:
 947         gfs2_quota_unlock(ip);
 948 out:
 949         gfs2_alloc_put(ip);
 950         return error;
 951 }
 952
 953
 954 /**
 955  * gfs2_block_truncate_page - Deal with zeroing out data for truncate
 956  *
 957  * This is partly borrowed from ext3.
 958  */
 959 static int gfs2_block_truncate_page(struct address_space *mapping)
 960 {
 961         struct inode *inode = mapping->host;
 962         struct gfs2_inode *ip = GFS2_I(inode);
 963         loff_t from = inode->i_size;
 964         unsigned long index = from >> PAGE_CACHE_SHIFT;
 965         unsigned offset = from & (PAGE_CACHE_SIZE-1);
 966         unsigned blocksize, iblock, length, pos;
 967         struct buffer_head *bh;
 968         struct page *page;
 969         int err;
 970
 971         page = grab_cache_page(mapping, index);
 972         if (!page)
 973                 return 0;
 974
 975         blocksize = inode->i_sb->s_blocksize;
 976         length = blocksize - (offset & (blocksize - 1));
 977         iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
 978
 979         if (!page_has_buffers(page))
 980                 create_empty_buffers(page, blocksize, 0);
 981
 982         /* Find the buffer that contains "offset" */
 983         bh = page_buffers(page);
 984         pos = blocksize;
 985         while (offset >= pos) {
 986                 bh = bh->b_this_page;
 987                 iblock++;
 988                 pos += blocksize;
 989         }
 990
 991         err = 0;
 992
 993         if (!buffer_mapped(bh)) {
 994                 gfs2_block_map(inode, iblock, bh, 0);
 995                 /* unmapped? It's a hole - nothing to do */
 996                 if (!buffer_mapped(bh))
 997                         goto unlock;
 998         }
 999
1000         /* Ok, it's mapped. Make sure it's up-to-date */
1001         if (PageUptodate(page))
1002                 set_buffer_uptodate(bh);
1003
1004         if (!buffer_uptodate(bh)) {
1005                 err = -EIO;
1006                 ll_rw_block(READ, 1, &bh);
1007                 wait_on_buffer(bh);
1008                 /* Uhhuh. Read error. Complain and punt. */
1009                 if (!buffer_uptodate(bh))
1010                         goto unlock;
1011                 err = 0;
1012         }
1013
1014         if (!gfs2_is_writeback(ip))
1015                 gfs2_trans_add_bh(ip->i_gl, bh, 0);
1016
1017         zero_user(page, offset, length);
1018         mark_buffer_dirty(bh);
1019 unlock:
1020         unlock_page(page);
1021         page_cache_release(page);
1022         return err;
1023 }
1024
1025 static int trunc_start(struct gfs2_inode *ip, u64 size)
1026 {
1027         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1028         struct buffer_head *dibh;
1029         int journaled = gfs2_is_jdata(ip);
1030         int error;
1031
1032         error = gfs2_trans_begin(sdp,
1033                                  RES_DINODE + (journaled ? RES_JDATA : 0), 0);
1034         if (error)
1035                 return error;
1036
1037         error = gfs2_meta_inode_buffer(ip, &dibh);
1038         if (error)
1039                 goto out;
1040
1041         if (gfs2_is_stuffed(ip)) {
1042                 ip->i_disksize = size;
1043                 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1044                 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1045                 gfs2_dinode_out(ip, dibh->b_data);
1046                 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + size);
1047                 error = 1;
1048
1049         } else {
1050                 if (size & (u64)(sdp->sd_sb.sb_bsize - 1))
1051                         error = gfs2_block_truncate_page(ip->i_inode.i_mapping);
1052
1053                 if (!error) {
1054                         ip->i_disksize = size;
1055                         ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1056                         ip->i_diskflags |= GFS2_DIF_TRUNC_IN_PROG;
1057                         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1058                         gfs2_dinode_out(ip, dibh->b_data);
1059                 }
1060         }
1061
1062         brelse(dibh);
1063
1064 out:
1065         gfs2_trans_end(sdp);
1066         return error;
1067 }
1068
1069 static int trunc_dealloc(struct gfs2_inode *ip, u64 size)
1070 {
1071         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1072         unsigned int height = ip->i_height;
1073         u64 lblock;
1074         struct metapath mp;
1075         int error;
1076
1077         if (!size)
1078                 lblock = 0;
1079         else
1080                 lblock = (size - 1) >> sdp->sd_sb.sb_bsize_shift;
1081
1082         find_metapath(sdp, lblock, &mp, ip->i_height);
1083         if (!gfs2_alloc_get(ip))
1084                 return -ENOMEM;
1085
1086         error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1087         if (error)
1088                 goto out;
1089
1090         while (height--) {
1091                 struct strip_mine sm;
1092                 sm.sm_first = !!size;
1093                 sm.sm_height = height;
1094
1095                 error = recursive_scan(ip, NULL, &mp, 0, 0, 1, do_strip, &sm);
1096                 if (error)
1097                         break;
1098         }
1099
1100         gfs2_quota_unhold(ip);
1101
1102 out:
1103         gfs2_alloc_put(ip);
1104         return error;
1105 }
1106
1107 static int trunc_end(struct gfs2_inode *ip)
1108 {
1109         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1110         struct buffer_head *dibh;
1111         int error;
1112
1113         error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1114         if (error)
1115                 return error;
1116
1117         down_write(&ip->i_rw_mutex);
1118
1119         error = gfs2_meta_inode_buffer(ip, &dibh);
1120         if (error)
1121                 goto out;
1122
1123         if (!ip->i_disksize) {
1124                 ip->i_height = 0;
1125                 ip->i_goal = ip->i_no_addr;
1126                 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
1127         }
1128         ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1129         ip->i_diskflags &= ~GFS2_DIF_TRUNC_IN_PROG;
1130
1131         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1132         gfs2_dinode_out(ip, dibh->b_data);
1133         brelse(dibh);
1134
1135 out:
1136         up_write(&ip->i_rw_mutex);
1137         gfs2_trans_end(sdp);
1138         return error;
1139 }
1140
1141 /**
1142  * do_shrink - make a file smaller
1143  * @ip: the inode
1144  * @size: the size to make the file
1145  * @truncator: function to truncate the last partial block
1146  *
1147  * Called with an exclusive lock on @ip.
1148  *
1149  * Returns: errno
1150  */
1151
1152 static int do_shrink(struct gfs2_inode *ip, u64 size)
1153 {
1154         int error;
1155
1156         error = trunc_start(ip, size);
1157         if (error < 0)
1158                 return error;
1159         if (error > 0)
1160                 return 0;
1161
1162         error = trunc_dealloc(ip, size);
1163         if (!error)
1164                 error = trunc_end(ip);
1165
1166         return error;
1167 }
1168
1169 static int do_touch(struct gfs2_inode *ip, u64 size)
1170 {
1171         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1172         struct buffer_head *dibh;
1173         int error;
1174
1175         error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1176         if (error)
1177                 return error;
1178
1179         down_write(&ip->i_rw_mutex);
1180
1181         error = gfs2_meta_inode_buffer(ip, &dibh);
1182         if (error)
1183                 goto do_touch_out;
1184
1185         ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1186         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1187         gfs2_dinode_out(ip, dibh->b_data);
1188         brelse(dibh);
1189
1190 do_touch_out:
1191         up_write(&ip->i_rw_mutex);
1192         gfs2_trans_end(sdp);
1193         return error;
1194 }
1195
1196 /**
1197  * gfs2_truncatei - make a file a given size
1198  * @ip: the inode
1199  * @size: the size to make the file
1200  * @truncator: function to truncate the last partial block
1201  *
1202  * The file size can grow, shrink, or stay the same size.
1203  *
1204  * Returns: errno
1205  */
1206
1207 int gfs2_truncatei(struct gfs2_inode *ip, u64 size)
1208 {
1209         int error;
1210
1211         if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), S_ISREG(ip->i_inode.i_mode)))
1212                 return -EINVAL;
1213
1214         if (size > ip->i_disksize)
1215                 error = do_grow(ip, size);
1216         else if (size < ip->i_disksize)
1217                 error = do_shrink(ip, size);
1218         else
1219                 /* update time stamps */
1220                 error = do_touch(ip, size);
1221
1222         return error;
1223 }
1224
1225 int gfs2_truncatei_resume(struct gfs2_inode *ip)
1226 {
1227         int error;
1228         error = trunc_dealloc(ip, ip->i_disksize);
1229         if (!error)
1230                 error = trunc_end(ip);
1231         return error;
1232 }
1233
1234 int gfs2_file_dealloc(struct gfs2_inode *ip)
1235 {
1236         return trunc_dealloc(ip, 0);
1237 }
1238
1239 /**
1240  * gfs2_write_alloc_required - figure out if a write will require an allocation
1241  * @ip: the file being written to
1242  * @offset: the offset to write to
1243  * @len: the number of bytes being written
1244  * @alloc_required: set to 1 if an alloc is required, 0 otherwise
1245  *
1246  * Returns: errno
1247  */
1248
1249 int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset,
1250                               unsigned int len, int *alloc_required)
1251 {
1252         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1253         struct buffer_head bh;
1254         unsigned int shift;
1255         u64 lblock, lblock_stop, size;
1256         u64 end_of_file;
1257
1258         *alloc_required = 0;
1259
1260         if (!len)
1261                 return 0;
1262
1263         if (gfs2_is_stuffed(ip)) {
1264                 if (offset + len >
1265                     sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
1266                         *alloc_required = 1;
1267                 return 0;
1268         }
1269
1270         *alloc_required = 1;
1271         shift = sdp->sd_sb.sb_bsize_shift;
1272         BUG_ON(gfs2_is_dir(ip));
1273         end_of_file = (ip->i_disksize + sdp->sd_sb.sb_bsize - 1) >> shift;
1274         lblock = offset >> shift;
1275         lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift;
1276         if (lblock_stop > end_of_file)
1277                 return 0;
1278
1279         size = (lblock_stop - lblock) << shift;
1280         do {
1281                 bh.b_state = 0;
1282                 bh.b_size = size;
1283                 gfs2_block_map(&ip->i_inode, lblock, &bh, 0);
1284                 if (!buffer_mapped(&bh))
1285                         return 0;
1286                 size -= bh.b_size;
1287                 lblock += (bh.b_size >> ip->i_inode.i_blkbits);
1288         } while(size > 0);
1289
1290         *alloc_required = 0;
1291         return 0;
1292 }
1293