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) || generic_drop_inode(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->i_state & 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 atomic_read(&inode->i_count)); 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 int updated = 0; 653 654 now = inode_set_ctime_current(inode); 655 if (!timespec64_equal(&now, &ctime)) 656 updated |= S_CTIME; 657 658 inode_set_mtime_to_ts(inode, attr->ia_mtime); 659 if (!timespec64_equal(&now, &mtime)) 660 updated |= S_MTIME; 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_timestamps(struct inode *inode, unsigned int ia_valid) 673 { 674 enum file_time_flags time_flags = 0; 675 unsigned int cache_flags = 0; 676 677 if (ia_valid & ATTR_MTIME) { 678 time_flags |= S_MTIME | S_CTIME; 679 cache_flags |= NFS_INO_INVALID_CTIME | NFS_INO_INVALID_MTIME; 680 } 681 if (ia_valid & ATTR_ATIME) { 682 time_flags |= S_ATIME; 683 cache_flags |= NFS_INO_INVALID_ATIME; 684 } 685 inode_update_timestamps(inode, time_flags); 686 NFS_I(inode)->cache_validity &= ~cache_flags; 687 } 688 689 void nfs_update_delegated_atime(struct inode *inode) 690 { 691 spin_lock(&inode->i_lock); 692 if (nfs_have_delegated_atime(inode)) 693 nfs_update_timestamps(inode, ATTR_ATIME); 694 spin_unlock(&inode->i_lock); 695 } 696 697 void nfs_update_delegated_mtime_locked(struct inode *inode) 698 { 699 if (nfs_have_delegated_mtime(inode)) 700 nfs_update_timestamps(inode, ATTR_MTIME); 701 } 702 703 void nfs_update_delegated_mtime(struct inode *inode) 704 { 705 spin_lock(&inode->i_lock); 706 nfs_update_delegated_mtime_locked(inode); 707 spin_unlock(&inode->i_lock); 708 } 709 EXPORT_SYMBOL_GPL(nfs_update_delegated_mtime); 710 711 #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) 712 713 int 714 nfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry, 715 struct iattr *attr) 716 { 717 struct inode *inode = d_inode(dentry); 718 struct nfs_fattr *fattr; 719 loff_t oldsize = i_size_read(inode); 720 int error = 0; 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 (attr->ia_valid & ATTR_SIZE) { 729 BUG_ON(!S_ISREG(inode->i_mode)); 730 731 error = inode_newsize_ok(inode, attr->ia_size); 732 if (error) 733 return error; 734 735 if (attr->ia_size == oldsize) 736 attr->ia_valid &= ~ATTR_SIZE; 737 } 738 739 if (nfs_have_delegated_mtime(inode) && attr->ia_valid & ATTR_MTIME) { 740 spin_lock(&inode->i_lock); 741 if (attr->ia_valid & ATTR_MTIME_SET) { 742 nfs_set_timestamps_to_ts(inode, attr); 743 attr->ia_valid &= ~(ATTR_MTIME|ATTR_MTIME_SET| 744 ATTR_ATIME|ATTR_ATIME_SET); 745 } else { 746 nfs_update_timestamps(inode, attr->ia_valid); 747 attr->ia_valid &= ~(ATTR_MTIME|ATTR_ATIME); 748 } 749 spin_unlock(&inode->i_lock); 750 } else if (nfs_have_delegated_atime(inode) && 751 attr->ia_valid & ATTR_ATIME && 752 !(attr->ia_valid & ATTR_MTIME)) { 753 if (attr->ia_valid & ATTR_ATIME_SET) { 754 spin_lock(&inode->i_lock); 755 nfs_set_timestamps_to_ts(inode, attr); 756 spin_unlock(&inode->i_lock); 757 attr->ia_valid &= ~(ATTR_ATIME|ATTR_ATIME_SET); 758 } else { 759 nfs_update_delegated_atime(inode); 760 attr->ia_valid &= ~ATTR_ATIME; 761 } 762 } 763 764 /* Optimization: if the end result is no change, don't RPC */ 765 if (((attr->ia_valid & NFS_VALID_ATTRS) & ~(ATTR_FILE|ATTR_OPEN)) == 0) 766 return 0; 767 768 trace_nfs_setattr_enter(inode); 769 770 /* Write all dirty data */ 771 if (S_ISREG(inode->i_mode)) { 772 nfs_file_block_o_direct(NFS_I(inode)); 773 nfs_sync_inode(inode); 774 } 775 776 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode)); 777 if (fattr == NULL) { 778 error = -ENOMEM; 779 goto out; 780 } 781 782 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr); 783 if (error == 0) { 784 if (attr->ia_valid & ATTR_SIZE) 785 nfs_truncate_last_folio(inode->i_mapping, oldsize, 786 attr->ia_size); 787 error = nfs_refresh_inode(inode, fattr); 788 } 789 nfs_free_fattr(fattr); 790 out: 791 trace_nfs_setattr_exit(inode, error); 792 return error; 793 } 794 EXPORT_SYMBOL_GPL(nfs_setattr); 795 796 /** 797 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall 798 * @inode: inode of the file used 799 * @offset: file offset to start truncating 800 * 801 * This is a copy of the common vmtruncate, but with the locking 802 * corrected to take into account the fact that NFS requires 803 * inode->i_size to be updated under the inode->i_lock. 804 * Note: must be called with inode->i_lock held! 805 */ 806 static int nfs_vmtruncate(struct inode * inode, loff_t offset) 807 { 808 int err; 809 810 err = inode_newsize_ok(inode, offset); 811 if (err) 812 goto out; 813 814 trace_nfs_size_truncate(inode, offset); 815 i_size_write(inode, offset); 816 /* Optimisation */ 817 if (offset == 0) { 818 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA; 819 nfs_ooo_clear(NFS_I(inode)); 820 } 821 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE; 822 823 spin_unlock(&inode->i_lock); 824 truncate_pagecache(inode, offset); 825 nfs_update_delegated_mtime_locked(inode); 826 spin_lock(&inode->i_lock); 827 out: 828 return err; 829 } 830 831 /** 832 * nfs_setattr_update_inode - Update inode metadata after a setattr call. 833 * @inode: pointer to struct inode 834 * @attr: pointer to struct iattr 835 * @fattr: pointer to struct nfs_fattr 836 * 837 * Note: we do this in the *proc.c in order to ensure that 838 * it works for things like exclusive creates too. 839 */ 840 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr, 841 struct nfs_fattr *fattr) 842 { 843 /* Barrier: bump the attribute generation count. */ 844 nfs_fattr_set_barrier(fattr); 845 846 spin_lock(&inode->i_lock); 847 NFS_I(inode)->attr_gencount = fattr->gencount; 848 if ((attr->ia_valid & ATTR_SIZE) != 0) { 849 if (!nfs_have_delegated_mtime(inode)) 850 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME); 851 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS); 852 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC); 853 nfs_vmtruncate(inode, attr->ia_size); 854 } 855 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) { 856 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME; 857 if ((attr->ia_valid & ATTR_KILL_SUID) != 0 && 858 inode->i_mode & S_ISUID) 859 inode->i_mode &= ~S_ISUID; 860 if (setattr_should_drop_sgid(&nop_mnt_idmap, inode)) 861 inode->i_mode &= ~S_ISGID; 862 if ((attr->ia_valid & ATTR_MODE) != 0) { 863 int mode = attr->ia_mode & S_IALLUGO; 864 mode |= inode->i_mode & ~S_IALLUGO; 865 inode->i_mode = mode; 866 } 867 if ((attr->ia_valid & ATTR_UID) != 0) 868 inode->i_uid = attr->ia_uid; 869 if ((attr->ia_valid & ATTR_GID) != 0) 870 inode->i_gid = attr->ia_gid; 871 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 872 inode_set_ctime_to_ts(inode, fattr->ctime); 873 else 874 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE 875 | NFS_INO_INVALID_CTIME); 876 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS 877 | NFS_INO_INVALID_ACL); 878 } 879 if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) { 880 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME 881 | NFS_INO_INVALID_CTIME); 882 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 883 inode_set_atime_to_ts(inode, fattr->atime); 884 else if (attr->ia_valid & ATTR_ATIME_SET) 885 inode_set_atime_to_ts(inode, attr->ia_atime); 886 else 887 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME); 888 889 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 890 inode_set_ctime_to_ts(inode, fattr->ctime); 891 else 892 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE 893 | NFS_INO_INVALID_CTIME); 894 } 895 if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) { 896 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME 897 | NFS_INO_INVALID_CTIME); 898 if (fattr->valid & NFS_ATTR_FATTR_MTIME) 899 inode_set_mtime_to_ts(inode, fattr->mtime); 900 else if (attr->ia_valid & ATTR_MTIME_SET) 901 inode_set_mtime_to_ts(inode, attr->ia_mtime); 902 else 903 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME); 904 905 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 906 inode_set_ctime_to_ts(inode, fattr->ctime); 907 else 908 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE 909 | NFS_INO_INVALID_CTIME); 910 } 911 if (fattr->valid) 912 nfs_update_inode(inode, fattr); 913 spin_unlock(&inode->i_lock); 914 } 915 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode); 916 917 /* 918 * Don't request help from readdirplus if the file is being written to, 919 * or if attribute caching is turned off 920 */ 921 static bool nfs_getattr_readdirplus_enable(const struct inode *inode) 922 { 923 return nfs_server_capable(inode, NFS_CAP_READDIRPLUS) && 924 !nfs_have_writebacks(inode) && NFS_MAXATTRTIMEO(inode) > 5 * HZ; 925 } 926 927 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry) 928 { 929 if (!IS_ROOT(dentry)) { 930 struct dentry *parent = dget_parent(dentry); 931 nfs_readdir_record_entry_cache_miss(d_inode(parent)); 932 dput(parent); 933 } 934 } 935 936 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry) 937 { 938 if (!IS_ROOT(dentry)) { 939 struct dentry *parent = dget_parent(dentry); 940 nfs_readdir_record_entry_cache_hit(d_inode(parent)); 941 dput(parent); 942 } 943 } 944 945 static u32 nfs_get_valid_attrmask(struct inode *inode) 946 { 947 u64 fattr_valid = NFS_SERVER(inode)->fattr_valid; 948 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity); 949 u32 reply_mask = STATX_INO | STATX_TYPE; 950 951 if (!(cache_validity & NFS_INO_INVALID_ATIME)) 952 reply_mask |= STATX_ATIME; 953 if (!(cache_validity & NFS_INO_INVALID_CTIME)) 954 reply_mask |= STATX_CTIME; 955 if (!(cache_validity & NFS_INO_INVALID_MTIME)) 956 reply_mask |= STATX_MTIME; 957 if (!(cache_validity & NFS_INO_INVALID_SIZE)) 958 reply_mask |= STATX_SIZE; 959 if (!(cache_validity & NFS_INO_INVALID_NLINK)) 960 reply_mask |= STATX_NLINK; 961 if (!(cache_validity & NFS_INO_INVALID_MODE)) 962 reply_mask |= STATX_MODE; 963 if (!(cache_validity & NFS_INO_INVALID_OTHER)) 964 reply_mask |= STATX_UID | STATX_GID; 965 if (!(cache_validity & NFS_INO_INVALID_BLOCKS)) 966 reply_mask |= STATX_BLOCKS; 967 if (!(cache_validity & NFS_INO_INVALID_BTIME) && 968 (fattr_valid & NFS_ATTR_FATTR_BTIME)) 969 reply_mask |= STATX_BTIME; 970 if (!(cache_validity & NFS_INO_INVALID_CHANGE)) 971 reply_mask |= STATX_CHANGE_COOKIE; 972 return reply_mask; 973 } 974 975 int nfs_getattr(struct mnt_idmap *idmap, const struct path *path, 976 struct kstat *stat, u32 request_mask, unsigned int query_flags) 977 { 978 struct inode *inode = d_inode(path->dentry); 979 struct nfs_server *server = NFS_SERVER(inode); 980 u64 fattr_valid = server->fattr_valid; 981 unsigned long cache_validity; 982 int err = 0; 983 bool force_sync = query_flags & AT_STATX_FORCE_SYNC; 984 bool do_update = false; 985 bool readdirplus_enabled = nfs_getattr_readdirplus_enable(inode); 986 987 trace_nfs_getattr_enter(inode); 988 989 request_mask &= STATX_TYPE | STATX_MODE | STATX_NLINK | STATX_UID | 990 STATX_GID | STATX_ATIME | STATX_MTIME | STATX_CTIME | 991 STATX_INO | STATX_SIZE | STATX_BLOCKS | STATX_BTIME | 992 STATX_CHANGE_COOKIE; 993 994 if (!(fattr_valid & NFS_ATTR_FATTR_BTIME)) 995 request_mask &= ~STATX_BTIME; 996 997 if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) { 998 if (readdirplus_enabled) 999 nfs_readdirplus_parent_cache_hit(path->dentry); 1000 goto out_no_revalidate; 1001 } 1002 1003 /* Flush out writes to the server in order to update c/mtime/version. */ 1004 if ((request_mask & (STATX_CTIME | STATX_MTIME | STATX_CHANGE_COOKIE)) && 1005 S_ISREG(inode->i_mode)) { 1006 if (nfs_have_delegated_mtime(inode)) 1007 filemap_fdatawrite(inode->i_mapping); 1008 else 1009 filemap_write_and_wait(inode->i_mapping); 1010 } 1011 1012 /* 1013 * We may force a getattr if the user cares about atime. 1014 * 1015 * Note that we only have to check the vfsmount flags here: 1016 * - NFS always sets S_NOATIME by so checking it would give a 1017 * bogus result 1018 * - NFS never sets SB_NOATIME or SB_NODIRATIME so there is 1019 * no point in checking those. 1020 */ 1021 if ((path->mnt->mnt_flags & MNT_NOATIME) || 1022 ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) 1023 request_mask &= ~STATX_ATIME; 1024 1025 /* Is the user requesting attributes that might need revalidation? */ 1026 if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME| 1027 STATX_MTIME|STATX_UID|STATX_GID| 1028 STATX_SIZE|STATX_BLOCKS|STATX_BTIME| 1029 STATX_CHANGE_COOKIE))) 1030 goto out_no_revalidate; 1031 1032 /* Check whether the cached attributes are stale */ 1033 do_update |= force_sync || nfs_attribute_cache_expired(inode); 1034 cache_validity = READ_ONCE(NFS_I(inode)->cache_validity); 1035 do_update |= cache_validity & NFS_INO_INVALID_CHANGE; 1036 if (request_mask & STATX_ATIME) 1037 do_update |= cache_validity & NFS_INO_INVALID_ATIME; 1038 if (request_mask & STATX_CTIME) 1039 do_update |= cache_validity & NFS_INO_INVALID_CTIME; 1040 if (request_mask & STATX_MTIME) 1041 do_update |= cache_validity & NFS_INO_INVALID_MTIME; 1042 if (request_mask & STATX_SIZE) 1043 do_update |= cache_validity & NFS_INO_INVALID_SIZE; 1044 if (request_mask & STATX_NLINK) 1045 do_update |= cache_validity & NFS_INO_INVALID_NLINK; 1046 if (request_mask & STATX_MODE) 1047 do_update |= cache_validity & NFS_INO_INVALID_MODE; 1048 if (request_mask & (STATX_UID | STATX_GID)) 1049 do_update |= cache_validity & NFS_INO_INVALID_OTHER; 1050 if (request_mask & STATX_BLOCKS) 1051 do_update |= cache_validity & NFS_INO_INVALID_BLOCKS; 1052 if (request_mask & STATX_BTIME) 1053 do_update |= cache_validity & NFS_INO_INVALID_BTIME; 1054 1055 if (do_update) { 1056 if (readdirplus_enabled) 1057 nfs_readdirplus_parent_cache_miss(path->dentry); 1058 err = __nfs_revalidate_inode(server, inode); 1059 if (err) 1060 goto out; 1061 } else if (readdirplus_enabled) 1062 nfs_readdirplus_parent_cache_hit(path->dentry); 1063 out_no_revalidate: 1064 /* Only return attributes that were revalidated. */ 1065 stat->result_mask = nfs_get_valid_attrmask(inode) | request_mask; 1066 1067 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat); 1068 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode)); 1069 stat->change_cookie = inode_peek_iversion_raw(inode); 1070 stat->attributes_mask |= STATX_ATTR_CHANGE_MONOTONIC; 1071 if (server->change_attr_type != NFS4_CHANGE_TYPE_IS_UNDEFINED) 1072 stat->attributes |= STATX_ATTR_CHANGE_MONOTONIC; 1073 if (S_ISDIR(inode->i_mode)) 1074 stat->blksize = NFS_SERVER(inode)->dtsize; 1075 stat->btime = NFS_I(inode)->btime; 1076 out: 1077 trace_nfs_getattr_exit(inode, err); 1078 return err; 1079 } 1080 EXPORT_SYMBOL_GPL(nfs_getattr); 1081 1082 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx) 1083 { 1084 refcount_set(&l_ctx->count, 1); 1085 l_ctx->lockowner = current->files; 1086 INIT_LIST_HEAD(&l_ctx->list); 1087 atomic_set(&l_ctx->io_count, 0); 1088 } 1089 1090 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx) 1091 { 1092 struct nfs_lock_context *pos; 1093 1094 list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) { 1095 if (pos->lockowner != current->files) 1096 continue; 1097 if (refcount_inc_not_zero(&pos->count)) 1098 return pos; 1099 } 1100 return NULL; 1101 } 1102 1103 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx) 1104 { 1105 struct nfs_lock_context *res, *new = NULL; 1106 struct inode *inode = d_inode(ctx->dentry); 1107 1108 rcu_read_lock(); 1109 res = __nfs_find_lock_context(ctx); 1110 rcu_read_unlock(); 1111 if (res == NULL) { 1112 new = kmalloc(sizeof(*new), GFP_KERNEL_ACCOUNT); 1113 if (new == NULL) 1114 return ERR_PTR(-ENOMEM); 1115 nfs_init_lock_context(new); 1116 spin_lock(&inode->i_lock); 1117 res = __nfs_find_lock_context(ctx); 1118 if (res == NULL) { 1119 new->open_context = get_nfs_open_context(ctx); 1120 if (new->open_context) { 1121 list_add_tail_rcu(&new->list, 1122 &ctx->lock_context.list); 1123 res = new; 1124 new = NULL; 1125 } else 1126 res = ERR_PTR(-EBADF); 1127 } 1128 spin_unlock(&inode->i_lock); 1129 kfree(new); 1130 } 1131 return res; 1132 } 1133 EXPORT_SYMBOL_GPL(nfs_get_lock_context); 1134 1135 void nfs_put_lock_context(struct nfs_lock_context *l_ctx) 1136 { 1137 struct nfs_open_context *ctx = l_ctx->open_context; 1138 struct inode *inode = d_inode(ctx->dentry); 1139 1140 if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock)) 1141 return; 1142 list_del_rcu(&l_ctx->list); 1143 spin_unlock(&inode->i_lock); 1144 put_nfs_open_context(ctx); 1145 kfree_rcu(l_ctx, rcu_head); 1146 } 1147 EXPORT_SYMBOL_GPL(nfs_put_lock_context); 1148 1149 /** 1150 * nfs_close_context - Common close_context() routine NFSv2/v3 1151 * @ctx: pointer to context 1152 * @is_sync: is this a synchronous close 1153 * 1154 * Ensure that the attributes are up to date if we're mounted 1155 * with close-to-open semantics and we have cached data that will 1156 * need to be revalidated on open. 1157 */ 1158 void nfs_close_context(struct nfs_open_context *ctx, int is_sync) 1159 { 1160 struct nfs_inode *nfsi; 1161 struct inode *inode; 1162 1163 if (!(ctx->mode & FMODE_WRITE)) 1164 return; 1165 if (!is_sync) 1166 return; 1167 inode = d_inode(ctx->dentry); 1168 if (nfs_have_read_or_write_delegation(inode)) 1169 return; 1170 nfsi = NFS_I(inode); 1171 if (inode->i_mapping->nrpages == 0) 1172 return; 1173 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 1174 return; 1175 if (!list_empty(&nfsi->open_files)) 1176 return; 1177 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NOCTO) 1178 return; 1179 nfs_revalidate_inode(inode, 1180 NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE); 1181 } 1182 EXPORT_SYMBOL_GPL(nfs_close_context); 1183 1184 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, 1185 fmode_t f_mode, 1186 struct file *filp) 1187 { 1188 struct nfs_open_context *ctx; 1189 1190 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL_ACCOUNT); 1191 if (!ctx) 1192 return ERR_PTR(-ENOMEM); 1193 nfs_sb_active(dentry->d_sb); 1194 ctx->dentry = dget(dentry); 1195 if (filp) 1196 ctx->cred = get_cred(filp->f_cred); 1197 else 1198 ctx->cred = get_current_cred(); 1199 rcu_assign_pointer(ctx->ll_cred, NULL); 1200 ctx->state = NULL; 1201 ctx->mode = f_mode; 1202 ctx->flags = 0; 1203 ctx->error = 0; 1204 ctx->flock_owner = (fl_owner_t)filp; 1205 nfs_init_lock_context(&ctx->lock_context); 1206 ctx->lock_context.open_context = ctx; 1207 INIT_LIST_HEAD(&ctx->list); 1208 ctx->mdsthreshold = NULL; 1209 nfs_localio_file_init(&ctx->nfl); 1210 1211 return ctx; 1212 } 1213 EXPORT_SYMBOL_GPL(alloc_nfs_open_context); 1214 1215 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx) 1216 { 1217 if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count)) 1218 return ctx; 1219 return NULL; 1220 } 1221 EXPORT_SYMBOL_GPL(get_nfs_open_context); 1222 1223 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync) 1224 { 1225 struct inode *inode = d_inode(ctx->dentry); 1226 struct super_block *sb = ctx->dentry->d_sb; 1227 1228 if (!refcount_dec_and_test(&ctx->lock_context.count)) 1229 return; 1230 if (!list_empty(&ctx->list)) { 1231 spin_lock(&inode->i_lock); 1232 list_del_rcu(&ctx->list); 1233 spin_unlock(&inode->i_lock); 1234 } 1235 if (inode != NULL) 1236 NFS_PROTO(inode)->close_context(ctx, is_sync); 1237 put_cred(ctx->cred); 1238 dput(ctx->dentry); 1239 nfs_sb_deactive(sb); 1240 put_rpccred(rcu_dereference_protected(ctx->ll_cred, 1)); 1241 kfree(ctx->mdsthreshold); 1242 nfs_close_local_fh(&ctx->nfl); 1243 kfree_rcu(ctx, rcu_head); 1244 } 1245 1246 void put_nfs_open_context(struct nfs_open_context *ctx) 1247 { 1248 __put_nfs_open_context(ctx, 0); 1249 } 1250 EXPORT_SYMBOL_GPL(put_nfs_open_context); 1251 1252 static void put_nfs_open_context_sync(struct nfs_open_context *ctx) 1253 { 1254 __put_nfs_open_context(ctx, 1); 1255 } 1256 1257 /* 1258 * Ensure that mmap has a recent RPC credential for use when writing out 1259 * shared pages 1260 */ 1261 void nfs_inode_attach_open_context(struct nfs_open_context *ctx) 1262 { 1263 struct inode *inode = d_inode(ctx->dentry); 1264 struct nfs_inode *nfsi = NFS_I(inode); 1265 1266 spin_lock(&inode->i_lock); 1267 if (list_empty(&nfsi->open_files) && 1268 nfs_ooo_test(nfsi)) 1269 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA | 1270 NFS_INO_REVAL_FORCED); 1271 list_add_tail_rcu(&ctx->list, &nfsi->open_files); 1272 spin_unlock(&inode->i_lock); 1273 } 1274 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context); 1275 1276 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx) 1277 { 1278 filp->private_data = get_nfs_open_context(ctx); 1279 set_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags); 1280 if (list_empty(&ctx->list)) 1281 nfs_inode_attach_open_context(ctx); 1282 } 1283 EXPORT_SYMBOL_GPL(nfs_file_set_open_context); 1284 1285 /* 1286 * Given an inode, search for an open context with the desired characteristics 1287 */ 1288 struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode) 1289 { 1290 struct nfs_inode *nfsi = NFS_I(inode); 1291 struct nfs_open_context *pos, *ctx = NULL; 1292 1293 rcu_read_lock(); 1294 list_for_each_entry_rcu(pos, &nfsi->open_files, list) { 1295 if (cred != NULL && cred_fscmp(pos->cred, cred) != 0) 1296 continue; 1297 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode) 1298 continue; 1299 if (!test_bit(NFS_CONTEXT_FILE_OPEN, &pos->flags)) 1300 continue; 1301 ctx = get_nfs_open_context(pos); 1302 if (ctx) 1303 break; 1304 } 1305 rcu_read_unlock(); 1306 return ctx; 1307 } 1308 1309 void nfs_file_clear_open_context(struct file *filp) 1310 { 1311 struct nfs_open_context *ctx = nfs_file_open_context(filp); 1312 1313 if (ctx) { 1314 struct inode *inode = d_inode(ctx->dentry); 1315 1316 clear_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags); 1317 /* 1318 * We fatal error on write before. Try to writeback 1319 * every page again. 1320 */ 1321 if (ctx->error < 0) 1322 invalidate_inode_pages2(inode->i_mapping); 1323 filp->private_data = NULL; 1324 put_nfs_open_context_sync(ctx); 1325 } 1326 } 1327 1328 /* 1329 * These allocate and release file read/write context information. 1330 */ 1331 int nfs_open(struct inode *inode, struct file *filp) 1332 { 1333 struct nfs_open_context *ctx; 1334 1335 ctx = alloc_nfs_open_context(file_dentry(filp), 1336 flags_to_mode(filp->f_flags), filp); 1337 if (IS_ERR(ctx)) 1338 return PTR_ERR(ctx); 1339 nfs_file_set_open_context(filp, ctx); 1340 put_nfs_open_context(ctx); 1341 nfs_fscache_open_file(inode, filp); 1342 return 0; 1343 } 1344 1345 /* 1346 * This function is called whenever some part of NFS notices that 1347 * the cached attributes have to be refreshed. 1348 */ 1349 int 1350 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) 1351 { 1352 int status = -ESTALE; 1353 struct nfs_fattr *fattr = NULL; 1354 struct nfs_inode *nfsi = NFS_I(inode); 1355 1356 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n", 1357 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode)); 1358 1359 trace_nfs_revalidate_inode_enter(inode); 1360 1361 if (is_bad_inode(inode)) 1362 goto out; 1363 if (NFS_STALE(inode)) 1364 goto out; 1365 1366 /* pNFS: Attributes aren't updated until we layoutcommit */ 1367 if (S_ISREG(inode->i_mode)) { 1368 status = pnfs_sync_inode(inode, false); 1369 if (status) 1370 goto out; 1371 } 1372 1373 status = -ENOMEM; 1374 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode)); 1375 if (fattr == NULL) 1376 goto out; 1377 1378 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE); 1379 1380 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, inode); 1381 if (status != 0) { 1382 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n", 1383 inode->i_sb->s_id, 1384 (unsigned long long)NFS_FILEID(inode), status); 1385 switch (status) { 1386 case -ETIMEDOUT: 1387 /* A soft timeout occurred. Use cached information? */ 1388 if (server->flags & NFS_MOUNT_SOFTREVAL) 1389 status = 0; 1390 break; 1391 case -ESTALE: 1392 if (!S_ISDIR(inode->i_mode)) 1393 nfs_set_inode_stale(inode); 1394 else 1395 nfs_zap_caches(inode); 1396 } 1397 goto out; 1398 } 1399 1400 status = nfs_refresh_inode(inode, fattr); 1401 if (status) { 1402 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n", 1403 inode->i_sb->s_id, 1404 (unsigned long long)NFS_FILEID(inode), status); 1405 goto out; 1406 } 1407 1408 if (nfsi->cache_validity & NFS_INO_INVALID_ACL) 1409 nfs_zap_acl_cache(inode); 1410 1411 nfs_setsecurity(inode, fattr); 1412 1413 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n", 1414 inode->i_sb->s_id, 1415 (unsigned long long)NFS_FILEID(inode)); 1416 1417 out: 1418 nfs_free_fattr(fattr); 1419 trace_nfs_revalidate_inode_exit(inode, status); 1420 return status; 1421 } 1422 1423 int nfs_attribute_cache_expired(struct inode *inode) 1424 { 1425 if (nfs_have_delegated_attributes(inode)) 1426 return 0; 1427 return nfs_attribute_timeout(inode); 1428 } 1429 1430 /** 1431 * nfs_revalidate_inode - Revalidate the inode attributes 1432 * @inode: pointer to inode struct 1433 * @flags: cache flags to check 1434 * 1435 * Updates inode attribute information by retrieving the data from the server. 1436 */ 1437 int nfs_revalidate_inode(struct inode *inode, unsigned long flags) 1438 { 1439 if (!nfs_check_cache_invalid(inode, flags)) 1440 return NFS_STALE(inode) ? -ESTALE : 0; 1441 return __nfs_revalidate_inode(NFS_SERVER(inode), inode); 1442 } 1443 EXPORT_SYMBOL_GPL(nfs_revalidate_inode); 1444 1445 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping) 1446 { 1447 int ret; 1448 1449 nfs_fscache_invalidate(inode, 0); 1450 if (mapping->nrpages != 0) { 1451 if (S_ISREG(inode->i_mode)) { 1452 ret = nfs_sync_mapping(mapping); 1453 if (ret < 0) 1454 return ret; 1455 } 1456 ret = invalidate_inode_pages2(mapping); 1457 if (ret < 0) 1458 return ret; 1459 } 1460 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE); 1461 1462 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n", 1463 inode->i_sb->s_id, 1464 (unsigned long long)NFS_FILEID(inode)); 1465 return 0; 1466 } 1467 1468 /** 1469 * nfs_clear_invalid_mapping - Conditionally clear a mapping 1470 * @mapping: pointer to mapping 1471 * 1472 * If the NFS_INO_INVALID_DATA inode flag is set, clear the mapping. 1473 */ 1474 int nfs_clear_invalid_mapping(struct address_space *mapping) 1475 { 1476 struct inode *inode = mapping->host; 1477 struct nfs_inode *nfsi = NFS_I(inode); 1478 unsigned long *bitlock = &nfsi->flags; 1479 int ret = 0; 1480 1481 /* 1482 * We must clear NFS_INO_INVALID_DATA first to ensure that 1483 * invalidations that come in while we're shooting down the mappings 1484 * are respected. But, that leaves a race window where one revalidator 1485 * can clear the flag, and then another checks it before the mapping 1486 * gets invalidated. Fix that by serializing access to this part of 1487 * the function. 1488 * 1489 * At the same time, we need to allow other tasks to see whether we 1490 * might be in the middle of invalidating the pages, so we only set 1491 * the bit lock here if it looks like we're going to be doing that. 1492 */ 1493 for (;;) { 1494 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING, 1495 nfs_wait_bit_killable, 1496 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE); 1497 if (ret) 1498 goto out; 1499 smp_rmb(); /* pairs with smp_wmb() below */ 1500 if (test_bit(NFS_INO_INVALIDATING, bitlock)) 1501 continue; 1502 /* pairs with nfs_set_cache_invalid()'s smp_store_release() */ 1503 if (!(smp_load_acquire(&nfsi->cache_validity) & NFS_INO_INVALID_DATA)) 1504 goto out; 1505 /* Slow-path that double-checks with spinlock held */ 1506 spin_lock(&inode->i_lock); 1507 if (test_bit(NFS_INO_INVALIDATING, bitlock)) { 1508 spin_unlock(&inode->i_lock); 1509 continue; 1510 } 1511 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 1512 break; 1513 spin_unlock(&inode->i_lock); 1514 goto out; 1515 } 1516 1517 set_bit(NFS_INO_INVALIDATING, bitlock); 1518 smp_wmb(); 1519 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA; 1520 nfs_ooo_clear(nfsi); 1521 spin_unlock(&inode->i_lock); 1522 trace_nfs_invalidate_mapping_enter(inode); 1523 ret = nfs_invalidate_mapping(inode, mapping); 1524 trace_nfs_invalidate_mapping_exit(inode, ret); 1525 1526 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock); 1527 smp_mb__after_atomic(); 1528 wake_up_bit(bitlock, NFS_INO_INVALIDATING); 1529 out: 1530 return ret; 1531 } 1532 1533 bool nfs_mapping_need_revalidate_inode(struct inode *inode) 1534 { 1535 return nfs_check_cache_invalid(inode, NFS_INO_INVALID_CHANGE) || 1536 NFS_STALE(inode); 1537 } 1538 1539 int nfs_revalidate_mapping_rcu(struct inode *inode) 1540 { 1541 struct nfs_inode *nfsi = NFS_I(inode); 1542 unsigned long *bitlock = &nfsi->flags; 1543 int ret = 0; 1544 1545 if (IS_SWAPFILE(inode)) 1546 goto out; 1547 if (nfs_mapping_need_revalidate_inode(inode)) { 1548 ret = -ECHILD; 1549 goto out; 1550 } 1551 spin_lock(&inode->i_lock); 1552 if (test_bit(NFS_INO_INVALIDATING, bitlock) || 1553 (nfsi->cache_validity & NFS_INO_INVALID_DATA)) 1554 ret = -ECHILD; 1555 spin_unlock(&inode->i_lock); 1556 out: 1557 return ret; 1558 } 1559 1560 /** 1561 * nfs_revalidate_mapping - Revalidate the pagecache 1562 * @inode: pointer to host inode 1563 * @mapping: pointer to mapping 1564 */ 1565 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping) 1566 { 1567 /* swapfiles are not supposed to be shared. */ 1568 if (IS_SWAPFILE(inode)) 1569 return 0; 1570 1571 if (nfs_mapping_need_revalidate_inode(inode)) { 1572 int ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 1573 if (ret < 0) 1574 return ret; 1575 } 1576 1577 return nfs_clear_invalid_mapping(mapping); 1578 } 1579 1580 static bool nfs_file_has_writers(struct nfs_inode *nfsi) 1581 { 1582 struct inode *inode = &nfsi->vfs_inode; 1583 1584 if (!S_ISREG(inode->i_mode)) 1585 return false; 1586 if (list_empty(&nfsi->open_files)) 1587 return false; 1588 return inode_is_open_for_write(inode); 1589 } 1590 1591 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi) 1592 { 1593 return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi); 1594 } 1595 1596 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1597 { 1598 struct timespec64 ts; 1599 1600 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE) 1601 && (fattr->valid & NFS_ATTR_FATTR_CHANGE) 1602 && inode_eq_iversion_raw(inode, fattr->pre_change_attr)) { 1603 inode_set_iversion_raw(inode, fattr->change_attr); 1604 if (S_ISDIR(inode->i_mode)) 1605 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA); 1606 else if (nfs_server_capable(inode, NFS_CAP_XATTR)) 1607 nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR); 1608 } 1609 /* If we have atomic WCC data, we may update some attributes */ 1610 ts = inode_get_ctime(inode); 1611 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME) 1612 && (fattr->valid & NFS_ATTR_FATTR_CTIME) 1613 && timespec64_equal(&ts, &fattr->pre_ctime)) { 1614 inode_set_ctime_to_ts(inode, fattr->ctime); 1615 } 1616 1617 ts = inode_get_mtime(inode); 1618 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME) 1619 && (fattr->valid & NFS_ATTR_FATTR_MTIME) 1620 && timespec64_equal(&ts, &fattr->pre_mtime)) { 1621 inode_set_mtime_to_ts(inode, fattr->mtime); 1622 } 1623 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE) 1624 && (fattr->valid & NFS_ATTR_FATTR_SIZE) 1625 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size) 1626 && !nfs_have_writebacks(inode)) { 1627 trace_nfs_size_wcc(inode, fattr->size); 1628 i_size_write(inode, nfs_size_to_loff_t(fattr->size)); 1629 } 1630 } 1631 1632 /** 1633 * nfs_check_inode_attributes - verify consistency of the inode attribute cache 1634 * @inode: pointer to inode 1635 * @fattr: updated attributes 1636 * 1637 * Verifies the attribute cache. If we have just changed the attributes, 1638 * so that fattr carries weak cache consistency data, then it may 1639 * also update the ctime/mtime/change_attribute. 1640 */ 1641 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr) 1642 { 1643 struct nfs_inode *nfsi = NFS_I(inode); 1644 loff_t cur_size, new_isize; 1645 unsigned long invalid = 0; 1646 struct timespec64 ts; 1647 1648 if (nfs_have_delegated_attributes(inode)) 1649 return 0; 1650 1651 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) { 1652 /* Only a mounted-on-fileid? Just exit */ 1653 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) 1654 return 0; 1655 /* Has the inode gone and changed behind our back? */ 1656 } else if (nfsi->fileid != fattr->fileid) { 1657 /* Is this perhaps the mounted-on fileid? */ 1658 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) && 1659 nfsi->fileid == fattr->mounted_on_fileid) 1660 return 0; 1661 return -ESTALE; 1662 } 1663 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) 1664 return -ESTALE; 1665 1666 1667 if (!nfs_file_has_buffered_writers(nfsi)) { 1668 /* Verify a few of the more important attributes */ 1669 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr)) 1670 invalid |= NFS_INO_INVALID_CHANGE; 1671 1672 ts = inode_get_mtime(inode); 1673 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime)) 1674 invalid |= NFS_INO_INVALID_MTIME; 1675 1676 ts = inode_get_ctime(inode); 1677 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime)) 1678 invalid |= NFS_INO_INVALID_CTIME; 1679 1680 if (fattr->valid & NFS_ATTR_FATTR_SIZE) { 1681 cur_size = i_size_read(inode); 1682 new_isize = nfs_size_to_loff_t(fattr->size); 1683 if (cur_size != new_isize) 1684 invalid |= NFS_INO_INVALID_SIZE; 1685 } 1686 } 1687 1688 /* Have any file permissions changed? */ 1689 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) 1690 invalid |= NFS_INO_INVALID_MODE; 1691 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid)) 1692 invalid |= NFS_INO_INVALID_OTHER; 1693 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid)) 1694 invalid |= NFS_INO_INVALID_OTHER; 1695 1696 /* Has the link count changed? */ 1697 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink) 1698 invalid |= NFS_INO_INVALID_NLINK; 1699 1700 ts = inode_get_atime(inode); 1701 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime)) 1702 invalid |= NFS_INO_INVALID_ATIME; 1703 1704 if (invalid != 0) 1705 nfs_set_cache_invalid(inode, invalid); 1706 1707 nfsi->read_cache_jiffies = fattr->time_start; 1708 return 0; 1709 } 1710 1711 static atomic_long_t nfs_attr_generation_counter; 1712 1713 static unsigned long nfs_read_attr_generation_counter(void) 1714 { 1715 return atomic_long_read(&nfs_attr_generation_counter); 1716 } 1717 1718 unsigned long nfs_inc_attr_generation_counter(void) 1719 { 1720 return atomic_long_inc_return(&nfs_attr_generation_counter); 1721 } 1722 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter); 1723 1724 void nfs_fattr_init(struct nfs_fattr *fattr) 1725 { 1726 fattr->valid = 0; 1727 fattr->time_start = jiffies; 1728 fattr->gencount = nfs_inc_attr_generation_counter(); 1729 fattr->owner_name = NULL; 1730 fattr->group_name = NULL; 1731 fattr->mdsthreshold = NULL; 1732 } 1733 EXPORT_SYMBOL_GPL(nfs_fattr_init); 1734 1735 /** 1736 * nfs_fattr_set_barrier 1737 * @fattr: attributes 1738 * 1739 * Used to set a barrier after an attribute was updated. This 1740 * barrier ensures that older attributes from RPC calls that may 1741 * have raced with our update cannot clobber these new values. 1742 * Note that you are still responsible for ensuring that other 1743 * operations which change the attribute on the server do not 1744 * collide. 1745 */ 1746 void nfs_fattr_set_barrier(struct nfs_fattr *fattr) 1747 { 1748 fattr->gencount = nfs_inc_attr_generation_counter(); 1749 } 1750 1751 struct nfs_fattr *nfs_alloc_fattr(void) 1752 { 1753 struct nfs_fattr *fattr; 1754 1755 fattr = kmalloc(sizeof(*fattr), GFP_KERNEL); 1756 if (fattr != NULL) { 1757 nfs_fattr_init(fattr); 1758 fattr->label = NULL; 1759 } 1760 return fattr; 1761 } 1762 EXPORT_SYMBOL_GPL(nfs_alloc_fattr); 1763 1764 struct nfs_fattr *nfs_alloc_fattr_with_label(struct nfs_server *server) 1765 { 1766 struct nfs_fattr *fattr = nfs_alloc_fattr(); 1767 1768 if (!fattr) 1769 return NULL; 1770 1771 fattr->label = nfs4_label_alloc(server, GFP_KERNEL); 1772 if (IS_ERR(fattr->label)) { 1773 kfree(fattr); 1774 return NULL; 1775 } 1776 1777 return fattr; 1778 } 1779 EXPORT_SYMBOL_GPL(nfs_alloc_fattr_with_label); 1780 1781 struct nfs_fh *nfs_alloc_fhandle(void) 1782 { 1783 struct nfs_fh *fh; 1784 1785 fh = kmalloc(sizeof(struct nfs_fh), GFP_KERNEL); 1786 if (fh != NULL) 1787 fh->size = 0; 1788 return fh; 1789 } 1790 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle); 1791 1792 #ifdef NFS_DEBUG 1793 /* 1794 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle 1795 * in the same way that wireshark does 1796 * 1797 * @fh: file handle 1798 * 1799 * For debugging only. 1800 */ 1801 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh) 1802 { 1803 /* wireshark uses 32-bit AUTODIN crc and does a bitwise 1804 * not on the result */ 1805 return nfs_fhandle_hash(fh); 1806 } 1807 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash); 1808 1809 /* 1810 * _nfs_display_fhandle - display an NFS file handle on the console 1811 * 1812 * @fh: file handle to display 1813 * @caption: display caption 1814 * 1815 * For debugging only. 1816 */ 1817 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption) 1818 { 1819 unsigned short i; 1820 1821 if (fh == NULL || fh->size == 0) { 1822 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh); 1823 return; 1824 } 1825 1826 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n", 1827 caption, fh, fh->size, _nfs_display_fhandle_hash(fh)); 1828 for (i = 0; i < fh->size; i += 16) { 1829 __be32 *pos = (__be32 *)&fh->data[i]; 1830 1831 switch ((fh->size - i - 1) >> 2) { 1832 case 0: 1833 printk(KERN_DEFAULT " %08x\n", 1834 be32_to_cpup(pos)); 1835 break; 1836 case 1: 1837 printk(KERN_DEFAULT " %08x %08x\n", 1838 be32_to_cpup(pos), be32_to_cpup(pos + 1)); 1839 break; 1840 case 2: 1841 printk(KERN_DEFAULT " %08x %08x %08x\n", 1842 be32_to_cpup(pos), be32_to_cpup(pos + 1), 1843 be32_to_cpup(pos + 2)); 1844 break; 1845 default: 1846 printk(KERN_DEFAULT " %08x %08x %08x %08x\n", 1847 be32_to_cpup(pos), be32_to_cpup(pos + 1), 1848 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3)); 1849 } 1850 } 1851 } 1852 EXPORT_SYMBOL_GPL(_nfs_display_fhandle); 1853 #endif 1854 1855 /** 1856 * nfs_inode_attrs_cmp_generic - compare attributes 1857 * @fattr: attributes 1858 * @inode: pointer to inode 1859 * 1860 * Attempt to divine whether or not an RPC call reply carrying stale 1861 * attributes got scheduled after another call carrying updated ones. 1862 * Note also the check for wraparound of 'attr_gencount' 1863 * 1864 * The function returns '1' if it thinks the attributes in @fattr are 1865 * more recent than the ones cached in @inode. Otherwise it returns 1866 * the value '0'. 1867 */ 1868 static int nfs_inode_attrs_cmp_generic(const struct nfs_fattr *fattr, 1869 const struct inode *inode) 1870 { 1871 unsigned long attr_gencount = NFS_I(inode)->attr_gencount; 1872 1873 return (long)(fattr->gencount - attr_gencount) > 0 || 1874 (long)(attr_gencount - nfs_read_attr_generation_counter()) > 0; 1875 } 1876 1877 /** 1878 * nfs_inode_attrs_cmp_monotonic - compare attributes 1879 * @fattr: attributes 1880 * @inode: pointer to inode 1881 * 1882 * Attempt to divine whether or not an RPC call reply carrying stale 1883 * attributes got scheduled after another call carrying updated ones. 1884 * 1885 * We assume that the server observes monotonic semantics for 1886 * the change attribute, so a larger value means that the attributes in 1887 * @fattr are more recent, in which case the function returns the 1888 * value '1'. 1889 * A return value of '0' indicates no measurable change 1890 * A return value of '-1' means that the attributes in @inode are 1891 * more recent. 1892 */ 1893 static int nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr *fattr, 1894 const struct inode *inode) 1895 { 1896 s64 diff = fattr->change_attr - inode_peek_iversion_raw(inode); 1897 if (diff > 0) 1898 return 1; 1899 return diff == 0 ? 0 : -1; 1900 } 1901 1902 /** 1903 * nfs_inode_attrs_cmp_strict_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 strictly 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 '-1' means that the attributes in @inode are 1915 * more recent or unchanged. 1916 */ 1917 static int nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr *fattr, 1918 const struct inode *inode) 1919 { 1920 return nfs_inode_attrs_cmp_monotonic(fattr, inode) > 0 ? 1 : -1; 1921 } 1922 1923 /** 1924 * nfs_inode_attrs_cmp - compare attributes 1925 * @fattr: attributes 1926 * @inode: pointer to inode 1927 * 1928 * This function returns '1' if it thinks the attributes in @fattr are 1929 * more recent than the ones cached in @inode. It returns '-1' if 1930 * the attributes in @inode are more recent than the ones in @fattr, 1931 * and it returns 0 if not sure. 1932 */ 1933 static int nfs_inode_attrs_cmp(const struct nfs_fattr *fattr, 1934 const struct inode *inode) 1935 { 1936 if (nfs_inode_attrs_cmp_generic(fattr, inode) > 0) 1937 return 1; 1938 switch (NFS_SERVER(inode)->change_attr_type) { 1939 case NFS4_CHANGE_TYPE_IS_UNDEFINED: 1940 break; 1941 case NFS4_CHANGE_TYPE_IS_TIME_METADATA: 1942 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE)) 1943 break; 1944 return nfs_inode_attrs_cmp_monotonic(fattr, inode); 1945 default: 1946 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE)) 1947 break; 1948 return nfs_inode_attrs_cmp_strict_monotonic(fattr, inode); 1949 } 1950 return 0; 1951 } 1952 1953 /** 1954 * nfs_inode_finish_partial_attr_update - complete a previous inode update 1955 * @fattr: attributes 1956 * @inode: pointer to inode 1957 * 1958 * Returns '1' if the last attribute update left the inode cached 1959 * attributes in a partially unrevalidated state, and @fattr 1960 * matches the change attribute of that partial update. 1961 * Otherwise returns '0'. 1962 */ 1963 static int nfs_inode_finish_partial_attr_update(const struct nfs_fattr *fattr, 1964 const struct inode *inode) 1965 { 1966 const unsigned long check_valid = 1967 NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME | 1968 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE | 1969 NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_OTHER | 1970 NFS_INO_INVALID_NLINK | NFS_INO_INVALID_BTIME; 1971 unsigned long cache_validity = NFS_I(inode)->cache_validity; 1972 enum nfs4_change_attr_type ctype = NFS_SERVER(inode)->change_attr_type; 1973 1974 if (ctype != NFS4_CHANGE_TYPE_IS_UNDEFINED && 1975 !(cache_validity & NFS_INO_INVALID_CHANGE) && 1976 (cache_validity & check_valid) != 0 && 1977 (fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && 1978 nfs_inode_attrs_cmp_monotonic(fattr, inode) == 0) 1979 return 1; 1980 return 0; 1981 } 1982 1983 static void nfs_ooo_merge(struct nfs_inode *nfsi, 1984 u64 start, u64 end) 1985 { 1986 int i, cnt; 1987 1988 if (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER) 1989 /* No point merging anything */ 1990 return; 1991 1992 if (!nfsi->ooo) { 1993 nfsi->ooo = kmalloc(sizeof(*nfsi->ooo), GFP_ATOMIC); 1994 if (!nfsi->ooo) { 1995 nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER; 1996 return; 1997 } 1998 nfsi->ooo->cnt = 0; 1999 } 2000 2001 /* add this range, merging if possible */ 2002 cnt = nfsi->ooo->cnt; 2003 for (i = 0; i < cnt; i++) { 2004 if (end == nfsi->ooo->gap[i].start) 2005 end = nfsi->ooo->gap[i].end; 2006 else if (start == nfsi->ooo->gap[i].end) 2007 start = nfsi->ooo->gap[i].start; 2008 else 2009 continue; 2010 /* Remove 'i' from table and loop to insert the new range */ 2011 cnt -= 1; 2012 nfsi->ooo->gap[i] = nfsi->ooo->gap[cnt]; 2013 i = -1; 2014 } 2015 if (start != end) { 2016 if (cnt >= ARRAY_SIZE(nfsi->ooo->gap)) { 2017 nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER; 2018 kfree(nfsi->ooo); 2019 nfsi->ooo = NULL; 2020 return; 2021 } 2022 nfsi->ooo->gap[cnt].start = start; 2023 nfsi->ooo->gap[cnt].end = end; 2024 cnt += 1; 2025 } 2026 nfsi->ooo->cnt = cnt; 2027 } 2028 2029 static void nfs_ooo_record(struct nfs_inode *nfsi, 2030 struct nfs_fattr *fattr) 2031 { 2032 /* This reply was out-of-order, so record in the 2033 * pre/post change id, possibly cancelling 2034 * gaps created when iversion was jumpped forward. 2035 */ 2036 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) && 2037 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE)) 2038 nfs_ooo_merge(nfsi, 2039 fattr->change_attr, 2040 fattr->pre_change_attr); 2041 } 2042 2043 static int nfs_refresh_inode_locked(struct inode *inode, 2044 struct nfs_fattr *fattr) 2045 { 2046 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode); 2047 int ret = 0; 2048 2049 trace_nfs_refresh_inode_enter(inode); 2050 2051 if (attr_cmp > 0 || nfs_inode_finish_partial_attr_update(fattr, inode)) 2052 ret = nfs_update_inode(inode, fattr); 2053 else { 2054 nfs_ooo_record(NFS_I(inode), fattr); 2055 2056 if (attr_cmp == 0) 2057 ret = nfs_check_inode_attributes(inode, fattr); 2058 } 2059 2060 trace_nfs_refresh_inode_exit(inode, ret); 2061 return ret; 2062 } 2063 2064 /** 2065 * nfs_refresh_inode - try to update the inode attribute cache 2066 * @inode: pointer to inode 2067 * @fattr: updated attributes 2068 * 2069 * Check that an RPC call that returned attributes has not overlapped with 2070 * other recent updates of the inode metadata, then decide whether it is 2071 * safe to do a full update of the inode attributes, or whether just to 2072 * call nfs_check_inode_attributes. 2073 */ 2074 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr) 2075 { 2076 int status; 2077 2078 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 2079 return 0; 2080 spin_lock(&inode->i_lock); 2081 status = nfs_refresh_inode_locked(inode, fattr); 2082 spin_unlock(&inode->i_lock); 2083 2084 return status; 2085 } 2086 EXPORT_SYMBOL_GPL(nfs_refresh_inode); 2087 2088 static int nfs_post_op_update_inode_locked(struct inode *inode, 2089 struct nfs_fattr *fattr, unsigned int invalid) 2090 { 2091 if (S_ISDIR(inode->i_mode)) 2092 invalid |= NFS_INO_INVALID_DATA; 2093 nfs_set_cache_invalid(inode, invalid); 2094 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 2095 return 0; 2096 return nfs_refresh_inode_locked(inode, fattr); 2097 } 2098 2099 /** 2100 * nfs_post_op_update_inode - try to update the inode attribute cache 2101 * @inode: pointer to inode 2102 * @fattr: updated attributes 2103 * 2104 * After an operation that has changed the inode metadata, mark the 2105 * attribute cache as being invalid, then try to update it. 2106 * 2107 * NB: if the server didn't return any post op attributes, this 2108 * function will force the retrieval of attributes before the next 2109 * NFS request. Thus it should be used only for operations that 2110 * are expected to change one or more attributes, to avoid 2111 * unnecessary NFS requests and trips through nfs_update_inode(). 2112 */ 2113 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr) 2114 { 2115 int status; 2116 2117 spin_lock(&inode->i_lock); 2118 nfs_fattr_set_barrier(fattr); 2119 status = nfs_post_op_update_inode_locked(inode, fattr, 2120 NFS_INO_INVALID_CHANGE 2121 | NFS_INO_INVALID_CTIME 2122 | NFS_INO_REVAL_FORCED); 2123 spin_unlock(&inode->i_lock); 2124 2125 return status; 2126 } 2127 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode); 2128 2129 /** 2130 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache 2131 * @inode: pointer to inode 2132 * @fattr: updated attributes 2133 * 2134 * After an operation that has changed the inode metadata, mark the 2135 * attribute cache as being invalid, then try to update it. Fake up 2136 * weak cache consistency data, if none exist. 2137 * 2138 * This function is mainly designed to be used by the ->write_done() functions. 2139 */ 2140 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr) 2141 { 2142 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode); 2143 int status; 2144 2145 /* Don't do a WCC update if these attributes are already stale */ 2146 if (attr_cmp < 0) 2147 return 0; 2148 if ((fattr->valid & NFS_ATTR_FATTR) == 0 || !attr_cmp) { 2149 /* Record the pre/post change info before clearing PRECHANGE */ 2150 nfs_ooo_record(NFS_I(inode), fattr); 2151 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE 2152 | NFS_ATTR_FATTR_PRESIZE 2153 | NFS_ATTR_FATTR_PREMTIME 2154 | NFS_ATTR_FATTR_PRECTIME); 2155 goto out_noforce; 2156 } 2157 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && 2158 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) { 2159 fattr->pre_change_attr = inode_peek_iversion_raw(inode); 2160 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE; 2161 } 2162 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 && 2163 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) { 2164 fattr->pre_ctime = inode_get_ctime(inode); 2165 fattr->valid |= NFS_ATTR_FATTR_PRECTIME; 2166 } 2167 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 && 2168 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) { 2169 fattr->pre_mtime = inode_get_mtime(inode); 2170 fattr->valid |= NFS_ATTR_FATTR_PREMTIME; 2171 } 2172 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 && 2173 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) { 2174 fattr->pre_size = i_size_read(inode); 2175 fattr->valid |= NFS_ATTR_FATTR_PRESIZE; 2176 } 2177 out_noforce: 2178 status = nfs_post_op_update_inode_locked(inode, fattr, 2179 NFS_INO_INVALID_CHANGE 2180 | NFS_INO_INVALID_CTIME 2181 | NFS_INO_INVALID_MTIME 2182 | NFS_INO_INVALID_BLOCKS); 2183 return status; 2184 } 2185 2186 /** 2187 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache 2188 * @inode: pointer to inode 2189 * @fattr: updated attributes 2190 * 2191 * After an operation that has changed the inode metadata, mark the 2192 * attribute cache as being invalid, then try to update it. Fake up 2193 * weak cache consistency data, if none exist. 2194 * 2195 * This function is mainly designed to be used by the ->write_done() functions. 2196 */ 2197 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr) 2198 { 2199 int status; 2200 2201 spin_lock(&inode->i_lock); 2202 nfs_fattr_set_barrier(fattr); 2203 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 2204 spin_unlock(&inode->i_lock); 2205 return status; 2206 } 2207 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc); 2208 2209 2210 /* 2211 * Many nfs protocol calls return the new file attributes after 2212 * an operation. Here we update the inode to reflect the state 2213 * of the server's inode. 2214 * 2215 * This is a bit tricky because we have to make sure all dirty pages 2216 * have been sent off to the server before calling invalidate_inode_pages. 2217 * To make sure no other process adds more write requests while we try 2218 * our best to flush them, we make them sleep during the attribute refresh. 2219 * 2220 * A very similar scenario holds for the dir cache. 2221 */ 2222 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr) 2223 { 2224 struct nfs_server *server = NFS_SERVER(inode); 2225 struct nfs_inode *nfsi = NFS_I(inode); 2226 loff_t cur_isize, new_isize; 2227 u64 fattr_supported = server->fattr_valid; 2228 unsigned long invalid = 0; 2229 unsigned long now = jiffies; 2230 unsigned long save_cache_validity; 2231 bool have_writers = nfs_file_has_buffered_writers(nfsi); 2232 bool cache_revalidated = true; 2233 bool attr_changed = false; 2234 bool have_delegation; 2235 2236 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%llx)\n", 2237 __func__, inode->i_sb->s_id, inode->i_ino, 2238 nfs_display_fhandle_hash(NFS_FH(inode)), 2239 atomic_read(&inode->i_count), fattr->valid); 2240 2241 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) { 2242 /* Only a mounted-on-fileid? Just exit */ 2243 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) 2244 return 0; 2245 /* Has the inode gone and changed behind our back? */ 2246 } else if (nfsi->fileid != fattr->fileid) { 2247 /* Is this perhaps the mounted-on fileid? */ 2248 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) && 2249 nfsi->fileid == fattr->mounted_on_fileid) 2250 return 0; 2251 printk(KERN_ERR "NFS: server %s error: fileid changed\n" 2252 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n", 2253 NFS_SERVER(inode)->nfs_client->cl_hostname, 2254 inode->i_sb->s_id, (long long)nfsi->fileid, 2255 (long long)fattr->fileid); 2256 goto out_err; 2257 } 2258 2259 /* 2260 * Make sure the inode's type hasn't changed. 2261 */ 2262 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) { 2263 /* 2264 * Big trouble! The inode has become a different object. 2265 */ 2266 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n", 2267 __func__, inode->i_ino, inode->i_mode, fattr->mode); 2268 goto out_err; 2269 } 2270 2271 /* Update the fsid? */ 2272 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) && 2273 !nfs_fsid_equal(&server->fsid, &fattr->fsid) && 2274 !IS_AUTOMOUNT(inode)) 2275 server->fsid = fattr->fsid; 2276 2277 /* Save the delegation state before clearing cache_validity */ 2278 have_delegation = nfs_have_delegated_attributes(inode); 2279 2280 /* 2281 * Update the read time so we don't revalidate too often. 2282 */ 2283 nfsi->read_cache_jiffies = fattr->time_start; 2284 2285 /* Fix up any delegated attributes in the struct nfs_fattr */ 2286 nfs_fattr_fixup_delegated(inode, fattr); 2287 2288 save_cache_validity = nfsi->cache_validity; 2289 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR 2290 | NFS_INO_INVALID_ATIME 2291 | NFS_INO_REVAL_FORCED 2292 | NFS_INO_INVALID_BLOCKS); 2293 2294 /* Do atomic weak cache consistency updates */ 2295 nfs_wcc_update_inode(inode, fattr); 2296 2297 if (pnfs_layoutcommit_outstanding(inode)) { 2298 nfsi->cache_validity |= 2299 save_cache_validity & 2300 (NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME | 2301 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE | 2302 NFS_INO_INVALID_BLOCKS); 2303 cache_revalidated = false; 2304 } 2305 2306 /* More cache consistency checks */ 2307 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) { 2308 if (!have_writers && nfsi->ooo && nfsi->ooo->cnt == 1 && 2309 nfsi->ooo->gap[0].end == inode_peek_iversion_raw(inode)) { 2310 /* There is one remaining gap that hasn't been 2311 * merged into iversion - do that now. 2312 */ 2313 inode_set_iversion_raw(inode, nfsi->ooo->gap[0].start); 2314 kfree(nfsi->ooo); 2315 nfsi->ooo = NULL; 2316 } 2317 if (!inode_eq_iversion_raw(inode, fattr->change_attr)) { 2318 /* Could it be a race with writeback? */ 2319 if (!(have_writers || have_delegation)) { 2320 invalid |= NFS_INO_INVALID_DATA 2321 | NFS_INO_INVALID_ACCESS 2322 | NFS_INO_INVALID_ACL 2323 | NFS_INO_INVALID_XATTR; 2324 /* Force revalidate of all attributes */ 2325 save_cache_validity |= NFS_INO_INVALID_CTIME 2326 | NFS_INO_INVALID_MTIME 2327 | NFS_INO_INVALID_SIZE 2328 | NFS_INO_INVALID_BLOCKS 2329 | NFS_INO_INVALID_NLINK 2330 | NFS_INO_INVALID_MODE 2331 | NFS_INO_INVALID_OTHER 2332 | NFS_INO_INVALID_BTIME; 2333 if (S_ISDIR(inode->i_mode)) 2334 nfs_force_lookup_revalidate(inode); 2335 attr_changed = true; 2336 dprintk("NFS: change_attr change on server for file %s/%ld\n", 2337 inode->i_sb->s_id, 2338 inode->i_ino); 2339 } else if (!have_delegation) { 2340 nfs_ooo_record(nfsi, fattr); 2341 nfs_ooo_merge(nfsi, inode_peek_iversion_raw(inode), 2342 fattr->change_attr); 2343 } 2344 inode_set_iversion_raw(inode, fattr->change_attr); 2345 } 2346 } else { 2347 nfsi->cache_validity |= 2348 save_cache_validity & NFS_INO_INVALID_CHANGE; 2349 if (!have_delegation || 2350 (nfsi->cache_validity & NFS_INO_INVALID_CHANGE) != 0) 2351 cache_revalidated = false; 2352 } 2353 2354 if (fattr->valid & NFS_ATTR_FATTR_MTIME) 2355 inode_set_mtime_to_ts(inode, fattr->mtime); 2356 else if (fattr_supported & NFS_ATTR_FATTR_MTIME) 2357 nfsi->cache_validity |= 2358 save_cache_validity & NFS_INO_INVALID_MTIME; 2359 2360 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 2361 inode_set_ctime_to_ts(inode, fattr->ctime); 2362 else if (fattr_supported & NFS_ATTR_FATTR_CTIME) 2363 nfsi->cache_validity |= 2364 save_cache_validity & NFS_INO_INVALID_CTIME; 2365 2366 if (fattr->valid & NFS_ATTR_FATTR_BTIME) 2367 nfsi->btime = fattr->btime; 2368 else if (fattr_supported & NFS_ATTR_FATTR_BTIME) 2369 nfsi->cache_validity |= 2370 save_cache_validity & NFS_INO_INVALID_BTIME; 2371 2372 /* Check if our cached file size is stale */ 2373 if (fattr->valid & NFS_ATTR_FATTR_SIZE) { 2374 new_isize = nfs_size_to_loff_t(fattr->size); 2375 cur_isize = i_size_read(inode); 2376 if (new_isize != cur_isize && !have_delegation) { 2377 /* Do we perhaps have any outstanding writes, or has 2378 * the file grown beyond our last write? */ 2379 if (!nfs_have_writebacks(inode) || new_isize > cur_isize) { 2380 trace_nfs_size_update(inode, new_isize); 2381 i_size_write(inode, new_isize); 2382 if (!have_writers) 2383 invalid |= NFS_INO_INVALID_DATA; 2384 } 2385 } 2386 if (new_isize == 0 && 2387 !(fattr->valid & (NFS_ATTR_FATTR_SPACE_USED | 2388 NFS_ATTR_FATTR_BLOCKS_USED))) { 2389 fattr->du.nfs3.used = 0; 2390 fattr->valid |= NFS_ATTR_FATTR_SPACE_USED; 2391 } 2392 } else 2393 nfsi->cache_validity |= 2394 save_cache_validity & NFS_INO_INVALID_SIZE; 2395 2396 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 2397 inode_set_atime_to_ts(inode, fattr->atime); 2398 else if (fattr_supported & NFS_ATTR_FATTR_ATIME) 2399 nfsi->cache_validity |= 2400 save_cache_validity & NFS_INO_INVALID_ATIME; 2401 2402 if (fattr->valid & NFS_ATTR_FATTR_MODE) { 2403 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) { 2404 umode_t newmode = inode->i_mode & S_IFMT; 2405 newmode |= fattr->mode & S_IALLUGO; 2406 inode->i_mode = newmode; 2407 invalid |= NFS_INO_INVALID_ACCESS 2408 | NFS_INO_INVALID_ACL; 2409 } 2410 } else if (fattr_supported & NFS_ATTR_FATTR_MODE) 2411 nfsi->cache_validity |= 2412 save_cache_validity & NFS_INO_INVALID_MODE; 2413 2414 if (fattr->valid & NFS_ATTR_FATTR_OWNER) { 2415 if (!uid_eq(inode->i_uid, fattr->uid)) { 2416 invalid |= NFS_INO_INVALID_ACCESS 2417 | NFS_INO_INVALID_ACL; 2418 inode->i_uid = fattr->uid; 2419 } 2420 } else if (fattr_supported & NFS_ATTR_FATTR_OWNER) 2421 nfsi->cache_validity |= 2422 save_cache_validity & NFS_INO_INVALID_OTHER; 2423 2424 if (fattr->valid & NFS_ATTR_FATTR_GROUP) { 2425 if (!gid_eq(inode->i_gid, fattr->gid)) { 2426 invalid |= NFS_INO_INVALID_ACCESS 2427 | NFS_INO_INVALID_ACL; 2428 inode->i_gid = fattr->gid; 2429 } 2430 } else if (fattr_supported & NFS_ATTR_FATTR_GROUP) 2431 nfsi->cache_validity |= 2432 save_cache_validity & NFS_INO_INVALID_OTHER; 2433 2434 if (fattr->valid & NFS_ATTR_FATTR_NLINK) { 2435 if (inode->i_nlink != fattr->nlink) 2436 set_nlink(inode, fattr->nlink); 2437 } else if (fattr_supported & NFS_ATTR_FATTR_NLINK) 2438 nfsi->cache_validity |= 2439 save_cache_validity & NFS_INO_INVALID_NLINK; 2440 2441 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) { 2442 /* 2443 * report the blocks in 512byte units 2444 */ 2445 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 2446 } else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED) 2447 nfsi->cache_validity |= 2448 save_cache_validity & NFS_INO_INVALID_BLOCKS; 2449 2450 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED) 2451 inode->i_blocks = fattr->du.nfs2.blocks; 2452 else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED) 2453 nfsi->cache_validity |= 2454 save_cache_validity & NFS_INO_INVALID_BLOCKS; 2455 2456 /* Update attrtimeo value if we're out of the unstable period */ 2457 if (attr_changed) { 2458 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 2459 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 2460 nfsi->attrtimeo_timestamp = now; 2461 /* Set barrier to be more recent than all outstanding updates */ 2462 nfsi->attr_gencount = nfs_inc_attr_generation_counter(); 2463 } else { 2464 if (cache_revalidated) { 2465 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, 2466 nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) { 2467 nfsi->attrtimeo <<= 1; 2468 if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode)) 2469 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode); 2470 } 2471 nfsi->attrtimeo_timestamp = now; 2472 } 2473 /* Set the barrier to be more recent than this fattr */ 2474 if ((long)(fattr->gencount - nfsi->attr_gencount) > 0) 2475 nfsi->attr_gencount = fattr->gencount; 2476 } 2477 2478 /* Don't invalidate the data if we were to blame */ 2479 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) 2480 || S_ISLNK(inode->i_mode))) 2481 invalid &= ~NFS_INO_INVALID_DATA; 2482 nfs_set_cache_invalid(inode, invalid); 2483 2484 return 0; 2485 out_err: 2486 /* 2487 * No need to worry about unhashing the dentry, as the 2488 * lookup validation will know that the inode is bad. 2489 * (But we fall through to invalidate the caches.) 2490 */ 2491 nfs_set_inode_stale_locked(inode); 2492 return -ESTALE; 2493 } 2494 2495 struct inode *nfs_alloc_inode(struct super_block *sb) 2496 { 2497 struct nfs_inode *nfsi; 2498 nfsi = alloc_inode_sb(sb, nfs_inode_cachep, GFP_KERNEL); 2499 if (!nfsi) 2500 return NULL; 2501 nfsi->flags = 0UL; 2502 nfsi->cache_validity = 0UL; 2503 nfsi->ooo = NULL; 2504 #if IS_ENABLED(CONFIG_NFS_V4) 2505 nfsi->nfs4_acl = NULL; 2506 #endif /* CONFIG_NFS_V4 */ 2507 #ifdef CONFIG_NFS_V4_2 2508 nfsi->xattr_cache = NULL; 2509 #endif 2510 nfs_netfs_inode_init(nfsi); 2511 2512 return &nfsi->vfs_inode; 2513 } 2514 EXPORT_SYMBOL_GPL(nfs_alloc_inode); 2515 2516 void nfs_free_inode(struct inode *inode) 2517 { 2518 kfree(NFS_I(inode)->ooo); 2519 kmem_cache_free(nfs_inode_cachep, NFS_I(inode)); 2520 } 2521 EXPORT_SYMBOL_GPL(nfs_free_inode); 2522 2523 static inline void nfs4_init_once(struct nfs_inode *nfsi) 2524 { 2525 #if IS_ENABLED(CONFIG_NFS_V4) 2526 INIT_LIST_HEAD(&nfsi->open_states); 2527 nfsi->delegation = NULL; 2528 init_rwsem(&nfsi->rwsem); 2529 nfsi->layout = NULL; 2530 #endif 2531 } 2532 2533 static void init_once(void *foo) 2534 { 2535 struct nfs_inode *nfsi = foo; 2536 2537 inode_init_once(&nfsi->vfs_inode); 2538 INIT_LIST_HEAD(&nfsi->open_files); 2539 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru); 2540 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru); 2541 nfs4_init_once(nfsi); 2542 } 2543 2544 static int __init nfs_init_inodecache(void) 2545 { 2546 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache", 2547 sizeof(struct nfs_inode), 2548 0, (SLAB_RECLAIM_ACCOUNT| 2549 SLAB_ACCOUNT), 2550 init_once); 2551 if (nfs_inode_cachep == NULL) 2552 return -ENOMEM; 2553 2554 return 0; 2555 } 2556 2557 static void nfs_destroy_inodecache(void) 2558 { 2559 /* 2560 * Make sure all delayed rcu free inodes are flushed before we 2561 * destroy cache. 2562 */ 2563 rcu_barrier(); 2564 kmem_cache_destroy(nfs_inode_cachep); 2565 } 2566 2567 struct workqueue_struct *nfslocaliod_workqueue; 2568 struct workqueue_struct *nfsiod_workqueue; 2569 EXPORT_SYMBOL_GPL(nfsiod_workqueue); 2570 2571 /* 2572 * Destroy the nfsiod workqueues 2573 */ 2574 static void nfsiod_stop(void) 2575 { 2576 struct workqueue_struct *wq; 2577 2578 wq = nfsiod_workqueue; 2579 if (wq != NULL) { 2580 nfsiod_workqueue = NULL; 2581 destroy_workqueue(wq); 2582 } 2583 #if IS_ENABLED(CONFIG_NFS_LOCALIO) 2584 wq = nfslocaliod_workqueue; 2585 if (wq != NULL) { 2586 nfslocaliod_workqueue = NULL; 2587 destroy_workqueue(wq); 2588 } 2589 #endif /* CONFIG_NFS_LOCALIO */ 2590 } 2591 2592 /* 2593 * Start the nfsiod workqueues 2594 */ 2595 static int nfsiod_start(void) 2596 { 2597 dprintk("RPC: creating workqueue nfsiod\n"); 2598 nfsiod_workqueue = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0); 2599 if (nfsiod_workqueue == NULL) 2600 return -ENOMEM; 2601 #if IS_ENABLED(CONFIG_NFS_LOCALIO) 2602 /* 2603 * localio writes need to use a normal (non-memreclaim) workqueue. 2604 * When we start getting low on space, XFS goes and calls flush_work() on 2605 * a non-memreclaim work queue, which causes a priority inversion problem. 2606 */ 2607 dprintk("RPC: creating workqueue nfslocaliod\n"); 2608 nfslocaliod_workqueue = alloc_workqueue("nfslocaliod", WQ_UNBOUND, 0); 2609 if (unlikely(nfslocaliod_workqueue == NULL)) { 2610 nfsiod_stop(); 2611 return -ENOMEM; 2612 } 2613 #endif /* CONFIG_NFS_LOCALIO */ 2614 return 0; 2615 } 2616 2617 unsigned int nfs_net_id; 2618 EXPORT_SYMBOL_GPL(nfs_net_id); 2619 2620 static int nfs_net_init(struct net *net) 2621 { 2622 struct nfs_net *nn = net_generic(net, nfs_net_id); 2623 int err; 2624 2625 nfs_clients_init(net); 2626 2627 if (!rpc_proc_register(net, &nn->rpcstats)) { 2628 err = -ENOMEM; 2629 goto err_proc_rpc; 2630 } 2631 2632 err = nfs_fs_proc_net_init(net); 2633 if (err) 2634 goto err_proc_nfs; 2635 2636 return 0; 2637 2638 err_proc_nfs: 2639 rpc_proc_unregister(net, "nfs"); 2640 err_proc_rpc: 2641 nfs_clients_exit(net); 2642 return err; 2643 } 2644 2645 static void nfs_net_exit(struct net *net) 2646 { 2647 rpc_proc_unregister(net, "nfs"); 2648 nfs_fs_proc_net_exit(net); 2649 nfs_clients_exit(net); 2650 } 2651 2652 static struct pernet_operations nfs_net_ops = { 2653 .init = nfs_net_init, 2654 .exit = nfs_net_exit, 2655 .id = &nfs_net_id, 2656 .size = sizeof(struct nfs_net), 2657 }; 2658 2659 #ifdef CONFIG_KEYS 2660 static struct key *nfs_keyring; 2661 2662 static int __init nfs_init_keyring(void) 2663 { 2664 nfs_keyring = keyring_alloc(".nfs", 2665 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, 2666 current_cred(), 2667 (KEY_POS_ALL & ~KEY_POS_SETATTR) | 2668 (KEY_USR_ALL & ~KEY_USR_SETATTR), 2669 KEY_ALLOC_NOT_IN_QUOTA, NULL, NULL); 2670 return PTR_ERR_OR_ZERO(nfs_keyring); 2671 } 2672 2673 static void nfs_exit_keyring(void) 2674 { 2675 key_put(nfs_keyring); 2676 } 2677 #else 2678 static inline int nfs_init_keyring(void) 2679 { 2680 return 0; 2681 } 2682 2683 static inline void nfs_exit_keyring(void) 2684 { 2685 } 2686 #endif /* CONFIG_KEYS */ 2687 2688 /* 2689 * Initialize NFS 2690 */ 2691 static int __init init_nfs_fs(void) 2692 { 2693 int err; 2694 2695 err = nfs_init_keyring(); 2696 if (err) 2697 return err; 2698 2699 err = nfs_sysfs_init(); 2700 if (err < 0) 2701 goto out10; 2702 2703 err = register_pernet_subsys(&nfs_net_ops); 2704 if (err < 0) 2705 goto out9; 2706 2707 err = nfsiod_start(); 2708 if (err) 2709 goto out7; 2710 2711 err = nfs_fs_proc_init(); 2712 if (err) 2713 goto out6; 2714 2715 err = nfs_init_nfspagecache(); 2716 if (err) 2717 goto out5; 2718 2719 err = nfs_init_inodecache(); 2720 if (err) 2721 goto out4; 2722 2723 err = nfs_init_readpagecache(); 2724 if (err) 2725 goto out3; 2726 2727 err = nfs_init_writepagecache(); 2728 if (err) 2729 goto out2; 2730 2731 err = nfs_init_directcache(); 2732 if (err) 2733 goto out1; 2734 2735 err = register_nfs_fs(); 2736 if (err) 2737 goto out0; 2738 2739 return 0; 2740 out0: 2741 nfs_destroy_directcache(); 2742 out1: 2743 nfs_destroy_writepagecache(); 2744 out2: 2745 nfs_destroy_readpagecache(); 2746 out3: 2747 nfs_destroy_inodecache(); 2748 out4: 2749 nfs_destroy_nfspagecache(); 2750 out5: 2751 nfs_fs_proc_exit(); 2752 out6: 2753 nfsiod_stop(); 2754 out7: 2755 unregister_pernet_subsys(&nfs_net_ops); 2756 out9: 2757 nfs_sysfs_exit(); 2758 out10: 2759 nfs_exit_keyring(); 2760 return err; 2761 } 2762 2763 static void __exit exit_nfs_fs(void) 2764 { 2765 nfs_destroy_directcache(); 2766 nfs_destroy_writepagecache(); 2767 nfs_destroy_readpagecache(); 2768 nfs_destroy_inodecache(); 2769 nfs_destroy_nfspagecache(); 2770 unregister_pernet_subsys(&nfs_net_ops); 2771 unregister_nfs_fs(); 2772 nfs_fs_proc_exit(); 2773 nfsiod_stop(); 2774 nfs_sysfs_exit(); 2775 nfs_exit_keyring(); 2776 } 2777 2778 /* Not quite true; I just maintain it */ 2779 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>"); 2780 MODULE_DESCRIPTION("NFS client support"); 2781 MODULE_LICENSE("GPL"); 2782 module_param(enable_ino64, bool, 0644); 2783 2784 module_init(init_nfs_fs) 2785 module_exit(exit_nfs_fs) 2786