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 static int nfs_wait_killable(int mode) 76 { 77 freezable_schedule_unsafe(); 78 if (signal_pending_state(mode, current)) 79 return -ERESTARTSYS; 80 return 0; 81 } 82 83 int nfs_wait_bit_killable(struct wait_bit_key *key, int mode) 84 { 85 return nfs_wait_killable(mode); 86 } 87 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable); 88 89 /** 90 * nfs_compat_user_ino64 - returns the user-visible inode number 91 * @fileid: 64-bit fileid 92 * 93 * This function returns a 32-bit inode number if the boot parameter 94 * nfs.enable_ino64 is zero. 95 */ 96 u64 nfs_compat_user_ino64(u64 fileid) 97 { 98 #ifdef CONFIG_COMPAT 99 compat_ulong_t ino; 100 #else 101 unsigned long ino; 102 #endif 103 104 if (enable_ino64) 105 return fileid; 106 ino = fileid; 107 if (sizeof(ino) < sizeof(fileid)) 108 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8; 109 return ino; 110 } 111 112 int nfs_drop_inode(struct inode *inode) 113 { 114 return NFS_STALE(inode) || generic_drop_inode(inode); 115 } 116 EXPORT_SYMBOL_GPL(nfs_drop_inode); 117 118 void nfs_clear_inode(struct inode *inode) 119 { 120 /* 121 * The following should never happen... 122 */ 123 WARN_ON_ONCE(nfs_have_writebacks(inode)); 124 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files)); 125 nfs_zap_acl_cache(inode); 126 nfs_access_zap_cache(inode); 127 nfs_fscache_clear_inode(inode); 128 } 129 EXPORT_SYMBOL_GPL(nfs_clear_inode); 130 131 void nfs_evict_inode(struct inode *inode) 132 { 133 truncate_inode_pages_final(&inode->i_data); 134 clear_inode(inode); 135 nfs_clear_inode(inode); 136 } 137 138 int nfs_sync_inode(struct inode *inode) 139 { 140 inode_dio_wait(inode); 141 return nfs_wb_all(inode); 142 } 143 EXPORT_SYMBOL_GPL(nfs_sync_inode); 144 145 /** 146 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk 147 * @mapping: pointer to struct address_space 148 */ 149 int nfs_sync_mapping(struct address_space *mapping) 150 { 151 int ret = 0; 152 153 if (mapping->nrpages != 0) { 154 unmap_mapping_range(mapping, 0, 0, 0); 155 ret = nfs_wb_all(mapping->host); 156 } 157 return ret; 158 } 159 160 static int nfs_attribute_timeout(struct inode *inode) 161 { 162 struct nfs_inode *nfsi = NFS_I(inode); 163 164 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo); 165 } 166 167 static bool nfs_check_cache_flags_invalid(struct inode *inode, 168 unsigned long flags) 169 { 170 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity); 171 172 return (cache_validity & flags) != 0; 173 } 174 175 bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags) 176 { 177 if (nfs_check_cache_flags_invalid(inode, flags)) 178 return true; 179 return nfs_attribute_cache_expired(inode); 180 } 181 EXPORT_SYMBOL_GPL(nfs_check_cache_invalid); 182 183 #ifdef CONFIG_NFS_V4_2 184 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi) 185 { 186 return nfsi->xattr_cache != NULL; 187 } 188 #else 189 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi) 190 { 191 return false; 192 } 193 #endif 194 195 void nfs_set_cache_invalid(struct inode *inode, unsigned long flags) 196 { 197 struct nfs_inode *nfsi = NFS_I(inode); 198 bool have_delegation = NFS_PROTO(inode)->have_delegation(inode, FMODE_READ); 199 200 if (have_delegation) { 201 if (!(flags & NFS_INO_REVAL_FORCED)) 202 flags &= ~(NFS_INO_INVALID_MODE | 203 NFS_INO_INVALID_OTHER | 204 NFS_INO_INVALID_XATTR); 205 flags &= ~(NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE); 206 } else if (flags & NFS_INO_REVAL_PAGECACHE) 207 flags |= NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE; 208 209 if (!nfs_has_xattr_cache(nfsi)) 210 flags &= ~NFS_INO_INVALID_XATTR; 211 if (flags & NFS_INO_INVALID_DATA) 212 nfs_fscache_invalidate(inode, 0); 213 flags &= ~(NFS_INO_REVAL_PAGECACHE | NFS_INO_REVAL_FORCED); 214 215 nfsi->cache_validity |= flags; 216 217 if (inode->i_mapping->nrpages == 0) 218 nfsi->cache_validity &= ~(NFS_INO_INVALID_DATA | 219 NFS_INO_DATA_INVAL_DEFER); 220 else if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 221 nfsi->cache_validity &= ~NFS_INO_DATA_INVAL_DEFER; 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 | NFS_INO_REVAL_PAGECACHE); 246 } else 247 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR 248 | NFS_INO_INVALID_ACCESS 249 | NFS_INO_INVALID_ACL 250 | NFS_INO_INVALID_XATTR 251 | NFS_INO_REVAL_PAGECACHE); 252 nfs_zap_label_cache_locked(nfsi); 253 } 254 255 void nfs_zap_caches(struct inode *inode) 256 { 257 spin_lock(&inode->i_lock); 258 nfs_zap_caches_locked(inode); 259 spin_unlock(&inode->i_lock); 260 } 261 262 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping) 263 { 264 if (mapping->nrpages != 0) { 265 spin_lock(&inode->i_lock); 266 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA); 267 spin_unlock(&inode->i_lock); 268 } 269 } 270 271 void nfs_zap_acl_cache(struct inode *inode) 272 { 273 void (*clear_acl_cache)(struct inode *); 274 275 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache; 276 if (clear_acl_cache != NULL) 277 clear_acl_cache(inode); 278 spin_lock(&inode->i_lock); 279 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL; 280 spin_unlock(&inode->i_lock); 281 } 282 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache); 283 284 void nfs_invalidate_atime(struct inode *inode) 285 { 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 = (struct nfs_find_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 = (struct nfs_find_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 INIT_LIST_HEAD(&nfsi->commit_info.list); 433 atomic_long_set(&nfsi->commit_info.ncommit, 0); 434 atomic_set(&nfsi->commit_info.rpcs_out, 0); 435 mutex_init(&nfsi->commit_mutex); 436 } 437 438 static void nfs_inode_init_dir(struct nfs_inode *nfsi) 439 { 440 nfsi->cache_change_attribute = 0; 441 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf)); 442 init_rwsem(&nfsi->rmdir_sem); 443 } 444 445 /* 446 * This is our front-end to iget that looks up inodes by file handle 447 * instead of inode number. 448 */ 449 struct inode * 450 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr) 451 { 452 struct nfs_find_desc desc = { 453 .fh = fh, 454 .fattr = fattr 455 }; 456 struct inode *inode = ERR_PTR(-ENOENT); 457 u64 fattr_supported = NFS_SB(sb)->fattr_valid; 458 unsigned long hash; 459 460 nfs_attr_check_mountpoint(sb, fattr); 461 462 if (nfs_attr_use_mounted_on_fileid(fattr)) 463 fattr->fileid = fattr->mounted_on_fileid; 464 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0) 465 goto out_no_inode; 466 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0) 467 goto out_no_inode; 468 469 hash = nfs_fattr_to_ino_t(fattr); 470 471 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc); 472 if (inode == NULL) { 473 inode = ERR_PTR(-ENOMEM); 474 goto out_no_inode; 475 } 476 477 if (inode->i_state & I_NEW) { 478 struct nfs_inode *nfsi = NFS_I(inode); 479 unsigned long now = jiffies; 480 481 /* We set i_ino for the few things that still rely on it, 482 * such as stat(2) */ 483 inode->i_ino = hash; 484 485 /* We can't support update_atime(), since the server will reset it */ 486 inode->i_flags |= S_NOATIME|S_NOCMTIME; 487 inode->i_mode = fattr->mode; 488 nfsi->cache_validity = 0; 489 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0 490 && (fattr_supported & NFS_ATTR_FATTR_MODE)) 491 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE); 492 /* Why so? Because we want revalidate for devices/FIFOs, and 493 * that's precisely what we have in nfs_file_inode_operations. 494 */ 495 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops; 496 if (S_ISREG(inode->i_mode)) { 497 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops; 498 inode->i_data.a_ops = &nfs_file_aops; 499 nfs_inode_init_regular(nfsi); 500 } else if (S_ISDIR(inode->i_mode)) { 501 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops; 502 inode->i_fop = &nfs_dir_operations; 503 inode->i_data.a_ops = &nfs_dir_aops; 504 nfs_inode_init_dir(nfsi); 505 /* Deal with crossing mountpoints */ 506 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT || 507 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) { 508 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) 509 inode->i_op = &nfs_referral_inode_operations; 510 else 511 inode->i_op = &nfs_mountpoint_inode_operations; 512 inode->i_fop = NULL; 513 inode->i_flags |= S_AUTOMOUNT; 514 } 515 } else if (S_ISLNK(inode->i_mode)) { 516 inode->i_op = &nfs_symlink_inode_operations; 517 inode_nohighmem(inode); 518 } else 519 init_special_inode(inode, inode->i_mode, fattr->rdev); 520 521 memset(&inode->i_atime, 0, sizeof(inode->i_atime)); 522 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime)); 523 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime)); 524 inode_set_iversion_raw(inode, 0); 525 inode->i_size = 0; 526 clear_nlink(inode); 527 inode->i_uid = make_kuid(&init_user_ns, -2); 528 inode->i_gid = make_kgid(&init_user_ns, -2); 529 inode->i_blocks = 0; 530 nfsi->write_io = 0; 531 nfsi->read_io = 0; 532 533 nfsi->read_cache_jiffies = fattr->time_start; 534 nfsi->attr_gencount = fattr->gencount; 535 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 536 inode->i_atime = fattr->atime; 537 else if (fattr_supported & NFS_ATTR_FATTR_ATIME) 538 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME); 539 if (fattr->valid & NFS_ATTR_FATTR_MTIME) 540 inode->i_mtime = fattr->mtime; 541 else if (fattr_supported & NFS_ATTR_FATTR_MTIME) 542 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME); 543 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 544 inode->i_ctime = fattr->ctime; 545 else if (fattr_supported & NFS_ATTR_FATTR_CTIME) 546 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME); 547 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) 548 inode_set_iversion_raw(inode, fattr->change_attr); 549 else 550 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE); 551 if (fattr->valid & NFS_ATTR_FATTR_SIZE) 552 inode->i_size = nfs_size_to_loff_t(fattr->size); 553 else 554 nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE); 555 if (fattr->valid & NFS_ATTR_FATTR_NLINK) 556 set_nlink(inode, fattr->nlink); 557 else if (fattr_supported & NFS_ATTR_FATTR_NLINK) 558 nfs_set_cache_invalid(inode, NFS_INO_INVALID_NLINK); 559 if (fattr->valid & NFS_ATTR_FATTR_OWNER) 560 inode->i_uid = fattr->uid; 561 else if (fattr_supported & NFS_ATTR_FATTR_OWNER) 562 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER); 563 if (fattr->valid & NFS_ATTR_FATTR_GROUP) 564 inode->i_gid = fattr->gid; 565 else if (fattr_supported & NFS_ATTR_FATTR_GROUP) 566 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER); 567 if (nfs_server_capable(inode, NFS_CAP_XATTR)) 568 nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR); 569 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED) 570 inode->i_blocks = fattr->du.nfs2.blocks; 571 else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED && 572 fattr->size != 0) 573 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS); 574 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) { 575 /* 576 * report the blocks in 512byte units 577 */ 578 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 579 } else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED && 580 fattr->size != 0) 581 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS); 582 583 nfs_setsecurity(inode, fattr); 584 585 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 586 nfsi->attrtimeo_timestamp = now; 587 nfsi->access_cache = RB_ROOT; 588 589 nfs_fscache_init_inode(inode); 590 591 unlock_new_inode(inode); 592 } else { 593 int err = nfs_refresh_inode(inode, fattr); 594 if (err < 0) { 595 iput(inode); 596 inode = ERR_PTR(err); 597 goto out_no_inode; 598 } 599 } 600 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n", 601 inode->i_sb->s_id, 602 (unsigned long long)NFS_FILEID(inode), 603 nfs_display_fhandle_hash(fh), 604 atomic_read(&inode->i_count)); 605 606 out: 607 return inode; 608 609 out_no_inode: 610 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode)); 611 goto out; 612 } 613 EXPORT_SYMBOL_GPL(nfs_fhget); 614 615 #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) 616 617 int 618 nfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry, 619 struct iattr *attr) 620 { 621 struct inode *inode = d_inode(dentry); 622 struct nfs_fattr *fattr; 623 int error = 0; 624 625 nfs_inc_stats(inode, NFSIOS_VFSSETATTR); 626 627 /* skip mode change if it's just for clearing setuid/setgid */ 628 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID)) 629 attr->ia_valid &= ~ATTR_MODE; 630 631 if (attr->ia_valid & ATTR_SIZE) { 632 BUG_ON(!S_ISREG(inode->i_mode)); 633 634 error = inode_newsize_ok(inode, attr->ia_size); 635 if (error) 636 return error; 637 638 if (attr->ia_size == i_size_read(inode)) 639 attr->ia_valid &= ~ATTR_SIZE; 640 } 641 642 /* Optimization: if the end result is no change, don't RPC */ 643 if (((attr->ia_valid & NFS_VALID_ATTRS) & ~(ATTR_FILE|ATTR_OPEN)) == 0) 644 return 0; 645 646 trace_nfs_setattr_enter(inode); 647 648 /* Write all dirty data */ 649 if (S_ISREG(inode->i_mode)) 650 nfs_sync_inode(inode); 651 652 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode)); 653 if (fattr == NULL) { 654 error = -ENOMEM; 655 goto out; 656 } 657 658 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr); 659 if (error == 0) 660 error = nfs_refresh_inode(inode, fattr); 661 nfs_free_fattr(fattr); 662 out: 663 trace_nfs_setattr_exit(inode, error); 664 return error; 665 } 666 EXPORT_SYMBOL_GPL(nfs_setattr); 667 668 /** 669 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall 670 * @inode: inode of the file used 671 * @offset: file offset to start truncating 672 * 673 * This is a copy of the common vmtruncate, but with the locking 674 * corrected to take into account the fact that NFS requires 675 * inode->i_size to be updated under the inode->i_lock. 676 * Note: must be called with inode->i_lock held! 677 */ 678 static int nfs_vmtruncate(struct inode * inode, loff_t offset) 679 { 680 int err; 681 682 err = inode_newsize_ok(inode, offset); 683 if (err) 684 goto out; 685 686 trace_nfs_size_truncate(inode, offset); 687 i_size_write(inode, offset); 688 /* Optimisation */ 689 if (offset == 0) 690 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_DATA | 691 NFS_INO_DATA_INVAL_DEFER); 692 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE; 693 694 spin_unlock(&inode->i_lock); 695 truncate_pagecache(inode, offset); 696 spin_lock(&inode->i_lock); 697 out: 698 return err; 699 } 700 701 /** 702 * nfs_setattr_update_inode - Update inode metadata after a setattr call. 703 * @inode: pointer to struct inode 704 * @attr: pointer to struct iattr 705 * @fattr: pointer to struct nfs_fattr 706 * 707 * Note: we do this in the *proc.c in order to ensure that 708 * it works for things like exclusive creates too. 709 */ 710 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr, 711 struct nfs_fattr *fattr) 712 { 713 /* Barrier: bump the attribute generation count. */ 714 nfs_fattr_set_barrier(fattr); 715 716 spin_lock(&inode->i_lock); 717 NFS_I(inode)->attr_gencount = fattr->gencount; 718 if ((attr->ia_valid & ATTR_SIZE) != 0) { 719 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME | 720 NFS_INO_INVALID_BLOCKS); 721 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC); 722 nfs_vmtruncate(inode, attr->ia_size); 723 } 724 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) { 725 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME; 726 if ((attr->ia_valid & ATTR_KILL_SUID) != 0 && 727 inode->i_mode & S_ISUID) 728 inode->i_mode &= ~S_ISUID; 729 if ((attr->ia_valid & ATTR_KILL_SGID) != 0 && 730 (inode->i_mode & (S_ISGID | S_IXGRP)) == 731 (S_ISGID | S_IXGRP)) 732 inode->i_mode &= ~S_ISGID; 733 if ((attr->ia_valid & ATTR_MODE) != 0) { 734 int mode = attr->ia_mode & S_IALLUGO; 735 mode |= inode->i_mode & ~S_IALLUGO; 736 inode->i_mode = mode; 737 } 738 if ((attr->ia_valid & ATTR_UID) != 0) 739 inode->i_uid = attr->ia_uid; 740 if ((attr->ia_valid & ATTR_GID) != 0) 741 inode->i_gid = attr->ia_gid; 742 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 743 inode->i_ctime = fattr->ctime; 744 else 745 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE 746 | NFS_INO_INVALID_CTIME); 747 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS 748 | NFS_INO_INVALID_ACL); 749 } 750 if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) { 751 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME 752 | NFS_INO_INVALID_CTIME); 753 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 754 inode->i_atime = fattr->atime; 755 else if (attr->ia_valid & ATTR_ATIME_SET) 756 inode->i_atime = attr->ia_atime; 757 else 758 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME); 759 760 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 761 inode->i_ctime = fattr->ctime; 762 else 763 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE 764 | NFS_INO_INVALID_CTIME); 765 } 766 if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) { 767 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME 768 | NFS_INO_INVALID_CTIME); 769 if (fattr->valid & NFS_ATTR_FATTR_MTIME) 770 inode->i_mtime = fattr->mtime; 771 else if (attr->ia_valid & ATTR_MTIME_SET) 772 inode->i_mtime = attr->ia_mtime; 773 else 774 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME); 775 776 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 777 inode->i_ctime = fattr->ctime; 778 else 779 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE 780 | NFS_INO_INVALID_CTIME); 781 } 782 if (fattr->valid) 783 nfs_update_inode(inode, fattr); 784 spin_unlock(&inode->i_lock); 785 } 786 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode); 787 788 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry) 789 { 790 struct dentry *parent; 791 792 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS)) 793 return; 794 parent = dget_parent(dentry); 795 nfs_force_use_readdirplus(d_inode(parent)); 796 dput(parent); 797 } 798 799 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry) 800 { 801 struct dentry *parent; 802 803 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS)) 804 return; 805 parent = dget_parent(dentry); 806 nfs_advise_use_readdirplus(d_inode(parent)); 807 dput(parent); 808 } 809 810 static u32 nfs_get_valid_attrmask(struct inode *inode) 811 { 812 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity); 813 u32 reply_mask = STATX_INO | STATX_TYPE; 814 815 if (!(cache_validity & NFS_INO_INVALID_ATIME)) 816 reply_mask |= STATX_ATIME; 817 if (!(cache_validity & NFS_INO_INVALID_CTIME)) 818 reply_mask |= STATX_CTIME; 819 if (!(cache_validity & NFS_INO_INVALID_MTIME)) 820 reply_mask |= STATX_MTIME; 821 if (!(cache_validity & NFS_INO_INVALID_SIZE)) 822 reply_mask |= STATX_SIZE; 823 if (!(cache_validity & NFS_INO_INVALID_NLINK)) 824 reply_mask |= STATX_NLINK; 825 if (!(cache_validity & NFS_INO_INVALID_MODE)) 826 reply_mask |= STATX_MODE; 827 if (!(cache_validity & NFS_INO_INVALID_OTHER)) 828 reply_mask |= STATX_UID | STATX_GID; 829 if (!(cache_validity & NFS_INO_INVALID_BLOCKS)) 830 reply_mask |= STATX_BLOCKS; 831 return reply_mask; 832 } 833 834 int nfs_getattr(struct user_namespace *mnt_userns, const struct path *path, 835 struct kstat *stat, u32 request_mask, unsigned int query_flags) 836 { 837 struct inode *inode = d_inode(path->dentry); 838 struct nfs_server *server = NFS_SERVER(inode); 839 unsigned long cache_validity; 840 int err = 0; 841 bool force_sync = query_flags & AT_STATX_FORCE_SYNC; 842 bool do_update = false; 843 844 trace_nfs_getattr_enter(inode); 845 846 request_mask &= STATX_TYPE | STATX_MODE | STATX_NLINK | STATX_UID | 847 STATX_GID | STATX_ATIME | STATX_MTIME | STATX_CTIME | 848 STATX_INO | STATX_SIZE | STATX_BLOCKS; 849 850 if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) { 851 nfs_readdirplus_parent_cache_hit(path->dentry); 852 goto out_no_revalidate; 853 } 854 855 /* Flush out writes to the server in order to update c/mtime. */ 856 if ((request_mask & (STATX_CTIME | STATX_MTIME)) && 857 S_ISREG(inode->i_mode)) 858 filemap_write_and_wait(inode->i_mapping); 859 860 /* 861 * We may force a getattr if the user cares about atime. 862 * 863 * Note that we only have to check the vfsmount flags here: 864 * - NFS always sets S_NOATIME by so checking it would give a 865 * bogus result 866 * - NFS never sets SB_NOATIME or SB_NODIRATIME so there is 867 * no point in checking those. 868 */ 869 if ((path->mnt->mnt_flags & MNT_NOATIME) || 870 ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) 871 request_mask &= ~STATX_ATIME; 872 873 /* Is the user requesting attributes that might need revalidation? */ 874 if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME| 875 STATX_MTIME|STATX_UID|STATX_GID| 876 STATX_SIZE|STATX_BLOCKS))) 877 goto out_no_revalidate; 878 879 /* Check whether the cached attributes are stale */ 880 do_update |= force_sync || nfs_attribute_cache_expired(inode); 881 cache_validity = READ_ONCE(NFS_I(inode)->cache_validity); 882 do_update |= cache_validity & NFS_INO_INVALID_CHANGE; 883 if (request_mask & STATX_ATIME) 884 do_update |= cache_validity & NFS_INO_INVALID_ATIME; 885 if (request_mask & STATX_CTIME) 886 do_update |= cache_validity & NFS_INO_INVALID_CTIME; 887 if (request_mask & STATX_MTIME) 888 do_update |= cache_validity & NFS_INO_INVALID_MTIME; 889 if (request_mask & STATX_SIZE) 890 do_update |= cache_validity & NFS_INO_INVALID_SIZE; 891 if (request_mask & STATX_NLINK) 892 do_update |= cache_validity & NFS_INO_INVALID_NLINK; 893 if (request_mask & STATX_MODE) 894 do_update |= cache_validity & NFS_INO_INVALID_MODE; 895 if (request_mask & (STATX_UID | STATX_GID)) 896 do_update |= cache_validity & NFS_INO_INVALID_OTHER; 897 if (request_mask & STATX_BLOCKS) 898 do_update |= cache_validity & NFS_INO_INVALID_BLOCKS; 899 900 if (do_update) { 901 /* Update the attribute cache */ 902 if (!(server->flags & NFS_MOUNT_NOAC)) 903 nfs_readdirplus_parent_cache_miss(path->dentry); 904 else 905 nfs_readdirplus_parent_cache_hit(path->dentry); 906 err = __nfs_revalidate_inode(server, inode); 907 if (err) 908 goto out; 909 } else 910 nfs_readdirplus_parent_cache_hit(path->dentry); 911 out_no_revalidate: 912 /* Only return attributes that were revalidated. */ 913 stat->result_mask = nfs_get_valid_attrmask(inode) | request_mask; 914 915 generic_fillattr(&init_user_ns, inode, stat); 916 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode)); 917 if (S_ISDIR(inode->i_mode)) 918 stat->blksize = NFS_SERVER(inode)->dtsize; 919 out: 920 trace_nfs_getattr_exit(inode, err); 921 return err; 922 } 923 EXPORT_SYMBOL_GPL(nfs_getattr); 924 925 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx) 926 { 927 refcount_set(&l_ctx->count, 1); 928 l_ctx->lockowner = current->files; 929 INIT_LIST_HEAD(&l_ctx->list); 930 atomic_set(&l_ctx->io_count, 0); 931 } 932 933 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx) 934 { 935 struct nfs_lock_context *pos; 936 937 list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) { 938 if (pos->lockowner != current->files) 939 continue; 940 if (refcount_inc_not_zero(&pos->count)) 941 return pos; 942 } 943 return NULL; 944 } 945 946 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx) 947 { 948 struct nfs_lock_context *res, *new = NULL; 949 struct inode *inode = d_inode(ctx->dentry); 950 951 rcu_read_lock(); 952 res = __nfs_find_lock_context(ctx); 953 rcu_read_unlock(); 954 if (res == NULL) { 955 new = kmalloc(sizeof(*new), GFP_KERNEL); 956 if (new == NULL) 957 return ERR_PTR(-ENOMEM); 958 nfs_init_lock_context(new); 959 spin_lock(&inode->i_lock); 960 res = __nfs_find_lock_context(ctx); 961 if (res == NULL) { 962 new->open_context = get_nfs_open_context(ctx); 963 if (new->open_context) { 964 list_add_tail_rcu(&new->list, 965 &ctx->lock_context.list); 966 res = new; 967 new = NULL; 968 } else 969 res = ERR_PTR(-EBADF); 970 } 971 spin_unlock(&inode->i_lock); 972 kfree(new); 973 } 974 return res; 975 } 976 EXPORT_SYMBOL_GPL(nfs_get_lock_context); 977 978 void nfs_put_lock_context(struct nfs_lock_context *l_ctx) 979 { 980 struct nfs_open_context *ctx = l_ctx->open_context; 981 struct inode *inode = d_inode(ctx->dentry); 982 983 if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock)) 984 return; 985 list_del_rcu(&l_ctx->list); 986 spin_unlock(&inode->i_lock); 987 put_nfs_open_context(ctx); 988 kfree_rcu(l_ctx, rcu_head); 989 } 990 EXPORT_SYMBOL_GPL(nfs_put_lock_context); 991 992 /** 993 * nfs_close_context - Common close_context() routine NFSv2/v3 994 * @ctx: pointer to context 995 * @is_sync: is this a synchronous close 996 * 997 * Ensure that the attributes are up to date if we're mounted 998 * with close-to-open semantics and we have cached data that will 999 * need to be revalidated on open. 1000 */ 1001 void nfs_close_context(struct nfs_open_context *ctx, int is_sync) 1002 { 1003 struct nfs_inode *nfsi; 1004 struct inode *inode; 1005 1006 if (!(ctx->mode & FMODE_WRITE)) 1007 return; 1008 if (!is_sync) 1009 return; 1010 inode = d_inode(ctx->dentry); 1011 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) 1012 return; 1013 nfsi = NFS_I(inode); 1014 if (inode->i_mapping->nrpages == 0) 1015 return; 1016 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 1017 return; 1018 if (!list_empty(&nfsi->open_files)) 1019 return; 1020 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NOCTO) 1021 return; 1022 nfs_revalidate_inode(inode, 1023 NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE); 1024 } 1025 EXPORT_SYMBOL_GPL(nfs_close_context); 1026 1027 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, 1028 fmode_t f_mode, 1029 struct file *filp) 1030 { 1031 struct nfs_open_context *ctx; 1032 1033 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); 1034 if (!ctx) 1035 return ERR_PTR(-ENOMEM); 1036 nfs_sb_active(dentry->d_sb); 1037 ctx->dentry = dget(dentry); 1038 if (filp) 1039 ctx->cred = get_cred(filp->f_cred); 1040 else 1041 ctx->cred = get_current_cred(); 1042 rcu_assign_pointer(ctx->ll_cred, NULL); 1043 ctx->state = NULL; 1044 ctx->mode = f_mode; 1045 ctx->flags = 0; 1046 ctx->error = 0; 1047 ctx->flock_owner = (fl_owner_t)filp; 1048 nfs_init_lock_context(&ctx->lock_context); 1049 ctx->lock_context.open_context = ctx; 1050 INIT_LIST_HEAD(&ctx->list); 1051 ctx->mdsthreshold = NULL; 1052 return ctx; 1053 } 1054 EXPORT_SYMBOL_GPL(alloc_nfs_open_context); 1055 1056 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx) 1057 { 1058 if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count)) 1059 return ctx; 1060 return NULL; 1061 } 1062 EXPORT_SYMBOL_GPL(get_nfs_open_context); 1063 1064 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync) 1065 { 1066 struct inode *inode = d_inode(ctx->dentry); 1067 struct super_block *sb = ctx->dentry->d_sb; 1068 1069 if (!refcount_dec_and_test(&ctx->lock_context.count)) 1070 return; 1071 if (!list_empty(&ctx->list)) { 1072 spin_lock(&inode->i_lock); 1073 list_del_rcu(&ctx->list); 1074 spin_unlock(&inode->i_lock); 1075 } 1076 if (inode != NULL) 1077 NFS_PROTO(inode)->close_context(ctx, is_sync); 1078 put_cred(ctx->cred); 1079 dput(ctx->dentry); 1080 nfs_sb_deactive(sb); 1081 put_rpccred(rcu_dereference_protected(ctx->ll_cred, 1)); 1082 kfree(ctx->mdsthreshold); 1083 kfree_rcu(ctx, rcu_head); 1084 } 1085 1086 void put_nfs_open_context(struct nfs_open_context *ctx) 1087 { 1088 __put_nfs_open_context(ctx, 0); 1089 } 1090 EXPORT_SYMBOL_GPL(put_nfs_open_context); 1091 1092 static void put_nfs_open_context_sync(struct nfs_open_context *ctx) 1093 { 1094 __put_nfs_open_context(ctx, 1); 1095 } 1096 1097 /* 1098 * Ensure that mmap has a recent RPC credential for use when writing out 1099 * shared pages 1100 */ 1101 void nfs_inode_attach_open_context(struct nfs_open_context *ctx) 1102 { 1103 struct inode *inode = d_inode(ctx->dentry); 1104 struct nfs_inode *nfsi = NFS_I(inode); 1105 1106 spin_lock(&inode->i_lock); 1107 if (list_empty(&nfsi->open_files) && 1108 (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER)) 1109 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA | 1110 NFS_INO_REVAL_FORCED); 1111 list_add_tail_rcu(&ctx->list, &nfsi->open_files); 1112 spin_unlock(&inode->i_lock); 1113 } 1114 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context); 1115 1116 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx) 1117 { 1118 filp->private_data = get_nfs_open_context(ctx); 1119 set_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags); 1120 if (list_empty(&ctx->list)) 1121 nfs_inode_attach_open_context(ctx); 1122 } 1123 EXPORT_SYMBOL_GPL(nfs_file_set_open_context); 1124 1125 /* 1126 * Given an inode, search for an open context with the desired characteristics 1127 */ 1128 struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode) 1129 { 1130 struct nfs_inode *nfsi = NFS_I(inode); 1131 struct nfs_open_context *pos, *ctx = NULL; 1132 1133 rcu_read_lock(); 1134 list_for_each_entry_rcu(pos, &nfsi->open_files, list) { 1135 if (cred != NULL && cred_fscmp(pos->cred, cred) != 0) 1136 continue; 1137 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode) 1138 continue; 1139 if (!test_bit(NFS_CONTEXT_FILE_OPEN, &pos->flags)) 1140 continue; 1141 ctx = get_nfs_open_context(pos); 1142 if (ctx) 1143 break; 1144 } 1145 rcu_read_unlock(); 1146 return ctx; 1147 } 1148 1149 void nfs_file_clear_open_context(struct file *filp) 1150 { 1151 struct nfs_open_context *ctx = nfs_file_open_context(filp); 1152 1153 if (ctx) { 1154 struct inode *inode = d_inode(ctx->dentry); 1155 1156 clear_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags); 1157 /* 1158 * We fatal error on write before. Try to writeback 1159 * every page again. 1160 */ 1161 if (ctx->error < 0) 1162 invalidate_inode_pages2(inode->i_mapping); 1163 filp->private_data = NULL; 1164 put_nfs_open_context_sync(ctx); 1165 } 1166 } 1167 1168 /* 1169 * These allocate and release file read/write context information. 1170 */ 1171 int nfs_open(struct inode *inode, struct file *filp) 1172 { 1173 struct nfs_open_context *ctx; 1174 1175 ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp); 1176 if (IS_ERR(ctx)) 1177 return PTR_ERR(ctx); 1178 nfs_file_set_open_context(filp, ctx); 1179 put_nfs_open_context(ctx); 1180 nfs_fscache_open_file(inode, filp); 1181 return 0; 1182 } 1183 EXPORT_SYMBOL_GPL(nfs_open); 1184 1185 /* 1186 * This function is called whenever some part of NFS notices that 1187 * the cached attributes have to be refreshed. 1188 */ 1189 int 1190 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) 1191 { 1192 int status = -ESTALE; 1193 struct nfs_fattr *fattr = NULL; 1194 struct nfs_inode *nfsi = NFS_I(inode); 1195 1196 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n", 1197 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode)); 1198 1199 trace_nfs_revalidate_inode_enter(inode); 1200 1201 if (is_bad_inode(inode)) 1202 goto out; 1203 if (NFS_STALE(inode)) 1204 goto out; 1205 1206 /* pNFS: Attributes aren't updated until we layoutcommit */ 1207 if (S_ISREG(inode->i_mode)) { 1208 status = pnfs_sync_inode(inode, false); 1209 if (status) 1210 goto out; 1211 } 1212 1213 status = -ENOMEM; 1214 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode)); 1215 if (fattr == NULL) 1216 goto out; 1217 1218 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE); 1219 1220 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, inode); 1221 if (status != 0) { 1222 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n", 1223 inode->i_sb->s_id, 1224 (unsigned long long)NFS_FILEID(inode), status); 1225 switch (status) { 1226 case -ETIMEDOUT: 1227 /* A soft timeout occurred. Use cached information? */ 1228 if (server->flags & NFS_MOUNT_SOFTREVAL) 1229 status = 0; 1230 break; 1231 case -ESTALE: 1232 if (!S_ISDIR(inode->i_mode)) 1233 nfs_set_inode_stale(inode); 1234 else 1235 nfs_zap_caches(inode); 1236 } 1237 goto out; 1238 } 1239 1240 status = nfs_refresh_inode(inode, fattr); 1241 if (status) { 1242 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n", 1243 inode->i_sb->s_id, 1244 (unsigned long long)NFS_FILEID(inode), status); 1245 goto out; 1246 } 1247 1248 if (nfsi->cache_validity & NFS_INO_INVALID_ACL) 1249 nfs_zap_acl_cache(inode); 1250 1251 nfs_setsecurity(inode, fattr); 1252 1253 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n", 1254 inode->i_sb->s_id, 1255 (unsigned long long)NFS_FILEID(inode)); 1256 1257 out: 1258 nfs_free_fattr(fattr); 1259 trace_nfs_revalidate_inode_exit(inode, status); 1260 return status; 1261 } 1262 1263 int nfs_attribute_cache_expired(struct inode *inode) 1264 { 1265 if (nfs_have_delegated_attributes(inode)) 1266 return 0; 1267 return nfs_attribute_timeout(inode); 1268 } 1269 1270 /** 1271 * nfs_revalidate_inode - Revalidate the inode attributes 1272 * @inode: pointer to inode struct 1273 * @flags: cache flags to check 1274 * 1275 * Updates inode attribute information by retrieving the data from the server. 1276 */ 1277 int nfs_revalidate_inode(struct inode *inode, unsigned long flags) 1278 { 1279 if (!nfs_check_cache_invalid(inode, flags)) 1280 return NFS_STALE(inode) ? -ESTALE : 0; 1281 return __nfs_revalidate_inode(NFS_SERVER(inode), inode); 1282 } 1283 EXPORT_SYMBOL_GPL(nfs_revalidate_inode); 1284 1285 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping) 1286 { 1287 int ret; 1288 1289 nfs_fscache_invalidate(inode, 0); 1290 if (mapping->nrpages != 0) { 1291 if (S_ISREG(inode->i_mode)) { 1292 ret = nfs_sync_mapping(mapping); 1293 if (ret < 0) 1294 return ret; 1295 } 1296 ret = invalidate_inode_pages2(mapping); 1297 if (ret < 0) 1298 return ret; 1299 } 1300 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE); 1301 1302 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n", 1303 inode->i_sb->s_id, 1304 (unsigned long long)NFS_FILEID(inode)); 1305 return 0; 1306 } 1307 1308 /** 1309 * nfs_clear_invalid_mapping - Conditionally clear a mapping 1310 * @mapping: pointer to mapping 1311 * 1312 * If the NFS_INO_INVALID_DATA inode flag is set, clear the mapping. 1313 */ 1314 int nfs_clear_invalid_mapping(struct address_space *mapping) 1315 { 1316 struct inode *inode = mapping->host; 1317 struct nfs_inode *nfsi = NFS_I(inode); 1318 unsigned long *bitlock = &nfsi->flags; 1319 int ret = 0; 1320 1321 /* 1322 * We must clear NFS_INO_INVALID_DATA first to ensure that 1323 * invalidations that come in while we're shooting down the mappings 1324 * are respected. But, that leaves a race window where one revalidator 1325 * can clear the flag, and then another checks it before the mapping 1326 * gets invalidated. Fix that by serializing access to this part of 1327 * the function. 1328 * 1329 * At the same time, we need to allow other tasks to see whether we 1330 * might be in the middle of invalidating the pages, so we only set 1331 * the bit lock here if it looks like we're going to be doing that. 1332 */ 1333 for (;;) { 1334 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING, 1335 nfs_wait_bit_killable, TASK_KILLABLE); 1336 if (ret) 1337 goto out; 1338 spin_lock(&inode->i_lock); 1339 if (test_bit(NFS_INO_INVALIDATING, bitlock)) { 1340 spin_unlock(&inode->i_lock); 1341 continue; 1342 } 1343 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 1344 break; 1345 spin_unlock(&inode->i_lock); 1346 goto out; 1347 } 1348 1349 set_bit(NFS_INO_INVALIDATING, bitlock); 1350 smp_wmb(); 1351 nfsi->cache_validity &= 1352 ~(NFS_INO_INVALID_DATA | NFS_INO_DATA_INVAL_DEFER); 1353 spin_unlock(&inode->i_lock); 1354 trace_nfs_invalidate_mapping_enter(inode); 1355 ret = nfs_invalidate_mapping(inode, mapping); 1356 trace_nfs_invalidate_mapping_exit(inode, ret); 1357 1358 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock); 1359 smp_mb__after_atomic(); 1360 wake_up_bit(bitlock, NFS_INO_INVALIDATING); 1361 out: 1362 return ret; 1363 } 1364 1365 bool nfs_mapping_need_revalidate_inode(struct inode *inode) 1366 { 1367 return nfs_check_cache_invalid(inode, NFS_INO_INVALID_CHANGE) || 1368 NFS_STALE(inode); 1369 } 1370 1371 int nfs_revalidate_mapping_rcu(struct inode *inode) 1372 { 1373 struct nfs_inode *nfsi = NFS_I(inode); 1374 unsigned long *bitlock = &nfsi->flags; 1375 int ret = 0; 1376 1377 if (IS_SWAPFILE(inode)) 1378 goto out; 1379 if (nfs_mapping_need_revalidate_inode(inode)) { 1380 ret = -ECHILD; 1381 goto out; 1382 } 1383 spin_lock(&inode->i_lock); 1384 if (test_bit(NFS_INO_INVALIDATING, bitlock) || 1385 (nfsi->cache_validity & NFS_INO_INVALID_DATA)) 1386 ret = -ECHILD; 1387 spin_unlock(&inode->i_lock); 1388 out: 1389 return ret; 1390 } 1391 1392 /** 1393 * nfs_revalidate_mapping - Revalidate the pagecache 1394 * @inode: pointer to host inode 1395 * @mapping: pointer to mapping 1396 */ 1397 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping) 1398 { 1399 /* swapfiles are not supposed to be shared. */ 1400 if (IS_SWAPFILE(inode)) 1401 return 0; 1402 1403 if (nfs_mapping_need_revalidate_inode(inode)) { 1404 int ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 1405 if (ret < 0) 1406 return ret; 1407 } 1408 1409 return nfs_clear_invalid_mapping(mapping); 1410 } 1411 1412 static bool nfs_file_has_writers(struct nfs_inode *nfsi) 1413 { 1414 struct inode *inode = &nfsi->vfs_inode; 1415 1416 if (!S_ISREG(inode->i_mode)) 1417 return false; 1418 if (list_empty(&nfsi->open_files)) 1419 return false; 1420 return inode_is_open_for_write(inode); 1421 } 1422 1423 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi) 1424 { 1425 return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi); 1426 } 1427 1428 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1429 { 1430 struct timespec64 ts; 1431 1432 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE) 1433 && (fattr->valid & NFS_ATTR_FATTR_CHANGE) 1434 && inode_eq_iversion_raw(inode, fattr->pre_change_attr)) { 1435 inode_set_iversion_raw(inode, fattr->change_attr); 1436 if (S_ISDIR(inode->i_mode)) 1437 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA); 1438 else if (nfs_server_capable(inode, NFS_CAP_XATTR)) 1439 nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR); 1440 } 1441 /* If we have atomic WCC data, we may update some attributes */ 1442 ts = inode->i_ctime; 1443 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME) 1444 && (fattr->valid & NFS_ATTR_FATTR_CTIME) 1445 && timespec64_equal(&ts, &fattr->pre_ctime)) { 1446 inode->i_ctime = fattr->ctime; 1447 } 1448 1449 ts = inode->i_mtime; 1450 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME) 1451 && (fattr->valid & NFS_ATTR_FATTR_MTIME) 1452 && timespec64_equal(&ts, &fattr->pre_mtime)) { 1453 inode->i_mtime = fattr->mtime; 1454 } 1455 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE) 1456 && (fattr->valid & NFS_ATTR_FATTR_SIZE) 1457 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size) 1458 && !nfs_have_writebacks(inode)) { 1459 trace_nfs_size_wcc(inode, fattr->size); 1460 i_size_write(inode, nfs_size_to_loff_t(fattr->size)); 1461 } 1462 } 1463 1464 /** 1465 * nfs_check_inode_attributes - verify consistency of the inode attribute cache 1466 * @inode: pointer to inode 1467 * @fattr: updated attributes 1468 * 1469 * Verifies the attribute cache. If we have just changed the attributes, 1470 * so that fattr carries weak cache consistency data, then it may 1471 * also update the ctime/mtime/change_attribute. 1472 */ 1473 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr) 1474 { 1475 struct nfs_inode *nfsi = NFS_I(inode); 1476 loff_t cur_size, new_isize; 1477 unsigned long invalid = 0; 1478 struct timespec64 ts; 1479 1480 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) 1481 return 0; 1482 1483 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) { 1484 /* Only a mounted-on-fileid? Just exit */ 1485 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) 1486 return 0; 1487 /* Has the inode gone and changed behind our back? */ 1488 } else if (nfsi->fileid != fattr->fileid) { 1489 /* Is this perhaps the mounted-on fileid? */ 1490 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) && 1491 nfsi->fileid == fattr->mounted_on_fileid) 1492 return 0; 1493 return -ESTALE; 1494 } 1495 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) 1496 return -ESTALE; 1497 1498 1499 if (!nfs_file_has_buffered_writers(nfsi)) { 1500 /* Verify a few of the more important attributes */ 1501 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr)) 1502 invalid |= NFS_INO_INVALID_CHANGE; 1503 1504 ts = inode->i_mtime; 1505 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime)) 1506 invalid |= NFS_INO_INVALID_MTIME; 1507 1508 ts = inode->i_ctime; 1509 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime)) 1510 invalid |= NFS_INO_INVALID_CTIME; 1511 1512 if (fattr->valid & NFS_ATTR_FATTR_SIZE) { 1513 cur_size = i_size_read(inode); 1514 new_isize = nfs_size_to_loff_t(fattr->size); 1515 if (cur_size != new_isize) 1516 invalid |= NFS_INO_INVALID_SIZE; 1517 } 1518 } 1519 1520 /* Have any file permissions changed? */ 1521 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) 1522 invalid |= NFS_INO_INVALID_MODE; 1523 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid)) 1524 invalid |= NFS_INO_INVALID_OTHER; 1525 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid)) 1526 invalid |= NFS_INO_INVALID_OTHER; 1527 1528 /* Has the link count changed? */ 1529 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink) 1530 invalid |= NFS_INO_INVALID_NLINK; 1531 1532 ts = inode->i_atime; 1533 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime)) 1534 invalid |= NFS_INO_INVALID_ATIME; 1535 1536 if (invalid != 0) 1537 nfs_set_cache_invalid(inode, invalid); 1538 1539 nfsi->read_cache_jiffies = fattr->time_start; 1540 return 0; 1541 } 1542 1543 static atomic_long_t nfs_attr_generation_counter; 1544 1545 static unsigned long nfs_read_attr_generation_counter(void) 1546 { 1547 return atomic_long_read(&nfs_attr_generation_counter); 1548 } 1549 1550 unsigned long nfs_inc_attr_generation_counter(void) 1551 { 1552 return atomic_long_inc_return(&nfs_attr_generation_counter); 1553 } 1554 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter); 1555 1556 void nfs_fattr_init(struct nfs_fattr *fattr) 1557 { 1558 fattr->valid = 0; 1559 fattr->time_start = jiffies; 1560 fattr->gencount = nfs_inc_attr_generation_counter(); 1561 fattr->owner_name = NULL; 1562 fattr->group_name = NULL; 1563 } 1564 EXPORT_SYMBOL_GPL(nfs_fattr_init); 1565 1566 /** 1567 * nfs_fattr_set_barrier 1568 * @fattr: attributes 1569 * 1570 * Used to set a barrier after an attribute was updated. This 1571 * barrier ensures that older attributes from RPC calls that may 1572 * have raced with our update cannot clobber these new values. 1573 * Note that you are still responsible for ensuring that other 1574 * operations which change the attribute on the server do not 1575 * collide. 1576 */ 1577 void nfs_fattr_set_barrier(struct nfs_fattr *fattr) 1578 { 1579 fattr->gencount = nfs_inc_attr_generation_counter(); 1580 } 1581 1582 struct nfs_fattr *nfs_alloc_fattr(void) 1583 { 1584 struct nfs_fattr *fattr; 1585 1586 fattr = kmalloc(sizeof(*fattr), GFP_NOFS); 1587 if (fattr != NULL) { 1588 nfs_fattr_init(fattr); 1589 fattr->label = NULL; 1590 } 1591 return fattr; 1592 } 1593 EXPORT_SYMBOL_GPL(nfs_alloc_fattr); 1594 1595 struct nfs_fattr *nfs_alloc_fattr_with_label(struct nfs_server *server) 1596 { 1597 struct nfs_fattr *fattr = nfs_alloc_fattr(); 1598 1599 if (!fattr) 1600 return NULL; 1601 1602 fattr->label = nfs4_label_alloc(server, GFP_NOFS); 1603 if (IS_ERR(fattr->label)) { 1604 kfree(fattr); 1605 return NULL; 1606 } 1607 1608 return fattr; 1609 } 1610 EXPORT_SYMBOL_GPL(nfs_alloc_fattr_with_label); 1611 1612 struct nfs_fh *nfs_alloc_fhandle(void) 1613 { 1614 struct nfs_fh *fh; 1615 1616 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS); 1617 if (fh != NULL) 1618 fh->size = 0; 1619 return fh; 1620 } 1621 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle); 1622 1623 #ifdef NFS_DEBUG 1624 /* 1625 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle 1626 * in the same way that wireshark does 1627 * 1628 * @fh: file handle 1629 * 1630 * For debugging only. 1631 */ 1632 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh) 1633 { 1634 /* wireshark uses 32-bit AUTODIN crc and does a bitwise 1635 * not on the result */ 1636 return nfs_fhandle_hash(fh); 1637 } 1638 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash); 1639 1640 /* 1641 * _nfs_display_fhandle - display an NFS file handle on the console 1642 * 1643 * @fh: file handle to display 1644 * @caption: display caption 1645 * 1646 * For debugging only. 1647 */ 1648 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption) 1649 { 1650 unsigned short i; 1651 1652 if (fh == NULL || fh->size == 0) { 1653 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh); 1654 return; 1655 } 1656 1657 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n", 1658 caption, fh, fh->size, _nfs_display_fhandle_hash(fh)); 1659 for (i = 0; i < fh->size; i += 16) { 1660 __be32 *pos = (__be32 *)&fh->data[i]; 1661 1662 switch ((fh->size - i - 1) >> 2) { 1663 case 0: 1664 printk(KERN_DEFAULT " %08x\n", 1665 be32_to_cpup(pos)); 1666 break; 1667 case 1: 1668 printk(KERN_DEFAULT " %08x %08x\n", 1669 be32_to_cpup(pos), be32_to_cpup(pos + 1)); 1670 break; 1671 case 2: 1672 printk(KERN_DEFAULT " %08x %08x %08x\n", 1673 be32_to_cpup(pos), be32_to_cpup(pos + 1), 1674 be32_to_cpup(pos + 2)); 1675 break; 1676 default: 1677 printk(KERN_DEFAULT " %08x %08x %08x %08x\n", 1678 be32_to_cpup(pos), be32_to_cpup(pos + 1), 1679 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3)); 1680 } 1681 } 1682 } 1683 EXPORT_SYMBOL_GPL(_nfs_display_fhandle); 1684 #endif 1685 1686 /** 1687 * nfs_inode_attrs_cmp_generic - compare attributes 1688 * @fattr: attributes 1689 * @inode: pointer to inode 1690 * 1691 * Attempt to divine whether or not an RPC call reply carrying stale 1692 * attributes got scheduled after another call carrying updated ones. 1693 * Note also the check for wraparound of 'attr_gencount' 1694 * 1695 * The function returns '1' if it thinks the attributes in @fattr are 1696 * more recent than the ones cached in @inode. Otherwise it returns 1697 * the value '0'. 1698 */ 1699 static int nfs_inode_attrs_cmp_generic(const struct nfs_fattr *fattr, 1700 const struct inode *inode) 1701 { 1702 unsigned long attr_gencount = NFS_I(inode)->attr_gencount; 1703 1704 return (long)(fattr->gencount - attr_gencount) > 0 || 1705 (long)(attr_gencount - nfs_read_attr_generation_counter()) > 0; 1706 } 1707 1708 /** 1709 * nfs_inode_attrs_cmp_monotonic - compare attributes 1710 * @fattr: attributes 1711 * @inode: pointer to inode 1712 * 1713 * Attempt to divine whether or not an RPC call reply carrying stale 1714 * attributes got scheduled after another call carrying updated ones. 1715 * 1716 * We assume that the server observes monotonic semantics for 1717 * the change attribute, so a larger value means that the attributes in 1718 * @fattr are more recent, in which case the function returns the 1719 * value '1'. 1720 * A return value of '0' indicates no measurable change 1721 * A return value of '-1' means that the attributes in @inode are 1722 * more recent. 1723 */ 1724 static int nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr *fattr, 1725 const struct inode *inode) 1726 { 1727 s64 diff = fattr->change_attr - inode_peek_iversion_raw(inode); 1728 if (diff > 0) 1729 return 1; 1730 return diff == 0 ? 0 : -1; 1731 } 1732 1733 /** 1734 * nfs_inode_attrs_cmp_strict_monotonic - compare attributes 1735 * @fattr: attributes 1736 * @inode: pointer to inode 1737 * 1738 * Attempt to divine whether or not an RPC call reply carrying stale 1739 * attributes got scheduled after another call carrying updated ones. 1740 * 1741 * We assume that the server observes strictly monotonic semantics for 1742 * the change attribute, so a larger value means that the attributes in 1743 * @fattr are more recent, in which case the function returns the 1744 * value '1'. 1745 * A return value of '-1' means that the attributes in @inode are 1746 * more recent or unchanged. 1747 */ 1748 static int nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr *fattr, 1749 const struct inode *inode) 1750 { 1751 return nfs_inode_attrs_cmp_monotonic(fattr, inode) > 0 ? 1 : -1; 1752 } 1753 1754 /** 1755 * nfs_inode_attrs_cmp - compare attributes 1756 * @fattr: attributes 1757 * @inode: pointer to inode 1758 * 1759 * This function returns '1' if it thinks the attributes in @fattr are 1760 * more recent than the ones cached in @inode. It returns '-1' if 1761 * the attributes in @inode are more recent than the ones in @fattr, 1762 * and it returns 0 if not sure. 1763 */ 1764 static int nfs_inode_attrs_cmp(const struct nfs_fattr *fattr, 1765 const struct inode *inode) 1766 { 1767 if (nfs_inode_attrs_cmp_generic(fattr, inode) > 0) 1768 return 1; 1769 switch (NFS_SERVER(inode)->change_attr_type) { 1770 case NFS4_CHANGE_TYPE_IS_UNDEFINED: 1771 break; 1772 case NFS4_CHANGE_TYPE_IS_TIME_METADATA: 1773 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE)) 1774 break; 1775 return nfs_inode_attrs_cmp_monotonic(fattr, inode); 1776 default: 1777 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE)) 1778 break; 1779 return nfs_inode_attrs_cmp_strict_monotonic(fattr, inode); 1780 } 1781 return 0; 1782 } 1783 1784 /** 1785 * nfs_inode_finish_partial_attr_update - complete a previous inode update 1786 * @fattr: attributes 1787 * @inode: pointer to inode 1788 * 1789 * Returns '1' if the last attribute update left the inode cached 1790 * attributes in a partially unrevalidated state, and @fattr 1791 * matches the change attribute of that partial update. 1792 * Otherwise returns '0'. 1793 */ 1794 static int nfs_inode_finish_partial_attr_update(const struct nfs_fattr *fattr, 1795 const struct inode *inode) 1796 { 1797 const unsigned long check_valid = 1798 NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME | 1799 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE | 1800 NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_OTHER | 1801 NFS_INO_INVALID_NLINK; 1802 unsigned long cache_validity = NFS_I(inode)->cache_validity; 1803 enum nfs4_change_attr_type ctype = NFS_SERVER(inode)->change_attr_type; 1804 1805 if (ctype != NFS4_CHANGE_TYPE_IS_UNDEFINED && 1806 !(cache_validity & NFS_INO_INVALID_CHANGE) && 1807 (cache_validity & check_valid) != 0 && 1808 (fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && 1809 nfs_inode_attrs_cmp_monotonic(fattr, inode) == 0) 1810 return 1; 1811 return 0; 1812 } 1813 1814 static int nfs_refresh_inode_locked(struct inode *inode, 1815 struct nfs_fattr *fattr) 1816 { 1817 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode); 1818 int ret = 0; 1819 1820 trace_nfs_refresh_inode_enter(inode); 1821 1822 if (attr_cmp > 0 || nfs_inode_finish_partial_attr_update(fattr, inode)) 1823 ret = nfs_update_inode(inode, fattr); 1824 else if (attr_cmp == 0) 1825 ret = nfs_check_inode_attributes(inode, fattr); 1826 1827 trace_nfs_refresh_inode_exit(inode, ret); 1828 return ret; 1829 } 1830 1831 /** 1832 * nfs_refresh_inode - try to update the inode attribute cache 1833 * @inode: pointer to inode 1834 * @fattr: updated attributes 1835 * 1836 * Check that an RPC call that returned attributes has not overlapped with 1837 * other recent updates of the inode metadata, then decide whether it is 1838 * safe to do a full update of the inode attributes, or whether just to 1839 * call nfs_check_inode_attributes. 1840 */ 1841 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr) 1842 { 1843 int status; 1844 1845 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 1846 return 0; 1847 spin_lock(&inode->i_lock); 1848 status = nfs_refresh_inode_locked(inode, fattr); 1849 spin_unlock(&inode->i_lock); 1850 1851 return status; 1852 } 1853 EXPORT_SYMBOL_GPL(nfs_refresh_inode); 1854 1855 static int nfs_post_op_update_inode_locked(struct inode *inode, 1856 struct nfs_fattr *fattr, unsigned int invalid) 1857 { 1858 if (S_ISDIR(inode->i_mode)) 1859 invalid |= NFS_INO_INVALID_DATA; 1860 nfs_set_cache_invalid(inode, invalid); 1861 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 1862 return 0; 1863 return nfs_refresh_inode_locked(inode, fattr); 1864 } 1865 1866 /** 1867 * nfs_post_op_update_inode - try to update the inode attribute cache 1868 * @inode: pointer to inode 1869 * @fattr: updated attributes 1870 * 1871 * After an operation that has changed the inode metadata, mark the 1872 * attribute cache as being invalid, then try to update it. 1873 * 1874 * NB: if the server didn't return any post op attributes, this 1875 * function will force the retrieval of attributes before the next 1876 * NFS request. Thus it should be used only for operations that 1877 * are expected to change one or more attributes, to avoid 1878 * unnecessary NFS requests and trips through nfs_update_inode(). 1879 */ 1880 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1881 { 1882 int status; 1883 1884 spin_lock(&inode->i_lock); 1885 nfs_fattr_set_barrier(fattr); 1886 status = nfs_post_op_update_inode_locked(inode, fattr, 1887 NFS_INO_INVALID_CHANGE 1888 | NFS_INO_INVALID_CTIME 1889 | NFS_INO_REVAL_FORCED); 1890 spin_unlock(&inode->i_lock); 1891 1892 return status; 1893 } 1894 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode); 1895 1896 /** 1897 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache 1898 * @inode: pointer to inode 1899 * @fattr: updated attributes 1900 * 1901 * After an operation that has changed the inode metadata, mark the 1902 * attribute cache as being invalid, then try to update it. Fake up 1903 * weak cache consistency data, if none exist. 1904 * 1905 * This function is mainly designed to be used by the ->write_done() functions. 1906 */ 1907 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr) 1908 { 1909 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode); 1910 int status; 1911 1912 /* Don't do a WCC update if these attributes are already stale */ 1913 if (attr_cmp < 0) 1914 return 0; 1915 if ((fattr->valid & NFS_ATTR_FATTR) == 0 || !attr_cmp) { 1916 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE 1917 | NFS_ATTR_FATTR_PRESIZE 1918 | NFS_ATTR_FATTR_PREMTIME 1919 | NFS_ATTR_FATTR_PRECTIME); 1920 goto out_noforce; 1921 } 1922 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && 1923 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) { 1924 fattr->pre_change_attr = inode_peek_iversion_raw(inode); 1925 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE; 1926 } 1927 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 && 1928 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) { 1929 fattr->pre_ctime = inode->i_ctime; 1930 fattr->valid |= NFS_ATTR_FATTR_PRECTIME; 1931 } 1932 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 && 1933 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) { 1934 fattr->pre_mtime = inode->i_mtime; 1935 fattr->valid |= NFS_ATTR_FATTR_PREMTIME; 1936 } 1937 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 && 1938 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) { 1939 fattr->pre_size = i_size_read(inode); 1940 fattr->valid |= NFS_ATTR_FATTR_PRESIZE; 1941 } 1942 out_noforce: 1943 status = nfs_post_op_update_inode_locked(inode, fattr, 1944 NFS_INO_INVALID_CHANGE 1945 | NFS_INO_INVALID_CTIME 1946 | NFS_INO_INVALID_MTIME 1947 | NFS_INO_INVALID_BLOCKS); 1948 return status; 1949 } 1950 1951 /** 1952 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache 1953 * @inode: pointer to inode 1954 * @fattr: updated attributes 1955 * 1956 * After an operation that has changed the inode metadata, mark the 1957 * attribute cache as being invalid, then try to update it. Fake up 1958 * weak cache consistency data, if none exist. 1959 * 1960 * This function is mainly designed to be used by the ->write_done() functions. 1961 */ 1962 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr) 1963 { 1964 int status; 1965 1966 spin_lock(&inode->i_lock); 1967 nfs_fattr_set_barrier(fattr); 1968 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 1969 spin_unlock(&inode->i_lock); 1970 return status; 1971 } 1972 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc); 1973 1974 1975 /* 1976 * Many nfs protocol calls return the new file attributes after 1977 * an operation. Here we update the inode to reflect the state 1978 * of the server's inode. 1979 * 1980 * This is a bit tricky because we have to make sure all dirty pages 1981 * have been sent off to the server before calling invalidate_inode_pages. 1982 * To make sure no other process adds more write requests while we try 1983 * our best to flush them, we make them sleep during the attribute refresh. 1984 * 1985 * A very similar scenario holds for the dir cache. 1986 */ 1987 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1988 { 1989 struct nfs_server *server = NFS_SERVER(inode); 1990 struct nfs_inode *nfsi = NFS_I(inode); 1991 loff_t cur_isize, new_isize; 1992 u64 fattr_supported = server->fattr_valid; 1993 unsigned long invalid = 0; 1994 unsigned long now = jiffies; 1995 unsigned long save_cache_validity; 1996 bool have_writers = nfs_file_has_buffered_writers(nfsi); 1997 bool cache_revalidated = true; 1998 bool attr_changed = false; 1999 bool have_delegation; 2000 2001 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n", 2002 __func__, inode->i_sb->s_id, inode->i_ino, 2003 nfs_display_fhandle_hash(NFS_FH(inode)), 2004 atomic_read(&inode->i_count), fattr->valid); 2005 2006 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) { 2007 /* Only a mounted-on-fileid? Just exit */ 2008 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) 2009 return 0; 2010 /* Has the inode gone and changed behind our back? */ 2011 } else if (nfsi->fileid != fattr->fileid) { 2012 /* Is this perhaps the mounted-on fileid? */ 2013 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) && 2014 nfsi->fileid == fattr->mounted_on_fileid) 2015 return 0; 2016 printk(KERN_ERR "NFS: server %s error: fileid changed\n" 2017 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n", 2018 NFS_SERVER(inode)->nfs_client->cl_hostname, 2019 inode->i_sb->s_id, (long long)nfsi->fileid, 2020 (long long)fattr->fileid); 2021 goto out_err; 2022 } 2023 2024 /* 2025 * Make sure the inode's type hasn't changed. 2026 */ 2027 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) { 2028 /* 2029 * Big trouble! The inode has become a different object. 2030 */ 2031 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n", 2032 __func__, inode->i_ino, inode->i_mode, fattr->mode); 2033 goto out_err; 2034 } 2035 2036 /* Update the fsid? */ 2037 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) && 2038 !nfs_fsid_equal(&server->fsid, &fattr->fsid) && 2039 !IS_AUTOMOUNT(inode)) 2040 server->fsid = fattr->fsid; 2041 2042 /* Save the delegation state before clearing cache_validity */ 2043 have_delegation = nfs_have_delegated_attributes(inode); 2044 2045 /* 2046 * Update the read time so we don't revalidate too often. 2047 */ 2048 nfsi->read_cache_jiffies = fattr->time_start; 2049 2050 save_cache_validity = nfsi->cache_validity; 2051 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR 2052 | NFS_INO_INVALID_ATIME 2053 | NFS_INO_REVAL_FORCED 2054 | NFS_INO_INVALID_BLOCKS); 2055 2056 /* Do atomic weak cache consistency updates */ 2057 nfs_wcc_update_inode(inode, fattr); 2058 2059 if (pnfs_layoutcommit_outstanding(inode)) { 2060 nfsi->cache_validity |= 2061 save_cache_validity & 2062 (NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME | 2063 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE | 2064 NFS_INO_INVALID_BLOCKS); 2065 cache_revalidated = false; 2066 } 2067 2068 /* More cache consistency checks */ 2069 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) { 2070 if (!inode_eq_iversion_raw(inode, fattr->change_attr)) { 2071 /* Could it be a race with writeback? */ 2072 if (!(have_writers || have_delegation)) { 2073 invalid |= NFS_INO_INVALID_DATA 2074 | NFS_INO_INVALID_ACCESS 2075 | NFS_INO_INVALID_ACL 2076 | NFS_INO_INVALID_XATTR; 2077 /* Force revalidate of all attributes */ 2078 save_cache_validity |= NFS_INO_INVALID_CTIME 2079 | NFS_INO_INVALID_MTIME 2080 | NFS_INO_INVALID_SIZE 2081 | NFS_INO_INVALID_BLOCKS 2082 | NFS_INO_INVALID_NLINK 2083 | NFS_INO_INVALID_MODE 2084 | NFS_INO_INVALID_OTHER; 2085 if (S_ISDIR(inode->i_mode)) 2086 nfs_force_lookup_revalidate(inode); 2087 attr_changed = true; 2088 dprintk("NFS: change_attr change on server for file %s/%ld\n", 2089 inode->i_sb->s_id, 2090 inode->i_ino); 2091 } else if (!have_delegation) 2092 nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER; 2093 inode_set_iversion_raw(inode, fattr->change_attr); 2094 } 2095 } else { 2096 nfsi->cache_validity |= 2097 save_cache_validity & NFS_INO_INVALID_CHANGE; 2098 if (!have_delegation || 2099 (nfsi->cache_validity & NFS_INO_INVALID_CHANGE) != 0) 2100 cache_revalidated = false; 2101 } 2102 2103 if (fattr->valid & NFS_ATTR_FATTR_MTIME) 2104 inode->i_mtime = fattr->mtime; 2105 else if (fattr_supported & NFS_ATTR_FATTR_MTIME) 2106 nfsi->cache_validity |= 2107 save_cache_validity & NFS_INO_INVALID_MTIME; 2108 2109 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 2110 inode->i_ctime = fattr->ctime; 2111 else if (fattr_supported & NFS_ATTR_FATTR_CTIME) 2112 nfsi->cache_validity |= 2113 save_cache_validity & NFS_INO_INVALID_CTIME; 2114 2115 /* Check if our cached file size is stale */ 2116 if (fattr->valid & NFS_ATTR_FATTR_SIZE) { 2117 new_isize = nfs_size_to_loff_t(fattr->size); 2118 cur_isize = i_size_read(inode); 2119 if (new_isize != cur_isize && !have_delegation) { 2120 /* Do we perhaps have any outstanding writes, or has 2121 * the file grown beyond our last write? */ 2122 if (!nfs_have_writebacks(inode) || new_isize > cur_isize) { 2123 trace_nfs_size_update(inode, new_isize); 2124 i_size_write(inode, new_isize); 2125 if (!have_writers) 2126 invalid |= NFS_INO_INVALID_DATA; 2127 } 2128 } 2129 if (new_isize == 0 && 2130 !(fattr->valid & (NFS_ATTR_FATTR_SPACE_USED | 2131 NFS_ATTR_FATTR_BLOCKS_USED))) { 2132 fattr->du.nfs3.used = 0; 2133 fattr->valid |= NFS_ATTR_FATTR_SPACE_USED; 2134 } 2135 } else 2136 nfsi->cache_validity |= 2137 save_cache_validity & NFS_INO_INVALID_SIZE; 2138 2139 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 2140 inode->i_atime = fattr->atime; 2141 else if (fattr_supported & NFS_ATTR_FATTR_ATIME) 2142 nfsi->cache_validity |= 2143 save_cache_validity & NFS_INO_INVALID_ATIME; 2144 2145 if (fattr->valid & NFS_ATTR_FATTR_MODE) { 2146 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) { 2147 umode_t newmode = inode->i_mode & S_IFMT; 2148 newmode |= fattr->mode & S_IALLUGO; 2149 inode->i_mode = newmode; 2150 invalid |= NFS_INO_INVALID_ACCESS 2151 | NFS_INO_INVALID_ACL; 2152 } 2153 } else if (fattr_supported & NFS_ATTR_FATTR_MODE) 2154 nfsi->cache_validity |= 2155 save_cache_validity & NFS_INO_INVALID_MODE; 2156 2157 if (fattr->valid & NFS_ATTR_FATTR_OWNER) { 2158 if (!uid_eq(inode->i_uid, fattr->uid)) { 2159 invalid |= NFS_INO_INVALID_ACCESS 2160 | NFS_INO_INVALID_ACL; 2161 inode->i_uid = fattr->uid; 2162 } 2163 } else if (fattr_supported & NFS_ATTR_FATTR_OWNER) 2164 nfsi->cache_validity |= 2165 save_cache_validity & NFS_INO_INVALID_OTHER; 2166 2167 if (fattr->valid & NFS_ATTR_FATTR_GROUP) { 2168 if (!gid_eq(inode->i_gid, fattr->gid)) { 2169 invalid |= NFS_INO_INVALID_ACCESS 2170 | NFS_INO_INVALID_ACL; 2171 inode->i_gid = fattr->gid; 2172 } 2173 } else if (fattr_supported & NFS_ATTR_FATTR_GROUP) 2174 nfsi->cache_validity |= 2175 save_cache_validity & NFS_INO_INVALID_OTHER; 2176 2177 if (fattr->valid & NFS_ATTR_FATTR_NLINK) { 2178 if (inode->i_nlink != fattr->nlink) 2179 set_nlink(inode, fattr->nlink); 2180 } else if (fattr_supported & NFS_ATTR_FATTR_NLINK) 2181 nfsi->cache_validity |= 2182 save_cache_validity & NFS_INO_INVALID_NLINK; 2183 2184 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) { 2185 /* 2186 * report the blocks in 512byte units 2187 */ 2188 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 2189 } else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED) 2190 nfsi->cache_validity |= 2191 save_cache_validity & NFS_INO_INVALID_BLOCKS; 2192 2193 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED) 2194 inode->i_blocks = fattr->du.nfs2.blocks; 2195 else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED) 2196 nfsi->cache_validity |= 2197 save_cache_validity & NFS_INO_INVALID_BLOCKS; 2198 2199 /* Update attrtimeo value if we're out of the unstable period */ 2200 if (attr_changed) { 2201 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 2202 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 2203 nfsi->attrtimeo_timestamp = now; 2204 /* Set barrier to be more recent than all outstanding updates */ 2205 nfsi->attr_gencount = nfs_inc_attr_generation_counter(); 2206 } else { 2207 if (cache_revalidated) { 2208 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, 2209 nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) { 2210 nfsi->attrtimeo <<= 1; 2211 if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode)) 2212 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode); 2213 } 2214 nfsi->attrtimeo_timestamp = now; 2215 } 2216 /* Set the barrier to be more recent than this fattr */ 2217 if ((long)(fattr->gencount - nfsi->attr_gencount) > 0) 2218 nfsi->attr_gencount = fattr->gencount; 2219 } 2220 2221 /* Don't invalidate the data if we were to blame */ 2222 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) 2223 || S_ISLNK(inode->i_mode))) 2224 invalid &= ~NFS_INO_INVALID_DATA; 2225 nfs_set_cache_invalid(inode, invalid); 2226 2227 return 0; 2228 out_err: 2229 /* 2230 * No need to worry about unhashing the dentry, as the 2231 * lookup validation will know that the inode is bad. 2232 * (But we fall through to invalidate the caches.) 2233 */ 2234 nfs_set_inode_stale_locked(inode); 2235 return -ESTALE; 2236 } 2237 2238 struct inode *nfs_alloc_inode(struct super_block *sb) 2239 { 2240 struct nfs_inode *nfsi; 2241 nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL); 2242 if (!nfsi) 2243 return NULL; 2244 nfsi->flags = 0UL; 2245 nfsi->cache_validity = 0UL; 2246 #if IS_ENABLED(CONFIG_NFS_V4) 2247 nfsi->nfs4_acl = NULL; 2248 #endif /* CONFIG_NFS_V4 */ 2249 #ifdef CONFIG_NFS_V4_2 2250 nfsi->xattr_cache = NULL; 2251 #endif 2252 return &nfsi->vfs_inode; 2253 } 2254 EXPORT_SYMBOL_GPL(nfs_alloc_inode); 2255 2256 void nfs_free_inode(struct inode *inode) 2257 { 2258 kmem_cache_free(nfs_inode_cachep, NFS_I(inode)); 2259 } 2260 EXPORT_SYMBOL_GPL(nfs_free_inode); 2261 2262 static inline void nfs4_init_once(struct nfs_inode *nfsi) 2263 { 2264 #if IS_ENABLED(CONFIG_NFS_V4) 2265 INIT_LIST_HEAD(&nfsi->open_states); 2266 nfsi->delegation = NULL; 2267 init_rwsem(&nfsi->rwsem); 2268 nfsi->layout = NULL; 2269 #endif 2270 } 2271 2272 static void init_once(void *foo) 2273 { 2274 struct nfs_inode *nfsi = (struct nfs_inode *) foo; 2275 2276 inode_init_once(&nfsi->vfs_inode); 2277 INIT_LIST_HEAD(&nfsi->open_files); 2278 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru); 2279 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru); 2280 nfs4_init_once(nfsi); 2281 } 2282 2283 static int __init nfs_init_inodecache(void) 2284 { 2285 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache", 2286 sizeof(struct nfs_inode), 2287 0, (SLAB_RECLAIM_ACCOUNT| 2288 SLAB_MEM_SPREAD|SLAB_ACCOUNT), 2289 init_once); 2290 if (nfs_inode_cachep == NULL) 2291 return -ENOMEM; 2292 2293 return 0; 2294 } 2295 2296 static void nfs_destroy_inodecache(void) 2297 { 2298 /* 2299 * Make sure all delayed rcu free inodes are flushed before we 2300 * destroy cache. 2301 */ 2302 rcu_barrier(); 2303 kmem_cache_destroy(nfs_inode_cachep); 2304 } 2305 2306 struct workqueue_struct *nfsiod_workqueue; 2307 EXPORT_SYMBOL_GPL(nfsiod_workqueue); 2308 2309 /* 2310 * start up the nfsiod workqueue 2311 */ 2312 static int nfsiod_start(void) 2313 { 2314 struct workqueue_struct *wq; 2315 dprintk("RPC: creating workqueue nfsiod\n"); 2316 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0); 2317 if (wq == NULL) 2318 return -ENOMEM; 2319 nfsiod_workqueue = wq; 2320 return 0; 2321 } 2322 2323 /* 2324 * Destroy the nfsiod workqueue 2325 */ 2326 static void nfsiod_stop(void) 2327 { 2328 struct workqueue_struct *wq; 2329 2330 wq = nfsiod_workqueue; 2331 if (wq == NULL) 2332 return; 2333 nfsiod_workqueue = NULL; 2334 destroy_workqueue(wq); 2335 } 2336 2337 unsigned int nfs_net_id; 2338 EXPORT_SYMBOL_GPL(nfs_net_id); 2339 2340 static int nfs_net_init(struct net *net) 2341 { 2342 nfs_clients_init(net); 2343 return nfs_fs_proc_net_init(net); 2344 } 2345 2346 static void nfs_net_exit(struct net *net) 2347 { 2348 nfs_fs_proc_net_exit(net); 2349 nfs_clients_exit(net); 2350 } 2351 2352 static struct pernet_operations nfs_net_ops = { 2353 .init = nfs_net_init, 2354 .exit = nfs_net_exit, 2355 .id = &nfs_net_id, 2356 .size = sizeof(struct nfs_net), 2357 }; 2358 2359 /* 2360 * Initialize NFS 2361 */ 2362 static int __init init_nfs_fs(void) 2363 { 2364 int err; 2365 2366 err = nfs_sysfs_init(); 2367 if (err < 0) 2368 goto out10; 2369 2370 err = register_pernet_subsys(&nfs_net_ops); 2371 if (err < 0) 2372 goto out9; 2373 2374 err = nfsiod_start(); 2375 if (err) 2376 goto out7; 2377 2378 err = nfs_fs_proc_init(); 2379 if (err) 2380 goto out6; 2381 2382 err = nfs_init_nfspagecache(); 2383 if (err) 2384 goto out5; 2385 2386 err = nfs_init_inodecache(); 2387 if (err) 2388 goto out4; 2389 2390 err = nfs_init_readpagecache(); 2391 if (err) 2392 goto out3; 2393 2394 err = nfs_init_writepagecache(); 2395 if (err) 2396 goto out2; 2397 2398 err = nfs_init_directcache(); 2399 if (err) 2400 goto out1; 2401 2402 rpc_proc_register(&init_net, &nfs_rpcstat); 2403 2404 err = register_nfs_fs(); 2405 if (err) 2406 goto out0; 2407 2408 return 0; 2409 out0: 2410 rpc_proc_unregister(&init_net, "nfs"); 2411 nfs_destroy_directcache(); 2412 out1: 2413 nfs_destroy_writepagecache(); 2414 out2: 2415 nfs_destroy_readpagecache(); 2416 out3: 2417 nfs_destroy_inodecache(); 2418 out4: 2419 nfs_destroy_nfspagecache(); 2420 out5: 2421 nfs_fs_proc_exit(); 2422 out6: 2423 nfsiod_stop(); 2424 out7: 2425 unregister_pernet_subsys(&nfs_net_ops); 2426 out9: 2427 nfs_sysfs_exit(); 2428 out10: 2429 return err; 2430 } 2431 2432 static void __exit exit_nfs_fs(void) 2433 { 2434 nfs_destroy_directcache(); 2435 nfs_destroy_writepagecache(); 2436 nfs_destroy_readpagecache(); 2437 nfs_destroy_inodecache(); 2438 nfs_destroy_nfspagecache(); 2439 unregister_pernet_subsys(&nfs_net_ops); 2440 rpc_proc_unregister(&init_net, "nfs"); 2441 unregister_nfs_fs(); 2442 nfs_fs_proc_exit(); 2443 nfsiod_stop(); 2444 nfs_sysfs_exit(); 2445 } 2446 2447 /* Not quite true; I just maintain it */ 2448 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>"); 2449 MODULE_LICENSE("GPL"); 2450 module_param(enable_ino64, bool, 0644); 2451 2452 module_init(init_nfs_fs) 2453 module_exit(exit_nfs_fs) 2454