fs/gfs2/dir.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 /*
  11  * Implements Extendible Hashing as described in:
  12  *   "Extendible Hashing" by Fagin, et al in
  13  *     __ACM Trans. on Database Systems__, Sept 1979.
  14  *
  15  *
  16  * Here's the layout of dirents which is essentially the same as that of ext2
  17  * within a single block. The field de_name_len is the number of bytes
  18  * actually required for the name (no null terminator). The field de_rec_len
  19  * is the number of bytes allocated to the dirent. The offset of the next
  20  * dirent in the block is (dirent + dirent->de_rec_len). When a dirent is
  21  * deleted, the preceding dirent inherits its allocated space, ie
  22  * prev->de_rec_len += deleted->de_rec_len. Since the next dirent is obtained
  23  * by adding de_rec_len to the current dirent, this essentially causes the
  24  * deleted dirent to get jumped over when iterating through all the dirents.
  25  *
  26  * When deleting the first dirent in a block, there is no previous dirent so
  27  * the field de_ino is set to zero to designate it as deleted. When allocating
  28  * a dirent, gfs2_dirent_alloc iterates through the dirents in a block. If the
  29  * first dirent has (de_ino == 0) and de_rec_len is large enough, this first
  30  * dirent is allocated. Otherwise it must go through all the 'used' dirents
  31  * searching for one in which the amount of total space minus the amount of
  32  * used space will provide enough space for the new dirent.
  33  *
  34  * There are two types of blocks in which dirents reside. In a stuffed dinode,
  35  * the dirents begin at offset sizeof(struct gfs2_dinode) from the beginning of
  36  * the block.  In leaves, they begin at offset sizeof(struct gfs2_leaf) from the
  37  * beginning of the leaf block. The dirents reside in leaves when
  38  *
  39  * dip->i_diskflags & GFS2_DIF_EXHASH is true
  40  *
  41  * Otherwise, the dirents are "linear", within a single stuffed dinode block.
  42  *
  43  * When the dirents are in leaves, the actual contents of the directory file are
  44  * used as an array of 64-bit block pointers pointing to the leaf blocks. The
  45  * dirents are NOT in the directory file itself. There can be more than one
  46  * block pointer in the array that points to the same leaf. In fact, when a
  47  * directory is first converted from linear to exhash, all of the pointers
  48  * point to the same leaf.
  49  *
  50  * When a leaf is completely full, the size of the hash table can be
  51  * doubled unless it is already at the maximum size which is hard coded into
  52  * GFS2_DIR_MAX_DEPTH. After that, leaves are chained together in a linked list,
  53  * but never before the maximum hash table size has been reached.
  54  */
  55
  56 #include <linux/slab.h>
  57 #include <linux/spinlock.h>
  58 #include <linux/buffer_head.h>
  59 #include <linux/sort.h>
  60 #include <linux/gfs2_ondisk.h>
  61 #include <linux/crc32.h>
  62 #include <linux/vmalloc.h>
  63
  64 #include "gfs2.h"
  65 #include "incore.h"
  66 #include "dir.h"
  67 #include "glock.h"
  68 #include "inode.h"
  69 #include "meta_io.h"
  70 #include "quota.h"
  71 #include "rgrp.h"
  72 #include "trans.h"
  73 #include "bmap.h"
  74 #include "util.h"
  75
  76 #define IS_LEAF     1 /* Hashed (leaf) directory */
  77 #define IS_DINODE   2 /* Linear (stuffed dinode block) directory */
  78
  79 #define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1)
  80 #define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1))
  81
  82 struct qstr gfs2_qdot __read_mostly;
  83 struct qstr gfs2_qdotdot __read_mostly;
  84
  85 typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent,
  86                             const struct qstr *name, void *opaque);
  87
  88 int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block,
  89                             struct buffer_head **bhp)
  90 {
  91         struct buffer_head *bh;
  92
  93         bh = gfs2_meta_new(ip->i_gl, block);
  94         gfs2_trans_add_bh(ip->i_gl, bh, 1);
  95         gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD);
  96         gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
  97         *bhp = bh;
  98         return 0;
  99 }
 100
 101 static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block,
 102                                         struct buffer_head **bhp)
 103 {
 104         struct buffer_head *bh;
 105         int error;
 106
 107         error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, &bh);
 108         if (error)
 109                 return error;
 110         if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) {
 111                 brelse(bh);
 112                 return -EIO;
 113         }
 114         *bhp = bh;
 115         return 0;
 116 }
 117
 118 static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf,
 119                                   unsigned int offset, unsigned int size)
 120 {
 121         struct buffer_head *dibh;
 122         int error;
 123
 124         error = gfs2_meta_inode_buffer(ip, &dibh);
 125         if (error)
 126                 return error;
 127
 128         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 129         memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size);
 130         if (ip->i_inode.i_size < offset + size)
 131                 i_size_write(&ip->i_inode, offset + size);
 132         ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
 133         gfs2_dinode_out(ip, dibh->b_data);
 134
 135         brelse(dibh);
 136
 137         return size;
 138 }
 139
 140
 141
 142 /**
 143  * gfs2_dir_write_data - Write directory information to the inode
 144  * @ip: The GFS2 inode
 145  * @buf: The buffer containing information to be written
 146  * @offset: The file offset to start writing at
 147  * @size: The amount of data to write
 148  *
 149  * Returns: The number of bytes correctly written or error code
 150  */
 151 static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf,
 152                                u64 offset, unsigned int size)
 153 {
 154         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 155         struct buffer_head *dibh;
 156         u64 lblock, dblock;
 157         u32 extlen = 0;
 158         unsigned int o;
 159         int copied = 0;
 160         int error = 0;
 161         int new = 0;
 162
 163         if (!size)
 164                 return 0;
 165
 166         if (gfs2_is_stuffed(ip) &&
 167             offset + size <= sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
 168                 return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset,
 169                                               size);
 170
 171         if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
 172                 return -EINVAL;
 173
 174         if (gfs2_is_stuffed(ip)) {
 175                 error = gfs2_unstuff_dinode(ip, NULL);
 176                 if (error)
 177                         return error;
 178         }
 179
 180         lblock = offset;
 181         o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
 182
 183         while (copied < size) {
 184                 unsigned int amount;
 185                 struct buffer_head *bh;
 186
 187                 amount = size - copied;
 188                 if (amount > sdp->sd_sb.sb_bsize - o)
 189                         amount = sdp->sd_sb.sb_bsize - o;
 190
 191                 if (!extlen) {
 192                         new = 1;
 193                         error = gfs2_extent_map(&ip->i_inode, lblock, &new,
 194                                                 &dblock, &extlen);
 195                         if (error)
 196                                 goto fail;
 197                         error = -EIO;
 198                         if (gfs2_assert_withdraw(sdp, dblock))
 199                                 goto fail;
 200                 }
 201
 202                 if (amount == sdp->sd_jbsize || new)
 203                         error = gfs2_dir_get_new_buffer(ip, dblock, &bh);
 204                 else
 205                         error = gfs2_dir_get_existing_buffer(ip, dblock, &bh);
 206
 207                 if (error)
 208                         goto fail;
 209
 210                 gfs2_trans_add_bh(ip->i_gl, bh, 1);
 211                 memcpy(bh->b_data + o, buf, amount);
 212                 brelse(bh);
 213
 214                 buf += amount;
 215                 copied += amount;
 216                 lblock++;
 217                 dblock++;
 218                 extlen--;
 219
 220                 o = sizeof(struct gfs2_meta_header);
 221         }
 222
 223 out:
 224         error = gfs2_meta_inode_buffer(ip, &dibh);
 225         if (error)
 226                 return error;
 227
 228         if (ip->i_inode.i_size < offset + copied)
 229                 i_size_write(&ip->i_inode, offset + copied);
 230         ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
 231
 232         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 233         gfs2_dinode_out(ip, dibh->b_data);
 234         brelse(dibh);
 235
 236         return copied;
 237 fail:
 238         if (copied)
 239                 goto out;
 240         return error;
 241 }
 242
 243 static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, char *buf,
 244                                  u64 offset, unsigned int size)
 245 {
 246         struct buffer_head *dibh;
 247         int error;
 248
 249         error = gfs2_meta_inode_buffer(ip, &dibh);
 250         if (!error) {
 251                 offset += sizeof(struct gfs2_dinode);
 252                 memcpy(buf, dibh->b_data + offset, size);
 253                 brelse(dibh);
 254         }
 255
 256         return (error) ? error : size;
 257 }
 258
 259
 260 /**
 261  * gfs2_dir_read_data - Read a data from a directory inode
 262  * @ip: The GFS2 Inode
 263  * @buf: The buffer to place result into
 264  * @offset: File offset to begin jdata_readng from
 265  * @size: Amount of data to transfer
 266  *
 267  * Returns: The amount of data actually copied or the error
 268  */
 269 static int gfs2_dir_read_data(struct gfs2_inode *ip, char *buf, u64 offset,
 270                               unsigned int size, unsigned ra)
 271 {
 272         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 273         u64 lblock, dblock;
 274         u32 extlen = 0;
 275         unsigned int o;
 276         int copied = 0;
 277         int error = 0;
 278         u64 disksize = i_size_read(&ip->i_inode);
 279
 280         if (offset >= disksize)
 281                 return 0;
 282
 283         if (offset + size > disksize)
 284                 size = disksize - offset;
 285
 286         if (!size)
 287                 return 0;
 288
 289         if (gfs2_is_stuffed(ip))
 290                 return gfs2_dir_read_stuffed(ip, buf, offset, size);
 291
 292         if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
 293                 return -EINVAL;
 294
 295         lblock = offset;
 296         o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
 297
 298         while (copied < size) {
 299                 unsigned int amount;
 300                 struct buffer_head *bh;
 301                 int new;
 302
 303                 amount = size - copied;
 304                 if (amount > sdp->sd_sb.sb_bsize - o)
 305                         amount = sdp->sd_sb.sb_bsize - o;
 306
 307                 if (!extlen) {
 308                         new = 0;
 309                         error = gfs2_extent_map(&ip->i_inode, lblock, &new,
 310                                                 &dblock, &extlen);
 311                         if (error || !dblock)
 312                                 goto fail;
 313                         BUG_ON(extlen < 1);
 314                         if (!ra)
 315                                 extlen = 1;
 316                         bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
 317                 } else {
 318                         error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, &bh);
 319                         if (error)
 320                                 goto fail;
 321                 }
 322                 error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD);
 323                 if (error) {
 324                         brelse(bh);
 325                         goto fail;
 326                 }
 327                 dblock++;
 328                 extlen--;
 329                 memcpy(buf, bh->b_data + o, amount);
 330                 brelse(bh);
 331                 buf += amount;
 332                 copied += amount;
 333                 lblock++;
 334                 o = sizeof(struct gfs2_meta_header);
 335         }
 336
 337         return copied;
 338 fail:
 339         return (copied) ? copied : error;
 340 }
 341
 342 /**
 343  * gfs2_dir_get_hash_table - Get pointer to the dir hash table
 344  * @ip: The inode in question
 345  *
 346  * Returns: The hash table or an error
 347  */
 348
 349 static __be64 *gfs2_dir_get_hash_table(struct gfs2_inode *ip)
 350 {
 351         struct inode *inode = &ip->i_inode;
 352         int ret;
 353         u32 hsize;
 354         __be64 *hc;
 355
 356         BUG_ON(!(ip->i_diskflags & GFS2_DIF_EXHASH));
 357
 358         hc = ip->i_hash_cache;
 359         if (hc)
 360                 return hc;
 361
 362         hsize = 1 << ip->i_depth;
 363         hsize *= sizeof(__be64);
 364         if (hsize != i_size_read(&ip->i_inode)) {
 365                 gfs2_consist_inode(ip);
 366                 return ERR_PTR(-EIO);
 367         }
 368
 369         hc = kmalloc(hsize, GFP_NOFS);
 370         ret = -ENOMEM;
 371         if (hc == NULL)
 372                 return ERR_PTR(-ENOMEM);
 373
 374         ret = gfs2_dir_read_data(ip, (char *)hc, 0, hsize, 1);
 375         if (ret < 0) {
 376                 kfree(hc);
 377                 return ERR_PTR(ret);
 378         }
 379
 380         spin_lock(&inode->i_lock);
 381         if (ip->i_hash_cache)
 382                 kfree(hc);
 383         else
 384                 ip->i_hash_cache = hc;
 385         spin_unlock(&inode->i_lock);
 386
 387         return ip->i_hash_cache;
 388 }
 389
 390 /**
 391  * gfs2_dir_hash_inval - Invalidate dir hash
 392  * @ip: The directory inode
 393  *
 394  * Must be called with an exclusive glock, or during glock invalidation.
 395  */
 396 void gfs2_dir_hash_inval(struct gfs2_inode *ip)
 397 {
 398         __be64 *hc = ip->i_hash_cache;
 399         ip->i_hash_cache = NULL;
 400         kfree(hc);
 401 }
 402
 403 static inline int gfs2_dirent_sentinel(const struct gfs2_dirent *dent)
 404 {
 405         return dent->de_inum.no_addr == 0 || dent->de_inum.no_formal_ino == 0;
 406 }
 407
 408 static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
 409                                      const struct qstr *name, int ret)
 410 {
 411         if (!gfs2_dirent_sentinel(dent) &&
 412             be32_to_cpu(dent->de_hash) == name->hash &&
 413             be16_to_cpu(dent->de_name_len) == name->len &&
 414             memcmp(dent+1, name->name, name->len) == 0)
 415                 return ret;
 416         return 0;
 417 }
 418
 419 static int gfs2_dirent_find(const struct gfs2_dirent *dent,
 420                             const struct qstr *name,
 421                             void *opaque)
 422 {
 423         return __gfs2_dirent_find(dent, name, 1);
 424 }
 425
 426 static int gfs2_dirent_prev(const struct gfs2_dirent *dent,
 427                             const struct qstr *name,
 428                             void *opaque)
 429 {
 430         return __gfs2_dirent_find(dent, name, 2);
 431 }
 432
 433 /*
 434  * name->name holds ptr to start of block.
 435  * name->len holds size of block.
 436  */
 437 static int gfs2_dirent_last(const struct gfs2_dirent *dent,
 438                             const struct qstr *name,
 439                             void *opaque)
 440 {
 441         const char *start = name->name;
 442         const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len);
 443         if (name->len == (end - start))
 444                 return 1;
 445         return 0;
 446 }
 447
 448 static int gfs2_dirent_find_space(const struct gfs2_dirent *dent,
 449                                   const struct qstr *name,
 450                                   void *opaque)
 451 {
 452         unsigned required = GFS2_DIRENT_SIZE(name->len);
 453         unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
 454         unsigned totlen = be16_to_cpu(dent->de_rec_len);
 455
 456         if (gfs2_dirent_sentinel(dent))
 457                 actual = 0;
 458         if (totlen - actual >= required)
 459                 return 1;
 460         return 0;
 461 }
 462
 463 struct dirent_gather {
 464         const struct gfs2_dirent **pdent;
 465         unsigned offset;
 466 };
 467
 468 static int gfs2_dirent_gather(const struct gfs2_dirent *dent,
 469                               const struct qstr *name,
 470                               void *opaque)
 471 {
 472         struct dirent_gather *g = opaque;
 473         if (!gfs2_dirent_sentinel(dent)) {
 474                 g->pdent[g->offset++] = dent;
 475         }
 476         return 0;
 477 }
 478
 479 /*
 480  * Other possible things to check:
 481  * - Inode located within filesystem size (and on valid block)
 482  * - Valid directory entry type
 483  * Not sure how heavy-weight we want to make this... could also check
 484  * hash is correct for example, but that would take a lot of extra time.
 485  * For now the most important thing is to check that the various sizes
 486  * are correct.
 487  */
 488 static int gfs2_check_dirent(struct gfs2_dirent *dent, unsigned int offset,
 489                              unsigned int size, unsigned int len, int first)
 490 {
 491         const char *msg = "gfs2_dirent too small";
 492         if (unlikely(size < sizeof(struct gfs2_dirent)))
 493                 goto error;
 494         msg = "gfs2_dirent misaligned";
 495         if (unlikely(offset & 0x7))
 496                 goto error;
 497         msg = "gfs2_dirent points beyond end of block";
 498         if (unlikely(offset + size > len))
 499                 goto error;
 500         msg = "zero inode number";
 501         if (unlikely(!first && gfs2_dirent_sentinel(dent)))
 502                 goto error;
 503         msg = "name length is greater than space in dirent";
 504         if (!gfs2_dirent_sentinel(dent) &&
 505             unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) >
 506                      size))
 507                 goto error;
 508         return 0;
 509 error:
 510         printk(KERN_WARNING "gfs2_check_dirent: %s (%s)\n", msg,
 511                first ? "first in block" : "not first in block");
 512         return -EIO;
 513 }
 514
 515 static int gfs2_dirent_offset(const void *buf)
 516 {
 517         const struct gfs2_meta_header *h = buf;
 518         int offset;
 519
 520         BUG_ON(buf == NULL);
 521
 522         switch(be32_to_cpu(h->mh_type)) {
 523         case GFS2_METATYPE_LF:
 524                 offset = sizeof(struct gfs2_leaf);
 525                 break;
 526         case GFS2_METATYPE_DI:
 527                 offset = sizeof(struct gfs2_dinode);
 528                 break;
 529         default:
 530                 goto wrong_type;
 531         }
 532         return offset;
 533 wrong_type:
 534         printk(KERN_WARNING "gfs2_scan_dirent: wrong block type %u\n",
 535                be32_to_cpu(h->mh_type));
 536         return -1;
 537 }
 538
 539 static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf,
 540                                             unsigned int len, gfs2_dscan_t scan,
 541                                             const struct qstr *name,
 542                                             void *opaque)
 543 {
 544         struct gfs2_dirent *dent, *prev;
 545         unsigned offset;
 546         unsigned size;
 547         int ret = 0;
 548
 549         ret = gfs2_dirent_offset(buf);
 550         if (ret < 0)
 551                 goto consist_inode;
 552
 553         offset = ret;
 554         prev = NULL;
 555         dent = buf + offset;
 556         size = be16_to_cpu(dent->de_rec_len);
 557         if (gfs2_check_dirent(dent, offset, size, len, 1))
 558                 goto consist_inode;
 559         do {
 560                 ret = scan(dent, name, opaque);
 561                 if (ret)
 562                         break;
 563                 offset += size;
 564                 if (offset == len)
 565                         break;
 566                 prev = dent;
 567                 dent = buf + offset;
 568                 size = be16_to_cpu(dent->de_rec_len);
 569                 if (gfs2_check_dirent(dent, offset, size, len, 0))
 570                         goto consist_inode;
 571         } while(1);
 572
 573         switch(ret) {
 574         case 0:
 575                 return NULL;
 576         case 1:
 577                 return dent;
 578         case 2:
 579                 return prev ? prev : dent;
 580         default:
 581                 BUG_ON(ret > 0);
 582                 return ERR_PTR(ret);
 583         }
 584
 585 consist_inode:
 586         gfs2_consist_inode(GFS2_I(inode));
 587         return ERR_PTR(-EIO);
 588 }
 589
 590 static int dirent_check_reclen(struct gfs2_inode *dip,
 591                                const struct gfs2_dirent *d, const void *end_p)
 592 {
 593         const void *ptr = d;
 594         u16 rec_len = be16_to_cpu(d->de_rec_len);
 595
 596         if (unlikely(rec_len < sizeof(struct gfs2_dirent)))
 597                 goto broken;
 598         ptr += rec_len;
 599         if (ptr < end_p)
 600                 return rec_len;
 601         if (ptr == end_p)
 602                 return -ENOENT;
 603 broken:
 604         gfs2_consist_inode(dip);
 605         return -EIO;
 606 }
 607
 608 /**
 609  * dirent_next - Next dirent
 610  * @dip: the directory
 611  * @bh: The buffer
 612  * @dent: Pointer to list of dirents
 613  *
 614  * Returns: 0 on success, error code otherwise
 615  */
 616
 617 static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh,
 618                        struct gfs2_dirent **dent)
 619 {
 620         struct gfs2_dirent *cur = *dent, *tmp;
 621         char *bh_end = bh->b_data + bh->b_size;
 622         int ret;
 623
 624         ret = dirent_check_reclen(dip, cur, bh_end);
 625         if (ret < 0)
 626                 return ret;
 627
 628         tmp = (void *)cur + ret;
 629         ret = dirent_check_reclen(dip, tmp, bh_end);
 630         if (ret == -EIO)
 631                 return ret;
 632
 633         /* Only the first dent could ever have de_inum.no_addr == 0 */
 634         if (gfs2_dirent_sentinel(tmp)) {
 635                 gfs2_consist_inode(dip);
 636                 return -EIO;
 637         }
 638
 639         *dent = tmp;
 640         return 0;
 641 }
 642
 643 /**
 644  * dirent_del - Delete a dirent
 645  * @dip: The GFS2 inode
 646  * @bh: The buffer
 647  * @prev: The previous dirent
 648  * @cur: The current dirent
 649  *
 650  */
 651
 652 static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh,
 653                        struct gfs2_dirent *prev, struct gfs2_dirent *cur)
 654 {
 655         u16 cur_rec_len, prev_rec_len;
 656
 657         if (gfs2_dirent_sentinel(cur)) {
 658                 gfs2_consist_inode(dip);
 659                 return;
 660         }
 661
 662         gfs2_trans_add_bh(dip->i_gl, bh, 1);
 663
 664         /* If there is no prev entry, this is the first entry in the block.
 665            The de_rec_len is already as big as it needs to be.  Just zero
 666            out the inode number and return.  */
 667
 668         if (!prev) {
 669                 cur->de_inum.no_addr = 0;
 670                 cur->de_inum.no_formal_ino = 0;
 671                 return;
 672         }
 673
 674         /*  Combine this dentry with the previous one.  */
 675
 676         prev_rec_len = be16_to_cpu(prev->de_rec_len);
 677         cur_rec_len = be16_to_cpu(cur->de_rec_len);
 678
 679         if ((char *)prev + prev_rec_len != (char *)cur)
 680                 gfs2_consist_inode(dip);
 681         if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size)
 682                 gfs2_consist_inode(dip);
 683
 684         prev_rec_len += cur_rec_len;
 685         prev->de_rec_len = cpu_to_be16(prev_rec_len);
 686 }
 687
 688 /*
 689  * Takes a dent from which to grab space as an argument. Returns the
 690  * newly created dent.
 691  */
 692 static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode,
 693                                             struct gfs2_dirent *dent,
 694                                             const struct qstr *name,
 695                                             struct buffer_head *bh)
 696 {
 697         struct gfs2_inode *ip = GFS2_I(inode);
 698         struct gfs2_dirent *ndent;
 699         unsigned offset = 0, totlen;
 700
 701         if (!gfs2_dirent_sentinel(dent))
 702                 offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
 703         totlen = be16_to_cpu(dent->de_rec_len);
 704         BUG_ON(offset + name->len > totlen);
 705         gfs2_trans_add_bh(ip->i_gl, bh, 1);
 706         ndent = (struct gfs2_dirent *)((char *)dent + offset);
 707         dent->de_rec_len = cpu_to_be16(offset);
 708         gfs2_qstr2dirent(name, totlen - offset, ndent);
 709         return ndent;
 710 }
 711
 712 static struct gfs2_dirent *gfs2_dirent_alloc(struct inode *inode,
 713                                              struct buffer_head *bh,
 714                                              const struct qstr *name)
 715 {
 716         struct gfs2_dirent *dent;
 717         dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
 718                                 gfs2_dirent_find_space, name, NULL);
 719         if (!dent || IS_ERR(dent))
 720                 return dent;
 721         return gfs2_init_dirent(inode, dent, name, bh);
 722 }
 723
 724 static int get_leaf(struct gfs2_inode *dip, u64 leaf_no,
 725                     struct buffer_head **bhp)
 726 {
 727         int error;
 728
 729         error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, bhp);
 730         if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) {
 731                 /* printk(KERN_INFO "block num=%llu\n", leaf_no); */
 732                 error = -EIO;
 733         }
 734
 735         return error;
 736 }
 737
 738 /**
 739  * get_leaf_nr - Get a leaf number associated with the index
 740  * @dip: The GFS2 inode
 741  * @index:
 742  * @leaf_out:
 743  *
 744  * Returns: 0 on success, error code otherwise
 745  */
 746
 747 static int get_leaf_nr(struct gfs2_inode *dip, u32 index,
 748                        u64 *leaf_out)
 749 {
 750         __be64 *hash;
 751
 752         hash = gfs2_dir_get_hash_table(dip);
 753         if (IS_ERR(hash))
 754                 return PTR_ERR(hash);
 755         *leaf_out = be64_to_cpu(*(hash + index));
 756         return 0;
 757 }
 758
 759 static int get_first_leaf(struct gfs2_inode *dip, u32 index,
 760                           struct buffer_head **bh_out)
 761 {
 762         u64 leaf_no;
 763         int error;
 764
 765         error = get_leaf_nr(dip, index, &leaf_no);
 766         if (!error)
 767                 error = get_leaf(dip, leaf_no, bh_out);
 768
 769         return error;
 770 }
 771
 772 static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode,
 773                                               const struct qstr *name,
 774                                               gfs2_dscan_t scan,
 775                                               struct buffer_head **pbh)
 776 {
 777         struct buffer_head *bh;
 778         struct gfs2_dirent *dent;
 779         struct gfs2_inode *ip = GFS2_I(inode);
 780         int error;
 781
 782         if (ip->i_diskflags & GFS2_DIF_EXHASH) {
 783                 struct gfs2_leaf *leaf;
 784                 unsigned hsize = 1 << ip->i_depth;
 785                 unsigned index;
 786                 u64 ln;
 787                 if (hsize * sizeof(u64) != i_size_read(inode)) {
 788                         gfs2_consist_inode(ip);
 789                         return ERR_PTR(-EIO);
 790                 }
 791
 792                 index = name->hash >> (32 - ip->i_depth);
 793                 error = get_first_leaf(ip, index, &bh);
 794                 if (error)
 795                         return ERR_PTR(error);
 796                 do {
 797                         dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
 798                                                 scan, name, NULL);
 799                         if (dent)
 800                                 goto got_dent;
 801                         leaf = (struct gfs2_leaf *)bh->b_data;
 802                         ln = be64_to_cpu(leaf->lf_next);
 803                         brelse(bh);
 804                         if (!ln)
 805                                 break;
 806
 807                         error = get_leaf(ip, ln, &bh);
 808                 } while(!error);
 809
 810                 return error ? ERR_PTR(error) : NULL;
 811         }
 812
 813
 814         error = gfs2_meta_inode_buffer(ip, &bh);
 815         if (error)
 816                 return ERR_PTR(error);
 817         dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL);
 818 got_dent:
 819         if (unlikely(dent == NULL || IS_ERR(dent))) {
 820                 brelse(bh);
 821                 bh = NULL;
 822         }
 823         *pbh = bh;
 824         return dent;
 825 }
 826
 827 static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth)
 828 {
 829         struct gfs2_inode *ip = GFS2_I(inode);
 830         unsigned int n = 1;
 831         u64 bn;
 832         int error;
 833         struct buffer_head *bh;
 834         struct gfs2_leaf *leaf;
 835         struct gfs2_dirent *dent;
 836         struct qstr name = { .name = "", .len = 0, .hash = 0 };
 837
 838         error = gfs2_alloc_block(ip, &bn, &n);
 839         if (error)
 840                 return NULL;
 841         bh = gfs2_meta_new(ip->i_gl, bn);
 842         if (!bh)
 843                 return NULL;
 844
 845         gfs2_trans_add_unrevoke(GFS2_SB(inode), bn, 1);
 846         gfs2_trans_add_bh(ip->i_gl, bh, 1);
 847         gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF);
 848         leaf = (struct gfs2_leaf *)bh->b_data;
 849         leaf->lf_depth = cpu_to_be16(depth);
 850         leaf->lf_entries = 0;
 851         leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE);
 852         leaf->lf_next = 0;
 853         memset(leaf->lf_reserved, 0, sizeof(leaf->lf_reserved));
 854         dent = (struct gfs2_dirent *)(leaf+1);
 855         gfs2_qstr2dirent(&name, bh->b_size - sizeof(struct gfs2_leaf), dent);
 856         *pbh = bh;
 857         return leaf;
 858 }
 859
 860 /**
 861  * dir_make_exhash - Convert a stuffed directory into an ExHash directory
 862  * @dip: The GFS2 inode
 863  *
 864  * Returns: 0 on success, error code otherwise
 865  */
 866
 867 static int dir_make_exhash(struct inode *inode)
 868 {
 869         struct gfs2_inode *dip = GFS2_I(inode);
 870         struct gfs2_sbd *sdp = GFS2_SB(inode);
 871         struct gfs2_dirent *dent;
 872         struct qstr args;
 873         struct buffer_head *bh, *dibh;
 874         struct gfs2_leaf *leaf;
 875         int y;
 876         u32 x;
 877         __be64 *lp;
 878         u64 bn;
 879         int error;
 880
 881         error = gfs2_meta_inode_buffer(dip, &dibh);
 882         if (error)
 883                 return error;
 884
 885         /*  Turn over a new leaf  */
 886
 887         leaf = new_leaf(inode, &bh, 0);
 888         if (!leaf)
 889                 return -ENOSPC;
 890         bn = bh->b_blocknr;
 891
 892         gfs2_assert(sdp, dip->i_entries < (1 << 16));
 893         leaf->lf_entries = cpu_to_be16(dip->i_entries);
 894
 895         /*  Copy dirents  */
 896
 897         gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh,
 898                              sizeof(struct gfs2_dinode));
 899
 900         /*  Find last entry  */
 901
 902         x = 0;
 903         args.len = bh->b_size - sizeof(struct gfs2_dinode) +
 904                    sizeof(struct gfs2_leaf);
 905         args.name = bh->b_data;
 906         dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size,
 907                                 gfs2_dirent_last, &args, NULL);
 908         if (!dent) {
 909                 brelse(bh);
 910                 brelse(dibh);
 911                 return -EIO;
 912         }
 913         if (IS_ERR(dent)) {
 914                 brelse(bh);
 915                 brelse(dibh);
 916                 return PTR_ERR(dent);
 917         }
 918
 919         /*  Adjust the last dirent's record length
 920            (Remember that dent still points to the last entry.)  */
 921
 922         dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) +
 923                 sizeof(struct gfs2_dinode) -
 924                 sizeof(struct gfs2_leaf));
 925
 926         brelse(bh);
 927
 928         /*  We're done with the new leaf block, now setup the new
 929             hash table.  */
 930
 931         gfs2_trans_add_bh(dip->i_gl, dibh, 1);
 932         gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
 933
 934         lp = (__be64 *)(dibh->b_data + sizeof(struct gfs2_dinode));
 935
 936         for (x = sdp->sd_hash_ptrs; x--; lp++)
 937                 *lp = cpu_to_be64(bn);
 938
 939         i_size_write(inode, sdp->sd_sb.sb_bsize / 2);
 940         gfs2_add_inode_blocks(&dip->i_inode, 1);
 941         dip->i_diskflags |= GFS2_DIF_EXHASH;
 942
 943         for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ;
 944         dip->i_depth = y;
 945
 946         gfs2_dinode_out(dip, dibh->b_data);
 947
 948         brelse(dibh);
 949
 950         return 0;
 951 }
 952
 953 /**
 954  * dir_split_leaf - Split a leaf block into two
 955  * @dip: The GFS2 inode
 956  * @index:
 957  * @leaf_no:
 958  *
 959  * Returns: 0 on success, error code on failure
 960  */
 961
 962 static int dir_split_leaf(struct inode *inode, const struct qstr *name)
 963 {
 964         struct gfs2_inode *dip = GFS2_I(inode);
 965         struct buffer_head *nbh, *obh, *dibh;
 966         struct gfs2_leaf *nleaf, *oleaf;
 967         struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new;
 968         u32 start, len, half_len, divider;
 969         u64 bn, leaf_no;
 970         __be64 *lp;
 971         u32 index;
 972         int x, moved = 0;
 973         int error;
 974
 975         index = name->hash >> (32 - dip->i_depth);
 976         error = get_leaf_nr(dip, index, &leaf_no);
 977         if (error)
 978                 return error;
 979
 980         /*  Get the old leaf block  */
 981         error = get_leaf(dip, leaf_no, &obh);
 982         if (error)
 983                 return error;
 984
 985         oleaf = (struct gfs2_leaf *)obh->b_data;
 986         if (dip->i_depth == be16_to_cpu(oleaf->lf_depth)) {
 987                 brelse(obh);
 988                 return 1; /* can't split */
 989         }
 990
 991         gfs2_trans_add_bh(dip->i_gl, obh, 1);
 992
 993         nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1);
 994         if (!nleaf) {
 995                 brelse(obh);
 996                 return -ENOSPC;
 997         }
 998         bn = nbh->b_blocknr;
 999
1000         /*  Compute the start and len of leaf pointers in the hash table.  */
1001         len = 1 << (dip->i_depth - be16_to_cpu(oleaf->lf_depth));
1002         half_len = len >> 1;
1003         if (!half_len) {
1004                 printk(KERN_WARNING "i_depth %u lf_depth %u index %u\n", dip->i_depth, be16_to_cpu(oleaf->lf_depth), index);
1005                 gfs2_consist_inode(dip);
1006                 error = -EIO;
1007                 goto fail_brelse;
1008         }
1009
1010         start = (index & ~(len - 1));
1011
1012         /* Change the pointers.
1013            Don't bother distinguishing stuffed from non-stuffed.
1014            This code is complicated enough already. */
1015         lp = kmalloc(half_len * sizeof(__be64), GFP_NOFS);
1016         if (!lp) {
1017                 error = -ENOMEM;
1018                 goto fail_brelse;
1019         }
1020
1021         /*  Change the pointers  */
1022         for (x = 0; x < half_len; x++)
1023                 lp[x] = cpu_to_be64(bn);
1024
1025         gfs2_dir_hash_inval(dip);
1026
1027         error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64),
1028                                     half_len * sizeof(u64));
1029         if (error != half_len * sizeof(u64)) {
1030                 if (error >= 0)
1031                         error = -EIO;
1032                 goto fail_lpfree;
1033         }
1034
1035         kfree(lp);
1036
1037         /*  Compute the divider  */
1038         divider = (start + half_len) << (32 - dip->i_depth);
1039
1040         /*  Copy the entries  */
1041         dent = (struct gfs2_dirent *)(obh->b_data + sizeof(struct gfs2_leaf));
1042
1043         do {
1044                 next = dent;
1045                 if (dirent_next(dip, obh, &next))
1046                         next = NULL;
1047
1048                 if (!gfs2_dirent_sentinel(dent) &&
1049                     be32_to_cpu(dent->de_hash) < divider) {
1050                         struct qstr str;
1051                         str.name = (char*)(dent+1);
1052                         str.len = be16_to_cpu(dent->de_name_len);
1053                         str.hash = be32_to_cpu(dent->de_hash);
1054                         new = gfs2_dirent_alloc(inode, nbh, &str);
1055                         if (IS_ERR(new)) {
1056                                 error = PTR_ERR(new);
1057                                 break;
1058                         }
1059
1060                         new->de_inum = dent->de_inum; /* No endian worries */
1061                         new->de_type = dent->de_type; /* No endian worries */
1062                         be16_add_cpu(&nleaf->lf_entries, 1);
1063
1064                         dirent_del(dip, obh, prev, dent);
1065
1066                         if (!oleaf->lf_entries)
1067                                 gfs2_consist_inode(dip);
1068                         be16_add_cpu(&oleaf->lf_entries, -1);
1069
1070                         if (!prev)
1071                                 prev = dent;
1072
1073                         moved = 1;
1074                 } else {
1075                         prev = dent;
1076                 }
1077                 dent = next;
1078         } while (dent);
1079
1080         oleaf->lf_depth = nleaf->lf_depth;
1081
1082         error = gfs2_meta_inode_buffer(dip, &dibh);
1083         if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) {
1084                 gfs2_trans_add_bh(dip->i_gl, dibh, 1);
1085                 gfs2_add_inode_blocks(&dip->i_inode, 1);
1086                 gfs2_dinode_out(dip, dibh->b_data);
1087                 brelse(dibh);
1088         }
1089
1090         brelse(obh);
1091         brelse(nbh);
1092
1093         return error;
1094
1095 fail_lpfree:
1096         kfree(lp);
1097
1098 fail_brelse:
1099         brelse(obh);
1100         brelse(nbh);
1101         return error;
1102 }
1103
1104 /**
1105  * dir_double_exhash - Double size of ExHash table
1106  * @dip: The GFS2 dinode
1107  *
1108  * Returns: 0 on success, error code on failure
1109  */
1110
1111 static int dir_double_exhash(struct gfs2_inode *dip)
1112 {
1113         struct buffer_head *dibh;
1114         u32 hsize;
1115         u32 hsize_bytes;
1116         __be64 *hc;
1117         __be64 *hc2, *h;
1118         int x;
1119         int error = 0;
1120
1121         hsize = 1 << dip->i_depth;
1122         hsize_bytes = hsize * sizeof(__be64);
1123
1124         hc = gfs2_dir_get_hash_table(dip);
1125         if (IS_ERR(hc))
1126                 return PTR_ERR(hc);
1127
1128         h = hc2 = kmalloc(hsize_bytes * 2, GFP_NOFS);
1129         if (!hc2)
1130                 return -ENOMEM;
1131
1132         error = gfs2_meta_inode_buffer(dip, &dibh);
1133         if (error)
1134                 goto out_kfree;
1135
1136         for (x = 0; x < hsize; x++) {
1137                 *h++ = *hc;
1138                 *h++ = *hc;
1139                 hc++;
1140         }
1141
1142         error = gfs2_dir_write_data(dip, (char *)hc2, 0, hsize_bytes * 2);
1143         if (error != (hsize_bytes * 2))
1144                 goto fail;
1145
1146         gfs2_dir_hash_inval(dip);
1147         dip->i_hash_cache = hc2;
1148         dip->i_depth++;
1149         gfs2_dinode_out(dip, dibh->b_data);
1150         brelse(dibh);
1151         return 0;
1152
1153 fail:
1154         /* Replace original hash table & size */
1155         gfs2_dir_write_data(dip, (char *)hc, 0, hsize_bytes);
1156         i_size_write(&dip->i_inode, hsize_bytes);
1157         gfs2_dinode_out(dip, dibh->b_data);
1158         brelse(dibh);
1159 out_kfree:
1160         kfree(hc2);
1161         return error;
1162 }
1163
1164 /**
1165  * compare_dents - compare directory entries by hash value
1166  * @a: first dent
1167  * @b: second dent
1168  *
1169  * When comparing the hash entries of @a to @b:
1170  *   gt: returns 1
1171  *   lt: returns -1
1172  *   eq: returns 0
1173  */
1174
1175 static int compare_dents(const void *a, const void *b)
1176 {
1177         const struct gfs2_dirent *dent_a, *dent_b;
1178         u32 hash_a, hash_b;
1179         int ret = 0;
1180
1181         dent_a = *(const struct gfs2_dirent **)a;
1182         hash_a = be32_to_cpu(dent_a->de_hash);
1183
1184         dent_b = *(const struct gfs2_dirent **)b;
1185         hash_b = be32_to_cpu(dent_b->de_hash);
1186
1187         if (hash_a > hash_b)
1188                 ret = 1;
1189         else if (hash_a < hash_b)
1190                 ret = -1;
1191         else {
1192                 unsigned int len_a = be16_to_cpu(dent_a->de_name_len);
1193                 unsigned int len_b = be16_to_cpu(dent_b->de_name_len);
1194
1195                 if (len_a > len_b)
1196                         ret = 1;
1197                 else if (len_a < len_b)
1198                         ret = -1;
1199                 else
1200                         ret = memcmp(dent_a + 1, dent_b + 1, len_a);
1201         }
1202
1203         return ret;
1204 }
1205
1206 /**
1207  * do_filldir_main - read out directory entries
1208  * @dip: The GFS2 inode
1209  * @offset: The offset in the file to read from
1210  * @opaque: opaque data to pass to filldir
1211  * @filldir: The function to pass entries to
1212  * @darr: an array of struct gfs2_dirent pointers to read
1213  * @entries: the number of entries in darr
1214  * @copied: pointer to int that's non-zero if a entry has been copied out
1215  *
1216  * Jump through some hoops to make sure that if there are hash collsions,
1217  * they are read out at the beginning of a buffer.  We want to minimize
1218  * the possibility that they will fall into different readdir buffers or
1219  * that someone will want to seek to that location.
1220  *
1221  * Returns: errno, >0 on exception from filldir
1222  */
1223
1224 static int do_filldir_main(struct gfs2_inode *dip, u64 *offset,
1225                            void *opaque, filldir_t filldir,
1226                            const struct gfs2_dirent **darr, u32 entries,
1227                            int *copied)
1228 {
1229         const struct gfs2_dirent *dent, *dent_next;
1230         u64 off, off_next;
1231         unsigned int x, y;
1232         int run = 0;
1233         int error = 0;
1234
1235         sort(darr, entries, sizeof(struct gfs2_dirent *), compare_dents, NULL);
1236
1237         dent_next = darr[0];
1238         off_next = be32_to_cpu(dent_next->de_hash);
1239         off_next = gfs2_disk_hash2offset(off_next);
1240
1241         for (x = 0, y = 1; x < entries; x++, y++) {
1242                 dent = dent_next;
1243                 off = off_next;
1244
1245                 if (y < entries) {
1246                         dent_next = darr[y];
1247                         off_next = be32_to_cpu(dent_next->de_hash);
1248                         off_next = gfs2_disk_hash2offset(off_next);
1249
1250                         if (off < *offset)
1251                                 continue;
1252                         *offset = off;
1253
1254                         if (off_next == off) {
1255                                 if (*copied && !run)
1256                                         return 1;
1257                                 run = 1;
1258                         } else
1259                                 run = 0;
1260                 } else {
1261                         if (off < *offset)
1262                                 continue;
1263                         *offset = off;
1264                 }
1265
1266                 error = filldir(opaque, (const char *)(dent + 1),
1267                                 be16_to_cpu(dent->de_name_len),
1268                                 off, be64_to_cpu(dent->de_inum.no_addr),
1269                                 be16_to_cpu(dent->de_type));
1270                 if (error)
1271                         return 1;
1272
1273                 *copied = 1;
1274         }
1275
1276         /* Increment the *offset by one, so the next time we come into the
1277            do_filldir fxn, we get the next entry instead of the last one in the
1278            current leaf */
1279
1280         (*offset)++;
1281
1282         return 0;
1283 }
1284
1285 static void *gfs2_alloc_sort_buffer(unsigned size)
1286 {
1287         void *ptr = NULL;
1288
1289         if (size < KMALLOC_MAX_SIZE)
1290                 ptr = kmalloc(size, GFP_NOFS | __GFP_NOWARN);
1291         if (!ptr)
1292                 ptr = __vmalloc(size, GFP_NOFS, PAGE_KERNEL);
1293         return ptr;
1294 }
1295
1296 static void gfs2_free_sort_buffer(void *ptr)
1297 {
1298         if (is_vmalloc_addr(ptr))
1299                 vfree(ptr);
1300         else
1301                 kfree(ptr);
1302 }
1303
1304 static int gfs2_dir_read_leaf(struct inode *inode, u64 *offset, void *opaque,
1305                               filldir_t filldir, int *copied, unsigned *depth,
1306                               u64 leaf_no)
1307 {
1308         struct gfs2_inode *ip = GFS2_I(inode);
1309         struct gfs2_sbd *sdp = GFS2_SB(inode);
1310         struct buffer_head *bh;
1311         struct gfs2_leaf *lf;
1312         unsigned entries = 0, entries2 = 0;
1313         unsigned leaves = 0;
1314         const struct gfs2_dirent **darr, *dent;
1315         struct dirent_gather g;
1316         struct buffer_head **larr;
1317         int leaf = 0;
1318         int error, i;
1319         u64 lfn = leaf_no;
1320
1321         do {
1322                 error = get_leaf(ip, lfn, &bh);
1323                 if (error)
1324                         goto out;
1325                 lf = (struct gfs2_leaf *)bh->b_data;
1326                 if (leaves == 0)
1327                         *depth = be16_to_cpu(lf->lf_depth);
1328                 entries += be16_to_cpu(lf->lf_entries);
1329                 leaves++;
1330                 lfn = be64_to_cpu(lf->lf_next);
1331                 brelse(bh);
1332         } while(lfn);
1333
1334         if (!entries)
1335                 return 0;
1336
1337         error = -ENOMEM;
1338         /*
1339          * The extra 99 entries are not normally used, but are a buffer
1340          * zone in case the number of entries in the leaf is corrupt.
1341          * 99 is the maximum number of entries that can fit in a single
1342          * leaf block.
1343          */
1344         larr = gfs2_alloc_sort_buffer((leaves + entries + 99) * sizeof(void *));
1345         if (!larr)
1346                 goto out;
1347         darr = (const struct gfs2_dirent **)(larr + leaves);
1348         g.pdent = darr;
1349         g.offset = 0;
1350         lfn = leaf_no;
1351
1352         do {
1353                 error = get_leaf(ip, lfn, &bh);
1354                 if (error)
1355                         goto out_free;
1356                 lf = (struct gfs2_leaf *)bh->b_data;
1357                 lfn = be64_to_cpu(lf->lf_next);
1358                 if (lf->lf_entries) {
1359                         entries2 += be16_to_cpu(lf->lf_entries);
1360                         dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
1361                                                 gfs2_dirent_gather, NULL, &g);
1362                         error = PTR_ERR(dent);
1363                         if (IS_ERR(dent))
1364                                 goto out_free;
1365                         if (entries2 != g.offset) {
1366                                 fs_warn(sdp, "Number of entries corrupt in dir "
1367                                                 "leaf %llu, entries2 (%u) != "
1368                                                 "g.offset (%u)\n",
1369                                         (unsigned long long)bh->b_blocknr,
1370                                         entries2, g.offset);
1371
1372                                 error = -EIO;
1373                                 goto out_free;
1374                         }
1375                         error = 0;
1376                         larr[leaf++] = bh;
1377                 } else {
1378                         brelse(bh);
1379                 }
1380         } while(lfn);
1381
1382         BUG_ON(entries2 != entries);
1383         error = do_filldir_main(ip, offset, opaque, filldir, darr,
1384                                 entries, copied);
1385 out_free:
1386         for(i = 0; i < leaf; i++)
1387                 brelse(larr[i]);
1388         gfs2_free_sort_buffer(larr);
1389 out:
1390         return error;
1391 }
1392
1393
1394 /**
1395  * dir_e_read - Reads the entries from a directory into a filldir buffer
1396  * @dip: dinode pointer
1397  * @offset: the hash of the last entry read shifted to the right once
1398  * @opaque: buffer for the filldir function to fill
1399  * @filldir: points to the filldir function to use
1400  *
1401  * Returns: errno
1402  */
1403
1404 static int dir_e_read(struct inode *inode, u64 *offset, void *opaque,
1405                       filldir_t filldir)
1406 {
1407         struct gfs2_inode *dip = GFS2_I(inode);
1408         u32 hsize, len = 0;
1409         u32 hash, index;
1410         __be64 *lp;
1411         int copied = 0;
1412         int error = 0;
1413         unsigned depth = 0;
1414
1415         hsize = 1 << dip->i_depth;
1416         hash = gfs2_dir_offset2hash(*offset);
1417         index = hash >> (32 - dip->i_depth);
1418
1419         lp = gfs2_dir_get_hash_table(dip);
1420         if (IS_ERR(lp))
1421                 return PTR_ERR(lp);
1422
1423         while (index < hsize) {
1424                 error = gfs2_dir_read_leaf(inode, offset, opaque, filldir,
1425                                            &copied, &depth,
1426                                            be64_to_cpu(lp[index]));
1427                 if (error)
1428                         break;
1429
1430                 len = 1 << (dip->i_depth - depth);
1431                 index = (index & ~(len - 1)) + len;
1432         }
1433
1434         if (error > 0)
1435                 error = 0;
1436         return error;
1437 }
1438
1439 int gfs2_dir_read(struct inode *inode, u64 *offset, void *opaque,
1440                   filldir_t filldir)
1441 {
1442         struct gfs2_inode *dip = GFS2_I(inode);
1443         struct gfs2_sbd *sdp = GFS2_SB(inode);
1444         struct dirent_gather g;
1445         const struct gfs2_dirent **darr, *dent;
1446         struct buffer_head *dibh;
1447         int copied = 0;
1448         int error;
1449
1450         if (!dip->i_entries)
1451                 return 0;
1452
1453         if (dip->i_diskflags & GFS2_DIF_EXHASH)
1454                 return dir_e_read(inode, offset, opaque, filldir);
1455
1456         if (!gfs2_is_stuffed(dip)) {
1457                 gfs2_consist_inode(dip);
1458                 return -EIO;
1459         }
1460
1461         error = gfs2_meta_inode_buffer(dip, &dibh);
1462         if (error)
1463                 return error;
1464
1465         error = -ENOMEM;
1466         /* 96 is max number of dirents which can be stuffed into an inode */
1467         darr = kmalloc(96 * sizeof(struct gfs2_dirent *), GFP_NOFS);
1468         if (darr) {
1469                 g.pdent = darr;
1470                 g.offset = 0;
1471                 dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size,
1472                                         gfs2_dirent_gather, NULL, &g);
1473                 if (IS_ERR(dent)) {
1474                         error = PTR_ERR(dent);
1475                         goto out;
1476                 }
1477                 if (dip->i_entries != g.offset) {
1478                         fs_warn(sdp, "Number of entries corrupt in dir %llu, "
1479                                 "ip->i_entries (%u) != g.offset (%u)\n",
1480                                 (unsigned long long)dip->i_no_addr,
1481                                 dip->i_entries,
1482                                 g.offset);
1483                         error = -EIO;
1484                         goto out;
1485                 }
1486                 error = do_filldir_main(dip, offset, opaque, filldir, darr,
1487                                         dip->i_entries, &copied);
1488 out:
1489                 kfree(darr);
1490         }
1491
1492         if (error > 0)
1493                 error = 0;
1494
1495         brelse(dibh);
1496
1497         return error;
1498 }
1499
1500 /**
1501  * gfs2_dir_search - Search a directory
1502  * @dip: The GFS2 inode
1503  * @filename:
1504  * @inode:
1505  *
1506  * This routine searches a directory for a file or another directory.
1507  * Assumes a glock is held on dip.
1508  *
1509  * Returns: errno
1510  */
1511
1512 struct inode *gfs2_dir_search(struct inode *dir, const struct qstr *name)
1513 {
1514         struct buffer_head *bh;
1515         struct gfs2_dirent *dent;
1516         struct inode *inode;
1517
1518         dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1519         if (dent) {
1520                 if (IS_ERR(dent))
1521                         return ERR_CAST(dent);
1522                 inode = gfs2_inode_lookup(dir->i_sb,
1523                                 be16_to_cpu(dent->de_type),
1524                                 be64_to_cpu(dent->de_inum.no_addr),
1525                                 be64_to_cpu(dent->de_inum.no_formal_ino), 0);
1526                 brelse(bh);
1527                 return inode;
1528         }
1529         return ERR_PTR(-ENOENT);
1530 }
1531
1532 int gfs2_dir_check(struct inode *dir, const struct qstr *name,
1533                    const struct gfs2_inode *ip)
1534 {
1535         struct buffer_head *bh;
1536         struct gfs2_dirent *dent;
1537         int ret = -ENOENT;
1538
1539         dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1540         if (dent) {
1541                 if (IS_ERR(dent))
1542                         return PTR_ERR(dent);
1543                 if (ip) {
1544                         if (be64_to_cpu(dent->de_inum.no_addr) != ip->i_no_addr)
1545                                 goto out;
1546                         if (be64_to_cpu(dent->de_inum.no_formal_ino) !=
1547                             ip->i_no_formal_ino)
1548                                 goto out;
1549                         if (unlikely(IF2DT(ip->i_inode.i_mode) !=
1550                             be16_to_cpu(dent->de_type))) {
1551                                 gfs2_consist_inode(GFS2_I(dir));
1552                                 ret = -EIO;
1553                                 goto out;
1554                         }
1555                 }
1556                 ret = 0;
1557 out:
1558                 brelse(bh);
1559         }
1560         return ret;
1561 }
1562
1563 static int dir_new_leaf(struct inode *inode, const struct qstr *name)
1564 {
1565         struct buffer_head *bh, *obh;
1566         struct gfs2_inode *ip = GFS2_I(inode);
1567         struct gfs2_leaf *leaf, *oleaf;
1568         int error;
1569         u32 index;
1570         u64 bn;
1571
1572         index = name->hash >> (32 - ip->i_depth);
1573         error = get_first_leaf(ip, index, &obh);
1574         if (error)
1575                 return error;
1576         do {
1577                 oleaf = (struct gfs2_leaf *)obh->b_data;
1578                 bn = be64_to_cpu(oleaf->lf_next);
1579                 if (!bn)
1580                         break;
1581                 brelse(obh);
1582                 error = get_leaf(ip, bn, &obh);
1583                 if (error)
1584                         return error;
1585         } while(1);
1586
1587         gfs2_trans_add_bh(ip->i_gl, obh, 1);
1588
1589         leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth));
1590         if (!leaf) {
1591                 brelse(obh);
1592                 return -ENOSPC;
1593         }
1594         oleaf->lf_next = cpu_to_be64(bh->b_blocknr);
1595         brelse(bh);
1596         brelse(obh);
1597
1598         error = gfs2_meta_inode_buffer(ip, &bh);
1599         if (error)
1600                 return error;
1601         gfs2_trans_add_bh(ip->i_gl, bh, 1);
1602         gfs2_add_inode_blocks(&ip->i_inode, 1);
1603         gfs2_dinode_out(ip, bh->b_data);
1604         brelse(bh);
1605         return 0;
1606 }
1607
1608 /**
1609  * gfs2_dir_add - Add new filename into directory
1610  * @dip: The GFS2 inode
1611  * @filename: The new name
1612  * @inode: The inode number of the entry
1613  * @type: The type of the entry
1614  *
1615  * Returns: 0 on success, error code on failure
1616  */
1617
1618 int gfs2_dir_add(struct inode *inode, const struct qstr *name,
1619                  const struct gfs2_inode *nip)
1620 {
1621         struct gfs2_inode *ip = GFS2_I(inode);
1622         struct buffer_head *bh;
1623         struct gfs2_dirent *dent;
1624         struct gfs2_leaf *leaf;
1625         int error;
1626
1627         while(1) {
1628                 dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space,
1629                                           &bh);
1630                 if (dent) {
1631                         if (IS_ERR(dent))
1632                                 return PTR_ERR(dent);
1633                         dent = gfs2_init_dirent(inode, dent, name, bh);
1634                         gfs2_inum_out(nip, dent);
1635                         dent->de_type = cpu_to_be16(IF2DT(nip->i_inode.i_mode));
1636                         if (ip->i_diskflags & GFS2_DIF_EXHASH) {
1637                                 leaf = (struct gfs2_leaf *)bh->b_data;
1638                                 be16_add_cpu(&leaf->lf_entries, 1);
1639                         }
1640                         brelse(bh);
1641                         error = gfs2_meta_inode_buffer(ip, &bh);
1642                         if (error)
1643                                 break;
1644                         gfs2_trans_add_bh(ip->i_gl, bh, 1);
1645                         ip->i_entries++;
1646                         ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1647                         if (S_ISDIR(nip->i_inode.i_mode))
1648                                 inc_nlink(&ip->i_inode);
1649                         gfs2_dinode_out(ip, bh->b_data);
1650                         brelse(bh);
1651                         error = 0;
1652                         break;
1653                 }
1654                 if (!(ip->i_diskflags & GFS2_DIF_EXHASH)) {
1655                         error = dir_make_exhash(inode);
1656                         if (error)
1657                                 break;
1658                         continue;
1659                 }
1660                 error = dir_split_leaf(inode, name);
1661                 if (error == 0)
1662                         continue;
1663                 if (error < 0)
1664                         break;
1665                 if (ip->i_depth < GFS2_DIR_MAX_DEPTH) {
1666                         error = dir_double_exhash(ip);
1667                         if (error)
1668                                 break;
1669                         error = dir_split_leaf(inode, name);
1670                         if (error < 0)
1671                                 break;
1672                         if (error == 0)
1673                                 continue;
1674                 }
1675                 error = dir_new_leaf(inode, name);
1676                 if (!error)
1677                         continue;
1678                 error = -ENOSPC;
1679                 break;
1680         }
1681         return error;
1682 }
1683
1684
1685 /**
1686  * gfs2_dir_del - Delete a directory entry
1687  * @dip: The GFS2 inode
1688  * @filename: The filename
1689  *
1690  * Returns: 0 on success, error code on failure
1691  */
1692
1693 int gfs2_dir_del(struct gfs2_inode *dip, const struct dentry *dentry)
1694 {
1695         const struct qstr *name = &dentry->d_name;
1696         struct gfs2_dirent *dent, *prev = NULL;
1697         struct buffer_head *bh;
1698         int error;
1699
1700         /* Returns _either_ the entry (if its first in block) or the
1701            previous entry otherwise */
1702         dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh);
1703         if (!dent) {
1704                 gfs2_consist_inode(dip);
1705                 return -EIO;
1706         }
1707         if (IS_ERR(dent)) {
1708                 gfs2_consist_inode(dip);
1709                 return PTR_ERR(dent);
1710         }
1711         /* If not first in block, adjust pointers accordingly */
1712         if (gfs2_dirent_find(dent, name, NULL) == 0) {
1713                 prev = dent;
1714                 dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len));
1715         }
1716
1717         dirent_del(dip, bh, prev, dent);
1718         if (dip->i_diskflags & GFS2_DIF_EXHASH) {
1719                 struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data;
1720                 u16 entries = be16_to_cpu(leaf->lf_entries);
1721                 if (!entries)
1722                         gfs2_consist_inode(dip);
1723                 leaf->lf_entries = cpu_to_be16(--entries);
1724         }
1725         brelse(bh);
1726
1727         error = gfs2_meta_inode_buffer(dip, &bh);
1728         if (error)
1729                 return error;
1730
1731         if (!dip->i_entries)
1732                 gfs2_consist_inode(dip);
1733         gfs2_trans_add_bh(dip->i_gl, bh, 1);
1734         dip->i_entries--;
1735         dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME;
1736         if (S_ISDIR(dentry->d_inode->i_mode))
1737                 drop_nlink(&dip->i_inode);
1738         gfs2_dinode_out(dip, bh->b_data);
1739         brelse(bh);
1740         mark_inode_dirty(&dip->i_inode);
1741
1742         return error;
1743 }
1744
1745 /**
1746  * gfs2_dir_mvino - Change inode number of directory entry
1747  * @dip: The GFS2 inode
1748  * @filename:
1749  * @new_inode:
1750  *
1751  * This routine changes the inode number of a directory entry.  It's used
1752  * by rename to change ".." when a directory is moved.
1753  * Assumes a glock is held on dvp.
1754  *
1755  * Returns: errno
1756  */
1757
1758 int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename,
1759                    const struct gfs2_inode *nip, unsigned int new_type)
1760 {
1761         struct buffer_head *bh;
1762         struct gfs2_dirent *dent;
1763         int error;
1764
1765         dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh);
1766         if (!dent) {
1767                 gfs2_consist_inode(dip);
1768                 return -EIO;
1769         }
1770         if (IS_ERR(dent))
1771                 return PTR_ERR(dent);
1772
1773         gfs2_trans_add_bh(dip->i_gl, bh, 1);
1774         gfs2_inum_out(nip, dent);
1775         dent->de_type = cpu_to_be16(new_type);
1776
1777         if (dip->i_diskflags & GFS2_DIF_EXHASH) {
1778                 brelse(bh);
1779                 error = gfs2_meta_inode_buffer(dip, &bh);
1780                 if (error)
1781                         return error;
1782                 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1783         }
1784
1785         dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME;
1786         gfs2_dinode_out(dip, bh->b_data);
1787         brelse(bh);
1788         return 0;
1789 }
1790
1791 /**
1792  * leaf_dealloc - Deallocate a directory leaf
1793  * @dip: the directory
1794  * @index: the hash table offset in the directory
1795  * @len: the number of pointers to this leaf
1796  * @leaf_no: the leaf number
1797  * @leaf_bh: buffer_head for the starting leaf
1798  * last_dealloc: 1 if this is the final dealloc for the leaf, else 0
1799  *
1800  * Returns: errno
1801  */
1802
1803 static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len,
1804                         u64 leaf_no, struct buffer_head *leaf_bh,
1805                         int last_dealloc)
1806 {
1807         struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1808         struct gfs2_leaf *tmp_leaf;
1809         struct gfs2_rgrp_list rlist;
1810         struct buffer_head *bh, *dibh;
1811         u64 blk, nblk;
1812         unsigned int rg_blocks = 0, l_blocks = 0;
1813         char *ht;
1814         unsigned int x, size = len * sizeof(u64);
1815         int error;
1816
1817         memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
1818
1819         ht = kzalloc(size, GFP_NOFS);
1820         if (!ht)
1821                 return -ENOMEM;
1822
1823         if (!gfs2_alloc_get(dip)) {
1824                 error = -ENOMEM;
1825                 goto out;
1826         }
1827
1828         error = gfs2_quota_hold(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1829         if (error)
1830                 goto out_put;
1831
1832         error = gfs2_rindex_hold(sdp, &dip->i_alloc->al_ri_gh);
1833         if (error)
1834                 goto out_qs;
1835
1836         /*  Count the number of leaves  */
1837         bh = leaf_bh;
1838
1839         for (blk = leaf_no; blk; blk = nblk) {
1840                 if (blk != leaf_no) {
1841                         error = get_leaf(dip, blk, &bh);
1842                         if (error)
1843                                 goto out_rlist;
1844                 }
1845                 tmp_leaf = (struct gfs2_leaf *)bh->b_data;
1846                 nblk = be64_to_cpu(tmp_leaf->lf_next);
1847                 if (blk != leaf_no)
1848                         brelse(bh);
1849
1850                 gfs2_rlist_add(sdp, &rlist, blk);
1851                 l_blocks++;
1852         }
1853
1854         gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE);
1855
1856         for (x = 0; x < rlist.rl_rgrps; x++) {
1857                 struct gfs2_rgrpd *rgd;
1858                 rgd = rlist.rl_ghs[x].gh_gl->gl_object;
1859                 rg_blocks += rgd->rd_length;
1860         }
1861
1862         error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
1863         if (error)
1864                 goto out_rlist;
1865
1866         error = gfs2_trans_begin(sdp,
1867                         rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) +
1868                         RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks);
1869         if (error)
1870                 goto out_rg_gunlock;
1871
1872         bh = leaf_bh;
1873
1874         for (blk = leaf_no; blk; blk = nblk) {
1875                 if (blk != leaf_no) {
1876                         error = get_leaf(dip, blk, &bh);
1877                         if (error)
1878                                 goto out_end_trans;
1879                 }
1880                 tmp_leaf = (struct gfs2_leaf *)bh->b_data;
1881                 nblk = be64_to_cpu(tmp_leaf->lf_next);
1882                 if (blk != leaf_no)
1883                         brelse(bh);
1884
1885                 gfs2_free_meta(dip, blk, 1);
1886                 gfs2_add_inode_blocks(&dip->i_inode, -1);
1887         }
1888
1889         error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size);
1890         if (error != size) {
1891                 if (error >= 0)
1892                         error = -EIO;
1893                 goto out_end_trans;
1894         }
1895
1896         error = gfs2_meta_inode_buffer(dip, &dibh);
1897         if (error)
1898                 goto out_end_trans;
1899
1900         gfs2_trans_add_bh(dip->i_gl, dibh, 1);
1901         /* On the last dealloc, make this a regular file in case we crash.
1902            (We don't want to free these blocks a second time.)  */
1903         if (last_dealloc)
1904                 dip->i_inode.i_mode = S_IFREG;
1905         gfs2_dinode_out(dip, dibh->b_data);
1906         brelse(dibh);
1907
1908 out_end_trans:
1909         gfs2_trans_end(sdp);
1910 out_rg_gunlock:
1911         gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
1912 out_rlist:
1913         gfs2_rlist_free(&rlist);
1914         gfs2_glock_dq_uninit(&dip->i_alloc->al_ri_gh);
1915 out_qs:
1916         gfs2_quota_unhold(dip);
1917 out_put:
1918         gfs2_alloc_put(dip);
1919 out:
1920         kfree(ht);
1921         return error;
1922 }
1923
1924 /**
1925  * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
1926  * @dip: the directory
1927  *
1928  * Dealloc all on-disk directory leaves to FREEMETA state
1929  * Change on-disk inode type to "regular file"
1930  *
1931  * Returns: errno
1932  */
1933
1934 int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip)
1935 {
1936         struct buffer_head *bh;
1937         struct gfs2_leaf *leaf;
1938         u32 hsize, len;
1939         u32 index = 0, next_index;
1940         __be64 *lp;
1941         u64 leaf_no;
1942         int error = 0, last;
1943
1944         hsize = 1 << dip->i_depth;
1945
1946         lp = gfs2_dir_get_hash_table(dip);
1947         if (IS_ERR(lp))
1948                 return PTR_ERR(lp);
1949
1950         while (index < hsize) {
1951                 leaf_no = be64_to_cpu(lp[index]);
1952                 if (leaf_no) {
1953                         error = get_leaf(dip, leaf_no, &bh);
1954                         if (error)
1955                                 goto out;
1956                         leaf = (struct gfs2_leaf *)bh->b_data;
1957                         len = 1 << (dip->i_depth - be16_to_cpu(leaf->lf_depth));
1958
1959                         next_index = (index & ~(len - 1)) + len;
1960                         last = ((next_index >= hsize) ? 1 : 0);
1961                         error = leaf_dealloc(dip, index, len, leaf_no, bh,
1962                                              last);
1963                         brelse(bh);
1964                         if (error)
1965                                 goto out;
1966                         index = next_index;
1967                 } else
1968                         index++;
1969         }
1970
1971         if (index != hsize) {
1972                 gfs2_consist_inode(dip);
1973                 error = -EIO;
1974         }
1975
1976 out:
1977
1978         return error;
1979 }
1980
1981 /**
1982  * gfs2_diradd_alloc_required - find if adding entry will require an allocation
1983  * @ip: the file being written to
1984  * @filname: the filename that's going to be added
1985  *
1986  * Returns: 1 if alloc required, 0 if not, -ve on error
1987  */
1988
1989 int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name)
1990 {
1991         struct gfs2_dirent *dent;
1992         struct buffer_head *bh;
1993
1994         dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh);
1995         if (!dent) {
1996                 return 1;
1997         }
1998         if (IS_ERR(dent))
1999                 return PTR_ERR(dent);
2000         brelse(bh);
2001         return 0;
2002 }
2003