| 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/module.h> |
| 23 | #include <linux/bio.h> |
| 24 | #include <linux/slab.h> |
| 25 | #include <linux/init.h> |
| 26 | #include <linux/buffer_head.h> |
| 27 | #include <linux/mempool.h> |
| 28 | #include <linux/seq_file.h> |
| 29 | #include "jfs_incore.h" |
| 30 | #include "jfs_superblock.h" |
| 31 | #include "jfs_filsys.h" |
| 32 | #include "jfs_metapage.h" |
| 33 | #include "jfs_txnmgr.h" |
| 34 | #include "jfs_debug.h" |
| 35 | |
| 36 | #ifdef CONFIG_JFS_STATISTICS |
| 37 | static struct { |
| 38 | uint pagealloc; /* # of page allocations */ |
| 39 | uint pagefree; /* # of page frees */ |
| 40 | uint lockwait; /* # of sleeping lock_metapage() calls */ |
| 41 | } mpStat; |
| 42 | #endif |
| 43 | |
| 44 | #define metapage_locked(mp) test_bit(META_locked, &(mp)->flag) |
| 45 | #define trylock_metapage(mp) test_and_set_bit_lock(META_locked, &(mp)->flag) |
| 46 | |
| 47 | static inline void unlock_metapage(struct metapage *mp) |
| 48 | { |
| 49 | clear_bit_unlock(META_locked, &mp->flag); |
| 50 | wake_up(&mp->wait); |
| 51 | } |
| 52 | |
| 53 | static inline void __lock_metapage(struct metapage *mp) |
| 54 | { |
| 55 | DECLARE_WAITQUEUE(wait, current); |
| 56 | INCREMENT(mpStat.lockwait); |
| 57 | add_wait_queue_exclusive(&mp->wait, &wait); |
| 58 | do { |
| 59 | set_current_state(TASK_UNINTERRUPTIBLE); |
| 60 | if (metapage_locked(mp)) { |
| 61 | unlock_page(mp->page); |
| 62 | io_schedule(); |
| 63 | lock_page(mp->page); |
| 64 | } |
| 65 | } while (trylock_metapage(mp)); |
| 66 | __set_current_state(TASK_RUNNING); |
| 67 | remove_wait_queue(&mp->wait, &wait); |
| 68 | } |
| 69 | |
| 70 | /* |
| 71 | * Must have mp->page locked |
| 72 | */ |
| 73 | static inline void lock_metapage(struct metapage *mp) |
| 74 | { |
| 75 | if (trylock_metapage(mp)) |
| 76 | __lock_metapage(mp); |
| 77 | } |
| 78 | |
| 79 | #define METAPOOL_MIN_PAGES 32 |
| 80 | static struct kmem_cache *metapage_cache; |
| 81 | static mempool_t *metapage_mempool; |
| 82 | |
| 83 | #define MPS_PER_PAGE (PAGE_CACHE_SIZE >> L2PSIZE) |
| 84 | |
| 85 | #if MPS_PER_PAGE > 1 |
| 86 | |
| 87 | struct meta_anchor { |
| 88 | int mp_count; |
| 89 | atomic_t io_count; |
| 90 | struct metapage *mp[MPS_PER_PAGE]; |
| 91 | }; |
| 92 | #define mp_anchor(page) ((struct meta_anchor *)page_private(page)) |
| 93 | |
| 94 | static inline struct metapage *page_to_mp(struct page *page, int offset) |
| 95 | { |
| 96 | if (!PagePrivate(page)) |
| 97 | return NULL; |
| 98 | return mp_anchor(page)->mp[offset >> L2PSIZE]; |
| 99 | } |
| 100 | |
| 101 | static inline int insert_metapage(struct page *page, struct metapage *mp) |
| 102 | { |
| 103 | struct meta_anchor *a; |
| 104 | int index; |
| 105 | int l2mp_blocks; /* log2 blocks per metapage */ |
| 106 | |
| 107 | if (PagePrivate(page)) |
| 108 | a = mp_anchor(page); |
| 109 | else { |
| 110 | a = kzalloc(sizeof(struct meta_anchor), GFP_NOFS); |
| 111 | if (!a) |
| 112 | return -ENOMEM; |
| 113 | set_page_private(page, (unsigned long)a); |
| 114 | SetPagePrivate(page); |
| 115 | kmap(page); |
| 116 | } |
| 117 | |
| 118 | if (mp) { |
| 119 | l2mp_blocks = L2PSIZE - page->mapping->host->i_blkbits; |
| 120 | index = (mp->index >> l2mp_blocks) & (MPS_PER_PAGE - 1); |
| 121 | a->mp_count++; |
| 122 | a->mp[index] = mp; |
| 123 | } |
| 124 | |
| 125 | return 0; |
| 126 | } |
| 127 | |
| 128 | static inline void remove_metapage(struct page *page, struct metapage *mp) |
| 129 | { |
| 130 | struct meta_anchor *a = mp_anchor(page); |
| 131 | int l2mp_blocks = L2PSIZE - page->mapping->host->i_blkbits; |
| 132 | int index; |
| 133 | |
| 134 | index = (mp->index >> l2mp_blocks) & (MPS_PER_PAGE - 1); |
| 135 | |
| 136 | BUG_ON(a->mp[index] != mp); |
| 137 | |
| 138 | a->mp[index] = NULL; |
| 139 | if (--a->mp_count == 0) { |
| 140 | kfree(a); |
| 141 | set_page_private(page, 0); |
| 142 | ClearPagePrivate(page); |
| 143 | kunmap(page); |
| 144 | } |
| 145 | } |
| 146 | |
| 147 | static inline void inc_io(struct page *page) |
| 148 | { |
| 149 | atomic_inc(&mp_anchor(page)->io_count); |
| 150 | } |
| 151 | |
| 152 | static inline void dec_io(struct page *page, void (*handler) (struct page *)) |
| 153 | { |
| 154 | if (atomic_dec_and_test(&mp_anchor(page)->io_count)) |
| 155 | handler(page); |
| 156 | } |
| 157 | |
| 158 | #else |
| 159 | static inline struct metapage *page_to_mp(struct page *page, int offset) |
| 160 | { |
| 161 | return PagePrivate(page) ? (struct metapage *)page_private(page) : NULL; |
| 162 | } |
| 163 | |
| 164 | static inline int insert_metapage(struct page *page, struct metapage *mp) |
| 165 | { |
| 166 | if (mp) { |
| 167 | set_page_private(page, (unsigned long)mp); |
| 168 | SetPagePrivate(page); |
| 169 | kmap(page); |
| 170 | } |
| 171 | return 0; |
| 172 | } |
| 173 | |
| 174 | static inline void remove_metapage(struct page *page, struct metapage *mp) |
| 175 | { |
| 176 | set_page_private(page, 0); |
| 177 | ClearPagePrivate(page); |
| 178 | kunmap(page); |
| 179 | } |
| 180 | |
| 181 | #define inc_io(page) do {} while(0) |
| 182 | #define dec_io(page, handler) handler(page) |
| 183 | |
| 184 | #endif |
| 185 | |
| 186 | static void init_once(void *foo) |
| 187 | { |
| 188 | struct metapage *mp = (struct metapage *)foo; |
| 189 | |
| 190 | mp->lid = 0; |
| 191 | mp->lsn = 0; |
| 192 | mp->flag = 0; |
| 193 | mp->data = NULL; |
| 194 | mp->clsn = 0; |
| 195 | mp->log = NULL; |
| 196 | set_bit(META_free, &mp->flag); |
| 197 | init_waitqueue_head(&mp->wait); |
| 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); |
| 220 | if (metapage_cache == NULL) |
| 221 | return -ENOMEM; |
| 222 | |
| 223 | metapage_mempool = mempool_create_slab_pool(METAPOOL_MIN_PAGES, |
| 224 | 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 | int *len) |
| 256 | { |
| 257 | int rc = 0; |
| 258 | int xflag; |
| 259 | s64 xaddr; |
| 260 | sector_t file_blocks = (inode->i_size + inode->i_sb->s_blocksize - 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 void metapage_read_end_io(struct bio *bio, int err) |
| 287 | { |
| 288 | struct page *page = bio->bi_private; |
| 289 | |
| 290 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) { |
| 291 | printk(KERN_ERR "metapage_read_end_io: I/O error\n"); |
| 292 | SetPageError(page); |
| 293 | } |
| 294 | |
| 295 | dec_io(page, last_read_complete); |
| 296 | bio_put(bio); |
| 297 | } |
| 298 | |
| 299 | static void remove_from_logsync(struct metapage *mp) |
| 300 | { |
| 301 | struct jfs_log *log = mp->log; |
| 302 | unsigned long flags; |
| 303 | /* |
| 304 | * This can race. Recheck that log hasn't been set to null, and after |
| 305 | * acquiring logsync lock, recheck lsn |
| 306 | */ |
| 307 | if (!log) |
| 308 | return; |
| 309 | |
| 310 | LOGSYNC_LOCK(log, flags); |
| 311 | if (mp->lsn) { |
| 312 | mp->log = NULL; |
| 313 | mp->lsn = 0; |
| 314 | mp->clsn = 0; |
| 315 | log->count--; |
| 316 | list_del(&mp->synclist); |
| 317 | } |
| 318 | LOGSYNC_UNLOCK(log, flags); |
| 319 | } |
| 320 | |
| 321 | static void last_write_complete(struct page *page) |
| 322 | { |
| 323 | struct metapage *mp; |
| 324 | unsigned int offset; |
| 325 | |
| 326 | for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) { |
| 327 | mp = page_to_mp(page, offset); |
| 328 | if (mp && test_bit(META_io, &mp->flag)) { |
| 329 | if (mp->lsn) |
| 330 | remove_from_logsync(mp); |
| 331 | clear_bit(META_io, &mp->flag); |
| 332 | } |
| 333 | /* |
| 334 | * I'd like to call drop_metapage here, but I don't think it's |
| 335 | * safe unless I have the page locked |
| 336 | */ |
| 337 | } |
| 338 | end_page_writeback(page); |
| 339 | } |
| 340 | |
| 341 | static void metapage_write_end_io(struct bio *bio, int err) |
| 342 | { |
| 343 | struct page *page = bio->bi_private; |
| 344 | |
| 345 | BUG_ON(!PagePrivate(page)); |
| 346 | |
| 347 | if (! test_bit(BIO_UPTODATE, &bio->bi_flags)) { |
| 348 | printk(KERN_ERR "metapage_write_end_io: I/O error\n"); |
| 349 | SetPageError(page); |
| 350 | } |
| 351 | dec_io(page, last_write_complete); |
| 352 | bio_put(bio); |
| 353 | } |
| 354 | |
| 355 | static int metapage_writepage(struct page *page, struct writeback_control *wbc) |
| 356 | { |
| 357 | struct bio *bio = NULL; |
| 358 | int block_offset; /* block offset of mp within page */ |
| 359 | struct inode *inode = page->mapping->host; |
| 360 | int blocks_per_mp = JFS_SBI(inode->i_sb)->nbperpage; |
| 361 | int len; |
| 362 | int xlen; |
| 363 | struct metapage *mp; |
| 364 | int redirty = 0; |
| 365 | sector_t lblock; |
| 366 | int nr_underway = 0; |
| 367 | sector_t pblock; |
| 368 | sector_t next_block = 0; |
| 369 | sector_t page_start; |
| 370 | unsigned long bio_bytes = 0; |
| 371 | unsigned long bio_offset = 0; |
| 372 | int offset; |
| 373 | int bad_blocks = 0; |
| 374 | |
| 375 | page_start = (sector_t)page->index << |
| 376 | (PAGE_CACHE_SHIFT - inode->i_blkbits); |
| 377 | BUG_ON(!PageLocked(page)); |
| 378 | BUG_ON(PageWriteback(page)); |
| 379 | set_page_writeback(page); |
| 380 | |
| 381 | for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) { |
| 382 | mp = page_to_mp(page, offset); |
| 383 | |
| 384 | if (!mp || !test_bit(META_dirty, &mp->flag)) |
| 385 | continue; |
| 386 | |
| 387 | if (mp->nohomeok && !test_bit(META_forcewrite, &mp->flag)) { |
| 388 | redirty = 1; |
| 389 | /* |
| 390 | * Make sure this page isn't blocked indefinitely. |
| 391 | * If the journal isn't undergoing I/O, push it |
| 392 | */ |
| 393 | if (mp->log && !(mp->log->cflag & logGC_PAGEOUT)) |
| 394 | jfs_flush_journal(mp->log, 0); |
| 395 | continue; |
| 396 | } |
| 397 | |
| 398 | clear_bit(META_dirty, &mp->flag); |
| 399 | set_bit(META_io, &mp->flag); |
| 400 | block_offset = offset >> inode->i_blkbits; |
| 401 | lblock = page_start + block_offset; |
| 402 | if (bio) { |
| 403 | if (xlen && lblock == next_block) { |
| 404 | /* Contiguous, in memory & on disk */ |
| 405 | len = min(xlen, blocks_per_mp); |
| 406 | xlen -= len; |
| 407 | bio_bytes += len << inode->i_blkbits; |
| 408 | continue; |
| 409 | } |
| 410 | /* Not contiguous */ |
| 411 | if (bio_add_page(bio, page, bio_bytes, bio_offset) < |
| 412 | bio_bytes) |
| 413 | goto add_failed; |
| 414 | /* |
| 415 | * Increment counter before submitting i/o to keep |
| 416 | * count from hitting zero before we're through |
| 417 | */ |
| 418 | inc_io(page); |
| 419 | if (!bio->bi_size) |
| 420 | goto dump_bio; |
| 421 | submit_bio(WRITE, bio); |
| 422 | nr_underway++; |
| 423 | bio = NULL; |
| 424 | } else |
| 425 | inc_io(page); |
| 426 | xlen = (PAGE_CACHE_SIZE - offset) >> inode->i_blkbits; |
| 427 | pblock = metapage_get_blocks(inode, lblock, &xlen); |
| 428 | if (!pblock) { |
| 429 | printk(KERN_ERR "JFS: metapage_get_blocks failed\n"); |
| 430 | /* |
| 431 | * We already called inc_io(), but can't cancel it |
| 432 | * with dec_io() until we're done with the page |
| 433 | */ |
| 434 | bad_blocks++; |
| 435 | continue; |
| 436 | } |
| 437 | len = min(xlen, (int)JFS_SBI(inode->i_sb)->nbperpage); |
| 438 | |
| 439 | bio = bio_alloc(GFP_NOFS, 1); |
| 440 | bio->bi_bdev = inode->i_sb->s_bdev; |
| 441 | bio->bi_sector = pblock << (inode->i_blkbits - 9); |
| 442 | bio->bi_end_io = metapage_write_end_io; |
| 443 | bio->bi_private = page; |
| 444 | |
| 445 | /* Don't call bio_add_page yet, we may add to this vec */ |
| 446 | bio_offset = offset; |
| 447 | bio_bytes = len << inode->i_blkbits; |
| 448 | |
| 449 | xlen -= len; |
| 450 | next_block = lblock + len; |
| 451 | } |
| 452 | if (bio) { |
| 453 | if (bio_add_page(bio, page, bio_bytes, bio_offset) < bio_bytes) |
| 454 | goto add_failed; |
| 455 | if (!bio->bi_size) |
| 456 | goto dump_bio; |
| 457 | |
| 458 | submit_bio(WRITE, bio); |
| 459 | nr_underway++; |
| 460 | } |
| 461 | if (redirty) |
| 462 | redirty_page_for_writepage(wbc, page); |
| 463 | |
| 464 | unlock_page(page); |
| 465 | |
| 466 | if (bad_blocks) |
| 467 | goto err_out; |
| 468 | |
| 469 | if (nr_underway == 0) |
| 470 | end_page_writeback(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 | print_hex_dump(KERN_ERR, "JFS: dump of bio: ", DUMP_PREFIX_ADDRESS, 16, |
| 479 | 4, bio, sizeof(*bio), 0); |
| 480 | skip: |
| 481 | bio_put(bio); |
| 482 | unlock_page(page); |
| 483 | dec_io(page, last_write_complete); |
| 484 | err_out: |
| 485 | while (bad_blocks--) |
| 486 | dec_io(page, last_write_complete); |
| 487 | return -EIO; |
| 488 | } |
| 489 | |
| 490 | static int metapage_readpage(struct file *fp, struct page *page) |
| 491 | { |
| 492 | struct inode *inode = page->mapping->host; |
| 493 | struct bio *bio = NULL; |
| 494 | int block_offset; |
| 495 | int blocks_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits; |
| 496 | sector_t page_start; /* address of page in fs blocks */ |
| 497 | sector_t pblock; |
| 498 | int xlen; |
| 499 | unsigned int len; |
| 500 | int offset; |
| 501 | |
| 502 | BUG_ON(!PageLocked(page)); |
| 503 | page_start = (sector_t)page->index << |
| 504 | (PAGE_CACHE_SHIFT - inode->i_blkbits); |
| 505 | |
| 506 | block_offset = 0; |
| 507 | while (block_offset < blocks_per_page) { |
| 508 | xlen = blocks_per_page - block_offset; |
| 509 | pblock = metapage_get_blocks(inode, page_start + block_offset, |
| 510 | &xlen); |
| 511 | if (pblock) { |
| 512 | if (!PagePrivate(page)) |
| 513 | insert_metapage(page, NULL); |
| 514 | inc_io(page); |
| 515 | if (bio) |
| 516 | submit_bio(READ, bio); |
| 517 | |
| 518 | bio = bio_alloc(GFP_NOFS, 1); |
| 519 | bio->bi_bdev = inode->i_sb->s_bdev; |
| 520 | bio->bi_sector = pblock << (inode->i_blkbits - 9); |
| 521 | bio->bi_end_io = metapage_read_end_io; |
| 522 | bio->bi_private = page; |
| 523 | len = xlen << inode->i_blkbits; |
| 524 | offset = block_offset << inode->i_blkbits; |
| 525 | if (bio_add_page(bio, page, len, offset) < len) |
| 526 | goto add_failed; |
| 527 | block_offset += xlen; |
| 528 | } else |
| 529 | block_offset++; |
| 530 | } |
| 531 | if (bio) |
| 532 | submit_bio(READ, bio); |
| 533 | else |
| 534 | unlock_page(page); |
| 535 | |
| 536 | return 0; |
| 537 | |
| 538 | add_failed: |
| 539 | printk(KERN_ERR "JFS: bio_add_page failed unexpectedly\n"); |
| 540 | bio_put(bio); |
| 541 | dec_io(page, last_read_complete); |
| 542 | return -EIO; |
| 543 | } |
| 544 | |
| 545 | static int metapage_releasepage(struct page *page, gfp_t gfp_mask) |
| 546 | { |
| 547 | struct metapage *mp; |
| 548 | int ret = 1; |
| 549 | int offset; |
| 550 | |
| 551 | for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) { |
| 552 | mp = page_to_mp(page, offset); |
| 553 | |
| 554 | if (!mp) |
| 555 | continue; |
| 556 | |
| 557 | jfs_info("metapage_releasepage: mp = 0x%p", mp); |
| 558 | if (mp->count || mp->nohomeok || |
| 559 | test_bit(META_dirty, &mp->flag)) { |
| 560 | jfs_info("count = %ld, nohomeok = %d", mp->count, |
| 561 | mp->nohomeok); |
| 562 | ret = 0; |
| 563 | continue; |
| 564 | } |
| 565 | if (mp->lsn) |
| 566 | remove_from_logsync(mp); |
| 567 | remove_metapage(page, mp); |
| 568 | INCREMENT(mpStat.pagefree); |
| 569 | free_metapage(mp); |
| 570 | } |
| 571 | return ret; |
| 572 | } |
| 573 | |
| 574 | static void metapage_invalidatepage(struct page *page, unsigned long offset) |
| 575 | { |
| 576 | BUG_ON(offset); |
| 577 | |
| 578 | BUG_ON(PageWriteback(page)); |
| 579 | |
| 580 | metapage_releasepage(page, 0); |
| 581 | } |
| 582 | |
| 583 | const struct address_space_operations jfs_metapage_aops = { |
| 584 | .readpage = metapage_readpage, |
| 585 | .writepage = metapage_writepage, |
| 586 | .releasepage = metapage_releasepage, |
| 587 | .invalidatepage = metapage_invalidatepage, |
| 588 | .set_page_dirty = __set_page_dirty_nobuffers, |
| 589 | }; |
| 590 | |
| 591 | struct metapage *__get_metapage(struct inode *inode, unsigned long lblock, |
| 592 | unsigned int size, int absolute, |
| 593 | unsigned long new) |
| 594 | { |
| 595 | int l2BlocksPerPage; |
| 596 | int l2bsize; |
| 597 | struct address_space *mapping; |
| 598 | struct metapage *mp = NULL; |
| 599 | struct page *page; |
| 600 | unsigned long page_index; |
| 601 | unsigned long page_offset; |
| 602 | |
| 603 | jfs_info("__get_metapage: ino = %ld, lblock = 0x%lx, abs=%d", |
| 604 | inode->i_ino, lblock, absolute); |
| 605 | |
| 606 | l2bsize = inode->i_blkbits; |
| 607 | l2BlocksPerPage = PAGE_CACHE_SHIFT - l2bsize; |
| 608 | page_index = lblock >> l2BlocksPerPage; |
| 609 | page_offset = (lblock - (page_index << l2BlocksPerPage)) << l2bsize; |
| 610 | if ((page_offset + size) > PAGE_CACHE_SIZE) { |
| 611 | jfs_err("MetaData crosses page boundary!!"); |
| 612 | jfs_err("lblock = %lx, size = %d", lblock, size); |
| 613 | dump_stack(); |
| 614 | return NULL; |
| 615 | } |
| 616 | if (absolute) |
| 617 | mapping = JFS_SBI(inode->i_sb)->direct_inode->i_mapping; |
| 618 | else { |
| 619 | /* |
| 620 | * If an nfs client tries to read an inode that is larger |
| 621 | * than any existing inodes, we may try to read past the |
| 622 | * end of the inode map |
| 623 | */ |
| 624 | if ((lblock << inode->i_blkbits) >= inode->i_size) |
| 625 | return NULL; |
| 626 | mapping = inode->i_mapping; |
| 627 | } |
| 628 | |
| 629 | if (new && (PSIZE == PAGE_CACHE_SIZE)) { |
| 630 | page = grab_cache_page(mapping, page_index); |
| 631 | if (!page) { |
| 632 | jfs_err("grab_cache_page failed!"); |
| 633 | return NULL; |
| 634 | } |
| 635 | SetPageUptodate(page); |
| 636 | } else { |
| 637 | page = read_mapping_page(mapping, page_index, NULL); |
| 638 | if (IS_ERR(page) || !PageUptodate(page)) { |
| 639 | jfs_err("read_mapping_page failed!"); |
| 640 | return NULL; |
| 641 | } |
| 642 | lock_page(page); |
| 643 | } |
| 644 | |
| 645 | mp = page_to_mp(page, page_offset); |
| 646 | if (mp) { |
| 647 | if (mp->logical_size != size) { |
| 648 | jfs_error(inode->i_sb, |
| 649 | "__get_metapage: mp->logical_size != size"); |
| 650 | jfs_err("logical_size = %d, size = %d", |
| 651 | mp->logical_size, size); |
| 652 | dump_stack(); |
| 653 | goto unlock; |
| 654 | } |
| 655 | mp->count++; |
| 656 | lock_metapage(mp); |
| 657 | if (test_bit(META_discard, &mp->flag)) { |
| 658 | if (!new) { |
| 659 | jfs_error(inode->i_sb, |
| 660 | "__get_metapage: using a " |
| 661 | "discarded metapage"); |
| 662 | discard_metapage(mp); |
| 663 | goto unlock; |
| 664 | } |
| 665 | clear_bit(META_discard, &mp->flag); |
| 666 | } |
| 667 | } else { |
| 668 | INCREMENT(mpStat.pagealloc); |
| 669 | mp = alloc_metapage(GFP_NOFS); |
| 670 | mp->page = page; |
| 671 | mp->flag = 0; |
| 672 | mp->xflag = COMMIT_PAGE; |
| 673 | mp->count = 1; |
| 674 | mp->nohomeok = 0; |
| 675 | mp->logical_size = size; |
| 676 | mp->data = page_address(page) + page_offset; |
| 677 | mp->index = lblock; |
| 678 | if (unlikely(insert_metapage(page, mp))) { |
| 679 | free_metapage(mp); |
| 680 | goto unlock; |
| 681 | } |
| 682 | lock_metapage(mp); |
| 683 | } |
| 684 | |
| 685 | if (new) { |
| 686 | jfs_info("zeroing mp = 0x%p", mp); |
| 687 | memset(mp->data, 0, PSIZE); |
| 688 | } |
| 689 | |
| 690 | unlock_page(page); |
| 691 | jfs_info("__get_metapage: returning = 0x%p data = 0x%p", mp, mp->data); |
| 692 | return mp; |
| 693 | |
| 694 | unlock: |
| 695 | unlock_page(page); |
| 696 | return NULL; |
| 697 | } |
| 698 | |
| 699 | void grab_metapage(struct metapage * mp) |
| 700 | { |
| 701 | jfs_info("grab_metapage: mp = 0x%p", mp); |
| 702 | page_cache_get(mp->page); |
| 703 | lock_page(mp->page); |
| 704 | mp->count++; |
| 705 | lock_metapage(mp); |
| 706 | unlock_page(mp->page); |
| 707 | } |
| 708 | |
| 709 | void force_metapage(struct metapage *mp) |
| 710 | { |
| 711 | struct page *page = mp->page; |
| 712 | jfs_info("force_metapage: mp = 0x%p", mp); |
| 713 | set_bit(META_forcewrite, &mp->flag); |
| 714 | clear_bit(META_sync, &mp->flag); |
| 715 | page_cache_get(page); |
| 716 | lock_page(page); |
| 717 | set_page_dirty(page); |
| 718 | write_one_page(page, 1); |
| 719 | clear_bit(META_forcewrite, &mp->flag); |
| 720 | page_cache_release(page); |
| 721 | } |
| 722 | |
| 723 | void hold_metapage(struct metapage *mp) |
| 724 | { |
| 725 | lock_page(mp->page); |
| 726 | } |
| 727 | |
| 728 | void put_metapage(struct metapage *mp) |
| 729 | { |
| 730 | if (mp->count || mp->nohomeok) { |
| 731 | /* Someone else will release this */ |
| 732 | unlock_page(mp->page); |
| 733 | return; |
| 734 | } |
| 735 | page_cache_get(mp->page); |
| 736 | mp->count++; |
| 737 | lock_metapage(mp); |
| 738 | unlock_page(mp->page); |
| 739 | release_metapage(mp); |
| 740 | } |
| 741 | |
| 742 | void release_metapage(struct metapage * mp) |
| 743 | { |
| 744 | struct page *page = mp->page; |
| 745 | jfs_info("release_metapage: mp = 0x%p, flag = 0x%lx", mp, mp->flag); |
| 746 | |
| 747 | BUG_ON(!page); |
| 748 | |
| 749 | lock_page(page); |
| 750 | unlock_metapage(mp); |
| 751 | |
| 752 | assert(mp->count); |
| 753 | if (--mp->count || mp->nohomeok) { |
| 754 | unlock_page(page); |
| 755 | page_cache_release(page); |
| 756 | return; |
| 757 | } |
| 758 | |
| 759 | if (test_bit(META_dirty, &mp->flag)) { |
| 760 | set_page_dirty(page); |
| 761 | if (test_bit(META_sync, &mp->flag)) { |
| 762 | clear_bit(META_sync, &mp->flag); |
| 763 | write_one_page(page, 1); |
| 764 | lock_page(page); /* write_one_page unlocks the page */ |
| 765 | } |
| 766 | } else if (mp->lsn) /* discard_metapage doesn't remove it */ |
| 767 | remove_from_logsync(mp); |
| 768 | |
| 769 | /* Try to keep metapages from using up too much memory */ |
| 770 | drop_metapage(page, mp); |
| 771 | |
| 772 | unlock_page(page); |
| 773 | page_cache_release(page); |
| 774 | } |
| 775 | |
| 776 | void __invalidate_metapages(struct inode *ip, s64 addr, int len) |
| 777 | { |
| 778 | sector_t lblock; |
| 779 | int l2BlocksPerPage = PAGE_CACHE_SHIFT - ip->i_blkbits; |
| 780 | int BlocksPerPage = 1 << l2BlocksPerPage; |
| 781 | /* All callers are interested in block device's mapping */ |
| 782 | struct address_space *mapping = |
| 783 | JFS_SBI(ip->i_sb)->direct_inode->i_mapping; |
| 784 | struct metapage *mp; |
| 785 | struct page *page; |
| 786 | unsigned int offset; |
| 787 | |
| 788 | /* |
| 789 | * Mark metapages to discard. They will eventually be |
| 790 | * released, but should not be written. |
| 791 | */ |
| 792 | for (lblock = addr & ~(BlocksPerPage - 1); lblock < addr + len; |
| 793 | lblock += BlocksPerPage) { |
| 794 | page = find_lock_page(mapping, lblock >> l2BlocksPerPage); |
| 795 | if (!page) |
| 796 | continue; |
| 797 | for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) { |
| 798 | mp = page_to_mp(page, offset); |
| 799 | if (!mp) |
| 800 | continue; |
| 801 | if (mp->index < addr) |
| 802 | continue; |
| 803 | if (mp->index >= addr + len) |
| 804 | break; |
| 805 | |
| 806 | clear_bit(META_dirty, &mp->flag); |
| 807 | set_bit(META_discard, &mp->flag); |
| 808 | if (mp->lsn) |
| 809 | remove_from_logsync(mp); |
| 810 | } |
| 811 | unlock_page(page); |
| 812 | page_cache_release(page); |
| 813 | } |
| 814 | } |
| 815 | |
| 816 | #ifdef CONFIG_JFS_STATISTICS |
| 817 | static int jfs_mpstat_proc_show(struct seq_file *m, void *v) |
| 818 | { |
| 819 | seq_printf(m, |
| 820 | "JFS Metapage statistics\n" |
| 821 | "=======================\n" |
| 822 | "page allocations = %d\n" |
| 823 | "page frees = %d\n" |
| 824 | "lock waits = %d\n", |
| 825 | mpStat.pagealloc, |
| 826 | mpStat.pagefree, |
| 827 | mpStat.lockwait); |
| 828 | return 0; |
| 829 | } |
| 830 | |
| 831 | static int jfs_mpstat_proc_open(struct inode *inode, struct file *file) |
| 832 | { |
| 833 | return single_open(file, jfs_mpstat_proc_show, NULL); |
| 834 | } |
| 835 | |
| 836 | const struct file_operations jfs_mpstat_proc_fops = { |
| 837 | .owner = THIS_MODULE, |
| 838 | .open = jfs_mpstat_proc_open, |
| 839 | .read = seq_read, |
| 840 | .llseek = seq_lseek, |
| 841 | .release = single_release, |
| 842 | }; |
| 843 | #endif |