1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * linux/fs/nfs/inode.c 4 * 5 * Copyright (C) 1992 Rick Sladkey 6 * 7 * nfs inode and superblock handling functions 8 * 9 * Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some 10 * experimental NFS changes. Modularisation taken straight from SYS5 fs. 11 * 12 * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts. 13 * J.S.Peatfield@damtp.cam.ac.uk 14 * 15 */ 16 17 #include <linux/module.h> 18 #include <linux/init.h> 19 #include <linux/sched/signal.h> 20 #include <linux/time.h> 21 #include <linux/kernel.h> 22 #include <linux/mm.h> 23 #include <linux/string.h> 24 #include <linux/stat.h> 25 #include <linux/errno.h> 26 #include <linux/unistd.h> 27 #include <linux/sunrpc/clnt.h> 28 #include <linux/sunrpc/stats.h> 29 #include <linux/sunrpc/metrics.h> 30 #include <linux/nfs_fs.h> 31 #include <linux/nfs_mount.h> 32 #include <linux/nfs4_mount.h> 33 #include <linux/lockd/bind.h> 34 #include <linux/seq_file.h> 35 #include <linux/mount.h> 36 #include <linux/vfs.h> 37 #include <linux/inet.h> 38 #include <linux/nfs_xdr.h> 39 #include <linux/slab.h> 40 #include <linux/compat.h> 41 #include <linux/freezer.h> 42 #include <linux/uaccess.h> 43 #include <linux/iversion.h> 44 45 #include "nfs4_fs.h" 46 #include "callback.h" 47 #include "delegation.h" 48 #include "iostat.h" 49 #include "internal.h" 50 #include "fscache.h" 51 #include "pnfs.h" 52 #include "nfs.h" 53 #include "netns.h" 54 #include "sysfs.h" 55 56 #include "nfstrace.h" 57 58 #define NFSDBG_FACILITY NFSDBG_VFS 59 60 #define NFS_64_BIT_INODE_NUMBERS_ENABLED 1 61 62 /* Default is to see 64-bit inode numbers */ 63 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED; 64 65 static int nfs_update_inode(struct inode *, struct nfs_fattr *); 66 67 static struct kmem_cache * nfs_inode_cachep; 68 69 static inline unsigned long 70 nfs_fattr_to_ino_t(struct nfs_fattr *fattr) 71 { 72 return nfs_fileid_to_ino_t(fattr->fileid); 73 } 74 75 int nfs_wait_bit_killable(struct wait_bit_key *key, int mode) 76 { 77 if (unlikely(nfs_current_task_exiting())) 78 return -EINTR; 79 schedule(); 80 if (signal_pending_state(mode, current)) 81 return -ERESTARTSYS; 82 return 0; 83 } 84 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable); 85 86 /** 87 * nfs_compat_user_ino64 - returns the user-visible inode number 88 * @fileid: 64-bit fileid 89 * 90 * This function returns a 32-bit inode number if the boot parameter 91 * nfs.enable_ino64 is zero. 92 */ 93 u64 nfs_compat_user_ino64(u64 fileid) 94 { 95 #ifdef CONFIG_COMPAT 96 compat_ulong_t ino; 97 #else 98 unsigned long ino; 99 #endif 100 101 if (enable_ino64) 102 return fileid; 103 ino = fileid; 104 if (sizeof(ino) < sizeof(fileid)) 105 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8; 106 return ino; 107 } 108 109 int nfs_drop_inode(struct inode *inode) 110 { 111 return NFS_STALE(inode) || inode_generic_drop(inode); 112 } 113 EXPORT_SYMBOL_GPL(nfs_drop_inode); 114 115 void nfs_clear_inode(struct inode *inode) 116 { 117 /* 118 * The following should never happen... 119 */ 120 WARN_ON_ONCE(nfs_have_writebacks(inode)); 121 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files)); 122 nfs_zap_acl_cache(inode); 123 nfs_access_zap_cache(inode); 124 nfs_fscache_clear_inode(inode); 125 } 126 EXPORT_SYMBOL_GPL(nfs_clear_inode); 127 128 void nfs_evict_inode(struct inode *inode) 129 { 130 truncate_inode_pages_final(&inode->i_data); 131 clear_inode(inode); 132 nfs_clear_inode(inode); 133 } 134 135 int nfs_sync_inode(struct inode *inode) 136 { 137 inode_dio_wait(inode); 138 return nfs_wb_all(inode); 139 } 140 EXPORT_SYMBOL_GPL(nfs_sync_inode); 141 142 /** 143 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk 144 * @mapping: pointer to struct address_space 145 */ 146 int nfs_sync_mapping(struct address_space *mapping) 147 { 148 int ret = 0; 149 150 if (mapping->nrpages != 0) { 151 unmap_mapping_range(mapping, 0, 0, 0); 152 ret = nfs_wb_all(mapping->host); 153 } 154 return ret; 155 } 156 157 static int nfs_attribute_timeout(struct inode *inode) 158 { 159 struct nfs_inode *nfsi = NFS_I(inode); 160 161 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo); 162 } 163 164 static bool nfs_check_cache_flags_invalid(struct inode *inode, 165 unsigned long flags) 166 { 167 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity); 168 169 return (cache_validity & flags) != 0; 170 } 171 172 bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags) 173 { 174 if (nfs_check_cache_flags_invalid(inode, flags)) 175 return true; 176 return nfs_attribute_cache_expired(inode); 177 } 178 EXPORT_SYMBOL_GPL(nfs_check_cache_invalid); 179 180 #ifdef CONFIG_NFS_V4_2 181 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi) 182 { 183 return nfsi->xattr_cache != NULL; 184 } 185 #else 186 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi) 187 { 188 return false; 189 } 190 #endif 191 192 void nfs_set_cache_invalid(struct inode *inode, unsigned long flags) 193 { 194 struct nfs_inode *nfsi = NFS_I(inode); 195 196 if (nfs_have_delegated_attributes(inode)) { 197 if (!(flags & NFS_INO_REVAL_FORCED)) 198 flags &= ~(NFS_INO_INVALID_MODE | 199 NFS_INO_INVALID_OTHER | 200 NFS_INO_INVALID_BTIME | 201 NFS_INO_INVALID_XATTR); 202 flags &= ~(NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE); 203 } 204 205 if (!nfs_has_xattr_cache(nfsi)) 206 flags &= ~NFS_INO_INVALID_XATTR; 207 if (flags & NFS_INO_INVALID_DATA) 208 nfs_fscache_invalidate(inode, 0); 209 flags &= ~NFS_INO_REVAL_FORCED; 210 211 flags |= nfsi->cache_validity; 212 if (inode->i_mapping->nrpages == 0) 213 flags &= ~NFS_INO_INVALID_DATA; 214 215 /* pairs with nfs_clear_invalid_mapping()'s smp_load_acquire() */ 216 smp_store_release(&nfsi->cache_validity, flags); 217 218 if (inode->i_mapping->nrpages == 0 || 219 nfsi->cache_validity & NFS_INO_INVALID_DATA) { 220 nfs_ooo_clear(nfsi); 221 } 222 trace_nfs_set_cache_invalid(inode, 0); 223 } 224 EXPORT_SYMBOL_GPL(nfs_set_cache_invalid); 225 226 /* 227 * Invalidate the local caches 228 */ 229 static void nfs_zap_caches_locked(struct inode *inode) 230 { 231 struct nfs_inode *nfsi = NFS_I(inode); 232 int mode = inode->i_mode; 233 234 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 235 236 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 237 nfsi->attrtimeo_timestamp = jiffies; 238 239 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) 240 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR | 241 NFS_INO_INVALID_DATA | 242 NFS_INO_INVALID_ACCESS | 243 NFS_INO_INVALID_ACL | 244 NFS_INO_INVALID_XATTR); 245 else 246 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR | 247 NFS_INO_INVALID_ACCESS | 248 NFS_INO_INVALID_ACL | 249 NFS_INO_INVALID_XATTR); 250 nfs_zap_label_cache_locked(nfsi); 251 } 252 253 void nfs_zap_caches(struct inode *inode) 254 { 255 spin_lock(&inode->i_lock); 256 nfs_zap_caches_locked(inode); 257 spin_unlock(&inode->i_lock); 258 } 259 260 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping) 261 { 262 if (mapping->nrpages != 0) { 263 spin_lock(&inode->i_lock); 264 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA); 265 spin_unlock(&inode->i_lock); 266 } 267 } 268 269 void nfs_zap_acl_cache(struct inode *inode) 270 { 271 void (*clear_acl_cache)(struct inode *); 272 273 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache; 274 if (clear_acl_cache != NULL) 275 clear_acl_cache(inode); 276 spin_lock(&inode->i_lock); 277 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL; 278 spin_unlock(&inode->i_lock); 279 } 280 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache); 281 282 void nfs_invalidate_atime(struct inode *inode) 283 { 284 if (nfs_have_delegated_atime(inode)) 285 return; 286 spin_lock(&inode->i_lock); 287 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME); 288 spin_unlock(&inode->i_lock); 289 } 290 EXPORT_SYMBOL_GPL(nfs_invalidate_atime); 291 292 /* 293 * Invalidate, but do not unhash, the inode. 294 * NB: must be called with inode->i_lock held! 295 */ 296 static void nfs_set_inode_stale_locked(struct inode *inode) 297 { 298 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); 299 nfs_zap_caches_locked(inode); 300 trace_nfs_set_inode_stale(inode); 301 } 302 303 void nfs_set_inode_stale(struct inode *inode) 304 { 305 spin_lock(&inode->i_lock); 306 nfs_set_inode_stale_locked(inode); 307 spin_unlock(&inode->i_lock); 308 } 309 310 struct nfs_find_desc { 311 struct nfs_fh *fh; 312 struct nfs_fattr *fattr; 313 }; 314 315 /* 316 * In NFSv3 we can have 64bit inode numbers. In order to support 317 * this, and re-exported directories (also seen in NFSv2) 318 * we are forced to allow 2 different inodes to have the same 319 * i_ino. 320 */ 321 static int 322 nfs_find_actor(struct inode *inode, void *opaque) 323 { 324 struct nfs_find_desc *desc = opaque; 325 struct nfs_fh *fh = desc->fh; 326 struct nfs_fattr *fattr = desc->fattr; 327 328 if (NFS_FILEID(inode) != fattr->fileid) 329 return 0; 330 if (inode_wrong_type(inode, fattr->mode)) 331 return 0; 332 if (nfs_compare_fh(NFS_FH(inode), fh)) 333 return 0; 334 if (is_bad_inode(inode) || NFS_STALE(inode)) 335 return 0; 336 return 1; 337 } 338 339 static int 340 nfs_init_locked(struct inode *inode, void *opaque) 341 { 342 struct nfs_find_desc *desc = opaque; 343 struct nfs_fattr *fattr = desc->fattr; 344 345 set_nfs_fileid(inode, fattr->fileid); 346 inode->i_mode = fattr->mode; 347 nfs_copy_fh(NFS_FH(inode), desc->fh); 348 return 0; 349 } 350 351 #ifdef CONFIG_NFS_V4_SECURITY_LABEL 352 static void nfs_clear_label_invalid(struct inode *inode) 353 { 354 spin_lock(&inode->i_lock); 355 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL; 356 spin_unlock(&inode->i_lock); 357 } 358 359 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr) 360 { 361 int error; 362 363 if (fattr->label == NULL) 364 return; 365 366 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) { 367 error = security_inode_notifysecctx(inode, fattr->label->label, 368 fattr->label->len); 369 if (error) 370 printk(KERN_ERR "%s() %s %d " 371 "security_inode_notifysecctx() %d\n", 372 __func__, 373 (char *)fattr->label->label, 374 fattr->label->len, error); 375 nfs_clear_label_invalid(inode); 376 } 377 } 378 379 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags) 380 { 381 struct nfs4_label *label; 382 383 if (!(server->caps & NFS_CAP_SECURITY_LABEL)) 384 return NULL; 385 386 label = kzalloc(sizeof(struct nfs4_label), flags); 387 if (label == NULL) 388 return ERR_PTR(-ENOMEM); 389 390 label->label = kzalloc(NFS4_MAXLABELLEN, flags); 391 if (label->label == NULL) { 392 kfree(label); 393 return ERR_PTR(-ENOMEM); 394 } 395 label->len = NFS4_MAXLABELLEN; 396 397 return label; 398 } 399 EXPORT_SYMBOL_GPL(nfs4_label_alloc); 400 #else 401 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr) 402 { 403 } 404 #endif 405 EXPORT_SYMBOL_GPL(nfs_setsecurity); 406 407 /* Search for inode identified by fh, fileid and i_mode in inode cache. */ 408 struct inode * 409 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh) 410 { 411 struct nfs_find_desc desc = { 412 .fh = fh, 413 .fattr = fattr, 414 }; 415 struct inode *inode; 416 unsigned long hash; 417 418 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) || 419 !(fattr->valid & NFS_ATTR_FATTR_TYPE)) 420 return NULL; 421 422 hash = nfs_fattr_to_ino_t(fattr); 423 inode = ilookup5(sb, hash, nfs_find_actor, &desc); 424 425 dprintk("%s: returning %p\n", __func__, inode); 426 return inode; 427 } 428 429 static void nfs_inode_init_regular(struct nfs_inode *nfsi) 430 { 431 atomic_long_set(&nfsi->nrequests, 0); 432 atomic_long_set(&nfsi->redirtied_pages, 0); 433 INIT_LIST_HEAD(&nfsi->commit_info.list); 434 atomic_long_set(&nfsi->commit_info.ncommit, 0); 435 atomic_set(&nfsi->commit_info.rpcs_out, 0); 436 mutex_init(&nfsi->commit_mutex); 437 } 438 439 static void nfs_inode_init_dir(struct nfs_inode *nfsi) 440 { 441 nfsi->cache_change_attribute = 0; 442 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf)); 443 init_rwsem(&nfsi->rmdir_sem); 444 } 445 446 /* 447 * This is our front-end to iget that looks up inodes by file handle 448 * instead of inode number. 449 */ 450 struct inode * 451 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr) 452 { 453 struct nfs_find_desc desc = { 454 .fh = fh, 455 .fattr = fattr 456 }; 457 struct inode *inode = ERR_PTR(-ENOENT); 458 u64 fattr_supported = NFS_SB(sb)->fattr_valid; 459 unsigned long hash; 460 461 nfs_attr_check_mountpoint(sb, fattr); 462 463 if (nfs_attr_use_mounted_on_fileid(fattr)) 464 fattr->fileid = fattr->mounted_on_fileid; 465 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0) 466 goto out_no_inode; 467 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0) 468 goto out_no_inode; 469 470 hash = nfs_fattr_to_ino_t(fattr); 471 472 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc); 473 if (inode == NULL) { 474 inode = ERR_PTR(-ENOMEM); 475 goto out_no_inode; 476 } 477 478 if (inode_state_read_once(inode) & I_NEW) { 479 struct nfs_inode *nfsi = NFS_I(inode); 480 unsigned long now = jiffies; 481 482 /* We set i_ino for the few things that still rely on it, 483 * such as stat(2) */ 484 inode->i_ino = hash; 485 486 /* We can't support update_atime(), since the server will reset it */ 487 inode->i_flags |= S_NOATIME|S_NOCMTIME; 488 inode->i_mode = fattr->mode; 489 nfsi->cache_validity = 0; 490 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0 491 && (fattr_supported & NFS_ATTR_FATTR_MODE)) 492 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE); 493 /* Why so? Because we want revalidate for devices/FIFOs, and 494 * that's precisely what we have in nfs_file_inode_operations. 495 */ 496 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops; 497 if (S_ISREG(inode->i_mode)) { 498 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops; 499 inode->i_data.a_ops = &nfs_file_aops; 500 nfs_inode_init_regular(nfsi); 501 mapping_set_large_folios(inode->i_mapping); 502 } else if (S_ISDIR(inode->i_mode)) { 503 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops; 504 inode->i_fop = &nfs_dir_operations; 505 inode->i_data.a_ops = &nfs_dir_aops; 506 nfs_inode_init_dir(nfsi); 507 /* Deal with crossing mountpoints */ 508 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT || 509 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) { 510 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) 511 inode->i_op = &nfs_referral_inode_operations; 512 else 513 inode->i_op = &nfs_mountpoint_inode_operations; 514 inode->i_fop = NULL; 515 inode->i_flags |= S_AUTOMOUNT; 516 } 517 } else if (S_ISLNK(inode->i_mode)) { 518 inode->i_op = &nfs_symlink_inode_operations; 519 inode_nohighmem(inode); 520 } else 521 init_special_inode(inode, inode->i_mode, fattr->rdev); 522 523 inode_set_atime(inode, 0, 0); 524 inode_set_mtime(inode, 0, 0); 525 inode_set_ctime(inode, 0, 0); 526 memset(&nfsi->btime, 0, sizeof(nfsi->btime)); 527 inode_set_iversion_raw(inode, 0); 528 inode->i_size = 0; 529 clear_nlink(inode); 530 inode->i_uid = make_kuid(&init_user_ns, -2); 531 inode->i_gid = make_kgid(&init_user_ns, -2); 532 inode->i_blocks = 0; 533 nfsi->write_io = 0; 534 nfsi->read_io = 0; 535 536 nfsi->read_cache_jiffies = fattr->time_start; 537 nfsi->attr_gencount = fattr->gencount; 538 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 539 inode_set_atime_to_ts(inode, fattr->atime); 540 else if (fattr_supported & NFS_ATTR_FATTR_ATIME) 541 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME); 542 if (fattr->valid & NFS_ATTR_FATTR_MTIME) 543 inode_set_mtime_to_ts(inode, fattr->mtime); 544 else if (fattr_supported & NFS_ATTR_FATTR_MTIME) 545 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME); 546 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 547 inode_set_ctime_to_ts(inode, fattr->ctime); 548 else if (fattr_supported & NFS_ATTR_FATTR_CTIME) 549 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME); 550 if (fattr->valid & NFS_ATTR_FATTR_BTIME) 551 nfsi->btime = fattr->btime; 552 else if (fattr_supported & NFS_ATTR_FATTR_BTIME) 553 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BTIME); 554 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) 555 inode_set_iversion_raw(inode, fattr->change_attr); 556 else 557 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE); 558 if (fattr->valid & NFS_ATTR_FATTR_SIZE) 559 inode->i_size = nfs_size_to_loff_t(fattr->size); 560 else 561 nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE); 562 if (fattr->valid & NFS_ATTR_FATTR_NLINK) 563 set_nlink(inode, fattr->nlink); 564 else if (fattr_supported & NFS_ATTR_FATTR_NLINK) 565 nfs_set_cache_invalid(inode, NFS_INO_INVALID_NLINK); 566 else 567 set_nlink(inode, 1); 568 if (fattr->valid & NFS_ATTR_FATTR_OWNER) 569 inode->i_uid = fattr->uid; 570 else if (fattr_supported & NFS_ATTR_FATTR_OWNER) 571 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER); 572 if (fattr->valid & NFS_ATTR_FATTR_GROUP) 573 inode->i_gid = fattr->gid; 574 else if (fattr_supported & NFS_ATTR_FATTR_GROUP) 575 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER); 576 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED) 577 inode->i_blocks = fattr->du.nfs2.blocks; 578 else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED && 579 fattr->size != 0) 580 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS); 581 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) { 582 /* 583 * report the blocks in 512byte units 584 */ 585 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 586 } else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED && 587 fattr->size != 0) 588 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS); 589 590 nfs_setsecurity(inode, fattr); 591 592 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 593 nfsi->attrtimeo_timestamp = now; 594 nfsi->access_cache = RB_ROOT; 595 596 nfs_fscache_init_inode(inode); 597 598 unlock_new_inode(inode); 599 } else { 600 int err = nfs_refresh_inode(inode, fattr); 601 if (err < 0) { 602 iput(inode); 603 inode = ERR_PTR(err); 604 goto out_no_inode; 605 } 606 } 607 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n", 608 inode->i_sb->s_id, 609 (unsigned long long)NFS_FILEID(inode), 610 nfs_display_fhandle_hash(fh), 611 icount_read(inode)); 612 613 out: 614 return inode; 615 616 out_no_inode: 617 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode)); 618 goto out; 619 } 620 EXPORT_SYMBOL_GPL(nfs_fhget); 621 622 static void 623 nfs_fattr_fixup_delegated(struct inode *inode, struct nfs_fattr *fattr) 624 { 625 unsigned long cache_validity = NFS_I(inode)->cache_validity; 626 627 if (nfs_have_delegated_mtime(inode)) { 628 if (!(cache_validity & NFS_INO_INVALID_CTIME)) 629 fattr->valid &= ~(NFS_ATTR_FATTR_PRECTIME | 630 NFS_ATTR_FATTR_CTIME); 631 632 if (!(cache_validity & NFS_INO_INVALID_MTIME)) 633 fattr->valid &= ~(NFS_ATTR_FATTR_PREMTIME | 634 NFS_ATTR_FATTR_MTIME); 635 636 if (!(cache_validity & NFS_INO_INVALID_ATIME)) 637 fattr->valid &= ~NFS_ATTR_FATTR_ATIME; 638 } else if (nfs_have_delegated_atime(inode)) { 639 if (!(cache_validity & NFS_INO_INVALID_ATIME)) 640 fattr->valid &= ~NFS_ATTR_FATTR_ATIME; 641 } 642 } 643 644 static void nfs_set_timestamps_to_ts(struct inode *inode, struct iattr *attr) 645 { 646 unsigned int cache_flags = 0; 647 648 if (attr->ia_valid & ATTR_MTIME_SET) { 649 struct timespec64 ctime = inode_get_ctime(inode); 650 struct timespec64 mtime = inode_get_mtime(inode); 651 struct timespec64 now; 652 bool updated = false; 653 654 now = inode_set_ctime_current(inode); 655 if (!timespec64_equal(&now, &ctime)) 656 updated = true; 657 658 inode_set_mtime_to_ts(inode, attr->ia_mtime); 659 if (!timespec64_equal(&now, &mtime)) 660 updated = true; 661 662 inode_maybe_inc_iversion(inode, updated); 663 cache_flags |= NFS_INO_INVALID_CTIME | NFS_INO_INVALID_MTIME; 664 } 665 if (attr->ia_valid & ATTR_ATIME_SET) { 666 inode_set_atime_to_ts(inode, attr->ia_atime); 667 cache_flags |= NFS_INO_INVALID_ATIME; 668 } 669 NFS_I(inode)->cache_validity &= ~cache_flags; 670 } 671 672 static void nfs_update_atime(struct inode *inode) 673 { 674 inode_update_time(inode, FS_UPD_ATIME, 0); 675 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ATIME; 676 } 677 678 static void nfs_update_mtime(struct inode *inode) 679 { 680 inode_update_time(inode, FS_UPD_CMTIME, 0); 681 NFS_I(inode)->cache_validity &= 682 ~(NFS_INO_INVALID_CTIME | NFS_INO_INVALID_MTIME); 683 } 684 685 void nfs_update_delegated_atime(struct inode *inode) 686 { 687 spin_lock(&inode->i_lock); 688 if (nfs_have_delegated_atime(inode)) 689 nfs_update_atime(inode); 690 spin_unlock(&inode->i_lock); 691 } 692 693 void nfs_update_delegated_mtime_locked(struct inode *inode) 694 { 695 if (nfs_have_delegated_mtime(inode)) 696 nfs_update_mtime(inode); 697 } 698 699 void nfs_update_delegated_mtime(struct inode *inode) 700 { 701 spin_lock(&inode->i_lock); 702 nfs_update_delegated_mtime_locked(inode); 703 spin_unlock(&inode->i_lock); 704 } 705 EXPORT_SYMBOL_GPL(nfs_update_delegated_mtime); 706 707 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE|ATTR_OPEN) 708 709 int 710 nfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry, 711 struct iattr *attr) 712 { 713 struct inode *inode = d_inode(dentry); 714 struct nfs_fattr *fattr; 715 loff_t oldsize; 716 int error = 0; 717 kuid_t task_uid = current_fsuid(); 718 kuid_t owner_uid = inode->i_uid; 719 720 nfs_inc_stats(inode, NFSIOS_VFSSETATTR); 721 722 /* skip mode change if it's just for clearing setuid/setgid */ 723 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID)) 724 attr->ia_valid &= ~ATTR_MODE; 725 726 if (S_ISREG(inode->i_mode)) 727 nfs_file_block_o_direct(NFS_I(inode)); 728 729 oldsize = i_size_read(inode); 730 if (attr->ia_valid & ATTR_SIZE) { 731 BUG_ON(!S_ISREG(inode->i_mode)); 732 733 error = inode_newsize_ok(inode, attr->ia_size); 734 if (error) 735 return error; 736 737 if (attr->ia_size == oldsize) 738 attr->ia_valid &= ~ATTR_SIZE; 739 } 740 741 if (nfs_have_delegated_mtime(inode) && attr->ia_valid & ATTR_MTIME) { 742 spin_lock(&inode->i_lock); 743 if (attr->ia_valid & ATTR_MTIME_SET) { 744 if (uid_eq(task_uid, owner_uid)) { 745 nfs_set_timestamps_to_ts(inode, attr); 746 attr->ia_valid &= ~(ATTR_MTIME|ATTR_MTIME_SET| 747 ATTR_ATIME|ATTR_ATIME_SET); 748 } 749 } else { 750 if (attr->ia_valid & ATTR_MTIME) 751 nfs_update_mtime(inode); 752 if (attr->ia_valid & ATTR_ATIME) 753 nfs_update_atime(inode); 754 attr->ia_valid &= ~(ATTR_MTIME|ATTR_ATIME); 755 } 756 spin_unlock(&inode->i_lock); 757 } else if (nfs_have_delegated_atime(inode) && 758 attr->ia_valid & ATTR_ATIME && 759 !(attr->ia_valid & ATTR_MTIME)) { 760 if (attr->ia_valid & ATTR_ATIME_SET) { 761 if (uid_eq(task_uid, owner_uid)) { 762 spin_lock(&inode->i_lock); 763 nfs_set_timestamps_to_ts(inode, attr); 764 spin_unlock(&inode->i_lock); 765 attr->ia_valid &= ~(ATTR_ATIME|ATTR_ATIME_SET); 766 } 767 } else { 768 nfs_update_delegated_atime(inode); 769 attr->ia_valid &= ~ATTR_ATIME; 770 } 771 } 772 773 /* Optimization: if the end result is no change, don't RPC */ 774 if (((attr->ia_valid & NFS_VALID_ATTRS) & ~(ATTR_FILE|ATTR_OPEN)) == 0) 775 return 0; 776 777 trace_nfs_setattr_enter(inode); 778 779 /* Write all dirty data */ 780 if (S_ISREG(inode->i_mode)) 781 nfs_sync_inode(inode); 782 783 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode)); 784 if (fattr == NULL) { 785 error = -ENOMEM; 786 goto out; 787 } 788 789 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr); 790 if (error == 0) { 791 if (attr->ia_valid & ATTR_SIZE) 792 nfs_truncate_last_folio(inode->i_mapping, oldsize, 793 attr->ia_size); 794 error = nfs_refresh_inode(inode, fattr); 795 } 796 nfs_free_fattr(fattr); 797 out: 798 trace_nfs_setattr_exit(inode, error); 799 return error; 800 } 801 EXPORT_SYMBOL_GPL(nfs_setattr); 802 803 /** 804 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall 805 * @inode: inode of the file used 806 * @offset: file offset to start truncating 807 * 808 * This is a copy of the common vmtruncate, but with the locking 809 * corrected to take into account the fact that NFS requires 810 * inode->i_size to be updated under the inode->i_lock. 811 * Note: must be called with inode->i_lock held! 812 */ 813 static int nfs_vmtruncate(struct inode * inode, loff_t offset) 814 { 815 int err; 816 817 err = inode_newsize_ok(inode, offset); 818 if (err) 819 goto out; 820 821 trace_nfs_size_truncate(inode, offset); 822 i_size_write(inode, offset); 823 /* Optimisation */ 824 if (offset == 0) { 825 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA; 826 nfs_ooo_clear(NFS_I(inode)); 827 } 828 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE; 829 830 spin_unlock(&inode->i_lock); 831 truncate_pagecache(inode, offset); 832 nfs_update_delegated_mtime_locked(inode); 833 spin_lock(&inode->i_lock); 834 out: 835 return err; 836 } 837 838 /** 839 * nfs_setattr_update_inode - Update inode metadata after a setattr call. 840 * @inode: pointer to struct inode 841 * @attr: pointer to struct iattr 842 * @fattr: pointer to struct nfs_fattr 843 * 844 * Note: we do this in the *proc.c in order to ensure that 845 * it works for things like exclusive creates too. 846 */ 847 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr, 848 struct nfs_fattr *fattr) 849 { 850 /* Barrier: bump the attribute generation count. */ 851 nfs_fattr_set_barrier(fattr); 852 853 spin_lock(&inode->i_lock); 854 NFS_I(inode)->attr_gencount = fattr->gencount; 855 if ((attr->ia_valid & ATTR_SIZE) != 0) { 856 if (!nfs_have_delegated_mtime(inode)) 857 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME); 858 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS); 859 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC); 860 nfs_vmtruncate(inode, attr->ia_size); 861 } 862 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) { 863 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME; 864 if ((attr->ia_valid & ATTR_KILL_SUID) != 0 && 865 inode->i_mode & S_ISUID) 866 inode->i_mode &= ~S_ISUID; 867 if (setattr_should_drop_sgid(&nop_mnt_idmap, inode)) 868 inode->i_mode &= ~S_ISGID; 869 if ((attr->ia_valid & ATTR_MODE) != 0) { 870 int mode = attr->ia_mode & S_IALLUGO; 871 mode |= inode->i_mode & ~S_IALLUGO; 872 inode->i_mode = mode; 873 } 874 if ((attr->ia_valid & ATTR_UID) != 0) 875 inode->i_uid = attr->ia_uid; 876 if ((attr->ia_valid & ATTR_GID) != 0) 877 inode->i_gid = attr->ia_gid; 878 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 879 inode_set_ctime_to_ts(inode, fattr->ctime); 880 else 881 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE 882 | NFS_INO_INVALID_CTIME); 883 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS 884 | NFS_INO_INVALID_ACL); 885 } 886 if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) { 887 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME 888 | NFS_INO_INVALID_CTIME); 889 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 890 inode_set_atime_to_ts(inode, fattr->atime); 891 else if (attr->ia_valid & ATTR_ATIME_SET) 892 inode_set_atime_to_ts(inode, attr->ia_atime); 893 else 894 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME); 895 896 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 897 inode_set_ctime_to_ts(inode, fattr->ctime); 898 else 899 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE 900 | NFS_INO_INVALID_CTIME); 901 } 902 if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) { 903 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME 904 | NFS_INO_INVALID_CTIME); 905 if (fattr->valid & NFS_ATTR_FATTR_MTIME) 906 inode_set_mtime_to_ts(inode, fattr->mtime); 907 else if (attr->ia_valid & ATTR_MTIME_SET) 908 inode_set_mtime_to_ts(inode, attr->ia_mtime); 909 else 910 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME); 911 912 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 913 inode_set_ctime_to_ts(inode, fattr->ctime); 914 else 915 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE 916 | NFS_INO_INVALID_CTIME); 917 } 918 if (fattr->valid) 919 nfs_update_inode(inode, fattr); 920 spin_unlock(&inode->i_lock); 921 } 922 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode); 923 924 /* 925 * Don't request help from readdirplus if the file is being written to, 926 * or if attribute caching is turned off 927 */ 928 static bool nfs_getattr_readdirplus_enable(const struct inode *inode) 929 { 930 return nfs_server_capable(inode, NFS_CAP_READDIRPLUS) && 931 !nfs_have_writebacks(inode) && NFS_MAXATTRTIMEO(inode) > 5 * HZ; 932 } 933 934 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry) 935 { 936 if (!IS_ROOT(dentry)) { 937 struct dentry *parent = dget_parent(dentry); 938 nfs_readdir_record_entry_cache_miss(d_inode(parent)); 939 dput(parent); 940 } 941 } 942 943 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry) 944 { 945 if (!IS_ROOT(dentry)) { 946 struct dentry *parent = dget_parent(dentry); 947 nfs_readdir_record_entry_cache_hit(d_inode(parent)); 948 dput(parent); 949 } 950 } 951 952 static u32 nfs_get_valid_attrmask(struct inode *inode) 953 { 954 u64 fattr_valid = NFS_SERVER(inode)->fattr_valid; 955 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity); 956 u32 reply_mask = STATX_INO | STATX_TYPE; 957 958 if (!(cache_validity & NFS_INO_INVALID_ATIME)) 959 reply_mask |= STATX_ATIME; 960 if (!(cache_validity & NFS_INO_INVALID_CTIME)) 961 reply_mask |= STATX_CTIME; 962 if (!(cache_validity & NFS_INO_INVALID_MTIME)) 963 reply_mask |= STATX_MTIME; 964 if (!(cache_validity & NFS_INO_INVALID_SIZE)) 965 reply_mask |= STATX_SIZE; 966 if (!(cache_validity & NFS_INO_INVALID_NLINK)) 967 reply_mask |= STATX_NLINK; 968 if (!(cache_validity & NFS_INO_INVALID_MODE)) 969 reply_mask |= STATX_MODE; 970 if (!(cache_validity & NFS_INO_INVALID_OTHER)) 971 reply_mask |= STATX_UID | STATX_GID; 972 if (!(cache_validity & NFS_INO_INVALID_BLOCKS)) 973 reply_mask |= STATX_BLOCKS; 974 if (!(cache_validity & NFS_INO_INVALID_BTIME) && 975 (fattr_valid & NFS_ATTR_FATTR_BTIME)) 976 reply_mask |= STATX_BTIME; 977 if (!(cache_validity & NFS_INO_INVALID_CHANGE)) 978 reply_mask |= STATX_CHANGE_COOKIE; 979 return reply_mask; 980 } 981 982 int nfs_getattr(struct mnt_idmap *idmap, const struct path *path, 983 struct kstat *stat, u32 request_mask, unsigned int query_flags) 984 { 985 struct inode *inode = d_inode(path->dentry); 986 struct nfs_server *server = NFS_SERVER(inode); 987 u64 fattr_valid = server->fattr_valid; 988 unsigned long cache_validity; 989 int err = 0; 990 bool force_sync = query_flags & AT_STATX_FORCE_SYNC; 991 bool do_update = false; 992 bool readdirplus_enabled = nfs_getattr_readdirplus_enable(inode); 993 994 trace_nfs_getattr_enter(inode); 995 996 request_mask &= STATX_TYPE | STATX_MODE | STATX_NLINK | STATX_UID | 997 STATX_GID | STATX_ATIME | STATX_MTIME | STATX_CTIME | 998 STATX_INO | STATX_SIZE | STATX_BLOCKS | STATX_BTIME | 999 STATX_CHANGE_COOKIE; 1000 1001 if (!(fattr_valid & NFS_ATTR_FATTR_BTIME)) 1002 request_mask &= ~STATX_BTIME; 1003 1004 if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) { 1005 if (readdirplus_enabled) 1006 nfs_readdirplus_parent_cache_hit(path->dentry); 1007 goto out_no_revalidate; 1008 } 1009 1010 /* Flush out writes to the server in order to update c/mtime/version. */ 1011 if ((request_mask & (STATX_CTIME | STATX_MTIME | STATX_CHANGE_COOKIE)) && 1012 S_ISREG(inode->i_mode)) { 1013 if (nfs_have_delegated_mtime(inode)) 1014 filemap_fdatawrite(inode->i_mapping); 1015 else 1016 filemap_write_and_wait(inode->i_mapping); 1017 } 1018 1019 /* 1020 * We may force a getattr if the user cares about atime. 1021 * 1022 * Note that we only have to check the vfsmount flags here: 1023 * - NFS always sets S_NOATIME by so checking it would give a 1024 * bogus result 1025 * - NFS never sets SB_NOATIME or SB_NODIRATIME so there is 1026 * no point in checking those. 1027 */ 1028 if ((path->mnt->mnt_flags & MNT_NOATIME) || 1029 ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) 1030 request_mask &= ~STATX_ATIME; 1031 1032 /* Is the user requesting attributes that might need revalidation? */ 1033 if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME| 1034 STATX_MTIME|STATX_UID|STATX_GID| 1035 STATX_SIZE|STATX_BLOCKS|STATX_BTIME| 1036 STATX_CHANGE_COOKIE))) 1037 goto out_no_revalidate; 1038 1039 /* Check whether the cached attributes are stale */ 1040 do_update |= force_sync || nfs_attribute_cache_expired(inode); 1041 cache_validity = READ_ONCE(NFS_I(inode)->cache_validity); 1042 do_update |= cache_validity & NFS_INO_INVALID_CHANGE; 1043 if (request_mask & STATX_ATIME) 1044 do_update |= cache_validity & NFS_INO_INVALID_ATIME; 1045 if (request_mask & STATX_CTIME) 1046 do_update |= cache_validity & NFS_INO_INVALID_CTIME; 1047 if (request_mask & STATX_MTIME) 1048 do_update |= cache_validity & NFS_INO_INVALID_MTIME; 1049 if (request_mask & STATX_SIZE) 1050 do_update |= cache_validity & NFS_INO_INVALID_SIZE; 1051 if (request_mask & STATX_NLINK) 1052 do_update |= cache_validity & NFS_INO_INVALID_NLINK; 1053 if (request_mask & STATX_MODE) 1054 do_update |= cache_validity & NFS_INO_INVALID_MODE; 1055 if (request_mask & (STATX_UID | STATX_GID)) 1056 do_update |= cache_validity & NFS_INO_INVALID_OTHER; 1057 if (request_mask & STATX_BLOCKS) 1058 do_update |= cache_validity & NFS_INO_INVALID_BLOCKS; 1059 if (request_mask & STATX_BTIME) 1060 do_update |= cache_validity & NFS_INO_INVALID_BTIME; 1061 1062 if (do_update) { 1063 if (readdirplus_enabled) 1064 nfs_readdirplus_parent_cache_miss(path->dentry); 1065 err = __nfs_revalidate_inode(server, inode); 1066 if (err) 1067 goto out; 1068 } else if (readdirplus_enabled) 1069 nfs_readdirplus_parent_cache_hit(path->dentry); 1070 out_no_revalidate: 1071 /* Only return attributes that were revalidated. */ 1072 stat->result_mask = nfs_get_valid_attrmask(inode) | request_mask; 1073 1074 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat); 1075 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode)); 1076 stat->change_cookie = inode_peek_iversion_raw(inode); 1077 stat->attributes_mask |= STATX_ATTR_CHANGE_MONOTONIC; 1078 if (server->change_attr_type != NFS4_CHANGE_TYPE_IS_UNDEFINED) 1079 stat->attributes |= STATX_ATTR_CHANGE_MONOTONIC; 1080 if (S_ISDIR(inode->i_mode)) 1081 stat->blksize = NFS_SERVER(inode)->dtsize; 1082 stat->btime = NFS_I(inode)->btime; 1083 1084 /* Special handling for STATX_DIOALIGN and STATX_DIO_READ_ALIGN 1085 * - NFS doesn't have DIO alignment constraints, avoid getting 1086 * these DIO attrs from remote and just respond with most 1087 * accommodating limits (so client will issue supported DIO). 1088 * - this is unintuitive, but the most coarse-grained 1089 * dio_offset_align is the most accommodating. 1090 */ 1091 if ((request_mask & (STATX_DIOALIGN | STATX_DIO_READ_ALIGN)) && 1092 S_ISREG(inode->i_mode)) { 1093 stat->result_mask |= STATX_DIOALIGN | STATX_DIO_READ_ALIGN; 1094 stat->dio_mem_align = 4; /* 4-byte alignment */ 1095 stat->dio_offset_align = PAGE_SIZE; 1096 stat->dio_read_offset_align = stat->dio_offset_align; 1097 } 1098 out: 1099 trace_nfs_getattr_exit(inode, err); 1100 return err; 1101 } 1102 EXPORT_SYMBOL_GPL(nfs_getattr); 1103 1104 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx) 1105 { 1106 refcount_set(&l_ctx->count, 1); 1107 l_ctx->lockowner = current->files; 1108 INIT_LIST_HEAD(&l_ctx->list); 1109 atomic_set(&l_ctx->io_count, 0); 1110 } 1111 1112 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx) 1113 { 1114 struct nfs_lock_context *pos; 1115 1116 list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) { 1117 if (pos->lockowner != current->files) 1118 continue; 1119 if (refcount_inc_not_zero(&pos->count)) 1120 return pos; 1121 } 1122 return NULL; 1123 } 1124 1125 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx) 1126 { 1127 struct nfs_lock_context *res, *new = NULL; 1128 struct inode *inode = d_inode(ctx->dentry); 1129 1130 rcu_read_lock(); 1131 res = __nfs_find_lock_context(ctx); 1132 rcu_read_unlock(); 1133 if (res == NULL) { 1134 new = kmalloc(sizeof(*new), GFP_KERNEL_ACCOUNT); 1135 if (new == NULL) 1136 return ERR_PTR(-ENOMEM); 1137 nfs_init_lock_context(new); 1138 spin_lock(&inode->i_lock); 1139 res = __nfs_find_lock_context(ctx); 1140 if (res == NULL) { 1141 new->open_context = get_nfs_open_context(ctx); 1142 if (new->open_context) { 1143 list_add_tail_rcu(&new->list, 1144 &ctx->lock_context.list); 1145 res = new; 1146 new = NULL; 1147 } else 1148 res = ERR_PTR(-EBADF); 1149 } 1150 spin_unlock(&inode->i_lock); 1151 kfree(new); 1152 } 1153 return res; 1154 } 1155 EXPORT_SYMBOL_GPL(nfs_get_lock_context); 1156 1157 void nfs_put_lock_context(struct nfs_lock_context *l_ctx) 1158 { 1159 struct nfs_open_context *ctx = l_ctx->open_context; 1160 struct inode *inode = d_inode(ctx->dentry); 1161 1162 if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock)) 1163 return; 1164 list_del_rcu(&l_ctx->list); 1165 spin_unlock(&inode->i_lock); 1166 put_nfs_open_context(ctx); 1167 kfree_rcu(l_ctx, rcu_head); 1168 } 1169 EXPORT_SYMBOL_GPL(nfs_put_lock_context); 1170 1171 /** 1172 * nfs_close_context - Common close_context() routine NFSv2/v3 1173 * @ctx: pointer to context 1174 * @is_sync: is this a synchronous close 1175 * 1176 * Ensure that the attributes are up to date if we're mounted 1177 * with close-to-open semantics and we have cached data that will 1178 * need to be revalidated on open. 1179 */ 1180 void nfs_close_context(struct nfs_open_context *ctx, int is_sync) 1181 { 1182 struct nfs_inode *nfsi; 1183 struct inode *inode; 1184 1185 if (!(ctx->mode & FMODE_WRITE)) 1186 return; 1187 if (!is_sync) 1188 return; 1189 inode = d_inode(ctx->dentry); 1190 if (nfs_have_read_or_write_delegation(inode)) 1191 return; 1192 nfsi = NFS_I(inode); 1193 if (inode->i_mapping->nrpages == 0) 1194 return; 1195 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 1196 return; 1197 if (!list_empty(&nfsi->open_files)) 1198 return; 1199 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NOCTO) 1200 return; 1201 nfs_revalidate_inode(inode, 1202 NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE); 1203 } 1204 EXPORT_SYMBOL_GPL(nfs_close_context); 1205 1206 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, 1207 fmode_t f_mode, 1208 struct file *filp) 1209 { 1210 struct nfs_open_context *ctx; 1211 1212 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL_ACCOUNT); 1213 if (!ctx) 1214 return ERR_PTR(-ENOMEM); 1215 nfs_sb_active(dentry->d_sb); 1216 ctx->dentry = dget(dentry); 1217 if (filp) 1218 ctx->cred = get_cred(filp->f_cred); 1219 else 1220 ctx->cred = get_current_cred(); 1221 rcu_assign_pointer(ctx->ll_cred, NULL); 1222 ctx->state = NULL; 1223 ctx->mode = f_mode; 1224 ctx->flags = 0; 1225 ctx->error = 0; 1226 ctx->flock_owner = (fl_owner_t)filp; 1227 nfs_init_lock_context(&ctx->lock_context); 1228 ctx->lock_context.open_context = ctx; 1229 INIT_LIST_HEAD(&ctx->list); 1230 ctx->mdsthreshold = NULL; 1231 nfs_localio_file_init(&ctx->nfl); 1232 1233 return ctx; 1234 } 1235 EXPORT_SYMBOL_GPL(alloc_nfs_open_context); 1236 1237 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx) 1238 { 1239 if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count)) 1240 return ctx; 1241 return NULL; 1242 } 1243 EXPORT_SYMBOL_GPL(get_nfs_open_context); 1244 1245 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync) 1246 { 1247 struct inode *inode = d_inode(ctx->dentry); 1248 struct super_block *sb = ctx->dentry->d_sb; 1249 1250 if (!refcount_dec_and_test(&ctx->lock_context.count)) 1251 return; 1252 if (!list_empty(&ctx->list)) { 1253 spin_lock(&inode->i_lock); 1254 list_del_rcu(&ctx->list); 1255 spin_unlock(&inode->i_lock); 1256 } 1257 if (inode != NULL) 1258 NFS_PROTO(inode)->close_context(ctx, is_sync); 1259 put_cred(ctx->cred); 1260 dput(ctx->dentry); 1261 nfs_sb_deactive(sb); 1262 put_rpccred(rcu_dereference_protected(ctx->ll_cred, 1)); 1263 kfree(ctx->mdsthreshold); 1264 nfs_close_local_fh(&ctx->nfl); 1265 kfree_rcu(ctx, rcu_head); 1266 } 1267 1268 void put_nfs_open_context(struct nfs_open_context *ctx) 1269 { 1270 __put_nfs_open_context(ctx, 0); 1271 } 1272 EXPORT_SYMBOL_GPL(put_nfs_open_context); 1273 1274 static void put_nfs_open_context_sync(struct nfs_open_context *ctx) 1275 { 1276 __put_nfs_open_context(ctx, 1); 1277 } 1278 1279 /* 1280 * Ensure that mmap has a recent RPC credential for use when writing out 1281 * shared pages 1282 */ 1283 void nfs_inode_attach_open_context(struct nfs_open_context *ctx) 1284 { 1285 struct inode *inode = d_inode(ctx->dentry); 1286 struct nfs_inode *nfsi = NFS_I(inode); 1287 1288 spin_lock(&inode->i_lock); 1289 if (list_empty(&nfsi->open_files) && 1290 nfs_ooo_test(nfsi)) 1291 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA | 1292 NFS_INO_REVAL_FORCED); 1293 list_add_tail_rcu(&ctx->list, &nfsi->open_files); 1294 spin_unlock(&inode->i_lock); 1295 } 1296 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context); 1297 1298 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx) 1299 { 1300 filp->private_data = get_nfs_open_context(ctx); 1301 set_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags); 1302 if (list_empty(&ctx->list)) 1303 nfs_inode_attach_open_context(ctx); 1304 } 1305 EXPORT_SYMBOL_GPL(nfs_file_set_open_context); 1306 1307 /* 1308 * Given an inode, search for an open context with the desired characteristics 1309 */ 1310 struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode) 1311 { 1312 struct nfs_inode *nfsi = NFS_I(inode); 1313 struct nfs_open_context *pos, *ctx = NULL; 1314 1315 rcu_read_lock(); 1316 list_for_each_entry_rcu(pos, &nfsi->open_files, list) { 1317 if (cred != NULL && cred_fscmp(pos->cred, cred) != 0) 1318 continue; 1319 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode) 1320 continue; 1321 if (!test_bit(NFS_CONTEXT_FILE_OPEN, &pos->flags)) 1322 continue; 1323 ctx = get_nfs_open_context(pos); 1324 if (ctx) 1325 break; 1326 } 1327 rcu_read_unlock(); 1328 return ctx; 1329 } 1330 1331 void nfs_file_clear_open_context(struct file *filp) 1332 { 1333 struct nfs_open_context *ctx = nfs_file_open_context(filp); 1334 1335 if (ctx) { 1336 struct inode *inode = d_inode(ctx->dentry); 1337 1338 clear_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags); 1339 /* 1340 * We fatal error on write before. Try to writeback 1341 * every page again. 1342 */ 1343 if (ctx->error < 0) 1344 invalidate_inode_pages2(inode->i_mapping); 1345 filp->private_data = NULL; 1346 put_nfs_open_context_sync(ctx); 1347 } 1348 } 1349 1350 /* 1351 * These allocate and release file read/write context information. 1352 */ 1353 int nfs_open(struct inode *inode, struct file *filp) 1354 { 1355 struct nfs_open_context *ctx; 1356 1357 ctx = alloc_nfs_open_context(file_dentry(filp), 1358 flags_to_mode(filp->f_flags), filp); 1359 if (IS_ERR(ctx)) 1360 return PTR_ERR(ctx); 1361 nfs_file_set_open_context(filp, ctx); 1362 put_nfs_open_context(ctx); 1363 nfs_fscache_open_file(inode, filp); 1364 return 0; 1365 } 1366 1367 /* 1368 * This function is called whenever some part of NFS notices that 1369 * the cached attributes have to be refreshed. 1370 */ 1371 int 1372 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) 1373 { 1374 int status = -ESTALE; 1375 struct nfs_fattr *fattr = NULL; 1376 struct nfs_inode *nfsi = NFS_I(inode); 1377 1378 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n", 1379 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode)); 1380 1381 trace_nfs_revalidate_inode_enter(inode); 1382 1383 if (is_bad_inode(inode)) 1384 goto out; 1385 if (NFS_STALE(inode)) 1386 goto out; 1387 1388 /* pNFS: Attributes aren't updated until we layoutcommit */ 1389 if (S_ISREG(inode->i_mode)) { 1390 status = pnfs_sync_inode(inode, false); 1391 if (status) 1392 goto out; 1393 } else if (nfs_have_directory_delegation(inode)) { 1394 status = 0; 1395 goto out; 1396 } 1397 1398 status = -ENOMEM; 1399 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode)); 1400 if (fattr == NULL) 1401 goto out; 1402 1403 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE); 1404 1405 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, inode); 1406 if (status != 0) { 1407 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n", 1408 inode->i_sb->s_id, 1409 (unsigned long long)NFS_FILEID(inode), status); 1410 switch (status) { 1411 case -ETIMEDOUT: 1412 /* A soft timeout occurred. Use cached information? */ 1413 if (server->flags & NFS_MOUNT_SOFTREVAL) 1414 status = 0; 1415 break; 1416 case -ESTALE: 1417 if (!S_ISDIR(inode->i_mode)) 1418 nfs_set_inode_stale(inode); 1419 else 1420 nfs_zap_caches(inode); 1421 } 1422 goto out; 1423 } 1424 1425 status = nfs_refresh_inode(inode, fattr); 1426 if (status) { 1427 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n", 1428 inode->i_sb->s_id, 1429 (unsigned long long)NFS_FILEID(inode), status); 1430 goto out; 1431 } 1432 1433 if (nfsi->cache_validity & NFS_INO_INVALID_ACL) 1434 nfs_zap_acl_cache(inode); 1435 1436 nfs_setsecurity(inode, fattr); 1437 1438 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n", 1439 inode->i_sb->s_id, 1440 (unsigned long long)NFS_FILEID(inode)); 1441 1442 out: 1443 nfs_free_fattr(fattr); 1444 trace_nfs_revalidate_inode_exit(inode, status); 1445 return status; 1446 } 1447 1448 int nfs_attribute_cache_expired(struct inode *inode) 1449 { 1450 if (nfs_have_delegated_attributes(inode)) 1451 return 0; 1452 return nfs_attribute_timeout(inode); 1453 } 1454 1455 /** 1456 * nfs_revalidate_inode - Revalidate the inode attributes 1457 * @inode: pointer to inode struct 1458 * @flags: cache flags to check 1459 * 1460 * Updates inode attribute information by retrieving the data from the server. 1461 */ 1462 int nfs_revalidate_inode(struct inode *inode, unsigned long flags) 1463 { 1464 if (!nfs_check_cache_invalid(inode, flags)) 1465 return NFS_STALE(inode) ? -ESTALE : 0; 1466 return __nfs_revalidate_inode(NFS_SERVER(inode), inode); 1467 } 1468 EXPORT_SYMBOL_GPL(nfs_revalidate_inode); 1469 1470 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping) 1471 { 1472 int ret; 1473 1474 nfs_fscache_invalidate(inode, 0); 1475 if (mapping->nrpages != 0) { 1476 if (S_ISREG(inode->i_mode)) { 1477 ret = nfs_sync_mapping(mapping); 1478 if (ret < 0) 1479 return ret; 1480 } 1481 ret = invalidate_inode_pages2(mapping); 1482 if (ret < 0) 1483 return ret; 1484 } 1485 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE); 1486 1487 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n", 1488 inode->i_sb->s_id, 1489 (unsigned long long)NFS_FILEID(inode)); 1490 return 0; 1491 } 1492 1493 /** 1494 * nfs_clear_invalid_mapping - Conditionally clear a mapping 1495 * @mapping: pointer to mapping 1496 * 1497 * If the NFS_INO_INVALID_DATA inode flag is set, clear the mapping. 1498 */ 1499 int nfs_clear_invalid_mapping(struct address_space *mapping) 1500 { 1501 struct inode *inode = mapping->host; 1502 struct nfs_inode *nfsi = NFS_I(inode); 1503 unsigned long *bitlock = &nfsi->flags; 1504 int ret = 0; 1505 1506 /* 1507 * We must clear NFS_INO_INVALID_DATA first to ensure that 1508 * invalidations that come in while we're shooting down the mappings 1509 * are respected. But, that leaves a race window where one revalidator 1510 * can clear the flag, and then another checks it before the mapping 1511 * gets invalidated. Fix that by serializing access to this part of 1512 * the function. 1513 * 1514 * At the same time, we need to allow other tasks to see whether we 1515 * might be in the middle of invalidating the pages, so we only set 1516 * the bit lock here if it looks like we're going to be doing that. 1517 */ 1518 for (;;) { 1519 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING, 1520 nfs_wait_bit_killable, 1521 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE); 1522 if (ret) 1523 goto out; 1524 smp_rmb(); /* pairs with smp_wmb() below */ 1525 if (test_bit(NFS_INO_INVALIDATING, bitlock)) 1526 continue; 1527 /* pairs with nfs_set_cache_invalid()'s smp_store_release() */ 1528 if (!(smp_load_acquire(&nfsi->cache_validity) & NFS_INO_INVALID_DATA)) 1529 goto out; 1530 /* Slow-path that double-checks with spinlock held */ 1531 spin_lock(&inode->i_lock); 1532 if (test_bit(NFS_INO_INVALIDATING, bitlock)) { 1533 spin_unlock(&inode->i_lock); 1534 continue; 1535 } 1536 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 1537 break; 1538 spin_unlock(&inode->i_lock); 1539 goto out; 1540 } 1541 1542 set_bit(NFS_INO_INVALIDATING, bitlock); 1543 smp_wmb(); 1544 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA; 1545 nfs_ooo_clear(nfsi); 1546 spin_unlock(&inode->i_lock); 1547 trace_nfs_invalidate_mapping_enter(inode); 1548 ret = nfs_invalidate_mapping(inode, mapping); 1549 trace_nfs_invalidate_mapping_exit(inode, ret); 1550 1551 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock); 1552 smp_mb__after_atomic(); 1553 wake_up_bit(bitlock, NFS_INO_INVALIDATING); 1554 out: 1555 return ret; 1556 } 1557 1558 bool nfs_mapping_need_revalidate_inode(struct inode *inode) 1559 { 1560 return nfs_check_cache_invalid(inode, NFS_INO_INVALID_CHANGE) || 1561 NFS_STALE(inode); 1562 } 1563 1564 int nfs_revalidate_mapping_rcu(struct inode *inode) 1565 { 1566 struct nfs_inode *nfsi = NFS_I(inode); 1567 unsigned long *bitlock = &nfsi->flags; 1568 int ret = 0; 1569 1570 if (IS_SWAPFILE(inode)) 1571 goto out; 1572 if (nfs_mapping_need_revalidate_inode(inode)) { 1573 ret = -ECHILD; 1574 goto out; 1575 } 1576 spin_lock(&inode->i_lock); 1577 if (test_bit(NFS_INO_INVALIDATING, bitlock) || 1578 (nfsi->cache_validity & NFS_INO_INVALID_DATA)) 1579 ret = -ECHILD; 1580 spin_unlock(&inode->i_lock); 1581 out: 1582 return ret; 1583 } 1584 1585 /** 1586 * nfs_revalidate_mapping - Revalidate the pagecache 1587 * @inode: pointer to host inode 1588 * @mapping: pointer to mapping 1589 */ 1590 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping) 1591 { 1592 /* swapfiles are not supposed to be shared. */ 1593 if (IS_SWAPFILE(inode)) 1594 return 0; 1595 1596 if (nfs_mapping_need_revalidate_inode(inode)) { 1597 int ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 1598 if (ret < 0) 1599 return ret; 1600 } 1601 1602 return nfs_clear_invalid_mapping(mapping); 1603 } 1604 1605 static bool nfs_file_has_writers(struct nfs_inode *nfsi) 1606 { 1607 struct inode *inode = &nfsi->vfs_inode; 1608 1609 if (!S_ISREG(inode->i_mode)) 1610 return false; 1611 if (list_empty(&nfsi->open_files)) 1612 return false; 1613 return inode_is_open_for_write(inode); 1614 } 1615 1616 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi) 1617 { 1618 return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi); 1619 } 1620 1621 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1622 { 1623 struct timespec64 ts; 1624 1625 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE) 1626 && (fattr->valid & NFS_ATTR_FATTR_CHANGE) 1627 && inode_eq_iversion_raw(inode, fattr->pre_change_attr)) { 1628 inode_set_iversion_raw(inode, fattr->change_attr); 1629 if (S_ISDIR(inode->i_mode)) 1630 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA); 1631 else if (nfs_server_capable(inode, NFS_CAP_XATTR)) 1632 nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR); 1633 } 1634 /* If we have atomic WCC data, we may update some attributes */ 1635 ts = inode_get_ctime(inode); 1636 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME) 1637 && (fattr->valid & NFS_ATTR_FATTR_CTIME) 1638 && timespec64_equal(&ts, &fattr->pre_ctime)) { 1639 inode_set_ctime_to_ts(inode, fattr->ctime); 1640 } 1641 1642 ts = inode_get_mtime(inode); 1643 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME) 1644 && (fattr->valid & NFS_ATTR_FATTR_MTIME) 1645 && timespec64_equal(&ts, &fattr->pre_mtime)) { 1646 inode_set_mtime_to_ts(inode, fattr->mtime); 1647 } 1648 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE) 1649 && (fattr->valid & NFS_ATTR_FATTR_SIZE) 1650 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size) 1651 && !nfs_have_writebacks(inode)) { 1652 trace_nfs_size_wcc(inode, fattr->size); 1653 i_size_write(inode, nfs_size_to_loff_t(fattr->size)); 1654 } 1655 } 1656 1657 /** 1658 * nfs_check_inode_attributes - verify consistency of the inode attribute cache 1659 * @inode: pointer to inode 1660 * @fattr: updated attributes 1661 * 1662 * Verifies the attribute cache. If we have just changed the attributes, 1663 * so that fattr carries weak cache consistency data, then it may 1664 * also update the ctime/mtime/change_attribute. 1665 */ 1666 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr) 1667 { 1668 struct nfs_inode *nfsi = NFS_I(inode); 1669 loff_t cur_size, new_isize; 1670 unsigned long invalid = 0; 1671 struct timespec64 ts; 1672 1673 if (nfs_have_delegated_attributes(inode)) 1674 return 0; 1675 1676 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) { 1677 /* Only a mounted-on-fileid? Just exit */ 1678 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) 1679 return 0; 1680 /* Has the inode gone and changed behind our back? */ 1681 } else if (nfsi->fileid != fattr->fileid) { 1682 /* Is this perhaps the mounted-on fileid? */ 1683 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) && 1684 nfsi->fileid == fattr->mounted_on_fileid) 1685 return 0; 1686 return -ESTALE; 1687 } 1688 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) 1689 return -ESTALE; 1690 1691 1692 if (!nfs_file_has_buffered_writers(nfsi)) { 1693 /* Verify a few of the more important attributes */ 1694 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr)) 1695 invalid |= NFS_INO_INVALID_CHANGE; 1696 1697 ts = inode_get_mtime(inode); 1698 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime)) 1699 invalid |= NFS_INO_INVALID_MTIME; 1700 1701 ts = inode_get_ctime(inode); 1702 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime)) 1703 invalid |= NFS_INO_INVALID_CTIME; 1704 1705 if (fattr->valid & NFS_ATTR_FATTR_SIZE) { 1706 cur_size = i_size_read(inode); 1707 new_isize = nfs_size_to_loff_t(fattr->size); 1708 if (cur_size != new_isize) 1709 invalid |= NFS_INO_INVALID_SIZE; 1710 } 1711 } 1712 1713 /* Have any file permissions changed? */ 1714 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) 1715 invalid |= NFS_INO_INVALID_MODE; 1716 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid)) 1717 invalid |= NFS_INO_INVALID_OTHER; 1718 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid)) 1719 invalid |= NFS_INO_INVALID_OTHER; 1720 1721 /* Has the link count changed? */ 1722 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink) 1723 invalid |= NFS_INO_INVALID_NLINK; 1724 1725 ts = inode_get_atime(inode); 1726 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime)) 1727 invalid |= NFS_INO_INVALID_ATIME; 1728 1729 if (invalid != 0) 1730 nfs_set_cache_invalid(inode, invalid); 1731 1732 nfsi->read_cache_jiffies = fattr->time_start; 1733 return 0; 1734 } 1735 1736 static atomic_long_t nfs_attr_generation_counter; 1737 1738 static unsigned long nfs_read_attr_generation_counter(void) 1739 { 1740 return atomic_long_read(&nfs_attr_generation_counter); 1741 } 1742 1743 unsigned long nfs_inc_attr_generation_counter(void) 1744 { 1745 return atomic_long_inc_return(&nfs_attr_generation_counter); 1746 } 1747 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter); 1748 1749 void nfs_fattr_init(struct nfs_fattr *fattr) 1750 { 1751 fattr->valid = 0; 1752 fattr->time_start = jiffies; 1753 fattr->gencount = nfs_inc_attr_generation_counter(); 1754 fattr->owner_name = NULL; 1755 fattr->group_name = NULL; 1756 fattr->mdsthreshold = NULL; 1757 } 1758 EXPORT_SYMBOL_GPL(nfs_fattr_init); 1759 1760 /** 1761 * nfs_fattr_set_barrier 1762 * @fattr: attributes 1763 * 1764 * Used to set a barrier after an attribute was updated. This 1765 * barrier ensures that older attributes from RPC calls that may 1766 * have raced with our update cannot clobber these new values. 1767 * Note that you are still responsible for ensuring that other 1768 * operations which change the attribute on the server do not 1769 * collide. 1770 */ 1771 void nfs_fattr_set_barrier(struct nfs_fattr *fattr) 1772 { 1773 fattr->gencount = nfs_inc_attr_generation_counter(); 1774 } 1775 1776 struct nfs_fattr *nfs_alloc_fattr(void) 1777 { 1778 struct nfs_fattr *fattr; 1779 1780 fattr = kmalloc(sizeof(*fattr), GFP_KERNEL); 1781 if (fattr != NULL) { 1782 nfs_fattr_init(fattr); 1783 fattr->label = NULL; 1784 } 1785 return fattr; 1786 } 1787 EXPORT_SYMBOL_GPL(nfs_alloc_fattr); 1788 1789 struct nfs_fattr *nfs_alloc_fattr_with_label(struct nfs_server *server) 1790 { 1791 struct nfs_fattr *fattr = nfs_alloc_fattr(); 1792 1793 if (!fattr) 1794 return NULL; 1795 1796 fattr->label = nfs4_label_alloc(server, GFP_KERNEL); 1797 if (IS_ERR(fattr->label)) { 1798 kfree(fattr); 1799 return NULL; 1800 } 1801 1802 return fattr; 1803 } 1804 EXPORT_SYMBOL_GPL(nfs_alloc_fattr_with_label); 1805 1806 struct nfs_fh *nfs_alloc_fhandle(void) 1807 { 1808 struct nfs_fh *fh; 1809 1810 fh = kmalloc(sizeof(struct nfs_fh), GFP_KERNEL); 1811 if (fh != NULL) 1812 fh->size = 0; 1813 return fh; 1814 } 1815 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle); 1816 1817 #ifdef NFS_DEBUG 1818 /* 1819 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle 1820 * in the same way that wireshark does 1821 * 1822 * @fh: file handle 1823 * 1824 * For debugging only. 1825 */ 1826 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh) 1827 { 1828 /* wireshark uses 32-bit AUTODIN crc and does a bitwise 1829 * not on the result */ 1830 return nfs_fhandle_hash(fh); 1831 } 1832 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash); 1833 1834 /* 1835 * _nfs_display_fhandle - display an NFS file handle on the console 1836 * 1837 * @fh: file handle to display 1838 * @caption: display caption 1839 * 1840 * For debugging only. 1841 */ 1842 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption) 1843 { 1844 unsigned short i; 1845 1846 if (fh == NULL || fh->size == 0) { 1847 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh); 1848 return; 1849 } 1850 1851 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n", 1852 caption, fh, fh->size, _nfs_display_fhandle_hash(fh)); 1853 for (i = 0; i < fh->size; i += 16) { 1854 __be32 *pos = (__be32 *)&fh->data[i]; 1855 1856 switch ((fh->size - i - 1) >> 2) { 1857 case 0: 1858 printk(KERN_DEFAULT " %08x\n", 1859 be32_to_cpup(pos)); 1860 break; 1861 case 1: 1862 printk(KERN_DEFAULT " %08x %08x\n", 1863 be32_to_cpup(pos), be32_to_cpup(pos + 1)); 1864 break; 1865 case 2: 1866 printk(KERN_DEFAULT " %08x %08x %08x\n", 1867 be32_to_cpup(pos), be32_to_cpup(pos + 1), 1868 be32_to_cpup(pos + 2)); 1869 break; 1870 default: 1871 printk(KERN_DEFAULT " %08x %08x %08x %08x\n", 1872 be32_to_cpup(pos), be32_to_cpup(pos + 1), 1873 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3)); 1874 } 1875 } 1876 } 1877 EXPORT_SYMBOL_GPL(_nfs_display_fhandle); 1878 #endif 1879 1880 /** 1881 * nfs_inode_attrs_cmp_generic - compare attributes 1882 * @fattr: attributes 1883 * @inode: pointer to inode 1884 * 1885 * Attempt to divine whether or not an RPC call reply carrying stale 1886 * attributes got scheduled after another call carrying updated ones. 1887 * Note also the check for wraparound of 'attr_gencount' 1888 * 1889 * The function returns '1' if it thinks the attributes in @fattr are 1890 * more recent than the ones cached in @inode. Otherwise it returns 1891 * the value '0'. 1892 */ 1893 static int nfs_inode_attrs_cmp_generic(const struct nfs_fattr *fattr, 1894 const struct inode *inode) 1895 { 1896 unsigned long attr_gencount = NFS_I(inode)->attr_gencount; 1897 1898 return (long)(fattr->gencount - attr_gencount) > 0 || 1899 (long)(attr_gencount - nfs_read_attr_generation_counter()) > 0; 1900 } 1901 1902 /** 1903 * nfs_inode_attrs_cmp_monotonic - compare attributes 1904 * @fattr: attributes 1905 * @inode: pointer to inode 1906 * 1907 * Attempt to divine whether or not an RPC call reply carrying stale 1908 * attributes got scheduled after another call carrying updated ones. 1909 * 1910 * We assume that the server observes monotonic semantics for 1911 * the change attribute, so a larger value means that the attributes in 1912 * @fattr are more recent, in which case the function returns the 1913 * value '1'. 1914 * A return value of '0' indicates no measurable change 1915 * A return value of '-1' means that the attributes in @inode are 1916 * more recent. 1917 */ 1918 static int nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr *fattr, 1919 const struct inode *inode) 1920 { 1921 s64 diff = fattr->change_attr - inode_peek_iversion_raw(inode); 1922 if (diff > 0) 1923 return 1; 1924 return diff == 0 ? 0 : -1; 1925 } 1926 1927 /** 1928 * nfs_inode_attrs_cmp_strict_monotonic - compare attributes 1929 * @fattr: attributes 1930 * @inode: pointer to inode 1931 * 1932 * Attempt to divine whether or not an RPC call reply carrying stale 1933 * attributes got scheduled after another call carrying updated ones. 1934 * 1935 * We assume that the server observes strictly monotonic semantics for 1936 * the change attribute, so a larger value means that the attributes in 1937 * @fattr are more recent, in which case the function returns the 1938 * value '1'. 1939 * A return value of '-1' means that the attributes in @inode are 1940 * more recent or unchanged. 1941 */ 1942 static int nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr *fattr, 1943 const struct inode *inode) 1944 { 1945 return nfs_inode_attrs_cmp_monotonic(fattr, inode) > 0 ? 1 : -1; 1946 } 1947 1948 /** 1949 * nfs_inode_attrs_cmp - compare attributes 1950 * @fattr: attributes 1951 * @inode: pointer to inode 1952 * 1953 * This function returns '1' if it thinks the attributes in @fattr are 1954 * more recent than the ones cached in @inode. It returns '-1' if 1955 * the attributes in @inode are more recent than the ones in @fattr, 1956 * and it returns 0 if not sure. 1957 */ 1958 static int nfs_inode_attrs_cmp(const struct nfs_fattr *fattr, 1959 const struct inode *inode) 1960 { 1961 if (nfs_inode_attrs_cmp_generic(fattr, inode) > 0) 1962 return 1; 1963 switch (NFS_SERVER(inode)->change_attr_type) { 1964 case NFS4_CHANGE_TYPE_IS_UNDEFINED: 1965 break; 1966 case NFS4_CHANGE_TYPE_IS_TIME_METADATA: 1967 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE)) 1968 break; 1969 return nfs_inode_attrs_cmp_monotonic(fattr, inode); 1970 default: 1971 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE)) 1972 break; 1973 return nfs_inode_attrs_cmp_strict_monotonic(fattr, inode); 1974 } 1975 return 0; 1976 } 1977 1978 /** 1979 * nfs_inode_finish_partial_attr_update - complete a previous inode update 1980 * @fattr: attributes 1981 * @inode: pointer to inode 1982 * 1983 * Returns '1' if the last attribute update left the inode cached 1984 * attributes in a partially unrevalidated state, and @fattr 1985 * matches the change attribute of that partial update. 1986 * Otherwise returns '0'. 1987 */ 1988 static int nfs_inode_finish_partial_attr_update(const struct nfs_fattr *fattr, 1989 const struct inode *inode) 1990 { 1991 const unsigned long check_valid = 1992 NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME | 1993 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE | 1994 NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_OTHER | 1995 NFS_INO_INVALID_NLINK | NFS_INO_INVALID_BTIME; 1996 unsigned long cache_validity = NFS_I(inode)->cache_validity; 1997 enum nfs4_change_attr_type ctype = NFS_SERVER(inode)->change_attr_type; 1998 1999 if (ctype != NFS4_CHANGE_TYPE_IS_UNDEFINED && 2000 !(cache_validity & NFS_INO_INVALID_CHANGE) && 2001 (cache_validity & check_valid) != 0 && 2002 (fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && 2003 nfs_inode_attrs_cmp_monotonic(fattr, inode) == 0) 2004 return 1; 2005 return 0; 2006 } 2007 2008 static void nfs_ooo_merge(struct nfs_inode *nfsi, 2009 u64 start, u64 end) 2010 { 2011 int i, cnt; 2012 2013 if (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER) 2014 /* No point merging anything */ 2015 return; 2016 2017 if (!nfsi->ooo) { 2018 nfsi->ooo = kmalloc(sizeof(*nfsi->ooo), GFP_ATOMIC); 2019 if (!nfsi->ooo) { 2020 nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER; 2021 return; 2022 } 2023 nfsi->ooo->cnt = 0; 2024 } 2025 2026 /* add this range, merging if possible */ 2027 cnt = nfsi->ooo->cnt; 2028 for (i = 0; i < cnt; i++) { 2029 if (end == nfsi->ooo->gap[i].start) 2030 end = nfsi->ooo->gap[i].end; 2031 else if (start == nfsi->ooo->gap[i].end) 2032 start = nfsi->ooo->gap[i].start; 2033 else 2034 continue; 2035 /* Remove 'i' from table and loop to insert the new range */ 2036 cnt -= 1; 2037 nfsi->ooo->gap[i] = nfsi->ooo->gap[cnt]; 2038 i = -1; 2039 } 2040 if (start != end) { 2041 if (cnt >= ARRAY_SIZE(nfsi->ooo->gap)) { 2042 nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER; 2043 kfree(nfsi->ooo); 2044 nfsi->ooo = NULL; 2045 return; 2046 } 2047 nfsi->ooo->gap[cnt].start = start; 2048 nfsi->ooo->gap[cnt].end = end; 2049 cnt += 1; 2050 } 2051 nfsi->ooo->cnt = cnt; 2052 } 2053 2054 static void nfs_ooo_record(struct nfs_inode *nfsi, 2055 struct nfs_fattr *fattr) 2056 { 2057 /* This reply was out-of-order, so record in the 2058 * pre/post change id, possibly cancelling 2059 * gaps created when iversion was jumpped forward. 2060 */ 2061 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) && 2062 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE)) 2063 nfs_ooo_merge(nfsi, 2064 fattr->change_attr, 2065 fattr->pre_change_attr); 2066 } 2067 2068 static int nfs_refresh_inode_locked(struct inode *inode, 2069 struct nfs_fattr *fattr) 2070 { 2071 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode); 2072 int ret = 0; 2073 2074 trace_nfs_refresh_inode_enter(inode); 2075 2076 if (attr_cmp > 0 || nfs_inode_finish_partial_attr_update(fattr, inode)) 2077 ret = nfs_update_inode(inode, fattr); 2078 else { 2079 nfs_ooo_record(NFS_I(inode), fattr); 2080 2081 if (attr_cmp == 0) 2082 ret = nfs_check_inode_attributes(inode, fattr); 2083 } 2084 2085 trace_nfs_refresh_inode_exit(inode, ret); 2086 return ret; 2087 } 2088 2089 /** 2090 * nfs_refresh_inode - try to update the inode attribute cache 2091 * @inode: pointer to inode 2092 * @fattr: updated attributes 2093 * 2094 * Check that an RPC call that returned attributes has not overlapped with 2095 * other recent updates of the inode metadata, then decide whether it is 2096 * safe to do a full update of the inode attributes, or whether just to 2097 * call nfs_check_inode_attributes. 2098 */ 2099 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr) 2100 { 2101 int status; 2102 2103 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 2104 return 0; 2105 spin_lock(&inode->i_lock); 2106 status = nfs_refresh_inode_locked(inode, fattr); 2107 spin_unlock(&inode->i_lock); 2108 2109 return status; 2110 } 2111 EXPORT_SYMBOL_GPL(nfs_refresh_inode); 2112 2113 static int nfs_post_op_update_inode_locked(struct inode *inode, 2114 struct nfs_fattr *fattr, unsigned int invalid) 2115 { 2116 if (S_ISDIR(inode->i_mode)) 2117 invalid |= NFS_INO_INVALID_DATA; 2118 nfs_set_cache_invalid(inode, invalid); 2119 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 2120 return 0; 2121 return nfs_refresh_inode_locked(inode, fattr); 2122 } 2123 2124 /** 2125 * nfs_post_op_update_inode - try to update the inode attribute cache 2126 * @inode: pointer to inode 2127 * @fattr: updated attributes 2128 * 2129 * After an operation that has changed the inode metadata, mark the 2130 * attribute cache as being invalid, then try to update it. 2131 * 2132 * NB: if the server didn't return any post op attributes, this 2133 * function will force the retrieval of attributes before the next 2134 * NFS request. Thus it should be used only for operations that 2135 * are expected to change one or more attributes, to avoid 2136 * unnecessary NFS requests and trips through nfs_update_inode(). 2137 */ 2138 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr) 2139 { 2140 int status; 2141 2142 spin_lock(&inode->i_lock); 2143 nfs_fattr_set_barrier(fattr); 2144 status = nfs_post_op_update_inode_locked(inode, fattr, 2145 NFS_INO_INVALID_CHANGE 2146 | NFS_INO_INVALID_CTIME 2147 | NFS_INO_REVAL_FORCED); 2148 spin_unlock(&inode->i_lock); 2149 2150 return status; 2151 } 2152 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode); 2153 2154 /** 2155 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache 2156 * @inode: pointer to inode 2157 * @fattr: updated attributes 2158 * 2159 * After an operation that has changed the inode metadata, mark the 2160 * attribute cache as being invalid, then try to update it. Fake up 2161 * weak cache consistency data, if none exist. 2162 * 2163 * This function is mainly designed to be used by the ->write_done() functions. 2164 */ 2165 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr) 2166 { 2167 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode); 2168 int status; 2169 2170 /* Don't do a WCC update if these attributes are already stale */ 2171 if (attr_cmp < 0) 2172 return 0; 2173 if ((fattr->valid & NFS_ATTR_FATTR) == 0 || !attr_cmp) { 2174 /* Record the pre/post change info before clearing PRECHANGE */ 2175 nfs_ooo_record(NFS_I(inode), fattr); 2176 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE 2177 | NFS_ATTR_FATTR_PRESIZE 2178 | NFS_ATTR_FATTR_PREMTIME 2179 | NFS_ATTR_FATTR_PRECTIME); 2180 goto out_noforce; 2181 } 2182 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && 2183 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) { 2184 fattr->pre_change_attr = inode_peek_iversion_raw(inode); 2185 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE; 2186 } 2187 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 && 2188 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) { 2189 fattr->pre_ctime = inode_get_ctime(inode); 2190 fattr->valid |= NFS_ATTR_FATTR_PRECTIME; 2191 } 2192 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 && 2193 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) { 2194 fattr->pre_mtime = inode_get_mtime(inode); 2195 fattr->valid |= NFS_ATTR_FATTR_PREMTIME; 2196 } 2197 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 && 2198 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) { 2199 fattr->pre_size = i_size_read(inode); 2200 fattr->valid |= NFS_ATTR_FATTR_PRESIZE; 2201 } 2202 out_noforce: 2203 status = nfs_post_op_update_inode_locked(inode, fattr, 2204 NFS_INO_INVALID_CHANGE 2205 | NFS_INO_INVALID_CTIME 2206 | NFS_INO_INVALID_MTIME 2207 | NFS_INO_INVALID_BLOCKS); 2208 return status; 2209 } 2210 2211 /** 2212 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache 2213 * @inode: pointer to inode 2214 * @fattr: updated attributes 2215 * 2216 * After an operation that has changed the inode metadata, mark the 2217 * attribute cache as being invalid, then try to update it. Fake up 2218 * weak cache consistency data, if none exist. 2219 * 2220 * This function is mainly designed to be used by the ->write_done() functions. 2221 */ 2222 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr) 2223 { 2224 int status; 2225 2226 spin_lock(&inode->i_lock); 2227 nfs_fattr_set_barrier(fattr); 2228 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 2229 spin_unlock(&inode->i_lock); 2230 return status; 2231 } 2232 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc); 2233 2234 2235 /* 2236 * Many nfs protocol calls return the new file attributes after 2237 * an operation. Here we update the inode to reflect the state 2238 * of the server's inode. 2239 * 2240 * This is a bit tricky because we have to make sure all dirty pages 2241 * have been sent off to the server before calling invalidate_inode_pages. 2242 * To make sure no other process adds more write requests while we try 2243 * our best to flush them, we make them sleep during the attribute refresh. 2244 * 2245 * A very similar scenario holds for the dir cache. 2246 */ 2247 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr) 2248 { 2249 struct nfs_server *server = NFS_SERVER(inode); 2250 struct nfs_inode *nfsi = NFS_I(inode); 2251 loff_t cur_isize, new_isize; 2252 u64 fattr_supported = server->fattr_valid; 2253 unsigned long invalid = 0; 2254 unsigned long now = jiffies; 2255 unsigned long save_cache_validity; 2256 bool have_writers = nfs_file_has_buffered_writers(nfsi); 2257 bool cache_revalidated = true; 2258 bool attr_changed = false; 2259 bool have_delegation; 2260 2261 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%llx)\n", 2262 __func__, inode->i_sb->s_id, inode->i_ino, 2263 nfs_display_fhandle_hash(NFS_FH(inode)), 2264 icount_read(inode), fattr->valid); 2265 2266 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) { 2267 /* Only a mounted-on-fileid? Just exit */ 2268 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) 2269 return 0; 2270 /* Has the inode gone and changed behind our back? */ 2271 } else if (nfsi->fileid != fattr->fileid) { 2272 /* Is this perhaps the mounted-on fileid? */ 2273 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) && 2274 nfsi->fileid == fattr->mounted_on_fileid) 2275 return 0; 2276 printk(KERN_ERR "NFS: server %s error: fileid changed\n" 2277 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n", 2278 NFS_SERVER(inode)->nfs_client->cl_hostname, 2279 inode->i_sb->s_id, (long long)nfsi->fileid, 2280 (long long)fattr->fileid); 2281 goto out_err; 2282 } 2283 2284 /* 2285 * Make sure the inode's type hasn't changed. 2286 */ 2287 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) { 2288 /* 2289 * Big trouble! The inode has become a different object. 2290 */ 2291 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n", 2292 __func__, inode->i_ino, inode->i_mode, fattr->mode); 2293 goto out_err; 2294 } 2295 2296 /* Update the fsid? */ 2297 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) && 2298 !nfs_fsid_equal(&server->fsid, &fattr->fsid) && 2299 !IS_AUTOMOUNT(inode)) 2300 server->fsid = fattr->fsid; 2301 2302 /* Save the delegation state before clearing cache_validity */ 2303 have_delegation = nfs_have_delegated_attributes(inode); 2304 2305 /* 2306 * Update the read time so we don't revalidate too often. 2307 */ 2308 nfsi->read_cache_jiffies = fattr->time_start; 2309 2310 /* Fix up any delegated attributes in the struct nfs_fattr */ 2311 nfs_fattr_fixup_delegated(inode, fattr); 2312 2313 save_cache_validity = nfsi->cache_validity; 2314 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR 2315 | NFS_INO_INVALID_ATIME 2316 | NFS_INO_REVAL_FORCED 2317 | NFS_INO_INVALID_BLOCKS); 2318 2319 /* Do atomic weak cache consistency updates */ 2320 nfs_wcc_update_inode(inode, fattr); 2321 2322 if (pnfs_layoutcommit_outstanding(inode)) { 2323 nfsi->cache_validity |= 2324 save_cache_validity & 2325 (NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME | 2326 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE | 2327 NFS_INO_INVALID_BLOCKS); 2328 cache_revalidated = false; 2329 } 2330 2331 /* More cache consistency checks */ 2332 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) { 2333 if (!have_writers && nfsi->ooo && nfsi->ooo->cnt == 1 && 2334 nfsi->ooo->gap[0].end == inode_peek_iversion_raw(inode)) { 2335 /* There is one remaining gap that hasn't been 2336 * merged into iversion - do that now. 2337 */ 2338 inode_set_iversion_raw(inode, nfsi->ooo->gap[0].start); 2339 kfree(nfsi->ooo); 2340 nfsi->ooo = NULL; 2341 } 2342 if (!inode_eq_iversion_raw(inode, fattr->change_attr)) { 2343 /* Could it be a race with writeback? */ 2344 if (!(have_writers || have_delegation)) { 2345 invalid |= NFS_INO_INVALID_DATA 2346 | NFS_INO_INVALID_ACCESS 2347 | NFS_INO_INVALID_ACL 2348 | NFS_INO_INVALID_XATTR; 2349 /* Force revalidate of all attributes */ 2350 save_cache_validity |= NFS_INO_INVALID_CTIME 2351 | NFS_INO_INVALID_MTIME 2352 | NFS_INO_INVALID_SIZE 2353 | NFS_INO_INVALID_BLOCKS 2354 | NFS_INO_INVALID_NLINK 2355 | NFS_INO_INVALID_MODE 2356 | NFS_INO_INVALID_OTHER 2357 | NFS_INO_INVALID_BTIME; 2358 if (S_ISDIR(inode->i_mode)) 2359 nfs_force_lookup_revalidate(inode); 2360 attr_changed = true; 2361 dprintk("NFS: change_attr change on server for file %s/%ld\n", 2362 inode->i_sb->s_id, 2363 inode->i_ino); 2364 } else if (!have_delegation) { 2365 nfs_ooo_record(nfsi, fattr); 2366 nfs_ooo_merge(nfsi, inode_peek_iversion_raw(inode), 2367 fattr->change_attr); 2368 } 2369 inode_set_iversion_raw(inode, fattr->change_attr); 2370 } 2371 } else { 2372 nfsi->cache_validity |= 2373 save_cache_validity & NFS_INO_INVALID_CHANGE; 2374 if (!have_delegation || 2375 (nfsi->cache_validity & NFS_INO_INVALID_CHANGE) != 0) 2376 cache_revalidated = false; 2377 } 2378 2379 if (fattr->valid & NFS_ATTR_FATTR_MTIME) 2380 inode_set_mtime_to_ts(inode, fattr->mtime); 2381 else if (fattr_supported & NFS_ATTR_FATTR_MTIME) 2382 nfsi->cache_validity |= 2383 save_cache_validity & NFS_INO_INVALID_MTIME; 2384 2385 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 2386 inode_set_ctime_to_ts(inode, fattr->ctime); 2387 else if (fattr_supported & NFS_ATTR_FATTR_CTIME) 2388 nfsi->cache_validity |= 2389 save_cache_validity & NFS_INO_INVALID_CTIME; 2390 2391 if (fattr->valid & NFS_ATTR_FATTR_BTIME) 2392 nfsi->btime = fattr->btime; 2393 else if (fattr_supported & NFS_ATTR_FATTR_BTIME) 2394 nfsi->cache_validity |= 2395 save_cache_validity & NFS_INO_INVALID_BTIME; 2396 2397 /* Check if our cached file size is stale */ 2398 if (fattr->valid & NFS_ATTR_FATTR_SIZE) { 2399 new_isize = nfs_size_to_loff_t(fattr->size); 2400 cur_isize = i_size_read(inode); 2401 if (new_isize != cur_isize && !have_delegation) { 2402 /* Do we perhaps have any outstanding writes, or has 2403 * the file grown beyond our last write? */ 2404 if (!nfs_have_writebacks(inode) || new_isize > cur_isize) { 2405 trace_nfs_size_update(inode, new_isize); 2406 i_size_write(inode, new_isize); 2407 if (!have_writers) 2408 invalid |= NFS_INO_INVALID_DATA; 2409 } 2410 } 2411 if (new_isize == 0 && 2412 !(fattr->valid & (NFS_ATTR_FATTR_SPACE_USED | 2413 NFS_ATTR_FATTR_BLOCKS_USED))) { 2414 fattr->du.nfs3.used = 0; 2415 fattr->valid |= NFS_ATTR_FATTR_SPACE_USED; 2416 } 2417 } else 2418 nfsi->cache_validity |= 2419 save_cache_validity & NFS_INO_INVALID_SIZE; 2420 2421 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 2422 inode_set_atime_to_ts(inode, fattr->atime); 2423 else if (fattr_supported & NFS_ATTR_FATTR_ATIME) 2424 nfsi->cache_validity |= 2425 save_cache_validity & NFS_INO_INVALID_ATIME; 2426 2427 if (fattr->valid & NFS_ATTR_FATTR_MODE) { 2428 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) { 2429 umode_t newmode = inode->i_mode & S_IFMT; 2430 newmode |= fattr->mode & S_IALLUGO; 2431 inode->i_mode = newmode; 2432 invalid |= NFS_INO_INVALID_ACCESS 2433 | NFS_INO_INVALID_ACL; 2434 } 2435 } else if (fattr_supported & NFS_ATTR_FATTR_MODE) 2436 nfsi->cache_validity |= 2437 save_cache_validity & NFS_INO_INVALID_MODE; 2438 2439 if (fattr->valid & NFS_ATTR_FATTR_OWNER) { 2440 if (!uid_eq(inode->i_uid, fattr->uid)) { 2441 invalid |= NFS_INO_INVALID_ACCESS 2442 | NFS_INO_INVALID_ACL; 2443 inode->i_uid = fattr->uid; 2444 } 2445 } else if (fattr_supported & NFS_ATTR_FATTR_OWNER) 2446 nfsi->cache_validity |= 2447 save_cache_validity & NFS_INO_INVALID_OTHER; 2448 2449 if (fattr->valid & NFS_ATTR_FATTR_GROUP) { 2450 if (!gid_eq(inode->i_gid, fattr->gid)) { 2451 invalid |= NFS_INO_INVALID_ACCESS 2452 | NFS_INO_INVALID_ACL; 2453 inode->i_gid = fattr->gid; 2454 } 2455 } else if (fattr_supported & NFS_ATTR_FATTR_GROUP) 2456 nfsi->cache_validity |= 2457 save_cache_validity & NFS_INO_INVALID_OTHER; 2458 2459 if (fattr->valid & NFS_ATTR_FATTR_NLINK) { 2460 if (inode->i_nlink != fattr->nlink) 2461 set_nlink(inode, fattr->nlink); 2462 } else if (fattr_supported & NFS_ATTR_FATTR_NLINK) 2463 nfsi->cache_validity |= 2464 save_cache_validity & NFS_INO_INVALID_NLINK; 2465 2466 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) { 2467 /* 2468 * report the blocks in 512byte units 2469 */ 2470 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 2471 } else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED) 2472 nfsi->cache_validity |= 2473 save_cache_validity & NFS_INO_INVALID_BLOCKS; 2474 2475 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED) 2476 inode->i_blocks = fattr->du.nfs2.blocks; 2477 else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED) 2478 nfsi->cache_validity |= 2479 save_cache_validity & NFS_INO_INVALID_BLOCKS; 2480 2481 /* Update attrtimeo value if we're out of the unstable period */ 2482 if (attr_changed) { 2483 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 2484 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 2485 nfsi->attrtimeo_timestamp = now; 2486 /* Set barrier to be more recent than all outstanding updates */ 2487 nfsi->attr_gencount = nfs_inc_attr_generation_counter(); 2488 } else { 2489 if (cache_revalidated) { 2490 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, 2491 nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) { 2492 nfsi->attrtimeo <<= 1; 2493 if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode)) 2494 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode); 2495 } 2496 nfsi->attrtimeo_timestamp = now; 2497 } 2498 /* Set the barrier to be more recent than this fattr */ 2499 if ((long)(fattr->gencount - nfsi->attr_gencount) > 0) 2500 nfsi->attr_gencount = fattr->gencount; 2501 } 2502 2503 /* Don't invalidate the data if we were to blame */ 2504 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) 2505 || S_ISLNK(inode->i_mode))) 2506 invalid &= ~NFS_INO_INVALID_DATA; 2507 nfs_set_cache_invalid(inode, invalid); 2508 2509 return 0; 2510 out_err: 2511 /* 2512 * No need to worry about unhashing the dentry, as the 2513 * lookup validation will know that the inode is bad. 2514 * (But we fall through to invalidate the caches.) 2515 */ 2516 nfs_set_inode_stale_locked(inode); 2517 return -ESTALE; 2518 } 2519 2520 struct inode *nfs_alloc_inode(struct super_block *sb) 2521 { 2522 struct nfs_inode *nfsi; 2523 nfsi = alloc_inode_sb(sb, nfs_inode_cachep, GFP_KERNEL); 2524 if (!nfsi) 2525 return NULL; 2526 nfsi->flags = 0UL; 2527 nfsi->cache_validity = 0UL; 2528 nfsi->ooo = NULL; 2529 #if IS_ENABLED(CONFIG_NFS_V4) 2530 nfsi->nfs4_acl = NULL; 2531 #endif /* CONFIG_NFS_V4 */ 2532 #ifdef CONFIG_NFS_V4_2 2533 nfsi->xattr_cache = NULL; 2534 #endif 2535 nfs_netfs_inode_init(nfsi); 2536 2537 return &nfsi->vfs_inode; 2538 } 2539 EXPORT_SYMBOL_GPL(nfs_alloc_inode); 2540 2541 void nfs_free_inode(struct inode *inode) 2542 { 2543 kfree(NFS_I(inode)->ooo); 2544 kmem_cache_free(nfs_inode_cachep, NFS_I(inode)); 2545 } 2546 EXPORT_SYMBOL_GPL(nfs_free_inode); 2547 2548 static inline void nfs4_init_once(struct nfs_inode *nfsi) 2549 { 2550 #if IS_ENABLED(CONFIG_NFS_V4) 2551 INIT_LIST_HEAD(&nfsi->open_states); 2552 nfsi->delegation = NULL; 2553 init_rwsem(&nfsi->rwsem); 2554 nfsi->layout = NULL; 2555 #endif 2556 } 2557 2558 static void init_once(void *foo) 2559 { 2560 struct nfs_inode *nfsi = foo; 2561 2562 inode_init_once(&nfsi->vfs_inode); 2563 INIT_LIST_HEAD(&nfsi->open_files); 2564 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru); 2565 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru); 2566 nfs4_init_once(nfsi); 2567 } 2568 2569 static int __init nfs_init_inodecache(void) 2570 { 2571 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache", 2572 sizeof(struct nfs_inode), 2573 0, (SLAB_RECLAIM_ACCOUNT| 2574 SLAB_ACCOUNT), 2575 init_once); 2576 if (nfs_inode_cachep == NULL) 2577 return -ENOMEM; 2578 2579 return 0; 2580 } 2581 2582 static void nfs_destroy_inodecache(void) 2583 { 2584 /* 2585 * Make sure all delayed rcu free inodes are flushed before we 2586 * destroy cache. 2587 */ 2588 rcu_barrier(); 2589 kmem_cache_destroy(nfs_inode_cachep); 2590 } 2591 2592 struct workqueue_struct *nfslocaliod_workqueue; 2593 struct workqueue_struct *nfsiod_workqueue; 2594 EXPORT_SYMBOL_GPL(nfsiod_workqueue); 2595 2596 /* 2597 * Destroy the nfsiod workqueues 2598 */ 2599 static void nfsiod_stop(void) 2600 { 2601 struct workqueue_struct *wq; 2602 2603 wq = nfsiod_workqueue; 2604 if (wq != NULL) { 2605 nfsiod_workqueue = NULL; 2606 destroy_workqueue(wq); 2607 } 2608 #if IS_ENABLED(CONFIG_NFS_LOCALIO) 2609 wq = nfslocaliod_workqueue; 2610 if (wq != NULL) { 2611 nfslocaliod_workqueue = NULL; 2612 destroy_workqueue(wq); 2613 } 2614 #endif /* CONFIG_NFS_LOCALIO */ 2615 } 2616 2617 /* 2618 * Start the nfsiod workqueues 2619 */ 2620 static int nfsiod_start(void) 2621 { 2622 dprintk("RPC: creating workqueue nfsiod\n"); 2623 nfsiod_workqueue = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0); 2624 if (nfsiod_workqueue == NULL) 2625 return -ENOMEM; 2626 #if IS_ENABLED(CONFIG_NFS_LOCALIO) 2627 /* 2628 * localio writes need to use a normal (non-memreclaim) workqueue. 2629 * When we start getting low on space, XFS goes and calls flush_work() on 2630 * a non-memreclaim work queue, which causes a priority inversion problem. 2631 */ 2632 dprintk("RPC: creating workqueue nfslocaliod\n"); 2633 nfslocaliod_workqueue = alloc_workqueue("nfslocaliod", WQ_UNBOUND, 0); 2634 if (unlikely(nfslocaliod_workqueue == NULL)) { 2635 nfsiod_stop(); 2636 return -ENOMEM; 2637 } 2638 #endif /* CONFIG_NFS_LOCALIO */ 2639 return 0; 2640 } 2641 2642 unsigned int nfs_net_id; 2643 EXPORT_SYMBOL_GPL(nfs_net_id); 2644 2645 static int nfs_net_init(struct net *net) 2646 { 2647 struct nfs_net *nn = net_generic(net, nfs_net_id); 2648 int err; 2649 2650 nfs_clients_init(net); 2651 2652 if (!rpc_proc_register(net, &nn->rpcstats)) { 2653 err = -ENOMEM; 2654 goto err_proc_rpc; 2655 } 2656 2657 err = nfs_fs_proc_net_init(net); 2658 if (err) 2659 goto err_proc_nfs; 2660 2661 return 0; 2662 2663 err_proc_nfs: 2664 rpc_proc_unregister(net, "nfs"); 2665 err_proc_rpc: 2666 nfs_clients_exit(net); 2667 return err; 2668 } 2669 2670 static void nfs_net_exit(struct net *net) 2671 { 2672 rpc_proc_unregister(net, "nfs"); 2673 nfs_fs_proc_net_exit(net); 2674 nfs_clients_exit(net); 2675 } 2676 2677 static struct pernet_operations nfs_net_ops = { 2678 .init = nfs_net_init, 2679 .exit = nfs_net_exit, 2680 .id = &nfs_net_id, 2681 .size = sizeof(struct nfs_net), 2682 }; 2683 2684 #ifdef CONFIG_KEYS 2685 static struct key *nfs_keyring; 2686 2687 static int __init nfs_init_keyring(void) 2688 { 2689 nfs_keyring = keyring_alloc(".nfs", 2690 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, 2691 current_cred(), 2692 (KEY_POS_ALL & ~KEY_POS_SETATTR) | 2693 (KEY_USR_ALL & ~KEY_USR_SETATTR), 2694 KEY_ALLOC_NOT_IN_QUOTA, NULL, NULL); 2695 return PTR_ERR_OR_ZERO(nfs_keyring); 2696 } 2697 2698 static void nfs_exit_keyring(void) 2699 { 2700 key_put(nfs_keyring); 2701 } 2702 #else 2703 static inline int nfs_init_keyring(void) 2704 { 2705 return 0; 2706 } 2707 2708 static inline void nfs_exit_keyring(void) 2709 { 2710 } 2711 #endif /* CONFIG_KEYS */ 2712 2713 /* 2714 * Initialize NFS 2715 */ 2716 static int __init init_nfs_fs(void) 2717 { 2718 int err; 2719 2720 err = nfs_init_keyring(); 2721 if (err) 2722 return err; 2723 2724 err = nfs_sysfs_init(); 2725 if (err < 0) 2726 goto out10; 2727 2728 err = register_pernet_subsys(&nfs_net_ops); 2729 if (err < 0) 2730 goto out9; 2731 2732 err = nfsiod_start(); 2733 if (err) 2734 goto out7; 2735 2736 err = nfs_fs_proc_init(); 2737 if (err) 2738 goto out6; 2739 2740 err = nfs_init_nfspagecache(); 2741 if (err) 2742 goto out5; 2743 2744 err = nfs_init_inodecache(); 2745 if (err) 2746 goto out4; 2747 2748 err = nfs_init_readpagecache(); 2749 if (err) 2750 goto out3; 2751 2752 err = nfs_init_writepagecache(); 2753 if (err) 2754 goto out2; 2755 2756 err = nfs_init_directcache(); 2757 if (err) 2758 goto out1; 2759 2760 err = register_nfs_fs(); 2761 if (err) 2762 goto out0; 2763 2764 return 0; 2765 out0: 2766 nfs_destroy_directcache(); 2767 out1: 2768 nfs_destroy_writepagecache(); 2769 out2: 2770 nfs_destroy_readpagecache(); 2771 out3: 2772 nfs_destroy_inodecache(); 2773 out4: 2774 nfs_destroy_nfspagecache(); 2775 out5: 2776 nfs_fs_proc_exit(); 2777 out6: 2778 nfsiod_stop(); 2779 out7: 2780 unregister_pernet_subsys(&nfs_net_ops); 2781 out9: 2782 nfs_sysfs_exit(); 2783 out10: 2784 nfs_exit_keyring(); 2785 return err; 2786 } 2787 2788 static void __exit exit_nfs_fs(void) 2789 { 2790 nfs_destroy_directcache(); 2791 nfs_destroy_writepagecache(); 2792 nfs_destroy_readpagecache(); 2793 nfs_destroy_inodecache(); 2794 nfs_destroy_nfspagecache(); 2795 unregister_pernet_subsys(&nfs_net_ops); 2796 unregister_nfs_fs(); 2797 nfs_fs_proc_exit(); 2798 nfsiod_stop(); 2799 nfs_sysfs_exit(); 2800 nfs_exit_keyring(); 2801 } 2802 2803 /* Not quite true; I just maintain it */ 2804 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>"); 2805 MODULE_DESCRIPTION("NFS client support"); 2806 MODULE_LICENSE("GPL"); 2807 module_param(enable_ino64, bool, 0644); 2808 2809 module_init(init_nfs_fs) 2810 module_exit(exit_nfs_fs) 2811