| 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/module.h> |
| 21 | #include <linux/time.h> |
| 22 | #include <linux/errno.h> |
| 23 | #include <linux/stat.h> |
| 24 | #include <linux/fcntl.h> |
| 25 | #include <linux/string.h> |
| 26 | #include <linux/kernel.h> |
| 27 | #include <linux/slab.h> |
| 28 | #include <linux/mm.h> |
| 29 | #include <linux/sunrpc/clnt.h> |
| 30 | #include <linux/nfs_fs.h> |
| 31 | #include <linux/nfs_mount.h> |
| 32 | #include <linux/pagemap.h> |
| 33 | #include <linux/pagevec.h> |
| 34 | #include <linux/namei.h> |
| 35 | #include <linux/mount.h> |
| 36 | #include <linux/sched.h> |
| 37 | #include <linux/kmemleak.h> |
| 38 | #include <linux/xattr.h> |
| 39 | |
| 40 | #include "delegation.h" |
| 41 | #include "iostat.h" |
| 42 | #include "internal.h" |
| 43 | #include "fscache.h" |
| 44 | |
| 45 | /* #define NFS_DEBUG_VERBOSE 1 */ |
| 46 | |
| 47 | static int nfs_opendir(struct inode *, struct file *); |
| 48 | static int nfs_closedir(struct inode *, struct file *); |
| 49 | static int nfs_readdir(struct file *, void *, filldir_t); |
| 50 | static int nfs_fsync_dir(struct file *, loff_t, loff_t, int); |
| 51 | static loff_t nfs_llseek_dir(struct file *, loff_t, int); |
| 52 | static void nfs_readdir_clear_array(struct page*); |
| 53 | |
| 54 | const struct file_operations nfs_dir_operations = { |
| 55 | .llseek = nfs_llseek_dir, |
| 56 | .read = generic_read_dir, |
| 57 | .readdir = nfs_readdir, |
| 58 | .open = nfs_opendir, |
| 59 | .release = nfs_closedir, |
| 60 | .fsync = nfs_fsync_dir, |
| 61 | }; |
| 62 | |
| 63 | const struct address_space_operations nfs_dir_aops = { |
| 64 | .freepage = nfs_readdir_clear_array, |
| 65 | }; |
| 66 | |
| 67 | static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred) |
| 68 | { |
| 69 | struct nfs_open_dir_context *ctx; |
| 70 | ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); |
| 71 | if (ctx != NULL) { |
| 72 | ctx->duped = 0; |
| 73 | ctx->attr_gencount = NFS_I(dir)->attr_gencount; |
| 74 | ctx->dir_cookie = 0; |
| 75 | ctx->dup_cookie = 0; |
| 76 | ctx->cred = get_rpccred(cred); |
| 77 | return ctx; |
| 78 | } |
| 79 | return ERR_PTR(-ENOMEM); |
| 80 | } |
| 81 | |
| 82 | static void put_nfs_open_dir_context(struct nfs_open_dir_context *ctx) |
| 83 | { |
| 84 | put_rpccred(ctx->cred); |
| 85 | kfree(ctx); |
| 86 | } |
| 87 | |
| 88 | /* |
| 89 | * Open file |
| 90 | */ |
| 91 | static int |
| 92 | nfs_opendir(struct inode *inode, struct file *filp) |
| 93 | { |
| 94 | int res = 0; |
| 95 | struct nfs_open_dir_context *ctx; |
| 96 | struct rpc_cred *cred; |
| 97 | |
| 98 | dfprintk(FILE, "NFS: open dir(%s/%s)\n", |
| 99 | filp->f_path.dentry->d_parent->d_name.name, |
| 100 | filp->f_path.dentry->d_name.name); |
| 101 | |
| 102 | nfs_inc_stats(inode, NFSIOS_VFSOPEN); |
| 103 | |
| 104 | cred = rpc_lookup_cred(); |
| 105 | if (IS_ERR(cred)) |
| 106 | return PTR_ERR(cred); |
| 107 | ctx = alloc_nfs_open_dir_context(inode, cred); |
| 108 | if (IS_ERR(ctx)) { |
| 109 | res = PTR_ERR(ctx); |
| 110 | goto out; |
| 111 | } |
| 112 | filp->private_data = ctx; |
| 113 | if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) { |
| 114 | /* This is a mountpoint, so d_revalidate will never |
| 115 | * have been called, so we need to refresh the |
| 116 | * inode (for close-open consistency) ourselves. |
| 117 | */ |
| 118 | __nfs_revalidate_inode(NFS_SERVER(inode), inode); |
| 119 | } |
| 120 | out: |
| 121 | put_rpccred(cred); |
| 122 | return res; |
| 123 | } |
| 124 | |
| 125 | static int |
| 126 | nfs_closedir(struct inode *inode, struct file *filp) |
| 127 | { |
| 128 | put_nfs_open_dir_context(filp->private_data); |
| 129 | return 0; |
| 130 | } |
| 131 | |
| 132 | struct nfs_cache_array_entry { |
| 133 | u64 cookie; |
| 134 | u64 ino; |
| 135 | struct qstr string; |
| 136 | unsigned char d_type; |
| 137 | }; |
| 138 | |
| 139 | struct nfs_cache_array { |
| 140 | int size; |
| 141 | int eof_index; |
| 142 | u64 last_cookie; |
| 143 | struct nfs_cache_array_entry array[0]; |
| 144 | }; |
| 145 | |
| 146 | typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int); |
| 147 | typedef struct { |
| 148 | struct file *file; |
| 149 | struct page *page; |
| 150 | unsigned long page_index; |
| 151 | u64 *dir_cookie; |
| 152 | u64 last_cookie; |
| 153 | loff_t current_index; |
| 154 | decode_dirent_t decode; |
| 155 | |
| 156 | unsigned long timestamp; |
| 157 | unsigned long gencount; |
| 158 | unsigned int cache_entry_index; |
| 159 | unsigned int plus:1; |
| 160 | unsigned int eof:1; |
| 161 | } nfs_readdir_descriptor_t; |
| 162 | |
| 163 | /* |
| 164 | * The caller is responsible for calling nfs_readdir_release_array(page) |
| 165 | */ |
| 166 | static |
| 167 | struct nfs_cache_array *nfs_readdir_get_array(struct page *page) |
| 168 | { |
| 169 | void *ptr; |
| 170 | if (page == NULL) |
| 171 | return ERR_PTR(-EIO); |
| 172 | ptr = kmap(page); |
| 173 | if (ptr == NULL) |
| 174 | return ERR_PTR(-ENOMEM); |
| 175 | return ptr; |
| 176 | } |
| 177 | |
| 178 | static |
| 179 | void nfs_readdir_release_array(struct page *page) |
| 180 | { |
| 181 | kunmap(page); |
| 182 | } |
| 183 | |
| 184 | /* |
| 185 | * we are freeing strings created by nfs_add_to_readdir_array() |
| 186 | */ |
| 187 | static |
| 188 | void nfs_readdir_clear_array(struct page *page) |
| 189 | { |
| 190 | struct nfs_cache_array *array; |
| 191 | int i; |
| 192 | |
| 193 | array = kmap_atomic(page); |
| 194 | for (i = 0; i < array->size; i++) |
| 195 | kfree(array->array[i].string.name); |
| 196 | kunmap_atomic(array); |
| 197 | } |
| 198 | |
| 199 | /* |
| 200 | * the caller is responsible for freeing qstr.name |
| 201 | * when called by nfs_readdir_add_to_array, the strings will be freed in |
| 202 | * nfs_clear_readdir_array() |
| 203 | */ |
| 204 | static |
| 205 | int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len) |
| 206 | { |
| 207 | string->len = len; |
| 208 | string->name = kmemdup(name, len, GFP_KERNEL); |
| 209 | if (string->name == NULL) |
| 210 | return -ENOMEM; |
| 211 | /* |
| 212 | * Avoid a kmemleak false positive. The pointer to the name is stored |
| 213 | * in a page cache page which kmemleak does not scan. |
| 214 | */ |
| 215 | kmemleak_not_leak(string->name); |
| 216 | string->hash = full_name_hash(name, len); |
| 217 | return 0; |
| 218 | } |
| 219 | |
| 220 | static |
| 221 | int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page) |
| 222 | { |
| 223 | struct nfs_cache_array *array = nfs_readdir_get_array(page); |
| 224 | struct nfs_cache_array_entry *cache_entry; |
| 225 | int ret; |
| 226 | |
| 227 | if (IS_ERR(array)) |
| 228 | return PTR_ERR(array); |
| 229 | |
| 230 | cache_entry = &array->array[array->size]; |
| 231 | |
| 232 | /* Check that this entry lies within the page bounds */ |
| 233 | ret = -ENOSPC; |
| 234 | if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE) |
| 235 | goto out; |
| 236 | |
| 237 | cache_entry->cookie = entry->prev_cookie; |
| 238 | cache_entry->ino = entry->ino; |
| 239 | cache_entry->d_type = entry->d_type; |
| 240 | ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len); |
| 241 | if (ret) |
| 242 | goto out; |
| 243 | array->last_cookie = entry->cookie; |
| 244 | array->size++; |
| 245 | if (entry->eof != 0) |
| 246 | array->eof_index = array->size; |
| 247 | out: |
| 248 | nfs_readdir_release_array(page); |
| 249 | return ret; |
| 250 | } |
| 251 | |
| 252 | static |
| 253 | int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) |
| 254 | { |
| 255 | loff_t diff = desc->file->f_pos - desc->current_index; |
| 256 | unsigned int index; |
| 257 | |
| 258 | if (diff < 0) |
| 259 | goto out_eof; |
| 260 | if (diff >= array->size) { |
| 261 | if (array->eof_index >= 0) |
| 262 | goto out_eof; |
| 263 | return -EAGAIN; |
| 264 | } |
| 265 | |
| 266 | index = (unsigned int)diff; |
| 267 | *desc->dir_cookie = array->array[index].cookie; |
| 268 | desc->cache_entry_index = index; |
| 269 | return 0; |
| 270 | out_eof: |
| 271 | desc->eof = 1; |
| 272 | return -EBADCOOKIE; |
| 273 | } |
| 274 | |
| 275 | static |
| 276 | int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) |
| 277 | { |
| 278 | int i; |
| 279 | loff_t new_pos; |
| 280 | int status = -EAGAIN; |
| 281 | |
| 282 | for (i = 0; i < array->size; i++) { |
| 283 | if (array->array[i].cookie == *desc->dir_cookie) { |
| 284 | struct nfs_inode *nfsi = NFS_I(file_inode(desc->file)); |
| 285 | struct nfs_open_dir_context *ctx = desc->file->private_data; |
| 286 | |
| 287 | new_pos = desc->current_index + i; |
| 288 | if (ctx->attr_gencount != nfsi->attr_gencount |
| 289 | || (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))) { |
| 290 | ctx->duped = 0; |
| 291 | ctx->attr_gencount = nfsi->attr_gencount; |
| 292 | } else if (new_pos < desc->file->f_pos) { |
| 293 | if (ctx->duped > 0 |
| 294 | && ctx->dup_cookie == *desc->dir_cookie) { |
| 295 | if (printk_ratelimit()) { |
| 296 | pr_notice("NFS: directory %s/%s contains a readdir loop." |
| 297 | "Please contact your server vendor. " |
| 298 | "The file: %s has duplicate cookie %llu\n", |
| 299 | desc->file->f_dentry->d_parent->d_name.name, |
| 300 | desc->file->f_dentry->d_name.name, |
| 301 | array->array[i].string.name, |
| 302 | *desc->dir_cookie); |
| 303 | } |
| 304 | status = -ELOOP; |
| 305 | goto out; |
| 306 | } |
| 307 | ctx->dup_cookie = *desc->dir_cookie; |
| 308 | ctx->duped = -1; |
| 309 | } |
| 310 | desc->file->f_pos = new_pos; |
| 311 | desc->cache_entry_index = i; |
| 312 | return 0; |
| 313 | } |
| 314 | } |
| 315 | if (array->eof_index >= 0) { |
| 316 | status = -EBADCOOKIE; |
| 317 | if (*desc->dir_cookie == array->last_cookie) |
| 318 | desc->eof = 1; |
| 319 | } |
| 320 | out: |
| 321 | return status; |
| 322 | } |
| 323 | |
| 324 | static |
| 325 | int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc) |
| 326 | { |
| 327 | struct nfs_cache_array *array; |
| 328 | int status; |
| 329 | |
| 330 | array = nfs_readdir_get_array(desc->page); |
| 331 | if (IS_ERR(array)) { |
| 332 | status = PTR_ERR(array); |
| 333 | goto out; |
| 334 | } |
| 335 | |
| 336 | if (*desc->dir_cookie == 0) |
| 337 | status = nfs_readdir_search_for_pos(array, desc); |
| 338 | else |
| 339 | status = nfs_readdir_search_for_cookie(array, desc); |
| 340 | |
| 341 | if (status == -EAGAIN) { |
| 342 | desc->last_cookie = array->last_cookie; |
| 343 | desc->current_index += array->size; |
| 344 | desc->page_index++; |
| 345 | } |
| 346 | nfs_readdir_release_array(desc->page); |
| 347 | out: |
| 348 | return status; |
| 349 | } |
| 350 | |
| 351 | /* Fill a page with xdr information before transferring to the cache page */ |
| 352 | static |
| 353 | int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc, |
| 354 | struct nfs_entry *entry, struct file *file, struct inode *inode) |
| 355 | { |
| 356 | struct nfs_open_dir_context *ctx = file->private_data; |
| 357 | struct rpc_cred *cred = ctx->cred; |
| 358 | unsigned long timestamp, gencount; |
| 359 | int error; |
| 360 | |
| 361 | again: |
| 362 | timestamp = jiffies; |
| 363 | gencount = nfs_inc_attr_generation_counter(); |
| 364 | error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages, |
| 365 | NFS_SERVER(inode)->dtsize, desc->plus); |
| 366 | if (error < 0) { |
| 367 | /* We requested READDIRPLUS, but the server doesn't grok it */ |
| 368 | if (error == -ENOTSUPP && desc->plus) { |
| 369 | NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; |
| 370 | clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); |
| 371 | desc->plus = 0; |
| 372 | goto again; |
| 373 | } |
| 374 | goto error; |
| 375 | } |
| 376 | desc->timestamp = timestamp; |
| 377 | desc->gencount = gencount; |
| 378 | error: |
| 379 | return error; |
| 380 | } |
| 381 | |
| 382 | static int xdr_decode(nfs_readdir_descriptor_t *desc, |
| 383 | struct nfs_entry *entry, struct xdr_stream *xdr) |
| 384 | { |
| 385 | int error; |
| 386 | |
| 387 | error = desc->decode(xdr, entry, desc->plus); |
| 388 | if (error) |
| 389 | return error; |
| 390 | entry->fattr->time_start = desc->timestamp; |
| 391 | entry->fattr->gencount = desc->gencount; |
| 392 | return 0; |
| 393 | } |
| 394 | |
| 395 | static |
| 396 | int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry) |
| 397 | { |
| 398 | if (dentry->d_inode == NULL) |
| 399 | goto different; |
| 400 | if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0) |
| 401 | goto different; |
| 402 | return 1; |
| 403 | different: |
| 404 | return 0; |
| 405 | } |
| 406 | |
| 407 | static |
| 408 | bool nfs_use_readdirplus(struct inode *dir, struct file *filp) |
| 409 | { |
| 410 | if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS)) |
| 411 | return false; |
| 412 | if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags)) |
| 413 | return true; |
| 414 | if (filp->f_pos == 0) |
| 415 | return true; |
| 416 | return false; |
| 417 | } |
| 418 | |
| 419 | /* |
| 420 | * This function is called by the lookup code to request the use of |
| 421 | * readdirplus to accelerate any future lookups in the same |
| 422 | * directory. |
| 423 | */ |
| 424 | static |
| 425 | void nfs_advise_use_readdirplus(struct inode *dir) |
| 426 | { |
| 427 | set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags); |
| 428 | } |
| 429 | |
| 430 | static |
| 431 | void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry) |
| 432 | { |
| 433 | struct qstr filename = QSTR_INIT(entry->name, entry->len); |
| 434 | struct dentry *dentry; |
| 435 | struct dentry *alias; |
| 436 | struct inode *dir = parent->d_inode; |
| 437 | struct inode *inode; |
| 438 | |
| 439 | if (filename.name[0] == '.') { |
| 440 | if (filename.len == 1) |
| 441 | return; |
| 442 | if (filename.len == 2 && filename.name[1] == '.') |
| 443 | return; |
| 444 | } |
| 445 | filename.hash = full_name_hash(filename.name, filename.len); |
| 446 | |
| 447 | dentry = d_lookup(parent, &filename); |
| 448 | if (dentry != NULL) { |
| 449 | if (nfs_same_file(dentry, entry)) { |
| 450 | nfs_refresh_inode(dentry->d_inode, entry->fattr); |
| 451 | goto out; |
| 452 | } else { |
| 453 | if (d_invalidate(dentry) != 0) |
| 454 | goto out; |
| 455 | dput(dentry); |
| 456 | } |
| 457 | } |
| 458 | |
| 459 | dentry = d_alloc(parent, &filename); |
| 460 | if (dentry == NULL) |
| 461 | return; |
| 462 | |
| 463 | inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label); |
| 464 | if (IS_ERR(inode)) |
| 465 | goto out; |
| 466 | |
| 467 | alias = d_materialise_unique(dentry, inode); |
| 468 | if (IS_ERR(alias)) |
| 469 | goto out; |
| 470 | else if (alias) { |
| 471 | nfs_set_verifier(alias, nfs_save_change_attribute(dir)); |
| 472 | dput(alias); |
| 473 | } else |
| 474 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
| 475 | |
| 476 | out: |
| 477 | dput(dentry); |
| 478 | } |
| 479 | |
| 480 | /* Perform conversion from xdr to cache array */ |
| 481 | static |
| 482 | int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, |
| 483 | struct page **xdr_pages, struct page *page, unsigned int buflen) |
| 484 | { |
| 485 | struct xdr_stream stream; |
| 486 | struct xdr_buf buf; |
| 487 | struct page *scratch; |
| 488 | struct nfs_cache_array *array; |
| 489 | unsigned int count = 0; |
| 490 | int status; |
| 491 | |
| 492 | scratch = alloc_page(GFP_KERNEL); |
| 493 | if (scratch == NULL) |
| 494 | return -ENOMEM; |
| 495 | |
| 496 | xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen); |
| 497 | xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE); |
| 498 | |
| 499 | do { |
| 500 | status = xdr_decode(desc, entry, &stream); |
| 501 | if (status != 0) { |
| 502 | if (status == -EAGAIN) |
| 503 | status = 0; |
| 504 | break; |
| 505 | } |
| 506 | |
| 507 | count++; |
| 508 | |
| 509 | if (desc->plus != 0) |
| 510 | nfs_prime_dcache(desc->file->f_path.dentry, entry); |
| 511 | |
| 512 | status = nfs_readdir_add_to_array(entry, page); |
| 513 | if (status != 0) |
| 514 | break; |
| 515 | } while (!entry->eof); |
| 516 | |
| 517 | if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) { |
| 518 | array = nfs_readdir_get_array(page); |
| 519 | if (!IS_ERR(array)) { |
| 520 | array->eof_index = array->size; |
| 521 | status = 0; |
| 522 | nfs_readdir_release_array(page); |
| 523 | } else |
| 524 | status = PTR_ERR(array); |
| 525 | } |
| 526 | |
| 527 | put_page(scratch); |
| 528 | return status; |
| 529 | } |
| 530 | |
| 531 | static |
| 532 | void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages) |
| 533 | { |
| 534 | unsigned int i; |
| 535 | for (i = 0; i < npages; i++) |
| 536 | put_page(pages[i]); |
| 537 | } |
| 538 | |
| 539 | static |
| 540 | void nfs_readdir_free_large_page(void *ptr, struct page **pages, |
| 541 | unsigned int npages) |
| 542 | { |
| 543 | nfs_readdir_free_pagearray(pages, npages); |
| 544 | } |
| 545 | |
| 546 | /* |
| 547 | * nfs_readdir_large_page will allocate pages that must be freed with a call |
| 548 | * to nfs_readdir_free_large_page |
| 549 | */ |
| 550 | static |
| 551 | int nfs_readdir_large_page(struct page **pages, unsigned int npages) |
| 552 | { |
| 553 | unsigned int i; |
| 554 | |
| 555 | for (i = 0; i < npages; i++) { |
| 556 | struct page *page = alloc_page(GFP_KERNEL); |
| 557 | if (page == NULL) |
| 558 | goto out_freepages; |
| 559 | pages[i] = page; |
| 560 | } |
| 561 | return 0; |
| 562 | |
| 563 | out_freepages: |
| 564 | nfs_readdir_free_pagearray(pages, i); |
| 565 | return -ENOMEM; |
| 566 | } |
| 567 | |
| 568 | static |
| 569 | int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode) |
| 570 | { |
| 571 | struct page *pages[NFS_MAX_READDIR_PAGES]; |
| 572 | void *pages_ptr = NULL; |
| 573 | struct nfs_entry entry; |
| 574 | struct file *file = desc->file; |
| 575 | struct nfs_cache_array *array; |
| 576 | int status = -ENOMEM; |
| 577 | unsigned int array_size = ARRAY_SIZE(pages); |
| 578 | |
| 579 | entry.prev_cookie = 0; |
| 580 | entry.cookie = desc->last_cookie; |
| 581 | entry.eof = 0; |
| 582 | entry.fh = nfs_alloc_fhandle(); |
| 583 | entry.fattr = nfs_alloc_fattr(); |
| 584 | entry.server = NFS_SERVER(inode); |
| 585 | if (entry.fh == NULL || entry.fattr == NULL) |
| 586 | goto out; |
| 587 | |
| 588 | entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); |
| 589 | if (IS_ERR(entry.label)) { |
| 590 | status = PTR_ERR(entry.label); |
| 591 | goto out; |
| 592 | } |
| 593 | |
| 594 | array = nfs_readdir_get_array(page); |
| 595 | if (IS_ERR(array)) { |
| 596 | status = PTR_ERR(array); |
| 597 | goto out_label_free; |
| 598 | } |
| 599 | memset(array, 0, sizeof(struct nfs_cache_array)); |
| 600 | array->eof_index = -1; |
| 601 | |
| 602 | status = nfs_readdir_large_page(pages, array_size); |
| 603 | if (status < 0) |
| 604 | goto out_release_array; |
| 605 | do { |
| 606 | unsigned int pglen; |
| 607 | status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode); |
| 608 | |
| 609 | if (status < 0) |
| 610 | break; |
| 611 | pglen = status; |
| 612 | status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen); |
| 613 | if (status < 0) { |
| 614 | if (status == -ENOSPC) |
| 615 | status = 0; |
| 616 | break; |
| 617 | } |
| 618 | } while (array->eof_index < 0); |
| 619 | |
| 620 | nfs_readdir_free_large_page(pages_ptr, pages, array_size); |
| 621 | out_release_array: |
| 622 | nfs_readdir_release_array(page); |
| 623 | out_label_free: |
| 624 | nfs4_label_free(entry.label); |
| 625 | out: |
| 626 | nfs_free_fattr(entry.fattr); |
| 627 | nfs_free_fhandle(entry.fh); |
| 628 | return status; |
| 629 | } |
| 630 | |
| 631 | /* |
| 632 | * Now we cache directories properly, by converting xdr information |
| 633 | * to an array that can be used for lookups later. This results in |
| 634 | * fewer cache pages, since we can store more information on each page. |
| 635 | * We only need to convert from xdr once so future lookups are much simpler |
| 636 | */ |
| 637 | static |
| 638 | int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page) |
| 639 | { |
| 640 | struct inode *inode = file_inode(desc->file); |
| 641 | int ret; |
| 642 | |
| 643 | ret = nfs_readdir_xdr_to_array(desc, page, inode); |
| 644 | if (ret < 0) |
| 645 | goto error; |
| 646 | SetPageUptodate(page); |
| 647 | |
| 648 | if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) { |
| 649 | /* Should never happen */ |
| 650 | nfs_zap_mapping(inode, inode->i_mapping); |
| 651 | } |
| 652 | unlock_page(page); |
| 653 | return 0; |
| 654 | error: |
| 655 | unlock_page(page); |
| 656 | return ret; |
| 657 | } |
| 658 | |
| 659 | static |
| 660 | void cache_page_release(nfs_readdir_descriptor_t *desc) |
| 661 | { |
| 662 | if (!desc->page->mapping) |
| 663 | nfs_readdir_clear_array(desc->page); |
| 664 | page_cache_release(desc->page); |
| 665 | desc->page = NULL; |
| 666 | } |
| 667 | |
| 668 | static |
| 669 | struct page *get_cache_page(nfs_readdir_descriptor_t *desc) |
| 670 | { |
| 671 | return read_cache_page(file_inode(desc->file)->i_mapping, |
| 672 | desc->page_index, (filler_t *)nfs_readdir_filler, desc); |
| 673 | } |
| 674 | |
| 675 | /* |
| 676 | * Returns 0 if desc->dir_cookie was found on page desc->page_index |
| 677 | */ |
| 678 | static |
| 679 | int find_cache_page(nfs_readdir_descriptor_t *desc) |
| 680 | { |
| 681 | int res; |
| 682 | |
| 683 | desc->page = get_cache_page(desc); |
| 684 | if (IS_ERR(desc->page)) |
| 685 | return PTR_ERR(desc->page); |
| 686 | |
| 687 | res = nfs_readdir_search_array(desc); |
| 688 | if (res != 0) |
| 689 | cache_page_release(desc); |
| 690 | return res; |
| 691 | } |
| 692 | |
| 693 | /* Search for desc->dir_cookie from the beginning of the page cache */ |
| 694 | static inline |
| 695 | int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) |
| 696 | { |
| 697 | int res; |
| 698 | |
| 699 | if (desc->page_index == 0) { |
| 700 | desc->current_index = 0; |
| 701 | desc->last_cookie = 0; |
| 702 | } |
| 703 | do { |
| 704 | res = find_cache_page(desc); |
| 705 | } while (res == -EAGAIN); |
| 706 | return res; |
| 707 | } |
| 708 | |
| 709 | /* |
| 710 | * Once we've found the start of the dirent within a page: fill 'er up... |
| 711 | */ |
| 712 | static |
| 713 | int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent, |
| 714 | filldir_t filldir) |
| 715 | { |
| 716 | struct file *file = desc->file; |
| 717 | int i = 0; |
| 718 | int res = 0; |
| 719 | struct nfs_cache_array *array = NULL; |
| 720 | struct nfs_open_dir_context *ctx = file->private_data; |
| 721 | |
| 722 | array = nfs_readdir_get_array(desc->page); |
| 723 | if (IS_ERR(array)) { |
| 724 | res = PTR_ERR(array); |
| 725 | goto out; |
| 726 | } |
| 727 | |
| 728 | for (i = desc->cache_entry_index; i < array->size; i++) { |
| 729 | struct nfs_cache_array_entry *ent; |
| 730 | |
| 731 | ent = &array->array[i]; |
| 732 | if (filldir(dirent, ent->string.name, ent->string.len, |
| 733 | file->f_pos, nfs_compat_user_ino64(ent->ino), |
| 734 | ent->d_type) < 0) { |
| 735 | desc->eof = 1; |
| 736 | break; |
| 737 | } |
| 738 | file->f_pos++; |
| 739 | if (i < (array->size-1)) |
| 740 | *desc->dir_cookie = array->array[i+1].cookie; |
| 741 | else |
| 742 | *desc->dir_cookie = array->last_cookie; |
| 743 | if (ctx->duped != 0) |
| 744 | ctx->duped = 1; |
| 745 | } |
| 746 | if (array->eof_index >= 0) |
| 747 | desc->eof = 1; |
| 748 | |
| 749 | nfs_readdir_release_array(desc->page); |
| 750 | out: |
| 751 | cache_page_release(desc); |
| 752 | dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", |
| 753 | (unsigned long long)*desc->dir_cookie, res); |
| 754 | return res; |
| 755 | } |
| 756 | |
| 757 | /* |
| 758 | * If we cannot find a cookie in our cache, we suspect that this is |
| 759 | * because it points to a deleted file, so we ask the server to return |
| 760 | * whatever it thinks is the next entry. We then feed this to filldir. |
| 761 | * If all goes well, we should then be able to find our way round the |
| 762 | * cache on the next call to readdir_search_pagecache(); |
| 763 | * |
| 764 | * NOTE: we cannot add the anonymous page to the pagecache because |
| 765 | * the data it contains might not be page aligned. Besides, |
| 766 | * we should already have a complete representation of the |
| 767 | * directory in the page cache by the time we get here. |
| 768 | */ |
| 769 | static inline |
| 770 | int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent, |
| 771 | filldir_t filldir) |
| 772 | { |
| 773 | struct page *page = NULL; |
| 774 | int status; |
| 775 | struct inode *inode = file_inode(desc->file); |
| 776 | struct nfs_open_dir_context *ctx = desc->file->private_data; |
| 777 | |
| 778 | dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", |
| 779 | (unsigned long long)*desc->dir_cookie); |
| 780 | |
| 781 | page = alloc_page(GFP_HIGHUSER); |
| 782 | if (!page) { |
| 783 | status = -ENOMEM; |
| 784 | goto out; |
| 785 | } |
| 786 | |
| 787 | desc->page_index = 0; |
| 788 | desc->last_cookie = *desc->dir_cookie; |
| 789 | desc->page = page; |
| 790 | ctx->duped = 0; |
| 791 | |
| 792 | status = nfs_readdir_xdr_to_array(desc, page, inode); |
| 793 | if (status < 0) |
| 794 | goto out_release; |
| 795 | |
| 796 | status = nfs_do_filldir(desc, dirent, filldir); |
| 797 | |
| 798 | out: |
| 799 | dfprintk(DIRCACHE, "NFS: %s: returns %d\n", |
| 800 | __func__, status); |
| 801 | return status; |
| 802 | out_release: |
| 803 | cache_page_release(desc); |
| 804 | goto out; |
| 805 | } |
| 806 | |
| 807 | /* The file offset position represents the dirent entry number. A |
| 808 | last cookie cache takes care of the common case of reading the |
| 809 | whole directory. |
| 810 | */ |
| 811 | static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir) |
| 812 | { |
| 813 | struct dentry *dentry = filp->f_path.dentry; |
| 814 | struct inode *inode = dentry->d_inode; |
| 815 | nfs_readdir_descriptor_t my_desc, |
| 816 | *desc = &my_desc; |
| 817 | struct nfs_open_dir_context *dir_ctx = filp->private_data; |
| 818 | int res; |
| 819 | |
| 820 | dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n", |
| 821 | dentry->d_parent->d_name.name, dentry->d_name.name, |
| 822 | (long long)filp->f_pos); |
| 823 | nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); |
| 824 | |
| 825 | /* |
| 826 | * filp->f_pos points to the dirent entry number. |
| 827 | * *desc->dir_cookie has the cookie for the next entry. We have |
| 828 | * to either find the entry with the appropriate number or |
| 829 | * revalidate the cookie. |
| 830 | */ |
| 831 | memset(desc, 0, sizeof(*desc)); |
| 832 | |
| 833 | desc->file = filp; |
| 834 | desc->dir_cookie = &dir_ctx->dir_cookie; |
| 835 | desc->decode = NFS_PROTO(inode)->decode_dirent; |
| 836 | desc->plus = nfs_use_readdirplus(inode, filp) ? 1 : 0; |
| 837 | |
| 838 | nfs_block_sillyrename(dentry); |
| 839 | res = nfs_revalidate_mapping(inode, filp->f_mapping); |
| 840 | if (res < 0) |
| 841 | goto out; |
| 842 | |
| 843 | do { |
| 844 | res = readdir_search_pagecache(desc); |
| 845 | |
| 846 | if (res == -EBADCOOKIE) { |
| 847 | res = 0; |
| 848 | /* This means either end of directory */ |
| 849 | if (*desc->dir_cookie && desc->eof == 0) { |
| 850 | /* Or that the server has 'lost' a cookie */ |
| 851 | res = uncached_readdir(desc, dirent, filldir); |
| 852 | if (res == 0) |
| 853 | continue; |
| 854 | } |
| 855 | break; |
| 856 | } |
| 857 | if (res == -ETOOSMALL && desc->plus) { |
| 858 | clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); |
| 859 | nfs_zap_caches(inode); |
| 860 | desc->page_index = 0; |
| 861 | desc->plus = 0; |
| 862 | desc->eof = 0; |
| 863 | continue; |
| 864 | } |
| 865 | if (res < 0) |
| 866 | break; |
| 867 | |
| 868 | res = nfs_do_filldir(desc, dirent, filldir); |
| 869 | if (res < 0) |
| 870 | break; |
| 871 | } while (!desc->eof); |
| 872 | out: |
| 873 | nfs_unblock_sillyrename(dentry); |
| 874 | if (res > 0) |
| 875 | res = 0; |
| 876 | dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n", |
| 877 | dentry->d_parent->d_name.name, dentry->d_name.name, |
| 878 | res); |
| 879 | return res; |
| 880 | } |
| 881 | |
| 882 | static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence) |
| 883 | { |
| 884 | struct dentry *dentry = filp->f_path.dentry; |
| 885 | struct inode *inode = dentry->d_inode; |
| 886 | struct nfs_open_dir_context *dir_ctx = filp->private_data; |
| 887 | |
| 888 | dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n", |
| 889 | dentry->d_parent->d_name.name, |
| 890 | dentry->d_name.name, |
| 891 | offset, whence); |
| 892 | |
| 893 | mutex_lock(&inode->i_mutex); |
| 894 | switch (whence) { |
| 895 | case 1: |
| 896 | offset += filp->f_pos; |
| 897 | case 0: |
| 898 | if (offset >= 0) |
| 899 | break; |
| 900 | default: |
| 901 | offset = -EINVAL; |
| 902 | goto out; |
| 903 | } |
| 904 | if (offset != filp->f_pos) { |
| 905 | filp->f_pos = offset; |
| 906 | dir_ctx->dir_cookie = 0; |
| 907 | dir_ctx->duped = 0; |
| 908 | } |
| 909 | out: |
| 910 | mutex_unlock(&inode->i_mutex); |
| 911 | return offset; |
| 912 | } |
| 913 | |
| 914 | /* |
| 915 | * All directory operations under NFS are synchronous, so fsync() |
| 916 | * is a dummy operation. |
| 917 | */ |
| 918 | static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end, |
| 919 | int datasync) |
| 920 | { |
| 921 | struct dentry *dentry = filp->f_path.dentry; |
| 922 | struct inode *inode = dentry->d_inode; |
| 923 | |
| 924 | dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n", |
| 925 | dentry->d_parent->d_name.name, dentry->d_name.name, |
| 926 | datasync); |
| 927 | |
| 928 | mutex_lock(&inode->i_mutex); |
| 929 | nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC); |
| 930 | mutex_unlock(&inode->i_mutex); |
| 931 | return 0; |
| 932 | } |
| 933 | |
| 934 | /** |
| 935 | * nfs_force_lookup_revalidate - Mark the directory as having changed |
| 936 | * @dir - pointer to directory inode |
| 937 | * |
| 938 | * This forces the revalidation code in nfs_lookup_revalidate() to do a |
| 939 | * full lookup on all child dentries of 'dir' whenever a change occurs |
| 940 | * on the server that might have invalidated our dcache. |
| 941 | * |
| 942 | * The caller should be holding dir->i_lock |
| 943 | */ |
| 944 | void nfs_force_lookup_revalidate(struct inode *dir) |
| 945 | { |
| 946 | NFS_I(dir)->cache_change_attribute++; |
| 947 | } |
| 948 | EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate); |
| 949 | |
| 950 | /* |
| 951 | * A check for whether or not the parent directory has changed. |
| 952 | * In the case it has, we assume that the dentries are untrustworthy |
| 953 | * and may need to be looked up again. |
| 954 | */ |
| 955 | static int nfs_check_verifier(struct inode *dir, struct dentry *dentry) |
| 956 | { |
| 957 | if (IS_ROOT(dentry)) |
| 958 | return 1; |
| 959 | if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) |
| 960 | return 0; |
| 961 | if (!nfs_verify_change_attribute(dir, dentry->d_time)) |
| 962 | return 0; |
| 963 | /* Revalidate nfsi->cache_change_attribute before we declare a match */ |
| 964 | if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) |
| 965 | return 0; |
| 966 | if (!nfs_verify_change_attribute(dir, dentry->d_time)) |
| 967 | return 0; |
| 968 | return 1; |
| 969 | } |
| 970 | |
| 971 | /* |
| 972 | * Use intent information to check whether or not we're going to do |
| 973 | * an O_EXCL create using this path component. |
| 974 | */ |
| 975 | static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags) |
| 976 | { |
| 977 | if (NFS_PROTO(dir)->version == 2) |
| 978 | return 0; |
| 979 | return flags & LOOKUP_EXCL; |
| 980 | } |
| 981 | |
| 982 | /* |
| 983 | * Inode and filehandle revalidation for lookups. |
| 984 | * |
| 985 | * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, |
| 986 | * or if the intent information indicates that we're about to open this |
| 987 | * particular file and the "nocto" mount flag is not set. |
| 988 | * |
| 989 | */ |
| 990 | static |
| 991 | int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags) |
| 992 | { |
| 993 | struct nfs_server *server = NFS_SERVER(inode); |
| 994 | int ret; |
| 995 | |
| 996 | if (IS_AUTOMOUNT(inode)) |
| 997 | return 0; |
| 998 | /* VFS wants an on-the-wire revalidation */ |
| 999 | if (flags & LOOKUP_REVAL) |
| 1000 | goto out_force; |
| 1001 | /* This is an open(2) */ |
| 1002 | if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) && |
| 1003 | (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))) |
| 1004 | goto out_force; |
| 1005 | out: |
| 1006 | return (inode->i_nlink == 0) ? -ENOENT : 0; |
| 1007 | out_force: |
| 1008 | ret = __nfs_revalidate_inode(server, inode); |
| 1009 | if (ret != 0) |
| 1010 | return ret; |
| 1011 | goto out; |
| 1012 | } |
| 1013 | |
| 1014 | /* |
| 1015 | * We judge how long we want to trust negative |
| 1016 | * dentries by looking at the parent inode mtime. |
| 1017 | * |
| 1018 | * If parent mtime has changed, we revalidate, else we wait for a |
| 1019 | * period corresponding to the parent's attribute cache timeout value. |
| 1020 | */ |
| 1021 | static inline |
| 1022 | int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, |
| 1023 | unsigned int flags) |
| 1024 | { |
| 1025 | /* Don't revalidate a negative dentry if we're creating a new file */ |
| 1026 | if (flags & LOOKUP_CREATE) |
| 1027 | return 0; |
| 1028 | if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) |
| 1029 | return 1; |
| 1030 | return !nfs_check_verifier(dir, dentry); |
| 1031 | } |
| 1032 | |
| 1033 | /* |
| 1034 | * This is called every time the dcache has a lookup hit, |
| 1035 | * and we should check whether we can really trust that |
| 1036 | * lookup. |
| 1037 | * |
| 1038 | * NOTE! The hit can be a negative hit too, don't assume |
| 1039 | * we have an inode! |
| 1040 | * |
| 1041 | * If the parent directory is seen to have changed, we throw out the |
| 1042 | * cached dentry and do a new lookup. |
| 1043 | */ |
| 1044 | static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags) |
| 1045 | { |
| 1046 | struct inode *dir; |
| 1047 | struct inode *inode; |
| 1048 | struct dentry *parent; |
| 1049 | struct nfs_fh *fhandle = NULL; |
| 1050 | struct nfs_fattr *fattr = NULL; |
| 1051 | struct nfs4_label *label = NULL; |
| 1052 | int error; |
| 1053 | |
| 1054 | if (flags & LOOKUP_RCU) |
| 1055 | return -ECHILD; |
| 1056 | |
| 1057 | parent = dget_parent(dentry); |
| 1058 | dir = parent->d_inode; |
| 1059 | nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); |
| 1060 | inode = dentry->d_inode; |
| 1061 | |
| 1062 | if (!inode) { |
| 1063 | if (nfs_neg_need_reval(dir, dentry, flags)) |
| 1064 | goto out_bad; |
| 1065 | goto out_valid_noent; |
| 1066 | } |
| 1067 | |
| 1068 | if (is_bad_inode(inode)) { |
| 1069 | dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n", |
| 1070 | __func__, dentry->d_parent->d_name.name, |
| 1071 | dentry->d_name.name); |
| 1072 | goto out_bad; |
| 1073 | } |
| 1074 | |
| 1075 | if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ)) |
| 1076 | goto out_set_verifier; |
| 1077 | |
| 1078 | /* Force a full look up iff the parent directory has changed */ |
| 1079 | if (!nfs_is_exclusive_create(dir, flags) && nfs_check_verifier(dir, dentry)) { |
| 1080 | if (nfs_lookup_verify_inode(inode, flags)) |
| 1081 | goto out_zap_parent; |
| 1082 | goto out_valid; |
| 1083 | } |
| 1084 | |
| 1085 | if (NFS_STALE(inode)) |
| 1086 | goto out_bad; |
| 1087 | |
| 1088 | error = -ENOMEM; |
| 1089 | fhandle = nfs_alloc_fhandle(); |
| 1090 | fattr = nfs_alloc_fattr(); |
| 1091 | if (fhandle == NULL || fattr == NULL) |
| 1092 | goto out_error; |
| 1093 | |
| 1094 | label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); |
| 1095 | if (IS_ERR(label)) |
| 1096 | goto out_error; |
| 1097 | |
| 1098 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label); |
| 1099 | if (error) |
| 1100 | goto out_bad; |
| 1101 | if (nfs_compare_fh(NFS_FH(inode), fhandle)) |
| 1102 | goto out_bad; |
| 1103 | if ((error = nfs_refresh_inode(inode, fattr)) != 0) |
| 1104 | goto out_bad; |
| 1105 | |
| 1106 | nfs_free_fattr(fattr); |
| 1107 | nfs_free_fhandle(fhandle); |
| 1108 | nfs4_label_free(label); |
| 1109 | |
| 1110 | out_set_verifier: |
| 1111 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
| 1112 | out_valid: |
| 1113 | /* Success: notify readdir to use READDIRPLUS */ |
| 1114 | nfs_advise_use_readdirplus(dir); |
| 1115 | out_valid_noent: |
| 1116 | dput(parent); |
| 1117 | dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n", |
| 1118 | __func__, dentry->d_parent->d_name.name, |
| 1119 | dentry->d_name.name); |
| 1120 | return 1; |
| 1121 | out_zap_parent: |
| 1122 | nfs_zap_caches(dir); |
| 1123 | out_bad: |
| 1124 | nfs_free_fattr(fattr); |
| 1125 | nfs_free_fhandle(fhandle); |
| 1126 | nfs4_label_free(label); |
| 1127 | nfs_mark_for_revalidate(dir); |
| 1128 | if (inode && S_ISDIR(inode->i_mode)) { |
| 1129 | /* Purge readdir caches. */ |
| 1130 | nfs_zap_caches(inode); |
| 1131 | /* If we have submounts, don't unhash ! */ |
| 1132 | if (have_submounts(dentry)) |
| 1133 | goto out_valid; |
| 1134 | if (dentry->d_flags & DCACHE_DISCONNECTED) |
| 1135 | goto out_valid; |
| 1136 | shrink_dcache_parent(dentry); |
| 1137 | } |
| 1138 | d_drop(dentry); |
| 1139 | dput(parent); |
| 1140 | dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n", |
| 1141 | __func__, dentry->d_parent->d_name.name, |
| 1142 | dentry->d_name.name); |
| 1143 | return 0; |
| 1144 | out_error: |
| 1145 | nfs_free_fattr(fattr); |
| 1146 | nfs_free_fhandle(fhandle); |
| 1147 | nfs4_label_free(label); |
| 1148 | dput(parent); |
| 1149 | dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n", |
| 1150 | __func__, dentry->d_parent->d_name.name, |
| 1151 | dentry->d_name.name, error); |
| 1152 | return error; |
| 1153 | } |
| 1154 | |
| 1155 | /* |
| 1156 | * A weaker form of d_revalidate for revalidating just the dentry->d_inode |
| 1157 | * when we don't really care about the dentry name. This is called when a |
| 1158 | * pathwalk ends on a dentry that was not found via a normal lookup in the |
| 1159 | * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals). |
| 1160 | * |
| 1161 | * In this situation, we just want to verify that the inode itself is OK |
| 1162 | * since the dentry might have changed on the server. |
| 1163 | */ |
| 1164 | static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags) |
| 1165 | { |
| 1166 | int error; |
| 1167 | struct inode *inode = dentry->d_inode; |
| 1168 | |
| 1169 | /* |
| 1170 | * I believe we can only get a negative dentry here in the case of a |
| 1171 | * procfs-style symlink. Just assume it's correct for now, but we may |
| 1172 | * eventually need to do something more here. |
| 1173 | */ |
| 1174 | if (!inode) { |
| 1175 | dfprintk(LOOKUPCACHE, "%s: %s/%s has negative inode\n", |
| 1176 | __func__, dentry->d_parent->d_name.name, |
| 1177 | dentry->d_name.name); |
| 1178 | return 1; |
| 1179 | } |
| 1180 | |
| 1181 | if (is_bad_inode(inode)) { |
| 1182 | dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n", |
| 1183 | __func__, dentry->d_parent->d_name.name, |
| 1184 | dentry->d_name.name); |
| 1185 | return 0; |
| 1186 | } |
| 1187 | |
| 1188 | error = nfs_revalidate_inode(NFS_SERVER(inode), inode); |
| 1189 | dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n", |
| 1190 | __func__, inode->i_ino, error ? "invalid" : "valid"); |
| 1191 | return !error; |
| 1192 | } |
| 1193 | |
| 1194 | /* |
| 1195 | * This is called from dput() when d_count is going to 0. |
| 1196 | */ |
| 1197 | static int nfs_dentry_delete(const struct dentry *dentry) |
| 1198 | { |
| 1199 | dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n", |
| 1200 | dentry->d_parent->d_name.name, dentry->d_name.name, |
| 1201 | dentry->d_flags); |
| 1202 | |
| 1203 | /* Unhash any dentry with a stale inode */ |
| 1204 | if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode)) |
| 1205 | return 1; |
| 1206 | |
| 1207 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
| 1208 | /* Unhash it, so that ->d_iput() would be called */ |
| 1209 | return 1; |
| 1210 | } |
| 1211 | if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { |
| 1212 | /* Unhash it, so that ancestors of killed async unlink |
| 1213 | * files will be cleaned up during umount */ |
| 1214 | return 1; |
| 1215 | } |
| 1216 | return 0; |
| 1217 | |
| 1218 | } |
| 1219 | |
| 1220 | /* Ensure that we revalidate inode->i_nlink */ |
| 1221 | static void nfs_drop_nlink(struct inode *inode) |
| 1222 | { |
| 1223 | spin_lock(&inode->i_lock); |
| 1224 | /* drop the inode if we're reasonably sure this is the last link */ |
| 1225 | if (inode->i_nlink == 1) |
| 1226 | clear_nlink(inode); |
| 1227 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR; |
| 1228 | spin_unlock(&inode->i_lock); |
| 1229 | } |
| 1230 | |
| 1231 | /* |
| 1232 | * Called when the dentry loses inode. |
| 1233 | * We use it to clean up silly-renamed files. |
| 1234 | */ |
| 1235 | static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) |
| 1236 | { |
| 1237 | if (S_ISDIR(inode->i_mode)) |
| 1238 | /* drop any readdir cache as it could easily be old */ |
| 1239 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; |
| 1240 | |
| 1241 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
| 1242 | nfs_complete_unlink(dentry, inode); |
| 1243 | nfs_drop_nlink(inode); |
| 1244 | } |
| 1245 | iput(inode); |
| 1246 | } |
| 1247 | |
| 1248 | static void nfs_d_release(struct dentry *dentry) |
| 1249 | { |
| 1250 | /* free cached devname value, if it survived that far */ |
| 1251 | if (unlikely(dentry->d_fsdata)) { |
| 1252 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) |
| 1253 | WARN_ON(1); |
| 1254 | else |
| 1255 | kfree(dentry->d_fsdata); |
| 1256 | } |
| 1257 | } |
| 1258 | |
| 1259 | const struct dentry_operations nfs_dentry_operations = { |
| 1260 | .d_revalidate = nfs_lookup_revalidate, |
| 1261 | .d_weak_revalidate = nfs_weak_revalidate, |
| 1262 | .d_delete = nfs_dentry_delete, |
| 1263 | .d_iput = nfs_dentry_iput, |
| 1264 | .d_automount = nfs_d_automount, |
| 1265 | .d_release = nfs_d_release, |
| 1266 | }; |
| 1267 | EXPORT_SYMBOL_GPL(nfs_dentry_operations); |
| 1268 | |
| 1269 | struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags) |
| 1270 | { |
| 1271 | struct dentry *res; |
| 1272 | struct dentry *parent; |
| 1273 | struct inode *inode = NULL; |
| 1274 | struct nfs_fh *fhandle = NULL; |
| 1275 | struct nfs_fattr *fattr = NULL; |
| 1276 | struct nfs4_label *label = NULL; |
| 1277 | int error; |
| 1278 | |
| 1279 | dfprintk(VFS, "NFS: lookup(%s/%s)\n", |
| 1280 | dentry->d_parent->d_name.name, dentry->d_name.name); |
| 1281 | nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); |
| 1282 | |
| 1283 | res = ERR_PTR(-ENAMETOOLONG); |
| 1284 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) |
| 1285 | goto out; |
| 1286 | |
| 1287 | /* |
| 1288 | * If we're doing an exclusive create, optimize away the lookup |
| 1289 | * but don't hash the dentry. |
| 1290 | */ |
| 1291 | if (nfs_is_exclusive_create(dir, flags)) { |
| 1292 | d_instantiate(dentry, NULL); |
| 1293 | res = NULL; |
| 1294 | goto out; |
| 1295 | } |
| 1296 | |
| 1297 | res = ERR_PTR(-ENOMEM); |
| 1298 | fhandle = nfs_alloc_fhandle(); |
| 1299 | fattr = nfs_alloc_fattr(); |
| 1300 | if (fhandle == NULL || fattr == NULL) |
| 1301 | goto out; |
| 1302 | |
| 1303 | label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT); |
| 1304 | if (IS_ERR(label)) |
| 1305 | goto out; |
| 1306 | |
| 1307 | parent = dentry->d_parent; |
| 1308 | /* Protect against concurrent sillydeletes */ |
| 1309 | nfs_block_sillyrename(parent); |
| 1310 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label); |
| 1311 | if (error == -ENOENT) |
| 1312 | goto no_entry; |
| 1313 | if (error < 0) { |
| 1314 | res = ERR_PTR(error); |
| 1315 | goto out_unblock_sillyrename; |
| 1316 | } |
| 1317 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); |
| 1318 | res = ERR_CAST(inode); |
| 1319 | if (IS_ERR(res)) |
| 1320 | goto out_unblock_sillyrename; |
| 1321 | |
| 1322 | /* Success: notify readdir to use READDIRPLUS */ |
| 1323 | nfs_advise_use_readdirplus(dir); |
| 1324 | |
| 1325 | no_entry: |
| 1326 | res = d_materialise_unique(dentry, inode); |
| 1327 | if (res != NULL) { |
| 1328 | if (IS_ERR(res)) |
| 1329 | goto out_unblock_sillyrename; |
| 1330 | dentry = res; |
| 1331 | } |
| 1332 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
| 1333 | out_unblock_sillyrename: |
| 1334 | nfs_unblock_sillyrename(parent); |
| 1335 | nfs4_label_free(label); |
| 1336 | out: |
| 1337 | nfs_free_fattr(fattr); |
| 1338 | nfs_free_fhandle(fhandle); |
| 1339 | return res; |
| 1340 | } |
| 1341 | EXPORT_SYMBOL_GPL(nfs_lookup); |
| 1342 | |
| 1343 | #if IS_ENABLED(CONFIG_NFS_V4) |
| 1344 | static int nfs4_lookup_revalidate(struct dentry *, unsigned int); |
| 1345 | |
| 1346 | const struct dentry_operations nfs4_dentry_operations = { |
| 1347 | .d_revalidate = nfs4_lookup_revalidate, |
| 1348 | .d_delete = nfs_dentry_delete, |
| 1349 | .d_iput = nfs_dentry_iput, |
| 1350 | .d_automount = nfs_d_automount, |
| 1351 | .d_release = nfs_d_release, |
| 1352 | }; |
| 1353 | EXPORT_SYMBOL_GPL(nfs4_dentry_operations); |
| 1354 | |
| 1355 | static fmode_t flags_to_mode(int flags) |
| 1356 | { |
| 1357 | fmode_t res = (__force fmode_t)flags & FMODE_EXEC; |
| 1358 | if ((flags & O_ACCMODE) != O_WRONLY) |
| 1359 | res |= FMODE_READ; |
| 1360 | if ((flags & O_ACCMODE) != O_RDONLY) |
| 1361 | res |= FMODE_WRITE; |
| 1362 | return res; |
| 1363 | } |
| 1364 | |
| 1365 | static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags) |
| 1366 | { |
| 1367 | return alloc_nfs_open_context(dentry, flags_to_mode(open_flags)); |
| 1368 | } |
| 1369 | |
| 1370 | static int do_open(struct inode *inode, struct file *filp) |
| 1371 | { |
| 1372 | nfs_fscache_set_inode_cookie(inode, filp); |
| 1373 | return 0; |
| 1374 | } |
| 1375 | |
| 1376 | static int nfs_finish_open(struct nfs_open_context *ctx, |
| 1377 | struct dentry *dentry, |
| 1378 | struct file *file, unsigned open_flags, |
| 1379 | int *opened) |
| 1380 | { |
| 1381 | int err; |
| 1382 | |
| 1383 | if (ctx->dentry != dentry) { |
| 1384 | dput(ctx->dentry); |
| 1385 | ctx->dentry = dget(dentry); |
| 1386 | } |
| 1387 | |
| 1388 | /* If the open_intent is for execute, we have an extra check to make */ |
| 1389 | if (ctx->mode & FMODE_EXEC) { |
| 1390 | err = nfs_may_open(dentry->d_inode, ctx->cred, open_flags); |
| 1391 | if (err < 0) |
| 1392 | goto out; |
| 1393 | } |
| 1394 | |
| 1395 | err = finish_open(file, dentry, do_open, opened); |
| 1396 | if (err) |
| 1397 | goto out; |
| 1398 | nfs_file_set_open_context(file, ctx); |
| 1399 | |
| 1400 | out: |
| 1401 | put_nfs_open_context(ctx); |
| 1402 | return err; |
| 1403 | } |
| 1404 | |
| 1405 | int nfs_atomic_open(struct inode *dir, struct dentry *dentry, |
| 1406 | struct file *file, unsigned open_flags, |
| 1407 | umode_t mode, int *opened) |
| 1408 | { |
| 1409 | struct nfs_open_context *ctx; |
| 1410 | struct dentry *res; |
| 1411 | struct iattr attr = { .ia_valid = ATTR_OPEN }; |
| 1412 | struct inode *inode; |
| 1413 | int err; |
| 1414 | |
| 1415 | /* Expect a negative dentry */ |
| 1416 | BUG_ON(dentry->d_inode); |
| 1417 | |
| 1418 | dfprintk(VFS, "NFS: atomic_open(%s/%ld), %s\n", |
| 1419 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); |
| 1420 | |
| 1421 | /* NFS only supports OPEN on regular files */ |
| 1422 | if ((open_flags & O_DIRECTORY)) { |
| 1423 | if (!d_unhashed(dentry)) { |
| 1424 | /* |
| 1425 | * Hashed negative dentry with O_DIRECTORY: dentry was |
| 1426 | * revalidated and is fine, no need to perform lookup |
| 1427 | * again |
| 1428 | */ |
| 1429 | return -ENOENT; |
| 1430 | } |
| 1431 | goto no_open; |
| 1432 | } |
| 1433 | |
| 1434 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) |
| 1435 | return -ENAMETOOLONG; |
| 1436 | |
| 1437 | if (open_flags & O_CREAT) { |
| 1438 | attr.ia_valid |= ATTR_MODE; |
| 1439 | attr.ia_mode = mode & ~current_umask(); |
| 1440 | } |
| 1441 | if (open_flags & O_TRUNC) { |
| 1442 | attr.ia_valid |= ATTR_SIZE; |
| 1443 | attr.ia_size = 0; |
| 1444 | } |
| 1445 | |
| 1446 | ctx = create_nfs_open_context(dentry, open_flags); |
| 1447 | err = PTR_ERR(ctx); |
| 1448 | if (IS_ERR(ctx)) |
| 1449 | goto out; |
| 1450 | |
| 1451 | nfs_block_sillyrename(dentry->d_parent); |
| 1452 | inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr); |
| 1453 | d_drop(dentry); |
| 1454 | if (IS_ERR(inode)) { |
| 1455 | nfs_unblock_sillyrename(dentry->d_parent); |
| 1456 | put_nfs_open_context(ctx); |
| 1457 | err = PTR_ERR(inode); |
| 1458 | switch (err) { |
| 1459 | case -ENOENT: |
| 1460 | d_add(dentry, NULL); |
| 1461 | break; |
| 1462 | case -EISDIR: |
| 1463 | case -ENOTDIR: |
| 1464 | goto no_open; |
| 1465 | case -ELOOP: |
| 1466 | if (!(open_flags & O_NOFOLLOW)) |
| 1467 | goto no_open; |
| 1468 | break; |
| 1469 | /* case -EINVAL: */ |
| 1470 | default: |
| 1471 | break; |
| 1472 | } |
| 1473 | goto out; |
| 1474 | } |
| 1475 | res = d_add_unique(dentry, inode); |
| 1476 | if (res != NULL) |
| 1477 | dentry = res; |
| 1478 | |
| 1479 | nfs_unblock_sillyrename(dentry->d_parent); |
| 1480 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
| 1481 | |
| 1482 | err = nfs_finish_open(ctx, dentry, file, open_flags, opened); |
| 1483 | |
| 1484 | dput(res); |
| 1485 | out: |
| 1486 | return err; |
| 1487 | |
| 1488 | no_open: |
| 1489 | res = nfs_lookup(dir, dentry, 0); |
| 1490 | err = PTR_ERR(res); |
| 1491 | if (IS_ERR(res)) |
| 1492 | goto out; |
| 1493 | |
| 1494 | return finish_no_open(file, res); |
| 1495 | } |
| 1496 | EXPORT_SYMBOL_GPL(nfs_atomic_open); |
| 1497 | |
| 1498 | static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags) |
| 1499 | { |
| 1500 | struct dentry *parent = NULL; |
| 1501 | struct inode *inode; |
| 1502 | struct inode *dir; |
| 1503 | int ret = 0; |
| 1504 | |
| 1505 | if (flags & LOOKUP_RCU) |
| 1506 | return -ECHILD; |
| 1507 | |
| 1508 | if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY)) |
| 1509 | goto no_open; |
| 1510 | if (d_mountpoint(dentry)) |
| 1511 | goto no_open; |
| 1512 | if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1) |
| 1513 | goto no_open; |
| 1514 | |
| 1515 | inode = dentry->d_inode; |
| 1516 | parent = dget_parent(dentry); |
| 1517 | dir = parent->d_inode; |
| 1518 | |
| 1519 | /* We can't create new files in nfs_open_revalidate(), so we |
| 1520 | * optimize away revalidation of negative dentries. |
| 1521 | */ |
| 1522 | if (inode == NULL) { |
| 1523 | if (!nfs_neg_need_reval(dir, dentry, flags)) |
| 1524 | ret = 1; |
| 1525 | goto out; |
| 1526 | } |
| 1527 | |
| 1528 | /* NFS only supports OPEN on regular files */ |
| 1529 | if (!S_ISREG(inode->i_mode)) |
| 1530 | goto no_open_dput; |
| 1531 | /* We cannot do exclusive creation on a positive dentry */ |
| 1532 | if (flags & LOOKUP_EXCL) |
| 1533 | goto no_open_dput; |
| 1534 | |
| 1535 | /* Let f_op->open() actually open (and revalidate) the file */ |
| 1536 | ret = 1; |
| 1537 | |
| 1538 | out: |
| 1539 | dput(parent); |
| 1540 | return ret; |
| 1541 | |
| 1542 | no_open_dput: |
| 1543 | dput(parent); |
| 1544 | no_open: |
| 1545 | return nfs_lookup_revalidate(dentry, flags); |
| 1546 | } |
| 1547 | |
| 1548 | #endif /* CONFIG_NFSV4 */ |
| 1549 | |
| 1550 | /* |
| 1551 | * Code common to create, mkdir, and mknod. |
| 1552 | */ |
| 1553 | int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, |
| 1554 | struct nfs_fattr *fattr, |
| 1555 | struct nfs4_label *label) |
| 1556 | { |
| 1557 | struct dentry *parent = dget_parent(dentry); |
| 1558 | struct inode *dir = parent->d_inode; |
| 1559 | struct inode *inode; |
| 1560 | int error = -EACCES; |
| 1561 | |
| 1562 | d_drop(dentry); |
| 1563 | |
| 1564 | /* We may have been initialized further down */ |
| 1565 | if (dentry->d_inode) |
| 1566 | goto out; |
| 1567 | if (fhandle->size == 0) { |
| 1568 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL); |
| 1569 | if (error) |
| 1570 | goto out_error; |
| 1571 | } |
| 1572 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
| 1573 | if (!(fattr->valid & NFS_ATTR_FATTR)) { |
| 1574 | struct nfs_server *server = NFS_SB(dentry->d_sb); |
| 1575 | error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL); |
| 1576 | if (error < 0) |
| 1577 | goto out_error; |
| 1578 | } |
| 1579 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); |
| 1580 | error = PTR_ERR(inode); |
| 1581 | if (IS_ERR(inode)) |
| 1582 | goto out_error; |
| 1583 | d_add(dentry, inode); |
| 1584 | out: |
| 1585 | dput(parent); |
| 1586 | return 0; |
| 1587 | out_error: |
| 1588 | nfs_mark_for_revalidate(dir); |
| 1589 | dput(parent); |
| 1590 | return error; |
| 1591 | } |
| 1592 | EXPORT_SYMBOL_GPL(nfs_instantiate); |
| 1593 | |
| 1594 | /* |
| 1595 | * Following a failed create operation, we drop the dentry rather |
| 1596 | * than retain a negative dentry. This avoids a problem in the event |
| 1597 | * that the operation succeeded on the server, but an error in the |
| 1598 | * reply path made it appear to have failed. |
| 1599 | */ |
| 1600 | int nfs_create(struct inode *dir, struct dentry *dentry, |
| 1601 | umode_t mode, bool excl) |
| 1602 | { |
| 1603 | struct iattr attr; |
| 1604 | int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT; |
| 1605 | int error; |
| 1606 | |
| 1607 | dfprintk(VFS, "NFS: create(%s/%ld), %s\n", |
| 1608 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); |
| 1609 | |
| 1610 | attr.ia_mode = mode; |
| 1611 | attr.ia_valid = ATTR_MODE; |
| 1612 | |
| 1613 | error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags); |
| 1614 | if (error != 0) |
| 1615 | goto out_err; |
| 1616 | return 0; |
| 1617 | out_err: |
| 1618 | d_drop(dentry); |
| 1619 | return error; |
| 1620 | } |
| 1621 | EXPORT_SYMBOL_GPL(nfs_create); |
| 1622 | |
| 1623 | /* |
| 1624 | * See comments for nfs_proc_create regarding failed operations. |
| 1625 | */ |
| 1626 | int |
| 1627 | nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev) |
| 1628 | { |
| 1629 | struct iattr attr; |
| 1630 | int status; |
| 1631 | |
| 1632 | dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n", |
| 1633 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); |
| 1634 | |
| 1635 | if (!new_valid_dev(rdev)) |
| 1636 | return -EINVAL; |
| 1637 | |
| 1638 | attr.ia_mode = mode; |
| 1639 | attr.ia_valid = ATTR_MODE; |
| 1640 | |
| 1641 | status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); |
| 1642 | if (status != 0) |
| 1643 | goto out_err; |
| 1644 | return 0; |
| 1645 | out_err: |
| 1646 | d_drop(dentry); |
| 1647 | return status; |
| 1648 | } |
| 1649 | EXPORT_SYMBOL_GPL(nfs_mknod); |
| 1650 | |
| 1651 | /* |
| 1652 | * See comments for nfs_proc_create regarding failed operations. |
| 1653 | */ |
| 1654 | int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
| 1655 | { |
| 1656 | struct iattr attr; |
| 1657 | int error; |
| 1658 | |
| 1659 | dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n", |
| 1660 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); |
| 1661 | |
| 1662 | attr.ia_valid = ATTR_MODE; |
| 1663 | attr.ia_mode = mode | S_IFDIR; |
| 1664 | |
| 1665 | error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); |
| 1666 | if (error != 0) |
| 1667 | goto out_err; |
| 1668 | return 0; |
| 1669 | out_err: |
| 1670 | d_drop(dentry); |
| 1671 | return error; |
| 1672 | } |
| 1673 | EXPORT_SYMBOL_GPL(nfs_mkdir); |
| 1674 | |
| 1675 | static void nfs_dentry_handle_enoent(struct dentry *dentry) |
| 1676 | { |
| 1677 | if (dentry->d_inode != NULL && !d_unhashed(dentry)) |
| 1678 | d_delete(dentry); |
| 1679 | } |
| 1680 | |
| 1681 | int nfs_rmdir(struct inode *dir, struct dentry *dentry) |
| 1682 | { |
| 1683 | int error; |
| 1684 | |
| 1685 | dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n", |
| 1686 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); |
| 1687 | |
| 1688 | error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); |
| 1689 | /* Ensure the VFS deletes this inode */ |
| 1690 | if (error == 0 && dentry->d_inode != NULL) |
| 1691 | clear_nlink(dentry->d_inode); |
| 1692 | else if (error == -ENOENT) |
| 1693 | nfs_dentry_handle_enoent(dentry); |
| 1694 | |
| 1695 | return error; |
| 1696 | } |
| 1697 | EXPORT_SYMBOL_GPL(nfs_rmdir); |
| 1698 | |
| 1699 | /* |
| 1700 | * Remove a file after making sure there are no pending writes, |
| 1701 | * and after checking that the file has only one user. |
| 1702 | * |
| 1703 | * We invalidate the attribute cache and free the inode prior to the operation |
| 1704 | * to avoid possible races if the server reuses the inode. |
| 1705 | */ |
| 1706 | static int nfs_safe_remove(struct dentry *dentry) |
| 1707 | { |
| 1708 | struct inode *dir = dentry->d_parent->d_inode; |
| 1709 | struct inode *inode = dentry->d_inode; |
| 1710 | int error = -EBUSY; |
| 1711 | |
| 1712 | dfprintk(VFS, "NFS: safe_remove(%s/%s)\n", |
| 1713 | dentry->d_parent->d_name.name, dentry->d_name.name); |
| 1714 | |
| 1715 | /* If the dentry was sillyrenamed, we simply call d_delete() */ |
| 1716 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
| 1717 | error = 0; |
| 1718 | goto out; |
| 1719 | } |
| 1720 | |
| 1721 | if (inode != NULL) { |
| 1722 | NFS_PROTO(inode)->return_delegation(inode); |
| 1723 | error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); |
| 1724 | if (error == 0) |
| 1725 | nfs_drop_nlink(inode); |
| 1726 | } else |
| 1727 | error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); |
| 1728 | if (error == -ENOENT) |
| 1729 | nfs_dentry_handle_enoent(dentry); |
| 1730 | out: |
| 1731 | return error; |
| 1732 | } |
| 1733 | |
| 1734 | /* We do silly rename. In case sillyrename() returns -EBUSY, the inode |
| 1735 | * belongs to an active ".nfs..." file and we return -EBUSY. |
| 1736 | * |
| 1737 | * If sillyrename() returns 0, we do nothing, otherwise we unlink. |
| 1738 | */ |
| 1739 | int nfs_unlink(struct inode *dir, struct dentry *dentry) |
| 1740 | { |
| 1741 | int error; |
| 1742 | int need_rehash = 0; |
| 1743 | |
| 1744 | dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id, |
| 1745 | dir->i_ino, dentry->d_name.name); |
| 1746 | |
| 1747 | spin_lock(&dentry->d_lock); |
| 1748 | if (dentry->d_count > 1) { |
| 1749 | spin_unlock(&dentry->d_lock); |
| 1750 | /* Start asynchronous writeout of the inode */ |
| 1751 | write_inode_now(dentry->d_inode, 0); |
| 1752 | error = nfs_sillyrename(dir, dentry); |
| 1753 | return error; |
| 1754 | } |
| 1755 | if (!d_unhashed(dentry)) { |
| 1756 | __d_drop(dentry); |
| 1757 | need_rehash = 1; |
| 1758 | } |
| 1759 | spin_unlock(&dentry->d_lock); |
| 1760 | error = nfs_safe_remove(dentry); |
| 1761 | if (!error || error == -ENOENT) { |
| 1762 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
| 1763 | } else if (need_rehash) |
| 1764 | d_rehash(dentry); |
| 1765 | return error; |
| 1766 | } |
| 1767 | EXPORT_SYMBOL_GPL(nfs_unlink); |
| 1768 | |
| 1769 | /* |
| 1770 | * To create a symbolic link, most file systems instantiate a new inode, |
| 1771 | * add a page to it containing the path, then write it out to the disk |
| 1772 | * using prepare_write/commit_write. |
| 1773 | * |
| 1774 | * Unfortunately the NFS client can't create the in-core inode first |
| 1775 | * because it needs a file handle to create an in-core inode (see |
| 1776 | * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the |
| 1777 | * symlink request has completed on the server. |
| 1778 | * |
| 1779 | * So instead we allocate a raw page, copy the symname into it, then do |
| 1780 | * the SYMLINK request with the page as the buffer. If it succeeds, we |
| 1781 | * now have a new file handle and can instantiate an in-core NFS inode |
| 1782 | * and move the raw page into its mapping. |
| 1783 | */ |
| 1784 | int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) |
| 1785 | { |
| 1786 | struct pagevec lru_pvec; |
| 1787 | struct page *page; |
| 1788 | char *kaddr; |
| 1789 | struct iattr attr; |
| 1790 | unsigned int pathlen = strlen(symname); |
| 1791 | int error; |
| 1792 | |
| 1793 | dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id, |
| 1794 | dir->i_ino, dentry->d_name.name, symname); |
| 1795 | |
| 1796 | if (pathlen > PAGE_SIZE) |
| 1797 | return -ENAMETOOLONG; |
| 1798 | |
| 1799 | attr.ia_mode = S_IFLNK | S_IRWXUGO; |
| 1800 | attr.ia_valid = ATTR_MODE; |
| 1801 | |
| 1802 | page = alloc_page(GFP_HIGHUSER); |
| 1803 | if (!page) |
| 1804 | return -ENOMEM; |
| 1805 | |
| 1806 | kaddr = kmap_atomic(page); |
| 1807 | memcpy(kaddr, symname, pathlen); |
| 1808 | if (pathlen < PAGE_SIZE) |
| 1809 | memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); |
| 1810 | kunmap_atomic(kaddr); |
| 1811 | |
| 1812 | error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); |
| 1813 | if (error != 0) { |
| 1814 | dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n", |
| 1815 | dir->i_sb->s_id, dir->i_ino, |
| 1816 | dentry->d_name.name, symname, error); |
| 1817 | d_drop(dentry); |
| 1818 | __free_page(page); |
| 1819 | return error; |
| 1820 | } |
| 1821 | |
| 1822 | /* |
| 1823 | * No big deal if we can't add this page to the page cache here. |
| 1824 | * READLINK will get the missing page from the server if needed. |
| 1825 | */ |
| 1826 | pagevec_init(&lru_pvec, 0); |
| 1827 | if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0, |
| 1828 | GFP_KERNEL)) { |
| 1829 | pagevec_add(&lru_pvec, page); |
| 1830 | pagevec_lru_add_file(&lru_pvec); |
| 1831 | SetPageUptodate(page); |
| 1832 | unlock_page(page); |
| 1833 | } else |
| 1834 | __free_page(page); |
| 1835 | |
| 1836 | return 0; |
| 1837 | } |
| 1838 | EXPORT_SYMBOL_GPL(nfs_symlink); |
| 1839 | |
| 1840 | int |
| 1841 | nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) |
| 1842 | { |
| 1843 | struct inode *inode = old_dentry->d_inode; |
| 1844 | int error; |
| 1845 | |
| 1846 | dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n", |
| 1847 | old_dentry->d_parent->d_name.name, old_dentry->d_name.name, |
| 1848 | dentry->d_parent->d_name.name, dentry->d_name.name); |
| 1849 | |
| 1850 | NFS_PROTO(inode)->return_delegation(inode); |
| 1851 | |
| 1852 | d_drop(dentry); |
| 1853 | error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); |
| 1854 | if (error == 0) { |
| 1855 | ihold(inode); |
| 1856 | d_add(dentry, inode); |
| 1857 | } |
| 1858 | return error; |
| 1859 | } |
| 1860 | EXPORT_SYMBOL_GPL(nfs_link); |
| 1861 | |
| 1862 | /* |
| 1863 | * RENAME |
| 1864 | * FIXME: Some nfsds, like the Linux user space nfsd, may generate a |
| 1865 | * different file handle for the same inode after a rename (e.g. when |
| 1866 | * moving to a different directory). A fail-safe method to do so would |
| 1867 | * be to look up old_dir/old_name, create a link to new_dir/new_name and |
| 1868 | * rename the old file using the sillyrename stuff. This way, the original |
| 1869 | * file in old_dir will go away when the last process iput()s the inode. |
| 1870 | * |
| 1871 | * FIXED. |
| 1872 | * |
| 1873 | * It actually works quite well. One needs to have the possibility for |
| 1874 | * at least one ".nfs..." file in each directory the file ever gets |
| 1875 | * moved or linked to which happens automagically with the new |
| 1876 | * implementation that only depends on the dcache stuff instead of |
| 1877 | * using the inode layer |
| 1878 | * |
| 1879 | * Unfortunately, things are a little more complicated than indicated |
| 1880 | * above. For a cross-directory move, we want to make sure we can get |
| 1881 | * rid of the old inode after the operation. This means there must be |
| 1882 | * no pending writes (if it's a file), and the use count must be 1. |
| 1883 | * If these conditions are met, we can drop the dentries before doing |
| 1884 | * the rename. |
| 1885 | */ |
| 1886 | int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
| 1887 | struct inode *new_dir, struct dentry *new_dentry) |
| 1888 | { |
| 1889 | struct inode *old_inode = old_dentry->d_inode; |
| 1890 | struct inode *new_inode = new_dentry->d_inode; |
| 1891 | struct dentry *dentry = NULL, *rehash = NULL; |
| 1892 | int error = -EBUSY; |
| 1893 | |
| 1894 | dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n", |
| 1895 | old_dentry->d_parent->d_name.name, old_dentry->d_name.name, |
| 1896 | new_dentry->d_parent->d_name.name, new_dentry->d_name.name, |
| 1897 | new_dentry->d_count); |
| 1898 | |
| 1899 | /* |
| 1900 | * For non-directories, check whether the target is busy and if so, |
| 1901 | * make a copy of the dentry and then do a silly-rename. If the |
| 1902 | * silly-rename succeeds, the copied dentry is hashed and becomes |
| 1903 | * the new target. |
| 1904 | */ |
| 1905 | if (new_inode && !S_ISDIR(new_inode->i_mode)) { |
| 1906 | /* |
| 1907 | * To prevent any new references to the target during the |
| 1908 | * rename, we unhash the dentry in advance. |
| 1909 | */ |
| 1910 | if (!d_unhashed(new_dentry)) { |
| 1911 | d_drop(new_dentry); |
| 1912 | rehash = new_dentry; |
| 1913 | } |
| 1914 | |
| 1915 | if (new_dentry->d_count > 2) { |
| 1916 | int err; |
| 1917 | |
| 1918 | /* copy the target dentry's name */ |
| 1919 | dentry = d_alloc(new_dentry->d_parent, |
| 1920 | &new_dentry->d_name); |
| 1921 | if (!dentry) |
| 1922 | goto out; |
| 1923 | |
| 1924 | /* silly-rename the existing target ... */ |
| 1925 | err = nfs_sillyrename(new_dir, new_dentry); |
| 1926 | if (err) |
| 1927 | goto out; |
| 1928 | |
| 1929 | new_dentry = dentry; |
| 1930 | rehash = NULL; |
| 1931 | new_inode = NULL; |
| 1932 | } |
| 1933 | } |
| 1934 | |
| 1935 | NFS_PROTO(old_inode)->return_delegation(old_inode); |
| 1936 | if (new_inode != NULL) |
| 1937 | NFS_PROTO(new_inode)->return_delegation(new_inode); |
| 1938 | |
| 1939 | error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name, |
| 1940 | new_dir, &new_dentry->d_name); |
| 1941 | nfs_mark_for_revalidate(old_inode); |
| 1942 | out: |
| 1943 | if (rehash) |
| 1944 | d_rehash(rehash); |
| 1945 | if (!error) { |
| 1946 | if (new_inode != NULL) |
| 1947 | nfs_drop_nlink(new_inode); |
| 1948 | d_move(old_dentry, new_dentry); |
| 1949 | nfs_set_verifier(new_dentry, |
| 1950 | nfs_save_change_attribute(new_dir)); |
| 1951 | } else if (error == -ENOENT) |
| 1952 | nfs_dentry_handle_enoent(old_dentry); |
| 1953 | |
| 1954 | /* new dentry created? */ |
| 1955 | if (dentry) |
| 1956 | dput(dentry); |
| 1957 | return error; |
| 1958 | } |
| 1959 | EXPORT_SYMBOL_GPL(nfs_rename); |
| 1960 | |
| 1961 | static DEFINE_SPINLOCK(nfs_access_lru_lock); |
| 1962 | static LIST_HEAD(nfs_access_lru_list); |
| 1963 | static atomic_long_t nfs_access_nr_entries; |
| 1964 | |
| 1965 | static void nfs_access_free_entry(struct nfs_access_entry *entry) |
| 1966 | { |
| 1967 | put_rpccred(entry->cred); |
| 1968 | kfree(entry); |
| 1969 | smp_mb__before_atomic_dec(); |
| 1970 | atomic_long_dec(&nfs_access_nr_entries); |
| 1971 | smp_mb__after_atomic_dec(); |
| 1972 | } |
| 1973 | |
| 1974 | static void nfs_access_free_list(struct list_head *head) |
| 1975 | { |
| 1976 | struct nfs_access_entry *cache; |
| 1977 | |
| 1978 | while (!list_empty(head)) { |
| 1979 | cache = list_entry(head->next, struct nfs_access_entry, lru); |
| 1980 | list_del(&cache->lru); |
| 1981 | nfs_access_free_entry(cache); |
| 1982 | } |
| 1983 | } |
| 1984 | |
| 1985 | int nfs_access_cache_shrinker(struct shrinker *shrink, |
| 1986 | struct shrink_control *sc) |
| 1987 | { |
| 1988 | LIST_HEAD(head); |
| 1989 | struct nfs_inode *nfsi, *next; |
| 1990 | struct nfs_access_entry *cache; |
| 1991 | int nr_to_scan = sc->nr_to_scan; |
| 1992 | gfp_t gfp_mask = sc->gfp_mask; |
| 1993 | |
| 1994 | if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) |
| 1995 | return (nr_to_scan == 0) ? 0 : -1; |
| 1996 | |
| 1997 | spin_lock(&nfs_access_lru_lock); |
| 1998 | list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { |
| 1999 | struct inode *inode; |
| 2000 | |
| 2001 | if (nr_to_scan-- == 0) |
| 2002 | break; |
| 2003 | inode = &nfsi->vfs_inode; |
| 2004 | spin_lock(&inode->i_lock); |
| 2005 | if (list_empty(&nfsi->access_cache_entry_lru)) |
| 2006 | goto remove_lru_entry; |
| 2007 | cache = list_entry(nfsi->access_cache_entry_lru.next, |
| 2008 | struct nfs_access_entry, lru); |
| 2009 | list_move(&cache->lru, &head); |
| 2010 | rb_erase(&cache->rb_node, &nfsi->access_cache); |
| 2011 | if (!list_empty(&nfsi->access_cache_entry_lru)) |
| 2012 | list_move_tail(&nfsi->access_cache_inode_lru, |
| 2013 | &nfs_access_lru_list); |
| 2014 | else { |
| 2015 | remove_lru_entry: |
| 2016 | list_del_init(&nfsi->access_cache_inode_lru); |
| 2017 | smp_mb__before_clear_bit(); |
| 2018 | clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); |
| 2019 | smp_mb__after_clear_bit(); |
| 2020 | } |
| 2021 | spin_unlock(&inode->i_lock); |
| 2022 | } |
| 2023 | spin_unlock(&nfs_access_lru_lock); |
| 2024 | nfs_access_free_list(&head); |
| 2025 | return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure; |
| 2026 | } |
| 2027 | |
| 2028 | static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) |
| 2029 | { |
| 2030 | struct rb_root *root_node = &nfsi->access_cache; |
| 2031 | struct rb_node *n; |
| 2032 | struct nfs_access_entry *entry; |
| 2033 | |
| 2034 | /* Unhook entries from the cache */ |
| 2035 | while ((n = rb_first(root_node)) != NULL) { |
| 2036 | entry = rb_entry(n, struct nfs_access_entry, rb_node); |
| 2037 | rb_erase(n, root_node); |
| 2038 | list_move(&entry->lru, head); |
| 2039 | } |
| 2040 | nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; |
| 2041 | } |
| 2042 | |
| 2043 | void nfs_access_zap_cache(struct inode *inode) |
| 2044 | { |
| 2045 | LIST_HEAD(head); |
| 2046 | |
| 2047 | if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) |
| 2048 | return; |
| 2049 | /* Remove from global LRU init */ |
| 2050 | spin_lock(&nfs_access_lru_lock); |
| 2051 | if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) |
| 2052 | list_del_init(&NFS_I(inode)->access_cache_inode_lru); |
| 2053 | |
| 2054 | spin_lock(&inode->i_lock); |
| 2055 | __nfs_access_zap_cache(NFS_I(inode), &head); |
| 2056 | spin_unlock(&inode->i_lock); |
| 2057 | spin_unlock(&nfs_access_lru_lock); |
| 2058 | nfs_access_free_list(&head); |
| 2059 | } |
| 2060 | EXPORT_SYMBOL_GPL(nfs_access_zap_cache); |
| 2061 | |
| 2062 | static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) |
| 2063 | { |
| 2064 | struct rb_node *n = NFS_I(inode)->access_cache.rb_node; |
| 2065 | struct nfs_access_entry *entry; |
| 2066 | |
| 2067 | while (n != NULL) { |
| 2068 | entry = rb_entry(n, struct nfs_access_entry, rb_node); |
| 2069 | |
| 2070 | if (cred < entry->cred) |
| 2071 | n = n->rb_left; |
| 2072 | else if (cred > entry->cred) |
| 2073 | n = n->rb_right; |
| 2074 | else |
| 2075 | return entry; |
| 2076 | } |
| 2077 | return NULL; |
| 2078 | } |
| 2079 | |
| 2080 | static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) |
| 2081 | { |
| 2082 | struct nfs_inode *nfsi = NFS_I(inode); |
| 2083 | struct nfs_access_entry *cache; |
| 2084 | int err = -ENOENT; |
| 2085 | |
| 2086 | spin_lock(&inode->i_lock); |
| 2087 | if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) |
| 2088 | goto out_zap; |
| 2089 | cache = nfs_access_search_rbtree(inode, cred); |
| 2090 | if (cache == NULL) |
| 2091 | goto out; |
| 2092 | if (!nfs_have_delegated_attributes(inode) && |
| 2093 | !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo)) |
| 2094 | goto out_stale; |
| 2095 | res->jiffies = cache->jiffies; |
| 2096 | res->cred = cache->cred; |
| 2097 | res->mask = cache->mask; |
| 2098 | list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); |
| 2099 | err = 0; |
| 2100 | out: |
| 2101 | spin_unlock(&inode->i_lock); |
| 2102 | return err; |
| 2103 | out_stale: |
| 2104 | rb_erase(&cache->rb_node, &nfsi->access_cache); |
| 2105 | list_del(&cache->lru); |
| 2106 | spin_unlock(&inode->i_lock); |
| 2107 | nfs_access_free_entry(cache); |
| 2108 | return -ENOENT; |
| 2109 | out_zap: |
| 2110 | spin_unlock(&inode->i_lock); |
| 2111 | nfs_access_zap_cache(inode); |
| 2112 | return -ENOENT; |
| 2113 | } |
| 2114 | |
| 2115 | static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) |
| 2116 | { |
| 2117 | struct nfs_inode *nfsi = NFS_I(inode); |
| 2118 | struct rb_root *root_node = &nfsi->access_cache; |
| 2119 | struct rb_node **p = &root_node->rb_node; |
| 2120 | struct rb_node *parent = NULL; |
| 2121 | struct nfs_access_entry *entry; |
| 2122 | |
| 2123 | spin_lock(&inode->i_lock); |
| 2124 | while (*p != NULL) { |
| 2125 | parent = *p; |
| 2126 | entry = rb_entry(parent, struct nfs_access_entry, rb_node); |
| 2127 | |
| 2128 | if (set->cred < entry->cred) |
| 2129 | p = &parent->rb_left; |
| 2130 | else if (set->cred > entry->cred) |
| 2131 | p = &parent->rb_right; |
| 2132 | else |
| 2133 | goto found; |
| 2134 | } |
| 2135 | rb_link_node(&set->rb_node, parent, p); |
| 2136 | rb_insert_color(&set->rb_node, root_node); |
| 2137 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
| 2138 | spin_unlock(&inode->i_lock); |
| 2139 | return; |
| 2140 | found: |
| 2141 | rb_replace_node(parent, &set->rb_node, root_node); |
| 2142 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
| 2143 | list_del(&entry->lru); |
| 2144 | spin_unlock(&inode->i_lock); |
| 2145 | nfs_access_free_entry(entry); |
| 2146 | } |
| 2147 | |
| 2148 | void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) |
| 2149 | { |
| 2150 | struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); |
| 2151 | if (cache == NULL) |
| 2152 | return; |
| 2153 | RB_CLEAR_NODE(&cache->rb_node); |
| 2154 | cache->jiffies = set->jiffies; |
| 2155 | cache->cred = get_rpccred(set->cred); |
| 2156 | cache->mask = set->mask; |
| 2157 | |
| 2158 | nfs_access_add_rbtree(inode, cache); |
| 2159 | |
| 2160 | /* Update accounting */ |
| 2161 | smp_mb__before_atomic_inc(); |
| 2162 | atomic_long_inc(&nfs_access_nr_entries); |
| 2163 | smp_mb__after_atomic_inc(); |
| 2164 | |
| 2165 | /* Add inode to global LRU list */ |
| 2166 | if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { |
| 2167 | spin_lock(&nfs_access_lru_lock); |
| 2168 | if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) |
| 2169 | list_add_tail(&NFS_I(inode)->access_cache_inode_lru, |
| 2170 | &nfs_access_lru_list); |
| 2171 | spin_unlock(&nfs_access_lru_lock); |
| 2172 | } |
| 2173 | } |
| 2174 | EXPORT_SYMBOL_GPL(nfs_access_add_cache); |
| 2175 | |
| 2176 | void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result) |
| 2177 | { |
| 2178 | entry->mask = 0; |
| 2179 | if (access_result & NFS4_ACCESS_READ) |
| 2180 | entry->mask |= MAY_READ; |
| 2181 | if (access_result & |
| 2182 | (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE)) |
| 2183 | entry->mask |= MAY_WRITE; |
| 2184 | if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE)) |
| 2185 | entry->mask |= MAY_EXEC; |
| 2186 | } |
| 2187 | EXPORT_SYMBOL_GPL(nfs_access_set_mask); |
| 2188 | |
| 2189 | static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) |
| 2190 | { |
| 2191 | struct nfs_access_entry cache; |
| 2192 | int status; |
| 2193 | |
| 2194 | status = nfs_access_get_cached(inode, cred, &cache); |
| 2195 | if (status == 0) |
| 2196 | goto out; |
| 2197 | |
| 2198 | /* Be clever: ask server to check for all possible rights */ |
| 2199 | cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; |
| 2200 | cache.cred = cred; |
| 2201 | cache.jiffies = jiffies; |
| 2202 | status = NFS_PROTO(inode)->access(inode, &cache); |
| 2203 | if (status != 0) { |
| 2204 | if (status == -ESTALE) { |
| 2205 | nfs_zap_caches(inode); |
| 2206 | if (!S_ISDIR(inode->i_mode)) |
| 2207 | set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); |
| 2208 | } |
| 2209 | return status; |
| 2210 | } |
| 2211 | nfs_access_add_cache(inode, &cache); |
| 2212 | out: |
| 2213 | if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
| 2214 | return 0; |
| 2215 | return -EACCES; |
| 2216 | } |
| 2217 | |
| 2218 | static int nfs_open_permission_mask(int openflags) |
| 2219 | { |
| 2220 | int mask = 0; |
| 2221 | |
| 2222 | if (openflags & __FMODE_EXEC) { |
| 2223 | /* ONLY check exec rights */ |
| 2224 | mask = MAY_EXEC; |
| 2225 | } else { |
| 2226 | if ((openflags & O_ACCMODE) != O_WRONLY) |
| 2227 | mask |= MAY_READ; |
| 2228 | if ((openflags & O_ACCMODE) != O_RDONLY) |
| 2229 | mask |= MAY_WRITE; |
| 2230 | } |
| 2231 | |
| 2232 | return mask; |
| 2233 | } |
| 2234 | |
| 2235 | int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags) |
| 2236 | { |
| 2237 | return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); |
| 2238 | } |
| 2239 | EXPORT_SYMBOL_GPL(nfs_may_open); |
| 2240 | |
| 2241 | int nfs_permission(struct inode *inode, int mask) |
| 2242 | { |
| 2243 | struct rpc_cred *cred; |
| 2244 | int res = 0; |
| 2245 | |
| 2246 | if (mask & MAY_NOT_BLOCK) |
| 2247 | return -ECHILD; |
| 2248 | |
| 2249 | nfs_inc_stats(inode, NFSIOS_VFSACCESS); |
| 2250 | |
| 2251 | if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
| 2252 | goto out; |
| 2253 | /* Is this sys_access() ? */ |
| 2254 | if (mask & (MAY_ACCESS | MAY_CHDIR)) |
| 2255 | goto force_lookup; |
| 2256 | |
| 2257 | switch (inode->i_mode & S_IFMT) { |
| 2258 | case S_IFLNK: |
| 2259 | goto out; |
| 2260 | case S_IFREG: |
| 2261 | /* NFSv4 has atomic_open... */ |
| 2262 | if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN) |
| 2263 | && (mask & MAY_OPEN) |
| 2264 | && !(mask & MAY_EXEC)) |
| 2265 | goto out; |
| 2266 | break; |
| 2267 | case S_IFDIR: |
| 2268 | /* |
| 2269 | * Optimize away all write operations, since the server |
| 2270 | * will check permissions when we perform the op. |
| 2271 | */ |
| 2272 | if ((mask & MAY_WRITE) && !(mask & MAY_READ)) |
| 2273 | goto out; |
| 2274 | } |
| 2275 | |
| 2276 | force_lookup: |
| 2277 | if (!NFS_PROTO(inode)->access) |
| 2278 | goto out_notsup; |
| 2279 | |
| 2280 | cred = rpc_lookup_cred(); |
| 2281 | if (!IS_ERR(cred)) { |
| 2282 | res = nfs_do_access(inode, cred, mask); |
| 2283 | put_rpccred(cred); |
| 2284 | } else |
| 2285 | res = PTR_ERR(cred); |
| 2286 | out: |
| 2287 | if (!res && (mask & MAY_EXEC) && !execute_ok(inode)) |
| 2288 | res = -EACCES; |
| 2289 | |
| 2290 | dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n", |
| 2291 | inode->i_sb->s_id, inode->i_ino, mask, res); |
| 2292 | return res; |
| 2293 | out_notsup: |
| 2294 | res = nfs_revalidate_inode(NFS_SERVER(inode), inode); |
| 2295 | if (res == 0) |
| 2296 | res = generic_permission(inode, mask); |
| 2297 | goto out; |
| 2298 | } |
| 2299 | EXPORT_SYMBOL_GPL(nfs_permission); |
| 2300 | |
| 2301 | /* |
| 2302 | * Local variables: |
| 2303 | * version-control: t |
| 2304 | * kept-new-versions: 5 |
| 2305 | * End: |
| 2306 | */ |