1/* 2 * linux/fs/nfs/dir.c 3 * 4 * Copyright (C) 1992 Rick Sladkey 5 * 6 * nfs directory handling functions 7 * 8 * 10 Apr 1996 Added silly rename for unlink --okir 9 * 28 Sep 1996 Improved directory cache --okir 10 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de 11 * Re-implemented silly rename for unlink, newly implemented 12 * silly rename for nfs_rename() following the suggestions 13 * of Olaf Kirch (okir) found in this file. 14 * Following Linus comments on my original hack, this version 15 * depends only on the dcache stuff and doesn't touch the inode 16 * layer (iput() and friends). 17 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM 18 */ 19 20#include <linux/time.h> 21#include <linux/errno.h> 22#include <linux/stat.h> 23#include <linux/fcntl.h> 24#include <linux/string.h> 25#include <linux/kernel.h> 26#include <linux/slab.h> 27#include <linux/mm.h> 28#include <linux/sunrpc/clnt.h> 29#include <linux/nfs_fs.h> 30#include <linux/nfs_mount.h> 31#include <linux/pagemap.h> 32#include <linux/smp_lock.h> 33#include <linux/pagevec.h> 34#include <linux/namei.h> 35#include <linux/mount.h> 36#include <linux/sched.h> 37 38#include "nfs4_fs.h" 39#include "delegation.h" 40#include "iostat.h" 41 42/* #define NFS_DEBUG_VERBOSE 1 */ 43 44static int nfs_opendir(struct inode *, struct file *); 45static int nfs_readdir(struct file *, void *, filldir_t); 46static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *); 47static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *); 48static int nfs_mkdir(struct inode *, struct dentry *, int); 49static int nfs_rmdir(struct inode *, struct dentry *); 50static int nfs_unlink(struct inode *, struct dentry *); 51static int nfs_symlink(struct inode *, struct dentry *, const char *); 52static int nfs_link(struct dentry *, struct inode *, struct dentry *); 53static int nfs_mknod(struct inode *, struct dentry *, int, dev_t); 54static int nfs_rename(struct inode *, struct dentry *, 55 struct inode *, struct dentry *); 56static int nfs_fsync_dir(struct file *, struct dentry *, int); 57static loff_t nfs_llseek_dir(struct file *, loff_t, int); 58 59const struct file_operations nfs_dir_operations = { 60 .llseek = nfs_llseek_dir, 61 .read = generic_read_dir, 62 .readdir = nfs_readdir, 63 .open = nfs_opendir, 64 .release = nfs_release, 65 .fsync = nfs_fsync_dir, 66}; 67 68const struct inode_operations nfs_dir_inode_operations = { 69 .create = nfs_create, 70 .lookup = nfs_lookup, 71 .link = nfs_link, 72 .unlink = nfs_unlink, 73 .symlink = nfs_symlink, 74 .mkdir = nfs_mkdir, 75 .rmdir = nfs_rmdir, 76 .mknod = nfs_mknod, 77 .rename = nfs_rename, 78 .permission = nfs_permission, 79 .getattr = nfs_getattr, 80 .setattr = nfs_setattr, 81}; 82 83#ifdef CONFIG_NFS_V3 84const struct inode_operations nfs3_dir_inode_operations = { 85 .create = nfs_create, 86 .lookup = nfs_lookup, 87 .link = nfs_link, 88 .unlink = nfs_unlink, 89 .symlink = nfs_symlink, 90 .mkdir = nfs_mkdir, 91 .rmdir = nfs_rmdir, 92 .mknod = nfs_mknod, 93 .rename = nfs_rename, 94 .permission = nfs_permission, 95 .getattr = nfs_getattr, 96 .setattr = nfs_setattr, 97 .listxattr = nfs3_listxattr, 98 .getxattr = nfs3_getxattr, 99 .setxattr = nfs3_setxattr, 100 .removexattr = nfs3_removexattr, 101}; 102#endif /* CONFIG_NFS_V3 */ 103 104#ifdef CONFIG_NFS_V4 105 106static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *); 107const struct inode_operations nfs4_dir_inode_operations = { 108 .create = nfs_create, 109 .lookup = nfs_atomic_lookup, 110 .link = nfs_link, 111 .unlink = nfs_unlink, 112 .symlink = nfs_symlink, 113 .mkdir = nfs_mkdir, 114 .rmdir = nfs_rmdir, 115 .mknod = nfs_mknod, 116 .rename = nfs_rename, 117 .permission = nfs_permission, 118 .getattr = nfs_getattr, 119 .setattr = nfs_setattr, 120 .getxattr = nfs4_getxattr, 121 .setxattr = nfs4_setxattr, 122 .listxattr = nfs4_listxattr, 123}; 124 125#endif /* CONFIG_NFS_V4 */ 126 127/* 128 * Open file 129 */ 130static int 131nfs_opendir(struct inode *inode, struct file *filp) 132{ 133 int res; 134 135 dfprintk(VFS, "NFS: opendir(%s/%ld)\n", 136 inode->i_sb->s_id, inode->i_ino); 137 138 lock_kernel(); 139 /* Call generic open code in order to cache credentials */ 140 res = nfs_open(inode, filp); 141 unlock_kernel(); 142 return res; 143} 144 145typedef __be32 * (*decode_dirent_t)(__be32 *, struct nfs_entry *, int); 146typedef struct { 147 struct file *file; 148 struct page *page; 149 unsigned long page_index; 150 __be32 *ptr; 151 u64 *dir_cookie; 152 loff_t current_index; 153 struct nfs_entry *entry; 154 decode_dirent_t decode; 155 int plus; 156 int error; 157 unsigned long timestamp; 158 int timestamp_valid; 159} nfs_readdir_descriptor_t; 160 161/* Now we cache directories properly, by stuffing the dirent 162 * data directly in the page cache. 163 * 164 * Inode invalidation due to refresh etc. takes care of 165 * _everything_, no sloppy entry flushing logic, no extraneous 166 * copying, network direct to page cache, the way it was meant 167 * to be. 168 * 169 * NOTE: Dirent information verification is done always by the 170 * page-in of the RPC reply, nowhere else, this simplies 171 * things substantially. 172 */ 173static 174int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page) 175{ 176 struct file *file = desc->file; 177 struct inode *inode = file->f_path.dentry->d_inode; 178 struct rpc_cred *cred = nfs_file_cred(file); 179 unsigned long timestamp; 180 int error; 181 182 dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n", 183 __FUNCTION__, (long long)desc->entry->cookie, 184 page->index); 185 186 again: 187 timestamp = jiffies; 188 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, desc->entry->cookie, page, 189 NFS_SERVER(inode)->dtsize, desc->plus); 190 if (error < 0) { 191 /* We requested READDIRPLUS, but the server doesn't grok it */ 192 if (error == -ENOTSUPP && desc->plus) { 193 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; 194 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode)); 195 desc->plus = 0; 196 goto again; 197 } 198 goto error; 199 } 200 desc->timestamp = timestamp; 201 desc->timestamp_valid = 1; 202 SetPageUptodate(page); 203 spin_lock(&inode->i_lock); 204 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME; 205 spin_unlock(&inode->i_lock); 206 /* Ensure consistent page alignment of the data. 207 * Note: assumes we have exclusive access to this mapping either 208 * through inode->i_mutex or some other mechanism. 209 */ 210 if (page->index == 0 && invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1) < 0) { 211 /* Should never happen */ 212 nfs_zap_mapping(inode, inode->i_mapping); 213 } 214 unlock_page(page); 215 return 0; 216 error: 217 SetPageError(page); 218 unlock_page(page); 219 nfs_zap_caches(inode); 220 desc->error = error; 221 return -EIO; 222} 223 224static inline 225int dir_decode(nfs_readdir_descriptor_t *desc) 226{ 227 __be32 *p = desc->ptr; 228 p = desc->decode(p, desc->entry, desc->plus); 229 if (IS_ERR(p)) 230 return PTR_ERR(p); 231 desc->ptr = p; 232 if (desc->timestamp_valid) 233 desc->entry->fattr->time_start = desc->timestamp; 234 else 235 desc->entry->fattr->valid &= ~NFS_ATTR_FATTR; 236 return 0; 237} 238 239static inline 240void dir_page_release(nfs_readdir_descriptor_t *desc) 241{ 242 kunmap(desc->page); 243 page_cache_release(desc->page); 244 desc->page = NULL; 245 desc->ptr = NULL; 246} 247 248/* 249 * Given a pointer to a buffer that has already been filled by a call 250 * to readdir, find the next entry with cookie '*desc->dir_cookie'. 251 * 252 * If the end of the buffer has been reached, return -EAGAIN, if not, 253 * return the offset within the buffer of the next entry to be 254 * read. 255 */ 256static inline 257int find_dirent(nfs_readdir_descriptor_t *desc) 258{ 259 struct nfs_entry *entry = desc->entry; 260 int loop_count = 0, 261 status; 262 263 while((status = dir_decode(desc)) == 0) { 264 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n", 265 __FUNCTION__, (unsigned long long)entry->cookie); 266 if (entry->prev_cookie == *desc->dir_cookie) 267 break; 268 if (loop_count++ > 200) { 269 loop_count = 0; 270 schedule(); 271 } 272 } 273 return status; 274} 275 276/* 277 * Given a pointer to a buffer that has already been filled by a call 278 * to readdir, find the entry at offset 'desc->file->f_pos'. 279 * 280 * If the end of the buffer has been reached, return -EAGAIN, if not, 281 * return the offset within the buffer of the next entry to be 282 * read. 283 */ 284static inline 285int find_dirent_index(nfs_readdir_descriptor_t *desc) 286{ 287 struct nfs_entry *entry = desc->entry; 288 int loop_count = 0, 289 status; 290 291 for(;;) { 292 status = dir_decode(desc); 293 if (status) 294 break; 295 296 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n", 297 (unsigned long long)entry->cookie, desc->current_index); 298 299 if (desc->file->f_pos == desc->current_index) { 300 *desc->dir_cookie = entry->cookie; 301 break; 302 } 303 desc->current_index++; 304 if (loop_count++ > 200) { 305 loop_count = 0; 306 schedule(); 307 } 308 } 309 return status; 310} 311 312/* 313 * Find the given page, and call find_dirent() or find_dirent_index in 314 * order to try to return the next entry. 315 */ 316static inline 317int find_dirent_page(nfs_readdir_descriptor_t *desc) 318{ 319 struct inode *inode = desc->file->f_path.dentry->d_inode; 320 struct page *page; 321 int status; 322 323 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n", 324 __FUNCTION__, desc->page_index, 325 (long long) *desc->dir_cookie); 326 327 /* If we find the page in the page_cache, we cannot be sure 328 * how fresh the data is, so we will ignore readdir_plus attributes. 329 */ 330 desc->timestamp_valid = 0; 331 page = read_cache_page(inode->i_mapping, desc->page_index, 332 (filler_t *)nfs_readdir_filler, desc); 333 if (IS_ERR(page)) { 334 status = PTR_ERR(page); 335 goto out; 336 } 337 338 /* NOTE: Someone else may have changed the READDIRPLUS flag */ 339 desc->page = page; 340 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */ 341 if (*desc->dir_cookie != 0) 342 status = find_dirent(desc); 343 else 344 status = find_dirent_index(desc); 345 if (status < 0) 346 dir_page_release(desc); 347 out: 348 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, status); 349 return status; 350} 351 352/* 353 * Recurse through the page cache pages, and return a 354 * filled nfs_entry structure of the next directory entry if possible. 355 * 356 * The target for the search is '*desc->dir_cookie' if non-0, 357 * 'desc->file->f_pos' otherwise 358 */ 359static inline 360int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) 361{ 362 int loop_count = 0; 363 int res; 364 365 /* Always search-by-index from the beginning of the cache */ 366 if (*desc->dir_cookie == 0) { 367 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n", 368 (long long)desc->file->f_pos); 369 desc->page_index = 0; 370 desc->entry->cookie = desc->entry->prev_cookie = 0; 371 desc->entry->eof = 0; 372 desc->current_index = 0; 373 } else 374 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n", 375 (unsigned long long)*desc->dir_cookie); 376 377 for (;;) { 378 res = find_dirent_page(desc); 379 if (res != -EAGAIN) 380 break; 381 /* Align to beginning of next page */ 382 desc->page_index ++; 383 if (loop_count++ > 200) { 384 loop_count = 0; 385 schedule(); 386 } 387 } 388 389 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, res); 390 return res; 391} 392 393static inline unsigned int dt_type(struct inode *inode) 394{ 395 return (inode->i_mode >> 12) & 15; 396} 397 398static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc); 399 400/* 401 * Once we've found the start of the dirent within a page: fill 'er up... 402 */ 403static 404int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent, 405 filldir_t filldir) 406{ 407 struct file *file = desc->file; 408 struct nfs_entry *entry = desc->entry; 409 struct dentry *dentry = NULL; 410 unsigned long fileid; 411 int loop_count = 0, 412 res; 413 414 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n", 415 (unsigned long long)entry->cookie); 416 417 for(;;) { 418 unsigned d_type = DT_UNKNOWN; 419 /* Note: entry->prev_cookie contains the cookie for 420 * retrieving the current dirent on the server */ 421 fileid = nfs_fileid_to_ino_t(entry->ino); 422 423 /* Get a dentry if we have one */ 424 if (dentry != NULL) 425 dput(dentry); 426 dentry = nfs_readdir_lookup(desc); 427 428 /* Use readdirplus info */ 429 if (dentry != NULL && dentry->d_inode != NULL) { 430 d_type = dt_type(dentry->d_inode); 431 fileid = dentry->d_inode->i_ino; 432 } 433 434 res = filldir(dirent, entry->name, entry->len, 435 file->f_pos, fileid, d_type); 436 if (res < 0) 437 break; 438 file->f_pos++; 439 *desc->dir_cookie = entry->cookie; 440 if (dir_decode(desc) != 0) { 441 desc->page_index ++; 442 break; 443 } 444 if (loop_count++ > 200) { 445 loop_count = 0; 446 schedule(); 447 } 448 } 449 dir_page_release(desc); 450 if (dentry != NULL) 451 dput(dentry); 452 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", 453 (unsigned long long)*desc->dir_cookie, res); 454 return res; 455} 456 457/* 458 * If we cannot find a cookie in our cache, we suspect that this is 459 * because it points to a deleted file, so we ask the server to return 460 * whatever it thinks is the next entry. We then feed this to filldir. 461 * If all goes well, we should then be able to find our way round the 462 * cache on the next call to readdir_search_pagecache(); 463 * 464 * NOTE: we cannot add the anonymous page to the pagecache because 465 * the data it contains might not be page aligned. Besides, 466 * we should already have a complete representation of the 467 * directory in the page cache by the time we get here. 468 */ 469static inline 470int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent, 471 filldir_t filldir) 472{ 473 struct file *file = desc->file; 474 struct inode *inode = file->f_path.dentry->d_inode; 475 struct rpc_cred *cred = nfs_file_cred(file); 476 struct page *page = NULL; 477 int status; 478 unsigned long timestamp; 479 480 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", 481 (unsigned long long)*desc->dir_cookie); 482 483 page = alloc_page(GFP_HIGHUSER); 484 if (!page) { 485 status = -ENOMEM; 486 goto out; 487 } 488 timestamp = jiffies; 489 desc->error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, *desc->dir_cookie, 490 page, 491 NFS_SERVER(inode)->dtsize, 492 desc->plus); 493 spin_lock(&inode->i_lock); 494 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME; 495 spin_unlock(&inode->i_lock); 496 desc->page = page; 497 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */ 498 if (desc->error >= 0) { 499 desc->timestamp = timestamp; 500 desc->timestamp_valid = 1; 501 if ((status = dir_decode(desc)) == 0) 502 desc->entry->prev_cookie = *desc->dir_cookie; 503 } else 504 status = -EIO; 505 if (status < 0) 506 goto out_release; 507 508 status = nfs_do_filldir(desc, dirent, filldir); 509 510 /* Reset read descriptor so it searches the page cache from 511 * the start upon the next call to readdir_search_pagecache() */ 512 desc->page_index = 0; 513 desc->entry->cookie = desc->entry->prev_cookie = 0; 514 desc->entry->eof = 0; 515 out: 516 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", 517 __FUNCTION__, status); 518 return status; 519 out_release: 520 dir_page_release(desc); 521 goto out; 522} 523 524/* The file offset position represents the dirent entry number. A 525 last cookie cache takes care of the common case of reading the 526 whole directory. 527 */ 528static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir) 529{ 530 struct dentry *dentry = filp->f_path.dentry; 531 struct inode *inode = dentry->d_inode; 532 nfs_readdir_descriptor_t my_desc, 533 *desc = &my_desc; 534 struct nfs_entry my_entry; 535 struct nfs_fh fh; 536 struct nfs_fattr fattr; 537 long res; 538 539 dfprintk(VFS, "NFS: readdir(%s/%s) starting at cookie %Lu\n", 540 dentry->d_parent->d_name.name, dentry->d_name.name, 541 (long long)filp->f_pos); 542 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); 543 544 lock_kernel(); 545 546 res = nfs_revalidate_mapping_nolock(inode, filp->f_mapping); 547 if (res < 0) { 548 unlock_kernel(); 549 return res; 550 } 551 552 /* 553 * filp->f_pos points to the dirent entry number. 554 * *desc->dir_cookie has the cookie for the next entry. We have 555 * to either find the entry with the appropriate number or 556 * revalidate the cookie. 557 */ 558 memset(desc, 0, sizeof(*desc)); 559 560 desc->file = filp; 561 desc->dir_cookie = &((struct nfs_open_context *)filp->private_data)->dir_cookie; 562 desc->decode = NFS_PROTO(inode)->decode_dirent; 563 desc->plus = NFS_USE_READDIRPLUS(inode); 564 565 my_entry.cookie = my_entry.prev_cookie = 0; 566 my_entry.eof = 0; 567 my_entry.fh = &fh; 568 my_entry.fattr = &fattr; 569 nfs_fattr_init(&fattr); 570 desc->entry = &my_entry; 571 572 while(!desc->entry->eof) { 573 res = readdir_search_pagecache(desc); 574 575 if (res == -EBADCOOKIE) { 576 /* This means either end of directory */ 577 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) { 578 /* Or that the server has 'lost' a cookie */ 579 res = uncached_readdir(desc, dirent, filldir); 580 if (res >= 0) 581 continue; 582 } 583 res = 0; 584 break; 585 } 586 if (res == -ETOOSMALL && desc->plus) { 587 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode)); 588 nfs_zap_caches(inode); 589 desc->plus = 0; 590 desc->entry->eof = 0; 591 continue; 592 } 593 if (res < 0) 594 break; 595 596 res = nfs_do_filldir(desc, dirent, filldir); 597 if (res < 0) { 598 res = 0; 599 break; 600 } 601 } 602 unlock_kernel(); 603 if (res > 0) 604 res = 0; 605 dfprintk(VFS, "NFS: readdir(%s/%s) returns %ld\n", 606 dentry->d_parent->d_name.name, dentry->d_name.name, 607 res); 608 return res; 609} 610 611static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin) 612{ 613 mutex_lock(&filp->f_path.dentry->d_inode->i_mutex); 614 switch (origin) { 615 case 1: 616 offset += filp->f_pos; 617 case 0: 618 if (offset >= 0) 619 break; 620 default: 621 offset = -EINVAL; 622 goto out; 623 } 624 if (offset != filp->f_pos) { 625 filp->f_pos = offset; 626 ((struct nfs_open_context *)filp->private_data)->dir_cookie = 0; 627 } 628out: 629 mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex); 630 return offset; 631} 632 633/* 634 * All directory operations under NFS are synchronous, so fsync() 635 * is a dummy operation. 636 */ 637static int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync) 638{ 639 dfprintk(VFS, "NFS: fsync_dir(%s/%s) datasync %d\n", 640 dentry->d_parent->d_name.name, dentry->d_name.name, 641 datasync); 642 643 return 0; 644} 645 646/* 647 * A check for whether or not the parent directory has changed. 648 * In the case it has, we assume that the dentries are untrustworthy 649 * and may need to be looked up again. 650 */ 651static int nfs_check_verifier(struct inode *dir, struct dentry *dentry) 652{ 653 unsigned long verf; 654 655 if (IS_ROOT(dentry)) 656 return 1; 657 verf = (unsigned long)dentry->d_fsdata; 658 if (nfs_caches_unstable(dir) 659 || verf != NFS_I(dir)->cache_change_attribute) 660 return 0; 661 return 1; 662} 663 664static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf) 665{ 666 dentry->d_fsdata = (void *)verf; 667} 668 669static void nfs_refresh_verifier(struct dentry * dentry, unsigned long verf) 670{ 671 nfs_set_verifier(dentry, verf); 672} 673 674/* 675 * Whenever an NFS operation succeeds, we know that the dentry 676 * is valid, so we update the revalidation timestamp. 677 */ 678static inline void nfs_renew_times(struct dentry * dentry) 679{ 680 dentry->d_time = jiffies; 681} 682 683/* 684 * Return the intent data that applies to this particular path component 685 * 686 * Note that the current set of intents only apply to the very last 687 * component of the path. 688 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT. 689 */ 690static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask) 691{ 692 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT)) 693 return 0; 694 return nd->flags & mask; 695} 696 697/* 698 * Inode and filehandle revalidation for lookups. 699 * 700 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, 701 * or if the intent information indicates that we're about to open this 702 * particular file and the "nocto" mount flag is not set. 703 * 704 */ 705static inline 706int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd) 707{ 708 struct nfs_server *server = NFS_SERVER(inode); 709 710 if (nd != NULL) { 711 /* VFS wants an on-the-wire revalidation */ 712 if (nd->flags & LOOKUP_REVAL) 713 goto out_force; 714 /* This is an open(2) */ 715 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 && 716 !(server->flags & NFS_MOUNT_NOCTO) && 717 (S_ISREG(inode->i_mode) || 718 S_ISDIR(inode->i_mode))) 719 goto out_force; 720 } 721 return nfs_revalidate_inode(server, inode); 722out_force: 723 return __nfs_revalidate_inode(server, inode); 724} 725 726/* 727 * We judge how long we want to trust negative 728 * dentries by looking at the parent inode mtime. 729 * 730 * If parent mtime has changed, we revalidate, else we wait for a 731 * period corresponding to the parent's attribute cache timeout value. 732 */ 733static inline 734int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, 735 struct nameidata *nd) 736{ 737 /* Don't revalidate a negative dentry if we're creating a new file */ 738 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0) 739 return 0; 740 return !nfs_check_verifier(dir, dentry); 741} 742 743/* 744 * This is called every time the dcache has a lookup hit, 745 * and we should check whether we can really trust that 746 * lookup. 747 * 748 * NOTE! The hit can be a negative hit too, don't assume 749 * we have an inode! 750 * 751 * If the parent directory is seen to have changed, we throw out the 752 * cached dentry and do a new lookup. 753 */ 754static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd) 755{ 756 struct inode *dir; 757 struct inode *inode; 758 struct dentry *parent; 759 int error; 760 struct nfs_fh fhandle; 761 struct nfs_fattr fattr; 762 unsigned long verifier; 763 764 parent = dget_parent(dentry); 765 lock_kernel(); 766 dir = parent->d_inode; 767 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); 768 inode = dentry->d_inode; 769 770 /* Revalidate parent directory attribute cache */ 771 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) 772 goto out_zap_parent; 773 774 if (!inode) { 775 if (nfs_neg_need_reval(dir, dentry, nd)) 776 goto out_bad; 777 goto out_valid; 778 } 779 780 if (is_bad_inode(inode)) { 781 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n", 782 __FUNCTION__, dentry->d_parent->d_name.name, 783 dentry->d_name.name); 784 goto out_bad; 785 } 786 787 /* Force a full look up iff the parent directory has changed */ 788 if (nfs_check_verifier(dir, dentry)) { 789 if (nfs_lookup_verify_inode(inode, nd)) 790 goto out_zap_parent; 791 goto out_valid; 792 } 793 794 if (NFS_STALE(inode)) 795 goto out_bad; 796 797 verifier = nfs_save_change_attribute(dir); 798 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr); 799 if (error) 800 goto out_bad; 801 if (nfs_compare_fh(NFS_FH(inode), &fhandle)) 802 goto out_bad; 803 if ((error = nfs_refresh_inode(inode, &fattr)) != 0) 804 goto out_bad; 805 806 nfs_renew_times(dentry); 807 nfs_refresh_verifier(dentry, verifier); 808 out_valid: 809 unlock_kernel(); 810 dput(parent); 811 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n", 812 __FUNCTION__, dentry->d_parent->d_name.name, 813 dentry->d_name.name); 814 return 1; 815out_zap_parent: 816 nfs_zap_caches(dir); 817 out_bad: 818 NFS_CACHEINV(dir); 819 if (inode && S_ISDIR(inode->i_mode)) { 820 /* Purge readdir caches. */ 821 nfs_zap_caches(inode); 822 /* If we have submounts, don't unhash ! */ 823 if (have_submounts(dentry)) 824 goto out_valid; 825 shrink_dcache_parent(dentry); 826 } 827 d_drop(dentry); 828 unlock_kernel(); 829 dput(parent); 830 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n", 831 __FUNCTION__, dentry->d_parent->d_name.name, 832 dentry->d_name.name); 833 return 0; 834} 835 836/* 837 * This is called from dput() when d_count is going to 0. 838 */ 839static int nfs_dentry_delete(struct dentry *dentry) 840{ 841 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n", 842 dentry->d_parent->d_name.name, dentry->d_name.name, 843 dentry->d_flags); 844 845 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 846 /* Unhash it, so that ->d_iput() would be called */ 847 return 1; 848 } 849 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { 850 /* Unhash it, so that ancestors of killed async unlink 851 * files will be cleaned up during umount */ 852 return 1; 853 } 854 return 0; 855 856} 857 858/* 859 * Called when the dentry loses inode. 860 * We use it to clean up silly-renamed files. 861 */ 862static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) 863{ 864 nfs_inode_return_delegation(inode); 865 if (S_ISDIR(inode->i_mode)) 866 /* drop any readdir cache as it could easily be old */ 867 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; 868 869 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 870 lock_kernel(); 871 drop_nlink(inode); 872 nfs_complete_unlink(dentry); 873 unlock_kernel(); 874 } 875 /* When creating a negative dentry, we want to renew d_time */ 876 nfs_renew_times(dentry); 877 iput(inode); 878} 879 880struct dentry_operations nfs_dentry_operations = { 881 .d_revalidate = nfs_lookup_revalidate, 882 .d_delete = nfs_dentry_delete, 883 .d_iput = nfs_dentry_iput, 884}; 885 886/* 887 * Use intent information to check whether or not we're going to do 888 * an O_EXCL create using this path component. 889 */ 890static inline 891int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd) 892{ 893 if (NFS_PROTO(dir)->version == 2) 894 return 0; 895 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0) 896 return 0; 897 return (nd->intent.open.flags & O_EXCL) != 0; 898} 899 900static inline int nfs_reval_fsid(struct vfsmount *mnt, struct inode *dir, 901 struct nfs_fh *fh, struct nfs_fattr *fattr) 902{ 903 struct nfs_server *server = NFS_SERVER(dir); 904 905 if (!nfs_fsid_equal(&server->fsid, &fattr->fsid)) 906 /* Revalidate fsid on root dir */ 907 return __nfs_revalidate_inode(server, mnt->mnt_root->d_inode); 908 return 0; 909} 910 911static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd) 912{ 913 struct dentry *res; 914 struct inode *inode = NULL; 915 int error; 916 struct nfs_fh fhandle; 917 struct nfs_fattr fattr; 918 919 dfprintk(VFS, "NFS: lookup(%s/%s)\n", 920 dentry->d_parent->d_name.name, dentry->d_name.name); 921 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); 922 923 res = ERR_PTR(-ENAMETOOLONG); 924 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) 925 goto out; 926 927 res = ERR_PTR(-ENOMEM); 928 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 929 930 lock_kernel(); 931 932 /* 933 * If we're doing an exclusive create, optimize away the lookup 934 * but don't hash the dentry. 935 */ 936 if (nfs_is_exclusive_create(dir, nd)) { 937 d_instantiate(dentry, NULL); 938 res = NULL; 939 goto out_unlock; 940 } 941 942 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr); 943 if (error == -ENOENT) 944 goto no_entry; 945 if (error < 0) { 946 res = ERR_PTR(error); 947 goto out_unlock; 948 } 949 error = nfs_reval_fsid(nd->mnt, dir, &fhandle, &fattr); 950 if (error < 0) { 951 res = ERR_PTR(error); 952 goto out_unlock; 953 } 954 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr); 955 res = (struct dentry *)inode; 956 if (IS_ERR(res)) 957 goto out_unlock; 958 959no_entry: 960 res = d_materialise_unique(dentry, inode); 961 if (res != NULL) { 962 struct dentry *parent; 963 if (IS_ERR(res)) 964 goto out_unlock; 965 /* Was a directory renamed! */ 966 parent = dget_parent(res); 967 if (!IS_ROOT(parent)) 968 nfs_mark_for_revalidate(parent->d_inode); 969 dput(parent); 970 dentry = res; 971 } 972 nfs_renew_times(dentry); 973 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 974out_unlock: 975 unlock_kernel(); 976out: 977 return res; 978} 979 980#ifdef CONFIG_NFS_V4 981static int nfs_open_revalidate(struct dentry *, struct nameidata *); 982 983struct dentry_operations nfs4_dentry_operations = { 984 .d_revalidate = nfs_open_revalidate, 985 .d_delete = nfs_dentry_delete, 986 .d_iput = nfs_dentry_iput, 987}; 988 989/* 990 * Use intent information to determine whether we need to substitute 991 * the NFSv4-style stateful OPEN for the LOOKUP call 992 */ 993static int is_atomic_open(struct inode *dir, struct nameidata *nd) 994{ 995 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0) 996 return 0; 997 /* NFS does not (yet) have a stateful open for directories */ 998 if (nd->flags & LOOKUP_DIRECTORY) 999 return 0; 1000 /* Are we trying to write to a read only partition? */ 1001 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE))) 1002 return 0; 1003 return 1; 1004} 1005 1006static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd) 1007{ 1008 struct dentry *res = NULL; 1009 int error; 1010 1011 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n", 1012 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1013 1014 /* Check that we are indeed trying to open this file */ 1015 if (!is_atomic_open(dir, nd)) 1016 goto no_open; 1017 1018 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) { 1019 res = ERR_PTR(-ENAMETOOLONG); 1020 goto out; 1021 } 1022 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 1023 1024 /* Let vfs_create() deal with O_EXCL */ 1025 if (nd->intent.open.flags & O_EXCL) { 1026 d_add(dentry, NULL); 1027 goto out; 1028 } 1029 1030 /* Open the file on the server */ 1031 lock_kernel(); 1032 /* Revalidate parent directory attribute cache */ 1033 error = nfs_revalidate_inode(NFS_SERVER(dir), dir); 1034 if (error < 0) { 1035 res = ERR_PTR(error); 1036 unlock_kernel(); 1037 goto out; 1038 } 1039 1040 if (nd->intent.open.flags & O_CREAT) { 1041 nfs_begin_data_update(dir); 1042 res = nfs4_atomic_open(dir, dentry, nd); 1043 nfs_end_data_update(dir); 1044 } else 1045 res = nfs4_atomic_open(dir, dentry, nd); 1046 unlock_kernel(); 1047 if (IS_ERR(res)) { 1048 error = PTR_ERR(res); 1049 switch (error) { 1050 /* Make a negative dentry */ 1051 case -ENOENT: 1052 res = NULL; 1053 goto out; 1054 /* This turned out not to be a regular file */ 1055 case -EISDIR: 1056 case -ENOTDIR: 1057 goto no_open; 1058 case -ELOOP: 1059 if (!(nd->intent.open.flags & O_NOFOLLOW)) 1060 goto no_open; 1061 /* case -EINVAL: */ 1062 default: 1063 goto out; 1064 } 1065 } else if (res != NULL) 1066 dentry = res; 1067 nfs_renew_times(dentry); 1068 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1069out: 1070 return res; 1071no_open: 1072 return nfs_lookup(dir, dentry, nd); 1073} 1074 1075static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd) 1076{ 1077 struct dentry *parent = NULL; 1078 struct inode *inode = dentry->d_inode; 1079 struct inode *dir; 1080 unsigned long verifier; 1081 int openflags, ret = 0; 1082 1083 parent = dget_parent(dentry); 1084 dir = parent->d_inode; 1085 if (!is_atomic_open(dir, nd)) 1086 goto no_open; 1087 /* We can't create new files in nfs_open_revalidate(), so we 1088 * optimize away revalidation of negative dentries. 1089 */ 1090 if (inode == NULL) 1091 goto out; 1092 /* NFS only supports OPEN on regular files */ 1093 if (!S_ISREG(inode->i_mode)) 1094 goto no_open; 1095 openflags = nd->intent.open.flags; 1096 /* We cannot do exclusive creation on a positive dentry */ 1097 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL)) 1098 goto no_open; 1099 /* We can't create new files, or truncate existing ones here */ 1100 openflags &= ~(O_CREAT|O_TRUNC); 1101 1102 /* 1103 * Note: we're not holding inode->i_mutex and so may be racing with 1104 * operations that change the directory. We therefore save the 1105 * change attribute *before* we do the RPC call. 1106 */ 1107 lock_kernel(); 1108 verifier = nfs_save_change_attribute(dir); 1109 ret = nfs4_open_revalidate(dir, dentry, openflags, nd); 1110 if (!ret) 1111 nfs_refresh_verifier(dentry, verifier); 1112 unlock_kernel(); 1113out: 1114 dput(parent); 1115 if (!ret) 1116 d_drop(dentry); 1117 return ret; 1118no_open: 1119 dput(parent); 1120 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ)) 1121 return 1; 1122 return nfs_lookup_revalidate(dentry, nd); 1123} 1124#endif /* CONFIG_NFSV4 */ 1125 1126static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc) 1127{ 1128 struct dentry *parent = desc->file->f_path.dentry; 1129 struct inode *dir = parent->d_inode; 1130 struct nfs_entry *entry = desc->entry; 1131 struct dentry *dentry, *alias; 1132 struct qstr name = { 1133 .name = entry->name, 1134 .len = entry->len, 1135 }; 1136 struct inode *inode; 1137 1138 switch (name.len) { 1139 case 2: 1140 if (name.name[0] == '.' && name.name[1] == '.') 1141 return dget_parent(parent); 1142 break; 1143 case 1: 1144 if (name.name[0] == '.') 1145 return dget(parent); 1146 } 1147 name.hash = full_name_hash(name.name, name.len); 1148 dentry = d_lookup(parent, &name); 1149 if (dentry != NULL) { 1150 /* Is this a positive dentry that matches the readdir info? */ 1151 if (dentry->d_inode != NULL && 1152 (NFS_FILEID(dentry->d_inode) == entry->ino || 1153 d_mountpoint(dentry))) { 1154 if (!desc->plus || entry->fh->size == 0) 1155 return dentry; 1156 if (nfs_compare_fh(NFS_FH(dentry->d_inode), 1157 entry->fh) == 0) 1158 goto out_renew; 1159 } 1160 /* No, so d_drop to allow one to be created */ 1161 d_drop(dentry); 1162 dput(dentry); 1163 } 1164 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR)) 1165 return NULL; 1166 /* Note: caller is already holding the dir->i_mutex! */ 1167 dentry = d_alloc(parent, &name); 1168 if (dentry == NULL) 1169 return NULL; 1170 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 1171 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr); 1172 if (IS_ERR(inode)) { 1173 dput(dentry); 1174 return NULL; 1175 } 1176 1177 alias = d_materialise_unique(dentry, inode); 1178 if (alias != NULL) { 1179 dput(dentry); 1180 if (IS_ERR(alias)) 1181 return NULL; 1182 dentry = alias; 1183 } 1184 1185 nfs_renew_times(dentry); 1186 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1187 return dentry; 1188out_renew: 1189 nfs_renew_times(dentry); 1190 nfs_refresh_verifier(dentry, nfs_save_change_attribute(dir)); 1191 return dentry; 1192} 1193 1194/* 1195 * Code common to create, mkdir, and mknod. 1196 */ 1197int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, 1198 struct nfs_fattr *fattr) 1199{ 1200 struct inode *inode; 1201 int error = -EACCES; 1202 1203 /* We may have been initialized further down */ 1204 if (dentry->d_inode) 1205 return 0; 1206 if (fhandle->size == 0) { 1207 struct inode *dir = dentry->d_parent->d_inode; 1208 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); 1209 if (error) 1210 return error; 1211 } 1212 if (!(fattr->valid & NFS_ATTR_FATTR)) { 1213 struct nfs_server *server = NFS_SB(dentry->d_sb); 1214 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr); 1215 if (error < 0) 1216 return error; 1217 } 1218 inode = nfs_fhget(dentry->d_sb, fhandle, fattr); 1219 error = PTR_ERR(inode); 1220 if (IS_ERR(inode)) 1221 return error; 1222 d_instantiate(dentry, inode); 1223 if (d_unhashed(dentry)) 1224 d_rehash(dentry); 1225 return 0; 1226} 1227 1228/* 1229 * Following a failed create operation, we drop the dentry rather 1230 * than retain a negative dentry. This avoids a problem in the event 1231 * that the operation succeeded on the server, but an error in the 1232 * reply path made it appear to have failed. 1233 */ 1234static int nfs_create(struct inode *dir, struct dentry *dentry, int mode, 1235 struct nameidata *nd) 1236{ 1237 struct iattr attr; 1238 int error; 1239 int open_flags = 0; 1240 1241 dfprintk(VFS, "NFS: create(%s/%ld), %s\n", 1242 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1243 1244 attr.ia_mode = mode; 1245 attr.ia_valid = ATTR_MODE; 1246 1247 if (nd && (nd->flags & LOOKUP_CREATE)) 1248 open_flags = nd->intent.open.flags; 1249 1250 lock_kernel(); 1251 nfs_begin_data_update(dir); 1252 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd); 1253 nfs_end_data_update(dir); 1254 if (error != 0) 1255 goto out_err; 1256 nfs_renew_times(dentry); 1257 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1258 unlock_kernel(); 1259 return 0; 1260out_err: 1261 unlock_kernel(); 1262 d_drop(dentry); 1263 return error; 1264} 1265 1266/* 1267 * See comments for nfs_proc_create regarding failed operations. 1268 */ 1269static int 1270nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev) 1271{ 1272 struct iattr attr; 1273 int status; 1274 1275 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n", 1276 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1277 1278 if (!new_valid_dev(rdev)) 1279 return -EINVAL; 1280 1281 attr.ia_mode = mode; 1282 attr.ia_valid = ATTR_MODE; 1283 1284 lock_kernel(); 1285 nfs_begin_data_update(dir); 1286 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); 1287 nfs_end_data_update(dir); 1288 if (status != 0) 1289 goto out_err; 1290 nfs_renew_times(dentry); 1291 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1292 unlock_kernel(); 1293 return 0; 1294out_err: 1295 unlock_kernel(); 1296 d_drop(dentry); 1297 return status; 1298} 1299 1300/* 1301 * See comments for nfs_proc_create regarding failed operations. 1302 */ 1303static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1304{ 1305 struct iattr attr; 1306 int error; 1307 1308 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n", 1309 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1310 1311 attr.ia_valid = ATTR_MODE; 1312 attr.ia_mode = mode | S_IFDIR; 1313 1314 lock_kernel(); 1315 nfs_begin_data_update(dir); 1316 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); 1317 nfs_end_data_update(dir); 1318 if (error != 0) 1319 goto out_err; 1320 nfs_renew_times(dentry); 1321 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1322 unlock_kernel(); 1323 return 0; 1324out_err: 1325 d_drop(dentry); 1326 unlock_kernel(); 1327 return error; 1328} 1329 1330static int nfs_rmdir(struct inode *dir, struct dentry *dentry) 1331{ 1332 int error; 1333 1334 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n", 1335 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1336 1337 lock_kernel(); 1338 nfs_begin_data_update(dir); 1339 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 1340 /* Ensure the VFS deletes this inode */ 1341 if (error == 0 && dentry->d_inode != NULL) 1342 clear_nlink(dentry->d_inode); 1343 nfs_end_data_update(dir); 1344 unlock_kernel(); 1345 1346 return error; 1347} 1348 1349static int nfs_sillyrename(struct inode *dir, struct dentry *dentry) 1350{ 1351 static unsigned int sillycounter; 1352 const int i_inosize = sizeof(dir->i_ino)*2; 1353 const int countersize = sizeof(sillycounter)*2; 1354 const int slen = sizeof(".nfs") + i_inosize + countersize - 1; 1355 char silly[slen+1]; 1356 struct qstr qsilly; 1357 struct dentry *sdentry; 1358 int error = -EIO; 1359 1360 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n", 1361 dentry->d_parent->d_name.name, dentry->d_name.name, 1362 atomic_read(&dentry->d_count)); 1363 nfs_inc_stats(dir, NFSIOS_SILLYRENAME); 1364 1365 /* 1366 * We don't allow a dentry to be silly-renamed twice. 1367 */ 1368 error = -EBUSY; 1369 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) 1370 goto out; 1371 1372 sprintf(silly, ".nfs%*.*lx", 1373 i_inosize, i_inosize, dentry->d_inode->i_ino); 1374 1375 /* Return delegation in anticipation of the rename */ 1376 nfs_inode_return_delegation(dentry->d_inode); 1377 1378 sdentry = NULL; 1379 do { 1380 char *suffix = silly + slen - countersize; 1381 1382 dput(sdentry); 1383 sillycounter++; 1384 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter); 1385 1386 dfprintk(VFS, "NFS: trying to rename %s to %s\n", 1387 dentry->d_name.name, silly); 1388 1389 sdentry = lookup_one_len(silly, dentry->d_parent, slen); 1390 /* 1391 * N.B. Better to return EBUSY here ... it could be 1392 * dangerous to delete the file while it's in use. 1393 */ 1394 if (IS_ERR(sdentry)) 1395 goto out; 1396 } while(sdentry->d_inode != NULL); /* need negative lookup */ 1397 1398 qsilly.name = silly; 1399 qsilly.len = strlen(silly); 1400 nfs_begin_data_update(dir); 1401 if (dentry->d_inode) { 1402 nfs_begin_data_update(dentry->d_inode); 1403 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name, 1404 dir, &qsilly); 1405 nfs_mark_for_revalidate(dentry->d_inode); 1406 nfs_end_data_update(dentry->d_inode); 1407 } else 1408 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name, 1409 dir, &qsilly); 1410 nfs_end_data_update(dir); 1411 if (!error) { 1412 nfs_renew_times(dentry); 1413 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1414 d_move(dentry, sdentry); 1415 error = nfs_async_unlink(dentry); 1416 /* If we return 0 we don't unlink */ 1417 } 1418 dput(sdentry); 1419out: 1420 return error; 1421} 1422 1423/* 1424 * Remove a file after making sure there are no pending writes, 1425 * and after checking that the file has only one user. 1426 * 1427 * We invalidate the attribute cache and free the inode prior to the operation 1428 * to avoid possible races if the server reuses the inode. 1429 */ 1430static int nfs_safe_remove(struct dentry *dentry) 1431{ 1432 struct inode *dir = dentry->d_parent->d_inode; 1433 struct inode *inode = dentry->d_inode; 1434 int error = -EBUSY; 1435 1436 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n", 1437 dentry->d_parent->d_name.name, dentry->d_name.name); 1438 1439 /* If the dentry was sillyrenamed, we simply call d_delete() */ 1440 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1441 error = 0; 1442 goto out; 1443 } 1444 1445 nfs_begin_data_update(dir); 1446 if (inode != NULL) { 1447 nfs_inode_return_delegation(inode); 1448 nfs_begin_data_update(inode); 1449 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1450 /* The VFS may want to delete this inode */ 1451 if (error == 0) 1452 drop_nlink(inode); 1453 nfs_mark_for_revalidate(inode); 1454 nfs_end_data_update(inode); 1455 } else 1456 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1457 nfs_end_data_update(dir); 1458out: 1459 return error; 1460} 1461 1462/* We do silly rename. In case sillyrename() returns -EBUSY, the inode 1463 * belongs to an active ".nfs..." file and we return -EBUSY. 1464 * 1465 * If sillyrename() returns 0, we do nothing, otherwise we unlink. 1466 */ 1467static int nfs_unlink(struct inode *dir, struct dentry *dentry) 1468{ 1469 int error; 1470 int need_rehash = 0; 1471 1472 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id, 1473 dir->i_ino, dentry->d_name.name); 1474 1475 lock_kernel(); 1476 spin_lock(&dcache_lock); 1477 spin_lock(&dentry->d_lock); 1478 if (atomic_read(&dentry->d_count) > 1) { 1479 spin_unlock(&dentry->d_lock); 1480 spin_unlock(&dcache_lock); 1481 /* Start asynchronous writeout of the inode */ 1482 write_inode_now(dentry->d_inode, 0); 1483 error = nfs_sillyrename(dir, dentry); 1484 unlock_kernel(); 1485 return error; 1486 } 1487 if (!d_unhashed(dentry)) { 1488 __d_drop(dentry); 1489 need_rehash = 1; 1490 } 1491 spin_unlock(&dentry->d_lock); 1492 spin_unlock(&dcache_lock); 1493 error = nfs_safe_remove(dentry); 1494 if (!error) { 1495 nfs_renew_times(dentry); 1496 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1497 } else if (need_rehash) 1498 d_rehash(dentry); 1499 unlock_kernel(); 1500 return error; 1501} 1502 1503/* 1504 * To create a symbolic link, most file systems instantiate a new inode, 1505 * add a page to it containing the path, then write it out to the disk 1506 * using prepare_write/commit_write. 1507 * 1508 * Unfortunately the NFS client can't create the in-core inode first 1509 * because it needs a file handle to create an in-core inode (see 1510 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the 1511 * symlink request has completed on the server. 1512 * 1513 * So instead we allocate a raw page, copy the symname into it, then do 1514 * the SYMLINK request with the page as the buffer. If it succeeds, we 1515 * now have a new file handle and can instantiate an in-core NFS inode 1516 * and move the raw page into its mapping. 1517 */ 1518static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1519{ 1520 struct pagevec lru_pvec; 1521 struct page *page; 1522 char *kaddr; 1523 struct iattr attr; 1524 unsigned int pathlen = strlen(symname); 1525 int error; 1526 1527 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id, 1528 dir->i_ino, dentry->d_name.name, symname); 1529 1530 if (pathlen > PAGE_SIZE) 1531 return -ENAMETOOLONG; 1532 1533 attr.ia_mode = S_IFLNK | S_IRWXUGO; 1534 attr.ia_valid = ATTR_MODE; 1535 1536 lock_kernel(); 1537 1538 page = alloc_page(GFP_KERNEL); 1539 if (!page) { 1540 unlock_kernel(); 1541 return -ENOMEM; 1542 } 1543 1544 kaddr = kmap_atomic(page, KM_USER0); 1545 memcpy(kaddr, symname, pathlen); 1546 if (pathlen < PAGE_SIZE) 1547 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); 1548 kunmap_atomic(kaddr, KM_USER0); 1549 1550 nfs_begin_data_update(dir); 1551 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); 1552 nfs_end_data_update(dir); 1553 if (error != 0) { 1554 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n", 1555 dir->i_sb->s_id, dir->i_ino, 1556 dentry->d_name.name, symname, error); 1557 d_drop(dentry); 1558 __free_page(page); 1559 unlock_kernel(); 1560 return error; 1561 } 1562 1563 /* 1564 * No big deal if we can't add this page to the page cache here. 1565 * READLINK will get the missing page from the server if needed. 1566 */ 1567 pagevec_init(&lru_pvec, 0); 1568 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0, 1569 GFP_KERNEL)) { 1570 pagevec_add(&lru_pvec, page); 1571 pagevec_lru_add(&lru_pvec); 1572 SetPageUptodate(page); 1573 unlock_page(page); 1574 } else 1575 __free_page(page); 1576 1577 unlock_kernel(); 1578 return 0; 1579} 1580 1581static int 1582nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1583{ 1584 struct inode *inode = old_dentry->d_inode; 1585 int error; 1586 1587 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n", 1588 old_dentry->d_parent->d_name.name, old_dentry->d_name.name, 1589 dentry->d_parent->d_name.name, dentry->d_name.name); 1590 1591 lock_kernel(); 1592 nfs_begin_data_update(dir); 1593 nfs_begin_data_update(inode); 1594 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); 1595 if (error == 0) { 1596 atomic_inc(&inode->i_count); 1597 d_instantiate(dentry, inode); 1598 } 1599 nfs_end_data_update(inode); 1600 nfs_end_data_update(dir); 1601 unlock_kernel(); 1602 return error; 1603} 1604 1605static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, 1606 struct inode *new_dir, struct dentry *new_dentry) 1607{ 1608 struct inode *old_inode = old_dentry->d_inode; 1609 struct inode *new_inode = new_dentry->d_inode; 1610 struct dentry *dentry = NULL, *rehash = NULL; 1611 int error = -EBUSY; 1612 1613 /* 1614 * To prevent any new references to the target during the rename, 1615 * we unhash the dentry and free the inode in advance. 1616 */ 1617 lock_kernel(); 1618 if (!d_unhashed(new_dentry)) { 1619 d_drop(new_dentry); 1620 rehash = new_dentry; 1621 } 1622 1623 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n", 1624 old_dentry->d_parent->d_name.name, old_dentry->d_name.name, 1625 new_dentry->d_parent->d_name.name, new_dentry->d_name.name, 1626 atomic_read(&new_dentry->d_count)); 1627 1628 /* 1629 * First check whether the target is busy ... we can't 1630 * safely do _any_ rename if the target is in use. 1631 * 1632 * For files, make a copy of the dentry and then do a 1633 * silly-rename. If the silly-rename succeeds, the 1634 * copied dentry is hashed and becomes the new target. 1635 */ 1636 if (!new_inode) 1637 goto go_ahead; 1638 if (S_ISDIR(new_inode->i_mode)) { 1639 error = -EISDIR; 1640 if (!S_ISDIR(old_inode->i_mode)) 1641 goto out; 1642 } else if (atomic_read(&new_dentry->d_count) > 2) { 1643 int err; 1644 /* copy the target dentry's name */ 1645 dentry = d_alloc(new_dentry->d_parent, 1646 &new_dentry->d_name); 1647 if (!dentry) 1648 goto out; 1649 1650 /* silly-rename the existing target ... */ 1651 err = nfs_sillyrename(new_dir, new_dentry); 1652 if (!err) { 1653 new_dentry = rehash = dentry; 1654 new_inode = NULL; 1655 /* instantiate the replacement target */ 1656 d_instantiate(new_dentry, NULL); 1657 } else if (atomic_read(&new_dentry->d_count) > 1) 1658 /* dentry still busy? */ 1659 goto out; 1660 } else 1661 drop_nlink(new_inode); 1662 1663go_ahead: 1664 /* 1665 * ... prune child dentries and writebacks if needed. 1666 */ 1667 if (atomic_read(&old_dentry->d_count) > 1) { 1668 if (S_ISREG(old_inode->i_mode)) 1669 nfs_wb_all(old_inode); 1670 shrink_dcache_parent(old_dentry); 1671 } 1672 nfs_inode_return_delegation(old_inode); 1673 1674 if (new_inode != NULL) { 1675 nfs_inode_return_delegation(new_inode); 1676 d_delete(new_dentry); 1677 } 1678 1679 nfs_begin_data_update(old_dir); 1680 nfs_begin_data_update(new_dir); 1681 nfs_begin_data_update(old_inode); 1682 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name, 1683 new_dir, &new_dentry->d_name); 1684 nfs_mark_for_revalidate(old_inode); 1685 nfs_end_data_update(old_inode); 1686 nfs_end_data_update(new_dir); 1687 nfs_end_data_update(old_dir); 1688out: 1689 if (rehash) 1690 d_rehash(rehash); 1691 if (!error) { 1692 d_move(old_dentry, new_dentry); 1693 nfs_renew_times(new_dentry); 1694 nfs_refresh_verifier(new_dentry, nfs_save_change_attribute(new_dir)); 1695 } 1696 1697 /* new dentry created? */ 1698 if (dentry) 1699 dput(dentry); 1700 unlock_kernel(); 1701 return error; 1702} 1703 1704static DEFINE_SPINLOCK(nfs_access_lru_lock); 1705static LIST_HEAD(nfs_access_lru_list); 1706static atomic_long_t nfs_access_nr_entries; 1707 1708static void nfs_access_free_entry(struct nfs_access_entry *entry) 1709{ 1710 put_rpccred(entry->cred); 1711 kfree(entry); 1712 smp_mb__before_atomic_dec(); 1713 atomic_long_dec(&nfs_access_nr_entries); 1714 smp_mb__after_atomic_dec(); 1715} 1716 1717int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask) 1718{ 1719 LIST_HEAD(head); 1720 struct nfs_inode *nfsi; 1721 struct nfs_access_entry *cache; 1722 1723 spin_lock(&nfs_access_lru_lock); 1724restart: 1725 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) { 1726 struct inode *inode; 1727 1728 if (nr_to_scan-- == 0) 1729 break; 1730 inode = igrab(&nfsi->vfs_inode); 1731 if (inode == NULL) 1732 continue; 1733 spin_lock(&inode->i_lock); 1734 if (list_empty(&nfsi->access_cache_entry_lru)) 1735 goto remove_lru_entry; 1736 cache = list_entry(nfsi->access_cache_entry_lru.next, 1737 struct nfs_access_entry, lru); 1738 list_move(&cache->lru, &head); 1739 rb_erase(&cache->rb_node, &nfsi->access_cache); 1740 if (!list_empty(&nfsi->access_cache_entry_lru)) 1741 list_move_tail(&nfsi->access_cache_inode_lru, 1742 &nfs_access_lru_list); 1743 else { 1744remove_lru_entry: 1745 list_del_init(&nfsi->access_cache_inode_lru); 1746 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); 1747 } 1748 spin_unlock(&inode->i_lock); 1749 iput(inode); 1750 goto restart; 1751 } 1752 spin_unlock(&nfs_access_lru_lock); 1753 while (!list_empty(&head)) { 1754 cache = list_entry(head.next, struct nfs_access_entry, lru); 1755 list_del(&cache->lru); 1756 nfs_access_free_entry(cache); 1757 } 1758 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure; 1759} 1760 1761static void __nfs_access_zap_cache(struct inode *inode) 1762{ 1763 struct nfs_inode *nfsi = NFS_I(inode); 1764 struct rb_root *root_node = &nfsi->access_cache; 1765 struct rb_node *n, *dispose = NULL; 1766 struct nfs_access_entry *entry; 1767 1768 /* Unhook entries from the cache */ 1769 while ((n = rb_first(root_node)) != NULL) { 1770 entry = rb_entry(n, struct nfs_access_entry, rb_node); 1771 rb_erase(n, root_node); 1772 list_del(&entry->lru); 1773 n->rb_left = dispose; 1774 dispose = n; 1775 } 1776 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; 1777 spin_unlock(&inode->i_lock); 1778 1779 /* Now kill them all! */ 1780 while (dispose != NULL) { 1781 n = dispose; 1782 dispose = n->rb_left; 1783 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node)); 1784 } 1785} 1786 1787void nfs_access_zap_cache(struct inode *inode) 1788{ 1789 /* Remove from global LRU init */ 1790 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) { 1791 spin_lock(&nfs_access_lru_lock); 1792 list_del_init(&NFS_I(inode)->access_cache_inode_lru); 1793 spin_unlock(&nfs_access_lru_lock); 1794 } 1795 1796 spin_lock(&inode->i_lock); 1797 /* This will release the spinlock */ 1798 __nfs_access_zap_cache(inode); 1799} 1800 1801static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) 1802{ 1803 struct rb_node *n = NFS_I(inode)->access_cache.rb_node; 1804 struct nfs_access_entry *entry; 1805 1806 while (n != NULL) { 1807 entry = rb_entry(n, struct nfs_access_entry, rb_node); 1808 1809 if (cred < entry->cred) 1810 n = n->rb_left; 1811 else if (cred > entry->cred) 1812 n = n->rb_right; 1813 else 1814 return entry; 1815 } 1816 return NULL; 1817} 1818 1819int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) 1820{ 1821 struct nfs_inode *nfsi = NFS_I(inode); 1822 struct nfs_access_entry *cache; 1823 int err = -ENOENT; 1824 1825 spin_lock(&inode->i_lock); 1826 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 1827 goto out_zap; 1828 cache = nfs_access_search_rbtree(inode, cred); 1829 if (cache == NULL) 1830 goto out; 1831 if (time_after(jiffies, cache->jiffies + NFS_ATTRTIMEO(inode))) 1832 goto out_stale; 1833 res->jiffies = cache->jiffies; 1834 res->cred = cache->cred; 1835 res->mask = cache->mask; 1836 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); 1837 err = 0; 1838out: 1839 spin_unlock(&inode->i_lock); 1840 return err; 1841out_stale: 1842 rb_erase(&cache->rb_node, &nfsi->access_cache); 1843 list_del(&cache->lru); 1844 spin_unlock(&inode->i_lock); 1845 nfs_access_free_entry(cache); 1846 return -ENOENT; 1847out_zap: 1848 /* This will release the spinlock */ 1849 __nfs_access_zap_cache(inode); 1850 return -ENOENT; 1851} 1852 1853static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) 1854{ 1855 struct nfs_inode *nfsi = NFS_I(inode); 1856 struct rb_root *root_node = &nfsi->access_cache; 1857 struct rb_node **p = &root_node->rb_node; 1858 struct rb_node *parent = NULL; 1859 struct nfs_access_entry *entry; 1860 1861 spin_lock(&inode->i_lock); 1862 while (*p != NULL) { 1863 parent = *p; 1864 entry = rb_entry(parent, struct nfs_access_entry, rb_node); 1865 1866 if (set->cred < entry->cred) 1867 p = &parent->rb_left; 1868 else if (set->cred > entry->cred) 1869 p = &parent->rb_right; 1870 else 1871 goto found; 1872 } 1873 rb_link_node(&set->rb_node, parent, p); 1874 rb_insert_color(&set->rb_node, root_node); 1875 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 1876 spin_unlock(&inode->i_lock); 1877 return; 1878found: 1879 rb_replace_node(parent, &set->rb_node, root_node); 1880 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 1881 list_del(&entry->lru); 1882 spin_unlock(&inode->i_lock); 1883 nfs_access_free_entry(entry); 1884} 1885 1886void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) 1887{ 1888 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); 1889 if (cache == NULL) 1890 return; 1891 RB_CLEAR_NODE(&cache->rb_node); 1892 cache->jiffies = set->jiffies; 1893 cache->cred = get_rpccred(set->cred); 1894 cache->mask = set->mask; 1895 1896 nfs_access_add_rbtree(inode, cache); 1897 1898 /* Update accounting */ 1899 smp_mb__before_atomic_inc(); 1900 atomic_long_inc(&nfs_access_nr_entries); 1901 smp_mb__after_atomic_inc(); 1902 1903 /* Add inode to global LRU list */ 1904 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) { 1905 spin_lock(&nfs_access_lru_lock); 1906 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list); 1907 spin_unlock(&nfs_access_lru_lock); 1908 } 1909} 1910 1911static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) 1912{ 1913 struct nfs_access_entry cache; 1914 int status; 1915 1916 status = nfs_access_get_cached(inode, cred, &cache); 1917 if (status == 0) 1918 goto out; 1919 1920 /* Be clever: ask server to check for all possible rights */ 1921 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; 1922 cache.cred = cred; 1923 cache.jiffies = jiffies; 1924 status = NFS_PROTO(inode)->access(inode, &cache); 1925 if (status != 0) 1926 return status; 1927 nfs_access_add_cache(inode, &cache); 1928out: 1929 if ((cache.mask & mask) == mask) 1930 return 0; 1931 return -EACCES; 1932} 1933 1934int nfs_permission(struct inode *inode, int mask, struct nameidata *nd) 1935{ 1936 struct rpc_cred *cred; 1937 int res = 0; 1938 1939 nfs_inc_stats(inode, NFSIOS_VFSACCESS); 1940 1941 if (mask == 0) 1942 goto out; 1943 /* Is this sys_access() ? */ 1944 if (nd != NULL && (nd->flags & LOOKUP_ACCESS)) 1945 goto force_lookup; 1946 1947 switch (inode->i_mode & S_IFMT) { 1948 case S_IFLNK: 1949 goto out; 1950 case S_IFREG: 1951 /* NFSv4 has atomic_open... */ 1952 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN) 1953 && nd != NULL 1954 && (nd->flags & LOOKUP_OPEN)) 1955 goto out; 1956 break; 1957 case S_IFDIR: 1958 /* 1959 * Optimize away all write operations, since the server 1960 * will check permissions when we perform the op. 1961 */ 1962 if ((mask & MAY_WRITE) && !(mask & MAY_READ)) 1963 goto out; 1964 } 1965 1966force_lookup: 1967 lock_kernel(); 1968 1969 if (!NFS_PROTO(inode)->access) 1970 goto out_notsup; 1971 1972 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0); 1973 if (!IS_ERR(cred)) { 1974 res = nfs_do_access(inode, cred, mask); 1975 put_rpccred(cred); 1976 } else 1977 res = PTR_ERR(cred); 1978 unlock_kernel(); 1979out: 1980 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n", 1981 inode->i_sb->s_id, inode->i_ino, mask, res); 1982 return res; 1983out_notsup: 1984 res = nfs_revalidate_inode(NFS_SERVER(inode), inode); 1985 if (res == 0) 1986 res = generic_permission(inode, mask, NULL); 1987 unlock_kernel(); 1988 goto out; 1989} 1990 1991/* 1992 * Local variables: 1993 * version-control: t 1994 * kept-new-versions: 5 1995 * End: 1996 */ 1997