fs/jfs/jfs_metapage.c

   1 /*
   2  *   Copyright (C) International Business Machines Corp., 2000-2005
   3  *   Portions Copyright (C) Christoph Hellwig, 2001-2002
   4  *
   5  *   This program is free software;  you can redistribute it and/or modify
   6  *   it under the terms of the GNU General Public License as published by
   7  *   the Free Software Foundation; either version 2 of the License, or
   8  *   (at your option) any later version.
   9  *
  10  *   This program is distributed in the hope that it will be useful,
  11  *   but WITHOUT ANY WARRANTY;  without even the implied warranty of
  12  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See
  13  *   the GNU General Public License for more details.
  14  *
  15  *   You should have received a copy of the GNU General Public License
  16  *   along with this program;  if not, write to the Free Software
  17  *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  18  */
  19
  20 #include <linux/fs.h>
  21 #include <linux/mm.h>
  22 #include <linux/bio.h>
  23 #include <linux/init.h>
  24 #include <linux/buffer_head.h>
  25 #include <linux/mempool.h>
  26 #include "jfs_incore.h"
  27 #include "jfs_superblock.h"
  28 #include "jfs_filsys.h"
  29 #include "jfs_metapage.h"
  30 #include "jfs_txnmgr.h"
  31 #include "jfs_debug.h"
  32
  33 #ifdef CONFIG_JFS_STATISTICS
  34 static struct {
  35         uint    pagealloc;      /* # of page allocations */
  36         uint    pagefree;       /* # of page frees */
  37         uint    lockwait;       /* # of sleeping lock_metapage() calls */
  38 } mpStat;
  39 #endif
  40
  41 #define metapage_locked(mp) test_bit(META_locked, &(mp)->flag)
  42 #define trylock_metapage(mp) test_and_set_bit(META_locked, &(mp)->flag)
  43
  44 static inline void unlock_metapage(struct metapage *mp)
  45 {
  46         clear_bit(META_locked, &mp->flag);
  47         wake_up(&mp->wait);
  48 }
  49
  50 static inline void __lock_metapage(struct metapage *mp)
  51 {
  52         DECLARE_WAITQUEUE(wait, current);
  53         INCREMENT(mpStat.lockwait);
  54         add_wait_queue_exclusive(&mp->wait, &wait);
  55         do {
  56                 set_current_state(TASK_UNINTERRUPTIBLE);
  57                 if (metapage_locked(mp)) {
  58                         unlock_page(mp->page);
  59                         schedule();
  60                         lock_page(mp->page);
  61                 }
  62         } while (trylock_metapage(mp));
  63         __set_current_state(TASK_RUNNING);
  64         remove_wait_queue(&mp->wait, &wait);
  65 }
  66
  67 /*
  68  * Must have mp->page locked
  69  */
  70 static inline void lock_metapage(struct metapage *mp)
  71 {
  72         if (trylock_metapage(mp))
  73                 __lock_metapage(mp);
  74 }
  75
  76 #define METAPOOL_MIN_PAGES 32
  77 static kmem_cache_t *metapage_cache;
  78 static mempool_t *metapage_mempool;
  79
  80 #define MPS_PER_PAGE (PAGE_CACHE_SIZE >> L2PSIZE)
  81
  82 #if MPS_PER_PAGE > 1
  83
  84 struct meta_anchor {
  85         int mp_count;
  86         atomic_t io_count;
  87         struct metapage *mp[MPS_PER_PAGE];
  88 };
  89 #define mp_anchor(page) ((struct meta_anchor *)page_private(page))
  90
  91 static inline struct metapage *page_to_mp(struct page *page, uint offset)
  92 {
  93         if (!PagePrivate(page))
  94                 return NULL;
  95         return mp_anchor(page)->mp[offset >> L2PSIZE];
  96 }
  97
  98 static inline int insert_metapage(struct page *page, struct metapage *mp)
  99 {
 100         struct meta_anchor *a;
 101         int index;
 102         int l2mp_blocks;        /* log2 blocks per metapage */
 103
 104         if (PagePrivate(page))
 105                 a = mp_anchor(page);
 106         else {
 107                 a = kzalloc(sizeof(struct meta_anchor), GFP_NOFS);
 108                 if (!a)
 109                         return -ENOMEM;
 110                 set_page_private(page, (unsigned long)a);
 111                 SetPagePrivate(page);
 112                 kmap(page);
 113         }
 114
 115         if (mp) {
 116                 l2mp_blocks = L2PSIZE - page->mapping->host->i_blkbits;
 117                 index = (mp->index >> l2mp_blocks) & (MPS_PER_PAGE - 1);
 118                 a->mp_count++;
 119                 a->mp[index] = mp;
 120         }
 121
 122         return 0;
 123 }
 124
 125 static inline void remove_metapage(struct page *page, struct metapage *mp)
 126 {
 127         struct meta_anchor *a = mp_anchor(page);
 128         int l2mp_blocks = L2PSIZE - page->mapping->host->i_blkbits;
 129         int index;
 130
 131         index = (mp->index >> l2mp_blocks) & (MPS_PER_PAGE - 1);
 132
 133         BUG_ON(a->mp[index] != mp);
 134
 135         a->mp[index] = NULL;
 136         if (--a->mp_count == 0) {
 137                 kfree(a);
 138                 set_page_private(page, 0);
 139                 ClearPagePrivate(page);
 140                 kunmap(page);
 141         }
 142 }
 143
 144 static inline void inc_io(struct page *page)
 145 {
 146         atomic_inc(&mp_anchor(page)->io_count);
 147 }
 148
 149 static inline void dec_io(struct page *page, void (*handler) (struct page *))
 150 {
 151         if (atomic_dec_and_test(&mp_anchor(page)->io_count))
 152                 handler(page);
 153 }
 154
 155 #else
 156 static inline struct metapage *page_to_mp(struct page *page, uint offset)
 157 {
 158         return PagePrivate(page) ? (struct metapage *)page_private(page) : NULL;
 159 }
 160
 161 static inline int insert_metapage(struct page *page, struct metapage *mp)
 162 {
 163         if (mp) {
 164                 set_page_private(page, (unsigned long)mp);
 165                 SetPagePrivate(page);
 166                 kmap(page);
 167         }
 168         return 0;
 169 }
 170
 171 static inline void remove_metapage(struct page *page, struct metapage *mp)
 172 {
 173         set_page_private(page, 0);
 174         ClearPagePrivate(page);
 175         kunmap(page);
 176 }
 177
 178 #define inc_io(page) do {} while(0)
 179 #define dec_io(page, handler) handler(page)
 180
 181 #endif
 182
 183 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
 184 {
 185         struct metapage *mp = (struct metapage *)foo;
 186
 187         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 188             SLAB_CTOR_CONSTRUCTOR) {
 189                 mp->lid = 0;
 190                 mp->lsn = 0;
 191                 mp->flag = 0;
 192                 mp->data = NULL;
 193                 mp->clsn = 0;
 194                 mp->log = NULL;
 195                 set_bit(META_free, &mp->flag);
 196                 init_waitqueue_head(&mp->wait);
 197         }
 198 }
 199
 200 static inline struct metapage *alloc_metapage(gfp_t gfp_mask)
 201 {
 202         return mempool_alloc(metapage_mempool, gfp_mask);
 203 }
 204
 205 static inline void free_metapage(struct metapage *mp)
 206 {
 207         mp->flag = 0;
 208         set_bit(META_free, &mp->flag);
 209
 210         mempool_free(mp, metapage_mempool);
 211 }
 212
 213 int __init metapage_init(void)
 214 {
 215         /*
 216          * Allocate the metapage structures
 217          */
 218         metapage_cache = kmem_cache_create("jfs_mp", sizeof(struct metapage),
 219                                            0, 0, init_once, NULL);
 220         if (metapage_cache == NULL)
 221                 return -ENOMEM;
 222
 223         metapage_mempool = mempool_create(METAPOOL_MIN_PAGES, mempool_alloc_slab,
 224                                           mempool_free_slab, metapage_cache);
 225
 226         if (metapage_mempool == NULL) {
 227                 kmem_cache_destroy(metapage_cache);
 228                 return -ENOMEM;
 229         }
 230
 231         return 0;
 232 }
 233
 234 void metapage_exit(void)
 235 {
 236         mempool_destroy(metapage_mempool);
 237         kmem_cache_destroy(metapage_cache);
 238 }
 239
 240 static inline void drop_metapage(struct page *page, struct metapage *mp)
 241 {
 242         if (mp->count || mp->nohomeok || test_bit(META_dirty, &mp->flag) ||
 243             test_bit(META_io, &mp->flag))
 244                 return;
 245         remove_metapage(page, mp);
 246         INCREMENT(mpStat.pagefree);
 247         free_metapage(mp);
 248 }
 249
 250 /*
 251  * Metapage address space operations
 252  */
 253
 254 static sector_t metapage_get_blocks(struct inode *inode, sector_t lblock,
 255                                     unsigned int *len)
 256 {
 257         int rc = 0;
 258         int xflag;
 259         s64 xaddr;
 260         sector_t file_blocks = (inode->i_size + inode->i_blksize - 1) >>
 261                                inode->i_blkbits;
 262
 263         if (lblock >= file_blocks)
 264                 return 0;
 265         if (lblock + *len > file_blocks)
 266                 *len = file_blocks - lblock;
 267
 268         if (inode->i_ino) {
 269                 rc = xtLookup(inode, (s64)lblock, *len, &xflag, &xaddr, len, 0);
 270                 if ((rc == 0) && *len)
 271                         lblock = (sector_t)xaddr;
 272                 else
 273                         lblock = 0;
 274         } /* else no mapping */
 275
 276         return lblock;
 277 }
 278
 279 static void last_read_complete(struct page *page)
 280 {
 281         if (!PageError(page))
 282                 SetPageUptodate(page);
 283         unlock_page(page);
 284 }
 285
 286 static int metapage_read_end_io(struct bio *bio, unsigned int bytes_done,
 287                                 int err)
 288 {
 289         struct page *page = bio->bi_private;
 290
 291         if (bio->bi_size)
 292                 return 1;
 293
 294         if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
 295                 printk(KERN_ERR "metapage_read_end_io: I/O error\n");
 296                 SetPageError(page);
 297         }
 298
 299         dec_io(page, last_read_complete);
 300         bio_put(bio);
 301
 302         return 0;
 303 }
 304
 305 static void remove_from_logsync(struct metapage *mp)
 306 {
 307         struct jfs_log *log = mp->log;
 308         unsigned long flags;
 309 /*
 310  * This can race.  Recheck that log hasn't been set to null, and after
 311  * acquiring logsync lock, recheck lsn
 312  */
 313         if (!log)
 314                 return;
 315
 316         LOGSYNC_LOCK(log, flags);
 317         if (mp->lsn) {
 318                 mp->log = NULL;
 319                 mp->lsn = 0;
 320                 mp->clsn = 0;
 321                 log->count--;
 322                 list_del(&mp->synclist);
 323         }
 324         LOGSYNC_UNLOCK(log, flags);
 325 }
 326
 327 static void last_write_complete(struct page *page)
 328 {
 329         struct metapage *mp;
 330         unsigned int offset;
 331
 332         for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
 333                 mp = page_to_mp(page, offset);
 334                 if (mp && test_bit(META_io, &mp->flag)) {
 335                         if (mp->lsn)
 336                                 remove_from_logsync(mp);
 337                         clear_bit(META_io, &mp->flag);
 338                 }
 339                 /*
 340                  * I'd like to call drop_metapage here, but I don't think it's
 341                  * safe unless I have the page locked
 342                  */
 343         }
 344         end_page_writeback(page);
 345 }
 346
 347 static int metapage_write_end_io(struct bio *bio, unsigned int bytes_done,
 348                                  int err)
 349 {
 350         struct page *page = bio->bi_private;
 351
 352         BUG_ON(!PagePrivate(page));
 353
 354         if (bio->bi_size)
 355                 return 1;
 356
 357         if (! test_bit(BIO_UPTODATE, &bio->bi_flags)) {
 358                 printk(KERN_ERR "metapage_write_end_io: I/O error\n");
 359                 SetPageError(page);
 360         }
 361         dec_io(page, last_write_complete);
 362         bio_put(bio);
 363         return 0;
 364 }
 365
 366 static int metapage_writepage(struct page *page, struct writeback_control *wbc)
 367 {
 368         struct bio *bio = NULL;
 369         unsigned int block_offset;      /* block offset of mp within page */
 370         struct inode *inode = page->mapping->host;
 371         unsigned int blocks_per_mp = JFS_SBI(inode->i_sb)->nbperpage;
 372         unsigned int len;
 373         unsigned int xlen;
 374         struct metapage *mp;
 375         int redirty = 0;
 376         sector_t lblock;
 377         sector_t pblock;
 378         sector_t next_block = 0;
 379         sector_t page_start;
 380         unsigned long bio_bytes = 0;
 381         unsigned long bio_offset = 0;
 382         unsigned int offset;
 383
 384         page_start = (sector_t)page->index <<
 385                      (PAGE_CACHE_SHIFT - inode->i_blkbits);
 386         BUG_ON(!PageLocked(page));
 387         BUG_ON(PageWriteback(page));
 388
 389         for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
 390                 mp = page_to_mp(page, offset);
 391
 392                 if (!mp || !test_bit(META_dirty, &mp->flag))
 393                         continue;
 394
 395                 if (mp->nohomeok && !test_bit(META_forcewrite, &mp->flag)) {
 396                         redirty = 1;
 397                         /*
 398                          * Make sure this page isn't blocked indefinitely.
 399                          * If the journal isn't undergoing I/O, push it
 400                          */
 401                         if (mp->log && !(mp->log->cflag & logGC_PAGEOUT))
 402                                 jfs_flush_journal(mp->log, 0);
 403                         continue;
 404                 }
 405
 406                 clear_bit(META_dirty, &mp->flag);
 407                 block_offset = offset >> inode->i_blkbits;
 408                 lblock = page_start + block_offset;
 409                 if (bio) {
 410                         if (xlen && lblock == next_block) {
 411                                 /* Contiguous, in memory & on disk */
 412                                 len = min(xlen, blocks_per_mp);
 413                                 xlen -= len;
 414                                 bio_bytes += len << inode->i_blkbits;
 415                                 set_bit(META_io, &mp->flag);
 416                                 continue;
 417                         }
 418                         /* Not contiguous */
 419                         if (bio_add_page(bio, page, bio_bytes, bio_offset) <
 420                             bio_bytes)
 421                                 goto add_failed;
 422                         /*
 423                          * Increment counter before submitting i/o to keep
 424                          * count from hitting zero before we're through
 425                          */
 426                         inc_io(page);
 427                         if (!bio->bi_size)
 428                                 goto dump_bio;
 429                         submit_bio(WRITE, bio);
 430                         bio = NULL;
 431                 } else {
 432                         set_page_writeback(page);
 433                         inc_io(page);
 434                 }
 435                 xlen = (PAGE_CACHE_SIZE - offset) >> inode->i_blkbits;
 436                 pblock = metapage_get_blocks(inode, lblock, &xlen);
 437                 if (!pblock) {
 438                         /* Need better error handling */
 439                         printk(KERN_ERR "JFS: metapage_get_blocks failed\n");
 440                         dec_io(page, last_write_complete);
 441                         continue;
 442                 }
 443                 set_bit(META_io, &mp->flag);
 444                 len = min(xlen, (uint) JFS_SBI(inode->i_sb)->nbperpage);
 445
 446                 bio = bio_alloc(GFP_NOFS, 1);
 447                 bio->bi_bdev = inode->i_sb->s_bdev;
 448                 bio->bi_sector = pblock << (inode->i_blkbits - 9);
 449                 bio->bi_end_io = metapage_write_end_io;
 450                 bio->bi_private = page;
 451
 452                 /* Don't call bio_add_page yet, we may add to this vec */
 453                 bio_offset = offset;
 454                 bio_bytes = len << inode->i_blkbits;
 455
 456                 xlen -= len;
 457                 next_block = lblock + len;
 458         }
 459         if (bio) {
 460                 if (bio_add_page(bio, page, bio_bytes, bio_offset) < bio_bytes)
 461                                 goto add_failed;
 462                 if (!bio->bi_size)
 463                         goto dump_bio;
 464
 465                 submit_bio(WRITE, bio);
 466         }
 467         if (redirty)
 468                 redirty_page_for_writepage(wbc, page);
 469
 470         unlock_page(page);
 471
 472         return 0;
 473 add_failed:
 474         /* We should never reach here, since we're only adding one vec */
 475         printk(KERN_ERR "JFS: bio_add_page failed unexpectedly\n");
 476         goto skip;
 477 dump_bio:
 478         dump_mem("bio", bio, sizeof(*bio));
 479 skip:
 480         bio_put(bio);
 481         unlock_page(page);
 482         dec_io(page, last_write_complete);
 483
 484         return -EIO;
 485 }
 486
 487 static int metapage_readpage(struct file *fp, struct page *page)
 488 {
 489         struct inode *inode = page->mapping->host;
 490         struct bio *bio = NULL;
 491         unsigned int block_offset;
 492         unsigned int blocks_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits;
 493         sector_t page_start;    /* address of page in fs blocks */
 494         sector_t pblock;
 495         unsigned int xlen;
 496         unsigned int len;
 497         unsigned int offset;
 498
 499         BUG_ON(!PageLocked(page));
 500         page_start = (sector_t)page->index <<
 501                      (PAGE_CACHE_SHIFT - inode->i_blkbits);
 502
 503         block_offset = 0;
 504         while (block_offset < blocks_per_page) {
 505                 xlen = blocks_per_page - block_offset;
 506                 pblock = metapage_get_blocks(inode, page_start + block_offset,
 507                                              &xlen);
 508                 if (pblock) {
 509                         if (!PagePrivate(page))
 510                                 insert_metapage(page, NULL);
 511                         inc_io(page);
 512                         if (bio)
 513                                 submit_bio(READ, bio);
 514
 515                         bio = bio_alloc(GFP_NOFS, 1);
 516                         bio->bi_bdev = inode->i_sb->s_bdev;
 517                         bio->bi_sector = pblock << (inode->i_blkbits - 9);
 518                         bio->bi_end_io = metapage_read_end_io;
 519                         bio->bi_private = page;
 520                         len = xlen << inode->i_blkbits;
 521                         offset = block_offset << inode->i_blkbits;
 522                         if (bio_add_page(bio, page, len, offset) < len)
 523                                 goto add_failed;
 524                         block_offset += xlen;
 525                 } else
 526                         block_offset++;
 527         }
 528         if (bio)
 529                 submit_bio(READ, bio);
 530         else
 531                 unlock_page(page);
 532
 533         return 0;
 534
 535 add_failed:
 536         printk(KERN_ERR "JFS: bio_add_page failed unexpectedly\n");
 537         bio_put(bio);
 538         dec_io(page, last_read_complete);
 539         return -EIO;
 540 }
 541
 542 static int metapage_releasepage(struct page *page, gfp_t gfp_mask)
 543 {
 544         struct metapage *mp;
 545         int busy = 0;
 546         unsigned int offset;
 547
 548         for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
 549                 mp = page_to_mp(page, offset);
 550
 551                 if (!mp)
 552                         continue;
 553
 554                 jfs_info("metapage_releasepage: mp = 0x%p", mp);
 555                 if (mp->count || mp->nohomeok) {
 556                         jfs_info("count = %ld, nohomeok = %d", mp->count,
 557                                  mp->nohomeok);
 558                         busy = 1;
 559                         continue;
 560                 }
 561                 wait_on_page_writeback(page);
 562                 //WARN_ON(test_bit(META_dirty, &mp->flag));
 563                 if (test_bit(META_dirty, &mp->flag)) {
 564                         dump_mem("dirty mp in metapage_releasepage", mp,
 565                                  sizeof(struct metapage));
 566                         dump_mem("page", page, sizeof(struct page));
 567                         dump_stack();
 568                 }
 569                 if (mp->lsn)
 570                         remove_from_logsync(mp);
 571                 remove_metapage(page, mp);
 572                 INCREMENT(mpStat.pagefree);
 573                 free_metapage(mp);
 574         }
 575         if (busy)
 576                 return -1;
 577
 578         return 0;
 579 }
 580
 581 static int metapage_invalidatepage(struct page *page, unsigned long offset)
 582 {
 583         BUG_ON(offset);
 584
 585         if (PageWriteback(page))
 586                 return 0;
 587
 588         return metapage_releasepage(page, 0);
 589 }
 590
 591 struct address_space_operations jfs_metapage_aops = {
 592         .readpage       = metapage_readpage,
 593         .writepage      = metapage_writepage,
 594         .sync_page      = block_sync_page,
 595         .releasepage    = metapage_releasepage,
 596         .invalidatepage = metapage_invalidatepage,
 597         .set_page_dirty = __set_page_dirty_nobuffers,
 598 };
 599
 600 struct metapage *__get_metapage(struct inode *inode, unsigned long lblock,
 601                                 unsigned int size, int absolute,
 602                                 unsigned long new)
 603 {
 604         int l2BlocksPerPage;
 605         int l2bsize;
 606         struct address_space *mapping;
 607         struct metapage *mp = NULL;
 608         struct page *page;
 609         unsigned long page_index;
 610         unsigned long page_offset;
 611
 612         jfs_info("__get_metapage: ino = %ld, lblock = 0x%lx, abs=%d",
 613                  inode->i_ino, lblock, absolute);
 614
 615         l2bsize = inode->i_blkbits;
 616         l2BlocksPerPage = PAGE_CACHE_SHIFT - l2bsize;
 617         page_index = lblock >> l2BlocksPerPage;
 618         page_offset = (lblock - (page_index << l2BlocksPerPage)) << l2bsize;
 619         if ((page_offset + size) > PAGE_CACHE_SIZE) {
 620                 jfs_err("MetaData crosses page boundary!!");
 621                 jfs_err("lblock = %lx, size  = %d", lblock, size);
 622                 dump_stack();
 623                 return NULL;
 624         }
 625         if (absolute)
 626                 mapping = JFS_SBI(inode->i_sb)->direct_inode->i_mapping;
 627         else {
 628                 /*
 629                  * If an nfs client tries to read an inode that is larger
 630                  * than any existing inodes, we may try to read past the
 631                  * end of the inode map
 632                  */
 633                 if ((lblock << inode->i_blkbits) >= inode->i_size)
 634                         return NULL;
 635                 mapping = inode->i_mapping;
 636         }
 637
 638         if (new && (PSIZE == PAGE_CACHE_SIZE)) {
 639                 page = grab_cache_page(mapping, page_index);
 640                 if (!page) {
 641                         jfs_err("grab_cache_page failed!");
 642                         return NULL;
 643                 }
 644                 SetPageUptodate(page);
 645         } else {
 646                 page = read_cache_page(mapping, page_index,
 647                             (filler_t *)mapping->a_ops->readpage, NULL);
 648                 if (IS_ERR(page) || !PageUptodate(page)) {
 649                         jfs_err("read_cache_page failed!");
 650                         return NULL;
 651                 }
 652                 lock_page(page);
 653         }
 654
 655         mp = page_to_mp(page, page_offset);
 656         if (mp) {
 657                 if (mp->logical_size != size) {
 658                         jfs_error(inode->i_sb,
 659                                   "__get_metapage: mp->logical_size != size");
 660                         jfs_err("logical_size = %d, size = %d",
 661                                 mp->logical_size, size);
 662                         dump_stack();
 663                         goto unlock;
 664                 }
 665                 mp->count++;
 666                 lock_metapage(mp);
 667                 if (test_bit(META_discard, &mp->flag)) {
 668                         if (!new) {
 669                                 jfs_error(inode->i_sb,
 670                                           "__get_metapage: using a "
 671                                           "discarded metapage");
 672                                 discard_metapage(mp);
 673                                 goto unlock;
 674                         }
 675                         clear_bit(META_discard, &mp->flag);
 676                 }
 677         } else {
 678                 INCREMENT(mpStat.pagealloc);
 679                 mp = alloc_metapage(GFP_NOFS);
 680                 mp->page = page;
 681                 mp->flag = 0;
 682                 mp->xflag = COMMIT_PAGE;
 683                 mp->count = 1;
 684                 mp->nohomeok = 0;
 685                 mp->logical_size = size;
 686                 mp->data = page_address(page) + page_offset;
 687                 mp->index = lblock;
 688                 if (unlikely(insert_metapage(page, mp))) {
 689                         free_metapage(mp);
 690                         goto unlock;
 691                 }
 692                 lock_metapage(mp);
 693         }
 694
 695         if (new) {
 696                 jfs_info("zeroing mp = 0x%p", mp);
 697                 memset(mp->data, 0, PSIZE);
 698         }
 699
 700         unlock_page(page);
 701         jfs_info("__get_metapage: returning = 0x%p data = 0x%p", mp, mp->data);
 702         return mp;
 703
 704 unlock:
 705         unlock_page(page);
 706         return NULL;
 707 }
 708
 709 void grab_metapage(struct metapage * mp)
 710 {
 711         jfs_info("grab_metapage: mp = 0x%p", mp);
 712         page_cache_get(mp->page);
 713         lock_page(mp->page);
 714         mp->count++;
 715         lock_metapage(mp);
 716         unlock_page(mp->page);
 717 }
 718
 719 void force_metapage(struct metapage *mp)
 720 {
 721         struct page *page = mp->page;
 722         jfs_info("force_metapage: mp = 0x%p", mp);
 723         set_bit(META_forcewrite, &mp->flag);
 724         clear_bit(META_sync, &mp->flag);
 725         page_cache_get(page);
 726         lock_page(page);
 727         set_page_dirty(page);
 728         write_one_page(page, 1);
 729         clear_bit(META_forcewrite, &mp->flag);
 730         page_cache_release(page);
 731 }
 732
 733 void hold_metapage(struct metapage *mp)
 734 {
 735         lock_page(mp->page);
 736 }
 737
 738 void put_metapage(struct metapage *mp)
 739 {
 740         if (mp->count || mp->nohomeok) {
 741                 /* Someone else will release this */
 742                 unlock_page(mp->page);
 743                 return;
 744         }
 745         page_cache_get(mp->page);
 746         mp->count++;
 747         lock_metapage(mp);
 748         unlock_page(mp->page);
 749         release_metapage(mp);
 750 }
 751
 752 void release_metapage(struct metapage * mp)
 753 {
 754         struct page *page = mp->page;
 755         jfs_info("release_metapage: mp = 0x%p, flag = 0x%lx", mp, mp->flag);
 756
 757         BUG_ON(!page);
 758
 759         lock_page(page);
 760         unlock_metapage(mp);
 761
 762         assert(mp->count);
 763         if (--mp->count || mp->nohomeok) {
 764                 unlock_page(page);
 765                 page_cache_release(page);
 766                 return;
 767         }
 768
 769         if (test_bit(META_dirty, &mp->flag)) {
 770                 set_page_dirty(page);
 771                 if (test_bit(META_sync, &mp->flag)) {
 772                         clear_bit(META_sync, &mp->flag);
 773                         write_one_page(page, 1);
 774                         lock_page(page); /* write_one_page unlocks the page */
 775                 }
 776         } else if (mp->lsn)     /* discard_metapage doesn't remove it */
 777                 remove_from_logsync(mp);
 778
 779 #if MPS_PER_PAGE == 1
 780         /*
 781          * If we know this is the only thing in the page, we can throw
 782          * the page out of the page cache.  If pages are larger, we
 783          * don't want to do this.
 784          */
 785
 786         /* Retest mp->count since we may have released page lock */
 787         if (test_bit(META_discard, &mp->flag) && !mp->count) {
 788                 clear_page_dirty(page);
 789                 ClearPageUptodate(page);
 790         }
 791 #else
 792         /* Try to keep metapages from using up too much memory */
 793         drop_metapage(page, mp);
 794 #endif
 795         unlock_page(page);
 796         page_cache_release(page);
 797 }
 798
 799 void __invalidate_metapages(struct inode *ip, s64 addr, int len)
 800 {
 801         sector_t lblock;
 802         int l2BlocksPerPage = PAGE_CACHE_SHIFT - ip->i_blkbits;
 803         int BlocksPerPage = 1 << l2BlocksPerPage;
 804         /* All callers are interested in block device's mapping */
 805         struct address_space *mapping =
 806                 JFS_SBI(ip->i_sb)->direct_inode->i_mapping;
 807         struct metapage *mp;
 808         struct page *page;
 809         unsigned int offset;
 810
 811         /*
 812          * Mark metapages to discard.  They will eventually be
 813          * released, but should not be written.
 814          */
 815         for (lblock = addr & ~(BlocksPerPage - 1); lblock < addr + len;
 816              lblock += BlocksPerPage) {
 817                 page = find_lock_page(mapping, lblock >> l2BlocksPerPage);
 818                 if (!page)
 819                         continue;
 820                 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
 821                         mp = page_to_mp(page, offset);
 822                         if (!mp)
 823                                 continue;
 824                         if (mp->index < addr)
 825                                 continue;
 826                         if (mp->index >= addr + len)
 827                                 break;
 828
 829                         clear_bit(META_dirty, &mp->flag);
 830                         set_bit(META_discard, &mp->flag);
 831                         if (mp->lsn)
 832                                 remove_from_logsync(mp);
 833                 }
 834                 unlock_page(page);
 835                 page_cache_release(page);
 836         }
 837 }
 838
 839 #ifdef CONFIG_JFS_STATISTICS
 840 int jfs_mpstat_read(char *buffer, char **start, off_t offset, int length,
 841                     int *eof, void *data)
 842 {
 843         int len = 0;
 844         off_t begin;
 845
 846         len += sprintf(buffer,
 847                        "JFS Metapage statistics\n"
 848                        "=======================\n"
 849                        "page allocations = %d\n"
 850                        "page frees = %d\n"
 851                        "lock waits = %d\n",
 852                        mpStat.pagealloc,
 853                        mpStat.pagefree,
 854                        mpStat.lockwait);
 855
 856         begin = offset;
 857         *start = buffer + begin;
 858         len -= begin;
 859
 860         if (len > length)
 861                 len = length;
 862         else
 863                 *eof = 1;
 864
 865         if (len < 0)
 866                 len = 0;
 867
 868         return len;
 869 }
 870 #endif