rjw | 1f88458 | 2022-01-06 17:20:42 +0800 | [diff] [blame^] | 1 | /* AFS filesystem file handling |
| 2 | * |
| 3 | * Copyright (C) 2002, 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 License |
| 8 | * as published by the Free Software Foundation; either version |
| 9 | * 2 of the License, or (at your option) any later version. |
| 10 | */ |
| 11 | |
| 12 | #include <linux/kernel.h> |
| 13 | #include <linux/module.h> |
| 14 | #include <linux/init.h> |
| 15 | #include <linux/fs.h> |
| 16 | #include <linux/pagemap.h> |
| 17 | #include <linux/writeback.h> |
| 18 | #include <linux/gfp.h> |
| 19 | #include <linux/task_io_accounting_ops.h> |
| 20 | #include "internal.h" |
| 21 | |
| 22 | static int afs_readpage(struct file *file, struct page *page); |
| 23 | static void afs_invalidatepage(struct page *page, unsigned int offset, |
| 24 | unsigned int length); |
| 25 | static int afs_releasepage(struct page *page, gfp_t gfp_flags); |
| 26 | static int afs_launder_page(struct page *page); |
| 27 | |
| 28 | static int afs_readpages(struct file *filp, struct address_space *mapping, |
| 29 | struct list_head *pages, unsigned nr_pages); |
| 30 | |
| 31 | const struct file_operations afs_file_operations = { |
| 32 | .open = afs_open, |
| 33 | .flush = afs_flush, |
| 34 | .release = afs_release, |
| 35 | .llseek = generic_file_llseek, |
| 36 | .read_iter = generic_file_read_iter, |
| 37 | .write_iter = afs_file_write, |
| 38 | .mmap = generic_file_readonly_mmap, |
| 39 | .splice_read = generic_file_splice_read, |
| 40 | .fsync = afs_fsync, |
| 41 | .lock = afs_lock, |
| 42 | .flock = afs_flock, |
| 43 | }; |
| 44 | |
| 45 | const struct inode_operations afs_file_inode_operations = { |
| 46 | .getattr = afs_getattr, |
| 47 | .setattr = afs_setattr, |
| 48 | .permission = afs_permission, |
| 49 | .listxattr = afs_listxattr, |
| 50 | }; |
| 51 | |
| 52 | const struct address_space_operations afs_fs_aops = { |
| 53 | .readpage = afs_readpage, |
| 54 | .readpages = afs_readpages, |
| 55 | .set_page_dirty = afs_set_page_dirty, |
| 56 | .launder_page = afs_launder_page, |
| 57 | .releasepage = afs_releasepage, |
| 58 | .invalidatepage = afs_invalidatepage, |
| 59 | .write_begin = afs_write_begin, |
| 60 | .write_end = afs_write_end, |
| 61 | .writepage = afs_writepage, |
| 62 | .writepages = afs_writepages, |
| 63 | }; |
| 64 | |
| 65 | /* |
| 66 | * open an AFS file or directory and attach a key to it |
| 67 | */ |
| 68 | int afs_open(struct inode *inode, struct file *file) |
| 69 | { |
| 70 | struct afs_vnode *vnode = AFS_FS_I(inode); |
| 71 | struct key *key; |
| 72 | int ret; |
| 73 | |
| 74 | _enter("{%x:%u},", vnode->fid.vid, vnode->fid.vnode); |
| 75 | |
| 76 | key = afs_request_key(vnode->volume->cell); |
| 77 | if (IS_ERR(key)) { |
| 78 | _leave(" = %ld [key]", PTR_ERR(key)); |
| 79 | return PTR_ERR(key); |
| 80 | } |
| 81 | |
| 82 | ret = afs_validate(vnode, key); |
| 83 | if (ret < 0) { |
| 84 | _leave(" = %d [val]", ret); |
| 85 | return ret; |
| 86 | } |
| 87 | |
| 88 | file->private_data = key; |
| 89 | _leave(" = 0"); |
| 90 | return 0; |
| 91 | } |
| 92 | |
| 93 | /* |
| 94 | * release an AFS file or directory and discard its key |
| 95 | */ |
| 96 | int afs_release(struct inode *inode, struct file *file) |
| 97 | { |
| 98 | struct afs_vnode *vnode = AFS_FS_I(inode); |
| 99 | |
| 100 | _enter("{%x:%u},", vnode->fid.vid, vnode->fid.vnode); |
| 101 | |
| 102 | key_put(file->private_data); |
| 103 | _leave(" = 0"); |
| 104 | return 0; |
| 105 | } |
| 106 | |
| 107 | /* |
| 108 | * Dispose of a ref to a read record. |
| 109 | */ |
| 110 | void afs_put_read(struct afs_read *req) |
| 111 | { |
| 112 | int i; |
| 113 | |
| 114 | if (atomic_dec_and_test(&req->usage)) { |
| 115 | for (i = 0; i < req->nr_pages; i++) |
| 116 | if (req->pages[i]) |
| 117 | put_page(req->pages[i]); |
| 118 | kfree(req); |
| 119 | } |
| 120 | } |
| 121 | |
| 122 | #ifdef CONFIG_AFS_FSCACHE |
| 123 | /* |
| 124 | * deal with notification that a page was read from the cache |
| 125 | */ |
| 126 | static void afs_file_readpage_read_complete(struct page *page, |
| 127 | void *data, |
| 128 | int error) |
| 129 | { |
| 130 | _enter("%p,%p,%d", page, data, error); |
| 131 | |
| 132 | /* if the read completes with an error, we just unlock the page and let |
| 133 | * the VM reissue the readpage */ |
| 134 | if (!error) |
| 135 | SetPageUptodate(page); |
| 136 | unlock_page(page); |
| 137 | } |
| 138 | #endif |
| 139 | |
| 140 | /* |
| 141 | * read page from file, directory or symlink, given a key to use |
| 142 | */ |
| 143 | static int __afs_page_filler(struct key *key, struct page *page) |
| 144 | { |
| 145 | struct inode *inode = page->mapping->host; |
| 146 | struct afs_vnode *vnode = AFS_FS_I(inode); |
| 147 | struct afs_read *req; |
| 148 | int ret; |
| 149 | |
| 150 | _enter("{%x},{%lu},{%lu}", key_serial(key), inode->i_ino, page->index); |
| 151 | |
| 152 | BUG_ON(!PageLocked(page)); |
| 153 | |
| 154 | ret = -ESTALE; |
| 155 | if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) |
| 156 | goto error; |
| 157 | |
| 158 | /* is it cached? */ |
| 159 | #ifdef CONFIG_AFS_FSCACHE |
| 160 | ret = fscache_read_or_alloc_page(vnode->cache, |
| 161 | page, |
| 162 | afs_file_readpage_read_complete, |
| 163 | NULL, |
| 164 | GFP_KERNEL); |
| 165 | #else |
| 166 | ret = -ENOBUFS; |
| 167 | #endif |
| 168 | switch (ret) { |
| 169 | /* read BIO submitted (page in cache) */ |
| 170 | case 0: |
| 171 | break; |
| 172 | |
| 173 | /* page not yet cached */ |
| 174 | case -ENODATA: |
| 175 | _debug("cache said ENODATA"); |
| 176 | goto go_on; |
| 177 | |
| 178 | /* page will not be cached */ |
| 179 | case -ENOBUFS: |
| 180 | _debug("cache said ENOBUFS"); |
| 181 | default: |
| 182 | go_on: |
| 183 | req = kzalloc(sizeof(struct afs_read) + sizeof(struct page *), |
| 184 | GFP_KERNEL); |
| 185 | if (!req) |
| 186 | goto enomem; |
| 187 | |
| 188 | /* We request a full page. If the page is a partial one at the |
| 189 | * end of the file, the server will return a short read and the |
| 190 | * unmarshalling code will clear the unfilled space. |
| 191 | */ |
| 192 | atomic_set(&req->usage, 1); |
| 193 | req->pos = (loff_t)page->index << PAGE_SHIFT; |
| 194 | req->len = PAGE_SIZE; |
| 195 | req->nr_pages = 1; |
| 196 | req->pages[0] = page; |
| 197 | get_page(page); |
| 198 | |
| 199 | /* read the contents of the file from the server into the |
| 200 | * page */ |
| 201 | ret = afs_vnode_fetch_data(vnode, key, req); |
| 202 | afs_put_read(req); |
| 203 | if (ret < 0) { |
| 204 | if (ret == -ENOENT) { |
| 205 | _debug("got NOENT from server" |
| 206 | " - marking file deleted and stale"); |
| 207 | set_bit(AFS_VNODE_DELETED, &vnode->flags); |
| 208 | ret = -ESTALE; |
| 209 | } |
| 210 | |
| 211 | #ifdef CONFIG_AFS_FSCACHE |
| 212 | fscache_uncache_page(vnode->cache, page); |
| 213 | #endif |
| 214 | BUG_ON(PageFsCache(page)); |
| 215 | |
| 216 | if (ret == -EINTR || |
| 217 | ret == -ENOMEM || |
| 218 | ret == -ERESTARTSYS || |
| 219 | ret == -EAGAIN) |
| 220 | goto error; |
| 221 | goto io_error; |
| 222 | } |
| 223 | |
| 224 | SetPageUptodate(page); |
| 225 | |
| 226 | /* send the page to the cache */ |
| 227 | #ifdef CONFIG_AFS_FSCACHE |
| 228 | if (PageFsCache(page) && |
| 229 | fscache_write_page(vnode->cache, page, GFP_KERNEL) != 0) { |
| 230 | fscache_uncache_page(vnode->cache, page); |
| 231 | BUG_ON(PageFsCache(page)); |
| 232 | } |
| 233 | #endif |
| 234 | unlock_page(page); |
| 235 | } |
| 236 | |
| 237 | _leave(" = 0"); |
| 238 | return 0; |
| 239 | |
| 240 | io_error: |
| 241 | SetPageError(page); |
| 242 | goto error; |
| 243 | enomem: |
| 244 | ret = -ENOMEM; |
| 245 | error: |
| 246 | unlock_page(page); |
| 247 | _leave(" = %d", ret); |
| 248 | return ret; |
| 249 | } |
| 250 | |
| 251 | int afs_page_filler(struct file *data, struct page *page) |
| 252 | { |
| 253 | struct key *key = (struct key *)data; |
| 254 | |
| 255 | return __afs_page_filler(key, page); |
| 256 | } |
| 257 | |
| 258 | /* |
| 259 | * read page from file, directory or symlink, given a file to nominate the key |
| 260 | * to be used |
| 261 | */ |
| 262 | static int afs_readpage(struct file *file, struct page *page) |
| 263 | { |
| 264 | struct key *key; |
| 265 | int ret; |
| 266 | |
| 267 | if (file) { |
| 268 | key = file->private_data; |
| 269 | ASSERT(key != NULL); |
| 270 | ret = __afs_page_filler(key, page); |
| 271 | } else { |
| 272 | struct inode *inode = page->mapping->host; |
| 273 | key = afs_request_key(AFS_FS_S(inode->i_sb)->volume->cell); |
| 274 | if (IS_ERR(key)) { |
| 275 | ret = PTR_ERR(key); |
| 276 | } else { |
| 277 | ret = __afs_page_filler(key, page); |
| 278 | key_put(key); |
| 279 | } |
| 280 | } |
| 281 | return ret; |
| 282 | } |
| 283 | |
| 284 | /* |
| 285 | * Make pages available as they're filled. |
| 286 | */ |
| 287 | static void afs_readpages_page_done(struct afs_call *call, struct afs_read *req) |
| 288 | { |
| 289 | #ifdef CONFIG_AFS_FSCACHE |
| 290 | struct afs_vnode *vnode = call->reply; |
| 291 | #endif |
| 292 | struct page *page = req->pages[req->index]; |
| 293 | |
| 294 | req->pages[req->index] = NULL; |
| 295 | SetPageUptodate(page); |
| 296 | |
| 297 | /* send the page to the cache */ |
| 298 | #ifdef CONFIG_AFS_FSCACHE |
| 299 | if (PageFsCache(page) && |
| 300 | fscache_write_page(vnode->cache, page, GFP_KERNEL) != 0) { |
| 301 | fscache_uncache_page(vnode->cache, page); |
| 302 | BUG_ON(PageFsCache(page)); |
| 303 | } |
| 304 | #endif |
| 305 | unlock_page(page); |
| 306 | put_page(page); |
| 307 | } |
| 308 | |
| 309 | /* |
| 310 | * Read a contiguous set of pages. |
| 311 | */ |
| 312 | static int afs_readpages_one(struct file *file, struct address_space *mapping, |
| 313 | struct list_head *pages) |
| 314 | { |
| 315 | struct afs_vnode *vnode = AFS_FS_I(mapping->host); |
| 316 | struct afs_read *req; |
| 317 | struct list_head *p; |
| 318 | struct page *first, *page; |
| 319 | struct key *key = file->private_data; |
| 320 | pgoff_t index; |
| 321 | int ret, n, i; |
| 322 | |
| 323 | /* Count the number of contiguous pages at the front of the list. Note |
| 324 | * that the list goes prev-wards rather than next-wards. |
| 325 | */ |
| 326 | first = list_entry(pages->prev, struct page, lru); |
| 327 | index = first->index + 1; |
| 328 | n = 1; |
| 329 | for (p = first->lru.prev; p != pages; p = p->prev) { |
| 330 | page = list_entry(p, struct page, lru); |
| 331 | if (page->index != index) |
| 332 | break; |
| 333 | index++; |
| 334 | n++; |
| 335 | } |
| 336 | |
| 337 | req = kzalloc(sizeof(struct afs_read) + sizeof(struct page *) * n, |
| 338 | GFP_NOFS); |
| 339 | if (!req) |
| 340 | return -ENOMEM; |
| 341 | |
| 342 | atomic_set(&req->usage, 1); |
| 343 | req->page_done = afs_readpages_page_done; |
| 344 | req->pos = first->index; |
| 345 | req->pos <<= PAGE_SHIFT; |
| 346 | |
| 347 | /* Transfer the pages to the request. We add them in until one fails |
| 348 | * to add to the LRU and then we stop (as that'll make a hole in the |
| 349 | * contiguous run. |
| 350 | * |
| 351 | * Note that it's possible for the file size to change whilst we're |
| 352 | * doing this, but we rely on the server returning less than we asked |
| 353 | * for if the file shrank. We also rely on this to deal with a partial |
| 354 | * page at the end of the file. |
| 355 | */ |
| 356 | do { |
| 357 | page = list_entry(pages->prev, struct page, lru); |
| 358 | list_del(&page->lru); |
| 359 | index = page->index; |
| 360 | if (add_to_page_cache_lru(page, mapping, index, |
| 361 | readahead_gfp_mask(mapping))) { |
| 362 | #ifdef CONFIG_AFS_FSCACHE |
| 363 | fscache_uncache_page(vnode->cache, page); |
| 364 | #endif |
| 365 | put_page(page); |
| 366 | break; |
| 367 | } |
| 368 | |
| 369 | req->pages[req->nr_pages++] = page; |
| 370 | req->len += PAGE_SIZE; |
| 371 | } while (req->nr_pages < n); |
| 372 | |
| 373 | if (req->nr_pages == 0) { |
| 374 | kfree(req); |
| 375 | return 0; |
| 376 | } |
| 377 | |
| 378 | ret = afs_vnode_fetch_data(vnode, key, req); |
| 379 | if (ret < 0) |
| 380 | goto error; |
| 381 | |
| 382 | task_io_account_read(PAGE_SIZE * req->nr_pages); |
| 383 | afs_put_read(req); |
| 384 | return 0; |
| 385 | |
| 386 | error: |
| 387 | if (ret == -ENOENT) { |
| 388 | _debug("got NOENT from server" |
| 389 | " - marking file deleted and stale"); |
| 390 | set_bit(AFS_VNODE_DELETED, &vnode->flags); |
| 391 | ret = -ESTALE; |
| 392 | } |
| 393 | |
| 394 | for (i = 0; i < req->nr_pages; i++) { |
| 395 | page = req->pages[i]; |
| 396 | if (page) { |
| 397 | #ifdef CONFIG_AFS_FSCACHE |
| 398 | fscache_uncache_page(vnode->cache, page); |
| 399 | #endif |
| 400 | SetPageError(page); |
| 401 | unlock_page(page); |
| 402 | } |
| 403 | } |
| 404 | |
| 405 | afs_put_read(req); |
| 406 | return ret; |
| 407 | } |
| 408 | |
| 409 | /* |
| 410 | * read a set of pages |
| 411 | */ |
| 412 | static int afs_readpages(struct file *file, struct address_space *mapping, |
| 413 | struct list_head *pages, unsigned nr_pages) |
| 414 | { |
| 415 | struct key *key = file->private_data; |
| 416 | struct afs_vnode *vnode; |
| 417 | int ret = 0; |
| 418 | |
| 419 | _enter("{%d},{%lu},,%d", |
| 420 | key_serial(key), mapping->host->i_ino, nr_pages); |
| 421 | |
| 422 | ASSERT(key != NULL); |
| 423 | |
| 424 | vnode = AFS_FS_I(mapping->host); |
| 425 | if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) { |
| 426 | _leave(" = -ESTALE"); |
| 427 | return -ESTALE; |
| 428 | } |
| 429 | |
| 430 | /* attempt to read as many of the pages as possible */ |
| 431 | #ifdef CONFIG_AFS_FSCACHE |
| 432 | ret = fscache_read_or_alloc_pages(vnode->cache, |
| 433 | mapping, |
| 434 | pages, |
| 435 | &nr_pages, |
| 436 | afs_file_readpage_read_complete, |
| 437 | NULL, |
| 438 | mapping_gfp_mask(mapping)); |
| 439 | #else |
| 440 | ret = -ENOBUFS; |
| 441 | #endif |
| 442 | |
| 443 | switch (ret) { |
| 444 | /* all pages are being read from the cache */ |
| 445 | case 0: |
| 446 | BUG_ON(!list_empty(pages)); |
| 447 | BUG_ON(nr_pages != 0); |
| 448 | _leave(" = 0 [reading all]"); |
| 449 | return 0; |
| 450 | |
| 451 | /* there were pages that couldn't be read from the cache */ |
| 452 | case -ENODATA: |
| 453 | case -ENOBUFS: |
| 454 | break; |
| 455 | |
| 456 | /* other error */ |
| 457 | default: |
| 458 | _leave(" = %d", ret); |
| 459 | return ret; |
| 460 | } |
| 461 | |
| 462 | while (!list_empty(pages)) { |
| 463 | ret = afs_readpages_one(file, mapping, pages); |
| 464 | if (ret < 0) |
| 465 | break; |
| 466 | } |
| 467 | |
| 468 | _leave(" = %d [netting]", ret); |
| 469 | return ret; |
| 470 | } |
| 471 | |
| 472 | /* |
| 473 | * write back a dirty page |
| 474 | */ |
| 475 | static int afs_launder_page(struct page *page) |
| 476 | { |
| 477 | _enter("{%lu}", page->index); |
| 478 | |
| 479 | return 0; |
| 480 | } |
| 481 | |
| 482 | /* |
| 483 | * invalidate part or all of a page |
| 484 | * - release a page and clean up its private data if offset is 0 (indicating |
| 485 | * the entire page) |
| 486 | */ |
| 487 | static void afs_invalidatepage(struct page *page, unsigned int offset, |
| 488 | unsigned int length) |
| 489 | { |
| 490 | struct afs_writeback *wb = (struct afs_writeback *) page_private(page); |
| 491 | |
| 492 | _enter("{%lu},%u,%u", page->index, offset, length); |
| 493 | |
| 494 | BUG_ON(!PageLocked(page)); |
| 495 | |
| 496 | /* we clean up only if the entire page is being invalidated */ |
| 497 | if (offset == 0 && length == PAGE_SIZE) { |
| 498 | #ifdef CONFIG_AFS_FSCACHE |
| 499 | if (PageFsCache(page)) { |
| 500 | struct afs_vnode *vnode = AFS_FS_I(page->mapping->host); |
| 501 | fscache_wait_on_page_write(vnode->cache, page); |
| 502 | fscache_uncache_page(vnode->cache, page); |
| 503 | } |
| 504 | #endif |
| 505 | |
| 506 | if (PagePrivate(page)) { |
| 507 | if (wb && !PageWriteback(page)) { |
| 508 | set_page_private(page, 0); |
| 509 | afs_put_writeback(wb); |
| 510 | } |
| 511 | |
| 512 | if (!page_private(page)) |
| 513 | ClearPagePrivate(page); |
| 514 | } |
| 515 | } |
| 516 | |
| 517 | _leave(""); |
| 518 | } |
| 519 | |
| 520 | /* |
| 521 | * release a page and clean up its private state if it's not busy |
| 522 | * - return true if the page can now be released, false if not |
| 523 | */ |
| 524 | static int afs_releasepage(struct page *page, gfp_t gfp_flags) |
| 525 | { |
| 526 | struct afs_writeback *wb = (struct afs_writeback *) page_private(page); |
| 527 | struct afs_vnode *vnode = AFS_FS_I(page->mapping->host); |
| 528 | |
| 529 | _enter("{{%x:%u}[%lu],%lx},%x", |
| 530 | vnode->fid.vid, vnode->fid.vnode, page->index, page->flags, |
| 531 | gfp_flags); |
| 532 | |
| 533 | /* deny if page is being written to the cache and the caller hasn't |
| 534 | * elected to wait */ |
| 535 | #ifdef CONFIG_AFS_FSCACHE |
| 536 | if (!fscache_maybe_release_page(vnode->cache, page, gfp_flags)) { |
| 537 | _leave(" = F [cache busy]"); |
| 538 | return 0; |
| 539 | } |
| 540 | #endif |
| 541 | |
| 542 | if (PagePrivate(page)) { |
| 543 | if (wb) { |
| 544 | set_page_private(page, 0); |
| 545 | afs_put_writeback(wb); |
| 546 | } |
| 547 | ClearPagePrivate(page); |
| 548 | } |
| 549 | |
| 550 | /* indicate that the page can be released */ |
| 551 | _leave(" = T"); |
| 552 | return 1; |
| 553 | } |