Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame^] | 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/namei.h> |
| 34 | |
| 35 | #include "delegation.h" |
| 36 | |
| 37 | #define NFS_PARANOIA 1 |
| 38 | /* #define NFS_DEBUG_VERBOSE 1 */ |
| 39 | |
| 40 | static int nfs_opendir(struct inode *, struct file *); |
| 41 | static int nfs_readdir(struct file *, void *, filldir_t); |
| 42 | static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *); |
| 43 | static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *); |
| 44 | static int nfs_mkdir(struct inode *, struct dentry *, int); |
| 45 | static int nfs_rmdir(struct inode *, struct dentry *); |
| 46 | static int nfs_unlink(struct inode *, struct dentry *); |
| 47 | static int nfs_symlink(struct inode *, struct dentry *, const char *); |
| 48 | static int nfs_link(struct dentry *, struct inode *, struct dentry *); |
| 49 | static int nfs_mknod(struct inode *, struct dentry *, int, dev_t); |
| 50 | static int nfs_rename(struct inode *, struct dentry *, |
| 51 | struct inode *, struct dentry *); |
| 52 | static int nfs_fsync_dir(struct file *, struct dentry *, int); |
| 53 | |
| 54 | struct file_operations nfs_dir_operations = { |
| 55 | .read = generic_read_dir, |
| 56 | .readdir = nfs_readdir, |
| 57 | .open = nfs_opendir, |
| 58 | .release = nfs_release, |
| 59 | .fsync = nfs_fsync_dir, |
| 60 | }; |
| 61 | |
| 62 | struct inode_operations nfs_dir_inode_operations = { |
| 63 | .create = nfs_create, |
| 64 | .lookup = nfs_lookup, |
| 65 | .link = nfs_link, |
| 66 | .unlink = nfs_unlink, |
| 67 | .symlink = nfs_symlink, |
| 68 | .mkdir = nfs_mkdir, |
| 69 | .rmdir = nfs_rmdir, |
| 70 | .mknod = nfs_mknod, |
| 71 | .rename = nfs_rename, |
| 72 | .permission = nfs_permission, |
| 73 | .getattr = nfs_getattr, |
| 74 | .setattr = nfs_setattr, |
| 75 | }; |
| 76 | |
| 77 | #ifdef CONFIG_NFS_V4 |
| 78 | |
| 79 | static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *); |
| 80 | struct inode_operations nfs4_dir_inode_operations = { |
| 81 | .create = nfs_create, |
| 82 | .lookup = nfs_atomic_lookup, |
| 83 | .link = nfs_link, |
| 84 | .unlink = nfs_unlink, |
| 85 | .symlink = nfs_symlink, |
| 86 | .mkdir = nfs_mkdir, |
| 87 | .rmdir = nfs_rmdir, |
| 88 | .mknod = nfs_mknod, |
| 89 | .rename = nfs_rename, |
| 90 | .permission = nfs_permission, |
| 91 | .getattr = nfs_getattr, |
| 92 | .setattr = nfs_setattr, |
| 93 | }; |
| 94 | |
| 95 | #endif /* CONFIG_NFS_V4 */ |
| 96 | |
| 97 | /* |
| 98 | * Open file |
| 99 | */ |
| 100 | static int |
| 101 | nfs_opendir(struct inode *inode, struct file *filp) |
| 102 | { |
| 103 | int res = 0; |
| 104 | |
| 105 | lock_kernel(); |
| 106 | /* Call generic open code in order to cache credentials */ |
| 107 | if (!res) |
| 108 | res = nfs_open(inode, filp); |
| 109 | unlock_kernel(); |
| 110 | return res; |
| 111 | } |
| 112 | |
| 113 | typedef u32 * (*decode_dirent_t)(u32 *, struct nfs_entry *, int); |
| 114 | typedef struct { |
| 115 | struct file *file; |
| 116 | struct page *page; |
| 117 | unsigned long page_index; |
| 118 | u32 *ptr; |
| 119 | u64 target; |
| 120 | struct nfs_entry *entry; |
| 121 | decode_dirent_t decode; |
| 122 | int plus; |
| 123 | int error; |
| 124 | } nfs_readdir_descriptor_t; |
| 125 | |
| 126 | /* Now we cache directories properly, by stuffing the dirent |
| 127 | * data directly in the page cache. |
| 128 | * |
| 129 | * Inode invalidation due to refresh etc. takes care of |
| 130 | * _everything_, no sloppy entry flushing logic, no extraneous |
| 131 | * copying, network direct to page cache, the way it was meant |
| 132 | * to be. |
| 133 | * |
| 134 | * NOTE: Dirent information verification is done always by the |
| 135 | * page-in of the RPC reply, nowhere else, this simplies |
| 136 | * things substantially. |
| 137 | */ |
| 138 | static |
| 139 | int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page) |
| 140 | { |
| 141 | struct file *file = desc->file; |
| 142 | struct inode *inode = file->f_dentry->d_inode; |
| 143 | struct rpc_cred *cred = nfs_file_cred(file); |
| 144 | unsigned long timestamp; |
| 145 | int error; |
| 146 | |
| 147 | dfprintk(VFS, "NFS: nfs_readdir_filler() reading cookie %Lu into page %lu.\n", (long long)desc->entry->cookie, page->index); |
| 148 | |
| 149 | again: |
| 150 | timestamp = jiffies; |
| 151 | error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, desc->entry->cookie, page, |
| 152 | NFS_SERVER(inode)->dtsize, desc->plus); |
| 153 | if (error < 0) { |
| 154 | /* We requested READDIRPLUS, but the server doesn't grok it */ |
| 155 | if (error == -ENOTSUPP && desc->plus) { |
| 156 | NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; |
| 157 | NFS_FLAGS(inode) &= ~NFS_INO_ADVISE_RDPLUS; |
| 158 | desc->plus = 0; |
| 159 | goto again; |
| 160 | } |
| 161 | goto error; |
| 162 | } |
| 163 | SetPageUptodate(page); |
| 164 | NFS_FLAGS(inode) |= NFS_INO_INVALID_ATIME; |
| 165 | /* Ensure consistent page alignment of the data. |
| 166 | * Note: assumes we have exclusive access to this mapping either |
| 167 | * throught inode->i_sem or some other mechanism. |
| 168 | */ |
| 169 | if (page->index == 0) { |
| 170 | invalidate_inode_pages(inode->i_mapping); |
| 171 | NFS_I(inode)->readdir_timestamp = timestamp; |
| 172 | } |
| 173 | unlock_page(page); |
| 174 | return 0; |
| 175 | error: |
| 176 | SetPageError(page); |
| 177 | unlock_page(page); |
| 178 | nfs_zap_caches(inode); |
| 179 | desc->error = error; |
| 180 | return -EIO; |
| 181 | } |
| 182 | |
| 183 | static inline |
| 184 | int dir_decode(nfs_readdir_descriptor_t *desc) |
| 185 | { |
| 186 | u32 *p = desc->ptr; |
| 187 | p = desc->decode(p, desc->entry, desc->plus); |
| 188 | if (IS_ERR(p)) |
| 189 | return PTR_ERR(p); |
| 190 | desc->ptr = p; |
| 191 | return 0; |
| 192 | } |
| 193 | |
| 194 | static inline |
| 195 | void dir_page_release(nfs_readdir_descriptor_t *desc) |
| 196 | { |
| 197 | kunmap(desc->page); |
| 198 | page_cache_release(desc->page); |
| 199 | desc->page = NULL; |
| 200 | desc->ptr = NULL; |
| 201 | } |
| 202 | |
| 203 | /* |
| 204 | * Given a pointer to a buffer that has already been filled by a call |
| 205 | * to readdir, find the next entry. |
| 206 | * |
| 207 | * If the end of the buffer has been reached, return -EAGAIN, if not, |
| 208 | * return the offset within the buffer of the next entry to be |
| 209 | * read. |
| 210 | */ |
| 211 | static inline |
| 212 | int find_dirent(nfs_readdir_descriptor_t *desc, struct page *page) |
| 213 | { |
| 214 | struct nfs_entry *entry = desc->entry; |
| 215 | int loop_count = 0, |
| 216 | status; |
| 217 | |
| 218 | while((status = dir_decode(desc)) == 0) { |
| 219 | dfprintk(VFS, "NFS: found cookie %Lu\n", (long long)entry->cookie); |
| 220 | if (entry->prev_cookie == desc->target) |
| 221 | break; |
| 222 | if (loop_count++ > 200) { |
| 223 | loop_count = 0; |
| 224 | schedule(); |
| 225 | } |
| 226 | } |
| 227 | dfprintk(VFS, "NFS: find_dirent() returns %d\n", status); |
| 228 | return status; |
| 229 | } |
| 230 | |
| 231 | /* |
| 232 | * Find the given page, and call find_dirent() in order to try to |
| 233 | * return the next entry. |
| 234 | */ |
| 235 | static inline |
| 236 | int find_dirent_page(nfs_readdir_descriptor_t *desc) |
| 237 | { |
| 238 | struct inode *inode = desc->file->f_dentry->d_inode; |
| 239 | struct page *page; |
| 240 | int status; |
| 241 | |
| 242 | dfprintk(VFS, "NFS: find_dirent_page() searching directory page %ld\n", desc->page_index); |
| 243 | |
| 244 | page = read_cache_page(inode->i_mapping, desc->page_index, |
| 245 | (filler_t *)nfs_readdir_filler, desc); |
| 246 | if (IS_ERR(page)) { |
| 247 | status = PTR_ERR(page); |
| 248 | goto out; |
| 249 | } |
| 250 | if (!PageUptodate(page)) |
| 251 | goto read_error; |
| 252 | |
| 253 | /* NOTE: Someone else may have changed the READDIRPLUS flag */ |
| 254 | desc->page = page; |
| 255 | desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */ |
| 256 | status = find_dirent(desc, page); |
| 257 | if (status < 0) |
| 258 | dir_page_release(desc); |
| 259 | out: |
| 260 | dfprintk(VFS, "NFS: find_dirent_page() returns %d\n", status); |
| 261 | return status; |
| 262 | read_error: |
| 263 | page_cache_release(page); |
| 264 | return -EIO; |
| 265 | } |
| 266 | |
| 267 | /* |
| 268 | * Recurse through the page cache pages, and return a |
| 269 | * filled nfs_entry structure of the next directory entry if possible. |
| 270 | * |
| 271 | * The target for the search is 'desc->target'. |
| 272 | */ |
| 273 | static inline |
| 274 | int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) |
| 275 | { |
| 276 | int loop_count = 0; |
| 277 | int res; |
| 278 | |
| 279 | dfprintk(VFS, "NFS: readdir_search_pagecache() searching for cookie %Lu\n", (long long)desc->target); |
| 280 | for (;;) { |
| 281 | res = find_dirent_page(desc); |
| 282 | if (res != -EAGAIN) |
| 283 | break; |
| 284 | /* Align to beginning of next page */ |
| 285 | desc->page_index ++; |
| 286 | if (loop_count++ > 200) { |
| 287 | loop_count = 0; |
| 288 | schedule(); |
| 289 | } |
| 290 | } |
| 291 | dfprintk(VFS, "NFS: readdir_search_pagecache() returned %d\n", res); |
| 292 | return res; |
| 293 | } |
| 294 | |
| 295 | static inline unsigned int dt_type(struct inode *inode) |
| 296 | { |
| 297 | return (inode->i_mode >> 12) & 15; |
| 298 | } |
| 299 | |
| 300 | static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc); |
| 301 | |
| 302 | /* |
| 303 | * Once we've found the start of the dirent within a page: fill 'er up... |
| 304 | */ |
| 305 | static |
| 306 | int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent, |
| 307 | filldir_t filldir) |
| 308 | { |
| 309 | struct file *file = desc->file; |
| 310 | struct nfs_entry *entry = desc->entry; |
| 311 | struct dentry *dentry = NULL; |
| 312 | unsigned long fileid; |
| 313 | int loop_count = 0, |
| 314 | res; |
| 315 | |
| 316 | dfprintk(VFS, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n", (long long)desc->target); |
| 317 | |
| 318 | for(;;) { |
| 319 | unsigned d_type = DT_UNKNOWN; |
| 320 | /* Note: entry->prev_cookie contains the cookie for |
| 321 | * retrieving the current dirent on the server */ |
| 322 | fileid = nfs_fileid_to_ino_t(entry->ino); |
| 323 | |
| 324 | /* Get a dentry if we have one */ |
| 325 | if (dentry != NULL) |
| 326 | dput(dentry); |
| 327 | dentry = nfs_readdir_lookup(desc); |
| 328 | |
| 329 | /* Use readdirplus info */ |
| 330 | if (dentry != NULL && dentry->d_inode != NULL) { |
| 331 | d_type = dt_type(dentry->d_inode); |
| 332 | fileid = dentry->d_inode->i_ino; |
| 333 | } |
| 334 | |
| 335 | res = filldir(dirent, entry->name, entry->len, |
| 336 | entry->prev_cookie, fileid, d_type); |
| 337 | if (res < 0) |
| 338 | break; |
| 339 | file->f_pos = desc->target = entry->cookie; |
| 340 | if (dir_decode(desc) != 0) { |
| 341 | desc->page_index ++; |
| 342 | break; |
| 343 | } |
| 344 | if (loop_count++ > 200) { |
| 345 | loop_count = 0; |
| 346 | schedule(); |
| 347 | } |
| 348 | } |
| 349 | dir_page_release(desc); |
| 350 | if (dentry != NULL) |
| 351 | dput(dentry); |
| 352 | dfprintk(VFS, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", (long long)desc->target, res); |
| 353 | return res; |
| 354 | } |
| 355 | |
| 356 | /* |
| 357 | * If we cannot find a cookie in our cache, we suspect that this is |
| 358 | * because it points to a deleted file, so we ask the server to return |
| 359 | * whatever it thinks is the next entry. We then feed this to filldir. |
| 360 | * If all goes well, we should then be able to find our way round the |
| 361 | * cache on the next call to readdir_search_pagecache(); |
| 362 | * |
| 363 | * NOTE: we cannot add the anonymous page to the pagecache because |
| 364 | * the data it contains might not be page aligned. Besides, |
| 365 | * we should already have a complete representation of the |
| 366 | * directory in the page cache by the time we get here. |
| 367 | */ |
| 368 | static inline |
| 369 | int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent, |
| 370 | filldir_t filldir) |
| 371 | { |
| 372 | struct file *file = desc->file; |
| 373 | struct inode *inode = file->f_dentry->d_inode; |
| 374 | struct rpc_cred *cred = nfs_file_cred(file); |
| 375 | struct page *page = NULL; |
| 376 | int status; |
| 377 | |
| 378 | dfprintk(VFS, "NFS: uncached_readdir() searching for cookie %Lu\n", (long long)desc->target); |
| 379 | |
| 380 | page = alloc_page(GFP_HIGHUSER); |
| 381 | if (!page) { |
| 382 | status = -ENOMEM; |
| 383 | goto out; |
| 384 | } |
| 385 | desc->error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, desc->target, |
| 386 | page, |
| 387 | NFS_SERVER(inode)->dtsize, |
| 388 | desc->plus); |
| 389 | NFS_FLAGS(inode) |= NFS_INO_INVALID_ATIME; |
| 390 | desc->page = page; |
| 391 | desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */ |
| 392 | if (desc->error >= 0) { |
| 393 | if ((status = dir_decode(desc)) == 0) |
| 394 | desc->entry->prev_cookie = desc->target; |
| 395 | } else |
| 396 | status = -EIO; |
| 397 | if (status < 0) |
| 398 | goto out_release; |
| 399 | |
| 400 | status = nfs_do_filldir(desc, dirent, filldir); |
| 401 | |
| 402 | /* Reset read descriptor so it searches the page cache from |
| 403 | * the start upon the next call to readdir_search_pagecache() */ |
| 404 | desc->page_index = 0; |
| 405 | desc->entry->cookie = desc->entry->prev_cookie = 0; |
| 406 | desc->entry->eof = 0; |
| 407 | out: |
| 408 | dfprintk(VFS, "NFS: uncached_readdir() returns %d\n", status); |
| 409 | return status; |
| 410 | out_release: |
| 411 | dir_page_release(desc); |
| 412 | goto out; |
| 413 | } |
| 414 | |
| 415 | /* The file offset position is now represented as a true offset into the |
| 416 | * page cache as is the case in most of the other filesystems. |
| 417 | */ |
| 418 | static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir) |
| 419 | { |
| 420 | struct dentry *dentry = filp->f_dentry; |
| 421 | struct inode *inode = dentry->d_inode; |
| 422 | nfs_readdir_descriptor_t my_desc, |
| 423 | *desc = &my_desc; |
| 424 | struct nfs_entry my_entry; |
| 425 | struct nfs_fh fh; |
| 426 | struct nfs_fattr fattr; |
| 427 | long res; |
| 428 | |
| 429 | lock_kernel(); |
| 430 | |
| 431 | res = nfs_revalidate_inode(NFS_SERVER(inode), inode); |
| 432 | if (res < 0) { |
| 433 | unlock_kernel(); |
| 434 | return res; |
| 435 | } |
| 436 | |
| 437 | /* |
| 438 | * filp->f_pos points to the file offset in the page cache. |
| 439 | * but if the cache has meanwhile been zapped, we need to |
| 440 | * read from the last dirent to revalidate f_pos |
| 441 | * itself. |
| 442 | */ |
| 443 | memset(desc, 0, sizeof(*desc)); |
| 444 | |
| 445 | desc->file = filp; |
| 446 | desc->target = filp->f_pos; |
| 447 | desc->decode = NFS_PROTO(inode)->decode_dirent; |
| 448 | desc->plus = NFS_USE_READDIRPLUS(inode); |
| 449 | |
| 450 | my_entry.cookie = my_entry.prev_cookie = 0; |
| 451 | my_entry.eof = 0; |
| 452 | my_entry.fh = &fh; |
| 453 | my_entry.fattr = &fattr; |
| 454 | desc->entry = &my_entry; |
| 455 | |
| 456 | while(!desc->entry->eof) { |
| 457 | res = readdir_search_pagecache(desc); |
| 458 | if (res == -EBADCOOKIE) { |
| 459 | /* This means either end of directory */ |
| 460 | if (desc->entry->cookie != desc->target) { |
| 461 | /* Or that the server has 'lost' a cookie */ |
| 462 | res = uncached_readdir(desc, dirent, filldir); |
| 463 | if (res >= 0) |
| 464 | continue; |
| 465 | } |
| 466 | res = 0; |
| 467 | break; |
| 468 | } |
| 469 | if (res == -ETOOSMALL && desc->plus) { |
| 470 | NFS_FLAGS(inode) &= ~NFS_INO_ADVISE_RDPLUS; |
| 471 | nfs_zap_caches(inode); |
| 472 | desc->plus = 0; |
| 473 | desc->entry->eof = 0; |
| 474 | continue; |
| 475 | } |
| 476 | if (res < 0) |
| 477 | break; |
| 478 | |
| 479 | res = nfs_do_filldir(desc, dirent, filldir); |
| 480 | if (res < 0) { |
| 481 | res = 0; |
| 482 | break; |
| 483 | } |
| 484 | } |
| 485 | unlock_kernel(); |
| 486 | if (desc->error < 0) |
| 487 | return desc->error; |
| 488 | if (res < 0) |
| 489 | return res; |
| 490 | return 0; |
| 491 | } |
| 492 | |
| 493 | /* |
| 494 | * All directory operations under NFS are synchronous, so fsync() |
| 495 | * is a dummy operation. |
| 496 | */ |
| 497 | int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync) |
| 498 | { |
| 499 | return 0; |
| 500 | } |
| 501 | |
| 502 | /* |
| 503 | * A check for whether or not the parent directory has changed. |
| 504 | * In the case it has, we assume that the dentries are untrustworthy |
| 505 | * and may need to be looked up again. |
| 506 | */ |
| 507 | static inline int nfs_check_verifier(struct inode *dir, struct dentry *dentry) |
| 508 | { |
| 509 | if (IS_ROOT(dentry)) |
| 510 | return 1; |
| 511 | if ((NFS_FLAGS(dir) & NFS_INO_INVALID_ATTR) != 0 |
| 512 | || nfs_attribute_timeout(dir)) |
| 513 | return 0; |
| 514 | return nfs_verify_change_attribute(dir, (unsigned long)dentry->d_fsdata); |
| 515 | } |
| 516 | |
| 517 | static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf) |
| 518 | { |
| 519 | dentry->d_fsdata = (void *)verf; |
| 520 | } |
| 521 | |
| 522 | /* |
| 523 | * Whenever an NFS operation succeeds, we know that the dentry |
| 524 | * is valid, so we update the revalidation timestamp. |
| 525 | */ |
| 526 | static inline void nfs_renew_times(struct dentry * dentry) |
| 527 | { |
| 528 | dentry->d_time = jiffies; |
| 529 | } |
| 530 | |
| 531 | static inline |
| 532 | int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd) |
| 533 | { |
| 534 | struct nfs_server *server = NFS_SERVER(inode); |
| 535 | |
| 536 | if (nd != NULL) { |
| 537 | int ndflags = nd->flags; |
| 538 | /* VFS wants an on-the-wire revalidation */ |
| 539 | if (ndflags & LOOKUP_REVAL) |
| 540 | goto out_force; |
| 541 | /* This is an open(2) */ |
| 542 | if ((ndflags & LOOKUP_OPEN) && |
| 543 | !(ndflags & LOOKUP_CONTINUE) && |
| 544 | !(server->flags & NFS_MOUNT_NOCTO)) |
| 545 | goto out_force; |
| 546 | } |
| 547 | return nfs_revalidate_inode(server, inode); |
| 548 | out_force: |
| 549 | return __nfs_revalidate_inode(server, inode); |
| 550 | } |
| 551 | |
| 552 | /* |
| 553 | * We judge how long we want to trust negative |
| 554 | * dentries by looking at the parent inode mtime. |
| 555 | * |
| 556 | * If parent mtime has changed, we revalidate, else we wait for a |
| 557 | * period corresponding to the parent's attribute cache timeout value. |
| 558 | */ |
| 559 | static inline |
| 560 | int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, |
| 561 | struct nameidata *nd) |
| 562 | { |
| 563 | int ndflags = 0; |
| 564 | |
| 565 | if (nd) |
| 566 | ndflags = nd->flags; |
| 567 | /* Don't revalidate a negative dentry if we're creating a new file */ |
| 568 | if ((ndflags & LOOKUP_CREATE) && !(ndflags & LOOKUP_CONTINUE)) |
| 569 | return 0; |
| 570 | return !nfs_check_verifier(dir, dentry); |
| 571 | } |
| 572 | |
| 573 | /* |
| 574 | * This is called every time the dcache has a lookup hit, |
| 575 | * and we should check whether we can really trust that |
| 576 | * lookup. |
| 577 | * |
| 578 | * NOTE! The hit can be a negative hit too, don't assume |
| 579 | * we have an inode! |
| 580 | * |
| 581 | * If the parent directory is seen to have changed, we throw out the |
| 582 | * cached dentry and do a new lookup. |
| 583 | */ |
| 584 | static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd) |
| 585 | { |
| 586 | struct inode *dir; |
| 587 | struct inode *inode; |
| 588 | struct dentry *parent; |
| 589 | int error; |
| 590 | struct nfs_fh fhandle; |
| 591 | struct nfs_fattr fattr; |
| 592 | unsigned long verifier; |
| 593 | |
| 594 | parent = dget_parent(dentry); |
| 595 | lock_kernel(); |
| 596 | dir = parent->d_inode; |
| 597 | inode = dentry->d_inode; |
| 598 | |
| 599 | if (!inode) { |
| 600 | if (nfs_neg_need_reval(dir, dentry, nd)) |
| 601 | goto out_bad; |
| 602 | goto out_valid; |
| 603 | } |
| 604 | |
| 605 | if (is_bad_inode(inode)) { |
| 606 | dfprintk(VFS, "nfs_lookup_validate: %s/%s has dud inode\n", |
| 607 | dentry->d_parent->d_name.name, dentry->d_name.name); |
| 608 | goto out_bad; |
| 609 | } |
| 610 | |
| 611 | /* Revalidate parent directory attribute cache */ |
| 612 | if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) |
| 613 | goto out_zap_parent; |
| 614 | |
| 615 | /* Force a full look up iff the parent directory has changed */ |
| 616 | if (nfs_check_verifier(dir, dentry)) { |
| 617 | if (nfs_lookup_verify_inode(inode, nd)) |
| 618 | goto out_zap_parent; |
| 619 | goto out_valid; |
| 620 | } |
| 621 | |
| 622 | if (NFS_STALE(inode)) |
| 623 | goto out_bad; |
| 624 | |
| 625 | verifier = nfs_save_change_attribute(dir); |
| 626 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr); |
| 627 | if (error) |
| 628 | goto out_bad; |
| 629 | if (nfs_compare_fh(NFS_FH(inode), &fhandle)) |
| 630 | goto out_bad; |
| 631 | if ((error = nfs_refresh_inode(inode, &fattr)) != 0) |
| 632 | goto out_bad; |
| 633 | |
| 634 | nfs_renew_times(dentry); |
| 635 | nfs_set_verifier(dentry, verifier); |
| 636 | out_valid: |
| 637 | unlock_kernel(); |
| 638 | dput(parent); |
| 639 | return 1; |
| 640 | out_zap_parent: |
| 641 | nfs_zap_caches(dir); |
| 642 | out_bad: |
| 643 | NFS_CACHEINV(dir); |
| 644 | if (inode && S_ISDIR(inode->i_mode)) { |
| 645 | /* Purge readdir caches. */ |
| 646 | nfs_zap_caches(inode); |
| 647 | /* If we have submounts, don't unhash ! */ |
| 648 | if (have_submounts(dentry)) |
| 649 | goto out_valid; |
| 650 | shrink_dcache_parent(dentry); |
| 651 | } |
| 652 | d_drop(dentry); |
| 653 | unlock_kernel(); |
| 654 | dput(parent); |
| 655 | return 0; |
| 656 | } |
| 657 | |
| 658 | /* |
| 659 | * This is called from dput() when d_count is going to 0. |
| 660 | */ |
| 661 | static int nfs_dentry_delete(struct dentry *dentry) |
| 662 | { |
| 663 | dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n", |
| 664 | dentry->d_parent->d_name.name, dentry->d_name.name, |
| 665 | dentry->d_flags); |
| 666 | |
| 667 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
| 668 | /* Unhash it, so that ->d_iput() would be called */ |
| 669 | return 1; |
| 670 | } |
| 671 | if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { |
| 672 | /* Unhash it, so that ancestors of killed async unlink |
| 673 | * files will be cleaned up during umount */ |
| 674 | return 1; |
| 675 | } |
| 676 | return 0; |
| 677 | |
| 678 | } |
| 679 | |
| 680 | /* |
| 681 | * Called when the dentry loses inode. |
| 682 | * We use it to clean up silly-renamed files. |
| 683 | */ |
| 684 | static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) |
| 685 | { |
| 686 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
| 687 | lock_kernel(); |
| 688 | inode->i_nlink--; |
| 689 | nfs_complete_unlink(dentry); |
| 690 | unlock_kernel(); |
| 691 | } |
| 692 | /* When creating a negative dentry, we want to renew d_time */ |
| 693 | nfs_renew_times(dentry); |
| 694 | iput(inode); |
| 695 | } |
| 696 | |
| 697 | struct dentry_operations nfs_dentry_operations = { |
| 698 | .d_revalidate = nfs_lookup_revalidate, |
| 699 | .d_delete = nfs_dentry_delete, |
| 700 | .d_iput = nfs_dentry_iput, |
| 701 | }; |
| 702 | |
| 703 | static inline |
| 704 | int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd) |
| 705 | { |
| 706 | if (NFS_PROTO(dir)->version == 2) |
| 707 | return 0; |
| 708 | if (!nd || (nd->flags & LOOKUP_CONTINUE) || !(nd->flags & LOOKUP_CREATE)) |
| 709 | return 0; |
| 710 | return (nd->intent.open.flags & O_EXCL) != 0; |
| 711 | } |
| 712 | |
| 713 | static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd) |
| 714 | { |
| 715 | struct dentry *res; |
| 716 | struct inode *inode = NULL; |
| 717 | int error; |
| 718 | struct nfs_fh fhandle; |
| 719 | struct nfs_fattr fattr; |
| 720 | |
| 721 | dfprintk(VFS, "NFS: lookup(%s/%s)\n", |
| 722 | dentry->d_parent->d_name.name, dentry->d_name.name); |
| 723 | |
| 724 | res = ERR_PTR(-ENAMETOOLONG); |
| 725 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) |
| 726 | goto out; |
| 727 | |
| 728 | res = ERR_PTR(-ENOMEM); |
| 729 | dentry->d_op = NFS_PROTO(dir)->dentry_ops; |
| 730 | |
| 731 | lock_kernel(); |
| 732 | /* Revalidate parent directory attribute cache */ |
| 733 | error = nfs_revalidate_inode(NFS_SERVER(dir), dir); |
| 734 | if (error < 0) { |
| 735 | res = ERR_PTR(error); |
| 736 | goto out_unlock; |
| 737 | } |
| 738 | |
| 739 | /* If we're doing an exclusive create, optimize away the lookup */ |
| 740 | if (nfs_is_exclusive_create(dir, nd)) |
| 741 | goto no_entry; |
| 742 | |
| 743 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr); |
| 744 | if (error == -ENOENT) |
| 745 | goto no_entry; |
| 746 | if (error < 0) { |
| 747 | res = ERR_PTR(error); |
| 748 | goto out_unlock; |
| 749 | } |
| 750 | res = ERR_PTR(-EACCES); |
| 751 | inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr); |
| 752 | if (!inode) |
| 753 | goto out_unlock; |
| 754 | no_entry: |
| 755 | res = d_add_unique(dentry, inode); |
| 756 | if (res != NULL) |
| 757 | dentry = res; |
| 758 | nfs_renew_times(dentry); |
| 759 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
| 760 | out_unlock: |
| 761 | unlock_kernel(); |
| 762 | out: |
| 763 | return res; |
| 764 | } |
| 765 | |
| 766 | #ifdef CONFIG_NFS_V4 |
| 767 | static int nfs_open_revalidate(struct dentry *, struct nameidata *); |
| 768 | |
| 769 | struct dentry_operations nfs4_dentry_operations = { |
| 770 | .d_revalidate = nfs_open_revalidate, |
| 771 | .d_delete = nfs_dentry_delete, |
| 772 | .d_iput = nfs_dentry_iput, |
| 773 | }; |
| 774 | |
| 775 | static int is_atomic_open(struct inode *dir, struct nameidata *nd) |
| 776 | { |
| 777 | if (!nd) |
| 778 | return 0; |
| 779 | /* Check that we are indeed trying to open this file */ |
| 780 | if ((nd->flags & LOOKUP_CONTINUE) || !(nd->flags & LOOKUP_OPEN)) |
| 781 | return 0; |
| 782 | /* NFS does not (yet) have a stateful open for directories */ |
| 783 | if (nd->flags & LOOKUP_DIRECTORY) |
| 784 | return 0; |
| 785 | /* Are we trying to write to a read only partition? */ |
| 786 | if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE))) |
| 787 | return 0; |
| 788 | return 1; |
| 789 | } |
| 790 | |
| 791 | static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd) |
| 792 | { |
| 793 | struct dentry *res = NULL; |
| 794 | struct inode *inode = NULL; |
| 795 | int error; |
| 796 | |
| 797 | /* Check that we are indeed trying to open this file */ |
| 798 | if (!is_atomic_open(dir, nd)) |
| 799 | goto no_open; |
| 800 | |
| 801 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) { |
| 802 | res = ERR_PTR(-ENAMETOOLONG); |
| 803 | goto out; |
| 804 | } |
| 805 | dentry->d_op = NFS_PROTO(dir)->dentry_ops; |
| 806 | |
| 807 | /* Let vfs_create() deal with O_EXCL */ |
| 808 | if (nd->intent.open.flags & O_EXCL) |
| 809 | goto no_entry; |
| 810 | |
| 811 | /* Open the file on the server */ |
| 812 | lock_kernel(); |
| 813 | /* Revalidate parent directory attribute cache */ |
| 814 | error = nfs_revalidate_inode(NFS_SERVER(dir), dir); |
| 815 | if (error < 0) { |
| 816 | res = ERR_PTR(error); |
| 817 | goto out; |
| 818 | } |
| 819 | |
| 820 | if (nd->intent.open.flags & O_CREAT) { |
| 821 | nfs_begin_data_update(dir); |
| 822 | inode = nfs4_atomic_open(dir, dentry, nd); |
| 823 | nfs_end_data_update(dir); |
| 824 | } else |
| 825 | inode = nfs4_atomic_open(dir, dentry, nd); |
| 826 | unlock_kernel(); |
| 827 | if (IS_ERR(inode)) { |
| 828 | error = PTR_ERR(inode); |
| 829 | switch (error) { |
| 830 | /* Make a negative dentry */ |
| 831 | case -ENOENT: |
| 832 | inode = NULL; |
| 833 | break; |
| 834 | /* This turned out not to be a regular file */ |
| 835 | case -ELOOP: |
| 836 | if (!(nd->intent.open.flags & O_NOFOLLOW)) |
| 837 | goto no_open; |
| 838 | /* case -EISDIR: */ |
| 839 | /* case -EINVAL: */ |
| 840 | default: |
| 841 | res = ERR_PTR(error); |
| 842 | goto out; |
| 843 | } |
| 844 | } |
| 845 | no_entry: |
| 846 | res = d_add_unique(dentry, inode); |
| 847 | if (res != NULL) |
| 848 | dentry = res; |
| 849 | nfs_renew_times(dentry); |
| 850 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
| 851 | out: |
| 852 | return res; |
| 853 | no_open: |
| 854 | return nfs_lookup(dir, dentry, nd); |
| 855 | } |
| 856 | |
| 857 | static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd) |
| 858 | { |
| 859 | struct dentry *parent = NULL; |
| 860 | struct inode *inode = dentry->d_inode; |
| 861 | struct inode *dir; |
| 862 | unsigned long verifier; |
| 863 | int openflags, ret = 0; |
| 864 | |
| 865 | parent = dget_parent(dentry); |
| 866 | dir = parent->d_inode; |
| 867 | if (!is_atomic_open(dir, nd)) |
| 868 | goto no_open; |
| 869 | /* We can't create new files in nfs_open_revalidate(), so we |
| 870 | * optimize away revalidation of negative dentries. |
| 871 | */ |
| 872 | if (inode == NULL) |
| 873 | goto out; |
| 874 | /* NFS only supports OPEN on regular files */ |
| 875 | if (!S_ISREG(inode->i_mode)) |
| 876 | goto no_open; |
| 877 | openflags = nd->intent.open.flags; |
| 878 | /* We cannot do exclusive creation on a positive dentry */ |
| 879 | if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL)) |
| 880 | goto no_open; |
| 881 | /* We can't create new files, or truncate existing ones here */ |
| 882 | openflags &= ~(O_CREAT|O_TRUNC); |
| 883 | |
| 884 | /* |
| 885 | * Note: we're not holding inode->i_sem and so may be racing with |
| 886 | * operations that change the directory. We therefore save the |
| 887 | * change attribute *before* we do the RPC call. |
| 888 | */ |
| 889 | lock_kernel(); |
| 890 | verifier = nfs_save_change_attribute(dir); |
| 891 | ret = nfs4_open_revalidate(dir, dentry, openflags); |
| 892 | if (!ret) |
| 893 | nfs_set_verifier(dentry, verifier); |
| 894 | unlock_kernel(); |
| 895 | out: |
| 896 | dput(parent); |
| 897 | if (!ret) |
| 898 | d_drop(dentry); |
| 899 | return ret; |
| 900 | no_open: |
| 901 | dput(parent); |
| 902 | if (inode != NULL && nfs_have_delegation(inode, FMODE_READ)) |
| 903 | return 1; |
| 904 | return nfs_lookup_revalidate(dentry, nd); |
| 905 | } |
| 906 | #endif /* CONFIG_NFSV4 */ |
| 907 | |
| 908 | static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc) |
| 909 | { |
| 910 | struct dentry *parent = desc->file->f_dentry; |
| 911 | struct inode *dir = parent->d_inode; |
| 912 | struct nfs_entry *entry = desc->entry; |
| 913 | struct dentry *dentry, *alias; |
| 914 | struct qstr name = { |
| 915 | .name = entry->name, |
| 916 | .len = entry->len, |
| 917 | }; |
| 918 | struct inode *inode; |
| 919 | |
| 920 | switch (name.len) { |
| 921 | case 2: |
| 922 | if (name.name[0] == '.' && name.name[1] == '.') |
| 923 | return dget_parent(parent); |
| 924 | break; |
| 925 | case 1: |
| 926 | if (name.name[0] == '.') |
| 927 | return dget(parent); |
| 928 | } |
| 929 | name.hash = full_name_hash(name.name, name.len); |
| 930 | dentry = d_lookup(parent, &name); |
| 931 | if (dentry != NULL) |
| 932 | return dentry; |
| 933 | if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR)) |
| 934 | return NULL; |
| 935 | /* Note: caller is already holding the dir->i_sem! */ |
| 936 | dentry = d_alloc(parent, &name); |
| 937 | if (dentry == NULL) |
| 938 | return NULL; |
| 939 | dentry->d_op = NFS_PROTO(dir)->dentry_ops; |
| 940 | inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr); |
| 941 | if (!inode) { |
| 942 | dput(dentry); |
| 943 | return NULL; |
| 944 | } |
| 945 | alias = d_add_unique(dentry, inode); |
| 946 | if (alias != NULL) { |
| 947 | dput(dentry); |
| 948 | dentry = alias; |
| 949 | } |
| 950 | nfs_renew_times(dentry); |
| 951 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
| 952 | return dentry; |
| 953 | } |
| 954 | |
| 955 | /* |
| 956 | * Code common to create, mkdir, and mknod. |
| 957 | */ |
| 958 | int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, |
| 959 | struct nfs_fattr *fattr) |
| 960 | { |
| 961 | struct inode *inode; |
| 962 | int error = -EACCES; |
| 963 | |
| 964 | /* We may have been initialized further down */ |
| 965 | if (dentry->d_inode) |
| 966 | return 0; |
| 967 | if (fhandle->size == 0) { |
| 968 | struct inode *dir = dentry->d_parent->d_inode; |
| 969 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); |
| 970 | if (error) |
| 971 | goto out_err; |
| 972 | } |
| 973 | if (!(fattr->valid & NFS_ATTR_FATTR)) { |
| 974 | struct nfs_server *server = NFS_SB(dentry->d_sb); |
| 975 | error = server->rpc_ops->getattr(server, fhandle, fattr); |
| 976 | if (error < 0) |
| 977 | goto out_err; |
| 978 | } |
| 979 | error = -ENOMEM; |
| 980 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr); |
| 981 | if (inode == NULL) |
| 982 | goto out_err; |
| 983 | d_instantiate(dentry, inode); |
| 984 | return 0; |
| 985 | out_err: |
| 986 | d_drop(dentry); |
| 987 | return error; |
| 988 | } |
| 989 | |
| 990 | /* |
| 991 | * Following a failed create operation, we drop the dentry rather |
| 992 | * than retain a negative dentry. This avoids a problem in the event |
| 993 | * that the operation succeeded on the server, but an error in the |
| 994 | * reply path made it appear to have failed. |
| 995 | */ |
| 996 | static int nfs_create(struct inode *dir, struct dentry *dentry, int mode, |
| 997 | struct nameidata *nd) |
| 998 | { |
| 999 | struct iattr attr; |
| 1000 | int error; |
| 1001 | int open_flags = 0; |
| 1002 | |
| 1003 | dfprintk(VFS, "NFS: create(%s/%ld, %s\n", dir->i_sb->s_id, |
| 1004 | dir->i_ino, dentry->d_name.name); |
| 1005 | |
| 1006 | attr.ia_mode = mode; |
| 1007 | attr.ia_valid = ATTR_MODE; |
| 1008 | |
| 1009 | if (nd && (nd->flags & LOOKUP_CREATE)) |
| 1010 | open_flags = nd->intent.open.flags; |
| 1011 | |
| 1012 | lock_kernel(); |
| 1013 | nfs_begin_data_update(dir); |
| 1014 | error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags); |
| 1015 | nfs_end_data_update(dir); |
| 1016 | if (error != 0) |
| 1017 | goto out_err; |
| 1018 | nfs_renew_times(dentry); |
| 1019 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
| 1020 | unlock_kernel(); |
| 1021 | return 0; |
| 1022 | out_err: |
| 1023 | unlock_kernel(); |
| 1024 | d_drop(dentry); |
| 1025 | return error; |
| 1026 | } |
| 1027 | |
| 1028 | /* |
| 1029 | * See comments for nfs_proc_create regarding failed operations. |
| 1030 | */ |
| 1031 | static int |
| 1032 | nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev) |
| 1033 | { |
| 1034 | struct iattr attr; |
| 1035 | int status; |
| 1036 | |
| 1037 | dfprintk(VFS, "NFS: mknod(%s/%ld, %s\n", dir->i_sb->s_id, |
| 1038 | dir->i_ino, dentry->d_name.name); |
| 1039 | |
| 1040 | if (!new_valid_dev(rdev)) |
| 1041 | return -EINVAL; |
| 1042 | |
| 1043 | attr.ia_mode = mode; |
| 1044 | attr.ia_valid = ATTR_MODE; |
| 1045 | |
| 1046 | lock_kernel(); |
| 1047 | nfs_begin_data_update(dir); |
| 1048 | status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); |
| 1049 | nfs_end_data_update(dir); |
| 1050 | if (status != 0) |
| 1051 | goto out_err; |
| 1052 | nfs_renew_times(dentry); |
| 1053 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
| 1054 | unlock_kernel(); |
| 1055 | return 0; |
| 1056 | out_err: |
| 1057 | unlock_kernel(); |
| 1058 | d_drop(dentry); |
| 1059 | return status; |
| 1060 | } |
| 1061 | |
| 1062 | /* |
| 1063 | * See comments for nfs_proc_create regarding failed operations. |
| 1064 | */ |
| 1065 | static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
| 1066 | { |
| 1067 | struct iattr attr; |
| 1068 | int error; |
| 1069 | |
| 1070 | dfprintk(VFS, "NFS: mkdir(%s/%ld, %s\n", dir->i_sb->s_id, |
| 1071 | dir->i_ino, dentry->d_name.name); |
| 1072 | |
| 1073 | attr.ia_valid = ATTR_MODE; |
| 1074 | attr.ia_mode = mode | S_IFDIR; |
| 1075 | |
| 1076 | lock_kernel(); |
| 1077 | nfs_begin_data_update(dir); |
| 1078 | error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); |
| 1079 | nfs_end_data_update(dir); |
| 1080 | if (error != 0) |
| 1081 | goto out_err; |
| 1082 | nfs_renew_times(dentry); |
| 1083 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
| 1084 | unlock_kernel(); |
| 1085 | return 0; |
| 1086 | out_err: |
| 1087 | d_drop(dentry); |
| 1088 | unlock_kernel(); |
| 1089 | return error; |
| 1090 | } |
| 1091 | |
| 1092 | static int nfs_rmdir(struct inode *dir, struct dentry *dentry) |
| 1093 | { |
| 1094 | int error; |
| 1095 | |
| 1096 | dfprintk(VFS, "NFS: rmdir(%s/%ld, %s\n", dir->i_sb->s_id, |
| 1097 | dir->i_ino, dentry->d_name.name); |
| 1098 | |
| 1099 | lock_kernel(); |
| 1100 | nfs_begin_data_update(dir); |
| 1101 | error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); |
| 1102 | /* Ensure the VFS deletes this inode */ |
| 1103 | if (error == 0 && dentry->d_inode != NULL) |
| 1104 | dentry->d_inode->i_nlink = 0; |
| 1105 | nfs_end_data_update(dir); |
| 1106 | unlock_kernel(); |
| 1107 | |
| 1108 | return error; |
| 1109 | } |
| 1110 | |
| 1111 | static int nfs_sillyrename(struct inode *dir, struct dentry *dentry) |
| 1112 | { |
| 1113 | static unsigned int sillycounter; |
| 1114 | const int i_inosize = sizeof(dir->i_ino)*2; |
| 1115 | const int countersize = sizeof(sillycounter)*2; |
| 1116 | const int slen = sizeof(".nfs") + i_inosize + countersize - 1; |
| 1117 | char silly[slen+1]; |
| 1118 | struct qstr qsilly; |
| 1119 | struct dentry *sdentry; |
| 1120 | int error = -EIO; |
| 1121 | |
| 1122 | dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n", |
| 1123 | dentry->d_parent->d_name.name, dentry->d_name.name, |
| 1124 | atomic_read(&dentry->d_count)); |
| 1125 | |
| 1126 | #ifdef NFS_PARANOIA |
| 1127 | if (!dentry->d_inode) |
| 1128 | printk("NFS: silly-renaming %s/%s, negative dentry??\n", |
| 1129 | dentry->d_parent->d_name.name, dentry->d_name.name); |
| 1130 | #endif |
| 1131 | /* |
| 1132 | * We don't allow a dentry to be silly-renamed twice. |
| 1133 | */ |
| 1134 | error = -EBUSY; |
| 1135 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) |
| 1136 | goto out; |
| 1137 | |
| 1138 | sprintf(silly, ".nfs%*.*lx", |
| 1139 | i_inosize, i_inosize, dentry->d_inode->i_ino); |
| 1140 | |
| 1141 | sdentry = NULL; |
| 1142 | do { |
| 1143 | char *suffix = silly + slen - countersize; |
| 1144 | |
| 1145 | dput(sdentry); |
| 1146 | sillycounter++; |
| 1147 | sprintf(suffix, "%*.*x", countersize, countersize, sillycounter); |
| 1148 | |
| 1149 | dfprintk(VFS, "trying to rename %s to %s\n", |
| 1150 | dentry->d_name.name, silly); |
| 1151 | |
| 1152 | sdentry = lookup_one_len(silly, dentry->d_parent, slen); |
| 1153 | /* |
| 1154 | * N.B. Better to return EBUSY here ... it could be |
| 1155 | * dangerous to delete the file while it's in use. |
| 1156 | */ |
| 1157 | if (IS_ERR(sdentry)) |
| 1158 | goto out; |
| 1159 | } while(sdentry->d_inode != NULL); /* need negative lookup */ |
| 1160 | |
| 1161 | qsilly.name = silly; |
| 1162 | qsilly.len = strlen(silly); |
| 1163 | nfs_begin_data_update(dir); |
| 1164 | if (dentry->d_inode) { |
| 1165 | nfs_begin_data_update(dentry->d_inode); |
| 1166 | error = NFS_PROTO(dir)->rename(dir, &dentry->d_name, |
| 1167 | dir, &qsilly); |
| 1168 | nfs_end_data_update(dentry->d_inode); |
| 1169 | } else |
| 1170 | error = NFS_PROTO(dir)->rename(dir, &dentry->d_name, |
| 1171 | dir, &qsilly); |
| 1172 | nfs_end_data_update(dir); |
| 1173 | if (!error) { |
| 1174 | nfs_renew_times(dentry); |
| 1175 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
| 1176 | d_move(dentry, sdentry); |
| 1177 | error = nfs_async_unlink(dentry); |
| 1178 | /* If we return 0 we don't unlink */ |
| 1179 | } |
| 1180 | dput(sdentry); |
| 1181 | out: |
| 1182 | return error; |
| 1183 | } |
| 1184 | |
| 1185 | /* |
| 1186 | * Remove a file after making sure there are no pending writes, |
| 1187 | * and after checking that the file has only one user. |
| 1188 | * |
| 1189 | * We invalidate the attribute cache and free the inode prior to the operation |
| 1190 | * to avoid possible races if the server reuses the inode. |
| 1191 | */ |
| 1192 | static int nfs_safe_remove(struct dentry *dentry) |
| 1193 | { |
| 1194 | struct inode *dir = dentry->d_parent->d_inode; |
| 1195 | struct inode *inode = dentry->d_inode; |
| 1196 | int error = -EBUSY; |
| 1197 | |
| 1198 | dfprintk(VFS, "NFS: safe_remove(%s/%s)\n", |
| 1199 | dentry->d_parent->d_name.name, dentry->d_name.name); |
| 1200 | |
| 1201 | /* If the dentry was sillyrenamed, we simply call d_delete() */ |
| 1202 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
| 1203 | error = 0; |
| 1204 | goto out; |
| 1205 | } |
| 1206 | |
| 1207 | nfs_begin_data_update(dir); |
| 1208 | if (inode != NULL) { |
| 1209 | nfs_begin_data_update(inode); |
| 1210 | error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); |
| 1211 | /* The VFS may want to delete this inode */ |
| 1212 | if (error == 0) |
| 1213 | inode->i_nlink--; |
| 1214 | nfs_end_data_update(inode); |
| 1215 | } else |
| 1216 | error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); |
| 1217 | nfs_end_data_update(dir); |
| 1218 | out: |
| 1219 | return error; |
| 1220 | } |
| 1221 | |
| 1222 | /* We do silly rename. In case sillyrename() returns -EBUSY, the inode |
| 1223 | * belongs to an active ".nfs..." file and we return -EBUSY. |
| 1224 | * |
| 1225 | * If sillyrename() returns 0, we do nothing, otherwise we unlink. |
| 1226 | */ |
| 1227 | static int nfs_unlink(struct inode *dir, struct dentry *dentry) |
| 1228 | { |
| 1229 | int error; |
| 1230 | int need_rehash = 0; |
| 1231 | |
| 1232 | dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id, |
| 1233 | dir->i_ino, dentry->d_name.name); |
| 1234 | |
| 1235 | lock_kernel(); |
| 1236 | spin_lock(&dcache_lock); |
| 1237 | spin_lock(&dentry->d_lock); |
| 1238 | if (atomic_read(&dentry->d_count) > 1) { |
| 1239 | spin_unlock(&dentry->d_lock); |
| 1240 | spin_unlock(&dcache_lock); |
| 1241 | error = nfs_sillyrename(dir, dentry); |
| 1242 | unlock_kernel(); |
| 1243 | return error; |
| 1244 | } |
| 1245 | if (!d_unhashed(dentry)) { |
| 1246 | __d_drop(dentry); |
| 1247 | need_rehash = 1; |
| 1248 | } |
| 1249 | spin_unlock(&dentry->d_lock); |
| 1250 | spin_unlock(&dcache_lock); |
| 1251 | error = nfs_safe_remove(dentry); |
| 1252 | if (!error) { |
| 1253 | nfs_renew_times(dentry); |
| 1254 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
| 1255 | } else if (need_rehash) |
| 1256 | d_rehash(dentry); |
| 1257 | unlock_kernel(); |
| 1258 | return error; |
| 1259 | } |
| 1260 | |
| 1261 | static int |
| 1262 | nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) |
| 1263 | { |
| 1264 | struct iattr attr; |
| 1265 | struct nfs_fattr sym_attr; |
| 1266 | struct nfs_fh sym_fh; |
| 1267 | struct qstr qsymname; |
| 1268 | int error; |
| 1269 | |
| 1270 | dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id, |
| 1271 | dir->i_ino, dentry->d_name.name, symname); |
| 1272 | |
| 1273 | #ifdef NFS_PARANOIA |
| 1274 | if (dentry->d_inode) |
| 1275 | printk("nfs_proc_symlink: %s/%s not negative!\n", |
| 1276 | dentry->d_parent->d_name.name, dentry->d_name.name); |
| 1277 | #endif |
| 1278 | /* |
| 1279 | * Fill in the sattr for the call. |
| 1280 | * Note: SunOS 4.1.2 crashes if the mode isn't initialized! |
| 1281 | */ |
| 1282 | attr.ia_valid = ATTR_MODE; |
| 1283 | attr.ia_mode = S_IFLNK | S_IRWXUGO; |
| 1284 | |
| 1285 | qsymname.name = symname; |
| 1286 | qsymname.len = strlen(symname); |
| 1287 | |
| 1288 | lock_kernel(); |
| 1289 | nfs_begin_data_update(dir); |
| 1290 | error = NFS_PROTO(dir)->symlink(dir, &dentry->d_name, &qsymname, |
| 1291 | &attr, &sym_fh, &sym_attr); |
| 1292 | nfs_end_data_update(dir); |
| 1293 | if (!error) { |
| 1294 | error = nfs_instantiate(dentry, &sym_fh, &sym_attr); |
| 1295 | } else { |
| 1296 | if (error == -EEXIST) |
| 1297 | printk("nfs_proc_symlink: %s/%s already exists??\n", |
| 1298 | dentry->d_parent->d_name.name, dentry->d_name.name); |
| 1299 | d_drop(dentry); |
| 1300 | } |
| 1301 | unlock_kernel(); |
| 1302 | return error; |
| 1303 | } |
| 1304 | |
| 1305 | static int |
| 1306 | nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) |
| 1307 | { |
| 1308 | struct inode *inode = old_dentry->d_inode; |
| 1309 | int error; |
| 1310 | |
| 1311 | dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n", |
| 1312 | old_dentry->d_parent->d_name.name, old_dentry->d_name.name, |
| 1313 | dentry->d_parent->d_name.name, dentry->d_name.name); |
| 1314 | |
| 1315 | /* |
| 1316 | * Drop the dentry in advance to force a new lookup. |
| 1317 | * Since nfs_proc_link doesn't return a file handle, |
| 1318 | * we can't use the existing dentry. |
| 1319 | */ |
| 1320 | lock_kernel(); |
| 1321 | d_drop(dentry); |
| 1322 | |
| 1323 | nfs_begin_data_update(dir); |
| 1324 | nfs_begin_data_update(inode); |
| 1325 | error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); |
| 1326 | nfs_end_data_update(inode); |
| 1327 | nfs_end_data_update(dir); |
| 1328 | unlock_kernel(); |
| 1329 | return error; |
| 1330 | } |
| 1331 | |
| 1332 | /* |
| 1333 | * RENAME |
| 1334 | * FIXME: Some nfsds, like the Linux user space nfsd, may generate a |
| 1335 | * different file handle for the same inode after a rename (e.g. when |
| 1336 | * moving to a different directory). A fail-safe method to do so would |
| 1337 | * be to look up old_dir/old_name, create a link to new_dir/new_name and |
| 1338 | * rename the old file using the sillyrename stuff. This way, the original |
| 1339 | * file in old_dir will go away when the last process iput()s the inode. |
| 1340 | * |
| 1341 | * FIXED. |
| 1342 | * |
| 1343 | * It actually works quite well. One needs to have the possibility for |
| 1344 | * at least one ".nfs..." file in each directory the file ever gets |
| 1345 | * moved or linked to which happens automagically with the new |
| 1346 | * implementation that only depends on the dcache stuff instead of |
| 1347 | * using the inode layer |
| 1348 | * |
| 1349 | * Unfortunately, things are a little more complicated than indicated |
| 1350 | * above. For a cross-directory move, we want to make sure we can get |
| 1351 | * rid of the old inode after the operation. This means there must be |
| 1352 | * no pending writes (if it's a file), and the use count must be 1. |
| 1353 | * If these conditions are met, we can drop the dentries before doing |
| 1354 | * the rename. |
| 1355 | */ |
| 1356 | static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
| 1357 | struct inode *new_dir, struct dentry *new_dentry) |
| 1358 | { |
| 1359 | struct inode *old_inode = old_dentry->d_inode; |
| 1360 | struct inode *new_inode = new_dentry->d_inode; |
| 1361 | struct dentry *dentry = NULL, *rehash = NULL; |
| 1362 | int error = -EBUSY; |
| 1363 | |
| 1364 | /* |
| 1365 | * To prevent any new references to the target during the rename, |
| 1366 | * we unhash the dentry and free the inode in advance. |
| 1367 | */ |
| 1368 | lock_kernel(); |
| 1369 | if (!d_unhashed(new_dentry)) { |
| 1370 | d_drop(new_dentry); |
| 1371 | rehash = new_dentry; |
| 1372 | } |
| 1373 | |
| 1374 | dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n", |
| 1375 | old_dentry->d_parent->d_name.name, old_dentry->d_name.name, |
| 1376 | new_dentry->d_parent->d_name.name, new_dentry->d_name.name, |
| 1377 | atomic_read(&new_dentry->d_count)); |
| 1378 | |
| 1379 | /* |
| 1380 | * First check whether the target is busy ... we can't |
| 1381 | * safely do _any_ rename if the target is in use. |
| 1382 | * |
| 1383 | * For files, make a copy of the dentry and then do a |
| 1384 | * silly-rename. If the silly-rename succeeds, the |
| 1385 | * copied dentry is hashed and becomes the new target. |
| 1386 | */ |
| 1387 | if (!new_inode) |
| 1388 | goto go_ahead; |
| 1389 | if (S_ISDIR(new_inode->i_mode)) |
| 1390 | goto out; |
| 1391 | else if (atomic_read(&new_dentry->d_count) > 2) { |
| 1392 | int err; |
| 1393 | /* copy the target dentry's name */ |
| 1394 | dentry = d_alloc(new_dentry->d_parent, |
| 1395 | &new_dentry->d_name); |
| 1396 | if (!dentry) |
| 1397 | goto out; |
| 1398 | |
| 1399 | /* silly-rename the existing target ... */ |
| 1400 | err = nfs_sillyrename(new_dir, new_dentry); |
| 1401 | if (!err) { |
| 1402 | new_dentry = rehash = dentry; |
| 1403 | new_inode = NULL; |
| 1404 | /* instantiate the replacement target */ |
| 1405 | d_instantiate(new_dentry, NULL); |
| 1406 | } else if (atomic_read(&new_dentry->d_count) > 1) { |
| 1407 | /* dentry still busy? */ |
| 1408 | #ifdef NFS_PARANOIA |
| 1409 | printk("nfs_rename: target %s/%s busy, d_count=%d\n", |
| 1410 | new_dentry->d_parent->d_name.name, |
| 1411 | new_dentry->d_name.name, |
| 1412 | atomic_read(&new_dentry->d_count)); |
| 1413 | #endif |
| 1414 | goto out; |
| 1415 | } |
| 1416 | } |
| 1417 | |
| 1418 | go_ahead: |
| 1419 | /* |
| 1420 | * ... prune child dentries and writebacks if needed. |
| 1421 | */ |
| 1422 | if (atomic_read(&old_dentry->d_count) > 1) { |
| 1423 | nfs_wb_all(old_inode); |
| 1424 | shrink_dcache_parent(old_dentry); |
| 1425 | } |
| 1426 | |
| 1427 | if (new_inode) |
| 1428 | d_delete(new_dentry); |
| 1429 | |
| 1430 | nfs_begin_data_update(old_dir); |
| 1431 | nfs_begin_data_update(new_dir); |
| 1432 | nfs_begin_data_update(old_inode); |
| 1433 | error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name, |
| 1434 | new_dir, &new_dentry->d_name); |
| 1435 | nfs_end_data_update(old_inode); |
| 1436 | nfs_end_data_update(new_dir); |
| 1437 | nfs_end_data_update(old_dir); |
| 1438 | out: |
| 1439 | if (rehash) |
| 1440 | d_rehash(rehash); |
| 1441 | if (!error) { |
| 1442 | if (!S_ISDIR(old_inode->i_mode)) |
| 1443 | d_move(old_dentry, new_dentry); |
| 1444 | nfs_renew_times(new_dentry); |
| 1445 | nfs_set_verifier(new_dentry, nfs_save_change_attribute(new_dir)); |
| 1446 | } |
| 1447 | |
| 1448 | /* new dentry created? */ |
| 1449 | if (dentry) |
| 1450 | dput(dentry); |
| 1451 | unlock_kernel(); |
| 1452 | return error; |
| 1453 | } |
| 1454 | |
| 1455 | int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) |
| 1456 | { |
| 1457 | struct nfs_access_entry *cache = &NFS_I(inode)->cache_access; |
| 1458 | |
| 1459 | if (cache->cred != cred |
| 1460 | || time_after(jiffies, cache->jiffies + NFS_ATTRTIMEO(inode)) |
| 1461 | || (NFS_FLAGS(inode) & NFS_INO_INVALID_ACCESS)) |
| 1462 | return -ENOENT; |
| 1463 | memcpy(res, cache, sizeof(*res)); |
| 1464 | return 0; |
| 1465 | } |
| 1466 | |
| 1467 | void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) |
| 1468 | { |
| 1469 | struct nfs_access_entry *cache = &NFS_I(inode)->cache_access; |
| 1470 | |
| 1471 | if (cache->cred != set->cred) { |
| 1472 | if (cache->cred) |
| 1473 | put_rpccred(cache->cred); |
| 1474 | cache->cred = get_rpccred(set->cred); |
| 1475 | } |
| 1476 | NFS_FLAGS(inode) &= ~NFS_INO_INVALID_ACCESS; |
| 1477 | cache->jiffies = set->jiffies; |
| 1478 | cache->mask = set->mask; |
| 1479 | } |
| 1480 | |
| 1481 | static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) |
| 1482 | { |
| 1483 | struct nfs_access_entry cache; |
| 1484 | int status; |
| 1485 | |
| 1486 | status = nfs_access_get_cached(inode, cred, &cache); |
| 1487 | if (status == 0) |
| 1488 | goto out; |
| 1489 | |
| 1490 | /* Be clever: ask server to check for all possible rights */ |
| 1491 | cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; |
| 1492 | cache.cred = cred; |
| 1493 | cache.jiffies = jiffies; |
| 1494 | status = NFS_PROTO(inode)->access(inode, &cache); |
| 1495 | if (status != 0) |
| 1496 | return status; |
| 1497 | nfs_access_add_cache(inode, &cache); |
| 1498 | out: |
| 1499 | if ((cache.mask & mask) == mask) |
| 1500 | return 0; |
| 1501 | return -EACCES; |
| 1502 | } |
| 1503 | |
| 1504 | int nfs_permission(struct inode *inode, int mask, struct nameidata *nd) |
| 1505 | { |
| 1506 | struct rpc_cred *cred; |
| 1507 | int res = 0; |
| 1508 | |
| 1509 | if (mask == 0) |
| 1510 | goto out; |
| 1511 | /* Is this sys_access() ? */ |
| 1512 | if (nd != NULL && (nd->flags & LOOKUP_ACCESS)) |
| 1513 | goto force_lookup; |
| 1514 | |
| 1515 | switch (inode->i_mode & S_IFMT) { |
| 1516 | case S_IFLNK: |
| 1517 | goto out; |
| 1518 | case S_IFREG: |
| 1519 | /* NFSv4 has atomic_open... */ |
| 1520 | if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN) |
| 1521 | && nd != NULL |
| 1522 | && (nd->flags & LOOKUP_OPEN)) |
| 1523 | goto out; |
| 1524 | break; |
| 1525 | case S_IFDIR: |
| 1526 | /* |
| 1527 | * Optimize away all write operations, since the server |
| 1528 | * will check permissions when we perform the op. |
| 1529 | */ |
| 1530 | if ((mask & MAY_WRITE) && !(mask & MAY_READ)) |
| 1531 | goto out; |
| 1532 | } |
| 1533 | |
| 1534 | force_lookup: |
| 1535 | lock_kernel(); |
| 1536 | |
| 1537 | if (!NFS_PROTO(inode)->access) |
| 1538 | goto out_notsup; |
| 1539 | |
| 1540 | cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0); |
| 1541 | if (!IS_ERR(cred)) { |
| 1542 | res = nfs_do_access(inode, cred, mask); |
| 1543 | put_rpccred(cred); |
| 1544 | } else |
| 1545 | res = PTR_ERR(cred); |
| 1546 | unlock_kernel(); |
| 1547 | out: |
| 1548 | return res; |
| 1549 | out_notsup: |
| 1550 | res = nfs_revalidate_inode(NFS_SERVER(inode), inode); |
| 1551 | if (res == 0) |
| 1552 | res = generic_permission(inode, mask, NULL); |
| 1553 | unlock_kernel(); |
| 1554 | return res; |
| 1555 | } |
| 1556 | |
| 1557 | /* |
| 1558 | * Local variables: |
| 1559 | * version-control: t |
| 1560 | * kept-new-versions: 5 |
| 1561 | * End: |
| 1562 | */ |