David Howells | 9ae326a | 2009-04-03 16:42:41 +0100 | [diff] [blame] | 1 | /* Storage object read/write |
| 2 | * |
| 3 | * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved. |
| 4 | * Written by David Howells (dhowells@redhat.com) |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or |
| 7 | * modify it under the terms of the GNU General Public Licence |
| 8 | * as published by the Free Software Foundation; either version |
| 9 | * 2 of the Licence, or (at your option) any later version. |
| 10 | */ |
| 11 | |
| 12 | #include <linux/mount.h> |
| 13 | #include <linux/file.h> |
| 14 | #include "internal.h" |
| 15 | |
| 16 | /* |
| 17 | * detect wake up events generated by the unlocking of pages in which we're |
| 18 | * interested |
| 19 | * - we use this to detect read completion of backing pages |
| 20 | * - the caller holds the waitqueue lock |
| 21 | */ |
| 22 | static int cachefiles_read_waiter(wait_queue_t *wait, unsigned mode, |
| 23 | int sync, void *_key) |
| 24 | { |
| 25 | struct cachefiles_one_read *monitor = |
| 26 | container_of(wait, struct cachefiles_one_read, monitor); |
| 27 | struct cachefiles_object *object; |
| 28 | struct wait_bit_key *key = _key; |
| 29 | struct page *page = wait->private; |
| 30 | |
| 31 | ASSERT(key); |
| 32 | |
| 33 | _enter("{%lu},%u,%d,{%p,%u}", |
| 34 | monitor->netfs_page->index, mode, sync, |
| 35 | key->flags, key->bit_nr); |
| 36 | |
| 37 | if (key->flags != &page->flags || |
| 38 | key->bit_nr != PG_locked) |
| 39 | return 0; |
| 40 | |
| 41 | _debug("--- monitor %p %lx ---", page, page->flags); |
| 42 | |
| 43 | if (!PageUptodate(page) && !PageError(page)) |
| 44 | dump_stack(); |
| 45 | |
| 46 | /* remove from the waitqueue */ |
| 47 | list_del(&wait->task_list); |
| 48 | |
| 49 | /* move onto the action list and queue for FS-Cache thread pool */ |
| 50 | ASSERT(monitor->op); |
| 51 | |
| 52 | object = container_of(monitor->op->op.object, |
| 53 | struct cachefiles_object, fscache); |
| 54 | |
| 55 | spin_lock(&object->work_lock); |
| 56 | list_add_tail(&monitor->op_link, &monitor->op->to_do); |
| 57 | spin_unlock(&object->work_lock); |
| 58 | |
| 59 | fscache_enqueue_retrieval(monitor->op); |
| 60 | return 0; |
| 61 | } |
| 62 | |
| 63 | /* |
| 64 | * copy data from backing pages to netfs pages to complete a read operation |
| 65 | * - driven by FS-Cache's thread pool |
| 66 | */ |
| 67 | static void cachefiles_read_copier(struct fscache_operation *_op) |
| 68 | { |
| 69 | struct cachefiles_one_read *monitor; |
| 70 | struct cachefiles_object *object; |
| 71 | struct fscache_retrieval *op; |
| 72 | struct pagevec pagevec; |
| 73 | int error, max; |
| 74 | |
| 75 | op = container_of(_op, struct fscache_retrieval, op); |
| 76 | object = container_of(op->op.object, |
| 77 | struct cachefiles_object, fscache); |
| 78 | |
| 79 | _enter("{ino=%lu}", object->backer->d_inode->i_ino); |
| 80 | |
| 81 | pagevec_init(&pagevec, 0); |
| 82 | |
| 83 | max = 8; |
| 84 | spin_lock_irq(&object->work_lock); |
| 85 | |
| 86 | while (!list_empty(&op->to_do)) { |
| 87 | monitor = list_entry(op->to_do.next, |
| 88 | struct cachefiles_one_read, op_link); |
| 89 | list_del(&monitor->op_link); |
| 90 | |
| 91 | spin_unlock_irq(&object->work_lock); |
| 92 | |
| 93 | _debug("- copy {%lu}", monitor->back_page->index); |
| 94 | |
| 95 | error = -EIO; |
| 96 | if (PageUptodate(monitor->back_page)) { |
| 97 | copy_highpage(monitor->netfs_page, monitor->back_page); |
| 98 | |
| 99 | pagevec_add(&pagevec, monitor->netfs_page); |
| 100 | fscache_mark_pages_cached(monitor->op, &pagevec); |
| 101 | error = 0; |
| 102 | } |
| 103 | |
| 104 | if (error) |
| 105 | cachefiles_io_error_obj( |
| 106 | object, |
| 107 | "Readpage failed on backing file %lx", |
| 108 | (unsigned long) monitor->back_page->flags); |
| 109 | |
| 110 | page_cache_release(monitor->back_page); |
| 111 | |
| 112 | fscache_end_io(op, monitor->netfs_page, error); |
| 113 | page_cache_release(monitor->netfs_page); |
| 114 | fscache_put_retrieval(op); |
| 115 | kfree(monitor); |
| 116 | |
| 117 | /* let the thread pool have some air occasionally */ |
| 118 | max--; |
| 119 | if (max < 0 || need_resched()) { |
| 120 | if (!list_empty(&op->to_do)) |
| 121 | fscache_enqueue_retrieval(op); |
| 122 | _leave(" [maxed out]"); |
| 123 | return; |
| 124 | } |
| 125 | |
| 126 | spin_lock_irq(&object->work_lock); |
| 127 | } |
| 128 | |
| 129 | spin_unlock_irq(&object->work_lock); |
| 130 | _leave(""); |
| 131 | } |
| 132 | |
| 133 | /* |
| 134 | * read the corresponding page to the given set from the backing file |
| 135 | * - an uncertain page is simply discarded, to be tried again another time |
| 136 | */ |
| 137 | static int cachefiles_read_backing_file_one(struct cachefiles_object *object, |
| 138 | struct fscache_retrieval *op, |
| 139 | struct page *netpage, |
| 140 | struct pagevec *pagevec) |
| 141 | { |
| 142 | struct cachefiles_one_read *monitor; |
| 143 | struct address_space *bmapping; |
| 144 | struct page *newpage, *backpage; |
| 145 | int ret; |
| 146 | |
| 147 | _enter(""); |
| 148 | |
| 149 | pagevec_reinit(pagevec); |
| 150 | |
| 151 | _debug("read back %p{%lu,%d}", |
| 152 | netpage, netpage->index, page_count(netpage)); |
| 153 | |
| 154 | monitor = kzalloc(sizeof(*monitor), GFP_KERNEL); |
| 155 | if (!monitor) |
| 156 | goto nomem; |
| 157 | |
| 158 | monitor->netfs_page = netpage; |
| 159 | monitor->op = fscache_get_retrieval(op); |
| 160 | |
| 161 | init_waitqueue_func_entry(&monitor->monitor, cachefiles_read_waiter); |
| 162 | |
| 163 | /* attempt to get hold of the backing page */ |
| 164 | bmapping = object->backer->d_inode->i_mapping; |
| 165 | newpage = NULL; |
| 166 | |
| 167 | for (;;) { |
| 168 | backpage = find_get_page(bmapping, netpage->index); |
| 169 | if (backpage) |
| 170 | goto backing_page_already_present; |
| 171 | |
| 172 | if (!newpage) { |
| 173 | newpage = page_cache_alloc_cold(bmapping); |
| 174 | if (!newpage) |
| 175 | goto nomem_monitor; |
| 176 | } |
| 177 | |
| 178 | ret = add_to_page_cache(newpage, bmapping, |
| 179 | netpage->index, GFP_KERNEL); |
| 180 | if (ret == 0) |
| 181 | goto installed_new_backing_page; |
| 182 | if (ret != -EEXIST) |
| 183 | goto nomem_page; |
| 184 | } |
| 185 | |
| 186 | /* we've installed a new backing page, so now we need to add it |
| 187 | * to the LRU list and start it reading */ |
| 188 | installed_new_backing_page: |
| 189 | _debug("- new %p", newpage); |
| 190 | |
| 191 | backpage = newpage; |
| 192 | newpage = NULL; |
| 193 | |
| 194 | page_cache_get(backpage); |
| 195 | pagevec_add(pagevec, backpage); |
| 196 | __pagevec_lru_add_file(pagevec); |
| 197 | |
| 198 | read_backing_page: |
| 199 | ret = bmapping->a_ops->readpage(NULL, backpage); |
| 200 | if (ret < 0) |
| 201 | goto read_error; |
| 202 | |
| 203 | /* set the monitor to transfer the data across */ |
| 204 | monitor_backing_page: |
| 205 | _debug("- monitor add"); |
| 206 | |
| 207 | /* install the monitor */ |
| 208 | page_cache_get(monitor->netfs_page); |
| 209 | page_cache_get(backpage); |
| 210 | monitor->back_page = backpage; |
| 211 | monitor->monitor.private = backpage; |
| 212 | add_page_wait_queue(backpage, &monitor->monitor); |
| 213 | monitor = NULL; |
| 214 | |
| 215 | /* but the page may have been read before the monitor was installed, so |
| 216 | * the monitor may miss the event - so we have to ensure that we do get |
| 217 | * one in such a case */ |
| 218 | if (trylock_page(backpage)) { |
| 219 | _debug("jumpstart %p {%lx}", backpage, backpage->flags); |
| 220 | unlock_page(backpage); |
| 221 | } |
| 222 | goto success; |
| 223 | |
| 224 | /* if the backing page is already present, it can be in one of |
| 225 | * three states: read in progress, read failed or read okay */ |
| 226 | backing_page_already_present: |
| 227 | _debug("- present"); |
| 228 | |
| 229 | if (newpage) { |
| 230 | page_cache_release(newpage); |
| 231 | newpage = NULL; |
| 232 | } |
| 233 | |
| 234 | if (PageError(backpage)) |
| 235 | goto io_error; |
| 236 | |
| 237 | if (PageUptodate(backpage)) |
| 238 | goto backing_page_already_uptodate; |
| 239 | |
| 240 | if (!trylock_page(backpage)) |
| 241 | goto monitor_backing_page; |
| 242 | _debug("read %p {%lx}", backpage, backpage->flags); |
| 243 | goto read_backing_page; |
| 244 | |
| 245 | /* the backing page is already up to date, attach the netfs |
| 246 | * page to the pagecache and LRU and copy the data across */ |
| 247 | backing_page_already_uptodate: |
| 248 | _debug("- uptodate"); |
| 249 | |
| 250 | pagevec_add(pagevec, netpage); |
| 251 | fscache_mark_pages_cached(op, pagevec); |
| 252 | |
| 253 | copy_highpage(netpage, backpage); |
| 254 | fscache_end_io(op, netpage, 0); |
| 255 | |
| 256 | success: |
| 257 | _debug("success"); |
| 258 | ret = 0; |
| 259 | |
| 260 | out: |
| 261 | if (backpage) |
| 262 | page_cache_release(backpage); |
| 263 | if (monitor) { |
| 264 | fscache_put_retrieval(monitor->op); |
| 265 | kfree(monitor); |
| 266 | } |
| 267 | _leave(" = %d", ret); |
| 268 | return ret; |
| 269 | |
| 270 | read_error: |
| 271 | _debug("read error %d", ret); |
| 272 | if (ret == -ENOMEM) |
| 273 | goto out; |
| 274 | io_error: |
| 275 | cachefiles_io_error_obj(object, "Page read error on backing file"); |
| 276 | ret = -ENOBUFS; |
| 277 | goto out; |
| 278 | |
| 279 | nomem_page: |
| 280 | page_cache_release(newpage); |
| 281 | nomem_monitor: |
| 282 | fscache_put_retrieval(monitor->op); |
| 283 | kfree(monitor); |
| 284 | nomem: |
| 285 | _leave(" = -ENOMEM"); |
| 286 | return -ENOMEM; |
| 287 | } |
| 288 | |
| 289 | /* |
| 290 | * read a page from the cache or allocate a block in which to store it |
| 291 | * - cache withdrawal is prevented by the caller |
| 292 | * - returns -EINTR if interrupted |
| 293 | * - returns -ENOMEM if ran out of memory |
| 294 | * - returns -ENOBUFS if no buffers can be made available |
| 295 | * - returns -ENOBUFS if page is beyond EOF |
| 296 | * - if the page is backed by a block in the cache: |
| 297 | * - a read will be started which will call the callback on completion |
| 298 | * - 0 will be returned |
| 299 | * - else if the page is unbacked: |
| 300 | * - the metadata will be retained |
| 301 | * - -ENODATA will be returned |
| 302 | */ |
| 303 | int cachefiles_read_or_alloc_page(struct fscache_retrieval *op, |
| 304 | struct page *page, |
| 305 | gfp_t gfp) |
| 306 | { |
| 307 | struct cachefiles_object *object; |
| 308 | struct cachefiles_cache *cache; |
| 309 | struct pagevec pagevec; |
| 310 | struct inode *inode; |
| 311 | sector_t block0, block; |
| 312 | unsigned shift; |
| 313 | int ret; |
| 314 | |
| 315 | object = container_of(op->op.object, |
| 316 | struct cachefiles_object, fscache); |
| 317 | cache = container_of(object->fscache.cache, |
| 318 | struct cachefiles_cache, cache); |
| 319 | |
| 320 | _enter("{%p},{%lx},,,", object, page->index); |
| 321 | |
| 322 | if (!object->backer) |
| 323 | return -ENOBUFS; |
| 324 | |
| 325 | inode = object->backer->d_inode; |
| 326 | ASSERT(S_ISREG(inode->i_mode)); |
| 327 | ASSERT(inode->i_mapping->a_ops->bmap); |
| 328 | ASSERT(inode->i_mapping->a_ops->readpages); |
| 329 | |
| 330 | /* calculate the shift required to use bmap */ |
| 331 | if (inode->i_sb->s_blocksize > PAGE_SIZE) |
| 332 | return -ENOBUFS; |
| 333 | |
| 334 | shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits; |
| 335 | |
David Howells | 4fbf429 | 2009-11-19 18:11:04 +0000 | [diff] [blame] | 336 | op->op.flags &= FSCACHE_OP_KEEP_FLAGS; |
| 337 | op->op.flags |= FSCACHE_OP_FAST; |
David Howells | 9ae326a | 2009-04-03 16:42:41 +0100 | [diff] [blame] | 338 | op->op.processor = cachefiles_read_copier; |
| 339 | |
| 340 | pagevec_init(&pagevec, 0); |
| 341 | |
| 342 | /* we assume the absence or presence of the first block is a good |
| 343 | * enough indication for the page as a whole |
| 344 | * - TODO: don't use bmap() for this as it is _not_ actually good |
| 345 | * enough for this as it doesn't indicate errors, but it's all we've |
| 346 | * got for the moment |
| 347 | */ |
| 348 | block0 = page->index; |
| 349 | block0 <<= shift; |
| 350 | |
| 351 | block = inode->i_mapping->a_ops->bmap(inode->i_mapping, block0); |
| 352 | _debug("%llx -> %llx", |
| 353 | (unsigned long long) block0, |
| 354 | (unsigned long long) block); |
| 355 | |
| 356 | if (block) { |
| 357 | /* submit the apparently valid page to the backing fs to be |
| 358 | * read from disk */ |
| 359 | ret = cachefiles_read_backing_file_one(object, op, page, |
| 360 | &pagevec); |
| 361 | } else if (cachefiles_has_space(cache, 0, 1) == 0) { |
| 362 | /* there's space in the cache we can use */ |
| 363 | pagevec_add(&pagevec, page); |
| 364 | fscache_mark_pages_cached(op, &pagevec); |
| 365 | ret = -ENODATA; |
| 366 | } else { |
| 367 | ret = -ENOBUFS; |
| 368 | } |
| 369 | |
| 370 | _leave(" = %d", ret); |
| 371 | return ret; |
| 372 | } |
| 373 | |
| 374 | /* |
| 375 | * read the corresponding pages to the given set from the backing file |
| 376 | * - any uncertain pages are simply discarded, to be tried again another time |
| 377 | */ |
| 378 | static int cachefiles_read_backing_file(struct cachefiles_object *object, |
| 379 | struct fscache_retrieval *op, |
| 380 | struct list_head *list, |
| 381 | struct pagevec *mark_pvec) |
| 382 | { |
| 383 | struct cachefiles_one_read *monitor = NULL; |
| 384 | struct address_space *bmapping = object->backer->d_inode->i_mapping; |
| 385 | struct pagevec lru_pvec; |
| 386 | struct page *newpage = NULL, *netpage, *_n, *backpage = NULL; |
| 387 | int ret = 0; |
| 388 | |
| 389 | _enter(""); |
| 390 | |
| 391 | pagevec_init(&lru_pvec, 0); |
| 392 | |
| 393 | list_for_each_entry_safe(netpage, _n, list, lru) { |
| 394 | list_del(&netpage->lru); |
| 395 | |
| 396 | _debug("read back %p{%lu,%d}", |
| 397 | netpage, netpage->index, page_count(netpage)); |
| 398 | |
| 399 | if (!monitor) { |
| 400 | monitor = kzalloc(sizeof(*monitor), GFP_KERNEL); |
| 401 | if (!monitor) |
| 402 | goto nomem; |
| 403 | |
| 404 | monitor->op = fscache_get_retrieval(op); |
| 405 | init_waitqueue_func_entry(&monitor->monitor, |
| 406 | cachefiles_read_waiter); |
| 407 | } |
| 408 | |
| 409 | for (;;) { |
| 410 | backpage = find_get_page(bmapping, netpage->index); |
| 411 | if (backpage) |
| 412 | goto backing_page_already_present; |
| 413 | |
| 414 | if (!newpage) { |
| 415 | newpage = page_cache_alloc_cold(bmapping); |
| 416 | if (!newpage) |
| 417 | goto nomem; |
| 418 | } |
| 419 | |
| 420 | ret = add_to_page_cache(newpage, bmapping, |
| 421 | netpage->index, GFP_KERNEL); |
| 422 | if (ret == 0) |
| 423 | goto installed_new_backing_page; |
| 424 | if (ret != -EEXIST) |
| 425 | goto nomem; |
| 426 | } |
| 427 | |
| 428 | /* we've installed a new backing page, so now we need to add it |
| 429 | * to the LRU list and start it reading */ |
| 430 | installed_new_backing_page: |
| 431 | _debug("- new %p", newpage); |
| 432 | |
| 433 | backpage = newpage; |
| 434 | newpage = NULL; |
| 435 | |
| 436 | page_cache_get(backpage); |
| 437 | if (!pagevec_add(&lru_pvec, backpage)) |
| 438 | __pagevec_lru_add_file(&lru_pvec); |
| 439 | |
| 440 | reread_backing_page: |
| 441 | ret = bmapping->a_ops->readpage(NULL, backpage); |
| 442 | if (ret < 0) |
| 443 | goto read_error; |
| 444 | |
| 445 | /* add the netfs page to the pagecache and LRU, and set the |
| 446 | * monitor to transfer the data across */ |
| 447 | monitor_backing_page: |
| 448 | _debug("- monitor add"); |
| 449 | |
| 450 | ret = add_to_page_cache(netpage, op->mapping, netpage->index, |
| 451 | GFP_KERNEL); |
| 452 | if (ret < 0) { |
| 453 | if (ret == -EEXIST) { |
| 454 | page_cache_release(netpage); |
| 455 | continue; |
| 456 | } |
| 457 | goto nomem; |
| 458 | } |
| 459 | |
| 460 | page_cache_get(netpage); |
| 461 | if (!pagevec_add(&lru_pvec, netpage)) |
| 462 | __pagevec_lru_add_file(&lru_pvec); |
| 463 | |
| 464 | /* install a monitor */ |
| 465 | page_cache_get(netpage); |
| 466 | monitor->netfs_page = netpage; |
| 467 | |
| 468 | page_cache_get(backpage); |
| 469 | monitor->back_page = backpage; |
| 470 | monitor->monitor.private = backpage; |
| 471 | add_page_wait_queue(backpage, &monitor->monitor); |
| 472 | monitor = NULL; |
| 473 | |
| 474 | /* but the page may have been read before the monitor was |
| 475 | * installed, so the monitor may miss the event - so we have to |
| 476 | * ensure that we do get one in such a case */ |
| 477 | if (trylock_page(backpage)) { |
| 478 | _debug("2unlock %p {%lx}", backpage, backpage->flags); |
| 479 | unlock_page(backpage); |
| 480 | } |
| 481 | |
| 482 | page_cache_release(backpage); |
| 483 | backpage = NULL; |
| 484 | |
| 485 | page_cache_release(netpage); |
| 486 | netpage = NULL; |
| 487 | continue; |
| 488 | |
| 489 | /* if the backing page is already present, it can be in one of |
| 490 | * three states: read in progress, read failed or read okay */ |
| 491 | backing_page_already_present: |
| 492 | _debug("- present %p", backpage); |
| 493 | |
| 494 | if (PageError(backpage)) |
| 495 | goto io_error; |
| 496 | |
| 497 | if (PageUptodate(backpage)) |
| 498 | goto backing_page_already_uptodate; |
| 499 | |
| 500 | _debug("- not ready %p{%lx}", backpage, backpage->flags); |
| 501 | |
| 502 | if (!trylock_page(backpage)) |
| 503 | goto monitor_backing_page; |
| 504 | |
| 505 | if (PageError(backpage)) { |
| 506 | _debug("error %lx", backpage->flags); |
| 507 | unlock_page(backpage); |
| 508 | goto io_error; |
| 509 | } |
| 510 | |
| 511 | if (PageUptodate(backpage)) |
| 512 | goto backing_page_already_uptodate_unlock; |
| 513 | |
| 514 | /* we've locked a page that's neither up to date nor erroneous, |
| 515 | * so we need to attempt to read it again */ |
| 516 | goto reread_backing_page; |
| 517 | |
| 518 | /* the backing page is already up to date, attach the netfs |
| 519 | * page to the pagecache and LRU and copy the data across */ |
| 520 | backing_page_already_uptodate_unlock: |
| 521 | _debug("uptodate %lx", backpage->flags); |
| 522 | unlock_page(backpage); |
| 523 | backing_page_already_uptodate: |
| 524 | _debug("- uptodate"); |
| 525 | |
| 526 | ret = add_to_page_cache(netpage, op->mapping, netpage->index, |
| 527 | GFP_KERNEL); |
| 528 | if (ret < 0) { |
| 529 | if (ret == -EEXIST) { |
| 530 | page_cache_release(netpage); |
| 531 | continue; |
| 532 | } |
| 533 | goto nomem; |
| 534 | } |
| 535 | |
| 536 | copy_highpage(netpage, backpage); |
| 537 | |
| 538 | page_cache_release(backpage); |
| 539 | backpage = NULL; |
| 540 | |
| 541 | if (!pagevec_add(mark_pvec, netpage)) |
| 542 | fscache_mark_pages_cached(op, mark_pvec); |
| 543 | |
| 544 | page_cache_get(netpage); |
| 545 | if (!pagevec_add(&lru_pvec, netpage)) |
| 546 | __pagevec_lru_add_file(&lru_pvec); |
| 547 | |
| 548 | fscache_end_io(op, netpage, 0); |
| 549 | page_cache_release(netpage); |
| 550 | netpage = NULL; |
| 551 | continue; |
| 552 | } |
| 553 | |
| 554 | netpage = NULL; |
| 555 | |
| 556 | _debug("out"); |
| 557 | |
| 558 | out: |
| 559 | /* tidy up */ |
| 560 | pagevec_lru_add_file(&lru_pvec); |
| 561 | |
| 562 | if (newpage) |
| 563 | page_cache_release(newpage); |
| 564 | if (netpage) |
| 565 | page_cache_release(netpage); |
| 566 | if (backpage) |
| 567 | page_cache_release(backpage); |
| 568 | if (monitor) { |
| 569 | fscache_put_retrieval(op); |
| 570 | kfree(monitor); |
| 571 | } |
| 572 | |
| 573 | list_for_each_entry_safe(netpage, _n, list, lru) { |
| 574 | list_del(&netpage->lru); |
| 575 | page_cache_release(netpage); |
| 576 | } |
| 577 | |
| 578 | _leave(" = %d", ret); |
| 579 | return ret; |
| 580 | |
| 581 | nomem: |
| 582 | _debug("nomem"); |
| 583 | ret = -ENOMEM; |
| 584 | goto out; |
| 585 | |
| 586 | read_error: |
| 587 | _debug("read error %d", ret); |
| 588 | if (ret == -ENOMEM) |
| 589 | goto out; |
| 590 | io_error: |
| 591 | cachefiles_io_error_obj(object, "Page read error on backing file"); |
| 592 | ret = -ENOBUFS; |
| 593 | goto out; |
| 594 | } |
| 595 | |
| 596 | /* |
| 597 | * read a list of pages from the cache or allocate blocks in which to store |
| 598 | * them |
| 599 | */ |
| 600 | int cachefiles_read_or_alloc_pages(struct fscache_retrieval *op, |
| 601 | struct list_head *pages, |
| 602 | unsigned *nr_pages, |
| 603 | gfp_t gfp) |
| 604 | { |
| 605 | struct cachefiles_object *object; |
| 606 | struct cachefiles_cache *cache; |
| 607 | struct list_head backpages; |
| 608 | struct pagevec pagevec; |
| 609 | struct inode *inode; |
| 610 | struct page *page, *_n; |
| 611 | unsigned shift, nrbackpages; |
| 612 | int ret, ret2, space; |
| 613 | |
| 614 | object = container_of(op->op.object, |
| 615 | struct cachefiles_object, fscache); |
| 616 | cache = container_of(object->fscache.cache, |
| 617 | struct cachefiles_cache, cache); |
| 618 | |
| 619 | _enter("{OBJ%x,%d},,%d,,", |
| 620 | object->fscache.debug_id, atomic_read(&op->op.usage), |
| 621 | *nr_pages); |
| 622 | |
| 623 | if (!object->backer) |
| 624 | return -ENOBUFS; |
| 625 | |
| 626 | space = 1; |
| 627 | if (cachefiles_has_space(cache, 0, *nr_pages) < 0) |
| 628 | space = 0; |
| 629 | |
| 630 | inode = object->backer->d_inode; |
| 631 | ASSERT(S_ISREG(inode->i_mode)); |
| 632 | ASSERT(inode->i_mapping->a_ops->bmap); |
| 633 | ASSERT(inode->i_mapping->a_ops->readpages); |
| 634 | |
| 635 | /* calculate the shift required to use bmap */ |
| 636 | if (inode->i_sb->s_blocksize > PAGE_SIZE) |
| 637 | return -ENOBUFS; |
| 638 | |
| 639 | shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits; |
| 640 | |
| 641 | pagevec_init(&pagevec, 0); |
| 642 | |
David Howells | 4fbf429 | 2009-11-19 18:11:04 +0000 | [diff] [blame] | 643 | op->op.flags &= FSCACHE_OP_KEEP_FLAGS; |
| 644 | op->op.flags |= FSCACHE_OP_FAST; |
David Howells | 9ae326a | 2009-04-03 16:42:41 +0100 | [diff] [blame] | 645 | op->op.processor = cachefiles_read_copier; |
| 646 | |
| 647 | INIT_LIST_HEAD(&backpages); |
| 648 | nrbackpages = 0; |
| 649 | |
| 650 | ret = space ? -ENODATA : -ENOBUFS; |
| 651 | list_for_each_entry_safe(page, _n, pages, lru) { |
| 652 | sector_t block0, block; |
| 653 | |
| 654 | /* we assume the absence or presence of the first block is a |
| 655 | * good enough indication for the page as a whole |
| 656 | * - TODO: don't use bmap() for this as it is _not_ actually |
| 657 | * good enough for this as it doesn't indicate errors, but |
| 658 | * it's all we've got for the moment |
| 659 | */ |
| 660 | block0 = page->index; |
| 661 | block0 <<= shift; |
| 662 | |
| 663 | block = inode->i_mapping->a_ops->bmap(inode->i_mapping, |
| 664 | block0); |
| 665 | _debug("%llx -> %llx", |
| 666 | (unsigned long long) block0, |
| 667 | (unsigned long long) block); |
| 668 | |
| 669 | if (block) { |
| 670 | /* we have data - add it to the list to give to the |
| 671 | * backing fs */ |
| 672 | list_move(&page->lru, &backpages); |
| 673 | (*nr_pages)--; |
| 674 | nrbackpages++; |
| 675 | } else if (space && pagevec_add(&pagevec, page) == 0) { |
| 676 | fscache_mark_pages_cached(op, &pagevec); |
| 677 | ret = -ENODATA; |
| 678 | } |
| 679 | } |
| 680 | |
| 681 | if (pagevec_count(&pagevec) > 0) |
| 682 | fscache_mark_pages_cached(op, &pagevec); |
| 683 | |
| 684 | if (list_empty(pages)) |
| 685 | ret = 0; |
| 686 | |
| 687 | /* submit the apparently valid pages to the backing fs to be read from |
| 688 | * disk */ |
| 689 | if (nrbackpages > 0) { |
| 690 | ret2 = cachefiles_read_backing_file(object, op, &backpages, |
| 691 | &pagevec); |
| 692 | if (ret2 == -ENOMEM || ret2 == -EINTR) |
| 693 | ret = ret2; |
| 694 | } |
| 695 | |
| 696 | if (pagevec_count(&pagevec) > 0) |
| 697 | fscache_mark_pages_cached(op, &pagevec); |
| 698 | |
| 699 | _leave(" = %d [nr=%u%s]", |
| 700 | ret, *nr_pages, list_empty(pages) ? " empty" : ""); |
| 701 | return ret; |
| 702 | } |
| 703 | |
| 704 | /* |
| 705 | * allocate a block in the cache in which to store a page |
| 706 | * - cache withdrawal is prevented by the caller |
| 707 | * - returns -EINTR if interrupted |
| 708 | * - returns -ENOMEM if ran out of memory |
| 709 | * - returns -ENOBUFS if no buffers can be made available |
| 710 | * - returns -ENOBUFS if page is beyond EOF |
| 711 | * - otherwise: |
| 712 | * - the metadata will be retained |
| 713 | * - 0 will be returned |
| 714 | */ |
| 715 | int cachefiles_allocate_page(struct fscache_retrieval *op, |
| 716 | struct page *page, |
| 717 | gfp_t gfp) |
| 718 | { |
| 719 | struct cachefiles_object *object; |
| 720 | struct cachefiles_cache *cache; |
| 721 | struct pagevec pagevec; |
| 722 | int ret; |
| 723 | |
| 724 | object = container_of(op->op.object, |
| 725 | struct cachefiles_object, fscache); |
| 726 | cache = container_of(object->fscache.cache, |
| 727 | struct cachefiles_cache, cache); |
| 728 | |
| 729 | _enter("%p,{%lx},", object, page->index); |
| 730 | |
| 731 | ret = cachefiles_has_space(cache, 0, 1); |
| 732 | if (ret == 0) { |
| 733 | pagevec_init(&pagevec, 0); |
| 734 | pagevec_add(&pagevec, page); |
| 735 | fscache_mark_pages_cached(op, &pagevec); |
| 736 | } else { |
| 737 | ret = -ENOBUFS; |
| 738 | } |
| 739 | |
| 740 | _leave(" = %d", ret); |
| 741 | return ret; |
| 742 | } |
| 743 | |
| 744 | /* |
| 745 | * allocate blocks in the cache in which to store a set of pages |
| 746 | * - cache withdrawal is prevented by the caller |
| 747 | * - returns -EINTR if interrupted |
| 748 | * - returns -ENOMEM if ran out of memory |
| 749 | * - returns -ENOBUFS if some buffers couldn't be made available |
| 750 | * - returns -ENOBUFS if some pages are beyond EOF |
| 751 | * - otherwise: |
| 752 | * - -ENODATA will be returned |
| 753 | * - metadata will be retained for any page marked |
| 754 | */ |
| 755 | int cachefiles_allocate_pages(struct fscache_retrieval *op, |
| 756 | struct list_head *pages, |
| 757 | unsigned *nr_pages, |
| 758 | gfp_t gfp) |
| 759 | { |
| 760 | struct cachefiles_object *object; |
| 761 | struct cachefiles_cache *cache; |
| 762 | struct pagevec pagevec; |
| 763 | struct page *page; |
| 764 | int ret; |
| 765 | |
| 766 | object = container_of(op->op.object, |
| 767 | struct cachefiles_object, fscache); |
| 768 | cache = container_of(object->fscache.cache, |
| 769 | struct cachefiles_cache, cache); |
| 770 | |
| 771 | _enter("%p,,,%d,", object, *nr_pages); |
| 772 | |
| 773 | ret = cachefiles_has_space(cache, 0, *nr_pages); |
| 774 | if (ret == 0) { |
| 775 | pagevec_init(&pagevec, 0); |
| 776 | |
| 777 | list_for_each_entry(page, pages, lru) { |
| 778 | if (pagevec_add(&pagevec, page) == 0) |
| 779 | fscache_mark_pages_cached(op, &pagevec); |
| 780 | } |
| 781 | |
| 782 | if (pagevec_count(&pagevec) > 0) |
| 783 | fscache_mark_pages_cached(op, &pagevec); |
| 784 | ret = -ENODATA; |
| 785 | } else { |
| 786 | ret = -ENOBUFS; |
| 787 | } |
| 788 | |
| 789 | _leave(" = %d", ret); |
| 790 | return ret; |
| 791 | } |
| 792 | |
| 793 | /* |
| 794 | * request a page be stored in the cache |
| 795 | * - cache withdrawal is prevented by the caller |
| 796 | * - this request may be ignored if there's no cache block available, in which |
| 797 | * case -ENOBUFS will be returned |
| 798 | * - if the op is in progress, 0 will be returned |
| 799 | */ |
| 800 | int cachefiles_write_page(struct fscache_storage *op, struct page *page) |
| 801 | { |
| 802 | struct cachefiles_object *object; |
| 803 | struct cachefiles_cache *cache; |
| 804 | mm_segment_t old_fs; |
| 805 | struct file *file; |
David Howells | a17754f | 2009-11-19 18:11:52 +0000 | [diff] [blame^] | 806 | loff_t pos, eof; |
| 807 | size_t len; |
David Howells | 9ae326a | 2009-04-03 16:42:41 +0100 | [diff] [blame] | 808 | void *data; |
| 809 | int ret; |
| 810 | |
| 811 | ASSERT(op != NULL); |
| 812 | ASSERT(page != NULL); |
| 813 | |
| 814 | object = container_of(op->op.object, |
| 815 | struct cachefiles_object, fscache); |
| 816 | |
| 817 | _enter("%p,%p{%lx},,,", object, page, page->index); |
| 818 | |
| 819 | if (!object->backer) { |
| 820 | _leave(" = -ENOBUFS"); |
| 821 | return -ENOBUFS; |
| 822 | } |
| 823 | |
| 824 | ASSERT(S_ISREG(object->backer->d_inode->i_mode)); |
| 825 | |
| 826 | cache = container_of(object->fscache.cache, |
| 827 | struct cachefiles_cache, cache); |
| 828 | |
| 829 | /* write the page to the backing filesystem and let it store it in its |
| 830 | * own time */ |
| 831 | dget(object->backer); |
| 832 | mntget(cache->mnt); |
| 833 | file = dentry_open(object->backer, cache->mnt, O_RDWR, |
| 834 | cache->cache_cred); |
| 835 | if (IS_ERR(file)) { |
| 836 | ret = PTR_ERR(file); |
| 837 | } else { |
| 838 | ret = -EIO; |
| 839 | if (file->f_op->write) { |
| 840 | pos = (loff_t) page->index << PAGE_SHIFT; |
David Howells | a17754f | 2009-11-19 18:11:52 +0000 | [diff] [blame^] | 841 | |
| 842 | /* we mustn't write more data than we have, so we have |
| 843 | * to beware of a partial page at EOF */ |
| 844 | eof = object->fscache.store_limit_l; |
| 845 | len = PAGE_SIZE; |
| 846 | if (eof & ~PAGE_MASK) { |
| 847 | ASSERTCMP(pos, <, eof); |
| 848 | if (eof - pos < PAGE_SIZE) { |
| 849 | _debug("cut short %llx to %llx", |
| 850 | pos, eof); |
| 851 | len = eof - pos; |
| 852 | ASSERTCMP(pos + len, ==, eof); |
| 853 | } |
| 854 | } |
| 855 | |
David Howells | 9ae326a | 2009-04-03 16:42:41 +0100 | [diff] [blame] | 856 | data = kmap(page); |
| 857 | old_fs = get_fs(); |
| 858 | set_fs(KERNEL_DS); |
| 859 | ret = file->f_op->write( |
David Howells | a17754f | 2009-11-19 18:11:52 +0000 | [diff] [blame^] | 860 | file, (const void __user *) data, len, &pos); |
David Howells | 9ae326a | 2009-04-03 16:42:41 +0100 | [diff] [blame] | 861 | set_fs(old_fs); |
| 862 | kunmap(page); |
David Howells | a17754f | 2009-11-19 18:11:52 +0000 | [diff] [blame^] | 863 | if (ret != len) |
David Howells | 9ae326a | 2009-04-03 16:42:41 +0100 | [diff] [blame] | 864 | ret = -EIO; |
| 865 | } |
| 866 | fput(file); |
| 867 | } |
| 868 | |
| 869 | if (ret < 0) { |
| 870 | if (ret == -EIO) |
| 871 | cachefiles_io_error_obj( |
| 872 | object, "Write page to backing file failed"); |
| 873 | ret = -ENOBUFS; |
| 874 | } |
| 875 | |
| 876 | _leave(" = %d", ret); |
| 877 | return ret; |
| 878 | } |
| 879 | |
| 880 | /* |
| 881 | * detach a backing block from a page |
| 882 | * - cache withdrawal is prevented by the caller |
| 883 | */ |
| 884 | void cachefiles_uncache_page(struct fscache_object *_object, struct page *page) |
| 885 | { |
| 886 | struct cachefiles_object *object; |
| 887 | struct cachefiles_cache *cache; |
| 888 | |
| 889 | object = container_of(_object, struct cachefiles_object, fscache); |
| 890 | cache = container_of(object->fscache.cache, |
| 891 | struct cachefiles_cache, cache); |
| 892 | |
| 893 | _enter("%p,{%lu}", object, page->index); |
| 894 | |
| 895 | spin_unlock(&object->fscache.cookie->lock); |
| 896 | } |