1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * linux/fs/nfs/file.c 4 * 5 * Copyright (C) 1992 Rick Sladkey 6 * 7 * Changes Copyright (C) 1994 by Florian La Roche 8 * - Do not copy data too often around in the kernel. 9 * - In nfs_file_read the return value of kmalloc wasn't checked. 10 * - Put in a better version of read look-ahead buffering. Original idea 11 * and implementation by Wai S Kok elekokws@ee.nus.sg. 12 * 13 * Expire cache on write to a file by Wai S Kok (Oct 1994). 14 * 15 * Total rewrite of read side for new NFS buffer cache.. Linus. 16 * 17 * nfs regular file handling functions 18 */ 19 20 #include <linux/module.h> 21 #include <linux/time.h> 22 #include <linux/kernel.h> 23 #include <linux/errno.h> 24 #include <linux/fcntl.h> 25 #include <linux/stat.h> 26 #include <linux/nfs_fs.h> 27 #include <linux/nfs_mount.h> 28 #include <linux/mm.h> 29 #include <linux/pagemap.h> 30 #include <linux/gfp.h> 31 #include <linux/swap.h> 32 33 #include <linux/uaccess.h> 34 #include <linux/filelock.h> 35 36 #include "delegation.h" 37 #include "internal.h" 38 #include "iostat.h" 39 #include "fscache.h" 40 #include "pnfs.h" 41 42 #include "nfstrace.h" 43 44 #define NFSDBG_FACILITY NFSDBG_FILE 45 46 static const struct vm_operations_struct nfs_file_vm_ops; 47 48 int nfs_check_flags(int flags) 49 { 50 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT)) 51 return -EINVAL; 52 53 return 0; 54 } 55 EXPORT_SYMBOL_GPL(nfs_check_flags); 56 57 /* 58 * Open file 59 */ 60 static int 61 nfs_file_open(struct inode *inode, struct file *filp) 62 { 63 int res; 64 65 dprintk("NFS: open file(%pD2)\n", filp); 66 67 nfs_inc_stats(inode, NFSIOS_VFSOPEN); 68 res = nfs_check_flags(filp->f_flags); 69 if (res) 70 return res; 71 72 res = nfs_open(inode, filp); 73 if (res == 0) 74 filp->f_mode |= FMODE_CAN_ODIRECT; 75 return res; 76 } 77 78 int 79 nfs_file_release(struct inode *inode, struct file *filp) 80 { 81 dprintk("NFS: release(%pD2)\n", filp); 82 83 nfs_inc_stats(inode, NFSIOS_VFSRELEASE); 84 nfs_file_clear_open_context(filp); 85 nfs_fscache_release_file(inode, filp); 86 return 0; 87 } 88 EXPORT_SYMBOL_GPL(nfs_file_release); 89 90 /** 91 * nfs_revalidate_file_size - Revalidate the file size 92 * @inode: pointer to inode struct 93 * @filp: pointer to struct file 94 * 95 * Revalidates the file length. This is basically a wrapper around 96 * nfs_revalidate_inode() that takes into account the fact that we may 97 * have cached writes (in which case we don't care about the server's 98 * idea of what the file length is), or O_DIRECT (in which case we 99 * shouldn't trust the cache). 100 */ 101 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp) 102 { 103 struct nfs_server *server = NFS_SERVER(inode); 104 105 if (filp->f_flags & O_DIRECT) 106 goto force_reval; 107 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_SIZE)) 108 goto force_reval; 109 return 0; 110 force_reval: 111 return __nfs_revalidate_inode(server, inode); 112 } 113 114 loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence) 115 { 116 dprintk("NFS: llseek file(%pD2, %lld, %d)\n", 117 filp, offset, whence); 118 119 /* 120 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate 121 * the cached file length 122 */ 123 if (whence != SEEK_SET && whence != SEEK_CUR) { 124 struct inode *inode = filp->f_mapping->host; 125 126 int retval = nfs_revalidate_file_size(inode, filp); 127 if (retval < 0) 128 return (loff_t)retval; 129 } 130 131 return generic_file_llseek(filp, offset, whence); 132 } 133 EXPORT_SYMBOL_GPL(nfs_file_llseek); 134 135 /* 136 * Flush all dirty pages, and check for write errors. 137 */ 138 static int 139 nfs_file_flush(struct file *file, fl_owner_t id) 140 { 141 struct inode *inode = file_inode(file); 142 errseq_t since; 143 144 dprintk("NFS: flush(%pD2)\n", file); 145 146 nfs_inc_stats(inode, NFSIOS_VFSFLUSH); 147 if ((file->f_mode & FMODE_WRITE) == 0) 148 return 0; 149 150 /* Flush writes to the server and return any errors */ 151 since = filemap_sample_wb_err(file->f_mapping); 152 nfs_wb_all(inode); 153 return filemap_check_wb_err(file->f_mapping, since); 154 } 155 156 ssize_t 157 nfs_file_read(struct kiocb *iocb, struct iov_iter *to) 158 { 159 struct inode *inode = file_inode(iocb->ki_filp); 160 ssize_t result; 161 162 if (iocb->ki_flags & IOCB_DIRECT) 163 return nfs_file_direct_read(iocb, to, false); 164 165 dprintk("NFS: read(%pD2, %zu@%lu)\n", 166 iocb->ki_filp, 167 iov_iter_count(to), (unsigned long) iocb->ki_pos); 168 169 nfs_start_io_read(inode); 170 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping); 171 if (!result) { 172 result = generic_file_read_iter(iocb, to); 173 if (result > 0) 174 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result); 175 } 176 nfs_end_io_read(inode); 177 return result; 178 } 179 EXPORT_SYMBOL_GPL(nfs_file_read); 180 181 int 182 nfs_file_mmap(struct file * file, struct vm_area_struct * vma) 183 { 184 struct inode *inode = file_inode(file); 185 int status; 186 187 dprintk("NFS: mmap(%pD2)\n", file); 188 189 /* Note: generic_file_mmap() returns ENOSYS on nommu systems 190 * so we call that before revalidating the mapping 191 */ 192 status = generic_file_mmap(file, vma); 193 if (!status) { 194 vma->vm_ops = &nfs_file_vm_ops; 195 status = nfs_revalidate_mapping(inode, file->f_mapping); 196 } 197 return status; 198 } 199 EXPORT_SYMBOL_GPL(nfs_file_mmap); 200 201 /* 202 * Flush any dirty pages for this process, and check for write errors. 203 * The return status from this call provides a reliable indication of 204 * whether any write errors occurred for this process. 205 */ 206 static int 207 nfs_file_fsync_commit(struct file *file, int datasync) 208 { 209 struct inode *inode = file_inode(file); 210 int ret, ret2; 211 212 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync); 213 214 nfs_inc_stats(inode, NFSIOS_VFSFSYNC); 215 ret = nfs_commit_inode(inode, FLUSH_SYNC); 216 ret2 = file_check_and_advance_wb_err(file); 217 if (ret2 < 0) 218 return ret2; 219 return ret; 220 } 221 222 int 223 nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync) 224 { 225 struct inode *inode = file_inode(file); 226 struct nfs_inode *nfsi = NFS_I(inode); 227 long save_nredirtied = atomic_long_read(&nfsi->redirtied_pages); 228 long nredirtied; 229 int ret; 230 231 trace_nfs_fsync_enter(inode); 232 233 for (;;) { 234 ret = file_write_and_wait_range(file, start, end); 235 if (ret != 0) 236 break; 237 ret = nfs_file_fsync_commit(file, datasync); 238 if (ret != 0) 239 break; 240 ret = pnfs_sync_inode(inode, !!datasync); 241 if (ret != 0) 242 break; 243 nredirtied = atomic_long_read(&nfsi->redirtied_pages); 244 if (nredirtied == save_nredirtied) 245 break; 246 save_nredirtied = nredirtied; 247 } 248 249 trace_nfs_fsync_exit(inode, ret); 250 return ret; 251 } 252 EXPORT_SYMBOL_GPL(nfs_file_fsync); 253 254 /* 255 * Decide whether a read/modify/write cycle may be more efficient 256 * then a modify/write/read cycle when writing to a page in the 257 * page cache. 258 * 259 * Some pNFS layout drivers can only read/write at a certain block 260 * granularity like all block devices and therefore we must perform 261 * read/modify/write whenever a page hasn't read yet and the data 262 * to be written there is not aligned to a block boundary and/or 263 * smaller than the block size. 264 * 265 * The modify/write/read cycle may occur if a page is read before 266 * being completely filled by the writer. In this situation, the 267 * page must be completely written to stable storage on the server 268 * before it can be refilled by reading in the page from the server. 269 * This can lead to expensive, small, FILE_SYNC mode writes being 270 * done. 271 * 272 * It may be more efficient to read the page first if the file is 273 * open for reading in addition to writing, the page is not marked 274 * as Uptodate, it is not dirty or waiting to be committed, 275 * indicating that it was previously allocated and then modified, 276 * that there were valid bytes of data in that range of the file, 277 * and that the new data won't completely replace the old data in 278 * that range of the file. 279 */ 280 static bool nfs_full_page_write(struct page *page, loff_t pos, unsigned int len) 281 { 282 unsigned int pglen = nfs_page_length(page); 283 unsigned int offset = pos & (PAGE_SIZE - 1); 284 unsigned int end = offset + len; 285 286 return !pglen || (end >= pglen && !offset); 287 } 288 289 static bool nfs_want_read_modify_write(struct file *file, struct page *page, 290 loff_t pos, unsigned int len) 291 { 292 /* 293 * Up-to-date pages, those with ongoing or full-page write 294 * don't need read/modify/write 295 */ 296 if (PageUptodate(page) || PagePrivate(page) || 297 nfs_full_page_write(page, pos, len)) 298 return false; 299 300 if (pnfs_ld_read_whole_page(file->f_mapping->host)) 301 return true; 302 /* Open for reading too? */ 303 if (file->f_mode & FMODE_READ) 304 return true; 305 return false; 306 } 307 308 /* 309 * This does the "real" work of the write. We must allocate and lock the 310 * page to be sent back to the generic routine, which then copies the 311 * data from user space. 312 * 313 * If the writer ends up delaying the write, the writer needs to 314 * increment the page use counts until he is done with the page. 315 */ 316 static int nfs_write_begin(struct file *file, struct address_space *mapping, 317 loff_t pos, unsigned len, 318 struct page **pagep, void **fsdata) 319 { 320 int ret; 321 pgoff_t index = pos >> PAGE_SHIFT; 322 struct page *page; 323 int once_thru = 0; 324 325 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n", 326 file, mapping->host->i_ino, len, (long long) pos); 327 328 start: 329 page = grab_cache_page_write_begin(mapping, index); 330 if (!page) 331 return -ENOMEM; 332 *pagep = page; 333 334 ret = nfs_flush_incompatible(file, page); 335 if (ret) { 336 unlock_page(page); 337 put_page(page); 338 } else if (!once_thru && 339 nfs_want_read_modify_write(file, page, pos, len)) { 340 once_thru = 1; 341 ret = nfs_read_folio(file, page_folio(page)); 342 put_page(page); 343 if (!ret) 344 goto start; 345 } 346 return ret; 347 } 348 349 static int nfs_write_end(struct file *file, struct address_space *mapping, 350 loff_t pos, unsigned len, unsigned copied, 351 struct page *page, void *fsdata) 352 { 353 unsigned offset = pos & (PAGE_SIZE - 1); 354 struct nfs_open_context *ctx = nfs_file_open_context(file); 355 int status; 356 357 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n", 358 file, mapping->host->i_ino, len, (long long) pos); 359 360 /* 361 * Zero any uninitialised parts of the page, and then mark the page 362 * as up to date if it turns out that we're extending the file. 363 */ 364 if (!PageUptodate(page)) { 365 unsigned pglen = nfs_page_length(page); 366 unsigned end = offset + copied; 367 368 if (pglen == 0) { 369 zero_user_segments(page, 0, offset, 370 end, PAGE_SIZE); 371 SetPageUptodate(page); 372 } else if (end >= pglen) { 373 zero_user_segment(page, end, PAGE_SIZE); 374 if (offset == 0) 375 SetPageUptodate(page); 376 } else 377 zero_user_segment(page, pglen, PAGE_SIZE); 378 } 379 380 status = nfs_updatepage(file, page, offset, copied); 381 382 unlock_page(page); 383 put_page(page); 384 385 if (status < 0) 386 return status; 387 NFS_I(mapping->host)->write_io += copied; 388 389 if (nfs_ctx_key_to_expire(ctx, mapping->host)) 390 nfs_wb_all(mapping->host); 391 392 return copied; 393 } 394 395 /* 396 * Partially or wholly invalidate a page 397 * - Release the private state associated with a page if undergoing complete 398 * page invalidation 399 * - Called if either PG_private or PG_fscache is set on the page 400 * - Caller holds page lock 401 */ 402 static void nfs_invalidate_folio(struct folio *folio, size_t offset, 403 size_t length) 404 { 405 dfprintk(PAGECACHE, "NFS: invalidate_folio(%lu, %zu, %zu)\n", 406 folio->index, offset, length); 407 408 if (offset != 0 || length < folio_size(folio)) 409 return; 410 /* Cancel any unstarted writes on this page */ 411 nfs_wb_folio_cancel(folio->mapping->host, folio); 412 folio_wait_fscache(folio); 413 } 414 415 /* 416 * Attempt to release the private state associated with a folio 417 * - Called if either private or fscache flags are set on the folio 418 * - Caller holds folio lock 419 * - Return true (may release folio) or false (may not) 420 */ 421 static bool nfs_release_folio(struct folio *folio, gfp_t gfp) 422 { 423 dfprintk(PAGECACHE, "NFS: release_folio(%p)\n", folio); 424 425 /* If the private flag is set, then the folio is not freeable */ 426 if (folio_test_private(folio)) 427 return false; 428 return nfs_fscache_release_folio(folio, gfp); 429 } 430 431 static void nfs_check_dirty_writeback(struct folio *folio, 432 bool *dirty, bool *writeback) 433 { 434 struct nfs_inode *nfsi; 435 struct address_space *mapping = folio->mapping; 436 437 /* 438 * Check if an unstable folio is currently being committed and 439 * if so, have the VM treat it as if the folio is under writeback 440 * so it will not block due to folios that will shortly be freeable. 441 */ 442 nfsi = NFS_I(mapping->host); 443 if (atomic_read(&nfsi->commit_info.rpcs_out)) { 444 *writeback = true; 445 return; 446 } 447 448 /* 449 * If the private flag is set, then the folio is not freeable 450 * and as the inode is not being committed, it's not going to 451 * be cleaned in the near future so treat it as dirty 452 */ 453 if (folio_test_private(folio)) 454 *dirty = true; 455 } 456 457 /* 458 * Attempt to clear the private state associated with a page when an error 459 * occurs that requires the cached contents of an inode to be written back or 460 * destroyed 461 * - Called if either PG_private or fscache is set on the page 462 * - Caller holds page lock 463 * - Return 0 if successful, -error otherwise 464 */ 465 static int nfs_launder_folio(struct folio *folio) 466 { 467 struct inode *inode = folio->mapping->host; 468 469 dfprintk(PAGECACHE, "NFS: launder_folio(%ld, %llu)\n", 470 inode->i_ino, folio_pos(folio)); 471 472 folio_wait_fscache(folio); 473 return nfs_wb_page(inode, &folio->page); 474 } 475 476 static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file, 477 sector_t *span) 478 { 479 unsigned long blocks; 480 long long isize; 481 int ret; 482 struct inode *inode = file_inode(file); 483 struct rpc_clnt *clnt = NFS_CLIENT(inode); 484 struct nfs_client *cl = NFS_SERVER(inode)->nfs_client; 485 486 spin_lock(&inode->i_lock); 487 blocks = inode->i_blocks; 488 isize = inode->i_size; 489 spin_unlock(&inode->i_lock); 490 if (blocks*512 < isize) { 491 pr_warn("swap activate: swapfile has holes\n"); 492 return -EINVAL; 493 } 494 495 ret = rpc_clnt_swap_activate(clnt); 496 if (ret) 497 return ret; 498 ret = add_swap_extent(sis, 0, sis->max, 0); 499 if (ret < 0) { 500 rpc_clnt_swap_deactivate(clnt); 501 return ret; 502 } 503 504 *span = sis->pages; 505 506 if (cl->rpc_ops->enable_swap) 507 cl->rpc_ops->enable_swap(inode); 508 509 sis->flags |= SWP_FS_OPS; 510 return ret; 511 } 512 513 static void nfs_swap_deactivate(struct file *file) 514 { 515 struct inode *inode = file_inode(file); 516 struct rpc_clnt *clnt = NFS_CLIENT(inode); 517 struct nfs_client *cl = NFS_SERVER(inode)->nfs_client; 518 519 rpc_clnt_swap_deactivate(clnt); 520 if (cl->rpc_ops->disable_swap) 521 cl->rpc_ops->disable_swap(file_inode(file)); 522 } 523 524 const struct address_space_operations nfs_file_aops = { 525 .read_folio = nfs_read_folio, 526 .readahead = nfs_readahead, 527 .dirty_folio = filemap_dirty_folio, 528 .writepage = nfs_writepage, 529 .writepages = nfs_writepages, 530 .write_begin = nfs_write_begin, 531 .write_end = nfs_write_end, 532 .invalidate_folio = nfs_invalidate_folio, 533 .release_folio = nfs_release_folio, 534 .migrate_folio = nfs_migrate_folio, 535 .launder_folio = nfs_launder_folio, 536 .is_dirty_writeback = nfs_check_dirty_writeback, 537 .error_remove_page = generic_error_remove_page, 538 .swap_activate = nfs_swap_activate, 539 .swap_deactivate = nfs_swap_deactivate, 540 .swap_rw = nfs_swap_rw, 541 }; 542 543 /* 544 * Notification that a PTE pointing to an NFS page is about to be made 545 * writable, implying that someone is about to modify the page through a 546 * shared-writable mapping 547 */ 548 static vm_fault_t nfs_vm_page_mkwrite(struct vm_fault *vmf) 549 { 550 struct page *page = vmf->page; 551 struct file *filp = vmf->vma->vm_file; 552 struct inode *inode = file_inode(filp); 553 unsigned pagelen; 554 vm_fault_t ret = VM_FAULT_NOPAGE; 555 struct address_space *mapping; 556 557 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n", 558 filp, filp->f_mapping->host->i_ino, 559 (long long)page_offset(page)); 560 561 sb_start_pagefault(inode->i_sb); 562 563 /* make sure the cache has finished storing the page */ 564 if (PageFsCache(page) && 565 wait_on_page_fscache_killable(vmf->page) < 0) { 566 ret = VM_FAULT_RETRY; 567 goto out; 568 } 569 570 wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING, 571 nfs_wait_bit_killable, 572 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE); 573 574 lock_page(page); 575 mapping = page_file_mapping(page); 576 if (mapping != inode->i_mapping) 577 goto out_unlock; 578 579 wait_on_page_writeback(page); 580 581 pagelen = nfs_page_length(page); 582 if (pagelen == 0) 583 goto out_unlock; 584 585 ret = VM_FAULT_LOCKED; 586 if (nfs_flush_incompatible(filp, page) == 0 && 587 nfs_updatepage(filp, page, 0, pagelen) == 0) 588 goto out; 589 590 ret = VM_FAULT_SIGBUS; 591 out_unlock: 592 unlock_page(page); 593 out: 594 sb_end_pagefault(inode->i_sb); 595 return ret; 596 } 597 598 static const struct vm_operations_struct nfs_file_vm_ops = { 599 .fault = filemap_fault, 600 .map_pages = filemap_map_pages, 601 .page_mkwrite = nfs_vm_page_mkwrite, 602 }; 603 604 ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from) 605 { 606 struct file *file = iocb->ki_filp; 607 struct inode *inode = file_inode(file); 608 unsigned int mntflags = NFS_SERVER(inode)->flags; 609 ssize_t result, written; 610 errseq_t since; 611 int error; 612 613 result = nfs_key_timeout_notify(file, inode); 614 if (result) 615 return result; 616 617 if (iocb->ki_flags & IOCB_DIRECT) 618 return nfs_file_direct_write(iocb, from, false); 619 620 dprintk("NFS: write(%pD2, %zu@%Ld)\n", 621 file, iov_iter_count(from), (long long) iocb->ki_pos); 622 623 if (IS_SWAPFILE(inode)) 624 goto out_swapfile; 625 /* 626 * O_APPEND implies that we must revalidate the file length. 627 */ 628 if (iocb->ki_flags & IOCB_APPEND || iocb->ki_pos > i_size_read(inode)) { 629 result = nfs_revalidate_file_size(inode, file); 630 if (result) 631 return result; 632 } 633 634 nfs_clear_invalid_mapping(file->f_mapping); 635 636 since = filemap_sample_wb_err(file->f_mapping); 637 nfs_start_io_write(inode); 638 result = generic_write_checks(iocb, from); 639 if (result > 0) { 640 current->backing_dev_info = inode_to_bdi(inode); 641 result = generic_perform_write(iocb, from); 642 current->backing_dev_info = NULL; 643 } 644 nfs_end_io_write(inode); 645 if (result <= 0) 646 goto out; 647 648 written = result; 649 iocb->ki_pos += written; 650 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written); 651 652 if (mntflags & NFS_MOUNT_WRITE_EAGER) { 653 result = filemap_fdatawrite_range(file->f_mapping, 654 iocb->ki_pos - written, 655 iocb->ki_pos - 1); 656 if (result < 0) 657 goto out; 658 } 659 if (mntflags & NFS_MOUNT_WRITE_WAIT) { 660 filemap_fdatawait_range(file->f_mapping, 661 iocb->ki_pos - written, 662 iocb->ki_pos - 1); 663 } 664 result = generic_write_sync(iocb, written); 665 if (result < 0) 666 return result; 667 668 out: 669 /* Return error values */ 670 error = filemap_check_wb_err(file->f_mapping, since); 671 switch (error) { 672 default: 673 break; 674 case -EDQUOT: 675 case -EFBIG: 676 case -ENOSPC: 677 nfs_wb_all(inode); 678 error = file_check_and_advance_wb_err(file); 679 if (error < 0) 680 result = error; 681 } 682 return result; 683 684 out_swapfile: 685 printk(KERN_INFO "NFS: attempt to write to active swap file!\n"); 686 return -ETXTBSY; 687 } 688 EXPORT_SYMBOL_GPL(nfs_file_write); 689 690 static int 691 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) 692 { 693 struct inode *inode = filp->f_mapping->host; 694 int status = 0; 695 unsigned int saved_type = fl->fl_type; 696 697 /* Try local locking first */ 698 posix_test_lock(filp, fl); 699 if (fl->fl_type != F_UNLCK) { 700 /* found a conflict */ 701 goto out; 702 } 703 fl->fl_type = saved_type; 704 705 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) 706 goto out_noconflict; 707 708 if (is_local) 709 goto out_noconflict; 710 711 status = NFS_PROTO(inode)->lock(filp, cmd, fl); 712 out: 713 return status; 714 out_noconflict: 715 fl->fl_type = F_UNLCK; 716 goto out; 717 } 718 719 static int 720 do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) 721 { 722 struct inode *inode = filp->f_mapping->host; 723 struct nfs_lock_context *l_ctx; 724 int status; 725 726 /* 727 * Flush all pending writes before doing anything 728 * with locks.. 729 */ 730 nfs_wb_all(inode); 731 732 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp)); 733 if (!IS_ERR(l_ctx)) { 734 status = nfs_iocounter_wait(l_ctx); 735 nfs_put_lock_context(l_ctx); 736 /* NOTE: special case 737 * If we're signalled while cleaning up locks on process exit, we 738 * still need to complete the unlock. 739 */ 740 if (status < 0 && !(fl->fl_flags & FL_CLOSE)) 741 return status; 742 } 743 744 /* 745 * Use local locking if mounted with "-onolock" or with appropriate 746 * "-olocal_lock=" 747 */ 748 if (!is_local) 749 status = NFS_PROTO(inode)->lock(filp, cmd, fl); 750 else 751 status = locks_lock_file_wait(filp, fl); 752 return status; 753 } 754 755 static int 756 do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) 757 { 758 struct inode *inode = filp->f_mapping->host; 759 int status; 760 761 /* 762 * Flush all pending writes before doing anything 763 * with locks.. 764 */ 765 status = nfs_sync_mapping(filp->f_mapping); 766 if (status != 0) 767 goto out; 768 769 /* 770 * Use local locking if mounted with "-onolock" or with appropriate 771 * "-olocal_lock=" 772 */ 773 if (!is_local) 774 status = NFS_PROTO(inode)->lock(filp, cmd, fl); 775 else 776 status = locks_lock_file_wait(filp, fl); 777 if (status < 0) 778 goto out; 779 780 /* 781 * Invalidate cache to prevent missing any changes. If 782 * the file is mapped, clear the page cache as well so 783 * those mappings will be loaded. 784 * 785 * This makes locking act as a cache coherency point. 786 */ 787 nfs_sync_mapping(filp->f_mapping); 788 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) { 789 nfs_zap_caches(inode); 790 if (mapping_mapped(filp->f_mapping)) 791 nfs_revalidate_mapping(inode, filp->f_mapping); 792 } 793 out: 794 return status; 795 } 796 797 /* 798 * Lock a (portion of) a file 799 */ 800 int nfs_lock(struct file *filp, int cmd, struct file_lock *fl) 801 { 802 struct inode *inode = filp->f_mapping->host; 803 int ret = -ENOLCK; 804 int is_local = 0; 805 806 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n", 807 filp, fl->fl_type, fl->fl_flags, 808 (long long)fl->fl_start, (long long)fl->fl_end); 809 810 nfs_inc_stats(inode, NFSIOS_VFSLOCK); 811 812 if (fl->fl_flags & FL_RECLAIM) 813 return -ENOGRACE; 814 815 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL) 816 is_local = 1; 817 818 if (NFS_PROTO(inode)->lock_check_bounds != NULL) { 819 ret = NFS_PROTO(inode)->lock_check_bounds(fl); 820 if (ret < 0) 821 goto out_err; 822 } 823 824 if (IS_GETLK(cmd)) 825 ret = do_getlk(filp, cmd, fl, is_local); 826 else if (fl->fl_type == F_UNLCK) 827 ret = do_unlk(filp, cmd, fl, is_local); 828 else 829 ret = do_setlk(filp, cmd, fl, is_local); 830 out_err: 831 return ret; 832 } 833 EXPORT_SYMBOL_GPL(nfs_lock); 834 835 /* 836 * Lock a (portion of) a file 837 */ 838 int nfs_flock(struct file *filp, int cmd, struct file_lock *fl) 839 { 840 struct inode *inode = filp->f_mapping->host; 841 int is_local = 0; 842 843 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n", 844 filp, fl->fl_type, fl->fl_flags); 845 846 if (!(fl->fl_flags & FL_FLOCK)) 847 return -ENOLCK; 848 849 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK) 850 is_local = 1; 851 852 /* We're simulating flock() locks using posix locks on the server */ 853 if (fl->fl_type == F_UNLCK) 854 return do_unlk(filp, cmd, fl, is_local); 855 return do_setlk(filp, cmd, fl, is_local); 856 } 857 EXPORT_SYMBOL_GPL(nfs_flock); 858 859 const struct file_operations nfs_file_operations = { 860 .llseek = nfs_file_llseek, 861 .read_iter = nfs_file_read, 862 .write_iter = nfs_file_write, 863 .mmap = nfs_file_mmap, 864 .open = nfs_file_open, 865 .flush = nfs_file_flush, 866 .release = nfs_file_release, 867 .fsync = nfs_file_fsync, 868 .lock = nfs_lock, 869 .flock = nfs_flock, 870 .splice_read = generic_file_splice_read, 871 .splice_write = iter_file_splice_write, 872 .check_flags = nfs_check_flags, 873 .setlease = simple_nosetlease, 874 }; 875 EXPORT_SYMBOL_GPL(nfs_file_operations); 876