linux/fs/gfs2/bmap.c
<<
>>
Prefs
   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/blkdev.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 "super.h"
  26#include "trans.h"
  27#include "dir.h"
  28#include "util.h"
  29#include "trace_gfs2.h"
  30
  31/* This doesn't need to be that large as max 64 bit pointers in a 4k
  32 * block is 512, so __u16 is fine for that. It saves stack space to
  33 * keep it small.
  34 */
  35struct metapath {
  36        struct buffer_head *mp_bh[GFS2_MAX_META_HEIGHT];
  37        __u16 mp_list[GFS2_MAX_META_HEIGHT];
  38};
  39
  40struct strip_mine {
  41        int sm_first;
  42        unsigned int sm_height;
  43};
  44
  45/**
  46 * gfs2_unstuffer_page - unstuff a stuffed inode into a block cached by a page
  47 * @ip: the inode
  48 * @dibh: the dinode buffer
  49 * @block: the block number that was allocated
  50 * @page: The (optional) page. This is looked up if @page is NULL
  51 *
  52 * Returns: errno
  53 */
  54
  55static int gfs2_unstuffer_page(struct gfs2_inode *ip, struct buffer_head *dibh,
  56                               u64 block, struct page *page)
  57{
  58        struct inode *inode = &ip->i_inode;
  59        struct buffer_head *bh;
  60        int release = 0;
  61
  62        if (!page || page->index) {
  63                page = find_or_create_page(inode->i_mapping, 0, GFP_NOFS);
  64                if (!page)
  65                        return -ENOMEM;
  66                release = 1;
  67        }
  68
  69        if (!PageUptodate(page)) {
  70                void *kaddr = kmap(page);
  71                u64 dsize = i_size_read(inode);
  72 
  73                if (dsize > (dibh->b_size - sizeof(struct gfs2_dinode)))
  74                        dsize = dibh->b_size - sizeof(struct gfs2_dinode);
  75
  76                memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
  77                memset(kaddr + dsize, 0, PAGE_CACHE_SIZE - dsize);
  78                kunmap(page);
  79
  80                SetPageUptodate(page);
  81        }
  82
  83        if (!page_has_buffers(page))
  84                create_empty_buffers(page, 1 << inode->i_blkbits,
  85                                     (1 << BH_Uptodate));
  86
  87        bh = page_buffers(page);
  88
  89        if (!buffer_mapped(bh))
  90                map_bh(bh, inode->i_sb, block);
  91
  92        set_buffer_uptodate(bh);
  93        if (!gfs2_is_jdata(ip))
  94                mark_buffer_dirty(bh);
  95        if (!gfs2_is_writeback(ip))
  96                gfs2_trans_add_bh(ip->i_gl, bh, 0);
  97
  98        if (release) {
  99                unlock_page(page);
 100                page_cache_release(page);
 101        }
 102
 103        return 0;
 104}
 105
 106/**
 107 * gfs2_unstuff_dinode - Unstuff a dinode when the data has grown too big
 108 * @ip: The GFS2 inode to unstuff
 109 * @page: The (optional) page. This is looked up if the @page is NULL
 110 *
 111 * This routine unstuffs a dinode and returns it to a "normal" state such
 112 * that the height can be grown in the traditional way.
 113 *
 114 * Returns: errno
 115 */
 116
 117int gfs2_unstuff_dinode(struct gfs2_inode *ip, struct page *page)
 118{
 119        struct buffer_head *bh, *dibh;
 120        struct gfs2_dinode *di;
 121        u64 block = 0;
 122        int isdir = gfs2_is_dir(ip);
 123        int error;
 124
 125        down_write(&ip->i_rw_mutex);
 126
 127        error = gfs2_meta_inode_buffer(ip, &dibh);
 128        if (error)
 129                goto out;
 130
 131        if (i_size_read(&ip->i_inode)) {
 132                /* Get a free block, fill it with the stuffed data,
 133                   and write it out to disk */
 134
 135                unsigned int n = 1;
 136                error = gfs2_alloc_blocks(ip, &block, &n, 0, NULL);
 137                if (error)
 138                        goto out_brelse;
 139                if (isdir) {
 140                        gfs2_trans_add_unrevoke(GFS2_SB(&ip->i_inode), block, 1);
 141                        error = gfs2_dir_get_new_buffer(ip, block, &bh);
 142                        if (error)
 143                                goto out_brelse;
 144                        gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_meta_header),
 145                                              dibh, sizeof(struct gfs2_dinode));
 146                        brelse(bh);
 147                } else {
 148                        error = gfs2_unstuffer_page(ip, dibh, block, page);
 149                        if (error)
 150                                goto out_brelse;
 151                }
 152        }
 153
 154        /*  Set up the pointer to the new block  */
 155
 156        gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 157        di = (struct gfs2_dinode *)dibh->b_data;
 158        gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
 159
 160        if (i_size_read(&ip->i_inode)) {
 161                *(__be64 *)(di + 1) = cpu_to_be64(block);
 162                gfs2_add_inode_blocks(&ip->i_inode, 1);
 163                di->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
 164        }
 165
 166        ip->i_height = 1;
 167        di->di_height = cpu_to_be16(1);
 168
 169out_brelse:
 170        brelse(dibh);
 171out:
 172        up_write(&ip->i_rw_mutex);
 173        return error;
 174}
 175
 176
 177/**
 178 * find_metapath - Find path through the metadata tree
 179 * @sdp: The superblock
 180 * @mp: The metapath to return the result in
 181 * @block: The disk block to look up
 182 * @height: The pre-calculated height of the metadata tree
 183 *
 184 *   This routine returns a struct metapath structure that defines a path
 185 *   through the metadata of inode "ip" to get to block "block".
 186 *
 187 *   Example:
 188 *   Given:  "ip" is a height 3 file, "offset" is 101342453, and this is a
 189 *   filesystem with a blocksize of 4096.
 190 *
 191 *   find_metapath() would return a struct metapath structure set to:
 192 *   mp_offset = 101342453, mp_height = 3, mp_list[0] = 0, mp_list[1] = 48,
 193 *   and mp_list[2] = 165.
 194 *
 195 *   That means that in order to get to the block containing the byte at
 196 *   offset 101342453, we would load the indirect block pointed to by pointer
 197 *   0 in the dinode.  We would then load the indirect block pointed to by
 198 *   pointer 48 in that indirect block.  We would then load the data block
 199 *   pointed to by pointer 165 in that indirect block.
 200 *
 201 *             ----------------------------------------
 202 *             | Dinode |                             |
 203 *             |        |                            4|
 204 *             |        |0 1 2 3 4 5                 9|
 205 *             |        |                            6|
 206 *             ----------------------------------------
 207 *                       |
 208 *                       |
 209 *                       V
 210 *             ----------------------------------------
 211 *             | Indirect Block                       |
 212 *             |                                     5|
 213 *             |            4 4 4 4 4 5 5            1|
 214 *             |0           5 6 7 8 9 0 1            2|
 215 *             ----------------------------------------
 216 *                                |
 217 *                                |
 218 *                                V
 219 *             ----------------------------------------
 220 *             | Indirect Block                       |
 221 *             |                         1 1 1 1 1   5|
 222 *             |                         6 6 6 6 6   1|
 223 *             |0                        3 4 5 6 7   2|
 224 *             ----------------------------------------
 225 *                                           |
 226 *                                           |
 227 *                                           V
 228 *             ----------------------------------------
 229 *             | Data block containing offset         |
 230 *             |            101342453                 |
 231 *             |                                      |
 232 *             |                                      |
 233 *             ----------------------------------------
 234 *
 235 */
 236
 237static void find_metapath(const struct gfs2_sbd *sdp, u64 block,
 238                          struct metapath *mp, unsigned int height)
 239{
 240        unsigned int i;
 241
 242        for (i = height; i--;)
 243                mp->mp_list[i] = do_div(block, sdp->sd_inptrs);
 244
 245}
 246
 247static inline unsigned int metapath_branch_start(const struct metapath *mp)
 248{
 249        if (mp->mp_list[0] == 0)
 250                return 2;
 251        return 1;
 252}
 253
 254/**
 255 * metapointer - Return pointer to start of metadata in a buffer
 256 * @height: The metadata height (0 = dinode)
 257 * @mp: The metapath
 258 *
 259 * Return a pointer to the block number of the next height of the metadata
 260 * tree given a buffer containing the pointer to the current height of the
 261 * metadata tree.
 262 */
 263
 264static inline __be64 *metapointer(unsigned int height, const struct metapath *mp)
 265{
 266        struct buffer_head *bh = mp->mp_bh[height];
 267        unsigned int head_size = (height > 0) ?
 268                sizeof(struct gfs2_meta_header) : sizeof(struct gfs2_dinode);
 269        return ((__be64 *)(bh->b_data + head_size)) + mp->mp_list[height];
 270}
 271
 272static void gfs2_metapath_ra(struct gfs2_glock *gl,
 273                             const struct buffer_head *bh, const __be64 *pos)
 274{
 275        struct buffer_head *rabh;
 276        const __be64 *endp = (const __be64 *)(bh->b_data + bh->b_size);
 277        const __be64 *t;
 278
 279        for (t = pos; t < endp; t++) {
 280                if (!*t)
 281                        continue;
 282
 283                rabh = gfs2_getbuf(gl, be64_to_cpu(*t), CREATE);
 284                if (trylock_buffer(rabh)) {
 285                        if (!buffer_uptodate(rabh)) {
 286                                rabh->b_end_io = end_buffer_read_sync;
 287                                submit_bh(READA | REQ_META, rabh);
 288                                continue;
 289                        }
 290                        unlock_buffer(rabh);
 291                }
 292                brelse(rabh);
 293        }
 294}
 295
 296/**
 297 * lookup_metapath - Walk the metadata tree to a specific point
 298 * @ip: The inode
 299 * @mp: The metapath
 300 *
 301 * Assumes that the inode's buffer has already been looked up and
 302 * hooked onto mp->mp_bh[0] and that the metapath has been initialised
 303 * by find_metapath().
 304 *
 305 * If this function encounters part of the tree which has not been
 306 * allocated, it returns the current height of the tree at the point
 307 * at which it found the unallocated block. Blocks which are found are
 308 * added to the mp->mp_bh[] list.
 309 *
 310 * Returns: error or height of metadata tree
 311 */
 312
 313static int lookup_metapath(struct gfs2_inode *ip, struct metapath *mp)
 314{
 315        unsigned int end_of_metadata = ip->i_height - 1;
 316        unsigned int x;
 317        __be64 *ptr;
 318        u64 dblock;
 319        int ret;
 320
 321        for (x = 0; x < end_of_metadata; x++) {
 322                ptr = metapointer(x, mp);
 323                dblock = be64_to_cpu(*ptr);
 324                if (!dblock)
 325                        return x + 1;
 326
 327                ret = gfs2_meta_indirect_buffer(ip, x+1, dblock, &mp->mp_bh[x+1]);
 328                if (ret)
 329                        return ret;
 330        }
 331
 332        return ip->i_height;
 333}
 334
 335static inline void release_metapath(struct metapath *mp)
 336{
 337        int i;
 338
 339        for (i = 0; i < GFS2_MAX_META_HEIGHT; i++) {
 340                if (mp->mp_bh[i] == NULL)
 341                        break;
 342                brelse(mp->mp_bh[i]);
 343        }
 344}
 345
 346/**
 347 * gfs2_extent_length - Returns length of an extent of blocks
 348 * @start: Start of the buffer
 349 * @len: Length of the buffer in bytes
 350 * @ptr: Current position in the buffer
 351 * @limit: Max extent length to return (0 = unlimited)
 352 * @eob: Set to 1 if we hit "end of block"
 353 *
 354 * If the first block is zero (unallocated) it will return the number of
 355 * unallocated blocks in the extent, otherwise it will return the number
 356 * of contiguous blocks in the extent.
 357 *
 358 * Returns: The length of the extent (minimum of one block)
 359 */
 360
 361static inline unsigned int gfs2_extent_length(void *start, unsigned int len, __be64 *ptr, unsigned limit, int *eob)
 362{
 363        const __be64 *end = (start + len);
 364        const __be64 *first = ptr;
 365        u64 d = be64_to_cpu(*ptr);
 366
 367        *eob = 0;
 368        do {
 369                ptr++;
 370                if (ptr >= end)
 371                        break;
 372                if (limit && --limit == 0)
 373                        break;
 374                if (d)
 375                        d++;
 376        } while(be64_to_cpu(*ptr) == d);
 377        if (ptr >= end)
 378                *eob = 1;
 379        return (ptr - first);
 380}
 381
 382static inline void bmap_lock(struct gfs2_inode *ip, int create)
 383{
 384        if (create)
 385                down_write(&ip->i_rw_mutex);
 386        else
 387                down_read(&ip->i_rw_mutex);
 388}
 389
 390static inline void bmap_unlock(struct gfs2_inode *ip, int create)
 391{
 392        if (create)
 393                up_write(&ip->i_rw_mutex);
 394        else
 395                up_read(&ip->i_rw_mutex);
 396}
 397
 398static inline __be64 *gfs2_indirect_init(struct metapath *mp,
 399                                         struct gfs2_glock *gl, unsigned int i,
 400                                         unsigned offset, u64 bn)
 401{
 402        __be64 *ptr = (__be64 *)(mp->mp_bh[i - 1]->b_data +
 403                       ((i > 1) ? sizeof(struct gfs2_meta_header) :
 404                                 sizeof(struct gfs2_dinode)));
 405        BUG_ON(i < 1);
 406        BUG_ON(mp->mp_bh[i] != NULL);
 407        mp->mp_bh[i] = gfs2_meta_new(gl, bn);
 408        gfs2_trans_add_bh(gl, mp->mp_bh[i], 1);
 409        gfs2_metatype_set(mp->mp_bh[i], GFS2_METATYPE_IN, GFS2_FORMAT_IN);
 410        gfs2_buffer_clear_tail(mp->mp_bh[i], sizeof(struct gfs2_meta_header));
 411        ptr += offset;
 412        *ptr = cpu_to_be64(bn);
 413        return ptr;
 414}
 415
 416enum alloc_state {
 417        ALLOC_DATA = 0,
 418        ALLOC_GROW_DEPTH = 1,
 419        ALLOC_GROW_HEIGHT = 2,
 420        /* ALLOC_UNSTUFF = 3,   TBD and rather complicated */
 421};
 422
 423/**
 424 * gfs2_bmap_alloc - Build a metadata tree of the requested height
 425 * @inode: The GFS2 inode
 426 * @lblock: The logical starting block of the extent
 427 * @bh_map: This is used to return the mapping details
 428 * @mp: The metapath
 429 * @sheight: The starting height (i.e. whats already mapped)
 430 * @height: The height to build to
 431 * @maxlen: The max number of data blocks to alloc
 432 *
 433 * In this routine we may have to alloc:
 434 *   i) Indirect blocks to grow the metadata tree height
 435 *  ii) Indirect blocks to fill in lower part of the metadata tree
 436 * iii) Data blocks
 437 *
 438 * The function is in two parts. The first part works out the total
 439 * number of blocks which we need. The second part does the actual
 440 * allocation asking for an extent at a time (if enough contiguous free
 441 * blocks are available, there will only be one request per bmap call)
 442 * and uses the state machine to initialise the blocks in order.
 443 *
 444 * Returns: errno on error
 445 */
 446
 447static int gfs2_bmap_alloc(struct inode *inode, const sector_t lblock,
 448                           struct buffer_head *bh_map, struct metapath *mp,
 449                           const unsigned int sheight,
 450                           const unsigned int height,
 451                           const unsigned int maxlen)
 452{
 453        struct gfs2_inode *ip = GFS2_I(inode);
 454        struct gfs2_sbd *sdp = GFS2_SB(inode);
 455        struct super_block *sb = sdp->sd_vfs;
 456        struct buffer_head *dibh = mp->mp_bh[0];
 457        u64 bn, dblock = 0;
 458        unsigned n, i, blks, alloced = 0, iblks = 0, branch_start = 0;
 459        unsigned dblks = 0;
 460        unsigned ptrs_per_blk;
 461        const unsigned end_of_metadata = height - 1;
 462        int ret;
 463        int eob = 0;
 464        enum alloc_state state;
 465        __be64 *ptr;
 466        __be64 zero_bn = 0;
 467
 468        BUG_ON(sheight < 1);
 469        BUG_ON(dibh == NULL);
 470
 471        gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 472
 473        if (height == sheight) {
 474                struct buffer_head *bh;
 475                /* Bottom indirect block exists, find unalloced extent size */
 476                ptr = metapointer(end_of_metadata, mp);
 477                bh = mp->mp_bh[end_of_metadata];
 478                dblks = gfs2_extent_length(bh->b_data, bh->b_size, ptr, maxlen,
 479                                           &eob);
 480                BUG_ON(dblks < 1);
 481                state = ALLOC_DATA;
 482        } else {
 483                /* Need to allocate indirect blocks */
 484                ptrs_per_blk = height > 1 ? sdp->sd_inptrs : sdp->sd_diptrs;
 485                dblks = min(maxlen, ptrs_per_blk - mp->mp_list[end_of_metadata]);
 486                if (height == ip->i_height) {
 487                        /* Writing into existing tree, extend tree down */
 488                        iblks = height - sheight;
 489                        state = ALLOC_GROW_DEPTH;
 490                } else {
 491                        /* Building up tree height */
 492                        state = ALLOC_GROW_HEIGHT;
 493                        iblks = height - ip->i_height;
 494                        branch_start = metapath_branch_start(mp);
 495                        iblks += (height - branch_start);
 496                }
 497        }
 498
 499        /* start of the second part of the function (state machine) */
 500
 501        blks = dblks + iblks;
 502        i = sheight;
 503        do {
 504                int error;
 505                n = blks - alloced;
 506                error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL);
 507                if (error)
 508                        return error;
 509                alloced += n;
 510                if (state != ALLOC_DATA || gfs2_is_jdata(ip))
 511                        gfs2_trans_add_unrevoke(sdp, bn, n);
 512                switch (state) {
 513                /* Growing height of tree */
 514                case ALLOC_GROW_HEIGHT:
 515                        if (i == 1) {
 516                                ptr = (__be64 *)(dibh->b_data +
 517                                                 sizeof(struct gfs2_dinode));
 518                                zero_bn = *ptr;
 519                        }
 520                        for (; i - 1 < height - ip->i_height && n > 0; i++, n--)
 521                                gfs2_indirect_init(mp, ip->i_gl, i, 0, bn++);
 522                        if (i - 1 == height - ip->i_height) {
 523                                i--;
 524                                gfs2_buffer_copy_tail(mp->mp_bh[i],
 525                                                sizeof(struct gfs2_meta_header),
 526                                                dibh, sizeof(struct gfs2_dinode));
 527                                gfs2_buffer_clear_tail(dibh,
 528                                                sizeof(struct gfs2_dinode) +
 529                                                sizeof(__be64));
 530                                ptr = (__be64 *)(mp->mp_bh[i]->b_data +
 531                                        sizeof(struct gfs2_meta_header));
 532                                *ptr = zero_bn;
 533                                state = ALLOC_GROW_DEPTH;
 534                                for(i = branch_start; i < height; i++) {
 535                                        if (mp->mp_bh[i] == NULL)
 536                                                break;
 537                                        brelse(mp->mp_bh[i]);
 538                                        mp->mp_bh[i] = NULL;
 539                                }
 540                                i = branch_start;
 541                        }
 542                        if (n == 0)
 543                                break;
 544                /* Branching from existing tree */
 545                case ALLOC_GROW_DEPTH:
 546                        if (i > 1 && i < height)
 547                                gfs2_trans_add_bh(ip->i_gl, mp->mp_bh[i-1], 1);
 548                        for (; i < height && n > 0; i++, n--)
 549                                gfs2_indirect_init(mp, ip->i_gl, i,
 550                                                   mp->mp_list[i-1], bn++);
 551                        if (i == height)
 552                                state = ALLOC_DATA;
 553                        if (n == 0)
 554                                break;
 555                /* Tree complete, adding data blocks */
 556                case ALLOC_DATA:
 557                        BUG_ON(n > dblks);
 558                        BUG_ON(mp->mp_bh[end_of_metadata] == NULL);
 559                        gfs2_trans_add_bh(ip->i_gl, mp->mp_bh[end_of_metadata], 1);
 560                        dblks = n;
 561                        ptr = metapointer(end_of_metadata, mp);
 562                        dblock = bn;
 563                        while (n-- > 0)
 564                                *ptr++ = cpu_to_be64(bn++);
 565                        if (buffer_zeronew(bh_map)) {
 566                                ret = sb_issue_zeroout(sb, dblock, dblks,
 567                                                       GFP_NOFS);
 568                                if (ret) {
 569                                        fs_err(sdp,
 570                                               "Failed to zero data buffers\n");
 571                                        clear_buffer_zeronew(bh_map);
 572                                }
 573                        }
 574                        break;
 575                }
 576        } while ((state != ALLOC_DATA) || !dblock);
 577
 578        ip->i_height = height;
 579        gfs2_add_inode_blocks(&ip->i_inode, alloced);
 580        gfs2_dinode_out(ip, mp->mp_bh[0]->b_data);
 581        map_bh(bh_map, inode->i_sb, dblock);
 582        bh_map->b_size = dblks << inode->i_blkbits;
 583        set_buffer_new(bh_map);
 584        return 0;
 585}
 586
 587/**
 588 * gfs2_block_map - Map a block from an inode to a disk block
 589 * @inode: The inode
 590 * @lblock: The logical block number
 591 * @bh_map: The bh to be mapped
 592 * @create: True if its ok to alloc blocks to satify the request
 593 *
 594 * Sets buffer_mapped() if successful, sets buffer_boundary() if a
 595 * read of metadata will be required before the next block can be
 596 * mapped. Sets buffer_new() if new blocks were allocated.
 597 *
 598 * Returns: errno
 599 */
 600
 601int gfs2_block_map(struct inode *inode, sector_t lblock,
 602                   struct buffer_head *bh_map, int create)
 603{
 604        struct gfs2_inode *ip = GFS2_I(inode);
 605        struct gfs2_sbd *sdp = GFS2_SB(inode);
 606        unsigned int bsize = sdp->sd_sb.sb_bsize;
 607        const unsigned int maxlen = bh_map->b_size >> inode->i_blkbits;
 608        const u64 *arr = sdp->sd_heightsize;
 609        __be64 *ptr;
 610        u64 size;
 611        struct metapath mp;
 612        int ret;
 613        int eob;
 614        unsigned int len;
 615        struct buffer_head *bh;
 616        u8 height;
 617
 618        BUG_ON(maxlen == 0);
 619
 620        memset(mp.mp_bh, 0, sizeof(mp.mp_bh));
 621        bmap_lock(ip, create);
 622        clear_buffer_mapped(bh_map);
 623        clear_buffer_new(bh_map);
 624        clear_buffer_boundary(bh_map);
 625        trace_gfs2_bmap(ip, bh_map, lblock, create, 1);
 626        if (gfs2_is_dir(ip)) {
 627                bsize = sdp->sd_jbsize;
 628                arr = sdp->sd_jheightsize;
 629        }
 630
 631        ret = gfs2_meta_inode_buffer(ip, &mp.mp_bh[0]);
 632        if (ret)
 633                goto out;
 634
 635        height = ip->i_height;
 636        size = (lblock + 1) * bsize;
 637        while (size > arr[height])
 638                height++;
 639        find_metapath(sdp, lblock, &mp, height);
 640        ret = 1;
 641        if (height > ip->i_height || gfs2_is_stuffed(ip))
 642                goto do_alloc;
 643        ret = lookup_metapath(ip, &mp);
 644        if (ret < 0)
 645                goto out;
 646        if (ret != ip->i_height)
 647                goto do_alloc;
 648        ptr = metapointer(ip->i_height - 1, &mp);
 649        if (*ptr == 0)
 650                goto do_alloc;
 651        map_bh(bh_map, inode->i_sb, be64_to_cpu(*ptr));
 652        bh = mp.mp_bh[ip->i_height - 1];
 653        len = gfs2_extent_length(bh->b_data, bh->b_size, ptr, maxlen, &eob);
 654        bh_map->b_size = (len << inode->i_blkbits);
 655        if (eob)
 656                set_buffer_boundary(bh_map);
 657        ret = 0;
 658out:
 659        release_metapath(&mp);
 660        trace_gfs2_bmap(ip, bh_map, lblock, create, ret);
 661        bmap_unlock(ip, create);
 662        return ret;
 663
 664do_alloc:
 665        /* All allocations are done here, firstly check create flag */
 666        if (!create) {
 667                BUG_ON(gfs2_is_stuffed(ip));
 668                ret = 0;
 669                goto out;
 670        }
 671
 672        /* At this point ret is the tree depth of already allocated blocks */
 673        ret = gfs2_bmap_alloc(inode, lblock, bh_map, &mp, ret, height, maxlen);
 674        goto out;
 675}
 676
 677/*
 678 * Deprecated: do not use in new code
 679 */
 680int gfs2_extent_map(struct inode *inode, u64 lblock, int *new, u64 *dblock, unsigned *extlen)
 681{
 682        struct buffer_head bh = { .b_state = 0, .b_blocknr = 0 };
 683        int ret;
 684        int create = *new;
 685
 686        BUG_ON(!extlen);
 687        BUG_ON(!dblock);
 688        BUG_ON(!new);
 689
 690        bh.b_size = 1 << (inode->i_blkbits + (create ? 0 : 5));
 691        ret = gfs2_block_map(inode, lblock, &bh, create);
 692        *extlen = bh.b_size >> inode->i_blkbits;
 693        *dblock = bh.b_blocknr;
 694        if (buffer_new(&bh))
 695                *new = 1;
 696        else
 697                *new = 0;
 698        return ret;
 699}
 700
 701/**
 702 * do_strip - Look for a layer a particular layer of the file and strip it off
 703 * @ip: the inode
 704 * @dibh: the dinode buffer
 705 * @bh: A buffer of pointers
 706 * @top: The first pointer in the buffer
 707 * @bottom: One more than the last pointer
 708 * @height: the height this buffer is at
 709 * @data: a pointer to a struct strip_mine
 710 *
 711 * Returns: errno
 712 */
 713
 714static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh,
 715                    struct buffer_head *bh, __be64 *top, __be64 *bottom,
 716                    unsigned int height, struct strip_mine *sm)
 717{
 718        struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 719        struct gfs2_rgrp_list rlist;
 720        u64 bn, bstart;
 721        u32 blen, btotal;
 722        __be64 *p;
 723        unsigned int rg_blocks = 0;
 724        int metadata;
 725        unsigned int revokes = 0;
 726        int x;
 727        int error;
 728
 729        error = gfs2_rindex_update(sdp);
 730        if (error)
 731                return error;
 732
 733        if (!*top)
 734                sm->sm_first = 0;
 735
 736        if (height != sm->sm_height)
 737                return 0;
 738
 739        if (sm->sm_first) {
 740                top++;
 741                sm->sm_first = 0;
 742        }
 743
 744        metadata = (height != ip->i_height - 1);
 745        if (metadata)
 746                revokes = (height) ? sdp->sd_inptrs : sdp->sd_diptrs;
 747        else if (ip->i_depth)
 748                revokes = sdp->sd_inptrs;
 749
 750        memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
 751        bstart = 0;
 752        blen = 0;
 753
 754        for (p = top; p < bottom; p++) {
 755                if (!*p)
 756                        continue;
 757
 758                bn = be64_to_cpu(*p);
 759
 760                if (bstart + blen == bn)
 761                        blen++;
 762                else {
 763                        if (bstart)
 764                                gfs2_rlist_add(ip, &rlist, bstart);
 765
 766                        bstart = bn;
 767                        blen = 1;
 768                }
 769        }
 770
 771        if (bstart)
 772                gfs2_rlist_add(ip, &rlist, bstart);
 773        else
 774                goto out; /* Nothing to do */
 775
 776        gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE);
 777
 778        for (x = 0; x < rlist.rl_rgrps; x++) {
 779                struct gfs2_rgrpd *rgd;
 780                rgd = rlist.rl_ghs[x].gh_gl->gl_object;
 781                rg_blocks += rgd->rd_length;
 782        }
 783
 784        error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
 785        if (error)
 786                goto out_rlist;
 787
 788        if (gfs2_rs_active(ip->i_res)) /* needs to be done with the rgrp glock held */
 789                gfs2_rs_deltree(ip->i_res);
 790
 791        error = gfs2_trans_begin(sdp, rg_blocks + RES_DINODE +
 792                                 RES_INDIRECT + RES_STATFS + RES_QUOTA,
 793                                 revokes);
 794        if (error)
 795                goto out_rg_gunlock;
 796
 797        down_write(&ip->i_rw_mutex);
 798
 799        gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 800        gfs2_trans_add_bh(ip->i_gl, bh, 1);
 801
 802        bstart = 0;
 803        blen = 0;
 804        btotal = 0;
 805
 806        for (p = top; p < bottom; p++) {
 807                if (!*p)
 808                        continue;
 809
 810                bn = be64_to_cpu(*p);
 811
 812                if (bstart + blen == bn)
 813                        blen++;
 814                else {
 815                        if (bstart) {
 816                                __gfs2_free_blocks(ip, bstart, blen, metadata);
 817                                btotal += blen;
 818                        }
 819
 820                        bstart = bn;
 821                        blen = 1;
 822                }
 823
 824                *p = 0;
 825                gfs2_add_inode_blocks(&ip->i_inode, -1);
 826        }
 827        if (bstart) {
 828                __gfs2_free_blocks(ip, bstart, blen, metadata);
 829                btotal += blen;
 830        }
 831
 832        gfs2_statfs_change(sdp, 0, +btotal, 0);
 833        gfs2_quota_change(ip, -(s64)btotal, ip->i_inode.i_uid,
 834                          ip->i_inode.i_gid);
 835
 836        ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
 837
 838        gfs2_dinode_out(ip, dibh->b_data);
 839
 840        up_write(&ip->i_rw_mutex);
 841
 842        gfs2_trans_end(sdp);
 843
 844out_rg_gunlock:
 845        gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
 846out_rlist:
 847        gfs2_rlist_free(&rlist);
 848out:
 849        return error;
 850}
 851
 852/**
 853 * recursive_scan - recursively scan through the end of a file
 854 * @ip: the inode
 855 * @dibh: the dinode buffer
 856 * @mp: the path through the metadata to the point to start
 857 * @height: the height the recursion is at
 858 * @block: the indirect block to look at
 859 * @first: 1 if this is the first block
 860 * @sm: data opaque to this function to pass to @bc
 861 *
 862 * When this is first called @height and @block should be zero and
 863 * @first should be 1.
 864 *
 865 * Returns: errno
 866 */
 867
 868static int recursive_scan(struct gfs2_inode *ip, struct buffer_head *dibh,
 869                          struct metapath *mp, unsigned int height,
 870                          u64 block, int first, struct strip_mine *sm)
 871{
 872        struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 873        struct buffer_head *bh = NULL;
 874        __be64 *top, *bottom;
 875        u64 bn;
 876        int error;
 877        int mh_size = sizeof(struct gfs2_meta_header);
 878
 879        if (!height) {
 880                error = gfs2_meta_inode_buffer(ip, &bh);
 881                if (error)
 882                        return error;
 883                dibh = bh;
 884
 885                top = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + mp->mp_list[0];
 886                bottom = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + sdp->sd_diptrs;
 887        } else {
 888                error = gfs2_meta_indirect_buffer(ip, height, block, &bh);
 889                if (error)
 890                        return error;
 891
 892                top = (__be64 *)(bh->b_data + mh_size) +
 893                                  (first ? mp->mp_list[height] : 0);
 894
 895                bottom = (__be64 *)(bh->b_data + mh_size) + sdp->sd_inptrs;
 896        }
 897
 898        error = do_strip(ip, dibh, bh, top, bottom, height, sm);
 899        if (error)
 900                goto out;
 901
 902        if (height < ip->i_height - 1) {
 903
 904                gfs2_metapath_ra(ip->i_gl, bh, top);
 905
 906                for (; top < bottom; top++, first = 0) {
 907                        if (!*top)
 908                                continue;
 909
 910                        bn = be64_to_cpu(*top);
 911
 912                        error = recursive_scan(ip, dibh, mp, height + 1, bn,
 913                                               first, sm);
 914                        if (error)
 915                                break;
 916                }
 917        }
 918out:
 919        brelse(bh);
 920        return error;
 921}
 922
 923
 924/**
 925 * gfs2_block_truncate_page - Deal with zeroing out data for truncate
 926 *
 927 * This is partly borrowed from ext3.
 928 */
 929static int gfs2_block_truncate_page(struct address_space *mapping, loff_t from)
 930{
 931        struct inode *inode = mapping->host;
 932        struct gfs2_inode *ip = GFS2_I(inode);
 933        unsigned long index = from >> PAGE_CACHE_SHIFT;
 934        unsigned offset = from & (PAGE_CACHE_SIZE-1);
 935        unsigned blocksize, iblock, length, pos;
 936        struct buffer_head *bh;
 937        struct page *page;
 938        int err;
 939
 940        page = find_or_create_page(mapping, index, GFP_NOFS);
 941        if (!page)
 942                return 0;
 943
 944        blocksize = inode->i_sb->s_blocksize;
 945        length = blocksize - (offset & (blocksize - 1));
 946        iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
 947
 948        if (!page_has_buffers(page))
 949                create_empty_buffers(page, blocksize, 0);
 950
 951        /* Find the buffer that contains "offset" */
 952        bh = page_buffers(page);
 953        pos = blocksize;
 954        while (offset >= pos) {
 955                bh = bh->b_this_page;
 956                iblock++;
 957                pos += blocksize;
 958        }
 959
 960        err = 0;
 961
 962        if (!buffer_mapped(bh)) {
 963                gfs2_block_map(inode, iblock, bh, 0);
 964                /* unmapped? It's a hole - nothing to do */
 965                if (!buffer_mapped(bh))
 966                        goto unlock;
 967        }
 968
 969        /* Ok, it's mapped. Make sure it's up-to-date */
 970        if (PageUptodate(page))
 971                set_buffer_uptodate(bh);
 972
 973        if (!buffer_uptodate(bh)) {
 974                err = -EIO;
 975                ll_rw_block(READ, 1, &bh);
 976                wait_on_buffer(bh);
 977                /* Uhhuh. Read error. Complain and punt. */
 978                if (!buffer_uptodate(bh))
 979                        goto unlock;
 980                err = 0;
 981        }
 982
 983        if (!gfs2_is_writeback(ip))
 984                gfs2_trans_add_bh(ip->i_gl, bh, 0);
 985
 986        zero_user(page, offset, length);
 987        mark_buffer_dirty(bh);
 988unlock:
 989        unlock_page(page);
 990        page_cache_release(page);
 991        return err;
 992}
 993
 994static int trunc_start(struct inode *inode, u64 oldsize, u64 newsize)
 995{
 996        struct gfs2_inode *ip = GFS2_I(inode);
 997        struct gfs2_sbd *sdp = GFS2_SB(inode);
 998        struct address_space *mapping = inode->i_mapping;
 999        struct buffer_head *dibh;
1000        int journaled = gfs2_is_jdata(ip);
1001        int error;
1002
1003        error = gfs2_trans_begin(sdp,
1004                                 RES_DINODE + (journaled ? RES_JDATA : 0), 0);
1005        if (error)
1006                return error;
1007
1008        error = gfs2_meta_inode_buffer(ip, &dibh);
1009        if (error)
1010                goto out;
1011
1012        gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1013
1014        if (gfs2_is_stuffed(ip)) {
1015                gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + newsize);
1016        } else {
1017                if (newsize & (u64)(sdp->sd_sb.sb_bsize - 1)) {
1018                        error = gfs2_block_truncate_page(mapping, newsize);
1019                        if (error)
1020                                goto out_brelse;
1021                }
1022                ip->i_diskflags |= GFS2_DIF_TRUNC_IN_PROG;
1023        }
1024
1025        i_size_write(inode, newsize);
1026        ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1027        gfs2_dinode_out(ip, dibh->b_data);
1028
1029        truncate_pagecache(inode, oldsize, newsize);
1030out_brelse:
1031        brelse(dibh);
1032out:
1033        gfs2_trans_end(sdp);
1034        return error;
1035}
1036
1037static int trunc_dealloc(struct gfs2_inode *ip, u64 size)
1038{
1039        struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1040        unsigned int height = ip->i_height;
1041        u64 lblock;
1042        struct metapath mp;
1043        int error;
1044
1045        if (!size)
1046                lblock = 0;
1047        else
1048                lblock = (size - 1) >> sdp->sd_sb.sb_bsize_shift;
1049
1050        find_metapath(sdp, lblock, &mp, ip->i_height);
1051        error = gfs2_rindex_update(sdp);
1052        if (error)
1053                return error;
1054
1055        error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1056        if (error)
1057                return error;
1058
1059        while (height--) {
1060                struct strip_mine sm;
1061                sm.sm_first = !!size;
1062                sm.sm_height = height;
1063
1064                error = recursive_scan(ip, NULL, &mp, 0, 0, 1, &sm);
1065                if (error)
1066                        break;
1067        }
1068
1069        gfs2_quota_unhold(ip);
1070
1071        return error;
1072}
1073
1074static int trunc_end(struct gfs2_inode *ip)
1075{
1076        struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1077        struct buffer_head *dibh;
1078        int error;
1079
1080        error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1081        if (error)
1082                return error;
1083
1084        down_write(&ip->i_rw_mutex);
1085
1086        error = gfs2_meta_inode_buffer(ip, &dibh);
1087        if (error)
1088                goto out;
1089
1090        if (!i_size_read(&ip->i_inode)) {
1091                ip->i_height = 0;
1092                ip->i_goal = ip->i_no_addr;
1093                gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
1094        }
1095        ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1096        ip->i_diskflags &= ~GFS2_DIF_TRUNC_IN_PROG;
1097
1098        gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1099        gfs2_dinode_out(ip, dibh->b_data);
1100        brelse(dibh);
1101
1102out:
1103        up_write(&ip->i_rw_mutex);
1104        gfs2_trans_end(sdp);
1105        return error;
1106}
1107
1108/**
1109 * do_shrink - make a file smaller
1110 * @inode: the inode
1111 * @oldsize: the current inode size
1112 * @newsize: the size to make the file
1113 *
1114 * Called with an exclusive lock on @inode. The @size must
1115 * be equal to or smaller than the current inode size.
1116 *
1117 * Returns: errno
1118 */
1119
1120static int do_shrink(struct inode *inode, u64 oldsize, u64 newsize)
1121{
1122        struct gfs2_inode *ip = GFS2_I(inode);
1123        int error;
1124
1125        error = trunc_start(inode, oldsize, newsize);
1126        if (error < 0)
1127                return error;
1128        if (gfs2_is_stuffed(ip))
1129                return 0;
1130
1131        error = trunc_dealloc(ip, newsize);
1132        if (error == 0)
1133                error = trunc_end(ip);
1134
1135        return error;
1136}
1137
1138void gfs2_trim_blocks(struct inode *inode)
1139{
1140        u64 size = inode->i_size;
1141        int ret;
1142
1143        ret = do_shrink(inode, size, size);
1144        WARN_ON(ret != 0);
1145}
1146
1147/**
1148 * do_grow - Touch and update inode size
1149 * @inode: The inode
1150 * @size: The new size
1151 *
1152 * This function updates the timestamps on the inode and
1153 * may also increase the size of the inode. This function
1154 * must not be called with @size any smaller than the current
1155 * inode size.
1156 *
1157 * Although it is not strictly required to unstuff files here,
1158 * earlier versions of GFS2 have a bug in the stuffed file reading
1159 * code which will result in a buffer overrun if the size is larger
1160 * than the max stuffed file size. In order to prevent this from
1161 * occurring, such files are unstuffed, but in other cases we can
1162 * just update the inode size directly.
1163 *
1164 * Returns: 0 on success, or -ve on error
1165 */
1166
1167static int do_grow(struct inode *inode, u64 size)
1168{
1169        struct gfs2_inode *ip = GFS2_I(inode);
1170        struct gfs2_sbd *sdp = GFS2_SB(inode);
1171        struct buffer_head *dibh;
1172        int error;
1173        int unstuff = 0;
1174
1175        if (gfs2_is_stuffed(ip) &&
1176            (size > (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)))) {
1177                error = gfs2_quota_lock_check(ip);
1178                if (error)
1179                        return error;
1180
1181                error = gfs2_inplace_reserve(ip, 1);
1182                if (error)
1183                        goto do_grow_qunlock;
1184                unstuff = 1;
1185        }
1186
1187        error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT, 0);
1188        if (error)
1189                goto do_grow_release;
1190
1191        if (unstuff) {
1192                error = gfs2_unstuff_dinode(ip, NULL);
1193                if (error)
1194                        goto do_end_trans;
1195        }
1196
1197        error = gfs2_meta_inode_buffer(ip, &dibh);
1198        if (error)
1199                goto do_end_trans;
1200
1201        i_size_write(inode, size);
1202        ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1203        gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1204        gfs2_dinode_out(ip, dibh->b_data);
1205        brelse(dibh);
1206
1207do_end_trans:
1208        gfs2_trans_end(sdp);
1209do_grow_release:
1210        if (unstuff) {
1211                gfs2_inplace_release(ip);
1212do_grow_qunlock:
1213                gfs2_quota_unlock(ip);
1214        }
1215        return error;
1216}
1217
1218/**
1219 * gfs2_setattr_size - make a file a given size
1220 * @inode: the inode
1221 * @newsize: the size to make the file
1222 *
1223 * The file size can grow, shrink, or stay the same size. This
1224 * is called holding i_mutex and an exclusive glock on the inode
1225 * in question.
1226 *
1227 * Returns: errno
1228 */
1229
1230int gfs2_setattr_size(struct inode *inode, u64 newsize)
1231{
1232        int ret;
1233        u64 oldsize;
1234
1235        BUG_ON(!S_ISREG(inode->i_mode));
1236
1237        ret = inode_newsize_ok(inode, newsize);
1238        if (ret)
1239                return ret;
1240
1241        inode_dio_wait(inode);
1242
1243        oldsize = inode->i_size;
1244        if (newsize >= oldsize)
1245                return do_grow(inode, newsize);
1246
1247        return do_shrink(inode, oldsize, newsize);
1248}
1249
1250int gfs2_truncatei_resume(struct gfs2_inode *ip)
1251{
1252        int error;
1253        error = trunc_dealloc(ip, i_size_read(&ip->i_inode));
1254        if (!error)
1255                error = trunc_end(ip);
1256        return error;
1257}
1258
1259int gfs2_file_dealloc(struct gfs2_inode *ip)
1260{
1261        return trunc_dealloc(ip, 0);
1262}
1263
1264/**
1265 * gfs2_write_alloc_required - figure out if a write will require an allocation
1266 * @ip: the file being written to
1267 * @offset: the offset to write to
1268 * @len: the number of bytes being written
1269 *
1270 * Returns: 1 if an alloc is required, 0 otherwise
1271 */
1272
1273int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset,
1274                              unsigned int len)
1275{
1276        struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1277        struct buffer_head bh;
1278        unsigned int shift;
1279        u64 lblock, lblock_stop, size;
1280        u64 end_of_file;
1281
1282        if (!len)
1283                return 0;
1284
1285        if (gfs2_is_stuffed(ip)) {
1286                if (offset + len >
1287                    sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
1288                        return 1;
1289                return 0;
1290        }
1291
1292        shift = sdp->sd_sb.sb_bsize_shift;
1293        BUG_ON(gfs2_is_dir(ip));
1294        end_of_file = (i_size_read(&ip->i_inode) + sdp->sd_sb.sb_bsize - 1) >> shift;
1295        lblock = offset >> shift;
1296        lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift;
1297        if (lblock_stop > end_of_file)
1298                return 1;
1299
1300        size = (lblock_stop - lblock) << shift;
1301        do {
1302                bh.b_state = 0;
1303                bh.b_size = size;
1304                gfs2_block_map(&ip->i_inode, lblock, &bh, 0);
1305                if (!buffer_mapped(&bh))
1306                        return 1;
1307                size -= bh.b_size;
1308                lblock += (bh.b_size >> ip->i_inode.i_blkbits);
1309        } while(size > 0);
1310
1311        return 0;
1312}
1313
1314
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.