1 /* 2 * linux/fs/nfs/file.c 3 * 4 * Copyright (C) 1992 Rick Sladkey 5 * 6 * Changes Copyright (C) 1994 by Florian La Roche 7 * - Do not copy data too often around in the kernel. 8 * - In nfs_file_read the return value of kmalloc wasn't checked. 9 * - Put in a better version of read look-ahead buffering. Original idea 10 * and implementation by Wai S Kok elekokws@ee.nus.sg. 11 * 12 * Expire cache on write to a file by Wai S Kok (Oct 1994). 13 * 14 * Total rewrite of read side for new NFS buffer cache.. Linus. 15 * 16 * nfs regular file handling functions 17 */ 18 19 #include <linux/time.h> 20 #include <linux/kernel.h> 21 #include <linux/errno.h> 22 #include <linux/fcntl.h> 23 #include <linux/stat.h> 24 #include <linux/nfs_fs.h> 25 #include <linux/nfs_mount.h> 26 #include <linux/mm.h> 27 #include <linux/slab.h> 28 #include <linux/pagemap.h> 29 #include <linux/aio.h> 30 31 #include <asm/uaccess.h> 32 #include <asm/system.h> 33 34 #include "delegation.h" 35 #include "internal.h" 36 #include "iostat.h" 37 #include "fscache.h" 38 39 #define NFSDBG_FACILITY NFSDBG_FILE 40 41 static int nfs_file_open(struct inode *, struct file *); 42 static int nfs_file_release(struct inode *, struct file *); 43 static loff_t nfs_file_llseek(struct file *file, loff_t offset, int origin); 44 static int nfs_file_mmap(struct file *, struct vm_area_struct *); 45 static ssize_t nfs_file_splice_read(struct file *filp, loff_t *ppos, 46 struct pipe_inode_info *pipe, 47 size_t count, unsigned int flags); 48 static ssize_t nfs_file_read(struct kiocb *, const struct iovec *iov, 49 unsigned long nr_segs, loff_t pos); 50 static ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe, 51 struct file *filp, loff_t *ppos, 52 size_t count, unsigned int flags); 53 static ssize_t nfs_file_write(struct kiocb *, const struct iovec *iov, 54 unsigned long nr_segs, loff_t pos); 55 static int nfs_file_flush(struct file *, fl_owner_t id); 56 static int nfs_file_fsync(struct file *, struct dentry *dentry, int datasync); 57 static int nfs_check_flags(int flags); 58 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl); 59 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl); 60 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl); 61 62 static const struct vm_operations_struct nfs_file_vm_ops; 63 64 const struct file_operations nfs_file_operations = { 65 .llseek = nfs_file_llseek, 66 .read = do_sync_read, 67 .write = do_sync_write, 68 .aio_read = nfs_file_read, 69 .aio_write = nfs_file_write, 70 .mmap = nfs_file_mmap, 71 .open = nfs_file_open, 72 .flush = nfs_file_flush, 73 .release = nfs_file_release, 74 .fsync = nfs_file_fsync, 75 .lock = nfs_lock, 76 .flock = nfs_flock, 77 .splice_read = nfs_file_splice_read, 78 .splice_write = nfs_file_splice_write, 79 .check_flags = nfs_check_flags, 80 .setlease = nfs_setlease, 81 }; 82 83 const struct inode_operations nfs_file_inode_operations = { 84 .permission = nfs_permission, 85 .getattr = nfs_getattr, 86 .setattr = nfs_setattr, 87 }; 88 89 #ifdef CONFIG_NFS_V3 90 const struct inode_operations nfs3_file_inode_operations = { 91 .permission = nfs_permission, 92 .getattr = nfs_getattr, 93 .setattr = nfs_setattr, 94 .listxattr = nfs3_listxattr, 95 .getxattr = nfs3_getxattr, 96 .setxattr = nfs3_setxattr, 97 .removexattr = nfs3_removexattr, 98 }; 99 #endif /* CONFIG_NFS_v3 */ 100 101 /* Hack for future NFS swap support */ 102 #ifndef IS_SWAPFILE 103 # define IS_SWAPFILE(inode) (0) 104 #endif 105 106 static int nfs_check_flags(int flags) 107 { 108 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT)) 109 return -EINVAL; 110 111 return 0; 112 } 113 114 /* 115 * Open file 116 */ 117 static int 118 nfs_file_open(struct inode *inode, struct file *filp) 119 { 120 int res; 121 122 dprintk("NFS: open file(%s/%s)\n", 123 filp->f_path.dentry->d_parent->d_name.name, 124 filp->f_path.dentry->d_name.name); 125 126 res = nfs_check_flags(filp->f_flags); 127 if (res) 128 return res; 129 130 nfs_inc_stats(inode, NFSIOS_VFSOPEN); 131 res = nfs_open(inode, filp); 132 return res; 133 } 134 135 static int 136 nfs_file_release(struct inode *inode, struct file *filp) 137 { 138 struct dentry *dentry = filp->f_path.dentry; 139 140 dprintk("NFS: release(%s/%s)\n", 141 dentry->d_parent->d_name.name, 142 dentry->d_name.name); 143 144 nfs_inc_stats(inode, NFSIOS_VFSRELEASE); 145 return nfs_release(inode, filp); 146 } 147 148 /** 149 * nfs_revalidate_size - Revalidate the file size 150 * @inode - pointer to inode struct 151 * @file - pointer to struct file 152 * 153 * Revalidates the file length. This is basically a wrapper around 154 * nfs_revalidate_inode() that takes into account the fact that we may 155 * have cached writes (in which case we don't care about the server's 156 * idea of what the file length is), or O_DIRECT (in which case we 157 * shouldn't trust the cache). 158 */ 159 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp) 160 { 161 struct nfs_server *server = NFS_SERVER(inode); 162 struct nfs_inode *nfsi = NFS_I(inode); 163 164 if (server->flags & NFS_MOUNT_NOAC) 165 goto force_reval; 166 if (filp->f_flags & O_DIRECT) 167 goto force_reval; 168 if (nfsi->npages != 0) 169 return 0; 170 if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode)) 171 return 0; 172 force_reval: 173 return __nfs_revalidate_inode(server, inode); 174 } 175 176 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin) 177 { 178 loff_t loff; 179 180 dprintk("NFS: llseek file(%s/%s, %lld, %d)\n", 181 filp->f_path.dentry->d_parent->d_name.name, 182 filp->f_path.dentry->d_name.name, 183 offset, origin); 184 185 /* origin == SEEK_END => we must revalidate the cached file length */ 186 if (origin == SEEK_END) { 187 struct inode *inode = filp->f_mapping->host; 188 189 int retval = nfs_revalidate_file_size(inode, filp); 190 if (retval < 0) 191 return (loff_t)retval; 192 193 spin_lock(&inode->i_lock); 194 loff = generic_file_llseek_unlocked(filp, offset, origin); 195 spin_unlock(&inode->i_lock); 196 } else 197 loff = generic_file_llseek_unlocked(filp, offset, origin); 198 return loff; 199 } 200 201 /* 202 * Helper for nfs_file_flush() and nfs_file_fsync() 203 * 204 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to 205 * disk, but it retrieves and clears ctx->error after synching, despite 206 * the two being set at the same time in nfs_context_set_write_error(). 207 * This is because the former is used to notify the _next_ call to 208 * nfs_file_write() that a write error occured, and hence cause it to 209 * fall back to doing a synchronous write. 210 */ 211 static int nfs_do_fsync(struct nfs_open_context *ctx, struct inode *inode) 212 { 213 int have_error, status; 214 int ret = 0; 215 216 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 217 status = nfs_wb_all(inode); 218 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 219 if (have_error) 220 ret = xchg(&ctx->error, 0); 221 if (!ret) 222 ret = status; 223 return ret; 224 } 225 226 /* 227 * Flush all dirty pages, and check for write errors. 228 */ 229 static int 230 nfs_file_flush(struct file *file, fl_owner_t id) 231 { 232 struct nfs_open_context *ctx = nfs_file_open_context(file); 233 struct dentry *dentry = file->f_path.dentry; 234 struct inode *inode = dentry->d_inode; 235 236 dprintk("NFS: flush(%s/%s)\n", 237 dentry->d_parent->d_name.name, 238 dentry->d_name.name); 239 240 if ((file->f_mode & FMODE_WRITE) == 0) 241 return 0; 242 nfs_inc_stats(inode, NFSIOS_VFSFLUSH); 243 244 /* Flush writes to the server and return any errors */ 245 return nfs_do_fsync(ctx, inode); 246 } 247 248 static ssize_t 249 nfs_file_read(struct kiocb *iocb, const struct iovec *iov, 250 unsigned long nr_segs, loff_t pos) 251 { 252 struct dentry * dentry = iocb->ki_filp->f_path.dentry; 253 struct inode * inode = dentry->d_inode; 254 ssize_t result; 255 size_t count = iov_length(iov, nr_segs); 256 257 if (iocb->ki_filp->f_flags & O_DIRECT) 258 return nfs_file_direct_read(iocb, iov, nr_segs, pos); 259 260 dprintk("NFS: read(%s/%s, %lu@%lu)\n", 261 dentry->d_parent->d_name.name, dentry->d_name.name, 262 (unsigned long) count, (unsigned long) pos); 263 264 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping); 265 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count); 266 if (!result) 267 result = generic_file_aio_read(iocb, iov, nr_segs, pos); 268 return result; 269 } 270 271 static ssize_t 272 nfs_file_splice_read(struct file *filp, loff_t *ppos, 273 struct pipe_inode_info *pipe, size_t count, 274 unsigned int flags) 275 { 276 struct dentry *dentry = filp->f_path.dentry; 277 struct inode *inode = dentry->d_inode; 278 ssize_t res; 279 280 dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n", 281 dentry->d_parent->d_name.name, dentry->d_name.name, 282 (unsigned long) count, (unsigned long long) *ppos); 283 284 res = nfs_revalidate_mapping(inode, filp->f_mapping); 285 if (!res) 286 res = generic_file_splice_read(filp, ppos, pipe, count, flags); 287 return res; 288 } 289 290 static int 291 nfs_file_mmap(struct file * file, struct vm_area_struct * vma) 292 { 293 struct dentry *dentry = file->f_path.dentry; 294 struct inode *inode = dentry->d_inode; 295 int status; 296 297 dprintk("NFS: mmap(%s/%s)\n", 298 dentry->d_parent->d_name.name, dentry->d_name.name); 299 300 /* Note: generic_file_mmap() returns ENOSYS on nommu systems 301 * so we call that before revalidating the mapping 302 */ 303 status = generic_file_mmap(file, vma); 304 if (!status) { 305 vma->vm_ops = &nfs_file_vm_ops; 306 status = nfs_revalidate_mapping(inode, file->f_mapping); 307 } 308 return status; 309 } 310 311 /* 312 * Flush any dirty pages for this process, and check for write errors. 313 * The return status from this call provides a reliable indication of 314 * whether any write errors occurred for this process. 315 */ 316 static int 317 nfs_file_fsync(struct file *file, struct dentry *dentry, int datasync) 318 { 319 struct nfs_open_context *ctx = nfs_file_open_context(file); 320 struct inode *inode = dentry->d_inode; 321 322 dprintk("NFS: fsync file(%s/%s) datasync %d\n", 323 dentry->d_parent->d_name.name, dentry->d_name.name, 324 datasync); 325 326 nfs_inc_stats(inode, NFSIOS_VFSFSYNC); 327 return nfs_do_fsync(ctx, inode); 328 } 329 330 /* 331 * Decide whether a read/modify/write cycle may be more efficient 332 * then a modify/write/read cycle when writing to a page in the 333 * page cache. 334 * 335 * The modify/write/read cycle may occur if a page is read before 336 * being completely filled by the writer. In this situation, the 337 * page must be completely written to stable storage on the server 338 * before it can be refilled by reading in the page from the server. 339 * This can lead to expensive, small, FILE_SYNC mode writes being 340 * done. 341 * 342 * It may be more efficient to read the page first if the file is 343 * open for reading in addition to writing, the page is not marked 344 * as Uptodate, it is not dirty or waiting to be committed, 345 * indicating that it was previously allocated and then modified, 346 * that there were valid bytes of data in that range of the file, 347 * and that the new data won't completely replace the old data in 348 * that range of the file. 349 */ 350 static int nfs_want_read_modify_write(struct file *file, struct page *page, 351 loff_t pos, unsigned len) 352 { 353 unsigned int pglen = nfs_page_length(page); 354 unsigned int offset = pos & (PAGE_CACHE_SIZE - 1); 355 unsigned int end = offset + len; 356 357 if ((file->f_mode & FMODE_READ) && /* open for read? */ 358 !PageUptodate(page) && /* Uptodate? */ 359 !PagePrivate(page) && /* i/o request already? */ 360 pglen && /* valid bytes of file? */ 361 (end < pglen || offset)) /* replace all valid bytes? */ 362 return 1; 363 return 0; 364 } 365 366 /* 367 * This does the "real" work of the write. We must allocate and lock the 368 * page to be sent back to the generic routine, which then copies the 369 * data from user space. 370 * 371 * If the writer ends up delaying the write, the writer needs to 372 * increment the page use counts until he is done with the page. 373 */ 374 static int nfs_write_begin(struct file *file, struct address_space *mapping, 375 loff_t pos, unsigned len, unsigned flags, 376 struct page **pagep, void **fsdata) 377 { 378 int ret; 379 pgoff_t index = pos >> PAGE_CACHE_SHIFT; 380 struct page *page; 381 int once_thru = 0; 382 383 dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n", 384 file->f_path.dentry->d_parent->d_name.name, 385 file->f_path.dentry->d_name.name, 386 mapping->host->i_ino, len, (long long) pos); 387 388 start: 389 /* 390 * Prevent starvation issues if someone is doing a consistency 391 * sync-to-disk 392 */ 393 ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING, 394 nfs_wait_bit_killable, TASK_KILLABLE); 395 if (ret) 396 return ret; 397 398 page = grab_cache_page_write_begin(mapping, index, flags); 399 if (!page) 400 return -ENOMEM; 401 *pagep = page; 402 403 ret = nfs_flush_incompatible(file, page); 404 if (ret) { 405 unlock_page(page); 406 page_cache_release(page); 407 } else if (!once_thru && 408 nfs_want_read_modify_write(file, page, pos, len)) { 409 once_thru = 1; 410 ret = nfs_readpage(file, page); 411 page_cache_release(page); 412 if (!ret) 413 goto start; 414 } 415 return ret; 416 } 417 418 static int nfs_write_end(struct file *file, struct address_space *mapping, 419 loff_t pos, unsigned len, unsigned copied, 420 struct page *page, void *fsdata) 421 { 422 unsigned offset = pos & (PAGE_CACHE_SIZE - 1); 423 int status; 424 425 dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n", 426 file->f_path.dentry->d_parent->d_name.name, 427 file->f_path.dentry->d_name.name, 428 mapping->host->i_ino, len, (long long) pos); 429 430 /* 431 * Zero any uninitialised parts of the page, and then mark the page 432 * as up to date if it turns out that we're extending the file. 433 */ 434 if (!PageUptodate(page)) { 435 unsigned pglen = nfs_page_length(page); 436 unsigned end = offset + len; 437 438 if (pglen == 0) { 439 zero_user_segments(page, 0, offset, 440 end, PAGE_CACHE_SIZE); 441 SetPageUptodate(page); 442 } else if (end >= pglen) { 443 zero_user_segment(page, end, PAGE_CACHE_SIZE); 444 if (offset == 0) 445 SetPageUptodate(page); 446 } else 447 zero_user_segment(page, pglen, PAGE_CACHE_SIZE); 448 } 449 450 status = nfs_updatepage(file, page, offset, copied); 451 452 unlock_page(page); 453 page_cache_release(page); 454 455 if (status < 0) 456 return status; 457 return copied; 458 } 459 460 /* 461 * Partially or wholly invalidate a page 462 * - Release the private state associated with a page if undergoing complete 463 * page invalidation 464 * - Called if either PG_private or PG_fscache is set on the page 465 * - Caller holds page lock 466 */ 467 static void nfs_invalidate_page(struct page *page, unsigned long offset) 468 { 469 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset); 470 471 if (offset != 0) 472 return; 473 /* Cancel any unstarted writes on this page */ 474 nfs_wb_page_cancel(page->mapping->host, page); 475 476 nfs_fscache_invalidate_page(page, page->mapping->host); 477 } 478 479 /* 480 * Attempt to release the private state associated with a page 481 * - Called if either PG_private or PG_fscache is set on the page 482 * - Caller holds page lock 483 * - Return true (may release page) or false (may not) 484 */ 485 static int nfs_release_page(struct page *page, gfp_t gfp) 486 { 487 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page); 488 489 /* If PagePrivate() is set, then the page is not freeable */ 490 if (PagePrivate(page)) 491 return 0; 492 return nfs_fscache_release_page(page, gfp); 493 } 494 495 /* 496 * Attempt to clear the private state associated with a page when an error 497 * occurs that requires the cached contents of an inode to be written back or 498 * destroyed 499 * - Called if either PG_private or fscache is set on the page 500 * - Caller holds page lock 501 * - Return 0 if successful, -error otherwise 502 */ 503 static int nfs_launder_page(struct page *page) 504 { 505 struct inode *inode = page->mapping->host; 506 struct nfs_inode *nfsi = NFS_I(inode); 507 508 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n", 509 inode->i_ino, (long long)page_offset(page)); 510 511 nfs_fscache_wait_on_page_write(nfsi, page); 512 return nfs_wb_page(inode, page); 513 } 514 515 const struct address_space_operations nfs_file_aops = { 516 .readpage = nfs_readpage, 517 .readpages = nfs_readpages, 518 .set_page_dirty = __set_page_dirty_nobuffers, 519 .writepage = nfs_writepage, 520 .writepages = nfs_writepages, 521 .write_begin = nfs_write_begin, 522 .write_end = nfs_write_end, 523 .invalidatepage = nfs_invalidate_page, 524 .releasepage = nfs_release_page, 525 .direct_IO = nfs_direct_IO, 526 .migratepage = nfs_migrate_page, 527 .launder_page = nfs_launder_page, 528 .error_remove_page = generic_error_remove_page, 529 }; 530 531 /* 532 * Notification that a PTE pointing to an NFS page is about to be made 533 * writable, implying that someone is about to modify the page through a 534 * shared-writable mapping 535 */ 536 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) 537 { 538 struct page *page = vmf->page; 539 struct file *filp = vma->vm_file; 540 struct dentry *dentry = filp->f_path.dentry; 541 unsigned pagelen; 542 int ret = -EINVAL; 543 struct address_space *mapping; 544 545 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n", 546 dentry->d_parent->d_name.name, dentry->d_name.name, 547 filp->f_mapping->host->i_ino, 548 (long long)page_offset(page)); 549 550 /* make sure the cache has finished storing the page */ 551 nfs_fscache_wait_on_page_write(NFS_I(dentry->d_inode), page); 552 553 lock_page(page); 554 mapping = page->mapping; 555 if (mapping != dentry->d_inode->i_mapping) 556 goto out_unlock; 557 558 ret = 0; 559 pagelen = nfs_page_length(page); 560 if (pagelen == 0) 561 goto out_unlock; 562 563 ret = nfs_flush_incompatible(filp, page); 564 if (ret != 0) 565 goto out_unlock; 566 567 ret = nfs_updatepage(filp, page, 0, pagelen); 568 out_unlock: 569 if (!ret) 570 return VM_FAULT_LOCKED; 571 unlock_page(page); 572 return VM_FAULT_SIGBUS; 573 } 574 575 static const struct vm_operations_struct nfs_file_vm_ops = { 576 .fault = filemap_fault, 577 .page_mkwrite = nfs_vm_page_mkwrite, 578 }; 579 580 static int nfs_need_sync_write(struct file *filp, struct inode *inode) 581 { 582 struct nfs_open_context *ctx; 583 584 if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC)) 585 return 1; 586 ctx = nfs_file_open_context(filp); 587 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags)) 588 return 1; 589 return 0; 590 } 591 592 static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov, 593 unsigned long nr_segs, loff_t pos) 594 { 595 struct dentry * dentry = iocb->ki_filp->f_path.dentry; 596 struct inode * inode = dentry->d_inode; 597 ssize_t result; 598 size_t count = iov_length(iov, nr_segs); 599 600 if (iocb->ki_filp->f_flags & O_DIRECT) 601 return nfs_file_direct_write(iocb, iov, nr_segs, pos); 602 603 dprintk("NFS: write(%s/%s, %lu@%Ld)\n", 604 dentry->d_parent->d_name.name, dentry->d_name.name, 605 (unsigned long) count, (long long) pos); 606 607 result = -EBUSY; 608 if (IS_SWAPFILE(inode)) 609 goto out_swapfile; 610 /* 611 * O_APPEND implies that we must revalidate the file length. 612 */ 613 if (iocb->ki_filp->f_flags & O_APPEND) { 614 result = nfs_revalidate_file_size(inode, iocb->ki_filp); 615 if (result) 616 goto out; 617 } 618 619 result = count; 620 if (!count) 621 goto out; 622 623 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count); 624 result = generic_file_aio_write(iocb, iov, nr_segs, pos); 625 /* Return error values for O_DSYNC and IS_SYNC() */ 626 if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) { 627 int err = nfs_do_fsync(nfs_file_open_context(iocb->ki_filp), inode); 628 if (err < 0) 629 result = err; 630 } 631 out: 632 return result; 633 634 out_swapfile: 635 printk(KERN_INFO "NFS: attempt to write to active swap file!\n"); 636 goto out; 637 } 638 639 static ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe, 640 struct file *filp, loff_t *ppos, 641 size_t count, unsigned int flags) 642 { 643 struct dentry *dentry = filp->f_path.dentry; 644 struct inode *inode = dentry->d_inode; 645 ssize_t ret; 646 647 dprintk("NFS splice_write(%s/%s, %lu@%llu)\n", 648 dentry->d_parent->d_name.name, dentry->d_name.name, 649 (unsigned long) count, (unsigned long long) *ppos); 650 651 /* 652 * The combination of splice and an O_APPEND destination is disallowed. 653 */ 654 655 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count); 656 657 ret = generic_file_splice_write(pipe, filp, ppos, count, flags); 658 if (ret >= 0 && nfs_need_sync_write(filp, inode)) { 659 int err = nfs_do_fsync(nfs_file_open_context(filp), inode); 660 if (err < 0) 661 ret = err; 662 } 663 return ret; 664 } 665 666 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl) 667 { 668 struct inode *inode = filp->f_mapping->host; 669 int status = 0; 670 671 /* Try local locking first */ 672 posix_test_lock(filp, fl); 673 if (fl->fl_type != F_UNLCK) { 674 /* found a conflict */ 675 goto out; 676 } 677 678 if (nfs_have_delegation(inode, FMODE_READ)) 679 goto out_noconflict; 680 681 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM) 682 goto out_noconflict; 683 684 status = NFS_PROTO(inode)->lock(filp, cmd, fl); 685 out: 686 return status; 687 out_noconflict: 688 fl->fl_type = F_UNLCK; 689 goto out; 690 } 691 692 static int do_vfs_lock(struct file *file, struct file_lock *fl) 693 { 694 int res = 0; 695 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) { 696 case FL_POSIX: 697 res = posix_lock_file_wait(file, fl); 698 break; 699 case FL_FLOCK: 700 res = flock_lock_file_wait(file, fl); 701 break; 702 default: 703 BUG(); 704 } 705 if (res < 0) 706 dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager" 707 " - error %d!\n", 708 __func__, res); 709 return res; 710 } 711 712 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl) 713 { 714 struct inode *inode = filp->f_mapping->host; 715 int status; 716 717 /* 718 * Flush all pending writes before doing anything 719 * with locks.. 720 */ 721 nfs_sync_mapping(filp->f_mapping); 722 723 /* NOTE: special case 724 * If we're signalled while cleaning up locks on process exit, we 725 * still need to complete the unlock. 726 */ 727 /* Use local locking if mounted with "-onolock" */ 728 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) 729 status = NFS_PROTO(inode)->lock(filp, cmd, fl); 730 else 731 status = do_vfs_lock(filp, fl); 732 return status; 733 } 734 735 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl) 736 { 737 struct inode *inode = filp->f_mapping->host; 738 int status; 739 740 /* 741 * Flush all pending writes before doing anything 742 * with locks.. 743 */ 744 status = nfs_sync_mapping(filp->f_mapping); 745 if (status != 0) 746 goto out; 747 748 /* Use local locking if mounted with "-onolock" */ 749 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) 750 status = NFS_PROTO(inode)->lock(filp, cmd, fl); 751 else 752 status = do_vfs_lock(filp, fl); 753 if (status < 0) 754 goto out; 755 /* 756 * Make sure we clear the cache whenever we try to get the lock. 757 * This makes locking act as a cache coherency point. 758 */ 759 nfs_sync_mapping(filp->f_mapping); 760 if (!nfs_have_delegation(inode, FMODE_READ)) 761 nfs_zap_caches(inode); 762 out: 763 return status; 764 } 765 766 /* 767 * Lock a (portion of) a file 768 */ 769 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl) 770 { 771 struct inode *inode = filp->f_mapping->host; 772 int ret = -ENOLCK; 773 774 dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n", 775 filp->f_path.dentry->d_parent->d_name.name, 776 filp->f_path.dentry->d_name.name, 777 fl->fl_type, fl->fl_flags, 778 (long long)fl->fl_start, (long long)fl->fl_end); 779 780 nfs_inc_stats(inode, NFSIOS_VFSLOCK); 781 782 /* No mandatory locks over NFS */ 783 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK) 784 goto out_err; 785 786 if (NFS_PROTO(inode)->lock_check_bounds != NULL) { 787 ret = NFS_PROTO(inode)->lock_check_bounds(fl); 788 if (ret < 0) 789 goto out_err; 790 } 791 792 if (IS_GETLK(cmd)) 793 ret = do_getlk(filp, cmd, fl); 794 else if (fl->fl_type == F_UNLCK) 795 ret = do_unlk(filp, cmd, fl); 796 else 797 ret = do_setlk(filp, cmd, fl); 798 out_err: 799 return ret; 800 } 801 802 /* 803 * Lock a (portion of) a file 804 */ 805 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl) 806 { 807 dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n", 808 filp->f_path.dentry->d_parent->d_name.name, 809 filp->f_path.dentry->d_name.name, 810 fl->fl_type, fl->fl_flags); 811 812 if (!(fl->fl_flags & FL_FLOCK)) 813 return -ENOLCK; 814 815 /* We're simulating flock() locks using posix locks on the server */ 816 fl->fl_owner = (fl_owner_t)filp; 817 fl->fl_start = 0; 818 fl->fl_end = OFFSET_MAX; 819 820 if (fl->fl_type == F_UNLCK) 821 return do_unlk(filp, cmd, fl); 822 return do_setlk(filp, cmd, fl); 823 } 824 825 /* 826 * There is no protocol support for leases, so we have no way to implement 827 * them correctly in the face of opens by other clients. 828 */ 829 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl) 830 { 831 dprintk("NFS: setlease(%s/%s, arg=%ld)\n", 832 file->f_path.dentry->d_parent->d_name.name, 833 file->f_path.dentry->d_name.name, arg); 834 835 return -EINVAL; 836 } 837