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 #include <linux/fileattr.h> 45 46 #include "nfs4_fs.h" 47 #include "callback.h" 48 #include "delegation.h" 49 #include "iostat.h" 50 #include "internal.h" 51 #include "fscache.h" 52 #include "pnfs.h" 53 #include "nfs.h" 54 #include "netns.h" 55 #include "sysfs.h" 56 57 #include "nfstrace.h" 58 59 #define NFSDBG_FACILITY NFSDBG_VFS 60 61 #define NFS_64_BIT_INODE_NUMBERS_ENABLED 1 62 63 /* Default is to see 64-bit inode numbers */ 64 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED; 65 66 static int nfs_update_inode(struct inode *, struct nfs_fattr *); 67 68 static struct kmem_cache * nfs_inode_cachep; 69 70 static inline unsigned long 71 nfs_fattr_to_ino_t(struct nfs_fattr *fattr) 72 { 73 return nfs_fileid_to_ino_t(fattr->fileid); 74 } 75 76 int nfs_wait_bit_killable(struct wait_bit_key *key, int mode) 77 { 78 if (unlikely(nfs_current_task_exiting())) 79 return -EINTR; 80 schedule(); 81 if (signal_pending_state(mode, current)) 82 return -ERESTARTSYS; 83 return 0; 84 } 85 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable); 86 87 /** 88 * nfs_compat_user_ino64 - returns the user-visible inode number 89 * @fileid: 64-bit fileid 90 * 91 * This function returns a 32-bit inode number if the boot parameter 92 * nfs.enable_ino64 is zero. 93 */ 94 u64 nfs_compat_user_ino64(u64 fileid) 95 { 96 #ifdef CONFIG_COMPAT 97 compat_ulong_t ino; 98 #else 99 unsigned long ino; 100 #endif 101 102 if (enable_ino64) 103 return fileid; 104 ino = fileid; 105 if (sizeof(ino) < sizeof(fileid)) 106 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8; 107 return ino; 108 } 109 110 int nfs_drop_inode(struct inode *inode) 111 { 112 return NFS_STALE(inode) || inode_generic_drop(inode); 113 } 114 EXPORT_SYMBOL_GPL(nfs_drop_inode); 115 116 void nfs_clear_inode(struct inode *inode) 117 { 118 /* 119 * The following should never happen... 120 */ 121 WARN_ON_ONCE(nfs_have_writebacks(inode)); 122 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files)); 123 nfs_zap_acl_cache(inode); 124 nfs_access_zap_cache(inode); 125 nfs_fscache_clear_inode(inode); 126 } 127 EXPORT_SYMBOL_GPL(nfs_clear_inode); 128 129 void nfs_evict_inode(struct inode *inode) 130 { 131 truncate_inode_pages_final(&inode->i_data); 132 clear_inode(inode); 133 nfs_clear_inode(inode); 134 } 135 136 int nfs_sync_inode(struct inode *inode) 137 { 138 inode_dio_wait(inode); 139 return nfs_wb_all(inode); 140 } 141 EXPORT_SYMBOL_GPL(nfs_sync_inode); 142 143 /** 144 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk 145 * @mapping: pointer to struct address_space 146 */ 147 int nfs_sync_mapping(struct address_space *mapping) 148 { 149 int ret = 0; 150 151 if (mapping->nrpages != 0) { 152 unmap_mapping_range(mapping, 0, 0, 0); 153 ret = nfs_wb_all(mapping->host); 154 } 155 return ret; 156 } 157 158 static int nfs_attribute_timeout(struct inode *inode) 159 { 160 struct nfs_inode *nfsi = NFS_I(inode); 161 162 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo); 163 } 164 165 static bool nfs_check_cache_flags_invalid(struct inode *inode, 166 unsigned long flags) 167 { 168 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity); 169 170 return (cache_validity & flags) != 0; 171 } 172 173 bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags) 174 { 175 if (nfs_check_cache_flags_invalid(inode, flags)) 176 return true; 177 return nfs_attribute_cache_expired(inode); 178 } 179 EXPORT_SYMBOL_GPL(nfs_check_cache_invalid); 180 181 #ifdef CONFIG_NFS_V4_2 182 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi) 183 { 184 return nfsi->xattr_cache != NULL; 185 } 186 #else 187 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi) 188 { 189 return false; 190 } 191 #endif 192 193 void nfs_set_cache_invalid(struct inode *inode, unsigned long flags) 194 { 195 struct nfs_inode *nfsi = NFS_I(inode); 196 197 if (nfs_have_delegated_attributes(inode)) { 198 if (!(flags & NFS_INO_REVAL_FORCED)) 199 flags &= ~(NFS_INO_INVALID_MODE | 200 NFS_INO_INVALID_OTHER | 201 NFS_INO_INVALID_BTIME | 202 NFS_INO_INVALID_XATTR); 203 flags &= ~(NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE); 204 } 205 206 if (!nfs_has_xattr_cache(nfsi)) 207 flags &= ~NFS_INO_INVALID_XATTR; 208 if (flags & NFS_INO_INVALID_DATA) 209 nfs_fscache_invalidate(inode, 0); 210 flags &= ~NFS_INO_REVAL_FORCED; 211 212 flags |= nfsi->cache_validity; 213 if (inode->i_mapping->nrpages == 0) 214 flags &= ~NFS_INO_INVALID_DATA; 215 216 /* pairs with nfs_clear_invalid_mapping()'s smp_load_acquire() */ 217 smp_store_release(&nfsi->cache_validity, flags); 218 219 if (inode->i_mapping->nrpages == 0 || 220 nfsi->cache_validity & NFS_INO_INVALID_DATA) { 221 nfs_ooo_clear(nfsi); 222 } 223 trace_nfs_set_cache_invalid(inode, 0); 224 } 225 EXPORT_SYMBOL_GPL(nfs_set_cache_invalid); 226 227 /* 228 * Invalidate the local caches 229 */ 230 static void nfs_zap_caches_locked(struct inode *inode) 231 { 232 struct nfs_inode *nfsi = NFS_I(inode); 233 int mode = inode->i_mode; 234 235 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 236 237 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 238 nfsi->attrtimeo_timestamp = jiffies; 239 240 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) 241 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR | 242 NFS_INO_INVALID_DATA | 243 NFS_INO_INVALID_ACCESS | 244 NFS_INO_INVALID_ACL | 245 NFS_INO_INVALID_XATTR); 246 else 247 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR | 248 NFS_INO_INVALID_ACCESS | 249 NFS_INO_INVALID_ACL | 250 NFS_INO_INVALID_XATTR); 251 nfs_zap_label_cache_locked(nfsi); 252 } 253 254 void nfs_zap_caches(struct inode *inode) 255 { 256 spin_lock(&inode->i_lock); 257 nfs_zap_caches_locked(inode); 258 spin_unlock(&inode->i_lock); 259 } 260 261 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping) 262 { 263 if (mapping->nrpages != 0) { 264 spin_lock(&inode->i_lock); 265 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA); 266 spin_unlock(&inode->i_lock); 267 } 268 } 269 270 void nfs_zap_acl_cache(struct inode *inode) 271 { 272 void (*clear_acl_cache)(struct inode *); 273 274 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache; 275 if (clear_acl_cache != NULL) 276 clear_acl_cache(inode); 277 spin_lock(&inode->i_lock); 278 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL; 279 spin_unlock(&inode->i_lock); 280 } 281 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache); 282 283 void nfs_invalidate_atime(struct inode *inode) 284 { 285 if (nfs_have_delegated_atime(inode)) 286 return; 287 spin_lock(&inode->i_lock); 288 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME); 289 spin_unlock(&inode->i_lock); 290 } 291 EXPORT_SYMBOL_GPL(nfs_invalidate_atime); 292 293 /* 294 * Invalidate, but do not unhash, the inode. 295 * NB: must be called with inode->i_lock held! 296 */ 297 static void nfs_set_inode_stale_locked(struct inode *inode) 298 { 299 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); 300 nfs_zap_caches_locked(inode); 301 trace_nfs_set_inode_stale(inode); 302 } 303 304 void nfs_set_inode_stale(struct inode *inode) 305 { 306 spin_lock(&inode->i_lock); 307 nfs_set_inode_stale_locked(inode); 308 spin_unlock(&inode->i_lock); 309 } 310 311 struct nfs_find_desc { 312 struct nfs_fh *fh; 313 struct nfs_fattr *fattr; 314 }; 315 316 /* 317 * In NFSv3 we can have 64bit inode numbers. In order to support 318 * this, and re-exported directories (also seen in NFSv2) 319 * we are forced to allow 2 different inodes to have the same 320 * i_ino. 321 */ 322 static int 323 nfs_find_actor(struct inode *inode, void *opaque) 324 { 325 struct nfs_find_desc *desc = opaque; 326 struct nfs_fh *fh = desc->fh; 327 struct nfs_fattr *fattr = desc->fattr; 328 329 if (NFS_FILEID(inode) != fattr->fileid) 330 return 0; 331 if (inode_wrong_type(inode, fattr->mode)) 332 return 0; 333 if (nfs_compare_fh(NFS_FH(inode), fh)) 334 return 0; 335 if (is_bad_inode(inode) || NFS_STALE(inode)) 336 return 0; 337 return 1; 338 } 339 340 static int 341 nfs_init_locked(struct inode *inode, void *opaque) 342 { 343 struct nfs_find_desc *desc = opaque; 344 struct nfs_fattr *fattr = desc->fattr; 345 346 set_nfs_fileid(inode, fattr->fileid); 347 inode->i_mode = fattr->mode; 348 nfs_copy_fh(NFS_FH(inode), desc->fh); 349 return 0; 350 } 351 352 #ifdef CONFIG_NFS_V4_SECURITY_LABEL 353 static void nfs_clear_label_invalid(struct inode *inode) 354 { 355 spin_lock(&inode->i_lock); 356 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL; 357 spin_unlock(&inode->i_lock); 358 } 359 360 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr) 361 { 362 int error; 363 364 if (fattr->label == NULL) 365 return; 366 367 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) { 368 error = security_inode_notifysecctx(inode, fattr->label->label, 369 fattr->label->len); 370 if (error) 371 printk(KERN_ERR "%s() %s %d " 372 "security_inode_notifysecctx() %d\n", 373 __func__, 374 (char *)fattr->label->label, 375 fattr->label->len, error); 376 nfs_clear_label_invalid(inode); 377 } 378 } 379 380 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags) 381 { 382 struct nfs4_label *label; 383 384 if (!(server->caps & NFS_CAP_SECURITY_LABEL)) 385 return NULL; 386 387 label = kzalloc_obj(struct nfs4_label, flags); 388 if (label == NULL) 389 return ERR_PTR(-ENOMEM); 390 391 label->label = kzalloc(NFS4_MAXLABELLEN, flags); 392 if (label->label == NULL) { 393 kfree(label); 394 return ERR_PTR(-ENOMEM); 395 } 396 label->len = NFS4_MAXLABELLEN; 397 398 return label; 399 } 400 EXPORT_SYMBOL_GPL(nfs4_label_alloc); 401 #else 402 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr) 403 { 404 } 405 #endif 406 EXPORT_SYMBOL_GPL(nfs_setsecurity); 407 408 /* Search for inode identified by fh, fileid and i_mode in inode cache. */ 409 struct inode * 410 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh) 411 { 412 struct nfs_find_desc desc = { 413 .fh = fh, 414 .fattr = fattr, 415 }; 416 struct inode *inode; 417 unsigned long hash; 418 419 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) || 420 !(fattr->valid & NFS_ATTR_FATTR_TYPE)) 421 return NULL; 422 423 hash = nfs_fattr_to_ino_t(fattr); 424 inode = ilookup5(sb, hash, nfs_find_actor, &desc); 425 426 dprintk("%s: returning %p\n", __func__, inode); 427 return inode; 428 } 429 430 static void nfs_inode_init_regular(struct nfs_inode *nfsi) 431 { 432 atomic_long_set(&nfsi->nrequests, 0); 433 atomic_long_set(&nfsi->redirtied_pages, 0); 434 INIT_LIST_HEAD(&nfsi->commit_info.list); 435 atomic_long_set(&nfsi->commit_info.ncommit, 0); 436 atomic_set(&nfsi->commit_info.rpcs_out, 0); 437 mutex_init(&nfsi->commit_mutex); 438 } 439 440 static void nfs_inode_init_dir(struct nfs_inode *nfsi) 441 { 442 nfsi->cache_change_attribute = 0; 443 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf)); 444 init_rwsem(&nfsi->rmdir_sem); 445 } 446 447 /* 448 * This is our front-end to iget that looks up inodes by file handle 449 * instead of inode number. 450 */ 451 struct inode * 452 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr) 453 { 454 struct nfs_find_desc desc = { 455 .fh = fh, 456 .fattr = fattr 457 }; 458 struct inode *inode = ERR_PTR(-ENOENT); 459 u64 fattr_supported = NFS_SB(sb)->fattr_valid; 460 unsigned long hash; 461 462 nfs_attr_check_mountpoint(sb, fattr); 463 464 if (nfs_attr_use_mounted_on_fileid(fattr)) 465 fattr->fileid = fattr->mounted_on_fileid; 466 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0) 467 goto out_no_inode; 468 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0) 469 goto out_no_inode; 470 471 hash = nfs_fattr_to_ino_t(fattr); 472 473 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc); 474 if (inode == NULL) { 475 inode = ERR_PTR(-ENOMEM); 476 goto out_no_inode; 477 } 478 479 if (inode_state_read_once(inode) & I_NEW) { 480 struct nfs_inode *nfsi = NFS_I(inode); 481 unsigned long now = jiffies; 482 483 /* We set i_ino for the few things that still rely on it, 484 * such as stat(2) */ 485 inode->i_ino = hash; 486 487 /* We can't support update_atime(), since the server will reset it */ 488 inode->i_flags |= S_NOATIME|S_NOCMTIME; 489 inode->i_mode = fattr->mode; 490 nfsi->cache_validity = 0; 491 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0 492 && (fattr_supported & NFS_ATTR_FATTR_MODE)) 493 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE); 494 /* Why so? Because we want revalidate for devices/FIFOs, and 495 * that's precisely what we have in nfs_file_inode_operations. 496 */ 497 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops; 498 if (S_ISREG(inode->i_mode)) { 499 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops; 500 inode->i_data.a_ops = &nfs_file_aops; 501 nfs_inode_init_regular(nfsi); 502 mapping_set_large_folios(inode->i_mapping); 503 } else if (S_ISDIR(inode->i_mode)) { 504 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops; 505 inode->i_fop = &nfs_dir_operations; 506 inode->i_data.a_ops = &nfs_dir_aops; 507 nfs_inode_init_dir(nfsi); 508 /* Deal with crossing mountpoints */ 509 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT || 510 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) { 511 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) 512 inode->i_op = &nfs_referral_inode_operations; 513 else 514 inode->i_op = &nfs_mountpoint_inode_operations; 515 inode->i_fop = NULL; 516 inode->i_flags |= S_AUTOMOUNT; 517 } 518 } else if (S_ISLNK(inode->i_mode)) { 519 inode->i_op = &nfs_symlink_inode_operations; 520 inode_nohighmem(inode); 521 } else 522 init_special_inode(inode, inode->i_mode, fattr->rdev); 523 524 inode_set_atime(inode, 0, 0); 525 inode_set_mtime(inode, 0, 0); 526 inode_set_ctime(inode, 0, 0); 527 memset(&nfsi->btime, 0, sizeof(nfsi->btime)); 528 inode_set_iversion_raw(inode, 0); 529 inode->i_size = 0; 530 clear_nlink(inode); 531 inode->i_uid = make_kuid(&init_user_ns, -2); 532 inode->i_gid = make_kgid(&init_user_ns, -2); 533 inode->i_blocks = 0; 534 nfsi->write_io = 0; 535 nfsi->read_io = 0; 536 537 nfsi->read_cache_jiffies = fattr->time_start; 538 nfsi->attr_gencount = fattr->gencount; 539 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 540 inode_set_atime_to_ts(inode, fattr->atime); 541 else if (fattr_supported & NFS_ATTR_FATTR_ATIME) 542 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME); 543 if (fattr->valid & NFS_ATTR_FATTR_MTIME) 544 inode_set_mtime_to_ts(inode, fattr->mtime); 545 else if (fattr_supported & NFS_ATTR_FATTR_MTIME) 546 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME); 547 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 548 inode_set_ctime_to_ts(inode, fattr->ctime); 549 else if (fattr_supported & NFS_ATTR_FATTR_CTIME) 550 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME); 551 if (fattr->valid & NFS_ATTR_FATTR_BTIME) 552 nfsi->btime = fattr->btime; 553 else if (fattr_supported & NFS_ATTR_FATTR_BTIME) 554 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BTIME); 555 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) 556 inode_set_iversion_raw(inode, fattr->change_attr); 557 else 558 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE); 559 if (fattr->valid & NFS_ATTR_FATTR_SIZE) 560 inode->i_size = nfs_size_to_loff_t(fattr->size); 561 else 562 nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE); 563 if (fattr->valid & NFS_ATTR_FATTR_NLINK) 564 set_nlink(inode, fattr->nlink); 565 else if (fattr_supported & NFS_ATTR_FATTR_NLINK) 566 nfs_set_cache_invalid(inode, NFS_INO_INVALID_NLINK); 567 else 568 set_nlink(inode, 1); 569 if (fattr->valid & NFS_ATTR_FATTR_OWNER) 570 inode->i_uid = fattr->uid; 571 else if (fattr_supported & NFS_ATTR_FATTR_OWNER) 572 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER); 573 if (fattr->valid & NFS_ATTR_FATTR_GROUP) 574 inode->i_gid = fattr->gid; 575 else if (fattr_supported & NFS_ATTR_FATTR_GROUP) 576 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER); 577 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED) 578 inode->i_blocks = fattr->du.nfs2.blocks; 579 else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED && 580 fattr->size != 0) 581 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS); 582 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) { 583 /* 584 * report the blocks in 512byte units 585 */ 586 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 587 } else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED && 588 fattr->size != 0) 589 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS); 590 591 nfs_setsecurity(inode, fattr); 592 593 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 594 nfsi->attrtimeo_timestamp = now; 595 nfsi->access_cache = RB_ROOT; 596 597 nfs_fscache_init_inode(inode); 598 599 unlock_new_inode(inode); 600 } else { 601 int err = nfs_refresh_inode(inode, fattr); 602 if (err < 0) { 603 iput(inode); 604 inode = ERR_PTR(err); 605 goto out_no_inode; 606 } 607 } 608 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n", 609 inode->i_sb->s_id, 610 (unsigned long long)NFS_FILEID(inode), 611 nfs_display_fhandle_hash(fh), 612 icount_read_once(inode)); 613 614 out: 615 return inode; 616 617 out_no_inode: 618 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode)); 619 goto out; 620 } 621 EXPORT_SYMBOL_GPL(nfs_fhget); 622 623 static void 624 nfs_fattr_fixup_delegated(struct inode *inode, struct nfs_fattr *fattr) 625 { 626 unsigned long cache_validity = NFS_I(inode)->cache_validity; 627 628 if (nfs_have_delegated_mtime(inode)) { 629 if (!(cache_validity & NFS_INO_INVALID_CTIME)) 630 fattr->valid &= ~(NFS_ATTR_FATTR_PRECTIME | 631 NFS_ATTR_FATTR_CTIME); 632 633 if (!(cache_validity & NFS_INO_INVALID_MTIME)) 634 fattr->valid &= ~(NFS_ATTR_FATTR_PREMTIME | 635 NFS_ATTR_FATTR_MTIME); 636 637 if (!(cache_validity & NFS_INO_INVALID_ATIME)) 638 fattr->valid &= ~NFS_ATTR_FATTR_ATIME; 639 } else if (nfs_have_delegated_atime(inode)) { 640 if (!(cache_validity & NFS_INO_INVALID_ATIME)) 641 fattr->valid &= ~NFS_ATTR_FATTR_ATIME; 642 } 643 } 644 645 static void nfs_set_timestamps_to_ts(struct inode *inode, struct iattr *attr) 646 { 647 unsigned int cache_flags = 0; 648 649 if (attr->ia_valid & ATTR_MTIME_SET) { 650 struct timespec64 ctime = inode_get_ctime(inode); 651 struct timespec64 mtime = inode_get_mtime(inode); 652 struct timespec64 now; 653 bool updated = false; 654 655 now = inode_set_ctime_current(inode); 656 if (!timespec64_equal(&now, &ctime)) 657 updated = true; 658 659 inode_set_mtime_to_ts(inode, attr->ia_mtime); 660 if (!timespec64_equal(&now, &mtime)) 661 updated = true; 662 663 inode_maybe_inc_iversion(inode, updated); 664 cache_flags |= NFS_INO_INVALID_CTIME | NFS_INO_INVALID_MTIME; 665 } 666 if (attr->ia_valid & ATTR_ATIME_SET) { 667 inode_set_atime_to_ts(inode, attr->ia_atime); 668 cache_flags |= NFS_INO_INVALID_ATIME; 669 } 670 NFS_I(inode)->cache_validity &= ~cache_flags; 671 } 672 673 static void nfs_update_atime(struct inode *inode) 674 { 675 inode_update_time(inode, FS_UPD_ATIME, 0); 676 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ATIME; 677 } 678 679 static void nfs_update_mtime(struct inode *inode) 680 { 681 inode_update_time(inode, FS_UPD_CMTIME, 0); 682 NFS_I(inode)->cache_validity &= 683 ~(NFS_INO_INVALID_CTIME | NFS_INO_INVALID_MTIME); 684 } 685 686 void nfs_update_delegated_atime(struct inode *inode) 687 { 688 spin_lock(&inode->i_lock); 689 if (nfs_have_delegated_atime(inode)) 690 nfs_update_atime(inode); 691 spin_unlock(&inode->i_lock); 692 } 693 694 void nfs_update_delegated_mtime_locked(struct inode *inode) 695 { 696 if (nfs_have_delegated_mtime(inode) || 697 nfs_have_directory_delegation(inode)) 698 nfs_update_mtime(inode); 699 } 700 701 void nfs_update_delegated_mtime(struct inode *inode) 702 { 703 spin_lock(&inode->i_lock); 704 nfs_update_delegated_mtime_locked(inode); 705 spin_unlock(&inode->i_lock); 706 } 707 EXPORT_SYMBOL_GPL(nfs_update_delegated_mtime); 708 709 #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) 710 711 int 712 nfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry, 713 struct iattr *attr) 714 { 715 struct inode *inode = d_inode(dentry); 716 struct nfs_fattr *fattr; 717 loff_t oldsize; 718 int error = 0; 719 kuid_t task_uid = current_fsuid(); 720 kuid_t owner_uid = inode->i_uid; 721 722 nfs_inc_stats(inode, NFSIOS_VFSSETATTR); 723 724 /* skip mode change if it's just for clearing setuid/setgid */ 725 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID)) 726 attr->ia_valid &= ~ATTR_MODE; 727 728 if (S_ISREG(inode->i_mode)) 729 nfs_file_block_o_direct(NFS_I(inode)); 730 731 oldsize = i_size_read(inode); 732 if (attr->ia_valid & ATTR_SIZE) { 733 BUG_ON(!S_ISREG(inode->i_mode)); 734 735 error = inode_newsize_ok(inode, attr->ia_size); 736 if (error) 737 return error; 738 739 if (attr->ia_size == oldsize) 740 attr->ia_valid &= ~ATTR_SIZE; 741 } 742 743 if (nfs_have_delegated_mtime(inode) && attr->ia_valid & ATTR_MTIME) { 744 spin_lock(&inode->i_lock); 745 if (attr->ia_valid & ATTR_MTIME_SET) { 746 if (uid_eq(task_uid, owner_uid)) { 747 nfs_set_timestamps_to_ts(inode, attr); 748 attr->ia_valid &= ~(ATTR_MTIME|ATTR_MTIME_SET| 749 ATTR_ATIME|ATTR_ATIME_SET); 750 } 751 } else { 752 if (attr->ia_valid & ATTR_MTIME) 753 nfs_update_mtime(inode); 754 if (attr->ia_valid & ATTR_ATIME) 755 nfs_update_atime(inode); 756 attr->ia_valid &= ~(ATTR_MTIME|ATTR_ATIME); 757 } 758 spin_unlock(&inode->i_lock); 759 } else if (nfs_have_delegated_atime(inode) && 760 attr->ia_valid & ATTR_ATIME && 761 !(attr->ia_valid & ATTR_MTIME)) { 762 if (!(attr->ia_valid & ATTR_ATIME_SET)) { 763 nfs_update_delegated_atime(inode); 764 attr->ia_valid &= ~ATTR_ATIME; 765 } 766 } 767 768 /* Optimization: if the end result is no change, don't RPC */ 769 if (((attr->ia_valid & NFS_VALID_ATTRS) & ~(ATTR_FILE|ATTR_OPEN)) == 0) 770 return 0; 771 772 trace_nfs_setattr_enter(inode); 773 774 /* Write all dirty data */ 775 if (S_ISREG(inode->i_mode)) 776 nfs_sync_inode(inode); 777 778 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode)); 779 if (fattr == NULL) { 780 error = -ENOMEM; 781 goto out; 782 } 783 784 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr); 785 if (error == 0) { 786 if (attr->ia_valid & ATTR_SIZE) 787 nfs_truncate_last_folio(inode->i_mapping, oldsize, 788 attr->ia_size); 789 error = nfs_refresh_inode(inode, fattr); 790 } 791 nfs_free_fattr(fattr); 792 out: 793 trace_nfs_setattr_exit(inode, error); 794 return error; 795 } 796 EXPORT_SYMBOL_GPL(nfs_setattr); 797 798 /** 799 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall 800 * @inode: inode of the file used 801 * @offset: file offset to start truncating 802 * 803 * This is a copy of the common vmtruncate, but with the locking 804 * corrected to take into account the fact that NFS requires 805 * inode->i_size to be updated under the inode->i_lock. 806 * Note: must be called with inode->i_lock held! 807 */ 808 static int nfs_vmtruncate(struct inode * inode, loff_t offset) 809 { 810 int err; 811 812 err = inode_newsize_ok(inode, offset); 813 if (err) 814 goto out; 815 816 trace_nfs_size_truncate(inode, offset); 817 i_size_write(inode, offset); 818 /* Optimisation */ 819 if (offset == 0) { 820 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA; 821 nfs_ooo_clear(NFS_I(inode)); 822 } 823 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE; 824 825 spin_unlock(&inode->i_lock); 826 truncate_pagecache(inode, offset); 827 nfs_update_delegated_mtime_locked(inode); 828 spin_lock(&inode->i_lock); 829 out: 830 return err; 831 } 832 833 /** 834 * nfs_setattr_update_inode - Update inode metadata after a setattr call. 835 * @inode: pointer to struct inode 836 * @attr: pointer to struct iattr 837 * @fattr: pointer to struct nfs_fattr 838 * 839 * Note: we do this in the *proc.c in order to ensure that 840 * it works for things like exclusive creates too. 841 */ 842 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr, 843 struct nfs_fattr *fattr) 844 { 845 /* Barrier: bump the attribute generation count. */ 846 nfs_fattr_set_barrier(fattr); 847 848 spin_lock(&inode->i_lock); 849 NFS_I(inode)->attr_gencount = fattr->gencount; 850 if ((attr->ia_valid & ATTR_SIZE) != 0) { 851 if (!nfs_have_delegated_mtime(inode)) 852 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME); 853 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS); 854 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC); 855 nfs_vmtruncate(inode, attr->ia_size); 856 } 857 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) { 858 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME; 859 if ((attr->ia_valid & ATTR_KILL_SUID) != 0 && 860 inode->i_mode & S_ISUID) 861 inode->i_mode &= ~S_ISUID; 862 if (setattr_should_drop_sgid(&nop_mnt_idmap, inode)) 863 inode->i_mode &= ~S_ISGID; 864 if ((attr->ia_valid & ATTR_MODE) != 0) { 865 int mode = attr->ia_mode & S_IALLUGO; 866 mode |= inode->i_mode & ~S_IALLUGO; 867 inode->i_mode = mode; 868 } 869 if ((attr->ia_valid & ATTR_UID) != 0) 870 inode->i_uid = attr->ia_uid; 871 if ((attr->ia_valid & ATTR_GID) != 0) 872 inode->i_gid = attr->ia_gid; 873 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 874 inode_set_ctime_to_ts(inode, fattr->ctime); 875 else 876 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE 877 | NFS_INO_INVALID_CTIME); 878 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS 879 | NFS_INO_INVALID_ACL); 880 } 881 if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) { 882 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME 883 | NFS_INO_INVALID_CTIME); 884 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 885 inode_set_atime_to_ts(inode, fattr->atime); 886 else if (attr->ia_valid & ATTR_ATIME_SET) 887 inode_set_atime_to_ts(inode, attr->ia_atime); 888 else 889 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME); 890 891 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 892 inode_set_ctime_to_ts(inode, fattr->ctime); 893 else 894 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE 895 | NFS_INO_INVALID_CTIME); 896 } 897 if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) { 898 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME 899 | NFS_INO_INVALID_CTIME); 900 if (fattr->valid & NFS_ATTR_FATTR_MTIME) 901 inode_set_mtime_to_ts(inode, fattr->mtime); 902 else if (attr->ia_valid & ATTR_MTIME_SET) 903 inode_set_mtime_to_ts(inode, attr->ia_mtime); 904 else 905 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME); 906 907 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 908 inode_set_ctime_to_ts(inode, fattr->ctime); 909 else 910 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE 911 | NFS_INO_INVALID_CTIME); 912 } 913 if (fattr->valid) 914 nfs_update_inode(inode, fattr); 915 spin_unlock(&inode->i_lock); 916 } 917 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode); 918 919 /* 920 * Don't request help from readdirplus if the file is being written to, 921 * or if attribute caching is turned off 922 */ 923 static bool nfs_getattr_readdirplus_enable(const struct inode *inode) 924 { 925 return nfs_server_capable(inode, NFS_CAP_READDIRPLUS) && 926 !nfs_have_writebacks(inode) && NFS_MAXATTRTIMEO(inode) > 5 * HZ; 927 } 928 929 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry) 930 { 931 if (!IS_ROOT(dentry)) { 932 struct dentry *parent = dget_parent(dentry); 933 nfs_readdir_record_entry_cache_miss(d_inode(parent)); 934 dput(parent); 935 } 936 } 937 938 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry) 939 { 940 if (!IS_ROOT(dentry)) { 941 struct dentry *parent = dget_parent(dentry); 942 nfs_readdir_record_entry_cache_hit(d_inode(parent)); 943 dput(parent); 944 } 945 } 946 947 static u32 nfs_get_valid_attrmask(struct inode *inode) 948 { 949 u64 fattr_valid = NFS_SERVER(inode)->fattr_valid; 950 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity); 951 u32 reply_mask = STATX_INO | STATX_TYPE; 952 953 if (!(cache_validity & NFS_INO_INVALID_ATIME)) 954 reply_mask |= STATX_ATIME; 955 if (!(cache_validity & NFS_INO_INVALID_CTIME)) 956 reply_mask |= STATX_CTIME; 957 if (!(cache_validity & NFS_INO_INVALID_MTIME)) 958 reply_mask |= STATX_MTIME; 959 if (!(cache_validity & NFS_INO_INVALID_SIZE)) 960 reply_mask |= STATX_SIZE; 961 if (!(cache_validity & NFS_INO_INVALID_NLINK)) 962 reply_mask |= STATX_NLINK; 963 if (!(cache_validity & NFS_INO_INVALID_MODE)) 964 reply_mask |= STATX_MODE; 965 if (!(cache_validity & NFS_INO_INVALID_OTHER)) 966 reply_mask |= STATX_UID | STATX_GID; 967 if (!(cache_validity & NFS_INO_INVALID_BLOCKS)) 968 reply_mask |= STATX_BLOCKS; 969 if (!(cache_validity & NFS_INO_INVALID_BTIME) && 970 (fattr_valid & NFS_ATTR_FATTR_BTIME)) 971 reply_mask |= STATX_BTIME; 972 if (!(cache_validity & NFS_INO_INVALID_CHANGE)) 973 reply_mask |= STATX_CHANGE_COOKIE; 974 return reply_mask; 975 } 976 977 int nfs_getattr(struct mnt_idmap *idmap, const struct path *path, 978 struct kstat *stat, u32 request_mask, unsigned int query_flags) 979 { 980 struct inode *inode = d_inode(path->dentry); 981 struct nfs_server *server = NFS_SERVER(inode); 982 u64 fattr_valid = server->fattr_valid; 983 unsigned long cache_validity; 984 int err = 0; 985 bool force_sync = query_flags & AT_STATX_FORCE_SYNC; 986 bool do_update = false; 987 bool readdirplus_enabled = nfs_getattr_readdirplus_enable(inode); 988 989 trace_nfs_getattr_enter(inode); 990 991 request_mask &= STATX_TYPE | STATX_MODE | STATX_NLINK | STATX_UID | 992 STATX_GID | STATX_ATIME | STATX_MTIME | STATX_CTIME | 993 STATX_INO | STATX_SIZE | STATX_BLOCKS | STATX_BTIME | 994 STATX_CHANGE_COOKIE; 995 996 if (!(fattr_valid & NFS_ATTR_FATTR_BTIME)) 997 request_mask &= ~STATX_BTIME; 998 999 if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) { 1000 if (readdirplus_enabled) 1001 nfs_readdirplus_parent_cache_hit(path->dentry); 1002 goto out_no_revalidate; 1003 } 1004 1005 /* Flush out writes to the server in order to update c/mtime/version. */ 1006 if ((request_mask & (STATX_CTIME | STATX_MTIME | STATX_CHANGE_COOKIE)) && 1007 S_ISREG(inode->i_mode)) { 1008 if (nfs_have_delegated_mtime(inode)) 1009 filemap_fdatawrite(inode->i_mapping); 1010 else 1011 filemap_write_and_wait(inode->i_mapping); 1012 } 1013 1014 /* 1015 * We may force a getattr if the user cares about atime. 1016 * 1017 * Note that we only have to check the vfsmount flags here: 1018 * - NFS always sets S_NOATIME by so checking it would give a 1019 * bogus result 1020 * - NFS never sets SB_NOATIME or SB_NODIRATIME so there is 1021 * no point in checking those. 1022 */ 1023 if ((path->mnt->mnt_flags & MNT_NOATIME) || 1024 ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) 1025 request_mask &= ~STATX_ATIME; 1026 1027 /* Is the user requesting attributes that might need revalidation? */ 1028 if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME| 1029 STATX_MTIME|STATX_UID|STATX_GID| 1030 STATX_SIZE|STATX_BLOCKS|STATX_BTIME| 1031 STATX_CHANGE_COOKIE))) 1032 goto out_no_revalidate; 1033 1034 /* Check whether the cached attributes are stale */ 1035 do_update |= force_sync || nfs_attribute_cache_expired(inode); 1036 cache_validity = READ_ONCE(NFS_I(inode)->cache_validity); 1037 do_update |= cache_validity & NFS_INO_INVALID_CHANGE; 1038 if (request_mask & STATX_ATIME) 1039 do_update |= cache_validity & NFS_INO_INVALID_ATIME; 1040 if (request_mask & STATX_CTIME) 1041 do_update |= cache_validity & NFS_INO_INVALID_CTIME; 1042 if (request_mask & STATX_MTIME) 1043 do_update |= cache_validity & NFS_INO_INVALID_MTIME; 1044 if (request_mask & STATX_SIZE) 1045 do_update |= cache_validity & NFS_INO_INVALID_SIZE; 1046 if (request_mask & STATX_NLINK) 1047 do_update |= cache_validity & NFS_INO_INVALID_NLINK; 1048 if (request_mask & STATX_MODE) 1049 do_update |= cache_validity & NFS_INO_INVALID_MODE; 1050 if (request_mask & (STATX_UID | STATX_GID)) 1051 do_update |= cache_validity & NFS_INO_INVALID_OTHER; 1052 if (request_mask & STATX_BLOCKS) 1053 do_update |= cache_validity & NFS_INO_INVALID_BLOCKS; 1054 if (request_mask & STATX_BTIME) 1055 do_update |= cache_validity & NFS_INO_INVALID_BTIME; 1056 1057 if (do_update) { 1058 if (readdirplus_enabled) 1059 nfs_readdirplus_parent_cache_miss(path->dentry); 1060 err = __nfs_revalidate_inode(server, inode); 1061 if (err) 1062 goto out; 1063 } else if (readdirplus_enabled) 1064 nfs_readdirplus_parent_cache_hit(path->dentry); 1065 out_no_revalidate: 1066 /* Only return attributes that were revalidated. */ 1067 stat->result_mask = nfs_get_valid_attrmask(inode) | request_mask; 1068 1069 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat); 1070 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode)); 1071 stat->change_cookie = inode_peek_iversion_raw(inode); 1072 stat->attributes_mask |= STATX_ATTR_CHANGE_MONOTONIC; 1073 if (server->change_attr_type != NFS4_CHANGE_TYPE_IS_UNDEFINED) 1074 stat->attributes |= STATX_ATTR_CHANGE_MONOTONIC; 1075 if (S_ISDIR(inode->i_mode)) 1076 stat->blksize = NFS_SERVER(inode)->dtsize; 1077 stat->btime = NFS_I(inode)->btime; 1078 1079 /* Special handling for STATX_DIOALIGN and STATX_DIO_READ_ALIGN 1080 * - NFS doesn't have DIO alignment constraints, avoid getting 1081 * these DIO attrs from remote and just respond with most 1082 * accommodating limits (so client will issue supported DIO). 1083 * - this is unintuitive, but the most coarse-grained 1084 * dio_offset_align is the most accommodating. 1085 */ 1086 if ((request_mask & (STATX_DIOALIGN | STATX_DIO_READ_ALIGN)) && 1087 S_ISREG(inode->i_mode)) { 1088 stat->result_mask |= STATX_DIOALIGN | STATX_DIO_READ_ALIGN; 1089 stat->dio_mem_align = 4; /* 4-byte alignment */ 1090 stat->dio_offset_align = PAGE_SIZE; 1091 stat->dio_read_offset_align = stat->dio_offset_align; 1092 } 1093 out: 1094 trace_nfs_getattr_exit(inode, err); 1095 return err; 1096 } 1097 EXPORT_SYMBOL_GPL(nfs_getattr); 1098 1099 int nfs_fileattr_get(struct dentry *dentry, struct file_kattr *fa) 1100 { 1101 struct inode *inode = d_inode(dentry); 1102 1103 if (nfs_server_capable(inode, NFS_CAP_CASE_INSENSITIVE)) { 1104 fa->fsx_xflags |= FS_XFLAG_CASEFOLD; 1105 fa->flags |= FS_CASEFOLD_FL; 1106 } 1107 if (nfs_server_capable(inode, NFS_CAP_CASE_NONPRESERVING)) 1108 fa->fsx_xflags |= FS_XFLAG_CASENONPRESERVING; 1109 return 0; 1110 } 1111 EXPORT_SYMBOL_GPL(nfs_fileattr_get); 1112 1113 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx) 1114 { 1115 refcount_set(&l_ctx->count, 1); 1116 l_ctx->lockowner = current->files; 1117 INIT_LIST_HEAD(&l_ctx->list); 1118 atomic_set(&l_ctx->io_count, 0); 1119 } 1120 1121 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx) 1122 { 1123 struct nfs_lock_context *pos; 1124 1125 list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) { 1126 if (pos->lockowner != current->files) 1127 continue; 1128 if (refcount_inc_not_zero(&pos->count)) 1129 return pos; 1130 } 1131 return NULL; 1132 } 1133 1134 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx) 1135 { 1136 struct nfs_lock_context *res, *new = NULL; 1137 struct inode *inode = d_inode(ctx->dentry); 1138 1139 rcu_read_lock(); 1140 res = __nfs_find_lock_context(ctx); 1141 rcu_read_unlock(); 1142 if (res == NULL) { 1143 new = kmalloc_obj(*new, GFP_KERNEL_ACCOUNT); 1144 if (new == NULL) 1145 return ERR_PTR(-ENOMEM); 1146 nfs_init_lock_context(new); 1147 spin_lock(&inode->i_lock); 1148 res = __nfs_find_lock_context(ctx); 1149 if (res == NULL) { 1150 new->open_context = get_nfs_open_context(ctx); 1151 if (new->open_context) { 1152 list_add_tail_rcu(&new->list, 1153 &ctx->lock_context.list); 1154 res = new; 1155 new = NULL; 1156 } else 1157 res = ERR_PTR(-EBADF); 1158 } 1159 spin_unlock(&inode->i_lock); 1160 kfree(new); 1161 } 1162 return res; 1163 } 1164 EXPORT_SYMBOL_GPL(nfs_get_lock_context); 1165 1166 void nfs_put_lock_context(struct nfs_lock_context *l_ctx) 1167 { 1168 struct nfs_open_context *ctx = l_ctx->open_context; 1169 struct inode *inode = d_inode(ctx->dentry); 1170 1171 if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock)) 1172 return; 1173 list_del_rcu(&l_ctx->list); 1174 spin_unlock(&inode->i_lock); 1175 put_nfs_open_context(ctx); 1176 kfree_rcu(l_ctx, rcu_head); 1177 } 1178 EXPORT_SYMBOL_GPL(nfs_put_lock_context); 1179 1180 /** 1181 * nfs_close_context - Common close_context() routine NFSv2/v3 1182 * @ctx: pointer to context 1183 * @is_sync: is this a synchronous close 1184 * 1185 * Ensure that the attributes are up to date if we're mounted 1186 * with close-to-open semantics and we have cached data that will 1187 * need to be revalidated on open. 1188 */ 1189 void nfs_close_context(struct nfs_open_context *ctx, int is_sync) 1190 { 1191 struct nfs_inode *nfsi; 1192 struct inode *inode; 1193 1194 if (!(ctx->mode & FMODE_WRITE)) 1195 return; 1196 if (!is_sync) 1197 return; 1198 inode = d_inode(ctx->dentry); 1199 if (nfs_have_read_or_write_delegation(inode)) 1200 return; 1201 nfsi = NFS_I(inode); 1202 if (inode->i_mapping->nrpages == 0) 1203 return; 1204 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 1205 return; 1206 if (!list_empty(&nfsi->open_files)) 1207 return; 1208 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NOCTO) 1209 return; 1210 nfs_revalidate_inode(inode, 1211 NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE); 1212 } 1213 EXPORT_SYMBOL_GPL(nfs_close_context); 1214 1215 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, 1216 fmode_t f_mode, 1217 struct file *filp) 1218 { 1219 struct nfs_open_context *ctx; 1220 1221 ctx = kmalloc_obj(*ctx, GFP_KERNEL_ACCOUNT); 1222 if (!ctx) 1223 return ERR_PTR(-ENOMEM); 1224 nfs_sb_active(dentry->d_sb); 1225 ctx->dentry = dget(dentry); 1226 if (filp) 1227 ctx->cred = get_cred(filp->f_cred); 1228 else 1229 ctx->cred = get_current_cred(); 1230 rcu_assign_pointer(ctx->ll_cred, NULL); 1231 ctx->state = NULL; 1232 ctx->mode = f_mode; 1233 ctx->flags = 0; 1234 ctx->error = 0; 1235 ctx->flock_owner = (fl_owner_t)filp; 1236 nfs_init_lock_context(&ctx->lock_context); 1237 ctx->lock_context.open_context = ctx; 1238 INIT_LIST_HEAD(&ctx->list); 1239 ctx->mdsthreshold = NULL; 1240 nfs_localio_file_init(&ctx->nfl); 1241 1242 return ctx; 1243 } 1244 EXPORT_SYMBOL_GPL(alloc_nfs_open_context); 1245 1246 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx) 1247 { 1248 if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count)) 1249 return ctx; 1250 return NULL; 1251 } 1252 EXPORT_SYMBOL_GPL(get_nfs_open_context); 1253 1254 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync) 1255 { 1256 struct inode *inode = d_inode(ctx->dentry); 1257 struct super_block *sb = ctx->dentry->d_sb; 1258 1259 if (!refcount_dec_and_test(&ctx->lock_context.count)) 1260 return; 1261 if (!list_empty(&ctx->list)) { 1262 spin_lock(&inode->i_lock); 1263 list_del_rcu(&ctx->list); 1264 spin_unlock(&inode->i_lock); 1265 } 1266 if (inode != NULL) 1267 NFS_PROTO(inode)->close_context(ctx, is_sync); 1268 put_cred(ctx->cred); 1269 dput(ctx->dentry); 1270 nfs_sb_deactive(sb); 1271 put_rpccred(rcu_dereference_protected(ctx->ll_cred, 1)); 1272 kfree(ctx->mdsthreshold); 1273 nfs_close_local_fh(&ctx->nfl); 1274 kfree_rcu(ctx, rcu_head); 1275 } 1276 1277 void put_nfs_open_context(struct nfs_open_context *ctx) 1278 { 1279 __put_nfs_open_context(ctx, 0); 1280 } 1281 EXPORT_SYMBOL_GPL(put_nfs_open_context); 1282 1283 static void put_nfs_open_context_sync(struct nfs_open_context *ctx) 1284 { 1285 __put_nfs_open_context(ctx, 1); 1286 } 1287 1288 /* 1289 * Ensure that mmap has a recent RPC credential for use when writing out 1290 * shared pages 1291 */ 1292 void nfs_inode_attach_open_context(struct nfs_open_context *ctx) 1293 { 1294 struct inode *inode = d_inode(ctx->dentry); 1295 struct nfs_inode *nfsi = NFS_I(inode); 1296 1297 spin_lock(&inode->i_lock); 1298 if (list_empty(&nfsi->open_files) && 1299 nfs_ooo_test(nfsi)) 1300 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA | 1301 NFS_INO_REVAL_FORCED); 1302 list_add_tail_rcu(&ctx->list, &nfsi->open_files); 1303 spin_unlock(&inode->i_lock); 1304 } 1305 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context); 1306 1307 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx) 1308 { 1309 filp->private_data = get_nfs_open_context(ctx); 1310 set_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags); 1311 if (list_empty(&ctx->list)) 1312 nfs_inode_attach_open_context(ctx); 1313 } 1314 EXPORT_SYMBOL_GPL(nfs_file_set_open_context); 1315 1316 /* 1317 * Given an inode, search for an open context with the desired characteristics 1318 */ 1319 struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode) 1320 { 1321 struct nfs_inode *nfsi = NFS_I(inode); 1322 struct nfs_open_context *pos, *ctx = NULL; 1323 1324 rcu_read_lock(); 1325 list_for_each_entry_rcu(pos, &nfsi->open_files, list) { 1326 if (cred != NULL && cred_fscmp(pos->cred, cred) != 0) 1327 continue; 1328 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode) 1329 continue; 1330 if (!test_bit(NFS_CONTEXT_FILE_OPEN, &pos->flags)) 1331 continue; 1332 ctx = get_nfs_open_context(pos); 1333 if (ctx) 1334 break; 1335 } 1336 rcu_read_unlock(); 1337 return ctx; 1338 } 1339 1340 void nfs_file_clear_open_context(struct file *filp) 1341 { 1342 struct nfs_open_context *ctx = nfs_file_open_context(filp); 1343 1344 if (ctx) { 1345 struct inode *inode = d_inode(ctx->dentry); 1346 1347 clear_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags); 1348 /* 1349 * We fatal error on write before. Try to writeback 1350 * every page again. 1351 */ 1352 if (ctx->error < 0) 1353 invalidate_inode_pages2(inode->i_mapping); 1354 filp->private_data = NULL; 1355 put_nfs_open_context_sync(ctx); 1356 } 1357 } 1358 1359 /* 1360 * These allocate and release file read/write context information. 1361 */ 1362 int nfs_open(struct inode *inode, struct file *filp) 1363 { 1364 struct nfs_open_context *ctx; 1365 1366 ctx = alloc_nfs_open_context(file_dentry(filp), 1367 flags_to_mode(filp->f_flags), filp); 1368 if (IS_ERR(ctx)) 1369 return PTR_ERR(ctx); 1370 nfs_file_set_open_context(filp, ctx); 1371 put_nfs_open_context(ctx); 1372 nfs_fscache_open_file(inode, filp); 1373 return 0; 1374 } 1375 1376 /* 1377 * This function is called whenever some part of NFS notices that 1378 * the cached attributes have to be refreshed. 1379 */ 1380 int 1381 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) 1382 { 1383 int status = -ESTALE; 1384 struct nfs_fattr *fattr = NULL; 1385 struct nfs_inode *nfsi = NFS_I(inode); 1386 1387 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n", 1388 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode)); 1389 1390 trace_nfs_revalidate_inode_enter(inode); 1391 1392 if (is_bad_inode(inode)) 1393 goto out; 1394 if (NFS_STALE(inode)) 1395 goto out; 1396 1397 /* pNFS: Attributes aren't updated until we layoutcommit */ 1398 if (S_ISREG(inode->i_mode)) { 1399 status = pnfs_sync_inode(inode, false); 1400 if (status) 1401 goto out; 1402 } else if (nfs_have_directory_delegation(inode)) { 1403 status = 0; 1404 goto out; 1405 } 1406 1407 status = -ENOMEM; 1408 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode)); 1409 if (fattr == NULL) 1410 goto out; 1411 1412 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE); 1413 1414 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, inode); 1415 if (status != 0) { 1416 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n", 1417 inode->i_sb->s_id, 1418 (unsigned long long)NFS_FILEID(inode), status); 1419 switch (status) { 1420 case -ETIMEDOUT: 1421 /* A soft timeout occurred. Use cached information? */ 1422 if (server->flags & NFS_MOUNT_SOFTREVAL) 1423 status = 0; 1424 break; 1425 case -ESTALE: 1426 if (!S_ISDIR(inode->i_mode)) 1427 nfs_set_inode_stale(inode); 1428 else 1429 nfs_zap_caches(inode); 1430 } 1431 goto out; 1432 } 1433 1434 status = nfs_refresh_inode(inode, fattr); 1435 if (status) { 1436 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n", 1437 inode->i_sb->s_id, 1438 (unsigned long long)NFS_FILEID(inode), status); 1439 goto out; 1440 } 1441 1442 if (nfsi->cache_validity & NFS_INO_INVALID_ACL) 1443 nfs_zap_acl_cache(inode); 1444 1445 nfs_setsecurity(inode, fattr); 1446 1447 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n", 1448 inode->i_sb->s_id, 1449 (unsigned long long)NFS_FILEID(inode)); 1450 1451 out: 1452 nfs_free_fattr(fattr); 1453 trace_nfs_revalidate_inode_exit(inode, status); 1454 return status; 1455 } 1456 1457 int nfs_attribute_cache_expired(struct inode *inode) 1458 { 1459 if (nfs_have_delegated_attributes(inode)) 1460 return 0; 1461 return nfs_attribute_timeout(inode); 1462 } 1463 1464 /** 1465 * nfs_revalidate_inode - Revalidate the inode attributes 1466 * @inode: pointer to inode struct 1467 * @flags: cache flags to check 1468 * 1469 * Updates inode attribute information by retrieving the data from the server. 1470 */ 1471 int nfs_revalidate_inode(struct inode *inode, unsigned long flags) 1472 { 1473 if (!nfs_check_cache_invalid(inode, flags)) 1474 return NFS_STALE(inode) ? -ESTALE : 0; 1475 return __nfs_revalidate_inode(NFS_SERVER(inode), inode); 1476 } 1477 EXPORT_SYMBOL_GPL(nfs_revalidate_inode); 1478 1479 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping) 1480 { 1481 int ret; 1482 1483 nfs_fscache_invalidate(inode, 0); 1484 if (mapping->nrpages != 0) { 1485 if (S_ISREG(inode->i_mode)) { 1486 ret = nfs_sync_mapping(mapping); 1487 if (ret < 0) 1488 return ret; 1489 } 1490 ret = invalidate_inode_pages2(mapping); 1491 if (ret < 0) 1492 return ret; 1493 } 1494 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE); 1495 1496 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n", 1497 inode->i_sb->s_id, 1498 (unsigned long long)NFS_FILEID(inode)); 1499 return 0; 1500 } 1501 1502 /** 1503 * nfs_clear_invalid_mapping - Conditionally clear a mapping 1504 * @mapping: pointer to mapping 1505 * 1506 * If the NFS_INO_INVALID_DATA inode flag is set, clear the mapping. 1507 */ 1508 int nfs_clear_invalid_mapping(struct address_space *mapping) 1509 { 1510 struct inode *inode = mapping->host; 1511 struct nfs_inode *nfsi = NFS_I(inode); 1512 unsigned long *bitlock = &nfsi->flags; 1513 int ret = 0; 1514 1515 /* 1516 * We must clear NFS_INO_INVALID_DATA first to ensure that 1517 * invalidations that come in while we're shooting down the mappings 1518 * are respected. But, that leaves a race window where one revalidator 1519 * can clear the flag, and then another checks it before the mapping 1520 * gets invalidated. Fix that by serializing access to this part of 1521 * the function. 1522 * 1523 * At the same time, we need to allow other tasks to see whether we 1524 * might be in the middle of invalidating the pages, so we only set 1525 * the bit lock here if it looks like we're going to be doing that. 1526 */ 1527 for (;;) { 1528 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING, 1529 nfs_wait_bit_killable, 1530 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE); 1531 if (ret) 1532 goto out; 1533 smp_rmb(); /* pairs with smp_wmb() below */ 1534 if (test_bit(NFS_INO_INVALIDATING, bitlock)) 1535 continue; 1536 /* pairs with nfs_set_cache_invalid()'s smp_store_release() */ 1537 if (!(smp_load_acquire(&nfsi->cache_validity) & NFS_INO_INVALID_DATA)) 1538 goto out; 1539 /* Slow-path that double-checks with spinlock held */ 1540 spin_lock(&inode->i_lock); 1541 if (test_bit(NFS_INO_INVALIDATING, bitlock)) { 1542 spin_unlock(&inode->i_lock); 1543 continue; 1544 } 1545 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 1546 break; 1547 spin_unlock(&inode->i_lock); 1548 goto out; 1549 } 1550 1551 set_bit(NFS_INO_INVALIDATING, bitlock); 1552 smp_wmb(); 1553 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA; 1554 nfs_ooo_clear(nfsi); 1555 spin_unlock(&inode->i_lock); 1556 trace_nfs_invalidate_mapping_enter(inode); 1557 ret = nfs_invalidate_mapping(inode, mapping); 1558 trace_nfs_invalidate_mapping_exit(inode, ret); 1559 1560 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock); 1561 smp_mb__after_atomic(); 1562 wake_up_bit(bitlock, NFS_INO_INVALIDATING); 1563 out: 1564 return ret; 1565 } 1566 1567 bool nfs_mapping_need_revalidate_inode(struct inode *inode) 1568 { 1569 return nfs_check_cache_invalid(inode, NFS_INO_INVALID_CHANGE) || 1570 NFS_STALE(inode); 1571 } 1572 1573 int nfs_revalidate_mapping_rcu(struct inode *inode) 1574 { 1575 struct nfs_inode *nfsi = NFS_I(inode); 1576 unsigned long *bitlock = &nfsi->flags; 1577 int ret = 0; 1578 1579 if (IS_SWAPFILE(inode)) 1580 goto out; 1581 if (nfs_mapping_need_revalidate_inode(inode)) { 1582 ret = -ECHILD; 1583 goto out; 1584 } 1585 spin_lock(&inode->i_lock); 1586 if (test_bit(NFS_INO_INVALIDATING, bitlock) || 1587 (nfsi->cache_validity & NFS_INO_INVALID_DATA)) 1588 ret = -ECHILD; 1589 spin_unlock(&inode->i_lock); 1590 out: 1591 return ret; 1592 } 1593 1594 /** 1595 * nfs_revalidate_mapping - Revalidate the pagecache 1596 * @inode: pointer to host inode 1597 * @mapping: pointer to mapping 1598 */ 1599 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping) 1600 { 1601 /* swapfiles are not supposed to be shared. */ 1602 if (IS_SWAPFILE(inode)) 1603 return 0; 1604 1605 if (nfs_mapping_need_revalidate_inode(inode)) { 1606 int ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 1607 if (ret < 0) 1608 return ret; 1609 } 1610 1611 return nfs_clear_invalid_mapping(mapping); 1612 } 1613 1614 static bool nfs_file_has_writers(struct nfs_inode *nfsi) 1615 { 1616 struct inode *inode = &nfsi->vfs_inode; 1617 1618 if (!S_ISREG(inode->i_mode)) 1619 return false; 1620 if (list_empty(&nfsi->open_files)) 1621 return false; 1622 return inode_is_open_for_write(inode); 1623 } 1624 1625 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi) 1626 { 1627 return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi); 1628 } 1629 1630 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1631 { 1632 struct timespec64 ts; 1633 1634 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE) 1635 && (fattr->valid & NFS_ATTR_FATTR_CHANGE) 1636 && inode_eq_iversion_raw(inode, fattr->pre_change_attr)) { 1637 inode_set_iversion_raw(inode, fattr->change_attr); 1638 if (S_ISDIR(inode->i_mode)) 1639 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA); 1640 else if (nfs_server_capable(inode, NFS_CAP_XATTR)) 1641 nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR); 1642 } 1643 /* If we have atomic WCC data, we may update some attributes */ 1644 ts = inode_get_ctime(inode); 1645 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME) 1646 && (fattr->valid & NFS_ATTR_FATTR_CTIME) 1647 && timespec64_equal(&ts, &fattr->pre_ctime)) { 1648 inode_set_ctime_to_ts(inode, fattr->ctime); 1649 } 1650 1651 ts = inode_get_mtime(inode); 1652 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME) 1653 && (fattr->valid & NFS_ATTR_FATTR_MTIME) 1654 && timespec64_equal(&ts, &fattr->pre_mtime)) { 1655 inode_set_mtime_to_ts(inode, fattr->mtime); 1656 } 1657 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE) 1658 && (fattr->valid & NFS_ATTR_FATTR_SIZE) 1659 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size) 1660 && !nfs_have_writebacks(inode)) { 1661 trace_nfs_size_wcc(inode, fattr->size); 1662 i_size_write(inode, nfs_size_to_loff_t(fattr->size)); 1663 } 1664 } 1665 1666 /** 1667 * nfs_check_inode_attributes - verify consistency of the inode attribute cache 1668 * @inode: pointer to inode 1669 * @fattr: updated attributes 1670 * 1671 * Verifies the attribute cache. If we have just changed the attributes, 1672 * so that fattr carries weak cache consistency data, then it may 1673 * also update the ctime/mtime/change_attribute. 1674 */ 1675 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr) 1676 { 1677 struct nfs_inode *nfsi = NFS_I(inode); 1678 loff_t cur_size, new_isize; 1679 unsigned long invalid = 0; 1680 struct timespec64 ts; 1681 1682 if (nfs_have_delegated_attributes(inode)) 1683 return 0; 1684 1685 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) { 1686 /* Only a mounted-on-fileid? Just exit */ 1687 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) 1688 return 0; 1689 /* Has the inode gone and changed behind our back? */ 1690 } else if (nfsi->fileid != fattr->fileid) { 1691 /* Is this perhaps the mounted-on fileid? */ 1692 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) && 1693 nfsi->fileid == fattr->mounted_on_fileid) 1694 return 0; 1695 return -ESTALE; 1696 } 1697 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) 1698 return -ESTALE; 1699 1700 1701 if (!nfs_file_has_buffered_writers(nfsi)) { 1702 /* Verify a few of the more important attributes */ 1703 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr)) 1704 invalid |= NFS_INO_INVALID_CHANGE; 1705 1706 ts = inode_get_mtime(inode); 1707 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime)) 1708 invalid |= NFS_INO_INVALID_MTIME; 1709 1710 ts = inode_get_ctime(inode); 1711 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime)) 1712 invalid |= NFS_INO_INVALID_CTIME; 1713 1714 if (fattr->valid & NFS_ATTR_FATTR_SIZE) { 1715 cur_size = i_size_read(inode); 1716 new_isize = nfs_size_to_loff_t(fattr->size); 1717 if (cur_size != new_isize) 1718 invalid |= NFS_INO_INVALID_SIZE; 1719 } 1720 } 1721 1722 /* Have any file permissions changed? */ 1723 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) 1724 invalid |= NFS_INO_INVALID_MODE; 1725 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid)) 1726 invalid |= NFS_INO_INVALID_OTHER; 1727 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid)) 1728 invalid |= NFS_INO_INVALID_OTHER; 1729 1730 /* Has the link count changed? */ 1731 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink) 1732 invalid |= NFS_INO_INVALID_NLINK; 1733 1734 ts = inode_get_atime(inode); 1735 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime)) 1736 invalid |= NFS_INO_INVALID_ATIME; 1737 1738 if (invalid != 0) 1739 nfs_set_cache_invalid(inode, invalid); 1740 1741 nfsi->read_cache_jiffies = fattr->time_start; 1742 return 0; 1743 } 1744 1745 static atomic_long_t nfs_attr_generation_counter; 1746 1747 static unsigned long nfs_read_attr_generation_counter(void) 1748 { 1749 return atomic_long_read(&nfs_attr_generation_counter); 1750 } 1751 1752 unsigned long nfs_inc_attr_generation_counter(void) 1753 { 1754 return atomic_long_inc_return(&nfs_attr_generation_counter); 1755 } 1756 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter); 1757 1758 void nfs_fattr_init(struct nfs_fattr *fattr) 1759 { 1760 fattr->valid = 0; 1761 fattr->time_start = jiffies; 1762 fattr->gencount = nfs_inc_attr_generation_counter(); 1763 fattr->owner_name = NULL; 1764 fattr->group_name = NULL; 1765 fattr->mdsthreshold = NULL; 1766 } 1767 EXPORT_SYMBOL_GPL(nfs_fattr_init); 1768 1769 /** 1770 * nfs_fattr_set_barrier 1771 * @fattr: attributes 1772 * 1773 * Used to set a barrier after an attribute was updated. This 1774 * barrier ensures that older attributes from RPC calls that may 1775 * have raced with our update cannot clobber these new values. 1776 * Note that you are still responsible for ensuring that other 1777 * operations which change the attribute on the server do not 1778 * collide. 1779 */ 1780 void nfs_fattr_set_barrier(struct nfs_fattr *fattr) 1781 { 1782 fattr->gencount = nfs_inc_attr_generation_counter(); 1783 } 1784 1785 struct nfs_fattr *nfs_alloc_fattr(void) 1786 { 1787 struct nfs_fattr *fattr; 1788 1789 fattr = kmalloc_obj(*fattr); 1790 if (fattr != NULL) { 1791 nfs_fattr_init(fattr); 1792 fattr->label = NULL; 1793 } 1794 return fattr; 1795 } 1796 EXPORT_SYMBOL_GPL(nfs_alloc_fattr); 1797 1798 struct nfs_fattr *nfs_alloc_fattr_with_label(struct nfs_server *server) 1799 { 1800 struct nfs_fattr *fattr = nfs_alloc_fattr(); 1801 1802 if (!fattr) 1803 return NULL; 1804 1805 fattr->label = nfs4_label_alloc(server, GFP_KERNEL); 1806 if (IS_ERR(fattr->label)) { 1807 kfree(fattr); 1808 return NULL; 1809 } 1810 1811 return fattr; 1812 } 1813 EXPORT_SYMBOL_GPL(nfs_alloc_fattr_with_label); 1814 1815 struct nfs_fh *nfs_alloc_fhandle(void) 1816 { 1817 struct nfs_fh *fh; 1818 1819 fh = kmalloc_obj(struct nfs_fh); 1820 if (fh != NULL) 1821 fh->size = 0; 1822 return fh; 1823 } 1824 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle); 1825 1826 #ifdef NFS_DEBUG 1827 /* 1828 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle 1829 * in the same way that wireshark does 1830 * 1831 * @fh: file handle 1832 * 1833 * For debugging only. 1834 */ 1835 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh) 1836 { 1837 /* wireshark uses 32-bit AUTODIN crc and does a bitwise 1838 * not on the result */ 1839 return nfs_fhandle_hash(fh); 1840 } 1841 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash); 1842 1843 /* 1844 * _nfs_display_fhandle - display an NFS file handle on the console 1845 * 1846 * @fh: file handle to display 1847 * @caption: display caption 1848 * 1849 * For debugging only. 1850 */ 1851 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption) 1852 { 1853 unsigned short i; 1854 1855 if (fh == NULL || fh->size == 0) { 1856 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh); 1857 return; 1858 } 1859 1860 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n", 1861 caption, fh, fh->size, _nfs_display_fhandle_hash(fh)); 1862 for (i = 0; i < fh->size; i += 16) { 1863 __be32 *pos = (__be32 *)&fh->data[i]; 1864 1865 switch ((fh->size - i - 1) >> 2) { 1866 case 0: 1867 printk(KERN_DEFAULT " %08x\n", 1868 be32_to_cpup(pos)); 1869 break; 1870 case 1: 1871 printk(KERN_DEFAULT " %08x %08x\n", 1872 be32_to_cpup(pos), be32_to_cpup(pos + 1)); 1873 break; 1874 case 2: 1875 printk(KERN_DEFAULT " %08x %08x %08x\n", 1876 be32_to_cpup(pos), be32_to_cpup(pos + 1), 1877 be32_to_cpup(pos + 2)); 1878 break; 1879 default: 1880 printk(KERN_DEFAULT " %08x %08x %08x %08x\n", 1881 be32_to_cpup(pos), be32_to_cpup(pos + 1), 1882 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3)); 1883 } 1884 } 1885 } 1886 EXPORT_SYMBOL_GPL(_nfs_display_fhandle); 1887 #endif 1888 1889 /** 1890 * nfs_inode_attrs_cmp_generic - compare attributes 1891 * @fattr: attributes 1892 * @inode: pointer to inode 1893 * 1894 * Attempt to divine whether or not an RPC call reply carrying stale 1895 * attributes got scheduled after another call carrying updated ones. 1896 * Note also the check for wraparound of 'attr_gencount' 1897 * 1898 * The function returns '1' if it thinks the attributes in @fattr are 1899 * more recent than the ones cached in @inode. Otherwise it returns 1900 * the value '0'. 1901 */ 1902 static int nfs_inode_attrs_cmp_generic(const struct nfs_fattr *fattr, 1903 const struct inode *inode) 1904 { 1905 unsigned long attr_gencount = NFS_I(inode)->attr_gencount; 1906 1907 return (long)(fattr->gencount - attr_gencount) > 0 || 1908 (long)(attr_gencount - nfs_read_attr_generation_counter()) > 0; 1909 } 1910 1911 /** 1912 * nfs_inode_attrs_cmp_monotonic - compare attributes 1913 * @fattr: attributes 1914 * @inode: pointer to inode 1915 * 1916 * Attempt to divine whether or not an RPC call reply carrying stale 1917 * attributes got scheduled after another call carrying updated ones. 1918 * 1919 * We assume that the server observes monotonic semantics for 1920 * the change attribute, so a larger value means that the attributes in 1921 * @fattr are more recent, in which case the function returns the 1922 * value '1'. 1923 * A return value of '0' indicates no measurable change 1924 * A return value of '-1' means that the attributes in @inode are 1925 * more recent. 1926 */ 1927 static int nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr *fattr, 1928 const struct inode *inode) 1929 { 1930 s64 diff = fattr->change_attr - inode_peek_iversion_raw(inode); 1931 if (diff > 0) 1932 return 1; 1933 return diff == 0 ? 0 : -1; 1934 } 1935 1936 /** 1937 * nfs_inode_attrs_cmp_strict_monotonic - compare attributes 1938 * @fattr: attributes 1939 * @inode: pointer to inode 1940 * 1941 * Attempt to divine whether or not an RPC call reply carrying stale 1942 * attributes got scheduled after another call carrying updated ones. 1943 * 1944 * We assume that the server observes strictly monotonic semantics for 1945 * the change attribute, so a larger value means that the attributes in 1946 * @fattr are more recent, in which case the function returns the 1947 * value '1'. 1948 * A return value of '-1' means that the attributes in @inode are 1949 * more recent or unchanged. 1950 */ 1951 static int nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr *fattr, 1952 const struct inode *inode) 1953 { 1954 return nfs_inode_attrs_cmp_monotonic(fattr, inode) > 0 ? 1 : -1; 1955 } 1956 1957 /** 1958 * nfs_inode_attrs_cmp - compare attributes 1959 * @fattr: attributes 1960 * @inode: pointer to inode 1961 * 1962 * This function returns '1' if it thinks the attributes in @fattr are 1963 * more recent than the ones cached in @inode. It returns '-1' if 1964 * the attributes in @inode are more recent than the ones in @fattr, 1965 * and it returns 0 if not sure. 1966 */ 1967 static int nfs_inode_attrs_cmp(const struct nfs_fattr *fattr, 1968 const struct inode *inode) 1969 { 1970 if (nfs_inode_attrs_cmp_generic(fattr, inode) > 0) 1971 return 1; 1972 switch (NFS_SERVER(inode)->change_attr_type) { 1973 case NFS4_CHANGE_TYPE_IS_UNDEFINED: 1974 break; 1975 case NFS4_CHANGE_TYPE_IS_TIME_METADATA: 1976 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE)) 1977 break; 1978 return nfs_inode_attrs_cmp_monotonic(fattr, inode); 1979 default: 1980 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE)) 1981 break; 1982 return nfs_inode_attrs_cmp_strict_monotonic(fattr, inode); 1983 } 1984 return 0; 1985 } 1986 1987 /** 1988 * nfs_inode_finish_partial_attr_update - complete a previous inode update 1989 * @fattr: attributes 1990 * @inode: pointer to inode 1991 * 1992 * Returns '1' if the last attribute update left the inode cached 1993 * attributes in a partially unrevalidated state, and @fattr 1994 * matches the change attribute of that partial update. 1995 * Otherwise returns '0'. 1996 */ 1997 static int nfs_inode_finish_partial_attr_update(const struct nfs_fattr *fattr, 1998 const struct inode *inode) 1999 { 2000 const unsigned long check_valid = 2001 NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME | 2002 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE | 2003 NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_OTHER | 2004 NFS_INO_INVALID_NLINK | NFS_INO_INVALID_BTIME; 2005 unsigned long cache_validity = NFS_I(inode)->cache_validity; 2006 enum nfs4_change_attr_type ctype = NFS_SERVER(inode)->change_attr_type; 2007 2008 if (ctype != NFS4_CHANGE_TYPE_IS_UNDEFINED && 2009 !(cache_validity & NFS_INO_INVALID_CHANGE) && 2010 (cache_validity & check_valid) != 0 && 2011 (fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && 2012 nfs_inode_attrs_cmp_monotonic(fattr, inode) == 0) 2013 return 1; 2014 return 0; 2015 } 2016 2017 static void nfs_ooo_merge(struct nfs_inode *nfsi, 2018 u64 start, u64 end) 2019 { 2020 int i, cnt; 2021 2022 if (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER) 2023 /* No point merging anything */ 2024 return; 2025 2026 if (!nfsi->ooo) { 2027 nfsi->ooo = kmalloc_obj(*nfsi->ooo, GFP_ATOMIC); 2028 if (!nfsi->ooo) { 2029 nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER; 2030 return; 2031 } 2032 nfsi->ooo->cnt = 0; 2033 } 2034 2035 /* add this range, merging if possible */ 2036 cnt = nfsi->ooo->cnt; 2037 for (i = 0; i < cnt; i++) { 2038 if (end == nfsi->ooo->gap[i].start) 2039 end = nfsi->ooo->gap[i].end; 2040 else if (start == nfsi->ooo->gap[i].end) 2041 start = nfsi->ooo->gap[i].start; 2042 else 2043 continue; 2044 /* Remove 'i' from table and loop to insert the new range */ 2045 cnt -= 1; 2046 nfsi->ooo->gap[i] = nfsi->ooo->gap[cnt]; 2047 i = -1; 2048 } 2049 if (start != end) { 2050 if (cnt >= ARRAY_SIZE(nfsi->ooo->gap)) { 2051 nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER; 2052 kfree(nfsi->ooo); 2053 nfsi->ooo = NULL; 2054 return; 2055 } 2056 nfsi->ooo->gap[cnt].start = start; 2057 nfsi->ooo->gap[cnt].end = end; 2058 cnt += 1; 2059 } 2060 nfsi->ooo->cnt = cnt; 2061 } 2062 2063 static void nfs_ooo_record(struct nfs_inode *nfsi, 2064 struct nfs_fattr *fattr) 2065 { 2066 /* This reply was out-of-order, so record in the 2067 * pre/post change id, possibly cancelling 2068 * gaps created when iversion was jumpped forward. 2069 */ 2070 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) && 2071 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE)) 2072 nfs_ooo_merge(nfsi, 2073 fattr->change_attr, 2074 fattr->pre_change_attr); 2075 } 2076 2077 static int nfs_refresh_inode_locked(struct inode *inode, 2078 struct nfs_fattr *fattr) 2079 { 2080 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode); 2081 int ret = 0; 2082 2083 trace_nfs_refresh_inode_enter(inode); 2084 2085 if (attr_cmp > 0 || nfs_inode_finish_partial_attr_update(fattr, inode)) 2086 ret = nfs_update_inode(inode, fattr); 2087 else { 2088 nfs_ooo_record(NFS_I(inode), fattr); 2089 2090 if (attr_cmp == 0) 2091 ret = nfs_check_inode_attributes(inode, fattr); 2092 } 2093 2094 trace_nfs_refresh_inode_exit(inode, ret); 2095 return ret; 2096 } 2097 2098 /** 2099 * nfs_refresh_inode - try to update the inode attribute cache 2100 * @inode: pointer to inode 2101 * @fattr: updated attributes 2102 * 2103 * Check that an RPC call that returned attributes has not overlapped with 2104 * other recent updates of the inode metadata, then decide whether it is 2105 * safe to do a full update of the inode attributes, or whether just to 2106 * call nfs_check_inode_attributes. 2107 */ 2108 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr) 2109 { 2110 int status; 2111 2112 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 2113 return 0; 2114 spin_lock(&inode->i_lock); 2115 status = nfs_refresh_inode_locked(inode, fattr); 2116 spin_unlock(&inode->i_lock); 2117 2118 return status; 2119 } 2120 EXPORT_SYMBOL_GPL(nfs_refresh_inode); 2121 2122 static int nfs_post_op_update_inode_locked(struct inode *inode, 2123 struct nfs_fattr *fattr, unsigned int invalid) 2124 { 2125 if (S_ISDIR(inode->i_mode)) 2126 invalid |= NFS_INO_INVALID_DATA; 2127 nfs_set_cache_invalid(inode, invalid); 2128 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 2129 return 0; 2130 return nfs_refresh_inode_locked(inode, fattr); 2131 } 2132 2133 /** 2134 * nfs_post_op_update_inode - try to update the inode attribute cache 2135 * @inode: pointer to inode 2136 * @fattr: updated attributes 2137 * 2138 * After an operation that has changed the inode metadata, mark the 2139 * attribute cache as being invalid, then try to update it. 2140 * 2141 * NB: if the server didn't return any post op attributes, this 2142 * function will force the retrieval of attributes before the next 2143 * NFS request. Thus it should be used only for operations that 2144 * are expected to change one or more attributes, to avoid 2145 * unnecessary NFS requests and trips through nfs_update_inode(). 2146 */ 2147 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr) 2148 { 2149 int status; 2150 2151 spin_lock(&inode->i_lock); 2152 nfs_fattr_set_barrier(fattr); 2153 status = nfs_post_op_update_inode_locked(inode, fattr, 2154 NFS_INO_INVALID_CHANGE 2155 | NFS_INO_INVALID_CTIME 2156 | NFS_INO_REVAL_FORCED); 2157 spin_unlock(&inode->i_lock); 2158 2159 return status; 2160 } 2161 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode); 2162 2163 /** 2164 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache 2165 * @inode: pointer to inode 2166 * @fattr: updated attributes 2167 * 2168 * After an operation that has changed the inode metadata, mark the 2169 * attribute cache as being invalid, then try to update it. Fake up 2170 * weak cache consistency data, if none exist. 2171 * 2172 * This function is mainly designed to be used by the ->write_done() functions. 2173 */ 2174 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr) 2175 { 2176 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode); 2177 int status; 2178 2179 /* Don't do a WCC update if these attributes are already stale */ 2180 if (attr_cmp < 0) 2181 return 0; 2182 if ((fattr->valid & NFS_ATTR_FATTR) == 0 || !attr_cmp) { 2183 /* Record the pre/post change info before clearing PRECHANGE */ 2184 nfs_ooo_record(NFS_I(inode), fattr); 2185 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE 2186 | NFS_ATTR_FATTR_PRESIZE 2187 | NFS_ATTR_FATTR_PREMTIME 2188 | NFS_ATTR_FATTR_PRECTIME); 2189 goto out_noforce; 2190 } 2191 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && 2192 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) { 2193 fattr->pre_change_attr = inode_peek_iversion_raw(inode); 2194 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE; 2195 } 2196 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 && 2197 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) { 2198 fattr->pre_ctime = inode_get_ctime(inode); 2199 fattr->valid |= NFS_ATTR_FATTR_PRECTIME; 2200 } 2201 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 && 2202 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) { 2203 fattr->pre_mtime = inode_get_mtime(inode); 2204 fattr->valid |= NFS_ATTR_FATTR_PREMTIME; 2205 } 2206 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 && 2207 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) { 2208 fattr->pre_size = i_size_read(inode); 2209 fattr->valid |= NFS_ATTR_FATTR_PRESIZE; 2210 } 2211 out_noforce: 2212 status = nfs_post_op_update_inode_locked(inode, fattr, 2213 NFS_INO_INVALID_CHANGE 2214 | NFS_INO_INVALID_CTIME 2215 | NFS_INO_INVALID_MTIME 2216 | NFS_INO_INVALID_BLOCKS); 2217 return status; 2218 } 2219 2220 /** 2221 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache 2222 * @inode: pointer to inode 2223 * @fattr: updated attributes 2224 * 2225 * After an operation that has changed the inode metadata, mark the 2226 * attribute cache as being invalid, then try to update it. Fake up 2227 * weak cache consistency data, if none exist. 2228 * 2229 * This function is mainly designed to be used by the ->write_done() functions. 2230 */ 2231 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr) 2232 { 2233 int status; 2234 2235 spin_lock(&inode->i_lock); 2236 nfs_fattr_set_barrier(fattr); 2237 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 2238 spin_unlock(&inode->i_lock); 2239 return status; 2240 } 2241 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc); 2242 2243 2244 /* 2245 * Many nfs protocol calls return the new file attributes after 2246 * an operation. Here we update the inode to reflect the state 2247 * of the server's inode. 2248 * 2249 * This is a bit tricky because we have to make sure all dirty pages 2250 * have been sent off to the server before calling invalidate_inode_pages. 2251 * To make sure no other process adds more write requests while we try 2252 * our best to flush them, we make them sleep during the attribute refresh. 2253 * 2254 * A very similar scenario holds for the dir cache. 2255 */ 2256 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr) 2257 { 2258 struct nfs_server *server = NFS_SERVER(inode); 2259 struct nfs_inode *nfsi = NFS_I(inode); 2260 loff_t cur_isize, new_isize; 2261 u64 fattr_supported = server->fattr_valid; 2262 unsigned long invalid = 0; 2263 unsigned long now = jiffies; 2264 unsigned long save_cache_validity; 2265 bool have_writers = nfs_file_has_buffered_writers(nfsi); 2266 bool cache_revalidated = true; 2267 bool attr_changed = false; 2268 bool have_delegation; 2269 2270 dfprintk(VFS, "NFS: %s(%s/%llu fh_crc=0x%08x ct=%d info=0x%llx)\n", 2271 __func__, inode->i_sb->s_id, inode->i_ino, 2272 nfs_display_fhandle_hash(NFS_FH(inode)), 2273 icount_read_once(inode), fattr->valid); 2274 2275 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) { 2276 /* Only a mounted-on-fileid? Just exit */ 2277 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) 2278 return 0; 2279 /* Has the inode gone and changed behind our back? */ 2280 } else if (nfsi->fileid != fattr->fileid) { 2281 /* Is this perhaps the mounted-on fileid? */ 2282 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) && 2283 nfsi->fileid == fattr->mounted_on_fileid) 2284 return 0; 2285 printk(KERN_ERR "NFS: server %s error: fileid changed\n" 2286 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n", 2287 NFS_SERVER(inode)->nfs_client->cl_hostname, 2288 inode->i_sb->s_id, (long long)nfsi->fileid, 2289 (long long)fattr->fileid); 2290 goto out_err; 2291 } 2292 2293 /* 2294 * Make sure the inode's type hasn't changed. 2295 */ 2296 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) { 2297 /* 2298 * Big trouble! The inode has become a different object. 2299 */ 2300 printk(KERN_DEBUG "NFS: %s: inode %llu mode changed, %07o to %07o\n", 2301 __func__, inode->i_ino, inode->i_mode, fattr->mode); 2302 goto out_err; 2303 } 2304 2305 /* Update the fsid? */ 2306 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) && 2307 !nfs_fsid_equal(&server->fsid, &fattr->fsid) && 2308 !IS_AUTOMOUNT(inode)) 2309 server->fsid = fattr->fsid; 2310 2311 /* Save the delegation state before clearing cache_validity */ 2312 have_delegation = nfs_have_delegated_attributes(inode); 2313 2314 /* 2315 * Update the read time so we don't revalidate too often. 2316 */ 2317 nfsi->read_cache_jiffies = fattr->time_start; 2318 2319 /* Fix up any delegated attributes in the struct nfs_fattr */ 2320 nfs_fattr_fixup_delegated(inode, fattr); 2321 2322 save_cache_validity = nfsi->cache_validity; 2323 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR 2324 | NFS_INO_INVALID_ATIME 2325 | NFS_INO_REVAL_FORCED 2326 | NFS_INO_INVALID_BLOCKS); 2327 2328 /* Do atomic weak cache consistency updates */ 2329 nfs_wcc_update_inode(inode, fattr); 2330 2331 if (pnfs_layoutcommit_outstanding(inode)) { 2332 nfsi->cache_validity |= 2333 save_cache_validity & 2334 (NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME | 2335 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE | 2336 NFS_INO_INVALID_BLOCKS); 2337 cache_revalidated = false; 2338 } 2339 2340 /* More cache consistency checks */ 2341 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) { 2342 if (!have_writers && nfsi->ooo && nfsi->ooo->cnt == 1 && 2343 nfsi->ooo->gap[0].end == inode_peek_iversion_raw(inode)) { 2344 /* There is one remaining gap that hasn't been 2345 * merged into iversion - do that now. 2346 */ 2347 inode_set_iversion_raw(inode, nfsi->ooo->gap[0].start); 2348 kfree(nfsi->ooo); 2349 nfsi->ooo = NULL; 2350 } 2351 if (!inode_eq_iversion_raw(inode, fattr->change_attr)) { 2352 /* Could it be a race with writeback? */ 2353 if (!(have_writers || have_delegation)) { 2354 invalid |= NFS_INO_INVALID_DATA 2355 | NFS_INO_INVALID_ACCESS 2356 | NFS_INO_INVALID_ACL 2357 | NFS_INO_INVALID_XATTR; 2358 /* Force revalidate of all attributes */ 2359 save_cache_validity |= NFS_INO_INVALID_CTIME 2360 | NFS_INO_INVALID_MTIME 2361 | NFS_INO_INVALID_SIZE 2362 | NFS_INO_INVALID_BLOCKS 2363 | NFS_INO_INVALID_NLINK 2364 | NFS_INO_INVALID_MODE 2365 | NFS_INO_INVALID_OTHER 2366 | NFS_INO_INVALID_BTIME; 2367 if (S_ISDIR(inode->i_mode)) 2368 nfs_force_lookup_revalidate(inode); 2369 attr_changed = true; 2370 dprintk("NFS: change_attr change on server for file %s/%llu\n", 2371 inode->i_sb->s_id, 2372 inode->i_ino); 2373 } else if (!have_delegation) { 2374 nfs_ooo_record(nfsi, fattr); 2375 nfs_ooo_merge(nfsi, inode_peek_iversion_raw(inode), 2376 fattr->change_attr); 2377 } 2378 inode_set_iversion_raw(inode, fattr->change_attr); 2379 } 2380 } else { 2381 nfsi->cache_validity |= 2382 save_cache_validity & NFS_INO_INVALID_CHANGE; 2383 if (!have_delegation || 2384 (nfsi->cache_validity & NFS_INO_INVALID_CHANGE) != 0) 2385 cache_revalidated = false; 2386 } 2387 2388 if (fattr->valid & NFS_ATTR_FATTR_MTIME) 2389 inode_set_mtime_to_ts(inode, fattr->mtime); 2390 else if (fattr_supported & NFS_ATTR_FATTR_MTIME) 2391 nfsi->cache_validity |= 2392 save_cache_validity & NFS_INO_INVALID_MTIME; 2393 2394 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 2395 inode_set_ctime_to_ts(inode, fattr->ctime); 2396 else if (fattr_supported & NFS_ATTR_FATTR_CTIME) 2397 nfsi->cache_validity |= 2398 save_cache_validity & NFS_INO_INVALID_CTIME; 2399 2400 if (fattr->valid & NFS_ATTR_FATTR_BTIME) 2401 nfsi->btime = fattr->btime; 2402 else if (fattr_supported & NFS_ATTR_FATTR_BTIME) 2403 nfsi->cache_validity |= 2404 save_cache_validity & NFS_INO_INVALID_BTIME; 2405 2406 /* Check if our cached file size is stale */ 2407 if (fattr->valid & NFS_ATTR_FATTR_SIZE) { 2408 new_isize = nfs_size_to_loff_t(fattr->size); 2409 cur_isize = i_size_read(inode); 2410 if (new_isize != cur_isize && !have_delegation) { 2411 /* Do we perhaps have any outstanding writes, or has 2412 * the file grown beyond our last write? */ 2413 if (!nfs_have_writebacks(inode) || new_isize > cur_isize) { 2414 trace_nfs_size_update(inode, new_isize); 2415 i_size_write(inode, new_isize); 2416 if (!have_writers) 2417 invalid |= NFS_INO_INVALID_DATA; 2418 } 2419 } 2420 if (new_isize == 0 && 2421 !(fattr->valid & (NFS_ATTR_FATTR_SPACE_USED | 2422 NFS_ATTR_FATTR_BLOCKS_USED))) { 2423 fattr->du.nfs3.used = 0; 2424 fattr->valid |= NFS_ATTR_FATTR_SPACE_USED; 2425 } 2426 } else 2427 nfsi->cache_validity |= 2428 save_cache_validity & NFS_INO_INVALID_SIZE; 2429 2430 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 2431 inode_set_atime_to_ts(inode, fattr->atime); 2432 else if (fattr_supported & NFS_ATTR_FATTR_ATIME) 2433 nfsi->cache_validity |= 2434 save_cache_validity & NFS_INO_INVALID_ATIME; 2435 2436 if (fattr->valid & NFS_ATTR_FATTR_MODE) { 2437 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) { 2438 umode_t newmode = inode->i_mode & S_IFMT; 2439 newmode |= fattr->mode & S_IALLUGO; 2440 inode->i_mode = newmode; 2441 invalid |= NFS_INO_INVALID_ACCESS 2442 | NFS_INO_INVALID_ACL; 2443 } 2444 } else if (fattr_supported & NFS_ATTR_FATTR_MODE) 2445 nfsi->cache_validity |= 2446 save_cache_validity & NFS_INO_INVALID_MODE; 2447 2448 if (fattr->valid & NFS_ATTR_FATTR_OWNER) { 2449 if (!uid_eq(inode->i_uid, fattr->uid)) { 2450 invalid |= NFS_INO_INVALID_ACCESS 2451 | NFS_INO_INVALID_ACL; 2452 inode->i_uid = fattr->uid; 2453 } 2454 } else if (fattr_supported & NFS_ATTR_FATTR_OWNER) 2455 nfsi->cache_validity |= 2456 save_cache_validity & NFS_INO_INVALID_OTHER; 2457 2458 if (fattr->valid & NFS_ATTR_FATTR_GROUP) { 2459 if (!gid_eq(inode->i_gid, fattr->gid)) { 2460 invalid |= NFS_INO_INVALID_ACCESS 2461 | NFS_INO_INVALID_ACL; 2462 inode->i_gid = fattr->gid; 2463 } 2464 } else if (fattr_supported & NFS_ATTR_FATTR_GROUP) 2465 nfsi->cache_validity |= 2466 save_cache_validity & NFS_INO_INVALID_OTHER; 2467 2468 if (fattr->valid & NFS_ATTR_FATTR_NLINK) { 2469 if (inode->i_nlink != fattr->nlink) 2470 set_nlink(inode, fattr->nlink); 2471 } else if (fattr_supported & NFS_ATTR_FATTR_NLINK) 2472 nfsi->cache_validity |= 2473 save_cache_validity & NFS_INO_INVALID_NLINK; 2474 2475 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) { 2476 /* 2477 * report the blocks in 512byte units 2478 */ 2479 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 2480 } else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED) 2481 nfsi->cache_validity |= 2482 save_cache_validity & NFS_INO_INVALID_BLOCKS; 2483 2484 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED) 2485 inode->i_blocks = fattr->du.nfs2.blocks; 2486 else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED) 2487 nfsi->cache_validity |= 2488 save_cache_validity & NFS_INO_INVALID_BLOCKS; 2489 2490 /* Update attrtimeo value if we're out of the unstable period */ 2491 if (attr_changed) { 2492 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 2493 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 2494 nfsi->attrtimeo_timestamp = now; 2495 /* Set barrier to be more recent than all outstanding updates */ 2496 nfsi->attr_gencount = nfs_inc_attr_generation_counter(); 2497 } else { 2498 if (cache_revalidated) { 2499 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, 2500 nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) { 2501 nfsi->attrtimeo <<= 1; 2502 if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode)) 2503 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode); 2504 } 2505 nfsi->attrtimeo_timestamp = now; 2506 } 2507 /* Set the barrier to be more recent than this fattr */ 2508 if ((long)(fattr->gencount - nfsi->attr_gencount) > 0) 2509 nfsi->attr_gencount = fattr->gencount; 2510 } 2511 2512 /* Don't invalidate the data if we were to blame */ 2513 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) 2514 || S_ISLNK(inode->i_mode))) 2515 invalid &= ~NFS_INO_INVALID_DATA; 2516 nfs_set_cache_invalid(inode, invalid); 2517 2518 return 0; 2519 out_err: 2520 /* 2521 * No need to worry about unhashing the dentry, as the 2522 * lookup validation will know that the inode is bad. 2523 * (But we fall through to invalidate the caches.) 2524 */ 2525 nfs_set_inode_stale_locked(inode); 2526 return -ESTALE; 2527 } 2528 2529 struct inode *nfs_alloc_inode(struct super_block *sb) 2530 { 2531 struct nfs_inode *nfsi; 2532 nfsi = alloc_inode_sb(sb, nfs_inode_cachep, GFP_KERNEL); 2533 if (!nfsi) 2534 return NULL; 2535 nfsi->flags = 0UL; 2536 nfsi->cache_validity = 0UL; 2537 nfsi->ooo = NULL; 2538 #if IS_ENABLED(CONFIG_NFS_V4) 2539 nfsi->nfs4_acl = NULL; 2540 #endif /* CONFIG_NFS_V4 */ 2541 #ifdef CONFIG_NFS_V4_2 2542 nfsi->xattr_cache = NULL; 2543 #endif 2544 nfs_netfs_inode_init(nfsi); 2545 2546 return &nfsi->vfs_inode; 2547 } 2548 EXPORT_SYMBOL_GPL(nfs_alloc_inode); 2549 2550 void nfs_free_inode(struct inode *inode) 2551 { 2552 kfree(NFS_I(inode)->ooo); 2553 kmem_cache_free(nfs_inode_cachep, NFS_I(inode)); 2554 } 2555 EXPORT_SYMBOL_GPL(nfs_free_inode); 2556 2557 static inline void nfs4_init_once(struct nfs_inode *nfsi) 2558 { 2559 #if IS_ENABLED(CONFIG_NFS_V4) 2560 INIT_LIST_HEAD(&nfsi->open_states); 2561 nfsi->delegation = NULL; 2562 init_rwsem(&nfsi->rwsem); 2563 nfsi->layout = NULL; 2564 #endif 2565 } 2566 2567 static void init_once(void *foo) 2568 { 2569 struct nfs_inode *nfsi = foo; 2570 2571 inode_init_once(&nfsi->vfs_inode); 2572 INIT_LIST_HEAD(&nfsi->open_files); 2573 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru); 2574 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru); 2575 nfs4_init_once(nfsi); 2576 } 2577 2578 static int __init nfs_init_inodecache(void) 2579 { 2580 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache", 2581 sizeof(struct nfs_inode), 2582 0, (SLAB_RECLAIM_ACCOUNT| 2583 SLAB_ACCOUNT), 2584 init_once); 2585 if (nfs_inode_cachep == NULL) 2586 return -ENOMEM; 2587 2588 return 0; 2589 } 2590 2591 static void nfs_destroy_inodecache(void) 2592 { 2593 /* 2594 * Make sure all delayed rcu free inodes are flushed before we 2595 * destroy cache. 2596 */ 2597 rcu_barrier(); 2598 kmem_cache_destroy(nfs_inode_cachep); 2599 } 2600 2601 struct workqueue_struct *nfslocaliod_workqueue; 2602 struct workqueue_struct *nfsiod_workqueue; 2603 EXPORT_SYMBOL_GPL(nfsiod_workqueue); 2604 2605 /* 2606 * Destroy the nfsiod workqueues 2607 */ 2608 static void nfsiod_stop(void) 2609 { 2610 struct workqueue_struct *wq; 2611 2612 wq = nfsiod_workqueue; 2613 if (wq != NULL) { 2614 nfsiod_workqueue = NULL; 2615 destroy_workqueue(wq); 2616 } 2617 #if IS_ENABLED(CONFIG_NFS_LOCALIO) 2618 wq = nfslocaliod_workqueue; 2619 if (wq != NULL) { 2620 nfslocaliod_workqueue = NULL; 2621 destroy_workqueue(wq); 2622 } 2623 #endif /* CONFIG_NFS_LOCALIO */ 2624 } 2625 2626 /* 2627 * Start the nfsiod workqueues 2628 */ 2629 static int nfsiod_start(void) 2630 { 2631 dprintk("RPC: creating workqueue nfsiod\n"); 2632 nfsiod_workqueue = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0); 2633 if (nfsiod_workqueue == NULL) 2634 return -ENOMEM; 2635 #if IS_ENABLED(CONFIG_NFS_LOCALIO) 2636 /* 2637 * localio writes need to use a normal (non-memreclaim) workqueue. 2638 * When we start getting low on space, XFS goes and calls flush_work() on 2639 * a non-memreclaim work queue, which causes a priority inversion problem. 2640 */ 2641 dprintk("RPC: creating workqueue nfslocaliod\n"); 2642 nfslocaliod_workqueue = alloc_workqueue("nfslocaliod", WQ_UNBOUND, 0); 2643 if (unlikely(nfslocaliod_workqueue == NULL)) { 2644 nfsiod_stop(); 2645 return -ENOMEM; 2646 } 2647 #endif /* CONFIG_NFS_LOCALIO */ 2648 return 0; 2649 } 2650 2651 unsigned int nfs_net_id; 2652 EXPORT_SYMBOL_GPL(nfs_net_id); 2653 2654 static int nfs_net_init(struct net *net) 2655 { 2656 struct nfs_net *nn = net_generic(net, nfs_net_id); 2657 int err; 2658 2659 nfs_clients_init(net); 2660 2661 if (!rpc_proc_register(net, &nn->rpcstats)) { 2662 err = -ENOMEM; 2663 goto err_proc_rpc; 2664 } 2665 2666 err = nfs_fs_proc_net_init(net); 2667 if (err) 2668 goto err_proc_nfs; 2669 2670 return 0; 2671 2672 err_proc_nfs: 2673 rpc_proc_unregister(net, "nfs"); 2674 err_proc_rpc: 2675 nfs_clients_exit(net); 2676 return err; 2677 } 2678 2679 static void nfs_net_exit(struct net *net) 2680 { 2681 rpc_proc_unregister(net, "nfs"); 2682 nfs_fs_proc_net_exit(net); 2683 nfs_clients_exit(net); 2684 } 2685 2686 static struct pernet_operations nfs_net_ops = { 2687 .init = nfs_net_init, 2688 .exit = nfs_net_exit, 2689 .id = &nfs_net_id, 2690 .size = sizeof(struct nfs_net), 2691 }; 2692 2693 #ifdef CONFIG_KEYS 2694 static struct key *nfs_keyring; 2695 2696 static int __init nfs_init_keyring(void) 2697 { 2698 nfs_keyring = keyring_alloc(".nfs", 2699 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, 2700 current_cred(), 2701 (KEY_POS_ALL & ~KEY_POS_SETATTR) | 2702 (KEY_USR_ALL & ~KEY_USR_SETATTR), 2703 KEY_ALLOC_NOT_IN_QUOTA, NULL, NULL); 2704 return PTR_ERR_OR_ZERO(nfs_keyring); 2705 } 2706 2707 static void nfs_exit_keyring(void) 2708 { 2709 key_put(nfs_keyring); 2710 } 2711 #else 2712 static inline int nfs_init_keyring(void) 2713 { 2714 return 0; 2715 } 2716 2717 static inline void nfs_exit_keyring(void) 2718 { 2719 } 2720 #endif /* CONFIG_KEYS */ 2721 2722 /* 2723 * Initialize NFS 2724 */ 2725 static int __init init_nfs_fs(void) 2726 { 2727 int err; 2728 2729 err = nfs_init_keyring(); 2730 if (err) 2731 return err; 2732 2733 err = nfs_sysfs_init(); 2734 if (err < 0) 2735 goto out10; 2736 2737 err = register_pernet_subsys(&nfs_net_ops); 2738 if (err < 0) 2739 goto out9; 2740 2741 err = nfsiod_start(); 2742 if (err) 2743 goto out7; 2744 2745 err = nfs_fs_proc_init(); 2746 if (err) 2747 goto out6; 2748 2749 err = nfs_init_nfspagecache(); 2750 if (err) 2751 goto out5; 2752 2753 err = nfs_init_inodecache(); 2754 if (err) 2755 goto out4; 2756 2757 err = nfs_init_readpagecache(); 2758 if (err) 2759 goto out3; 2760 2761 err = nfs_init_writepagecache(); 2762 if (err) 2763 goto out2; 2764 2765 err = nfs_init_directcache(); 2766 if (err) 2767 goto out1; 2768 2769 err = register_nfs_fs(); 2770 if (err) 2771 goto out0; 2772 2773 return 0; 2774 out0: 2775 nfs_destroy_directcache(); 2776 out1: 2777 nfs_destroy_writepagecache(); 2778 out2: 2779 nfs_destroy_readpagecache(); 2780 out3: 2781 nfs_destroy_inodecache(); 2782 out4: 2783 nfs_destroy_nfspagecache(); 2784 out5: 2785 nfs_fs_proc_exit(); 2786 out6: 2787 nfsiod_stop(); 2788 out7: 2789 unregister_pernet_subsys(&nfs_net_ops); 2790 out9: 2791 nfs_sysfs_exit(); 2792 out10: 2793 nfs_exit_keyring(); 2794 return err; 2795 } 2796 2797 static void __exit exit_nfs_fs(void) 2798 { 2799 nfs_destroy_directcache(); 2800 nfs_destroy_writepagecache(); 2801 nfs_destroy_readpagecache(); 2802 nfs_destroy_inodecache(); 2803 nfs_destroy_nfspagecache(); 2804 unregister_pernet_subsys(&nfs_net_ops); 2805 unregister_nfs_fs(); 2806 nfs_fs_proc_exit(); 2807 nfsiod_stop(); 2808 nfs_sysfs_exit(); 2809 nfs_exit_keyring(); 2810 } 2811 2812 /* Not quite true; I just maintain it */ 2813 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>"); 2814 MODULE_DESCRIPTION("NFS client support"); 2815 MODULE_LICENSE("GPL"); 2816 module_param(enable_ino64, bool, 0644); 2817 2818 module_init(init_nfs_fs) 2819 module_exit(exit_nfs_fs) 2820