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 } 207 208 if (!nfs_has_xattr_cache(nfsi)) 209 flags &= ~NFS_INO_INVALID_XATTR; 210 if (flags & NFS_INO_INVALID_DATA) 211 nfs_fscache_invalidate(inode, 0); 212 flags &= ~NFS_INO_REVAL_FORCED; 213 214 nfsi->cache_validity |= flags; 215 216 if (inode->i_mapping->nrpages == 0) 217 nfsi->cache_validity &= ~(NFS_INO_INVALID_DATA | 218 NFS_INO_DATA_INVAL_DEFER); 219 else if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 220 nfsi->cache_validity &= ~NFS_INO_DATA_INVAL_DEFER; 221 trace_nfs_set_cache_invalid(inode, 0); 222 } 223 EXPORT_SYMBOL_GPL(nfs_set_cache_invalid); 224 225 /* 226 * Invalidate the local caches 227 */ 228 static void nfs_zap_caches_locked(struct inode *inode) 229 { 230 struct nfs_inode *nfsi = NFS_I(inode); 231 int mode = inode->i_mode; 232 233 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 234 235 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 236 nfsi->attrtimeo_timestamp = jiffies; 237 238 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) 239 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR | 240 NFS_INO_INVALID_DATA | 241 NFS_INO_INVALID_ACCESS | 242 NFS_INO_INVALID_ACL | 243 NFS_INO_INVALID_XATTR); 244 else 245 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR | 246 NFS_INO_INVALID_ACCESS | 247 NFS_INO_INVALID_ACL | 248 NFS_INO_INVALID_XATTR); 249 nfs_zap_label_cache_locked(nfsi); 250 } 251 252 void nfs_zap_caches(struct inode *inode) 253 { 254 spin_lock(&inode->i_lock); 255 nfs_zap_caches_locked(inode); 256 spin_unlock(&inode->i_lock); 257 } 258 259 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping) 260 { 261 if (mapping->nrpages != 0) { 262 spin_lock(&inode->i_lock); 263 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA); 264 spin_unlock(&inode->i_lock); 265 } 266 } 267 268 void nfs_zap_acl_cache(struct inode *inode) 269 { 270 void (*clear_acl_cache)(struct inode *); 271 272 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache; 273 if (clear_acl_cache != NULL) 274 clear_acl_cache(inode); 275 spin_lock(&inode->i_lock); 276 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL; 277 spin_unlock(&inode->i_lock); 278 } 279 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache); 280 281 void nfs_invalidate_atime(struct inode *inode) 282 { 283 spin_lock(&inode->i_lock); 284 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME); 285 spin_unlock(&inode->i_lock); 286 } 287 EXPORT_SYMBOL_GPL(nfs_invalidate_atime); 288 289 /* 290 * Invalidate, but do not unhash, the inode. 291 * NB: must be called with inode->i_lock held! 292 */ 293 static void nfs_set_inode_stale_locked(struct inode *inode) 294 { 295 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); 296 nfs_zap_caches_locked(inode); 297 trace_nfs_set_inode_stale(inode); 298 } 299 300 void nfs_set_inode_stale(struct inode *inode) 301 { 302 spin_lock(&inode->i_lock); 303 nfs_set_inode_stale_locked(inode); 304 spin_unlock(&inode->i_lock); 305 } 306 307 struct nfs_find_desc { 308 struct nfs_fh *fh; 309 struct nfs_fattr *fattr; 310 }; 311 312 /* 313 * In NFSv3 we can have 64bit inode numbers. In order to support 314 * this, and re-exported directories (also seen in NFSv2) 315 * we are forced to allow 2 different inodes to have the same 316 * i_ino. 317 */ 318 static int 319 nfs_find_actor(struct inode *inode, void *opaque) 320 { 321 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque; 322 struct nfs_fh *fh = desc->fh; 323 struct nfs_fattr *fattr = desc->fattr; 324 325 if (NFS_FILEID(inode) != fattr->fileid) 326 return 0; 327 if (inode_wrong_type(inode, fattr->mode)) 328 return 0; 329 if (nfs_compare_fh(NFS_FH(inode), fh)) 330 return 0; 331 if (is_bad_inode(inode) || NFS_STALE(inode)) 332 return 0; 333 return 1; 334 } 335 336 static int 337 nfs_init_locked(struct inode *inode, void *opaque) 338 { 339 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque; 340 struct nfs_fattr *fattr = desc->fattr; 341 342 set_nfs_fileid(inode, fattr->fileid); 343 inode->i_mode = fattr->mode; 344 nfs_copy_fh(NFS_FH(inode), desc->fh); 345 return 0; 346 } 347 348 #ifdef CONFIG_NFS_V4_SECURITY_LABEL 349 static void nfs_clear_label_invalid(struct inode *inode) 350 { 351 spin_lock(&inode->i_lock); 352 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL; 353 spin_unlock(&inode->i_lock); 354 } 355 356 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr) 357 { 358 int error; 359 360 if (fattr->label == NULL) 361 return; 362 363 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) { 364 error = security_inode_notifysecctx(inode, fattr->label->label, 365 fattr->label->len); 366 if (error) 367 printk(KERN_ERR "%s() %s %d " 368 "security_inode_notifysecctx() %d\n", 369 __func__, 370 (char *)fattr->label->label, 371 fattr->label->len, error); 372 nfs_clear_label_invalid(inode); 373 } 374 } 375 376 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags) 377 { 378 struct nfs4_label *label; 379 380 if (!(server->caps & NFS_CAP_SECURITY_LABEL)) 381 return NULL; 382 383 label = kzalloc(sizeof(struct nfs4_label), flags); 384 if (label == NULL) 385 return ERR_PTR(-ENOMEM); 386 387 label->label = kzalloc(NFS4_MAXLABELLEN, flags); 388 if (label->label == NULL) { 389 kfree(label); 390 return ERR_PTR(-ENOMEM); 391 } 392 label->len = NFS4_MAXLABELLEN; 393 394 return label; 395 } 396 EXPORT_SYMBOL_GPL(nfs4_label_alloc); 397 #else 398 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr) 399 { 400 } 401 #endif 402 EXPORT_SYMBOL_GPL(nfs_setsecurity); 403 404 /* Search for inode identified by fh, fileid and i_mode in inode cache. */ 405 struct inode * 406 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh) 407 { 408 struct nfs_find_desc desc = { 409 .fh = fh, 410 .fattr = fattr, 411 }; 412 struct inode *inode; 413 unsigned long hash; 414 415 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) || 416 !(fattr->valid & NFS_ATTR_FATTR_TYPE)) 417 return NULL; 418 419 hash = nfs_fattr_to_ino_t(fattr); 420 inode = ilookup5(sb, hash, nfs_find_actor, &desc); 421 422 dprintk("%s: returning %p\n", __func__, inode); 423 return inode; 424 } 425 426 static void nfs_inode_init_regular(struct nfs_inode *nfsi) 427 { 428 atomic_long_set(&nfsi->nrequests, 0); 429 INIT_LIST_HEAD(&nfsi->commit_info.list); 430 atomic_long_set(&nfsi->commit_info.ncommit, 0); 431 atomic_set(&nfsi->commit_info.rpcs_out, 0); 432 mutex_init(&nfsi->commit_mutex); 433 } 434 435 static void nfs_inode_init_dir(struct nfs_inode *nfsi) 436 { 437 nfsi->cache_change_attribute = 0; 438 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf)); 439 init_rwsem(&nfsi->rmdir_sem); 440 } 441 442 /* 443 * This is our front-end to iget that looks up inodes by file handle 444 * instead of inode number. 445 */ 446 struct inode * 447 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr) 448 { 449 struct nfs_find_desc desc = { 450 .fh = fh, 451 .fattr = fattr 452 }; 453 struct inode *inode = ERR_PTR(-ENOENT); 454 u64 fattr_supported = NFS_SB(sb)->fattr_valid; 455 unsigned long hash; 456 457 nfs_attr_check_mountpoint(sb, fattr); 458 459 if (nfs_attr_use_mounted_on_fileid(fattr)) 460 fattr->fileid = fattr->mounted_on_fileid; 461 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0) 462 goto out_no_inode; 463 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0) 464 goto out_no_inode; 465 466 hash = nfs_fattr_to_ino_t(fattr); 467 468 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc); 469 if (inode == NULL) { 470 inode = ERR_PTR(-ENOMEM); 471 goto out_no_inode; 472 } 473 474 if (inode->i_state & I_NEW) { 475 struct nfs_inode *nfsi = NFS_I(inode); 476 unsigned long now = jiffies; 477 478 /* We set i_ino for the few things that still rely on it, 479 * such as stat(2) */ 480 inode->i_ino = hash; 481 482 /* We can't support update_atime(), since the server will reset it */ 483 inode->i_flags |= S_NOATIME|S_NOCMTIME; 484 inode->i_mode = fattr->mode; 485 nfsi->cache_validity = 0; 486 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0 487 && (fattr_supported & NFS_ATTR_FATTR_MODE)) 488 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE); 489 /* Why so? Because we want revalidate for devices/FIFOs, and 490 * that's precisely what we have in nfs_file_inode_operations. 491 */ 492 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops; 493 if (S_ISREG(inode->i_mode)) { 494 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops; 495 inode->i_data.a_ops = &nfs_file_aops; 496 nfs_inode_init_regular(nfsi); 497 } else if (S_ISDIR(inode->i_mode)) { 498 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops; 499 inode->i_fop = &nfs_dir_operations; 500 inode->i_data.a_ops = &nfs_dir_aops; 501 nfs_inode_init_dir(nfsi); 502 /* Deal with crossing mountpoints */ 503 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT || 504 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) { 505 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) 506 inode->i_op = &nfs_referral_inode_operations; 507 else 508 inode->i_op = &nfs_mountpoint_inode_operations; 509 inode->i_fop = NULL; 510 inode->i_flags |= S_AUTOMOUNT; 511 } 512 } else if (S_ISLNK(inode->i_mode)) { 513 inode->i_op = &nfs_symlink_inode_operations; 514 inode_nohighmem(inode); 515 } else 516 init_special_inode(inode, inode->i_mode, fattr->rdev); 517 518 memset(&inode->i_atime, 0, sizeof(inode->i_atime)); 519 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime)); 520 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime)); 521 inode_set_iversion_raw(inode, 0); 522 inode->i_size = 0; 523 clear_nlink(inode); 524 inode->i_uid = make_kuid(&init_user_ns, -2); 525 inode->i_gid = make_kgid(&init_user_ns, -2); 526 inode->i_blocks = 0; 527 nfsi->write_io = 0; 528 nfsi->read_io = 0; 529 530 nfsi->read_cache_jiffies = fattr->time_start; 531 nfsi->attr_gencount = fattr->gencount; 532 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 533 inode->i_atime = fattr->atime; 534 else if (fattr_supported & NFS_ATTR_FATTR_ATIME) 535 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME); 536 if (fattr->valid & NFS_ATTR_FATTR_MTIME) 537 inode->i_mtime = fattr->mtime; 538 else if (fattr_supported & NFS_ATTR_FATTR_MTIME) 539 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME); 540 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 541 inode->i_ctime = fattr->ctime; 542 else if (fattr_supported & NFS_ATTR_FATTR_CTIME) 543 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME); 544 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) 545 inode_set_iversion_raw(inode, fattr->change_attr); 546 else 547 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE); 548 if (fattr->valid & NFS_ATTR_FATTR_SIZE) 549 inode->i_size = nfs_size_to_loff_t(fattr->size); 550 else 551 nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE); 552 if (fattr->valid & NFS_ATTR_FATTR_NLINK) 553 set_nlink(inode, fattr->nlink); 554 else if (fattr_supported & NFS_ATTR_FATTR_NLINK) 555 nfs_set_cache_invalid(inode, NFS_INO_INVALID_NLINK); 556 if (fattr->valid & NFS_ATTR_FATTR_OWNER) 557 inode->i_uid = fattr->uid; 558 else if (fattr_supported & NFS_ATTR_FATTR_OWNER) 559 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER); 560 if (fattr->valid & NFS_ATTR_FATTR_GROUP) 561 inode->i_gid = fattr->gid; 562 else if (fattr_supported & NFS_ATTR_FATTR_GROUP) 563 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER); 564 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED) 565 inode->i_blocks = fattr->du.nfs2.blocks; 566 else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED && 567 fattr->size != 0) 568 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS); 569 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) { 570 /* 571 * report the blocks in 512byte units 572 */ 573 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 574 } else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED && 575 fattr->size != 0) 576 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS); 577 578 nfs_setsecurity(inode, fattr); 579 580 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 581 nfsi->attrtimeo_timestamp = now; 582 nfsi->access_cache = RB_ROOT; 583 584 nfs_fscache_init_inode(inode); 585 586 unlock_new_inode(inode); 587 } else { 588 int err = nfs_refresh_inode(inode, fattr); 589 if (err < 0) { 590 iput(inode); 591 inode = ERR_PTR(err); 592 goto out_no_inode; 593 } 594 } 595 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n", 596 inode->i_sb->s_id, 597 (unsigned long long)NFS_FILEID(inode), 598 nfs_display_fhandle_hash(fh), 599 atomic_read(&inode->i_count)); 600 601 out: 602 return inode; 603 604 out_no_inode: 605 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode)); 606 goto out; 607 } 608 EXPORT_SYMBOL_GPL(nfs_fhget); 609 610 #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) 611 612 int 613 nfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry, 614 struct iattr *attr) 615 { 616 struct inode *inode = d_inode(dentry); 617 struct nfs_fattr *fattr; 618 int error = 0; 619 620 nfs_inc_stats(inode, NFSIOS_VFSSETATTR); 621 622 /* skip mode change if it's just for clearing setuid/setgid */ 623 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID)) 624 attr->ia_valid &= ~ATTR_MODE; 625 626 if (attr->ia_valid & ATTR_SIZE) { 627 BUG_ON(!S_ISREG(inode->i_mode)); 628 629 error = inode_newsize_ok(inode, attr->ia_size); 630 if (error) 631 return error; 632 633 if (attr->ia_size == i_size_read(inode)) 634 attr->ia_valid &= ~ATTR_SIZE; 635 } 636 637 /* Optimization: if the end result is no change, don't RPC */ 638 if (((attr->ia_valid & NFS_VALID_ATTRS) & ~(ATTR_FILE|ATTR_OPEN)) == 0) 639 return 0; 640 641 trace_nfs_setattr_enter(inode); 642 643 /* Write all dirty data */ 644 if (S_ISREG(inode->i_mode)) 645 nfs_sync_inode(inode); 646 647 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode)); 648 if (fattr == NULL) { 649 error = -ENOMEM; 650 goto out; 651 } 652 653 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr); 654 if (error == 0) 655 error = nfs_refresh_inode(inode, fattr); 656 nfs_free_fattr(fattr); 657 out: 658 trace_nfs_setattr_exit(inode, error); 659 return error; 660 } 661 EXPORT_SYMBOL_GPL(nfs_setattr); 662 663 /** 664 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall 665 * @inode: inode of the file used 666 * @offset: file offset to start truncating 667 * 668 * This is a copy of the common vmtruncate, but with the locking 669 * corrected to take into account the fact that NFS requires 670 * inode->i_size to be updated under the inode->i_lock. 671 * Note: must be called with inode->i_lock held! 672 */ 673 static int nfs_vmtruncate(struct inode * inode, loff_t offset) 674 { 675 int err; 676 677 err = inode_newsize_ok(inode, offset); 678 if (err) 679 goto out; 680 681 trace_nfs_size_truncate(inode, offset); 682 i_size_write(inode, offset); 683 /* Optimisation */ 684 if (offset == 0) 685 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_DATA | 686 NFS_INO_DATA_INVAL_DEFER); 687 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE; 688 689 spin_unlock(&inode->i_lock); 690 truncate_pagecache(inode, offset); 691 spin_lock(&inode->i_lock); 692 out: 693 return err; 694 } 695 696 /** 697 * nfs_setattr_update_inode - Update inode metadata after a setattr call. 698 * @inode: pointer to struct inode 699 * @attr: pointer to struct iattr 700 * @fattr: pointer to struct nfs_fattr 701 * 702 * Note: we do this in the *proc.c in order to ensure that 703 * it works for things like exclusive creates too. 704 */ 705 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr, 706 struct nfs_fattr *fattr) 707 { 708 /* Barrier: bump the attribute generation count. */ 709 nfs_fattr_set_barrier(fattr); 710 711 spin_lock(&inode->i_lock); 712 NFS_I(inode)->attr_gencount = fattr->gencount; 713 if ((attr->ia_valid & ATTR_SIZE) != 0) { 714 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME | 715 NFS_INO_INVALID_BLOCKS); 716 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC); 717 nfs_vmtruncate(inode, attr->ia_size); 718 } 719 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) { 720 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME; 721 if ((attr->ia_valid & ATTR_KILL_SUID) != 0 && 722 inode->i_mode & S_ISUID) 723 inode->i_mode &= ~S_ISUID; 724 if ((attr->ia_valid & ATTR_KILL_SGID) != 0 && 725 (inode->i_mode & (S_ISGID | S_IXGRP)) == 726 (S_ISGID | S_IXGRP)) 727 inode->i_mode &= ~S_ISGID; 728 if ((attr->ia_valid & ATTR_MODE) != 0) { 729 int mode = attr->ia_mode & S_IALLUGO; 730 mode |= inode->i_mode & ~S_IALLUGO; 731 inode->i_mode = mode; 732 } 733 if ((attr->ia_valid & ATTR_UID) != 0) 734 inode->i_uid = attr->ia_uid; 735 if ((attr->ia_valid & ATTR_GID) != 0) 736 inode->i_gid = attr->ia_gid; 737 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 738 inode->i_ctime = fattr->ctime; 739 else 740 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE 741 | NFS_INO_INVALID_CTIME); 742 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS 743 | NFS_INO_INVALID_ACL); 744 } 745 if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) { 746 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME 747 | NFS_INO_INVALID_CTIME); 748 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 749 inode->i_atime = fattr->atime; 750 else if (attr->ia_valid & ATTR_ATIME_SET) 751 inode->i_atime = attr->ia_atime; 752 else 753 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME); 754 755 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 756 inode->i_ctime = fattr->ctime; 757 else 758 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE 759 | NFS_INO_INVALID_CTIME); 760 } 761 if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) { 762 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME 763 | NFS_INO_INVALID_CTIME); 764 if (fattr->valid & NFS_ATTR_FATTR_MTIME) 765 inode->i_mtime = fattr->mtime; 766 else if (attr->ia_valid & ATTR_MTIME_SET) 767 inode->i_mtime = attr->ia_mtime; 768 else 769 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME); 770 771 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 772 inode->i_ctime = fattr->ctime; 773 else 774 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE 775 | NFS_INO_INVALID_CTIME); 776 } 777 if (fattr->valid) 778 nfs_update_inode(inode, fattr); 779 spin_unlock(&inode->i_lock); 780 } 781 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode); 782 783 /* 784 * Don't request help from readdirplus if the file is being written to, 785 * or if attribute caching is turned off 786 */ 787 static bool nfs_getattr_readdirplus_enable(const struct inode *inode) 788 { 789 return nfs_server_capable(inode, NFS_CAP_READDIRPLUS) && 790 !nfs_have_writebacks(inode) && NFS_MAXATTRTIMEO(inode) > 5 * HZ; 791 } 792 793 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry) 794 { 795 if (!IS_ROOT(dentry)) { 796 struct dentry *parent = dget_parent(dentry); 797 nfs_readdir_record_entry_cache_miss(d_inode(parent)); 798 dput(parent); 799 } 800 } 801 802 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry) 803 { 804 if (!IS_ROOT(dentry)) { 805 struct dentry *parent = dget_parent(dentry); 806 nfs_readdir_record_entry_cache_hit(d_inode(parent)); 807 dput(parent); 808 } 809 } 810 811 static u32 nfs_get_valid_attrmask(struct inode *inode) 812 { 813 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity); 814 u32 reply_mask = STATX_INO | STATX_TYPE; 815 816 if (!(cache_validity & NFS_INO_INVALID_ATIME)) 817 reply_mask |= STATX_ATIME; 818 if (!(cache_validity & NFS_INO_INVALID_CTIME)) 819 reply_mask |= STATX_CTIME; 820 if (!(cache_validity & NFS_INO_INVALID_MTIME)) 821 reply_mask |= STATX_MTIME; 822 if (!(cache_validity & NFS_INO_INVALID_SIZE)) 823 reply_mask |= STATX_SIZE; 824 if (!(cache_validity & NFS_INO_INVALID_NLINK)) 825 reply_mask |= STATX_NLINK; 826 if (!(cache_validity & NFS_INO_INVALID_MODE)) 827 reply_mask |= STATX_MODE; 828 if (!(cache_validity & NFS_INO_INVALID_OTHER)) 829 reply_mask |= STATX_UID | STATX_GID; 830 if (!(cache_validity & NFS_INO_INVALID_BLOCKS)) 831 reply_mask |= STATX_BLOCKS; 832 return reply_mask; 833 } 834 835 int nfs_getattr(struct user_namespace *mnt_userns, const struct path *path, 836 struct kstat *stat, u32 request_mask, unsigned int query_flags) 837 { 838 struct inode *inode = d_inode(path->dentry); 839 struct nfs_server *server = NFS_SERVER(inode); 840 unsigned long cache_validity; 841 int err = 0; 842 bool force_sync = query_flags & AT_STATX_FORCE_SYNC; 843 bool do_update = false; 844 bool readdirplus_enabled = nfs_getattr_readdirplus_enable(inode); 845 846 trace_nfs_getattr_enter(inode); 847 848 request_mask &= STATX_TYPE | STATX_MODE | STATX_NLINK | STATX_UID | 849 STATX_GID | STATX_ATIME | STATX_MTIME | STATX_CTIME | 850 STATX_INO | STATX_SIZE | STATX_BLOCKS; 851 852 if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) { 853 if (readdirplus_enabled) 854 nfs_readdirplus_parent_cache_hit(path->dentry); 855 goto out_no_revalidate; 856 } 857 858 /* Flush out writes to the server in order to update c/mtime. */ 859 if ((request_mask & (STATX_CTIME | STATX_MTIME)) && 860 S_ISREG(inode->i_mode)) 861 filemap_write_and_wait(inode->i_mapping); 862 863 /* 864 * We may force a getattr if the user cares about atime. 865 * 866 * Note that we only have to check the vfsmount flags here: 867 * - NFS always sets S_NOATIME by so checking it would give a 868 * bogus result 869 * - NFS never sets SB_NOATIME or SB_NODIRATIME so there is 870 * no point in checking those. 871 */ 872 if ((path->mnt->mnt_flags & MNT_NOATIME) || 873 ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) 874 request_mask &= ~STATX_ATIME; 875 876 /* Is the user requesting attributes that might need revalidation? */ 877 if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME| 878 STATX_MTIME|STATX_UID|STATX_GID| 879 STATX_SIZE|STATX_BLOCKS))) 880 goto out_no_revalidate; 881 882 /* Check whether the cached attributes are stale */ 883 do_update |= force_sync || nfs_attribute_cache_expired(inode); 884 cache_validity = READ_ONCE(NFS_I(inode)->cache_validity); 885 do_update |= cache_validity & NFS_INO_INVALID_CHANGE; 886 if (request_mask & STATX_ATIME) 887 do_update |= cache_validity & NFS_INO_INVALID_ATIME; 888 if (request_mask & STATX_CTIME) 889 do_update |= cache_validity & NFS_INO_INVALID_CTIME; 890 if (request_mask & STATX_MTIME) 891 do_update |= cache_validity & NFS_INO_INVALID_MTIME; 892 if (request_mask & STATX_SIZE) 893 do_update |= cache_validity & NFS_INO_INVALID_SIZE; 894 if (request_mask & STATX_NLINK) 895 do_update |= cache_validity & NFS_INO_INVALID_NLINK; 896 if (request_mask & STATX_MODE) 897 do_update |= cache_validity & NFS_INO_INVALID_MODE; 898 if (request_mask & (STATX_UID | STATX_GID)) 899 do_update |= cache_validity & NFS_INO_INVALID_OTHER; 900 if (request_mask & STATX_BLOCKS) 901 do_update |= cache_validity & NFS_INO_INVALID_BLOCKS; 902 903 if (do_update) { 904 if (readdirplus_enabled) 905 nfs_readdirplus_parent_cache_miss(path->dentry); 906 err = __nfs_revalidate_inode(server, inode); 907 if (err) 908 goto out; 909 } else if (readdirplus_enabled) 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_ACCOUNT); 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_ACCOUNT); 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 1184 /* 1185 * This function is called whenever some part of NFS notices that 1186 * the cached attributes have to be refreshed. 1187 */ 1188 int 1189 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) 1190 { 1191 int status = -ESTALE; 1192 struct nfs_fattr *fattr = NULL; 1193 struct nfs_inode *nfsi = NFS_I(inode); 1194 1195 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n", 1196 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode)); 1197 1198 trace_nfs_revalidate_inode_enter(inode); 1199 1200 if (is_bad_inode(inode)) 1201 goto out; 1202 if (NFS_STALE(inode)) 1203 goto out; 1204 1205 /* pNFS: Attributes aren't updated until we layoutcommit */ 1206 if (S_ISREG(inode->i_mode)) { 1207 status = pnfs_sync_inode(inode, false); 1208 if (status) 1209 goto out; 1210 } 1211 1212 status = -ENOMEM; 1213 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode)); 1214 if (fattr == NULL) 1215 goto out; 1216 1217 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE); 1218 1219 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, inode); 1220 if (status != 0) { 1221 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n", 1222 inode->i_sb->s_id, 1223 (unsigned long long)NFS_FILEID(inode), status); 1224 switch (status) { 1225 case -ETIMEDOUT: 1226 /* A soft timeout occurred. Use cached information? */ 1227 if (server->flags & NFS_MOUNT_SOFTREVAL) 1228 status = 0; 1229 break; 1230 case -ESTALE: 1231 if (!S_ISDIR(inode->i_mode)) 1232 nfs_set_inode_stale(inode); 1233 else 1234 nfs_zap_caches(inode); 1235 } 1236 goto out; 1237 } 1238 1239 status = nfs_refresh_inode(inode, fattr); 1240 if (status) { 1241 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n", 1242 inode->i_sb->s_id, 1243 (unsigned long long)NFS_FILEID(inode), status); 1244 goto out; 1245 } 1246 1247 if (nfsi->cache_validity & NFS_INO_INVALID_ACL) 1248 nfs_zap_acl_cache(inode); 1249 1250 nfs_setsecurity(inode, fattr); 1251 1252 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n", 1253 inode->i_sb->s_id, 1254 (unsigned long long)NFS_FILEID(inode)); 1255 1256 out: 1257 nfs_free_fattr(fattr); 1258 trace_nfs_revalidate_inode_exit(inode, status); 1259 return status; 1260 } 1261 1262 int nfs_attribute_cache_expired(struct inode *inode) 1263 { 1264 if (nfs_have_delegated_attributes(inode)) 1265 return 0; 1266 return nfs_attribute_timeout(inode); 1267 } 1268 1269 /** 1270 * nfs_revalidate_inode - Revalidate the inode attributes 1271 * @inode: pointer to inode struct 1272 * @flags: cache flags to check 1273 * 1274 * Updates inode attribute information by retrieving the data from the server. 1275 */ 1276 int nfs_revalidate_inode(struct inode *inode, unsigned long flags) 1277 { 1278 if (!nfs_check_cache_invalid(inode, flags)) 1279 return NFS_STALE(inode) ? -ESTALE : 0; 1280 return __nfs_revalidate_inode(NFS_SERVER(inode), inode); 1281 } 1282 EXPORT_SYMBOL_GPL(nfs_revalidate_inode); 1283 1284 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping) 1285 { 1286 int ret; 1287 1288 nfs_fscache_invalidate(inode, 0); 1289 if (mapping->nrpages != 0) { 1290 if (S_ISREG(inode->i_mode)) { 1291 ret = nfs_sync_mapping(mapping); 1292 if (ret < 0) 1293 return ret; 1294 } 1295 ret = invalidate_inode_pages2(mapping); 1296 if (ret < 0) 1297 return ret; 1298 } 1299 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE); 1300 1301 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n", 1302 inode->i_sb->s_id, 1303 (unsigned long long)NFS_FILEID(inode)); 1304 return 0; 1305 } 1306 1307 /** 1308 * nfs_clear_invalid_mapping - Conditionally clear a mapping 1309 * @mapping: pointer to mapping 1310 * 1311 * If the NFS_INO_INVALID_DATA inode flag is set, clear the mapping. 1312 */ 1313 int nfs_clear_invalid_mapping(struct address_space *mapping) 1314 { 1315 struct inode *inode = mapping->host; 1316 struct nfs_inode *nfsi = NFS_I(inode); 1317 unsigned long *bitlock = &nfsi->flags; 1318 int ret = 0; 1319 1320 /* 1321 * We must clear NFS_INO_INVALID_DATA first to ensure that 1322 * invalidations that come in while we're shooting down the mappings 1323 * are respected. But, that leaves a race window where one revalidator 1324 * can clear the flag, and then another checks it before the mapping 1325 * gets invalidated. Fix that by serializing access to this part of 1326 * the function. 1327 * 1328 * At the same time, we need to allow other tasks to see whether we 1329 * might be in the middle of invalidating the pages, so we only set 1330 * the bit lock here if it looks like we're going to be doing that. 1331 */ 1332 for (;;) { 1333 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING, 1334 nfs_wait_bit_killable, TASK_KILLABLE); 1335 if (ret) 1336 goto out; 1337 spin_lock(&inode->i_lock); 1338 if (test_bit(NFS_INO_INVALIDATING, bitlock)) { 1339 spin_unlock(&inode->i_lock); 1340 continue; 1341 } 1342 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 1343 break; 1344 spin_unlock(&inode->i_lock); 1345 goto out; 1346 } 1347 1348 set_bit(NFS_INO_INVALIDATING, bitlock); 1349 smp_wmb(); 1350 nfsi->cache_validity &= 1351 ~(NFS_INO_INVALID_DATA | NFS_INO_DATA_INVAL_DEFER); 1352 spin_unlock(&inode->i_lock); 1353 trace_nfs_invalidate_mapping_enter(inode); 1354 ret = nfs_invalidate_mapping(inode, mapping); 1355 trace_nfs_invalidate_mapping_exit(inode, ret); 1356 1357 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock); 1358 smp_mb__after_atomic(); 1359 wake_up_bit(bitlock, NFS_INO_INVALIDATING); 1360 out: 1361 return ret; 1362 } 1363 1364 bool nfs_mapping_need_revalidate_inode(struct inode *inode) 1365 { 1366 return nfs_check_cache_invalid(inode, NFS_INO_INVALID_CHANGE) || 1367 NFS_STALE(inode); 1368 } 1369 1370 int nfs_revalidate_mapping_rcu(struct inode *inode) 1371 { 1372 struct nfs_inode *nfsi = NFS_I(inode); 1373 unsigned long *bitlock = &nfsi->flags; 1374 int ret = 0; 1375 1376 if (IS_SWAPFILE(inode)) 1377 goto out; 1378 if (nfs_mapping_need_revalidate_inode(inode)) { 1379 ret = -ECHILD; 1380 goto out; 1381 } 1382 spin_lock(&inode->i_lock); 1383 if (test_bit(NFS_INO_INVALIDATING, bitlock) || 1384 (nfsi->cache_validity & NFS_INO_INVALID_DATA)) 1385 ret = -ECHILD; 1386 spin_unlock(&inode->i_lock); 1387 out: 1388 return ret; 1389 } 1390 1391 /** 1392 * nfs_revalidate_mapping - Revalidate the pagecache 1393 * @inode: pointer to host inode 1394 * @mapping: pointer to mapping 1395 */ 1396 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping) 1397 { 1398 /* swapfiles are not supposed to be shared. */ 1399 if (IS_SWAPFILE(inode)) 1400 return 0; 1401 1402 if (nfs_mapping_need_revalidate_inode(inode)) { 1403 int ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 1404 if (ret < 0) 1405 return ret; 1406 } 1407 1408 return nfs_clear_invalid_mapping(mapping); 1409 } 1410 1411 static bool nfs_file_has_writers(struct nfs_inode *nfsi) 1412 { 1413 struct inode *inode = &nfsi->vfs_inode; 1414 1415 if (!S_ISREG(inode->i_mode)) 1416 return false; 1417 if (list_empty(&nfsi->open_files)) 1418 return false; 1419 return inode_is_open_for_write(inode); 1420 } 1421 1422 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi) 1423 { 1424 return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi); 1425 } 1426 1427 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1428 { 1429 struct timespec64 ts; 1430 1431 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE) 1432 && (fattr->valid & NFS_ATTR_FATTR_CHANGE) 1433 && inode_eq_iversion_raw(inode, fattr->pre_change_attr)) { 1434 inode_set_iversion_raw(inode, fattr->change_attr); 1435 if (S_ISDIR(inode->i_mode)) 1436 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA); 1437 else if (nfs_server_capable(inode, NFS_CAP_XATTR)) 1438 nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR); 1439 } 1440 /* If we have atomic WCC data, we may update some attributes */ 1441 ts = inode->i_ctime; 1442 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME) 1443 && (fattr->valid & NFS_ATTR_FATTR_CTIME) 1444 && timespec64_equal(&ts, &fattr->pre_ctime)) { 1445 inode->i_ctime = fattr->ctime; 1446 } 1447 1448 ts = inode->i_mtime; 1449 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME) 1450 && (fattr->valid & NFS_ATTR_FATTR_MTIME) 1451 && timespec64_equal(&ts, &fattr->pre_mtime)) { 1452 inode->i_mtime = fattr->mtime; 1453 } 1454 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE) 1455 && (fattr->valid & NFS_ATTR_FATTR_SIZE) 1456 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size) 1457 && !nfs_have_writebacks(inode)) { 1458 trace_nfs_size_wcc(inode, fattr->size); 1459 i_size_write(inode, nfs_size_to_loff_t(fattr->size)); 1460 } 1461 } 1462 1463 /** 1464 * nfs_check_inode_attributes - verify consistency of the inode attribute cache 1465 * @inode: pointer to inode 1466 * @fattr: updated attributes 1467 * 1468 * Verifies the attribute cache. If we have just changed the attributes, 1469 * so that fattr carries weak cache consistency data, then it may 1470 * also update the ctime/mtime/change_attribute. 1471 */ 1472 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr) 1473 { 1474 struct nfs_inode *nfsi = NFS_I(inode); 1475 loff_t cur_size, new_isize; 1476 unsigned long invalid = 0; 1477 struct timespec64 ts; 1478 1479 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) 1480 return 0; 1481 1482 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) { 1483 /* Only a mounted-on-fileid? Just exit */ 1484 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) 1485 return 0; 1486 /* Has the inode gone and changed behind our back? */ 1487 } else if (nfsi->fileid != fattr->fileid) { 1488 /* Is this perhaps the mounted-on fileid? */ 1489 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) && 1490 nfsi->fileid == fattr->mounted_on_fileid) 1491 return 0; 1492 return -ESTALE; 1493 } 1494 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) 1495 return -ESTALE; 1496 1497 1498 if (!nfs_file_has_buffered_writers(nfsi)) { 1499 /* Verify a few of the more important attributes */ 1500 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr)) 1501 invalid |= NFS_INO_INVALID_CHANGE; 1502 1503 ts = inode->i_mtime; 1504 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime)) 1505 invalid |= NFS_INO_INVALID_MTIME; 1506 1507 ts = inode->i_ctime; 1508 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime)) 1509 invalid |= NFS_INO_INVALID_CTIME; 1510 1511 if (fattr->valid & NFS_ATTR_FATTR_SIZE) { 1512 cur_size = i_size_read(inode); 1513 new_isize = nfs_size_to_loff_t(fattr->size); 1514 if (cur_size != new_isize) 1515 invalid |= NFS_INO_INVALID_SIZE; 1516 } 1517 } 1518 1519 /* Have any file permissions changed? */ 1520 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) 1521 invalid |= NFS_INO_INVALID_MODE; 1522 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid)) 1523 invalid |= NFS_INO_INVALID_OTHER; 1524 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid)) 1525 invalid |= NFS_INO_INVALID_OTHER; 1526 1527 /* Has the link count changed? */ 1528 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink) 1529 invalid |= NFS_INO_INVALID_NLINK; 1530 1531 ts = inode->i_atime; 1532 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime)) 1533 invalid |= NFS_INO_INVALID_ATIME; 1534 1535 if (invalid != 0) 1536 nfs_set_cache_invalid(inode, invalid); 1537 1538 nfsi->read_cache_jiffies = fattr->time_start; 1539 return 0; 1540 } 1541 1542 static atomic_long_t nfs_attr_generation_counter; 1543 1544 static unsigned long nfs_read_attr_generation_counter(void) 1545 { 1546 return atomic_long_read(&nfs_attr_generation_counter); 1547 } 1548 1549 unsigned long nfs_inc_attr_generation_counter(void) 1550 { 1551 return atomic_long_inc_return(&nfs_attr_generation_counter); 1552 } 1553 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter); 1554 1555 void nfs_fattr_init(struct nfs_fattr *fattr) 1556 { 1557 fattr->valid = 0; 1558 fattr->time_start = jiffies; 1559 fattr->gencount = nfs_inc_attr_generation_counter(); 1560 fattr->owner_name = NULL; 1561 fattr->group_name = NULL; 1562 } 1563 EXPORT_SYMBOL_GPL(nfs_fattr_init); 1564 1565 /** 1566 * nfs_fattr_set_barrier 1567 * @fattr: attributes 1568 * 1569 * Used to set a barrier after an attribute was updated. This 1570 * barrier ensures that older attributes from RPC calls that may 1571 * have raced with our update cannot clobber these new values. 1572 * Note that you are still responsible for ensuring that other 1573 * operations which change the attribute on the server do not 1574 * collide. 1575 */ 1576 void nfs_fattr_set_barrier(struct nfs_fattr *fattr) 1577 { 1578 fattr->gencount = nfs_inc_attr_generation_counter(); 1579 } 1580 1581 struct nfs_fattr *nfs_alloc_fattr(void) 1582 { 1583 struct nfs_fattr *fattr; 1584 1585 fattr = kmalloc(sizeof(*fattr), GFP_KERNEL); 1586 if (fattr != NULL) { 1587 nfs_fattr_init(fattr); 1588 fattr->label = NULL; 1589 } 1590 return fattr; 1591 } 1592 EXPORT_SYMBOL_GPL(nfs_alloc_fattr); 1593 1594 struct nfs_fattr *nfs_alloc_fattr_with_label(struct nfs_server *server) 1595 { 1596 struct nfs_fattr *fattr = nfs_alloc_fattr(); 1597 1598 if (!fattr) 1599 return NULL; 1600 1601 fattr->label = nfs4_label_alloc(server, GFP_KERNEL); 1602 if (IS_ERR(fattr->label)) { 1603 kfree(fattr); 1604 return NULL; 1605 } 1606 1607 return fattr; 1608 } 1609 EXPORT_SYMBOL_GPL(nfs_alloc_fattr_with_label); 1610 1611 struct nfs_fh *nfs_alloc_fhandle(void) 1612 { 1613 struct nfs_fh *fh; 1614 1615 fh = kmalloc(sizeof(struct nfs_fh), GFP_KERNEL); 1616 if (fh != NULL) 1617 fh->size = 0; 1618 return fh; 1619 } 1620 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle); 1621 1622 #ifdef NFS_DEBUG 1623 /* 1624 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle 1625 * in the same way that wireshark does 1626 * 1627 * @fh: file handle 1628 * 1629 * For debugging only. 1630 */ 1631 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh) 1632 { 1633 /* wireshark uses 32-bit AUTODIN crc and does a bitwise 1634 * not on the result */ 1635 return nfs_fhandle_hash(fh); 1636 } 1637 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash); 1638 1639 /* 1640 * _nfs_display_fhandle - display an NFS file handle on the console 1641 * 1642 * @fh: file handle to display 1643 * @caption: display caption 1644 * 1645 * For debugging only. 1646 */ 1647 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption) 1648 { 1649 unsigned short i; 1650 1651 if (fh == NULL || fh->size == 0) { 1652 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh); 1653 return; 1654 } 1655 1656 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n", 1657 caption, fh, fh->size, _nfs_display_fhandle_hash(fh)); 1658 for (i = 0; i < fh->size; i += 16) { 1659 __be32 *pos = (__be32 *)&fh->data[i]; 1660 1661 switch ((fh->size - i - 1) >> 2) { 1662 case 0: 1663 printk(KERN_DEFAULT " %08x\n", 1664 be32_to_cpup(pos)); 1665 break; 1666 case 1: 1667 printk(KERN_DEFAULT " %08x %08x\n", 1668 be32_to_cpup(pos), be32_to_cpup(pos + 1)); 1669 break; 1670 case 2: 1671 printk(KERN_DEFAULT " %08x %08x %08x\n", 1672 be32_to_cpup(pos), be32_to_cpup(pos + 1), 1673 be32_to_cpup(pos + 2)); 1674 break; 1675 default: 1676 printk(KERN_DEFAULT " %08x %08x %08x %08x\n", 1677 be32_to_cpup(pos), be32_to_cpup(pos + 1), 1678 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3)); 1679 } 1680 } 1681 } 1682 EXPORT_SYMBOL_GPL(_nfs_display_fhandle); 1683 #endif 1684 1685 /** 1686 * nfs_inode_attrs_cmp_generic - compare attributes 1687 * @fattr: attributes 1688 * @inode: pointer to inode 1689 * 1690 * Attempt to divine whether or not an RPC call reply carrying stale 1691 * attributes got scheduled after another call carrying updated ones. 1692 * Note also the check for wraparound of 'attr_gencount' 1693 * 1694 * The function returns '1' if it thinks the attributes in @fattr are 1695 * more recent than the ones cached in @inode. Otherwise it returns 1696 * the value '0'. 1697 */ 1698 static int nfs_inode_attrs_cmp_generic(const struct nfs_fattr *fattr, 1699 const struct inode *inode) 1700 { 1701 unsigned long attr_gencount = NFS_I(inode)->attr_gencount; 1702 1703 return (long)(fattr->gencount - attr_gencount) > 0 || 1704 (long)(attr_gencount - nfs_read_attr_generation_counter()) > 0; 1705 } 1706 1707 /** 1708 * nfs_inode_attrs_cmp_monotonic - compare attributes 1709 * @fattr: attributes 1710 * @inode: pointer to inode 1711 * 1712 * Attempt to divine whether or not an RPC call reply carrying stale 1713 * attributes got scheduled after another call carrying updated ones. 1714 * 1715 * We assume that the server observes monotonic semantics for 1716 * the change attribute, so a larger value means that the attributes in 1717 * @fattr are more recent, in which case the function returns the 1718 * value '1'. 1719 * A return value of '0' indicates no measurable change 1720 * A return value of '-1' means that the attributes in @inode are 1721 * more recent. 1722 */ 1723 static int nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr *fattr, 1724 const struct inode *inode) 1725 { 1726 s64 diff = fattr->change_attr - inode_peek_iversion_raw(inode); 1727 if (diff > 0) 1728 return 1; 1729 return diff == 0 ? 0 : -1; 1730 } 1731 1732 /** 1733 * nfs_inode_attrs_cmp_strict_monotonic - compare attributes 1734 * @fattr: attributes 1735 * @inode: pointer to inode 1736 * 1737 * Attempt to divine whether or not an RPC call reply carrying stale 1738 * attributes got scheduled after another call carrying updated ones. 1739 * 1740 * We assume that the server observes strictly monotonic semantics for 1741 * the change attribute, so a larger value means that the attributes in 1742 * @fattr are more recent, in which case the function returns the 1743 * value '1'. 1744 * A return value of '-1' means that the attributes in @inode are 1745 * more recent or unchanged. 1746 */ 1747 static int nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr *fattr, 1748 const struct inode *inode) 1749 { 1750 return nfs_inode_attrs_cmp_monotonic(fattr, inode) > 0 ? 1 : -1; 1751 } 1752 1753 /** 1754 * nfs_inode_attrs_cmp - compare attributes 1755 * @fattr: attributes 1756 * @inode: pointer to inode 1757 * 1758 * This function returns '1' if it thinks the attributes in @fattr are 1759 * more recent than the ones cached in @inode. It returns '-1' if 1760 * the attributes in @inode are more recent than the ones in @fattr, 1761 * and it returns 0 if not sure. 1762 */ 1763 static int nfs_inode_attrs_cmp(const struct nfs_fattr *fattr, 1764 const struct inode *inode) 1765 { 1766 if (nfs_inode_attrs_cmp_generic(fattr, inode) > 0) 1767 return 1; 1768 switch (NFS_SERVER(inode)->change_attr_type) { 1769 case NFS4_CHANGE_TYPE_IS_UNDEFINED: 1770 break; 1771 case NFS4_CHANGE_TYPE_IS_TIME_METADATA: 1772 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE)) 1773 break; 1774 return nfs_inode_attrs_cmp_monotonic(fattr, inode); 1775 default: 1776 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE)) 1777 break; 1778 return nfs_inode_attrs_cmp_strict_monotonic(fattr, inode); 1779 } 1780 return 0; 1781 } 1782 1783 /** 1784 * nfs_inode_finish_partial_attr_update - complete a previous inode update 1785 * @fattr: attributes 1786 * @inode: pointer to inode 1787 * 1788 * Returns '1' if the last attribute update left the inode cached 1789 * attributes in a partially unrevalidated state, and @fattr 1790 * matches the change attribute of that partial update. 1791 * Otherwise returns '0'. 1792 */ 1793 static int nfs_inode_finish_partial_attr_update(const struct nfs_fattr *fattr, 1794 const struct inode *inode) 1795 { 1796 const unsigned long check_valid = 1797 NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME | 1798 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE | 1799 NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_OTHER | 1800 NFS_INO_INVALID_NLINK; 1801 unsigned long cache_validity = NFS_I(inode)->cache_validity; 1802 enum nfs4_change_attr_type ctype = NFS_SERVER(inode)->change_attr_type; 1803 1804 if (ctype != NFS4_CHANGE_TYPE_IS_UNDEFINED && 1805 !(cache_validity & NFS_INO_INVALID_CHANGE) && 1806 (cache_validity & check_valid) != 0 && 1807 (fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && 1808 nfs_inode_attrs_cmp_monotonic(fattr, inode) == 0) 1809 return 1; 1810 return 0; 1811 } 1812 1813 static int nfs_refresh_inode_locked(struct inode *inode, 1814 struct nfs_fattr *fattr) 1815 { 1816 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode); 1817 int ret = 0; 1818 1819 trace_nfs_refresh_inode_enter(inode); 1820 1821 if (attr_cmp > 0 || nfs_inode_finish_partial_attr_update(fattr, inode)) 1822 ret = nfs_update_inode(inode, fattr); 1823 else if (attr_cmp == 0) 1824 ret = nfs_check_inode_attributes(inode, fattr); 1825 1826 trace_nfs_refresh_inode_exit(inode, ret); 1827 return ret; 1828 } 1829 1830 /** 1831 * nfs_refresh_inode - try to update the inode attribute cache 1832 * @inode: pointer to inode 1833 * @fattr: updated attributes 1834 * 1835 * Check that an RPC call that returned attributes has not overlapped with 1836 * other recent updates of the inode metadata, then decide whether it is 1837 * safe to do a full update of the inode attributes, or whether just to 1838 * call nfs_check_inode_attributes. 1839 */ 1840 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr) 1841 { 1842 int status; 1843 1844 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 1845 return 0; 1846 spin_lock(&inode->i_lock); 1847 status = nfs_refresh_inode_locked(inode, fattr); 1848 spin_unlock(&inode->i_lock); 1849 1850 return status; 1851 } 1852 EXPORT_SYMBOL_GPL(nfs_refresh_inode); 1853 1854 static int nfs_post_op_update_inode_locked(struct inode *inode, 1855 struct nfs_fattr *fattr, unsigned int invalid) 1856 { 1857 if (S_ISDIR(inode->i_mode)) 1858 invalid |= NFS_INO_INVALID_DATA; 1859 nfs_set_cache_invalid(inode, invalid); 1860 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 1861 return 0; 1862 return nfs_refresh_inode_locked(inode, fattr); 1863 } 1864 1865 /** 1866 * nfs_post_op_update_inode - try to update the inode attribute cache 1867 * @inode: pointer to inode 1868 * @fattr: updated attributes 1869 * 1870 * After an operation that has changed the inode metadata, mark the 1871 * attribute cache as being invalid, then try to update it. 1872 * 1873 * NB: if the server didn't return any post op attributes, this 1874 * function will force the retrieval of attributes before the next 1875 * NFS request. Thus it should be used only for operations that 1876 * are expected to change one or more attributes, to avoid 1877 * unnecessary NFS requests and trips through nfs_update_inode(). 1878 */ 1879 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1880 { 1881 int status; 1882 1883 spin_lock(&inode->i_lock); 1884 nfs_fattr_set_barrier(fattr); 1885 status = nfs_post_op_update_inode_locked(inode, fattr, 1886 NFS_INO_INVALID_CHANGE 1887 | NFS_INO_INVALID_CTIME 1888 | NFS_INO_REVAL_FORCED); 1889 spin_unlock(&inode->i_lock); 1890 1891 return status; 1892 } 1893 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode); 1894 1895 /** 1896 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache 1897 * @inode: pointer to inode 1898 * @fattr: updated attributes 1899 * 1900 * After an operation that has changed the inode metadata, mark the 1901 * attribute cache as being invalid, then try to update it. Fake up 1902 * weak cache consistency data, if none exist. 1903 * 1904 * This function is mainly designed to be used by the ->write_done() functions. 1905 */ 1906 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr) 1907 { 1908 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode); 1909 int status; 1910 1911 /* Don't do a WCC update if these attributes are already stale */ 1912 if (attr_cmp < 0) 1913 return 0; 1914 if ((fattr->valid & NFS_ATTR_FATTR) == 0 || !attr_cmp) { 1915 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE 1916 | NFS_ATTR_FATTR_PRESIZE 1917 | NFS_ATTR_FATTR_PREMTIME 1918 | NFS_ATTR_FATTR_PRECTIME); 1919 goto out_noforce; 1920 } 1921 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && 1922 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) { 1923 fattr->pre_change_attr = inode_peek_iversion_raw(inode); 1924 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE; 1925 } 1926 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 && 1927 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) { 1928 fattr->pre_ctime = inode->i_ctime; 1929 fattr->valid |= NFS_ATTR_FATTR_PRECTIME; 1930 } 1931 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 && 1932 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) { 1933 fattr->pre_mtime = inode->i_mtime; 1934 fattr->valid |= NFS_ATTR_FATTR_PREMTIME; 1935 } 1936 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 && 1937 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) { 1938 fattr->pre_size = i_size_read(inode); 1939 fattr->valid |= NFS_ATTR_FATTR_PRESIZE; 1940 } 1941 out_noforce: 1942 status = nfs_post_op_update_inode_locked(inode, fattr, 1943 NFS_INO_INVALID_CHANGE 1944 | NFS_INO_INVALID_CTIME 1945 | NFS_INO_INVALID_MTIME 1946 | NFS_INO_INVALID_BLOCKS); 1947 return status; 1948 } 1949 1950 /** 1951 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache 1952 * @inode: pointer to inode 1953 * @fattr: updated attributes 1954 * 1955 * After an operation that has changed the inode metadata, mark the 1956 * attribute cache as being invalid, then try to update it. Fake up 1957 * weak cache consistency data, if none exist. 1958 * 1959 * This function is mainly designed to be used by the ->write_done() functions. 1960 */ 1961 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr) 1962 { 1963 int status; 1964 1965 spin_lock(&inode->i_lock); 1966 nfs_fattr_set_barrier(fattr); 1967 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 1968 spin_unlock(&inode->i_lock); 1969 return status; 1970 } 1971 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc); 1972 1973 1974 /* 1975 * Many nfs protocol calls return the new file attributes after 1976 * an operation. Here we update the inode to reflect the state 1977 * of the server's inode. 1978 * 1979 * This is a bit tricky because we have to make sure all dirty pages 1980 * have been sent off to the server before calling invalidate_inode_pages. 1981 * To make sure no other process adds more write requests while we try 1982 * our best to flush them, we make them sleep during the attribute refresh. 1983 * 1984 * A very similar scenario holds for the dir cache. 1985 */ 1986 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1987 { 1988 struct nfs_server *server = NFS_SERVER(inode); 1989 struct nfs_inode *nfsi = NFS_I(inode); 1990 loff_t cur_isize, new_isize; 1991 u64 fattr_supported = server->fattr_valid; 1992 unsigned long invalid = 0; 1993 unsigned long now = jiffies; 1994 unsigned long save_cache_validity; 1995 bool have_writers = nfs_file_has_buffered_writers(nfsi); 1996 bool cache_revalidated = true; 1997 bool attr_changed = false; 1998 bool have_delegation; 1999 2000 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n", 2001 __func__, inode->i_sb->s_id, inode->i_ino, 2002 nfs_display_fhandle_hash(NFS_FH(inode)), 2003 atomic_read(&inode->i_count), fattr->valid); 2004 2005 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) { 2006 /* Only a mounted-on-fileid? Just exit */ 2007 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) 2008 return 0; 2009 /* Has the inode gone and changed behind our back? */ 2010 } else if (nfsi->fileid != fattr->fileid) { 2011 /* Is this perhaps the mounted-on fileid? */ 2012 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) && 2013 nfsi->fileid == fattr->mounted_on_fileid) 2014 return 0; 2015 printk(KERN_ERR "NFS: server %s error: fileid changed\n" 2016 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n", 2017 NFS_SERVER(inode)->nfs_client->cl_hostname, 2018 inode->i_sb->s_id, (long long)nfsi->fileid, 2019 (long long)fattr->fileid); 2020 goto out_err; 2021 } 2022 2023 /* 2024 * Make sure the inode's type hasn't changed. 2025 */ 2026 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) { 2027 /* 2028 * Big trouble! The inode has become a different object. 2029 */ 2030 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n", 2031 __func__, inode->i_ino, inode->i_mode, fattr->mode); 2032 goto out_err; 2033 } 2034 2035 /* Update the fsid? */ 2036 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) && 2037 !nfs_fsid_equal(&server->fsid, &fattr->fsid) && 2038 !IS_AUTOMOUNT(inode)) 2039 server->fsid = fattr->fsid; 2040 2041 /* Save the delegation state before clearing cache_validity */ 2042 have_delegation = nfs_have_delegated_attributes(inode); 2043 2044 /* 2045 * Update the read time so we don't revalidate too often. 2046 */ 2047 nfsi->read_cache_jiffies = fattr->time_start; 2048 2049 save_cache_validity = nfsi->cache_validity; 2050 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR 2051 | NFS_INO_INVALID_ATIME 2052 | NFS_INO_REVAL_FORCED 2053 | NFS_INO_INVALID_BLOCKS); 2054 2055 /* Do atomic weak cache consistency updates */ 2056 nfs_wcc_update_inode(inode, fattr); 2057 2058 if (pnfs_layoutcommit_outstanding(inode)) { 2059 nfsi->cache_validity |= 2060 save_cache_validity & 2061 (NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME | 2062 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE | 2063 NFS_INO_INVALID_BLOCKS); 2064 cache_revalidated = false; 2065 } 2066 2067 /* More cache consistency checks */ 2068 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) { 2069 if (!inode_eq_iversion_raw(inode, fattr->change_attr)) { 2070 /* Could it be a race with writeback? */ 2071 if (!(have_writers || have_delegation)) { 2072 invalid |= NFS_INO_INVALID_DATA 2073 | NFS_INO_INVALID_ACCESS 2074 | NFS_INO_INVALID_ACL 2075 | NFS_INO_INVALID_XATTR; 2076 /* Force revalidate of all attributes */ 2077 save_cache_validity |= NFS_INO_INVALID_CTIME 2078 | NFS_INO_INVALID_MTIME 2079 | NFS_INO_INVALID_SIZE 2080 | NFS_INO_INVALID_BLOCKS 2081 | NFS_INO_INVALID_NLINK 2082 | NFS_INO_INVALID_MODE 2083 | NFS_INO_INVALID_OTHER; 2084 if (S_ISDIR(inode->i_mode)) 2085 nfs_force_lookup_revalidate(inode); 2086 attr_changed = true; 2087 dprintk("NFS: change_attr change on server for file %s/%ld\n", 2088 inode->i_sb->s_id, 2089 inode->i_ino); 2090 } else if (!have_delegation) 2091 nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER; 2092 inode_set_iversion_raw(inode, fattr->change_attr); 2093 } 2094 } else { 2095 nfsi->cache_validity |= 2096 save_cache_validity & NFS_INO_INVALID_CHANGE; 2097 if (!have_delegation || 2098 (nfsi->cache_validity & NFS_INO_INVALID_CHANGE) != 0) 2099 cache_revalidated = false; 2100 } 2101 2102 if (fattr->valid & NFS_ATTR_FATTR_MTIME) 2103 inode->i_mtime = fattr->mtime; 2104 else if (fattr_supported & NFS_ATTR_FATTR_MTIME) 2105 nfsi->cache_validity |= 2106 save_cache_validity & NFS_INO_INVALID_MTIME; 2107 2108 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 2109 inode->i_ctime = fattr->ctime; 2110 else if (fattr_supported & NFS_ATTR_FATTR_CTIME) 2111 nfsi->cache_validity |= 2112 save_cache_validity & NFS_INO_INVALID_CTIME; 2113 2114 /* Check if our cached file size is stale */ 2115 if (fattr->valid & NFS_ATTR_FATTR_SIZE) { 2116 new_isize = nfs_size_to_loff_t(fattr->size); 2117 cur_isize = i_size_read(inode); 2118 if (new_isize != cur_isize && !have_delegation) { 2119 /* Do we perhaps have any outstanding writes, or has 2120 * the file grown beyond our last write? */ 2121 if (!nfs_have_writebacks(inode) || new_isize > cur_isize) { 2122 trace_nfs_size_update(inode, new_isize); 2123 i_size_write(inode, new_isize); 2124 if (!have_writers) 2125 invalid |= NFS_INO_INVALID_DATA; 2126 } 2127 } 2128 if (new_isize == 0 && 2129 !(fattr->valid & (NFS_ATTR_FATTR_SPACE_USED | 2130 NFS_ATTR_FATTR_BLOCKS_USED))) { 2131 fattr->du.nfs3.used = 0; 2132 fattr->valid |= NFS_ATTR_FATTR_SPACE_USED; 2133 } 2134 } else 2135 nfsi->cache_validity |= 2136 save_cache_validity & NFS_INO_INVALID_SIZE; 2137 2138 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 2139 inode->i_atime = fattr->atime; 2140 else if (fattr_supported & NFS_ATTR_FATTR_ATIME) 2141 nfsi->cache_validity |= 2142 save_cache_validity & NFS_INO_INVALID_ATIME; 2143 2144 if (fattr->valid & NFS_ATTR_FATTR_MODE) { 2145 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) { 2146 umode_t newmode = inode->i_mode & S_IFMT; 2147 newmode |= fattr->mode & S_IALLUGO; 2148 inode->i_mode = newmode; 2149 invalid |= NFS_INO_INVALID_ACCESS 2150 | NFS_INO_INVALID_ACL; 2151 } 2152 } else if (fattr_supported & NFS_ATTR_FATTR_MODE) 2153 nfsi->cache_validity |= 2154 save_cache_validity & NFS_INO_INVALID_MODE; 2155 2156 if (fattr->valid & NFS_ATTR_FATTR_OWNER) { 2157 if (!uid_eq(inode->i_uid, fattr->uid)) { 2158 invalid |= NFS_INO_INVALID_ACCESS 2159 | NFS_INO_INVALID_ACL; 2160 inode->i_uid = fattr->uid; 2161 } 2162 } else if (fattr_supported & NFS_ATTR_FATTR_OWNER) 2163 nfsi->cache_validity |= 2164 save_cache_validity & NFS_INO_INVALID_OTHER; 2165 2166 if (fattr->valid & NFS_ATTR_FATTR_GROUP) { 2167 if (!gid_eq(inode->i_gid, fattr->gid)) { 2168 invalid |= NFS_INO_INVALID_ACCESS 2169 | NFS_INO_INVALID_ACL; 2170 inode->i_gid = fattr->gid; 2171 } 2172 } else if (fattr_supported & NFS_ATTR_FATTR_GROUP) 2173 nfsi->cache_validity |= 2174 save_cache_validity & NFS_INO_INVALID_OTHER; 2175 2176 if (fattr->valid & NFS_ATTR_FATTR_NLINK) { 2177 if (inode->i_nlink != fattr->nlink) 2178 set_nlink(inode, fattr->nlink); 2179 } else if (fattr_supported & NFS_ATTR_FATTR_NLINK) 2180 nfsi->cache_validity |= 2181 save_cache_validity & NFS_INO_INVALID_NLINK; 2182 2183 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) { 2184 /* 2185 * report the blocks in 512byte units 2186 */ 2187 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 2188 } else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED) 2189 nfsi->cache_validity |= 2190 save_cache_validity & NFS_INO_INVALID_BLOCKS; 2191 2192 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED) 2193 inode->i_blocks = fattr->du.nfs2.blocks; 2194 else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED) 2195 nfsi->cache_validity |= 2196 save_cache_validity & NFS_INO_INVALID_BLOCKS; 2197 2198 /* Update attrtimeo value if we're out of the unstable period */ 2199 if (attr_changed) { 2200 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 2201 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 2202 nfsi->attrtimeo_timestamp = now; 2203 /* Set barrier to be more recent than all outstanding updates */ 2204 nfsi->attr_gencount = nfs_inc_attr_generation_counter(); 2205 } else { 2206 if (cache_revalidated) { 2207 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, 2208 nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) { 2209 nfsi->attrtimeo <<= 1; 2210 if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode)) 2211 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode); 2212 } 2213 nfsi->attrtimeo_timestamp = now; 2214 } 2215 /* Set the barrier to be more recent than this fattr */ 2216 if ((long)(fattr->gencount - nfsi->attr_gencount) > 0) 2217 nfsi->attr_gencount = fattr->gencount; 2218 } 2219 2220 /* Don't invalidate the data if we were to blame */ 2221 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) 2222 || S_ISLNK(inode->i_mode))) 2223 invalid &= ~NFS_INO_INVALID_DATA; 2224 nfs_set_cache_invalid(inode, invalid); 2225 2226 return 0; 2227 out_err: 2228 /* 2229 * No need to worry about unhashing the dentry, as the 2230 * lookup validation will know that the inode is bad. 2231 * (But we fall through to invalidate the caches.) 2232 */ 2233 nfs_set_inode_stale_locked(inode); 2234 return -ESTALE; 2235 } 2236 2237 struct inode *nfs_alloc_inode(struct super_block *sb) 2238 { 2239 struct nfs_inode *nfsi; 2240 nfsi = alloc_inode_sb(sb, nfs_inode_cachep, GFP_KERNEL); 2241 if (!nfsi) 2242 return NULL; 2243 nfsi->flags = 0UL; 2244 nfsi->cache_validity = 0UL; 2245 #if IS_ENABLED(CONFIG_NFS_V4) 2246 nfsi->nfs4_acl = NULL; 2247 #endif /* CONFIG_NFS_V4 */ 2248 #ifdef CONFIG_NFS_V4_2 2249 nfsi->xattr_cache = NULL; 2250 #endif 2251 return &nfsi->vfs_inode; 2252 } 2253 EXPORT_SYMBOL_GPL(nfs_alloc_inode); 2254 2255 void nfs_free_inode(struct inode *inode) 2256 { 2257 kmem_cache_free(nfs_inode_cachep, NFS_I(inode)); 2258 } 2259 EXPORT_SYMBOL_GPL(nfs_free_inode); 2260 2261 static inline void nfs4_init_once(struct nfs_inode *nfsi) 2262 { 2263 #if IS_ENABLED(CONFIG_NFS_V4) 2264 INIT_LIST_HEAD(&nfsi->open_states); 2265 nfsi->delegation = NULL; 2266 init_rwsem(&nfsi->rwsem); 2267 nfsi->layout = NULL; 2268 #endif 2269 } 2270 2271 static void init_once(void *foo) 2272 { 2273 struct nfs_inode *nfsi = (struct nfs_inode *) foo; 2274 2275 inode_init_once(&nfsi->vfs_inode); 2276 INIT_LIST_HEAD(&nfsi->open_files); 2277 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru); 2278 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru); 2279 nfs4_init_once(nfsi); 2280 } 2281 2282 static int __init nfs_init_inodecache(void) 2283 { 2284 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache", 2285 sizeof(struct nfs_inode), 2286 0, (SLAB_RECLAIM_ACCOUNT| 2287 SLAB_MEM_SPREAD|SLAB_ACCOUNT), 2288 init_once); 2289 if (nfs_inode_cachep == NULL) 2290 return -ENOMEM; 2291 2292 return 0; 2293 } 2294 2295 static void nfs_destroy_inodecache(void) 2296 { 2297 /* 2298 * Make sure all delayed rcu free inodes are flushed before we 2299 * destroy cache. 2300 */ 2301 rcu_barrier(); 2302 kmem_cache_destroy(nfs_inode_cachep); 2303 } 2304 2305 struct workqueue_struct *nfsiod_workqueue; 2306 EXPORT_SYMBOL_GPL(nfsiod_workqueue); 2307 2308 /* 2309 * start up the nfsiod workqueue 2310 */ 2311 static int nfsiod_start(void) 2312 { 2313 struct workqueue_struct *wq; 2314 dprintk("RPC: creating workqueue nfsiod\n"); 2315 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0); 2316 if (wq == NULL) 2317 return -ENOMEM; 2318 nfsiod_workqueue = wq; 2319 return 0; 2320 } 2321 2322 /* 2323 * Destroy the nfsiod workqueue 2324 */ 2325 static void nfsiod_stop(void) 2326 { 2327 struct workqueue_struct *wq; 2328 2329 wq = nfsiod_workqueue; 2330 if (wq == NULL) 2331 return; 2332 nfsiod_workqueue = NULL; 2333 destroy_workqueue(wq); 2334 } 2335 2336 unsigned int nfs_net_id; 2337 EXPORT_SYMBOL_GPL(nfs_net_id); 2338 2339 static int nfs_net_init(struct net *net) 2340 { 2341 nfs_clients_init(net); 2342 return nfs_fs_proc_net_init(net); 2343 } 2344 2345 static void nfs_net_exit(struct net *net) 2346 { 2347 nfs_fs_proc_net_exit(net); 2348 nfs_clients_exit(net); 2349 } 2350 2351 static struct pernet_operations nfs_net_ops = { 2352 .init = nfs_net_init, 2353 .exit = nfs_net_exit, 2354 .id = &nfs_net_id, 2355 .size = sizeof(struct nfs_net), 2356 }; 2357 2358 /* 2359 * Initialize NFS 2360 */ 2361 static int __init init_nfs_fs(void) 2362 { 2363 int err; 2364 2365 err = nfs_sysfs_init(); 2366 if (err < 0) 2367 goto out10; 2368 2369 err = register_pernet_subsys(&nfs_net_ops); 2370 if (err < 0) 2371 goto out9; 2372 2373 err = nfsiod_start(); 2374 if (err) 2375 goto out7; 2376 2377 err = nfs_fs_proc_init(); 2378 if (err) 2379 goto out6; 2380 2381 err = nfs_init_nfspagecache(); 2382 if (err) 2383 goto out5; 2384 2385 err = nfs_init_inodecache(); 2386 if (err) 2387 goto out4; 2388 2389 err = nfs_init_readpagecache(); 2390 if (err) 2391 goto out3; 2392 2393 err = nfs_init_writepagecache(); 2394 if (err) 2395 goto out2; 2396 2397 err = nfs_init_directcache(); 2398 if (err) 2399 goto out1; 2400 2401 rpc_proc_register(&init_net, &nfs_rpcstat); 2402 2403 err = register_nfs_fs(); 2404 if (err) 2405 goto out0; 2406 2407 return 0; 2408 out0: 2409 rpc_proc_unregister(&init_net, "nfs"); 2410 nfs_destroy_directcache(); 2411 out1: 2412 nfs_destroy_writepagecache(); 2413 out2: 2414 nfs_destroy_readpagecache(); 2415 out3: 2416 nfs_destroy_inodecache(); 2417 out4: 2418 nfs_destroy_nfspagecache(); 2419 out5: 2420 nfs_fs_proc_exit(); 2421 out6: 2422 nfsiod_stop(); 2423 out7: 2424 unregister_pernet_subsys(&nfs_net_ops); 2425 out9: 2426 nfs_sysfs_exit(); 2427 out10: 2428 return err; 2429 } 2430 2431 static void __exit exit_nfs_fs(void) 2432 { 2433 nfs_destroy_directcache(); 2434 nfs_destroy_writepagecache(); 2435 nfs_destroy_readpagecache(); 2436 nfs_destroy_inodecache(); 2437 nfs_destroy_nfspagecache(); 2438 unregister_pernet_subsys(&nfs_net_ops); 2439 rpc_proc_unregister(&init_net, "nfs"); 2440 unregister_nfs_fs(); 2441 nfs_fs_proc_exit(); 2442 nfsiod_stop(); 2443 nfs_sysfs_exit(); 2444 } 2445 2446 /* Not quite true; I just maintain it */ 2447 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>"); 2448 MODULE_LICENSE("GPL"); 2449 module_param(enable_ino64, bool, 0644); 2450 2451 module_init(init_nfs_fs) 2452 module_exit(exit_nfs_fs) 2453