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