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