fs/nilfs2/alloc.c

   1 /*
   2  * alloc.c - NILFS dat/inode allocator
   3  *
   4  * Copyright (C) 2006-2008 Nippon Telegraph and Telephone Corporation.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * Originally written by Koji Sato.
  17  * Two allocators were unified by Ryusuke Konishi and Amagai Yoshiji.
  18  */
  19
  20 #include <linux/types.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/fs.h>
  23 #include <linux/bitops.h>
  24 #include <linux/slab.h>
  25 #include "mdt.h"
  26 #include "alloc.h"
  27
  28
  29 /**
  30  * nilfs_palloc_groups_per_desc_block - get the number of groups that a group
  31  *                                      descriptor block can maintain
  32  * @inode: inode of metadata file using this allocator
  33  */
  34 static inline unsigned long
  35 nilfs_palloc_groups_per_desc_block(const struct inode *inode)
  36 {
  37         return i_blocksize(inode) /
  38                 sizeof(struct nilfs_palloc_group_desc);
  39 }
  40
  41 /**
  42  * nilfs_palloc_groups_count - get maximum number of groups
  43  * @inode: inode of metadata file using this allocator
  44  */
  45 static inline unsigned long
  46 nilfs_palloc_groups_count(const struct inode *inode)
  47 {
  48         return 1UL << (BITS_PER_LONG - (inode->i_blkbits + 3 /* log2(8) */));
  49 }
  50
  51 /**
  52  * nilfs_palloc_init_blockgroup - initialize private variables for allocator
  53  * @inode: inode of metadata file using this allocator
  54  * @entry_size: size of the persistent object
  55  */
  56 int nilfs_palloc_init_blockgroup(struct inode *inode, unsigned int entry_size)
  57 {
  58         struct nilfs_mdt_info *mi = NILFS_MDT(inode);
  59
  60         mi->mi_bgl = kmalloc(sizeof(*mi->mi_bgl), GFP_NOFS);
  61         if (!mi->mi_bgl)
  62                 return -ENOMEM;
  63
  64         bgl_lock_init(mi->mi_bgl);
  65
  66         nilfs_mdt_set_entry_size(inode, entry_size, 0);
  67
  68         mi->mi_blocks_per_group =
  69                 DIV_ROUND_UP(nilfs_palloc_entries_per_group(inode),
  70                              mi->mi_entries_per_block) + 1;
  71                 /*
  72                  * Number of blocks in a group including entry blocks
  73                  * and a bitmap block
  74                  */
  75         mi->mi_blocks_per_desc_block =
  76                 nilfs_palloc_groups_per_desc_block(inode) *
  77                 mi->mi_blocks_per_group + 1;
  78                 /*
  79                  * Number of blocks per descriptor including the
  80                  * descriptor block
  81                  */
  82         return 0;
  83 }
  84
  85 /**
  86  * nilfs_palloc_group - get group number and offset from an entry number
  87  * @inode: inode of metadata file using this allocator
  88  * @nr: serial number of the entry (e.g. inode number)
  89  * @offset: pointer to store offset number in the group
  90  */
  91 static unsigned long nilfs_palloc_group(const struct inode *inode, __u64 nr,
  92                                         unsigned long *offset)
  93 {
  94         __u64 group = nr;
  95
  96         *offset = do_div(group, nilfs_palloc_entries_per_group(inode));
  97         return group;
  98 }
  99
 100 /**
 101  * nilfs_palloc_desc_blkoff - get block offset of a group descriptor block
 102  * @inode: inode of metadata file using this allocator
 103  * @group: group number
 104  *
 105  * nilfs_palloc_desc_blkoff() returns block offset of the descriptor
 106  * block which contains a descriptor of the specified group.
 107  */
 108 static unsigned long
 109 nilfs_palloc_desc_blkoff(const struct inode *inode, unsigned long group)
 110 {
 111         unsigned long desc_block =
 112                 group / nilfs_palloc_groups_per_desc_block(inode);
 113         return desc_block * NILFS_MDT(inode)->mi_blocks_per_desc_block;
 114 }
 115
 116 /**
 117  * nilfs_palloc_bitmap_blkoff - get block offset of a bitmap block
 118  * @inode: inode of metadata file using this allocator
 119  * @group: group number
 120  *
 121  * nilfs_palloc_bitmap_blkoff() returns block offset of the bitmap
 122  * block used to allocate/deallocate entries in the specified group.
 123  */
 124 static unsigned long
 125 nilfs_palloc_bitmap_blkoff(const struct inode *inode, unsigned long group)
 126 {
 127         unsigned long desc_offset =
 128                 group % nilfs_palloc_groups_per_desc_block(inode);
 129         return nilfs_palloc_desc_blkoff(inode, group) + 1 +
 130                 desc_offset * NILFS_MDT(inode)->mi_blocks_per_group;
 131 }
 132
 133 /**
 134  * nilfs_palloc_group_desc_nfrees - get the number of free entries in a group
 135  * @desc: pointer to descriptor structure for the group
 136  * @lock: spin lock protecting @desc
 137  */
 138 static unsigned long
 139 nilfs_palloc_group_desc_nfrees(const struct nilfs_palloc_group_desc *desc,
 140                                spinlock_t *lock)
 141 {
 142         unsigned long nfree;
 143
 144         spin_lock(lock);
 145         nfree = le32_to_cpu(desc->pg_nfrees);
 146         spin_unlock(lock);
 147         return nfree;
 148 }
 149
 150 /**
 151  * nilfs_palloc_group_desc_add_entries - adjust count of free entries
 152  * @desc: pointer to descriptor structure for the group
 153  * @lock: spin lock protecting @desc
 154  * @n: delta to be added
 155  */
 156 static u32
 157 nilfs_palloc_group_desc_add_entries(struct nilfs_palloc_group_desc *desc,
 158                                     spinlock_t *lock, u32 n)
 159 {
 160         u32 nfree;
 161
 162         spin_lock(lock);
 163         le32_add_cpu(&desc->pg_nfrees, n);
 164         nfree = le32_to_cpu(desc->pg_nfrees);
 165         spin_unlock(lock);
 166         return nfree;
 167 }
 168
 169 /**
 170  * nilfs_palloc_entry_blkoff - get block offset of an entry block
 171  * @inode: inode of metadata file using this allocator
 172  * @nr: serial number of the entry (e.g. inode number)
 173  */
 174 static unsigned long
 175 nilfs_palloc_entry_blkoff(const struct inode *inode, __u64 nr)
 176 {
 177         unsigned long group, group_offset;
 178
 179         group = nilfs_palloc_group(inode, nr, &group_offset);
 180
 181         return nilfs_palloc_bitmap_blkoff(inode, group) + 1 +
 182                 group_offset / NILFS_MDT(inode)->mi_entries_per_block;
 183 }
 184
 185 /**
 186  * nilfs_palloc_desc_block_init - initialize buffer of a group descriptor block
 187  * @inode: inode of metadata file
 188  * @bh: buffer head of the buffer to be initialized
 189  * @kaddr: kernel address mapped for the page including the buffer
 190  */
 191 static void nilfs_palloc_desc_block_init(struct inode *inode,
 192                                          struct buffer_head *bh, void *kaddr)
 193 {
 194         struct nilfs_palloc_group_desc *desc = kaddr + bh_offset(bh);
 195         unsigned long n = nilfs_palloc_groups_per_desc_block(inode);
 196         __le32 nfrees;
 197
 198         nfrees = cpu_to_le32(nilfs_palloc_entries_per_group(inode));
 199         while (n-- > 0) {
 200                 desc->pg_nfrees = nfrees;
 201                 desc++;
 202         }
 203 }
 204
 205 static int nilfs_palloc_get_block(struct inode *inode, unsigned long blkoff,
 206                                   int create,
 207                                   void (*init_block)(struct inode *,
 208                                                      struct buffer_head *,
 209                                                      void *),
 210                                   struct buffer_head **bhp,
 211                                   struct nilfs_bh_assoc *prev,
 212                                   spinlock_t *lock)
 213 {
 214         int ret;
 215
 216         spin_lock(lock);
 217         if (prev->bh && blkoff == prev->blkoff) {
 218                 get_bh(prev->bh);
 219                 *bhp = prev->bh;
 220                 spin_unlock(lock);
 221                 return 0;
 222         }
 223         spin_unlock(lock);
 224
 225         ret = nilfs_mdt_get_block(inode, blkoff, create, init_block, bhp);
 226         if (!ret) {
 227                 spin_lock(lock);
 228                 /*
 229                  * The following code must be safe for change of the
 230                  * cache contents during the get block call.
 231                  */
 232                 brelse(prev->bh);
 233                 get_bh(*bhp);
 234                 prev->bh = *bhp;
 235                 prev->blkoff = blkoff;
 236                 spin_unlock(lock);
 237         }
 238         return ret;
 239 }
 240
 241 /**
 242  * nilfs_palloc_delete_block - delete a block on the persistent allocator file
 243  * @inode: inode of metadata file using this allocator
 244  * @blkoff: block offset
 245  * @prev: nilfs_bh_assoc struct of the last used buffer
 246  * @lock: spin lock protecting @prev
 247  */
 248 static int nilfs_palloc_delete_block(struct inode *inode, unsigned long blkoff,
 249                                      struct nilfs_bh_assoc *prev,
 250                                      spinlock_t *lock)
 251 {
 252         spin_lock(lock);
 253         if (prev->bh && blkoff == prev->blkoff) {
 254                 brelse(prev->bh);
 255                 prev->bh = NULL;
 256         }
 257         spin_unlock(lock);
 258         return nilfs_mdt_delete_block(inode, blkoff);
 259 }
 260
 261 /**
 262  * nilfs_palloc_get_desc_block - get buffer head of a group descriptor block
 263  * @inode: inode of metadata file using this allocator
 264  * @group: group number
 265  * @create: create flag
 266  * @bhp: pointer to store the resultant buffer head
 267  */
 268 static int nilfs_palloc_get_desc_block(struct inode *inode,
 269                                        unsigned long group,
 270                                        int create, struct buffer_head **bhp)
 271 {
 272         struct nilfs_palloc_cache *cache = NILFS_MDT(inode)->mi_palloc_cache;
 273
 274         return nilfs_palloc_get_block(inode,
 275                                       nilfs_palloc_desc_blkoff(inode, group),
 276                                       create, nilfs_palloc_desc_block_init,
 277                                       bhp, &cache->prev_desc, &cache->lock);
 278 }
 279
 280 /**
 281  * nilfs_palloc_get_bitmap_block - get buffer head of a bitmap block
 282  * @inode: inode of metadata file using this allocator
 283  * @group: group number
 284  * @create: create flag
 285  * @bhp: pointer to store the resultant buffer head
 286  */
 287 static int nilfs_palloc_get_bitmap_block(struct inode *inode,
 288                                          unsigned long group,
 289                                          int create, struct buffer_head **bhp)
 290 {
 291         struct nilfs_palloc_cache *cache = NILFS_MDT(inode)->mi_palloc_cache;
 292
 293         return nilfs_palloc_get_block(inode,
 294                                       nilfs_palloc_bitmap_blkoff(inode, group),
 295                                       create, NULL, bhp,
 296                                       &cache->prev_bitmap, &cache->lock);
 297 }
 298
 299 /**
 300  * nilfs_palloc_delete_bitmap_block - delete a bitmap block
 301  * @inode: inode of metadata file using this allocator
 302  * @group: group number
 303  */
 304 static int nilfs_palloc_delete_bitmap_block(struct inode *inode,
 305                                             unsigned long group)
 306 {
 307         struct nilfs_palloc_cache *cache = NILFS_MDT(inode)->mi_palloc_cache;
 308
 309         return nilfs_palloc_delete_block(inode,
 310                                          nilfs_palloc_bitmap_blkoff(inode,
 311                                                                     group),
 312                                          &cache->prev_bitmap, &cache->lock);
 313 }
 314
 315 /**
 316  * nilfs_palloc_get_entry_block - get buffer head of an entry block
 317  * @inode: inode of metadata file using this allocator
 318  * @nr: serial number of the entry (e.g. inode number)
 319  * @create: create flag
 320  * @bhp: pointer to store the resultant buffer head
 321  */
 322 int nilfs_palloc_get_entry_block(struct inode *inode, __u64 nr,
 323                                  int create, struct buffer_head **bhp)
 324 {
 325         struct nilfs_palloc_cache *cache = NILFS_MDT(inode)->mi_palloc_cache;
 326
 327         return nilfs_palloc_get_block(inode,
 328                                       nilfs_palloc_entry_blkoff(inode, nr),
 329                                       create, NULL, bhp,
 330                                       &cache->prev_entry, &cache->lock);
 331 }
 332
 333 /**
 334  * nilfs_palloc_delete_entry_block - delete an entry block
 335  * @inode: inode of metadata file using this allocator
 336  * @nr: serial number of the entry
 337  */
 338 static int nilfs_palloc_delete_entry_block(struct inode *inode, __u64 nr)
 339 {
 340         struct nilfs_palloc_cache *cache = NILFS_MDT(inode)->mi_palloc_cache;
 341
 342         return nilfs_palloc_delete_block(inode,
 343                                          nilfs_palloc_entry_blkoff(inode, nr),
 344                                          &cache->prev_entry, &cache->lock);
 345 }
 346
 347 /**
 348  * nilfs_palloc_block_get_group_desc - get kernel address of a group descriptor
 349  * @inode: inode of metadata file using this allocator
 350  * @group: group number
 351  * @bh: buffer head of the buffer storing the group descriptor block
 352  * @kaddr: kernel address mapped for the page including the buffer
 353  */
 354 static struct nilfs_palloc_group_desc *
 355 nilfs_palloc_block_get_group_desc(const struct inode *inode,
 356                                   unsigned long group,
 357                                   const struct buffer_head *bh, void *kaddr)
 358 {
 359         return (struct nilfs_palloc_group_desc *)(kaddr + bh_offset(bh)) +
 360                 group % nilfs_palloc_groups_per_desc_block(inode);
 361 }
 362
 363 /**
 364  * nilfs_palloc_block_get_entry - get kernel address of an entry
 365  * @inode: inode of metadata file using this allocator
 366  * @nr: serial number of the entry (e.g. inode number)
 367  * @bh: buffer head of the buffer storing the entry block
 368  * @kaddr: kernel address mapped for the page including the buffer
 369  */
 370 void *nilfs_palloc_block_get_entry(const struct inode *inode, __u64 nr,
 371                                    const struct buffer_head *bh, void *kaddr)
 372 {
 373         unsigned long entry_offset, group_offset;
 374
 375         nilfs_palloc_group(inode, nr, &group_offset);
 376         entry_offset = group_offset % NILFS_MDT(inode)->mi_entries_per_block;
 377
 378         return kaddr + bh_offset(bh) +
 379                 entry_offset * NILFS_MDT(inode)->mi_entry_size;
 380 }
 381
 382 /**
 383  * nilfs_palloc_find_available_slot - find available slot in a group
 384  * @bitmap: bitmap of the group
 385  * @target: offset number of an entry in the group (start point)
 386  * @bsize: size in bits
 387  * @lock: spin lock protecting @bitmap
 388  */
 389 static int nilfs_palloc_find_available_slot(unsigned char *bitmap,
 390                                             unsigned long target,
 391                                             unsigned int bsize,
 392                                             spinlock_t *lock)
 393 {
 394         int pos, end = bsize;
 395
 396         if (likely(target < bsize)) {
 397                 pos = target;
 398                 do {
 399                         pos = nilfs_find_next_zero_bit(bitmap, end, pos);
 400                         if (pos >= end)
 401                                 break;
 402                         if (!nilfs_set_bit_atomic(lock, pos, bitmap))
 403                                 return pos;
 404                 } while (++pos < end);
 405
 406                 end = target;
 407         }
 408
 409         /* wrap around */
 410         for (pos = 0; pos < end; pos++) {
 411                 pos = nilfs_find_next_zero_bit(bitmap, end, pos);
 412                 if (pos >= end)
 413                         break;
 414                 if (!nilfs_set_bit_atomic(lock, pos, bitmap))
 415                         return pos;
 416         }
 417
 418         return -ENOSPC;
 419 }
 420
 421 /**
 422  * nilfs_palloc_rest_groups_in_desc_block - get the remaining number of groups
 423  *                                          in a group descriptor block
 424  * @inode: inode of metadata file using this allocator
 425  * @curr: current group number
 426  * @max: maximum number of groups
 427  */
 428 static unsigned long
 429 nilfs_palloc_rest_groups_in_desc_block(const struct inode *inode,
 430                                        unsigned long curr, unsigned long max)
 431 {
 432         return min_t(unsigned long,
 433                      nilfs_palloc_groups_per_desc_block(inode) -
 434                      curr % nilfs_palloc_groups_per_desc_block(inode),
 435                      max - curr + 1);
 436 }
 437
 438 /**
 439  * nilfs_palloc_count_desc_blocks - count descriptor blocks number
 440  * @inode: inode of metadata file using this allocator
 441  * @desc_blocks: descriptor blocks number [out]
 442  */
 443 static int nilfs_palloc_count_desc_blocks(struct inode *inode,
 444                                             unsigned long *desc_blocks)
 445 {
 446         __u64 blknum;
 447         int ret;
 448
 449         ret = nilfs_bmap_last_key(NILFS_I(inode)->i_bmap, &blknum);
 450         if (likely(!ret))
 451                 *desc_blocks = DIV_ROUND_UP(
 452                         (unsigned long)blknum,
 453                         NILFS_MDT(inode)->mi_blocks_per_desc_block);
 454         return ret;
 455 }
 456
 457 /**
 458  * nilfs_palloc_mdt_file_can_grow - check potential opportunity for
 459  *                                      MDT file growing
 460  * @inode: inode of metadata file using this allocator
 461  * @desc_blocks: known current descriptor blocks count
 462  */
 463 static inline bool nilfs_palloc_mdt_file_can_grow(struct inode *inode,
 464                                                     unsigned long desc_blocks)
 465 {
 466         return (nilfs_palloc_groups_per_desc_block(inode) * desc_blocks) <
 467                         nilfs_palloc_groups_count(inode);
 468 }
 469
 470 /**
 471  * nilfs_palloc_count_max_entries - count max number of entries that can be
 472  *                                      described by descriptor blocks count
 473  * @inode: inode of metadata file using this allocator
 474  * @nused: current number of used entries
 475  * @nmaxp: max number of entries [out]
 476  */
 477 int nilfs_palloc_count_max_entries(struct inode *inode, u64 nused, u64 *nmaxp)
 478 {
 479         unsigned long desc_blocks = 0;
 480         u64 entries_per_desc_block, nmax;
 481         int err;
 482
 483         err = nilfs_palloc_count_desc_blocks(inode, &desc_blocks);
 484         if (unlikely(err))
 485                 return err;
 486
 487         entries_per_desc_block = (u64)nilfs_palloc_entries_per_group(inode) *
 488                                 nilfs_palloc_groups_per_desc_block(inode);
 489         nmax = entries_per_desc_block * desc_blocks;
 490
 491         if (nused == nmax &&
 492                         nilfs_palloc_mdt_file_can_grow(inode, desc_blocks))
 493                 nmax += entries_per_desc_block;
 494
 495         if (nused > nmax)
 496                 return -ERANGE;
 497
 498         *nmaxp = nmax;
 499         return 0;
 500 }
 501
 502 /**
 503  * nilfs_palloc_prepare_alloc_entry - prepare to allocate a persistent object
 504  * @inode: inode of metadata file using this allocator
 505  * @req: nilfs_palloc_req structure exchanged for the allocation
 506  */
 507 int nilfs_palloc_prepare_alloc_entry(struct inode *inode,
 508                                      struct nilfs_palloc_req *req)
 509 {
 510         struct buffer_head *desc_bh, *bitmap_bh;
 511         struct nilfs_palloc_group_desc *desc;
 512         unsigned char *bitmap;
 513         void *desc_kaddr, *bitmap_kaddr;
 514         unsigned long group, maxgroup, ngroups;
 515         unsigned long group_offset, maxgroup_offset;
 516         unsigned long n, entries_per_group;
 517         unsigned long i, j;
 518         spinlock_t *lock;
 519         int pos, ret;
 520
 521         ngroups = nilfs_palloc_groups_count(inode);
 522         maxgroup = ngroups - 1;
 523         group = nilfs_palloc_group(inode, req->pr_entry_nr, &group_offset);
 524         entries_per_group = nilfs_palloc_entries_per_group(inode);
 525
 526         for (i = 0; i < ngroups; i += n) {
 527                 if (group >= ngroups) {
 528                         /* wrap around */
 529                         group = 0;
 530                         maxgroup = nilfs_palloc_group(inode, req->pr_entry_nr,
 531                                                       &maxgroup_offset) - 1;
 532                 }
 533                 ret = nilfs_palloc_get_desc_block(inode, group, 1, &desc_bh);
 534                 if (ret < 0)
 535                         return ret;
 536                 desc_kaddr = kmap(desc_bh->b_page);
 537                 desc = nilfs_palloc_block_get_group_desc(
 538                         inode, group, desc_bh, desc_kaddr);
 539                 n = nilfs_palloc_rest_groups_in_desc_block(inode, group,
 540                                                            maxgroup);
 541                 for (j = 0; j < n; j++, desc++, group++) {
 542                         lock = nilfs_mdt_bgl_lock(inode, group);
 543                         if (nilfs_palloc_group_desc_nfrees(desc, lock) > 0) {
 544                                 ret = nilfs_palloc_get_bitmap_block(
 545                                         inode, group, 1, &bitmap_bh);
 546                                 if (ret < 0)
 547                                         goto out_desc;
 548                                 bitmap_kaddr = kmap(bitmap_bh->b_page);
 549                                 bitmap = bitmap_kaddr + bh_offset(bitmap_bh);
 550                                 pos = nilfs_palloc_find_available_slot(
 551                                         bitmap, group_offset,
 552                                         entries_per_group, lock);
 553                                 if (pos >= 0) {
 554                                         /* found a free entry */
 555                                         nilfs_palloc_group_desc_add_entries(
 556                                                 desc, lock, -1);
 557                                         req->pr_entry_nr =
 558                                                 entries_per_group * group + pos;
 559                                         kunmap(desc_bh->b_page);
 560                                         kunmap(bitmap_bh->b_page);
 561
 562                                         req->pr_desc_bh = desc_bh;
 563                                         req->pr_bitmap_bh = bitmap_bh;
 564                                         return 0;
 565                                 }
 566                                 kunmap(bitmap_bh->b_page);
 567                                 brelse(bitmap_bh);
 568                         }
 569
 570                         group_offset = 0;
 571                 }
 572
 573                 kunmap(desc_bh->b_page);
 574                 brelse(desc_bh);
 575         }
 576
 577         /* no entries left */
 578         return -ENOSPC;
 579
 580  out_desc:
 581         kunmap(desc_bh->b_page);
 582         brelse(desc_bh);
 583         return ret;
 584 }
 585
 586 /**
 587  * nilfs_palloc_commit_alloc_entry - finish allocation of a persistent object
 588  * @inode: inode of metadata file using this allocator
 589  * @req: nilfs_palloc_req structure exchanged for the allocation
 590  */
 591 void nilfs_palloc_commit_alloc_entry(struct inode *inode,
 592                                      struct nilfs_palloc_req *req)
 593 {
 594         mark_buffer_dirty(req->pr_bitmap_bh);
 595         mark_buffer_dirty(req->pr_desc_bh);
 596         nilfs_mdt_mark_dirty(inode);
 597
 598         brelse(req->pr_bitmap_bh);
 599         brelse(req->pr_desc_bh);
 600 }
 601
 602 /**
 603  * nilfs_palloc_commit_free_entry - finish deallocating a persistent object
 604  * @inode: inode of metadata file using this allocator
 605  * @req: nilfs_palloc_req structure exchanged for the removal
 606  */
 607 void nilfs_palloc_commit_free_entry(struct inode *inode,
 608                                     struct nilfs_palloc_req *req)
 609 {
 610         struct nilfs_palloc_group_desc *desc;
 611         unsigned long group, group_offset;
 612         unsigned char *bitmap;
 613         void *desc_kaddr, *bitmap_kaddr;
 614         spinlock_t *lock;
 615
 616         group = nilfs_palloc_group(inode, req->pr_entry_nr, &group_offset);
 617         desc_kaddr = kmap(req->pr_desc_bh->b_page);
 618         desc = nilfs_palloc_block_get_group_desc(inode, group,
 619                                                  req->pr_desc_bh, desc_kaddr);
 620         bitmap_kaddr = kmap(req->pr_bitmap_bh->b_page);
 621         bitmap = bitmap_kaddr + bh_offset(req->pr_bitmap_bh);
 622         lock = nilfs_mdt_bgl_lock(inode, group);
 623
 624         if (!nilfs_clear_bit_atomic(lock, group_offset, bitmap))
 625                 nilfs_msg(inode->i_sb, KERN_WARNING,
 626                           "%s (ino=%lu): entry number %llu already freed",
 627                           __func__, inode->i_ino,
 628                           (unsigned long long)req->pr_entry_nr);
 629         else
 630                 nilfs_palloc_group_desc_add_entries(desc, lock, 1);
 631
 632         kunmap(req->pr_bitmap_bh->b_page);
 633         kunmap(req->pr_desc_bh->b_page);
 634
 635         mark_buffer_dirty(req->pr_desc_bh);
 636         mark_buffer_dirty(req->pr_bitmap_bh);
 637         nilfs_mdt_mark_dirty(inode);
 638
 639         brelse(req->pr_bitmap_bh);
 640         brelse(req->pr_desc_bh);
 641 }
 642
 643 /**
 644  * nilfs_palloc_abort_alloc_entry - cancel allocation of a persistent object
 645  * @inode: inode of metadata file using this allocator
 646  * @req: nilfs_palloc_req structure exchanged for the allocation
 647  */
 648 void nilfs_palloc_abort_alloc_entry(struct inode *inode,
 649                                     struct nilfs_palloc_req *req)
 650 {
 651         struct nilfs_palloc_group_desc *desc;
 652         void *desc_kaddr, *bitmap_kaddr;
 653         unsigned char *bitmap;
 654         unsigned long group, group_offset;
 655         spinlock_t *lock;
 656
 657         group = nilfs_palloc_group(inode, req->pr_entry_nr, &group_offset);
 658         desc_kaddr = kmap(req->pr_desc_bh->b_page);
 659         desc = nilfs_palloc_block_get_group_desc(inode, group,
 660                                                  req->pr_desc_bh, desc_kaddr);
 661         bitmap_kaddr = kmap(req->pr_bitmap_bh->b_page);
 662         bitmap = bitmap_kaddr + bh_offset(req->pr_bitmap_bh);
 663         lock = nilfs_mdt_bgl_lock(inode, group);
 664
 665         if (!nilfs_clear_bit_atomic(lock, group_offset, bitmap))
 666                 nilfs_msg(inode->i_sb, KERN_WARNING,
 667                           "%s (ino=%lu): entry number %llu already freed",
 668                           __func__, inode->i_ino,
 669                           (unsigned long long)req->pr_entry_nr);
 670         else
 671                 nilfs_palloc_group_desc_add_entries(desc, lock, 1);
 672
 673         kunmap(req->pr_bitmap_bh->b_page);
 674         kunmap(req->pr_desc_bh->b_page);
 675
 676         brelse(req->pr_bitmap_bh);
 677         brelse(req->pr_desc_bh);
 678
 679         req->pr_entry_nr = 0;
 680         req->pr_bitmap_bh = NULL;
 681         req->pr_desc_bh = NULL;
 682 }
 683
 684 /**
 685  * nilfs_palloc_prepare_free_entry - prepare to deallocate a persistent object
 686  * @inode: inode of metadata file using this allocator
 687  * @req: nilfs_palloc_req structure exchanged for the removal
 688  */
 689 int nilfs_palloc_prepare_free_entry(struct inode *inode,
 690                                     struct nilfs_palloc_req *req)
 691 {
 692         struct buffer_head *desc_bh, *bitmap_bh;
 693         unsigned long group, group_offset;
 694         int ret;
 695
 696         group = nilfs_palloc_group(inode, req->pr_entry_nr, &group_offset);
 697         ret = nilfs_palloc_get_desc_block(inode, group, 1, &desc_bh);
 698         if (ret < 0)
 699                 return ret;
 700         ret = nilfs_palloc_get_bitmap_block(inode, group, 1, &bitmap_bh);
 701         if (ret < 0) {
 702                 brelse(desc_bh);
 703                 return ret;
 704         }
 705
 706         req->pr_desc_bh = desc_bh;
 707         req->pr_bitmap_bh = bitmap_bh;
 708         return 0;
 709 }
 710
 711 /**
 712  * nilfs_palloc_abort_free_entry - cancel deallocating a persistent object
 713  * @inode: inode of metadata file using this allocator
 714  * @req: nilfs_palloc_req structure exchanged for the removal
 715  */
 716 void nilfs_palloc_abort_free_entry(struct inode *inode,
 717                                    struct nilfs_palloc_req *req)
 718 {
 719         brelse(req->pr_bitmap_bh);
 720         brelse(req->pr_desc_bh);
 721
 722         req->pr_entry_nr = 0;
 723         req->pr_bitmap_bh = NULL;
 724         req->pr_desc_bh = NULL;
 725 }
 726
 727 /**
 728  * nilfs_palloc_freev - deallocate a set of persistent objects
 729  * @inode: inode of metadata file using this allocator
 730  * @entry_nrs: array of entry numbers to be deallocated
 731  * @nitems: number of entries stored in @entry_nrs
 732  */
 733 int nilfs_palloc_freev(struct inode *inode, __u64 *entry_nrs, size_t nitems)
 734 {
 735         struct buffer_head *desc_bh, *bitmap_bh;
 736         struct nilfs_palloc_group_desc *desc;
 737         unsigned char *bitmap;
 738         void *desc_kaddr, *bitmap_kaddr;
 739         unsigned long group, group_offset;
 740         __u64 group_min_nr, last_nrs[8];
 741         const unsigned long epg = nilfs_palloc_entries_per_group(inode);
 742         const unsigned int epb = NILFS_MDT(inode)->mi_entries_per_block;
 743         unsigned int entry_start, end, pos;
 744         spinlock_t *lock;
 745         int i, j, k, ret;
 746         u32 nfree;
 747
 748         for (i = 0; i < nitems; i = j) {
 749                 int change_group = false;
 750                 int nempties = 0, n = 0;
 751
 752                 group = nilfs_palloc_group(inode, entry_nrs[i], &group_offset);
 753                 ret = nilfs_palloc_get_desc_block(inode, group, 0, &desc_bh);
 754                 if (ret < 0)
 755                         return ret;
 756                 ret = nilfs_palloc_get_bitmap_block(inode, group, 0,
 757                                                     &bitmap_bh);
 758                 if (ret < 0) {
 759                         brelse(desc_bh);
 760                         return ret;
 761                 }
 762
 763                 /* Get the first entry number of the group */
 764                 group_min_nr = (__u64)group * epg;
 765
 766                 bitmap_kaddr = kmap(bitmap_bh->b_page);
 767                 bitmap = bitmap_kaddr + bh_offset(bitmap_bh);
 768                 lock = nilfs_mdt_bgl_lock(inode, group);
 769
 770                 j = i;
 771                 entry_start = rounddown(group_offset, epb);
 772                 do {
 773                         if (!nilfs_clear_bit_atomic(lock, group_offset,
 774                                                     bitmap)) {
 775                                 nilfs_msg(inode->i_sb, KERN_WARNING,
 776                                           "%s (ino=%lu): entry number %llu already freed",
 777                                           __func__, inode->i_ino,
 778                                           (unsigned long long)entry_nrs[j]);
 779                         } else {
 780                                 n++;
 781                         }
 782
 783                         j++;
 784                         if (j >= nitems || entry_nrs[j] < group_min_nr ||
 785                             entry_nrs[j] >= group_min_nr + epg) {
 786                                 change_group = true;
 787                         } else {
 788                                 group_offset = entry_nrs[j] - group_min_nr;
 789                                 if (group_offset >= entry_start &&
 790                                     group_offset < entry_start + epb) {
 791                                         /* This entry is in the same block */
 792                                         continue;
 793                                 }
 794                         }
 795
 796                         /* Test if the entry block is empty or not */
 797                         end = entry_start + epb;
 798                         pos = nilfs_find_next_bit(bitmap, end, entry_start);
 799                         if (pos >= end) {
 800                                 last_nrs[nempties++] = entry_nrs[j - 1];
 801                                 if (nempties >= ARRAY_SIZE(last_nrs))
 802                                         break;
 803                         }
 804
 805                         if (change_group)
 806                                 break;
 807
 808                         /* Go on to the next entry block */
 809                         entry_start = rounddown(group_offset, epb);
 810                 } while (true);
 811
 812                 kunmap(bitmap_bh->b_page);
 813                 mark_buffer_dirty(bitmap_bh);
 814                 brelse(bitmap_bh);
 815
 816                 for (k = 0; k < nempties; k++) {
 817                         ret = nilfs_palloc_delete_entry_block(inode,
 818                                                               last_nrs[k]);
 819                         if (ret && ret != -ENOENT)
 820                                 nilfs_msg(inode->i_sb, KERN_WARNING,
 821                                           "error %d deleting block that object (entry=%llu, ino=%lu) belongs to",
 822                                           ret, (unsigned long long)last_nrs[k],
 823                                           inode->i_ino);
 824                 }
 825
 826                 desc_kaddr = kmap_atomic(desc_bh->b_page);
 827                 desc = nilfs_palloc_block_get_group_desc(
 828                         inode, group, desc_bh, desc_kaddr);
 829                 nfree = nilfs_palloc_group_desc_add_entries(desc, lock, n);
 830                 kunmap_atomic(desc_kaddr);
 831                 mark_buffer_dirty(desc_bh);
 832                 nilfs_mdt_mark_dirty(inode);
 833                 brelse(desc_bh);
 834
 835                 if (nfree == nilfs_palloc_entries_per_group(inode)) {
 836                         ret = nilfs_palloc_delete_bitmap_block(inode, group);
 837                         if (ret && ret != -ENOENT)
 838                                 nilfs_msg(inode->i_sb, KERN_WARNING,
 839                                           "error %d deleting bitmap block of group=%lu, ino=%lu",
 840                                           ret, group, inode->i_ino);
 841                 }
 842         }
 843         return 0;
 844 }
 845
 846 void nilfs_palloc_setup_cache(struct inode *inode,
 847                               struct nilfs_palloc_cache *cache)
 848 {
 849         NILFS_MDT(inode)->mi_palloc_cache = cache;
 850         spin_lock_init(&cache->lock);
 851 }
 852
 853 void nilfs_palloc_clear_cache(struct inode *inode)
 854 {
 855         struct nilfs_palloc_cache *cache = NILFS_MDT(inode)->mi_palloc_cache;
 856
 857         spin_lock(&cache->lock);
 858         brelse(cache->prev_desc.bh);
 859         brelse(cache->prev_bitmap.bh);
 860         brelse(cache->prev_entry.bh);
 861         cache->prev_desc.bh = NULL;
 862         cache->prev_bitmap.bh = NULL;
 863         cache->prev_entry.bh = NULL;
 864         spin_unlock(&cache->lock);
 865 }
 866
 867 void nilfs_palloc_destroy_cache(struct inode *inode)
 868 {
 869         nilfs_palloc_clear_cache(inode);
 870         NILFS_MDT(inode)->mi_palloc_cache = NULL;
 871 }