1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * The NFSD open file cache. 4 * 5 * (c) 2015 - Jeff Layton <jeff.layton@primarydata.com> 6 * 7 * An nfsd_file object is a per-file collection of open state that binds 8 * together: 9 * - a struct file * 10 * - a user credential 11 * - a network namespace 12 * - a read-ahead context 13 * - monitoring for writeback errors 14 * 15 * nfsd_file objects are reference-counted. Consumers acquire a new 16 * object via the nfsd_file_acquire API. They manage their interest in 17 * the acquired object, and hence the object's reference count, via 18 * nfsd_file_get and nfsd_file_put. There are two varieties of nfsd_file 19 * object: 20 * 21 * * non-garbage-collected: When a consumer wants to precisely control 22 * the lifetime of a file's open state, it acquires a non-garbage- 23 * collected nfsd_file. The final nfsd_file_put releases the open 24 * state immediately. 25 * 26 * * garbage-collected: When a consumer does not control the lifetime 27 * of open state, it acquires a garbage-collected nfsd_file. The 28 * final nfsd_file_put allows the open state to linger for a period 29 * during which it may be re-used. 30 */ 31 32 #include <linux/hash.h> 33 #include <linux/slab.h> 34 #include <linux/file.h> 35 #include <linux/pagemap.h> 36 #include <linux/sched.h> 37 #include <linux/list_lru.h> 38 #include <linux/fsnotify_backend.h> 39 #include <linux/fsnotify.h> 40 #include <linux/seq_file.h> 41 #include <linux/rhashtable.h> 42 #include <linux/nfslocalio.h> 43 44 #include "vfs.h" 45 #include "nfsd.h" 46 #include "nfsfh.h" 47 #include "netns.h" 48 #include "filecache.h" 49 #include "trace.h" 50 51 #define NFSD_LAUNDRETTE_DELAY (2 * HZ) 52 53 #define NFSD_FILE_CACHE_UP (0) 54 55 /* We only care about NFSD_MAY_READ/WRITE for this cache */ 56 #define NFSD_FILE_MAY_MASK (NFSD_MAY_READ|NFSD_MAY_WRITE|NFSD_MAY_LOCALIO) 57 58 static DEFINE_PER_CPU(unsigned long, nfsd_file_cache_hits); 59 static DEFINE_PER_CPU(unsigned long, nfsd_file_acquisitions); 60 static DEFINE_PER_CPU(unsigned long, nfsd_file_allocations); 61 static DEFINE_PER_CPU(unsigned long, nfsd_file_releases); 62 static DEFINE_PER_CPU(unsigned long, nfsd_file_total_age); 63 static DEFINE_PER_CPU(unsigned long, nfsd_file_evictions); 64 65 struct nfsd_fcache_disposal { 66 spinlock_t lock; 67 struct list_head freeme; 68 }; 69 70 static struct kmem_cache *nfsd_file_slab; 71 static struct kmem_cache *nfsd_file_mark_slab; 72 static struct list_lru nfsd_file_lru; 73 static unsigned long nfsd_file_flags; 74 static struct fsnotify_group *nfsd_file_fsnotify_group; 75 static struct delayed_work nfsd_filecache_laundrette; 76 static struct rhltable nfsd_file_rhltable 77 ____cacheline_aligned_in_smp; 78 79 static bool 80 nfsd_match_cred(const struct cred *c1, const struct cred *c2) 81 { 82 int i; 83 84 if (!uid_eq(c1->fsuid, c2->fsuid)) 85 return false; 86 if (!gid_eq(c1->fsgid, c2->fsgid)) 87 return false; 88 if (c1->group_info == NULL || c2->group_info == NULL) 89 return c1->group_info == c2->group_info; 90 if (c1->group_info->ngroups != c2->group_info->ngroups) 91 return false; 92 for (i = 0; i < c1->group_info->ngroups; i++) { 93 if (!gid_eq(c1->group_info->gid[i], c2->group_info->gid[i])) 94 return false; 95 } 96 return true; 97 } 98 99 static const struct rhashtable_params nfsd_file_rhash_params = { 100 .key_len = sizeof_field(struct nfsd_file, nf_inode), 101 .key_offset = offsetof(struct nfsd_file, nf_inode), 102 .head_offset = offsetof(struct nfsd_file, nf_rlist), 103 104 /* 105 * Start with a single page hash table to reduce resizing churn 106 * on light workloads. 107 */ 108 .min_size = 256, 109 .automatic_shrinking = true, 110 }; 111 112 static void 113 nfsd_file_schedule_laundrette(void) 114 { 115 if (test_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags)) 116 queue_delayed_work(system_dfl_wq, &nfsd_filecache_laundrette, 117 NFSD_LAUNDRETTE_DELAY); 118 } 119 120 static void 121 nfsd_file_slab_free(struct rcu_head *rcu) 122 { 123 struct nfsd_file *nf = container_of(rcu, struct nfsd_file, nf_rcu); 124 125 put_cred(nf->nf_cred); 126 kmem_cache_free(nfsd_file_slab, nf); 127 } 128 129 static void 130 nfsd_file_mark_free(struct fsnotify_mark *mark) 131 { 132 struct nfsd_file_mark *nfm = container_of(mark, struct nfsd_file_mark, 133 nfm_mark); 134 135 kmem_cache_free(nfsd_file_mark_slab, nfm); 136 } 137 138 static struct nfsd_file_mark * 139 nfsd_file_mark_get(struct nfsd_file_mark *nfm) 140 { 141 if (!refcount_inc_not_zero(&nfm->nfm_ref)) 142 return NULL; 143 return nfm; 144 } 145 146 static void 147 nfsd_file_mark_put(struct nfsd_file_mark *nfm) 148 { 149 if (refcount_dec_and_test(&nfm->nfm_ref)) { 150 fsnotify_destroy_mark(&nfm->nfm_mark, nfsd_file_fsnotify_group); 151 fsnotify_put_mark(&nfm->nfm_mark); 152 } 153 } 154 155 static struct nfsd_file_mark * 156 nfsd_file_mark_find_or_create(struct inode *inode) 157 { 158 int err; 159 struct fsnotify_mark *mark; 160 struct nfsd_file_mark *nfm = NULL, *new; 161 162 do { 163 fsnotify_group_lock(nfsd_file_fsnotify_group); 164 mark = fsnotify_find_inode_mark(inode, 165 nfsd_file_fsnotify_group); 166 if (mark) { 167 nfm = nfsd_file_mark_get(container_of(mark, 168 struct nfsd_file_mark, 169 nfm_mark)); 170 fsnotify_group_unlock(nfsd_file_fsnotify_group); 171 if (nfm) { 172 fsnotify_put_mark(mark); 173 break; 174 } 175 /* Avoid soft lockup race with nfsd_file_mark_put() */ 176 fsnotify_destroy_mark(mark, nfsd_file_fsnotify_group); 177 fsnotify_put_mark(mark); 178 } else { 179 fsnotify_group_unlock(nfsd_file_fsnotify_group); 180 } 181 182 /* allocate a new nfm */ 183 new = kmem_cache_alloc(nfsd_file_mark_slab, GFP_KERNEL); 184 if (!new) 185 return NULL; 186 fsnotify_init_mark(&new->nfm_mark, nfsd_file_fsnotify_group); 187 new->nfm_mark.mask = FS_ATTRIB|FS_DELETE_SELF; 188 refcount_set(&new->nfm_ref, 1); 189 190 err = fsnotify_add_inode_mark(&new->nfm_mark, inode, 0); 191 192 /* 193 * If the add was successful, then return the object. 194 * Otherwise, we need to put the reference we hold on the 195 * nfm_mark. The fsnotify code will take a reference and put 196 * it on failure, so we can't just free it directly. It's also 197 * not safe to call fsnotify_destroy_mark on it as the 198 * mark->group will be NULL. Thus, we can't let the nfm_ref 199 * counter drive the destruction at this point. 200 */ 201 if (likely(!err)) 202 nfm = new; 203 else 204 fsnotify_put_mark(&new->nfm_mark); 205 } while (unlikely(err == -EEXIST)); 206 207 return nfm; 208 } 209 210 static struct nfsd_file * 211 nfsd_file_alloc(struct net *net, struct inode *inode, unsigned char need, 212 bool want_gc) 213 { 214 struct nfsd_file *nf; 215 216 nf = kmem_cache_alloc(nfsd_file_slab, GFP_KERNEL); 217 if (unlikely(!nf)) 218 return NULL; 219 220 this_cpu_inc(nfsd_file_allocations); 221 INIT_LIST_HEAD(&nf->nf_lru); 222 INIT_LIST_HEAD(&nf->nf_gc); 223 nf->nf_birthtime = ktime_get(); 224 nf->nf_file = NULL; 225 nf->nf_cred = get_current_cred(); 226 nf->nf_net = net; 227 nf->nf_flags = want_gc ? 228 BIT(NFSD_FILE_HASHED) | BIT(NFSD_FILE_PENDING) | BIT(NFSD_FILE_GC) : 229 BIT(NFSD_FILE_HASHED) | BIT(NFSD_FILE_PENDING); 230 nf->nf_inode = inode; 231 refcount_set(&nf->nf_ref, 1); 232 nf->nf_may = need; 233 nf->nf_mark = NULL; 234 nf->nf_dio_mem_align = 0; 235 nf->nf_dio_offset_align = 0; 236 nf->nf_dio_read_offset_align = 0; 237 return nf; 238 } 239 240 /** 241 * nfsd_file_check_write_error - check for writeback errors on a file 242 * @nf: nfsd_file to check for writeback errors 243 * 244 * Check whether a nfsd_file has an unseen error. Reset the write 245 * verifier if so. 246 */ 247 static void 248 nfsd_file_check_write_error(struct nfsd_file *nf) 249 { 250 struct file *file = nf->nf_file; 251 252 if ((file->f_mode & FMODE_WRITE) && 253 filemap_check_wb_err(file->f_mapping, READ_ONCE(file->f_wb_err))) 254 nfsd_reset_write_verifier(net_generic(nf->nf_net, nfsd_net_id)); 255 } 256 257 static void 258 nfsd_file_hash_remove(struct nfsd_file *nf) 259 { 260 trace_nfsd_file_unhash(nf); 261 rhltable_remove(&nfsd_file_rhltable, &nf->nf_rlist, 262 nfsd_file_rhash_params); 263 } 264 265 static bool 266 nfsd_file_unhash(struct nfsd_file *nf) 267 { 268 if (test_and_clear_bit(NFSD_FILE_HASHED, &nf->nf_flags)) { 269 nfsd_file_hash_remove(nf); 270 return true; 271 } 272 return false; 273 } 274 275 static void 276 nfsd_file_free(struct nfsd_file *nf) 277 { 278 s64 age = ktime_to_ms(ktime_sub(ktime_get(), nf->nf_birthtime)); 279 280 trace_nfsd_file_free(nf); 281 282 this_cpu_inc(nfsd_file_releases); 283 this_cpu_add(nfsd_file_total_age, age); 284 285 nfsd_file_unhash(nf); 286 if (nf->nf_mark) 287 nfsd_file_mark_put(nf->nf_mark); 288 if (nf->nf_file) { 289 nfsd_file_check_write_error(nf); 290 nfsd_filp_close(nf->nf_file); 291 } 292 293 /* 294 * If this item is still linked via nf_lru, that's a bug. 295 * WARN and leak it to preserve system stability. 296 */ 297 if (WARN_ON_ONCE(!list_empty(&nf->nf_lru))) 298 return; 299 300 call_rcu(&nf->nf_rcu, nfsd_file_slab_free); 301 } 302 303 static bool 304 nfsd_file_check_writeback(struct nfsd_file *nf) 305 { 306 struct file *file = nf->nf_file; 307 struct address_space *mapping; 308 309 /* File not open for write? */ 310 if (!(file->f_mode & FMODE_WRITE)) 311 return false; 312 313 /* 314 * Some filesystems (e.g. NFS) flush all dirty data on close. 315 * On others, there is no need to wait for writeback. 316 */ 317 if (!(file_inode(file)->i_sb->s_export_op->flags & EXPORT_OP_FLUSH_ON_CLOSE)) 318 return false; 319 320 mapping = file->f_mapping; 321 return mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) || 322 mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK); 323 } 324 325 static void nfsd_file_lru_add(struct nfsd_file *nf) 326 { 327 refcount_inc(&nf->nf_ref); 328 if (list_lru_add_obj(&nfsd_file_lru, &nf->nf_lru)) 329 trace_nfsd_file_lru_add(nf); 330 else 331 WARN_ON(1); 332 nfsd_file_schedule_laundrette(); 333 } 334 335 static bool nfsd_file_lru_remove(struct nfsd_file *nf) 336 { 337 if (list_lru_del_obj(&nfsd_file_lru, &nf->nf_lru)) { 338 trace_nfsd_file_lru_del(nf); 339 return true; 340 } 341 return false; 342 } 343 344 struct nfsd_file * 345 nfsd_file_get(struct nfsd_file *nf) 346 { 347 if (nf && refcount_inc_not_zero(&nf->nf_ref)) 348 return nf; 349 return NULL; 350 } 351 352 /** 353 * nfsd_file_put - put the reference to a nfsd_file 354 * @nf: nfsd_file of which to put the reference 355 * 356 * Put a reference to a nfsd_file. In the non-GC case, we just put the 357 * reference immediately. In the GC case, if the reference would be 358 * the last one, the put it on the LRU instead to be cleaned up later. 359 */ 360 void 361 nfsd_file_put(struct nfsd_file *nf) 362 { 363 might_sleep(); 364 trace_nfsd_file_put(nf); 365 366 if (test_bit(NFSD_FILE_GC, &nf->nf_flags) && 367 test_bit(NFSD_FILE_HASHED, &nf->nf_flags)) { 368 set_bit(NFSD_FILE_REFERENCED, &nf->nf_flags); 369 set_bit(NFSD_FILE_RECENT, &nf->nf_flags); 370 } 371 372 if (refcount_dec_and_test(&nf->nf_ref)) 373 nfsd_file_free(nf); 374 } 375 376 /** 377 * nfsd_file_put_local - put nfsd_file reference and arm nfsd_net_put in caller 378 * @pnf: nfsd_file of which to put the reference 379 * 380 * First save the associated net to return to caller, then put 381 * the reference of the nfsd_file. 382 */ 383 struct net * 384 nfsd_file_put_local(struct nfsd_file __rcu **pnf) 385 { 386 struct nfsd_file *nf; 387 struct net *net = NULL; 388 389 nf = unrcu_pointer(xchg(pnf, NULL)); 390 if (nf) { 391 net = nf->nf_net; 392 nfsd_file_put(nf); 393 } 394 return net; 395 } 396 397 /** 398 * nfsd_file_file - get the backing file of an nfsd_file 399 * @nf: nfsd_file of which to access the backing file. 400 * 401 * Return backing file for @nf. 402 */ 403 struct file * 404 nfsd_file_file(struct nfsd_file *nf) 405 { 406 return nf->nf_file; 407 } 408 409 static void 410 nfsd_file_dispose_list(struct list_head *dispose) 411 { 412 struct nfsd_file *nf; 413 414 while (!list_empty(dispose)) { 415 nf = list_first_entry(dispose, struct nfsd_file, nf_gc); 416 list_del_init(&nf->nf_gc); 417 nfsd_file_free(nf); 418 } 419 } 420 421 /** 422 * nfsd_file_dispose_list_delayed - move list of dead files to net's freeme list 423 * @dispose: list of nfsd_files to be disposed 424 * 425 * Transfers each file to the "freeme" list for its nfsd_net, to eventually 426 * be disposed of by the per-net garbage collector. 427 */ 428 static void 429 nfsd_file_dispose_list_delayed(struct list_head *dispose) 430 { 431 while(!list_empty(dispose)) { 432 struct nfsd_file *nf = list_first_entry(dispose, 433 struct nfsd_file, nf_gc); 434 struct nfsd_net *nn = net_generic(nf->nf_net, nfsd_net_id); 435 struct nfsd_fcache_disposal *l = nn->fcache_disposal; 436 struct svc_serv *serv; 437 438 spin_lock(&l->lock); 439 list_move_tail(&nf->nf_gc, &l->freeme); 440 spin_unlock(&l->lock); 441 442 /* 443 * The filecache laundrette is shut down after the 444 * nn->nfsd_serv pointer is cleared, but before the 445 * svc_serv is freed. 446 */ 447 serv = nn->nfsd_serv; 448 if (serv) 449 svc_wake_up(serv); 450 } 451 } 452 453 /** 454 * nfsd_file_net_dispose - deal with nfsd_files waiting to be disposed. 455 * @nn: nfsd_net in which to find files to be disposed. 456 * 457 * When files held open for nfsv3 are removed from the filecache, whether 458 * due to memory pressure or garbage collection, they are queued to 459 * a per-net-ns queue. This function completes the disposal, either 460 * directly or by waking another nfsd thread to help with the work. 461 */ 462 void nfsd_file_net_dispose(struct nfsd_net *nn) 463 { 464 struct nfsd_fcache_disposal *l = nn->fcache_disposal; 465 466 if (!list_empty(&l->freeme)) { 467 LIST_HEAD(dispose); 468 int i; 469 470 spin_lock(&l->lock); 471 for (i = 0; i < 8 && !list_empty(&l->freeme); i++) 472 list_move(l->freeme.next, &dispose); 473 spin_unlock(&l->lock); 474 if (!list_empty(&l->freeme)) 475 /* Wake up another thread to share the work 476 * *before* doing any actual disposing. 477 */ 478 svc_wake_up(nn->nfsd_serv); 479 nfsd_file_dispose_list(&dispose); 480 } 481 } 482 483 /** 484 * nfsd_file_lru_cb - Examine an entry on the LRU list 485 * @item: LRU entry to examine 486 * @lru: controlling LRU 487 * @arg: dispose list 488 * 489 * Return values: 490 * %LRU_REMOVED: @item was removed from the LRU 491 * %LRU_ROTATE: @item is to be moved to the LRU tail 492 * %LRU_SKIP: @item cannot be evicted 493 */ 494 static enum lru_status 495 nfsd_file_lru_cb(struct list_head *item, struct list_lru_one *lru, 496 void *arg) 497 { 498 struct list_head *head = arg; 499 struct nfsd_file *nf = list_entry(item, struct nfsd_file, nf_lru); 500 501 /* We should only be dealing with GC entries here */ 502 WARN_ON_ONCE(!test_bit(NFSD_FILE_GC, &nf->nf_flags)); 503 504 /* 505 * Don't throw out files that are still undergoing I/O or 506 * that have uncleared errors pending. 507 */ 508 if (nfsd_file_check_writeback(nf)) { 509 trace_nfsd_file_gc_writeback(nf); 510 return LRU_SKIP; 511 } 512 513 /* If it was recently added to the list, skip it */ 514 if (test_and_clear_bit(NFSD_FILE_REFERENCED, &nf->nf_flags)) { 515 trace_nfsd_file_gc_referenced(nf); 516 return LRU_ROTATE; 517 } 518 519 /* 520 * Put the reference held on behalf of the LRU if it is the last 521 * reference, else rotate. 522 */ 523 if (!refcount_dec_if_one(&nf->nf_ref)) { 524 trace_nfsd_file_gc_in_use(nf); 525 return LRU_ROTATE; 526 } 527 528 /* Refcount went to zero. Unhash it and queue it to the dispose list */ 529 nfsd_file_unhash(nf); 530 list_lru_isolate(lru, &nf->nf_lru); 531 list_add(&nf->nf_gc, head); 532 this_cpu_inc(nfsd_file_evictions); 533 trace_nfsd_file_gc_disposed(nf); 534 return LRU_REMOVED; 535 } 536 537 static enum lru_status 538 nfsd_file_gc_cb(struct list_head *item, struct list_lru_one *lru, 539 void *arg) 540 { 541 struct nfsd_file *nf = list_entry(item, struct nfsd_file, nf_lru); 542 543 if (test_and_clear_bit(NFSD_FILE_RECENT, &nf->nf_flags)) { 544 /* 545 * "REFERENCED" really means "should be at the end of the 546 * LRU. As we are putting it there we can clear the flag. 547 */ 548 clear_bit(NFSD_FILE_REFERENCED, &nf->nf_flags); 549 trace_nfsd_file_gc_aged(nf); 550 return LRU_ROTATE; 551 } 552 return nfsd_file_lru_cb(item, lru, arg); 553 } 554 555 /* If the shrinker runs between calls to list_lru_walk_node() in 556 * nfsd_file_gc(), the "remaining" count will be wrong. This could 557 * result in premature freeing of some files. This may not matter much 558 * but is easy to fix with this spinlock which temporarily disables 559 * the shrinker. 560 */ 561 static DEFINE_SPINLOCK(nfsd_gc_lock); 562 static void 563 nfsd_file_gc(void) 564 { 565 unsigned long ret = 0; 566 LIST_HEAD(dispose); 567 int nid; 568 569 spin_lock(&nfsd_gc_lock); 570 for_each_node_state(nid, N_NORMAL_MEMORY) { 571 unsigned long remaining = list_lru_count_node(&nfsd_file_lru, nid); 572 573 while (remaining > 0) { 574 unsigned long nr = min(remaining, NFSD_FILE_GC_BATCH); 575 576 remaining -= nr; 577 ret += list_lru_walk_node(&nfsd_file_lru, nid, nfsd_file_gc_cb, 578 &dispose, &nr); 579 if (nr) 580 /* walk aborted early */ 581 remaining = 0; 582 } 583 } 584 spin_unlock(&nfsd_gc_lock); 585 trace_nfsd_file_gc_removed(ret, list_lru_count(&nfsd_file_lru)); 586 nfsd_file_dispose_list_delayed(&dispose); 587 } 588 589 static void 590 nfsd_file_gc_worker(struct work_struct *work) 591 { 592 if (list_lru_count(&nfsd_file_lru)) 593 nfsd_file_gc(); 594 nfsd_file_schedule_laundrette(); 595 } 596 597 static unsigned long 598 nfsd_file_lru_count(struct shrinker *s, struct shrink_control *sc) 599 { 600 return list_lru_count(&nfsd_file_lru); 601 } 602 603 static unsigned long 604 nfsd_file_lru_scan(struct shrinker *s, struct shrink_control *sc) 605 { 606 LIST_HEAD(dispose); 607 unsigned long ret; 608 609 if (!spin_trylock(&nfsd_gc_lock)) 610 return SHRINK_STOP; 611 612 ret = list_lru_shrink_walk(&nfsd_file_lru, sc, 613 nfsd_file_lru_cb, &dispose); 614 spin_unlock(&nfsd_gc_lock); 615 trace_nfsd_file_shrinker_removed(ret, list_lru_count(&nfsd_file_lru)); 616 nfsd_file_dispose_list_delayed(&dispose); 617 return ret; 618 } 619 620 static struct shrinker *nfsd_file_shrinker; 621 622 /** 623 * nfsd_file_cond_queue - conditionally unhash and queue a nfsd_file 624 * @nf: nfsd_file to attempt to queue 625 * @dispose: private list to queue successfully-put objects 626 * 627 * Unhash an nfsd_file, try to get a reference to it, and then put that 628 * reference. If it's the last reference, queue it to the dispose list. 629 */ 630 static void 631 nfsd_file_cond_queue(struct nfsd_file *nf, struct list_head *dispose) 632 __must_hold(RCU) 633 { 634 int decrement = 1; 635 636 /* If we raced with someone else unhashing, ignore it */ 637 if (!nfsd_file_unhash(nf)) 638 return; 639 640 /* If we can't get a reference, ignore it */ 641 if (!nfsd_file_get(nf)) 642 return; 643 644 /* Extra decrement if we remove from the LRU */ 645 if (nfsd_file_lru_remove(nf)) 646 ++decrement; 647 648 /* If refcount goes to 0, then put on the dispose list */ 649 if (refcount_sub_and_test(decrement, &nf->nf_ref)) { 650 list_add(&nf->nf_gc, dispose); 651 trace_nfsd_file_closing(nf); 652 } 653 } 654 655 /** 656 * nfsd_file_queue_for_close: try to close out any open nfsd_files for an inode 657 * @inode: inode on which to close out nfsd_files 658 * @dispose: list on which to gather nfsd_files to close out 659 * 660 * An nfsd_file represents a struct file being held open on behalf of nfsd. 661 * An open file however can block other activity (such as leases), or cause 662 * undesirable behavior (e.g. spurious silly-renames when reexporting NFS). 663 * 664 * This function is intended to find open nfsd_files when this sort of 665 * conflicting access occurs and then attempt to close those files out. 666 * 667 * Populates the dispose list with entries that have already had their 668 * refcounts go to zero. The actual free of an nfsd_file can be expensive, 669 * so we leave it up to the caller whether it wants to wait or not. 670 */ 671 static void 672 nfsd_file_queue_for_close(struct inode *inode, struct list_head *dispose) 673 { 674 struct rhlist_head *tmp, *list; 675 struct nfsd_file *nf; 676 677 rcu_read_lock(); 678 list = rhltable_lookup(&nfsd_file_rhltable, &inode, 679 nfsd_file_rhash_params); 680 rhl_for_each_entry_rcu(nf, tmp, list, nf_rlist) { 681 if (!test_bit(NFSD_FILE_GC, &nf->nf_flags)) 682 continue; 683 nfsd_file_cond_queue(nf, dispose); 684 } 685 rcu_read_unlock(); 686 } 687 688 /** 689 * nfsd_file_close_inode - attempt a delayed close of a nfsd_file 690 * @inode: inode of the file to attempt to remove 691 * 692 * Close out any open nfsd_files that can be reaped for @inode. The 693 * actual freeing is deferred to the dispose_list_delayed infrastructure. 694 * 695 * This is used by the fsnotify callbacks and setlease notifier. 696 */ 697 static void 698 nfsd_file_close_inode(struct inode *inode) 699 { 700 LIST_HEAD(dispose); 701 702 nfsd_file_queue_for_close(inode, &dispose); 703 nfsd_file_dispose_list_delayed(&dispose); 704 } 705 706 /** 707 * nfsd_file_close_inode_sync - attempt to forcibly close a nfsd_file 708 * @inode: inode of the file to attempt to remove 709 * 710 * Close out any open nfsd_files that can be reaped for @inode. The 711 * nfsd_files are closed out synchronously. 712 * 713 * This is called from nfsd_rename and nfsd_unlink to avoid silly-renames 714 * when reexporting NFS. 715 */ 716 void 717 nfsd_file_close_inode_sync(struct inode *inode) 718 { 719 LIST_HEAD(dispose); 720 721 trace_nfsd_file_close(inode); 722 723 nfsd_file_queue_for_close(inode, &dispose); 724 nfsd_file_dispose_list(&dispose); 725 } 726 727 /** 728 * nfsd_file_close_export - close cached file handles for an export 729 * @net: net namespace in which to operate 730 * @path: export path whose cached files should be closed 731 * 732 * Close out GC-managed nfsd_file entries whose underlying file is on 733 * the same filesystem as, and a descendant of, @path. nfsd_file 734 * entries do not carry an export reference, so the check uses the 735 * file's dentry ancestry. False positives (closing a cached handle 736 * that did not originate from the target export) are harmless -- the 737 * handle is simply reopened on the next access. 738 * 739 * Called from the NFSD_CMD_UNLOCK_EXPORT handler before revoking 740 * NFSv4 state, to ensure that cached file handles do not hold the 741 * filesystem busy. 742 */ 743 void nfsd_file_close_export(struct net *net, const struct path *path) 744 { 745 struct rhashtable_iter iter; 746 struct nfsd_file *nf; 747 LIST_HEAD(dispose); 748 749 rhltable_walk_enter(&nfsd_file_rhltable, &iter); 750 do { 751 rhashtable_walk_start(&iter); 752 753 nf = rhashtable_walk_next(&iter); 754 while (!IS_ERR_OR_NULL(nf)) { 755 if (nf->nf_net == net && 756 test_bit(NFSD_FILE_GC, &nf->nf_flags) && 757 nf->nf_file && 758 file_inode(nf->nf_file)->i_sb == 759 path->dentry->d_sb && 760 is_subdir(nf->nf_file->f_path.dentry, 761 path->dentry)) 762 nfsd_file_cond_queue(nf, &dispose); 763 nf = rhashtable_walk_next(&iter); 764 } 765 766 rhashtable_walk_stop(&iter); 767 } while (nf == ERR_PTR(-EAGAIN)); 768 rhashtable_walk_exit(&iter); 769 770 nfsd_file_dispose_list(&dispose); 771 } 772 773 static int 774 nfsd_file_lease_notifier_call(struct notifier_block *nb, unsigned long arg, 775 void *data) 776 { 777 struct file_lease *fl = data; 778 779 /* Only close files for F_SETLEASE leases */ 780 if (fl->c.flc_flags & FL_LEASE) 781 nfsd_file_close_inode(file_inode(fl->c.flc_file)); 782 return 0; 783 } 784 785 static struct notifier_block nfsd_file_lease_notifier = { 786 .notifier_call = nfsd_file_lease_notifier_call, 787 }; 788 789 static int 790 nfsd_file_fsnotify_handle_event(struct fsnotify_mark *mark, u32 mask, 791 struct inode *inode, struct inode *dir, 792 const struct qstr *name, u32 cookie) 793 { 794 if (WARN_ON_ONCE(!inode)) 795 return 0; 796 797 trace_nfsd_file_fsnotify_handle_event(inode, mask); 798 799 /* Should be no marks on non-regular files */ 800 if (!S_ISREG(inode->i_mode)) { 801 WARN_ON_ONCE(1); 802 return 0; 803 } 804 805 /* don't close files if this was not the last link */ 806 if (mask & FS_ATTRIB) { 807 if (inode->i_nlink) 808 return 0; 809 } 810 811 nfsd_file_close_inode(inode); 812 return 0; 813 } 814 815 816 static const struct fsnotify_ops nfsd_file_fsnotify_ops = { 817 .handle_inode_event = nfsd_file_fsnotify_handle_event, 818 .free_mark = nfsd_file_mark_free, 819 }; 820 821 int 822 nfsd_file_cache_init(void) 823 { 824 int ret; 825 826 lockdep_assert_held(&nfsd_mutex); 827 if (test_and_set_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags) == 1) 828 return 0; 829 830 ret = rhltable_init(&nfsd_file_rhltable, &nfsd_file_rhash_params); 831 if (ret) 832 goto out; 833 834 ret = -ENOMEM; 835 nfsd_file_slab = KMEM_CACHE(nfsd_file, 0); 836 if (!nfsd_file_slab) { 837 pr_err("nfsd: unable to create nfsd_file_slab\n"); 838 goto out_err; 839 } 840 841 nfsd_file_mark_slab = KMEM_CACHE(nfsd_file_mark, 0); 842 if (!nfsd_file_mark_slab) { 843 pr_err("nfsd: unable to create nfsd_file_mark_slab\n"); 844 goto out_err; 845 } 846 847 ret = list_lru_init(&nfsd_file_lru); 848 if (ret) { 849 pr_err("nfsd: failed to init nfsd_file_lru: %d\n", ret); 850 goto out_err; 851 } 852 853 nfsd_file_shrinker = shrinker_alloc(0, "nfsd-filecache"); 854 if (!nfsd_file_shrinker) { 855 ret = -ENOMEM; 856 pr_err("nfsd: failed to allocate nfsd_file_shrinker\n"); 857 goto out_lru; 858 } 859 860 nfsd_file_shrinker->count_objects = nfsd_file_lru_count; 861 nfsd_file_shrinker->scan_objects = nfsd_file_lru_scan; 862 nfsd_file_shrinker->seeks = 1; 863 864 shrinker_register(nfsd_file_shrinker); 865 866 ret = lease_register_notifier(&nfsd_file_lease_notifier); 867 if (ret) { 868 pr_err("nfsd: unable to register lease notifier: %d\n", ret); 869 goto out_shrinker; 870 } 871 872 nfsd_file_fsnotify_group = fsnotify_alloc_group(&nfsd_file_fsnotify_ops, 873 0); 874 if (IS_ERR(nfsd_file_fsnotify_group)) { 875 pr_err("nfsd: unable to create fsnotify group: %ld\n", 876 PTR_ERR(nfsd_file_fsnotify_group)); 877 ret = PTR_ERR(nfsd_file_fsnotify_group); 878 nfsd_file_fsnotify_group = NULL; 879 goto out_notifier; 880 } 881 882 INIT_DELAYED_WORK(&nfsd_filecache_laundrette, nfsd_file_gc_worker); 883 out: 884 if (ret) 885 clear_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags); 886 return ret; 887 out_notifier: 888 lease_unregister_notifier(&nfsd_file_lease_notifier); 889 out_shrinker: 890 shrinker_free(nfsd_file_shrinker); 891 out_lru: 892 list_lru_destroy(&nfsd_file_lru); 893 out_err: 894 kmem_cache_destroy(nfsd_file_slab); 895 nfsd_file_slab = NULL; 896 kmem_cache_destroy(nfsd_file_mark_slab); 897 nfsd_file_mark_slab = NULL; 898 rhltable_destroy(&nfsd_file_rhltable); 899 goto out; 900 } 901 902 /** 903 * __nfsd_file_cache_purge: clean out the cache for shutdown 904 * @net: net-namespace to shut down the cache (may be NULL) 905 * 906 * Walk the nfsd_file cache and close out any that match @net. If @net is NULL, 907 * then close out everything. Called when an nfsd instance is being shut down, 908 * and when the exports table is flushed. 909 */ 910 static void 911 __nfsd_file_cache_purge(struct net *net) 912 { 913 struct rhashtable_iter iter; 914 struct nfsd_file *nf; 915 LIST_HEAD(dispose); 916 917 #if IS_ENABLED(CONFIG_NFS_LOCALIO) 918 if (net) { 919 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 920 nfs_localio_invalidate_clients(&nn->local_clients, 921 &nn->local_clients_lock); 922 } 923 #endif 924 925 rhltable_walk_enter(&nfsd_file_rhltable, &iter); 926 do { 927 rhashtable_walk_start(&iter); 928 929 nf = rhashtable_walk_next(&iter); 930 while (!IS_ERR_OR_NULL(nf)) { 931 if (!net || nf->nf_net == net) 932 nfsd_file_cond_queue(nf, &dispose); 933 nf = rhashtable_walk_next(&iter); 934 } 935 936 rhashtable_walk_stop(&iter); 937 } while (nf == ERR_PTR(-EAGAIN)); 938 rhashtable_walk_exit(&iter); 939 940 nfsd_file_dispose_list(&dispose); 941 } 942 943 static struct nfsd_fcache_disposal * 944 nfsd_alloc_fcache_disposal(void) 945 { 946 struct nfsd_fcache_disposal *l; 947 948 l = kmalloc_obj(*l); 949 if (!l) 950 return NULL; 951 spin_lock_init(&l->lock); 952 INIT_LIST_HEAD(&l->freeme); 953 return l; 954 } 955 956 static void 957 nfsd_free_fcache_disposal(struct nfsd_fcache_disposal *l) 958 { 959 nfsd_file_dispose_list(&l->freeme); 960 kfree(l); 961 } 962 963 static void 964 nfsd_free_fcache_disposal_net(struct net *net) 965 { 966 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 967 struct nfsd_fcache_disposal *l = nn->fcache_disposal; 968 969 nfsd_free_fcache_disposal(l); 970 } 971 972 int 973 nfsd_file_cache_start_net(struct net *net) 974 { 975 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 976 977 nn->fcache_disposal = nfsd_alloc_fcache_disposal(); 978 return nn->fcache_disposal ? 0 : -ENOMEM; 979 } 980 981 /** 982 * nfsd_file_cache_purge - Remove all cache items associated with @net 983 * @net: target net namespace 984 * 985 */ 986 void 987 nfsd_file_cache_purge(struct net *net) 988 { 989 lockdep_assert_held(&nfsd_mutex); 990 if (test_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags) == 1) 991 __nfsd_file_cache_purge(net); 992 } 993 994 void 995 nfsd_file_cache_shutdown_net(struct net *net) 996 { 997 nfsd_file_cache_purge(net); 998 nfsd_free_fcache_disposal_net(net); 999 } 1000 1001 void 1002 nfsd_file_cache_shutdown(void) 1003 { 1004 int i; 1005 1006 lockdep_assert_held(&nfsd_mutex); 1007 if (test_and_clear_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags) == 0) 1008 return; 1009 1010 lease_unregister_notifier(&nfsd_file_lease_notifier); 1011 shrinker_free(nfsd_file_shrinker); 1012 /* 1013 * make sure all callers of nfsd_file_lru_cb are done before 1014 * calling nfsd_file_cache_purge 1015 */ 1016 cancel_delayed_work_sync(&nfsd_filecache_laundrette); 1017 __nfsd_file_cache_purge(NULL); 1018 list_lru_destroy(&nfsd_file_lru); 1019 rcu_barrier(); 1020 fsnotify_put_group(nfsd_file_fsnotify_group); 1021 nfsd_file_fsnotify_group = NULL; 1022 kmem_cache_destroy(nfsd_file_slab); 1023 nfsd_file_slab = NULL; 1024 fsnotify_wait_marks_destroyed(); 1025 kmem_cache_destroy(nfsd_file_mark_slab); 1026 nfsd_file_mark_slab = NULL; 1027 rhltable_destroy(&nfsd_file_rhltable); 1028 1029 for_each_possible_cpu(i) { 1030 per_cpu(nfsd_file_cache_hits, i) = 0; 1031 per_cpu(nfsd_file_acquisitions, i) = 0; 1032 per_cpu(nfsd_file_allocations, i) = 0; 1033 per_cpu(nfsd_file_releases, i) = 0; 1034 per_cpu(nfsd_file_total_age, i) = 0; 1035 per_cpu(nfsd_file_evictions, i) = 0; 1036 } 1037 } 1038 1039 static struct nfsd_file * 1040 nfsd_file_lookup_locked(const struct net *net, const struct cred *cred, 1041 struct inode *inode, unsigned char need, 1042 bool want_gc) 1043 { 1044 struct rhlist_head *tmp, *list; 1045 struct nfsd_file *nf; 1046 1047 list = rhltable_lookup(&nfsd_file_rhltable, &inode, 1048 nfsd_file_rhash_params); 1049 rhl_for_each_entry_rcu(nf, tmp, list, nf_rlist) { 1050 if (nf->nf_may != need) 1051 continue; 1052 if (nf->nf_net != net) 1053 continue; 1054 if (!nfsd_match_cred(nf->nf_cred, cred)) 1055 continue; 1056 if (test_bit(NFSD_FILE_GC, &nf->nf_flags) != want_gc) 1057 continue; 1058 if (test_bit(NFSD_FILE_HASHED, &nf->nf_flags) == 0) 1059 continue; 1060 1061 if (!nfsd_file_get(nf)) 1062 continue; 1063 return nf; 1064 } 1065 return NULL; 1066 } 1067 1068 /** 1069 * nfsd_file_is_cached - are there any cached open files for this inode? 1070 * @inode: inode to check 1071 * 1072 * The lookup matches inodes in all net namespaces and is atomic wrt 1073 * nfsd_file_acquire(). 1074 * 1075 * Return values: 1076 * %true: filecache contains at least one file matching this inode 1077 * %false: filecache contains no files matching this inode 1078 */ 1079 bool 1080 nfsd_file_is_cached(struct inode *inode) 1081 { 1082 struct rhlist_head *tmp, *list; 1083 struct nfsd_file *nf; 1084 bool ret = false; 1085 1086 rcu_read_lock(); 1087 list = rhltable_lookup(&nfsd_file_rhltable, &inode, 1088 nfsd_file_rhash_params); 1089 rhl_for_each_entry_rcu(nf, tmp, list, nf_rlist) 1090 if (test_bit(NFSD_FILE_GC, &nf->nf_flags)) { 1091 ret = true; 1092 break; 1093 } 1094 rcu_read_unlock(); 1095 1096 trace_nfsd_file_is_cached(inode, (int)ret); 1097 return ret; 1098 } 1099 1100 static __be32 1101 nfsd_file_get_dio_attrs(const struct svc_fh *fhp, struct nfsd_file *nf) 1102 { 1103 struct inode *inode = file_inode(nf->nf_file); 1104 struct kstat stat; 1105 __be32 status; 1106 1107 /* Currently only need to get DIO alignment info for regular files */ 1108 if (!S_ISREG(inode->i_mode)) 1109 return nfs_ok; 1110 1111 status = fh_getattr(fhp, &stat); 1112 if (status != nfs_ok) 1113 return status; 1114 1115 trace_nfsd_file_get_dio_attrs(inode, &stat); 1116 1117 if (stat.result_mask & STATX_DIOALIGN) { 1118 nf->nf_dio_mem_align = stat.dio_mem_align; 1119 nf->nf_dio_offset_align = stat.dio_offset_align; 1120 } 1121 if (stat.result_mask & STATX_DIO_READ_ALIGN) 1122 nf->nf_dio_read_offset_align = stat.dio_read_offset_align; 1123 else 1124 nf->nf_dio_read_offset_align = nf->nf_dio_offset_align; 1125 1126 return nfs_ok; 1127 } 1128 1129 static __be32 1130 nfsd_file_do_acquire(struct svc_rqst *rqstp, struct net *net, 1131 struct svc_cred *cred, 1132 struct auth_domain *client, 1133 struct svc_fh *fhp, 1134 unsigned int may_flags, struct file *file, 1135 umode_t type, bool want_gc, struct nfsd_file **pnf) 1136 { 1137 unsigned char need = may_flags & NFSD_FILE_MAY_MASK; 1138 struct nfsd_file *new, *nf; 1139 bool stale_retry = true; 1140 bool open_retry = true; 1141 struct inode *inode; 1142 __be32 status; 1143 int ret; 1144 1145 retry: 1146 if (rqstp) 1147 status = fh_verify(rqstp, fhp, type, 1148 may_flags|NFSD_MAY_OWNER_OVERRIDE); 1149 else 1150 status = fh_verify_local(net, cred, client, fhp, type, 1151 may_flags|NFSD_MAY_OWNER_OVERRIDE); 1152 1153 if (status != nfs_ok) 1154 return status; 1155 inode = d_inode(fhp->fh_dentry); 1156 1157 rcu_read_lock(); 1158 nf = nfsd_file_lookup_locked(net, current_cred(), inode, need, want_gc); 1159 rcu_read_unlock(); 1160 1161 if (nf) 1162 goto wait_for_construction; 1163 1164 new = nfsd_file_alloc(net, inode, need, want_gc); 1165 if (!new) { 1166 status = nfserr_jukebox; 1167 goto out; 1168 } 1169 1170 rcu_read_lock(); 1171 spin_lock(&inode->i_lock); 1172 nf = nfsd_file_lookup_locked(net, current_cred(), inode, need, want_gc); 1173 if (unlikely(nf)) { 1174 spin_unlock(&inode->i_lock); 1175 rcu_read_unlock(); 1176 nfsd_file_free(new); 1177 goto wait_for_construction; 1178 } 1179 nf = new; 1180 ret = rhltable_insert(&nfsd_file_rhltable, &nf->nf_rlist, 1181 nfsd_file_rhash_params); 1182 spin_unlock(&inode->i_lock); 1183 rcu_read_unlock(); 1184 if (likely(ret == 0)) 1185 goto open_file; 1186 1187 trace_nfsd_file_insert_err(rqstp, inode, may_flags, ret); 1188 status = nfserr_jukebox; 1189 goto construction_err; 1190 1191 wait_for_construction: 1192 wait_on_bit(&nf->nf_flags, NFSD_FILE_PENDING, TASK_UNINTERRUPTIBLE); 1193 1194 /* Did construction of this file fail? */ 1195 if (!test_bit(NFSD_FILE_HASHED, &nf->nf_flags)) { 1196 trace_nfsd_file_cons_err(rqstp, inode, may_flags, nf); 1197 if (!open_retry) { 1198 status = nfserr_jukebox; 1199 goto construction_err; 1200 } 1201 nfsd_file_put(nf); 1202 open_retry = false; 1203 fh_put(fhp); 1204 goto retry; 1205 } 1206 this_cpu_inc(nfsd_file_cache_hits); 1207 1208 status = nfserrno(nfsd_open_break_lease(file_inode(nf->nf_file), may_flags)); 1209 if (status != nfs_ok) { 1210 nfsd_file_put(nf); 1211 nf = NULL; 1212 } 1213 1214 out: 1215 if (status == nfs_ok) { 1216 this_cpu_inc(nfsd_file_acquisitions); 1217 nfsd_file_check_write_error(nf); 1218 *pnf = nf; 1219 } 1220 trace_nfsd_file_acquire(rqstp, inode, may_flags, nf, status); 1221 return status; 1222 1223 open_file: 1224 trace_nfsd_file_alloc(nf); 1225 1226 if (type == S_IFREG) 1227 nf->nf_mark = nfsd_file_mark_find_or_create(inode); 1228 1229 if (type != S_IFREG || nf->nf_mark) { 1230 if (file) { 1231 get_file(file); 1232 nf->nf_file = file; 1233 status = nfs_ok; 1234 trace_nfsd_file_opened(nf, status); 1235 } else { 1236 ret = nfsd_open_verified(fhp, type, may_flags, &nf->nf_file); 1237 if (ret == -EOPENSTALE && stale_retry) { 1238 stale_retry = false; 1239 nfsd_file_unhash(nf); 1240 clear_and_wake_up_bit(NFSD_FILE_PENDING, 1241 &nf->nf_flags); 1242 if (refcount_dec_and_test(&nf->nf_ref)) 1243 nfsd_file_free(nf); 1244 nf = NULL; 1245 fh_put(fhp); 1246 goto retry; 1247 } 1248 status = nfserrno(ret); 1249 trace_nfsd_file_open(nf, status); 1250 if (status == nfs_ok) 1251 status = nfsd_file_get_dio_attrs(fhp, nf); 1252 } 1253 } else 1254 status = nfserr_jukebox; 1255 /* 1256 * If construction failed, or we raced with a call to unlink() 1257 * then unhash. 1258 */ 1259 if (status != nfs_ok || inode->i_nlink == 0) 1260 nfsd_file_unhash(nf); 1261 else if (want_gc) 1262 nfsd_file_lru_add(nf); 1263 1264 clear_and_wake_up_bit(NFSD_FILE_PENDING, &nf->nf_flags); 1265 if (status == nfs_ok) 1266 goto out; 1267 1268 construction_err: 1269 if (refcount_dec_and_test(&nf->nf_ref)) 1270 nfsd_file_free(nf); 1271 nf = NULL; 1272 goto out; 1273 } 1274 1275 /** 1276 * nfsd_file_acquire_gc - Get a struct nfsd_file with an open file 1277 * @rqstp: the RPC transaction being executed 1278 * @fhp: the NFS filehandle of the file to be opened 1279 * @may_flags: NFSD_MAY_ settings for the file 1280 * @pnf: OUT: new or found "struct nfsd_file" object 1281 * 1282 * The nfsd_file object returned by this API is reference-counted 1283 * and garbage-collected. The object is retained for a few 1284 * seconds after the final nfsd_file_put() in case the caller 1285 * wants to re-use it. 1286 * 1287 * Return values: 1288 * %nfs_ok - @pnf points to an nfsd_file with its reference 1289 * count boosted. 1290 * 1291 * On error, an nfsstat value in network byte order is returned. 1292 */ 1293 __be32 1294 nfsd_file_acquire_gc(struct svc_rqst *rqstp, struct svc_fh *fhp, 1295 unsigned int may_flags, struct nfsd_file **pnf) 1296 { 1297 return nfsd_file_do_acquire(rqstp, SVC_NET(rqstp), NULL, NULL, 1298 fhp, may_flags, NULL, S_IFREG, true, pnf); 1299 } 1300 1301 /** 1302 * nfsd_file_acquire - Get a struct nfsd_file with an open file 1303 * @rqstp: the RPC transaction being executed 1304 * @fhp: the NFS filehandle of the file to be opened 1305 * @may_flags: NFSD_MAY_ settings for the file 1306 * @pnf: OUT: new or found "struct nfsd_file" object 1307 * 1308 * The nfsd_file_object returned by this API is reference-counted 1309 * but not garbage-collected. The object is unhashed after the 1310 * final nfsd_file_put(). 1311 * 1312 * Return values: 1313 * %nfs_ok - @pnf points to an nfsd_file with its reference 1314 * count boosted. 1315 * 1316 * On error, an nfsstat value in network byte order is returned. 1317 */ 1318 __be32 1319 nfsd_file_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp, 1320 unsigned int may_flags, struct nfsd_file **pnf) 1321 { 1322 return nfsd_file_do_acquire(rqstp, SVC_NET(rqstp), NULL, NULL, 1323 fhp, may_flags, NULL, S_IFREG, false, pnf); 1324 } 1325 1326 /** 1327 * nfsd_file_acquire_local - Get a struct nfsd_file with an open file for localio 1328 * @net: The network namespace in which to perform a lookup 1329 * @cred: the user credential with which to validate access 1330 * @client: the auth_domain for LOCALIO lookup 1331 * @fhp: the NFS filehandle of the file to be opened 1332 * @may_flags: NFSD_MAY_ settings for the file 1333 * @pnf: OUT: new or found "struct nfsd_file" object 1334 * 1335 * This file lookup interface provide access to a file given the 1336 * filehandle and credential. No connection-based authorisation 1337 * is performed and in that way it is quite different to other 1338 * file access mediated by nfsd. It allows a kernel module such as the NFS 1339 * client to reach across network and filesystem namespaces to access 1340 * a file. The security implications of this should be carefully 1341 * considered before use. 1342 * 1343 * The nfsd_file_object returned by this API is reference-counted 1344 * but not garbage-collected. The object is unhashed after the 1345 * final nfsd_file_put(). 1346 * 1347 * Return values: 1348 * %nfs_ok - @pnf points to an nfsd_file with its reference 1349 * count boosted. 1350 * 1351 * On error, an nfsstat value in network byte order is returned. 1352 */ 1353 __be32 1354 nfsd_file_acquire_local(struct net *net, struct svc_cred *cred, 1355 struct auth_domain *client, struct svc_fh *fhp, 1356 unsigned int may_flags, struct nfsd_file **pnf) 1357 { 1358 /* 1359 * Save creds before calling nfsd_file_do_acquire() (which calls 1360 * nfsd_setuser). Important because caller (LOCALIO) is from 1361 * client context. 1362 */ 1363 const struct cred *save_cred = get_current_cred(); 1364 __be32 beres; 1365 1366 beres = nfsd_file_do_acquire(NULL, net, cred, client, fhp, may_flags, 1367 NULL, S_IFREG, false, pnf); 1368 put_cred(revert_creds(save_cred)); 1369 return beres; 1370 } 1371 1372 /** 1373 * nfsd_file_acquire_opened - Get a struct nfsd_file using existing open file 1374 * @rqstp: the RPC transaction being executed 1375 * @fhp: the NFS filehandle of the file just created 1376 * @may_flags: NFSD_MAY_ settings for the file 1377 * @file: cached, already-open file (may be NULL) 1378 * @pnf: OUT: new or found "struct nfsd_file" object 1379 * 1380 * Acquire a nfsd_file object that is not GC'ed. If one doesn't already exist, 1381 * and @file is non-NULL, use it to instantiate a new nfsd_file instead of 1382 * opening a new one. 1383 * 1384 * Return values: 1385 * %nfs_ok - @pnf points to an nfsd_file with its reference 1386 * count boosted. 1387 * 1388 * On error, an nfsstat value in network byte order is returned. 1389 */ 1390 __be32 1391 nfsd_file_acquire_opened(struct svc_rqst *rqstp, struct svc_fh *fhp, 1392 unsigned int may_flags, struct file *file, 1393 struct nfsd_file **pnf) 1394 { 1395 return nfsd_file_do_acquire(rqstp, SVC_NET(rqstp), NULL, NULL, 1396 fhp, may_flags, file, S_IFREG, false, pnf); 1397 } 1398 1399 /** 1400 * nfsd_file_acquire_dir - Get a struct nfsd_file with an open directory 1401 * @rqstp: the RPC transaction being executed 1402 * @fhp: the NFS filehandle of the file to be opened 1403 * @pnf: OUT: new or found "struct nfsd_file" object 1404 * 1405 * The nfsd_file_object returned by this API is reference-counted 1406 * but not garbage-collected. The object is unhashed after the 1407 * final nfsd_file_put(). This opens directories only, and only 1408 * in O_RDONLY mode. 1409 * 1410 * Return values: 1411 * %nfs_ok - @pnf points to an nfsd_file with its reference 1412 * count boosted. 1413 * 1414 * On error, an nfsstat value in network byte order is returned. 1415 */ 1416 __be32 1417 nfsd_file_acquire_dir(struct svc_rqst *rqstp, struct svc_fh *fhp, 1418 struct nfsd_file **pnf) 1419 { 1420 return nfsd_file_do_acquire(rqstp, SVC_NET(rqstp), NULL, NULL, fhp, 1421 NFSD_MAY_READ|NFSD_MAY_64BIT_COOKIE, 1422 NULL, S_IFDIR, false, pnf); 1423 } 1424 1425 /* 1426 * Note that fields may be added, removed or reordered in the future. Programs 1427 * scraping this file for info should test the labels to ensure they're 1428 * getting the correct field. 1429 */ 1430 int nfsd_file_cache_stats_show(struct seq_file *m, void *v) 1431 { 1432 unsigned long allocations = 0, releases = 0, evictions = 0; 1433 unsigned long hits = 0, acquisitions = 0; 1434 unsigned int i, count = 0, buckets = 0; 1435 unsigned long lru = 0, total_age = 0; 1436 1437 /* Serialize with server shutdown */ 1438 mutex_lock(&nfsd_mutex); 1439 if (test_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags) == 1) { 1440 struct bucket_table *tbl; 1441 struct rhashtable *ht; 1442 1443 lru = list_lru_count(&nfsd_file_lru); 1444 1445 rcu_read_lock(); 1446 ht = &nfsd_file_rhltable.ht; 1447 count = atomic_read(&ht->nelems); 1448 tbl = rht_dereference_rcu(ht->tbl, ht); 1449 buckets = tbl->size; 1450 rcu_read_unlock(); 1451 } 1452 mutex_unlock(&nfsd_mutex); 1453 1454 for_each_possible_cpu(i) { 1455 hits += per_cpu(nfsd_file_cache_hits, i); 1456 acquisitions += per_cpu(nfsd_file_acquisitions, i); 1457 allocations += per_cpu(nfsd_file_allocations, i); 1458 releases += per_cpu(nfsd_file_releases, i); 1459 total_age += per_cpu(nfsd_file_total_age, i); 1460 evictions += per_cpu(nfsd_file_evictions, i); 1461 } 1462 1463 seq_printf(m, "total inodes: %u\n", count); 1464 seq_printf(m, "hash buckets: %u\n", buckets); 1465 seq_printf(m, "lru entries: %lu\n", lru); 1466 seq_printf(m, "cache hits: %lu\n", hits); 1467 seq_printf(m, "acquisitions: %lu\n", acquisitions); 1468 seq_printf(m, "allocations: %lu\n", allocations); 1469 seq_printf(m, "releases: %lu\n", releases); 1470 seq_printf(m, "evictions: %lu\n", evictions); 1471 if (releases) 1472 seq_printf(m, "mean age (ms): %ld\n", total_age / releases); 1473 else 1474 seq_printf(m, "mean age (ms): -\n"); 1475 return 0; 1476 } 1477