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_invalid_delegated(struct inode *inode, unsigned long flags) 168 { 169 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity); 170 171 /* Special case for the pagecache or access cache */ 172 if (flags == NFS_INO_REVAL_PAGECACHE && 173 !(cache_validity & NFS_INO_REVAL_FORCED)) 174 return false; 175 return (cache_validity & flags) != 0; 176 } 177 178 static bool nfs_check_cache_invalid_not_delegated(struct inode *inode, unsigned long flags) 179 { 180 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity); 181 182 if ((cache_validity & flags) != 0) 183 return true; 184 if (nfs_attribute_timeout(inode)) 185 return true; 186 return false; 187 } 188 189 bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags) 190 { 191 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) 192 return nfs_check_cache_invalid_delegated(inode, flags); 193 194 return nfs_check_cache_invalid_not_delegated(inode, flags); 195 } 196 EXPORT_SYMBOL_GPL(nfs_check_cache_invalid); 197 198 #ifdef CONFIG_NFS_V4_2 199 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi) 200 { 201 return nfsi->xattr_cache != NULL; 202 } 203 #else 204 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi) 205 { 206 return false; 207 } 208 #endif 209 210 void nfs_set_cache_invalid(struct inode *inode, unsigned long flags) 211 { 212 struct nfs_inode *nfsi = NFS_I(inode); 213 bool have_delegation = NFS_PROTO(inode)->have_delegation(inode, FMODE_READ); 214 215 if (have_delegation) { 216 if (!(flags & NFS_INO_REVAL_FORCED)) 217 flags &= ~NFS_INO_INVALID_OTHER; 218 flags &= ~(NFS_INO_INVALID_CHANGE 219 | NFS_INO_INVALID_SIZE 220 | NFS_INO_REVAL_PAGECACHE 221 | NFS_INO_INVALID_XATTR); 222 } 223 224 if (!nfs_has_xattr_cache(nfsi)) 225 flags &= ~NFS_INO_INVALID_XATTR; 226 if (inode->i_mapping->nrpages == 0) 227 flags &= ~(NFS_INO_INVALID_DATA|NFS_INO_DATA_INVAL_DEFER); 228 nfsi->cache_validity |= flags; 229 if (flags & NFS_INO_INVALID_DATA) 230 nfs_fscache_invalidate(inode); 231 } 232 EXPORT_SYMBOL_GPL(nfs_set_cache_invalid); 233 234 /* 235 * Invalidate the local caches 236 */ 237 static void nfs_zap_caches_locked(struct inode *inode) 238 { 239 struct nfs_inode *nfsi = NFS_I(inode); 240 int mode = inode->i_mode; 241 242 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 243 244 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 245 nfsi->attrtimeo_timestamp = jiffies; 246 247 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) { 248 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR 249 | NFS_INO_INVALID_DATA 250 | NFS_INO_INVALID_ACCESS 251 | NFS_INO_INVALID_ACL 252 | NFS_INO_INVALID_XATTR 253 | NFS_INO_REVAL_PAGECACHE); 254 } else 255 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR 256 | NFS_INO_INVALID_ACCESS 257 | NFS_INO_INVALID_ACL 258 | NFS_INO_INVALID_XATTR 259 | NFS_INO_REVAL_PAGECACHE); 260 nfs_zap_label_cache_locked(nfsi); 261 } 262 263 void nfs_zap_caches(struct inode *inode) 264 { 265 spin_lock(&inode->i_lock); 266 nfs_zap_caches_locked(inode); 267 spin_unlock(&inode->i_lock); 268 } 269 270 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping) 271 { 272 if (mapping->nrpages != 0) { 273 spin_lock(&inode->i_lock); 274 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA); 275 spin_unlock(&inode->i_lock); 276 } 277 } 278 279 void nfs_zap_acl_cache(struct inode *inode) 280 { 281 void (*clear_acl_cache)(struct inode *); 282 283 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache; 284 if (clear_acl_cache != NULL) 285 clear_acl_cache(inode); 286 spin_lock(&inode->i_lock); 287 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL; 288 spin_unlock(&inode->i_lock); 289 } 290 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache); 291 292 void nfs_invalidate_atime(struct inode *inode) 293 { 294 spin_lock(&inode->i_lock); 295 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME); 296 spin_unlock(&inode->i_lock); 297 } 298 EXPORT_SYMBOL_GPL(nfs_invalidate_atime); 299 300 /* 301 * Invalidate, but do not unhash, the inode. 302 * NB: must be called with inode->i_lock held! 303 */ 304 static void nfs_set_inode_stale_locked(struct inode *inode) 305 { 306 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); 307 nfs_zap_caches_locked(inode); 308 trace_nfs_set_inode_stale(inode); 309 } 310 311 void nfs_set_inode_stale(struct inode *inode) 312 { 313 spin_lock(&inode->i_lock); 314 nfs_set_inode_stale_locked(inode); 315 spin_unlock(&inode->i_lock); 316 } 317 318 struct nfs_find_desc { 319 struct nfs_fh *fh; 320 struct nfs_fattr *fattr; 321 }; 322 323 /* 324 * In NFSv3 we can have 64bit inode numbers. In order to support 325 * this, and re-exported directories (also seen in NFSv2) 326 * we are forced to allow 2 different inodes to have the same 327 * i_ino. 328 */ 329 static int 330 nfs_find_actor(struct inode *inode, void *opaque) 331 { 332 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque; 333 struct nfs_fh *fh = desc->fh; 334 struct nfs_fattr *fattr = desc->fattr; 335 336 if (NFS_FILEID(inode) != fattr->fileid) 337 return 0; 338 if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode)) 339 return 0; 340 if (nfs_compare_fh(NFS_FH(inode), fh)) 341 return 0; 342 if (is_bad_inode(inode) || NFS_STALE(inode)) 343 return 0; 344 return 1; 345 } 346 347 static int 348 nfs_init_locked(struct inode *inode, void *opaque) 349 { 350 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque; 351 struct nfs_fattr *fattr = desc->fattr; 352 353 set_nfs_fileid(inode, fattr->fileid); 354 inode->i_mode = fattr->mode; 355 nfs_copy_fh(NFS_FH(inode), desc->fh); 356 return 0; 357 } 358 359 #ifdef CONFIG_NFS_V4_SECURITY_LABEL 360 static void nfs_clear_label_invalid(struct inode *inode) 361 { 362 spin_lock(&inode->i_lock); 363 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL; 364 spin_unlock(&inode->i_lock); 365 } 366 367 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr, 368 struct nfs4_label *label) 369 { 370 int error; 371 372 if (label == NULL) 373 return; 374 375 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) { 376 error = security_inode_notifysecctx(inode, label->label, 377 label->len); 378 if (error) 379 printk(KERN_ERR "%s() %s %d " 380 "security_inode_notifysecctx() %d\n", 381 __func__, 382 (char *)label->label, 383 label->len, error); 384 nfs_clear_label_invalid(inode); 385 } 386 } 387 388 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags) 389 { 390 struct nfs4_label *label = NULL; 391 int minor_version = server->nfs_client->cl_minorversion; 392 393 if (minor_version < 2) 394 return label; 395 396 if (!(server->caps & NFS_CAP_SECURITY_LABEL)) 397 return label; 398 399 label = kzalloc(sizeof(struct nfs4_label), flags); 400 if (label == NULL) 401 return ERR_PTR(-ENOMEM); 402 403 label->label = kzalloc(NFS4_MAXLABELLEN, flags); 404 if (label->label == NULL) { 405 kfree(label); 406 return ERR_PTR(-ENOMEM); 407 } 408 label->len = NFS4_MAXLABELLEN; 409 410 return label; 411 } 412 EXPORT_SYMBOL_GPL(nfs4_label_alloc); 413 #else 414 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr, 415 struct nfs4_label *label) 416 { 417 } 418 #endif 419 EXPORT_SYMBOL_GPL(nfs_setsecurity); 420 421 /* Search for inode identified by fh, fileid and i_mode in inode cache. */ 422 struct inode * 423 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh) 424 { 425 struct nfs_find_desc desc = { 426 .fh = fh, 427 .fattr = fattr, 428 }; 429 struct inode *inode; 430 unsigned long hash; 431 432 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) || 433 !(fattr->valid & NFS_ATTR_FATTR_TYPE)) 434 return NULL; 435 436 hash = nfs_fattr_to_ino_t(fattr); 437 inode = ilookup5(sb, hash, nfs_find_actor, &desc); 438 439 dprintk("%s: returning %p\n", __func__, inode); 440 return inode; 441 } 442 443 /* 444 * This is our front-end to iget that looks up inodes by file handle 445 * instead of inode number. 446 */ 447 struct inode * 448 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label) 449 { 450 struct nfs_find_desc desc = { 451 .fh = fh, 452 .fattr = fattr 453 }; 454 struct inode *inode = ERR_PTR(-ENOENT); 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 && nfs_server_capable(inode, NFS_CAP_MODE)) 488 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER); 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 } else if (S_ISDIR(inode->i_mode)) { 497 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops; 498 inode->i_fop = &nfs_dir_operations; 499 inode->i_data.a_ops = &nfs_dir_aops; 500 /* Deal with crossing mountpoints */ 501 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT || 502 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) { 503 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) 504 inode->i_op = &nfs_referral_inode_operations; 505 else 506 inode->i_op = &nfs_mountpoint_inode_operations; 507 inode->i_fop = NULL; 508 inode->i_flags |= S_AUTOMOUNT; 509 } 510 } else if (S_ISLNK(inode->i_mode)) { 511 inode->i_op = &nfs_symlink_inode_operations; 512 inode_nohighmem(inode); 513 } else 514 init_special_inode(inode, inode->i_mode, fattr->rdev); 515 516 memset(&inode->i_atime, 0, sizeof(inode->i_atime)); 517 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime)); 518 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime)); 519 inode_set_iversion_raw(inode, 0); 520 inode->i_size = 0; 521 clear_nlink(inode); 522 inode->i_uid = make_kuid(&init_user_ns, -2); 523 inode->i_gid = make_kgid(&init_user_ns, -2); 524 inode->i_blocks = 0; 525 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf)); 526 nfsi->write_io = 0; 527 nfsi->read_io = 0; 528 529 nfsi->read_cache_jiffies = fattr->time_start; 530 nfsi->attr_gencount = fattr->gencount; 531 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 532 inode->i_atime = fattr->atime; 533 else if (nfs_server_capable(inode, NFS_CAP_ATIME)) 534 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME); 535 if (fattr->valid & NFS_ATTR_FATTR_MTIME) 536 inode->i_mtime = fattr->mtime; 537 else if (nfs_server_capable(inode, NFS_CAP_MTIME)) 538 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME); 539 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 540 inode->i_ctime = fattr->ctime; 541 else if (nfs_server_capable(inode, NFS_CAP_CTIME)) 542 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME); 543 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) 544 inode_set_iversion_raw(inode, fattr->change_attr); 545 else 546 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE); 547 if (fattr->valid & NFS_ATTR_FATTR_SIZE) 548 inode->i_size = nfs_size_to_loff_t(fattr->size); 549 else 550 nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE); 551 if (fattr->valid & NFS_ATTR_FATTR_NLINK) 552 set_nlink(inode, fattr->nlink); 553 else if (nfs_server_capable(inode, NFS_CAP_NLINK)) 554 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER); 555 if (fattr->valid & NFS_ATTR_FATTR_OWNER) 556 inode->i_uid = fattr->uid; 557 else if (nfs_server_capable(inode, NFS_CAP_OWNER)) 558 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER); 559 if (fattr->valid & NFS_ATTR_FATTR_GROUP) 560 inode->i_gid = fattr->gid; 561 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP)) 562 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER); 563 if (nfs_server_capable(inode, NFS_CAP_XATTR)) 564 nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR); 565 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED) 566 inode->i_blocks = fattr->du.nfs2.blocks; 567 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) { 568 /* 569 * report the blocks in 512byte units 570 */ 571 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 572 } 573 574 if (nfsi->cache_validity != 0) 575 nfsi->cache_validity |= NFS_INO_REVAL_FORCED; 576 577 nfs_setsecurity(inode, fattr, label); 578 579 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 580 nfsi->attrtimeo_timestamp = now; 581 nfsi->access_cache = RB_ROOT; 582 583 nfs_fscache_init_inode(inode); 584 585 unlock_new_inode(inode); 586 } else { 587 int err = nfs_refresh_inode(inode, fattr); 588 if (err < 0) { 589 iput(inode); 590 inode = ERR_PTR(err); 591 goto out_no_inode; 592 } 593 } 594 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n", 595 inode->i_sb->s_id, 596 (unsigned long long)NFS_FILEID(inode), 597 nfs_display_fhandle_hash(fh), 598 atomic_read(&inode->i_count)); 599 600 out: 601 return inode; 602 603 out_no_inode: 604 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode)); 605 goto out; 606 } 607 EXPORT_SYMBOL_GPL(nfs_fhget); 608 609 #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) 610 611 int 612 nfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry, 613 struct iattr *attr) 614 { 615 struct inode *inode = d_inode(dentry); 616 struct nfs_fattr *fattr; 617 int error = 0; 618 619 nfs_inc_stats(inode, NFSIOS_VFSSETATTR); 620 621 /* skip mode change if it's just for clearing setuid/setgid */ 622 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID)) 623 attr->ia_valid &= ~ATTR_MODE; 624 625 if (attr->ia_valid & ATTR_SIZE) { 626 BUG_ON(!S_ISREG(inode->i_mode)); 627 628 error = inode_newsize_ok(inode, attr->ia_size); 629 if (error) 630 return error; 631 632 if (attr->ia_size == i_size_read(inode)) 633 attr->ia_valid &= ~ATTR_SIZE; 634 } 635 636 /* Optimization: if the end result is no change, don't RPC */ 637 attr->ia_valid &= NFS_VALID_ATTRS; 638 if ((attr->ia_valid & ~(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(); 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 i_size_write(inode, offset); 682 /* Optimisation */ 683 if (offset == 0) 684 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_DATA | 685 NFS_INO_DATA_INVAL_DEFER); 686 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE; 687 688 spin_unlock(&inode->i_lock); 689 truncate_pagecache(inode, offset); 690 spin_lock(&inode->i_lock); 691 out: 692 return err; 693 } 694 695 /** 696 * nfs_setattr_update_inode - Update inode metadata after a setattr call. 697 * @inode: pointer to struct inode 698 * @attr: pointer to struct iattr 699 * @fattr: pointer to struct nfs_fattr 700 * 701 * Note: we do this in the *proc.c in order to ensure that 702 * it works for things like exclusive creates too. 703 */ 704 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr, 705 struct nfs_fattr *fattr) 706 { 707 /* Barrier: bump the attribute generation count. */ 708 nfs_fattr_set_barrier(fattr); 709 710 spin_lock(&inode->i_lock); 711 NFS_I(inode)->attr_gencount = fattr->gencount; 712 if ((attr->ia_valid & ATTR_SIZE) != 0) { 713 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME); 714 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC); 715 nfs_vmtruncate(inode, attr->ia_size); 716 } 717 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) { 718 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME; 719 if ((attr->ia_valid & ATTR_MODE) != 0) { 720 int mode = attr->ia_mode & S_IALLUGO; 721 mode |= inode->i_mode & ~S_IALLUGO; 722 inode->i_mode = mode; 723 } 724 if ((attr->ia_valid & ATTR_UID) != 0) 725 inode->i_uid = attr->ia_uid; 726 if ((attr->ia_valid & ATTR_GID) != 0) 727 inode->i_gid = attr->ia_gid; 728 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 729 inode->i_ctime = fattr->ctime; 730 else 731 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE 732 | NFS_INO_INVALID_CTIME); 733 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS 734 | NFS_INO_INVALID_ACL); 735 } 736 if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) { 737 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME 738 | NFS_INO_INVALID_CTIME); 739 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 740 inode->i_atime = fattr->atime; 741 else if (attr->ia_valid & ATTR_ATIME_SET) 742 inode->i_atime = attr->ia_atime; 743 else 744 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME); 745 746 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 747 inode->i_ctime = fattr->ctime; 748 else 749 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE 750 | NFS_INO_INVALID_CTIME); 751 } 752 if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) { 753 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME 754 | NFS_INO_INVALID_CTIME); 755 if (fattr->valid & NFS_ATTR_FATTR_MTIME) 756 inode->i_mtime = fattr->mtime; 757 else if (attr->ia_valid & ATTR_MTIME_SET) 758 inode->i_mtime = attr->ia_mtime; 759 else 760 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME); 761 762 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 763 inode->i_ctime = fattr->ctime; 764 else 765 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE 766 | NFS_INO_INVALID_CTIME); 767 } 768 if (fattr->valid) 769 nfs_update_inode(inode, fattr); 770 spin_unlock(&inode->i_lock); 771 } 772 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode); 773 774 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry) 775 { 776 struct dentry *parent; 777 778 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS)) 779 return; 780 parent = dget_parent(dentry); 781 nfs_force_use_readdirplus(d_inode(parent)); 782 dput(parent); 783 } 784 785 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry) 786 { 787 struct dentry *parent; 788 789 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS)) 790 return; 791 parent = dget_parent(dentry); 792 nfs_advise_use_readdirplus(d_inode(parent)); 793 dput(parent); 794 } 795 796 static bool nfs_need_revalidate_inode(struct inode *inode) 797 { 798 if (NFS_I(inode)->cache_validity & 799 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL)) 800 return true; 801 if (nfs_attribute_cache_expired(inode)) 802 return true; 803 return false; 804 } 805 806 int nfs_getattr(struct user_namespace *mnt_userns, const struct path *path, 807 struct kstat *stat, u32 request_mask, unsigned int query_flags) 808 { 809 struct inode *inode = d_inode(path->dentry); 810 struct nfs_server *server = NFS_SERVER(inode); 811 unsigned long cache_validity; 812 int err = 0; 813 bool force_sync = query_flags & AT_STATX_FORCE_SYNC; 814 bool do_update = false; 815 816 trace_nfs_getattr_enter(inode); 817 818 if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) { 819 nfs_readdirplus_parent_cache_hit(path->dentry); 820 goto out_no_update; 821 } 822 823 /* Flush out writes to the server in order to update c/mtime. */ 824 if ((request_mask & (STATX_CTIME|STATX_MTIME)) && 825 S_ISREG(inode->i_mode)) { 826 err = filemap_write_and_wait(inode->i_mapping); 827 if (err) 828 goto out; 829 } 830 831 /* 832 * We may force a getattr if the user cares about atime. 833 * 834 * Note that we only have to check the vfsmount flags here: 835 * - NFS always sets S_NOATIME by so checking it would give a 836 * bogus result 837 * - NFS never sets SB_NOATIME or SB_NODIRATIME so there is 838 * no point in checking those. 839 */ 840 if ((path->mnt->mnt_flags & MNT_NOATIME) || 841 ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) 842 request_mask &= ~STATX_ATIME; 843 844 /* Is the user requesting attributes that might need revalidation? */ 845 if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME| 846 STATX_MTIME|STATX_UID|STATX_GID| 847 STATX_SIZE|STATX_BLOCKS))) 848 goto out_no_revalidate; 849 850 /* Check whether the cached attributes are stale */ 851 do_update |= force_sync || nfs_attribute_cache_expired(inode); 852 cache_validity = READ_ONCE(NFS_I(inode)->cache_validity); 853 do_update |= cache_validity & 854 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL); 855 if (request_mask & STATX_ATIME) 856 do_update |= cache_validity & NFS_INO_INVALID_ATIME; 857 if (request_mask & (STATX_CTIME|STATX_MTIME)) 858 do_update |= cache_validity & NFS_INO_REVAL_PAGECACHE; 859 if (request_mask & STATX_BLOCKS) 860 do_update |= cache_validity & NFS_INO_INVALID_BLOCKS; 861 if (do_update) { 862 /* Update the attribute cache */ 863 if (!(server->flags & NFS_MOUNT_NOAC)) 864 nfs_readdirplus_parent_cache_miss(path->dentry); 865 else 866 nfs_readdirplus_parent_cache_hit(path->dentry); 867 err = __nfs_revalidate_inode(server, inode); 868 if (err) 869 goto out; 870 } else 871 nfs_readdirplus_parent_cache_hit(path->dentry); 872 out_no_revalidate: 873 /* Only return attributes that were revalidated. */ 874 stat->result_mask &= request_mask; 875 out_no_update: 876 generic_fillattr(&init_user_ns, inode, stat); 877 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode)); 878 if (S_ISDIR(inode->i_mode)) 879 stat->blksize = NFS_SERVER(inode)->dtsize; 880 out: 881 trace_nfs_getattr_exit(inode, err); 882 return err; 883 } 884 EXPORT_SYMBOL_GPL(nfs_getattr); 885 886 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx) 887 { 888 refcount_set(&l_ctx->count, 1); 889 l_ctx->lockowner = current->files; 890 INIT_LIST_HEAD(&l_ctx->list); 891 atomic_set(&l_ctx->io_count, 0); 892 } 893 894 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx) 895 { 896 struct nfs_lock_context *pos; 897 898 list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) { 899 if (pos->lockowner != current->files) 900 continue; 901 if (refcount_inc_not_zero(&pos->count)) 902 return pos; 903 } 904 return NULL; 905 } 906 907 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx) 908 { 909 struct nfs_lock_context *res, *new = NULL; 910 struct inode *inode = d_inode(ctx->dentry); 911 912 rcu_read_lock(); 913 res = __nfs_find_lock_context(ctx); 914 rcu_read_unlock(); 915 if (res == NULL) { 916 new = kmalloc(sizeof(*new), GFP_KERNEL); 917 if (new == NULL) 918 return ERR_PTR(-ENOMEM); 919 nfs_init_lock_context(new); 920 spin_lock(&inode->i_lock); 921 res = __nfs_find_lock_context(ctx); 922 if (res == NULL) { 923 new->open_context = get_nfs_open_context(ctx); 924 if (new->open_context) { 925 list_add_tail_rcu(&new->list, 926 &ctx->lock_context.list); 927 res = new; 928 new = NULL; 929 } else 930 res = ERR_PTR(-EBADF); 931 } 932 spin_unlock(&inode->i_lock); 933 kfree(new); 934 } 935 return res; 936 } 937 EXPORT_SYMBOL_GPL(nfs_get_lock_context); 938 939 void nfs_put_lock_context(struct nfs_lock_context *l_ctx) 940 { 941 struct nfs_open_context *ctx = l_ctx->open_context; 942 struct inode *inode = d_inode(ctx->dentry); 943 944 if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock)) 945 return; 946 list_del_rcu(&l_ctx->list); 947 spin_unlock(&inode->i_lock); 948 put_nfs_open_context(ctx); 949 kfree_rcu(l_ctx, rcu_head); 950 } 951 EXPORT_SYMBOL_GPL(nfs_put_lock_context); 952 953 /** 954 * nfs_close_context - Common close_context() routine NFSv2/v3 955 * @ctx: pointer to context 956 * @is_sync: is this a synchronous close 957 * 958 * Ensure that the attributes are up to date if we're mounted 959 * with close-to-open semantics and we have cached data that will 960 * need to be revalidated on open. 961 */ 962 void nfs_close_context(struct nfs_open_context *ctx, int is_sync) 963 { 964 struct nfs_inode *nfsi; 965 struct inode *inode; 966 struct nfs_server *server; 967 968 if (!(ctx->mode & FMODE_WRITE)) 969 return; 970 if (!is_sync) 971 return; 972 inode = d_inode(ctx->dentry); 973 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) 974 return; 975 nfsi = NFS_I(inode); 976 if (inode->i_mapping->nrpages == 0) 977 return; 978 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 979 return; 980 if (!list_empty(&nfsi->open_files)) 981 return; 982 server = NFS_SERVER(inode); 983 if (server->flags & NFS_MOUNT_NOCTO) 984 return; 985 nfs_revalidate_inode(server, inode); 986 } 987 EXPORT_SYMBOL_GPL(nfs_close_context); 988 989 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, 990 fmode_t f_mode, 991 struct file *filp) 992 { 993 struct nfs_open_context *ctx; 994 995 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); 996 if (!ctx) 997 return ERR_PTR(-ENOMEM); 998 nfs_sb_active(dentry->d_sb); 999 ctx->dentry = dget(dentry); 1000 if (filp) 1001 ctx->cred = get_cred(filp->f_cred); 1002 else 1003 ctx->cred = get_current_cred(); 1004 ctx->ll_cred = NULL; 1005 ctx->state = NULL; 1006 ctx->mode = f_mode; 1007 ctx->flags = 0; 1008 ctx->error = 0; 1009 ctx->flock_owner = (fl_owner_t)filp; 1010 nfs_init_lock_context(&ctx->lock_context); 1011 ctx->lock_context.open_context = ctx; 1012 INIT_LIST_HEAD(&ctx->list); 1013 ctx->mdsthreshold = NULL; 1014 return ctx; 1015 } 1016 EXPORT_SYMBOL_GPL(alloc_nfs_open_context); 1017 1018 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx) 1019 { 1020 if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count)) 1021 return ctx; 1022 return NULL; 1023 } 1024 EXPORT_SYMBOL_GPL(get_nfs_open_context); 1025 1026 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync) 1027 { 1028 struct inode *inode = d_inode(ctx->dentry); 1029 struct super_block *sb = ctx->dentry->d_sb; 1030 1031 if (!refcount_dec_and_test(&ctx->lock_context.count)) 1032 return; 1033 if (!list_empty(&ctx->list)) { 1034 spin_lock(&inode->i_lock); 1035 list_del_rcu(&ctx->list); 1036 spin_unlock(&inode->i_lock); 1037 } 1038 if (inode != NULL) 1039 NFS_PROTO(inode)->close_context(ctx, is_sync); 1040 put_cred(ctx->cred); 1041 dput(ctx->dentry); 1042 nfs_sb_deactive(sb); 1043 put_rpccred(ctx->ll_cred); 1044 kfree(ctx->mdsthreshold); 1045 kfree_rcu(ctx, rcu_head); 1046 } 1047 1048 void put_nfs_open_context(struct nfs_open_context *ctx) 1049 { 1050 __put_nfs_open_context(ctx, 0); 1051 } 1052 EXPORT_SYMBOL_GPL(put_nfs_open_context); 1053 1054 static void put_nfs_open_context_sync(struct nfs_open_context *ctx) 1055 { 1056 __put_nfs_open_context(ctx, 1); 1057 } 1058 1059 /* 1060 * Ensure that mmap has a recent RPC credential for use when writing out 1061 * shared pages 1062 */ 1063 void nfs_inode_attach_open_context(struct nfs_open_context *ctx) 1064 { 1065 struct inode *inode = d_inode(ctx->dentry); 1066 struct nfs_inode *nfsi = NFS_I(inode); 1067 1068 spin_lock(&inode->i_lock); 1069 if (list_empty(&nfsi->open_files) && 1070 (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER)) 1071 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA | 1072 NFS_INO_REVAL_FORCED); 1073 list_add_tail_rcu(&ctx->list, &nfsi->open_files); 1074 spin_unlock(&inode->i_lock); 1075 } 1076 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context); 1077 1078 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx) 1079 { 1080 filp->private_data = get_nfs_open_context(ctx); 1081 if (list_empty(&ctx->list)) 1082 nfs_inode_attach_open_context(ctx); 1083 } 1084 EXPORT_SYMBOL_GPL(nfs_file_set_open_context); 1085 1086 /* 1087 * Given an inode, search for an open context with the desired characteristics 1088 */ 1089 struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode) 1090 { 1091 struct nfs_inode *nfsi = NFS_I(inode); 1092 struct nfs_open_context *pos, *ctx = NULL; 1093 1094 rcu_read_lock(); 1095 list_for_each_entry_rcu(pos, &nfsi->open_files, list) { 1096 if (cred != NULL && cred_fscmp(pos->cred, cred) != 0) 1097 continue; 1098 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode) 1099 continue; 1100 ctx = get_nfs_open_context(pos); 1101 if (ctx) 1102 break; 1103 } 1104 rcu_read_unlock(); 1105 return ctx; 1106 } 1107 1108 void nfs_file_clear_open_context(struct file *filp) 1109 { 1110 struct nfs_open_context *ctx = nfs_file_open_context(filp); 1111 1112 if (ctx) { 1113 struct inode *inode = d_inode(ctx->dentry); 1114 1115 /* 1116 * We fatal error on write before. Try to writeback 1117 * every page again. 1118 */ 1119 if (ctx->error < 0) 1120 invalidate_inode_pages2(inode->i_mapping); 1121 filp->private_data = NULL; 1122 put_nfs_open_context_sync(ctx); 1123 } 1124 } 1125 1126 /* 1127 * These allocate and release file read/write context information. 1128 */ 1129 int nfs_open(struct inode *inode, struct file *filp) 1130 { 1131 struct nfs_open_context *ctx; 1132 1133 ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp); 1134 if (IS_ERR(ctx)) 1135 return PTR_ERR(ctx); 1136 nfs_file_set_open_context(filp, ctx); 1137 put_nfs_open_context(ctx); 1138 nfs_fscache_open_file(inode, filp); 1139 return 0; 1140 } 1141 EXPORT_SYMBOL_GPL(nfs_open); 1142 1143 /* 1144 * This function is called whenever some part of NFS notices that 1145 * the cached attributes have to be refreshed. 1146 */ 1147 int 1148 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) 1149 { 1150 int status = -ESTALE; 1151 struct nfs4_label *label = NULL; 1152 struct nfs_fattr *fattr = NULL; 1153 struct nfs_inode *nfsi = NFS_I(inode); 1154 1155 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n", 1156 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode)); 1157 1158 trace_nfs_revalidate_inode_enter(inode); 1159 1160 if (is_bad_inode(inode)) 1161 goto out; 1162 if (NFS_STALE(inode)) 1163 goto out; 1164 1165 /* pNFS: Attributes aren't updated until we layoutcommit */ 1166 if (S_ISREG(inode->i_mode)) { 1167 status = pnfs_sync_inode(inode, false); 1168 if (status) 1169 goto out; 1170 } 1171 1172 status = -ENOMEM; 1173 fattr = nfs_alloc_fattr(); 1174 if (fattr == NULL) 1175 goto out; 1176 1177 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE); 1178 1179 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL); 1180 if (IS_ERR(label)) { 1181 status = PTR_ERR(label); 1182 goto out; 1183 } 1184 1185 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, 1186 label, inode); 1187 if (status != 0) { 1188 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n", 1189 inode->i_sb->s_id, 1190 (unsigned long long)NFS_FILEID(inode), status); 1191 switch (status) { 1192 case -ETIMEDOUT: 1193 /* A soft timeout occurred. Use cached information? */ 1194 if (server->flags & NFS_MOUNT_SOFTREVAL) 1195 status = 0; 1196 break; 1197 case -ESTALE: 1198 if (!S_ISDIR(inode->i_mode)) 1199 nfs_set_inode_stale(inode); 1200 else 1201 nfs_zap_caches(inode); 1202 } 1203 goto err_out; 1204 } 1205 1206 status = nfs_refresh_inode(inode, fattr); 1207 if (status) { 1208 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n", 1209 inode->i_sb->s_id, 1210 (unsigned long long)NFS_FILEID(inode), status); 1211 goto err_out; 1212 } 1213 1214 if (nfsi->cache_validity & NFS_INO_INVALID_ACL) 1215 nfs_zap_acl_cache(inode); 1216 1217 nfs_setsecurity(inode, fattr, label); 1218 1219 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n", 1220 inode->i_sb->s_id, 1221 (unsigned long long)NFS_FILEID(inode)); 1222 1223 err_out: 1224 nfs4_label_free(label); 1225 out: 1226 nfs_free_fattr(fattr); 1227 trace_nfs_revalidate_inode_exit(inode, status); 1228 return status; 1229 } 1230 1231 int nfs_attribute_cache_expired(struct inode *inode) 1232 { 1233 if (nfs_have_delegated_attributes(inode)) 1234 return 0; 1235 return nfs_attribute_timeout(inode); 1236 } 1237 1238 /** 1239 * nfs_revalidate_inode - Revalidate the inode attributes 1240 * @server: pointer to nfs_server struct 1241 * @inode: pointer to inode struct 1242 * 1243 * Updates inode attribute information by retrieving the data from the server. 1244 */ 1245 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) 1246 { 1247 if (!nfs_need_revalidate_inode(inode)) 1248 return NFS_STALE(inode) ? -ESTALE : 0; 1249 return __nfs_revalidate_inode(server, inode); 1250 } 1251 EXPORT_SYMBOL_GPL(nfs_revalidate_inode); 1252 1253 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping) 1254 { 1255 int ret; 1256 1257 if (mapping->nrpages != 0) { 1258 if (S_ISREG(inode->i_mode)) { 1259 ret = nfs_sync_mapping(mapping); 1260 if (ret < 0) 1261 return ret; 1262 } 1263 ret = invalidate_inode_pages2(mapping); 1264 if (ret < 0) 1265 return ret; 1266 } 1267 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE); 1268 nfs_fscache_wait_on_invalidate(inode); 1269 1270 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n", 1271 inode->i_sb->s_id, 1272 (unsigned long long)NFS_FILEID(inode)); 1273 return 0; 1274 } 1275 1276 /** 1277 * nfs_clear_invalid_mapping - Conditionally clear a mapping 1278 * @mapping: pointer to mapping 1279 * 1280 * If the NFS_INO_INVALID_DATA inode flag is set, clear the mapping. 1281 */ 1282 int nfs_clear_invalid_mapping(struct address_space *mapping) 1283 { 1284 struct inode *inode = mapping->host; 1285 struct nfs_inode *nfsi = NFS_I(inode); 1286 unsigned long *bitlock = &nfsi->flags; 1287 int ret = 0; 1288 1289 /* 1290 * We must clear NFS_INO_INVALID_DATA first to ensure that 1291 * invalidations that come in while we're shooting down the mappings 1292 * are respected. But, that leaves a race window where one revalidator 1293 * can clear the flag, and then another checks it before the mapping 1294 * gets invalidated. Fix that by serializing access to this part of 1295 * the function. 1296 * 1297 * At the same time, we need to allow other tasks to see whether we 1298 * might be in the middle of invalidating the pages, so we only set 1299 * the bit lock here if it looks like we're going to be doing that. 1300 */ 1301 for (;;) { 1302 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING, 1303 nfs_wait_bit_killable, TASK_KILLABLE); 1304 if (ret) 1305 goto out; 1306 spin_lock(&inode->i_lock); 1307 if (test_bit(NFS_INO_INVALIDATING, bitlock)) { 1308 spin_unlock(&inode->i_lock); 1309 continue; 1310 } 1311 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 1312 break; 1313 spin_unlock(&inode->i_lock); 1314 goto out; 1315 } 1316 1317 set_bit(NFS_INO_INVALIDATING, bitlock); 1318 smp_wmb(); 1319 nfsi->cache_validity &= 1320 ~(NFS_INO_INVALID_DATA | NFS_INO_DATA_INVAL_DEFER); 1321 spin_unlock(&inode->i_lock); 1322 trace_nfs_invalidate_mapping_enter(inode); 1323 ret = nfs_invalidate_mapping(inode, mapping); 1324 trace_nfs_invalidate_mapping_exit(inode, ret); 1325 1326 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock); 1327 smp_mb__after_atomic(); 1328 wake_up_bit(bitlock, NFS_INO_INVALIDATING); 1329 out: 1330 return ret; 1331 } 1332 1333 bool nfs_mapping_need_revalidate_inode(struct inode *inode) 1334 { 1335 return nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE) || 1336 NFS_STALE(inode); 1337 } 1338 1339 int nfs_revalidate_mapping_rcu(struct inode *inode) 1340 { 1341 struct nfs_inode *nfsi = NFS_I(inode); 1342 unsigned long *bitlock = &nfsi->flags; 1343 int ret = 0; 1344 1345 if (IS_SWAPFILE(inode)) 1346 goto out; 1347 if (nfs_mapping_need_revalidate_inode(inode)) { 1348 ret = -ECHILD; 1349 goto out; 1350 } 1351 spin_lock(&inode->i_lock); 1352 if (test_bit(NFS_INO_INVALIDATING, bitlock) || 1353 (nfsi->cache_validity & NFS_INO_INVALID_DATA)) 1354 ret = -ECHILD; 1355 spin_unlock(&inode->i_lock); 1356 out: 1357 return ret; 1358 } 1359 1360 /** 1361 * nfs_revalidate_mapping - Revalidate the pagecache 1362 * @inode: pointer to host inode 1363 * @mapping: pointer to mapping 1364 */ 1365 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping) 1366 { 1367 /* swapfiles are not supposed to be shared. */ 1368 if (IS_SWAPFILE(inode)) 1369 return 0; 1370 1371 if (nfs_mapping_need_revalidate_inode(inode)) { 1372 int ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 1373 if (ret < 0) 1374 return ret; 1375 } 1376 1377 return nfs_clear_invalid_mapping(mapping); 1378 } 1379 1380 static bool nfs_file_has_writers(struct nfs_inode *nfsi) 1381 { 1382 struct inode *inode = &nfsi->vfs_inode; 1383 1384 if (!S_ISREG(inode->i_mode)) 1385 return false; 1386 if (list_empty(&nfsi->open_files)) 1387 return false; 1388 return inode_is_open_for_write(inode); 1389 } 1390 1391 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi) 1392 { 1393 return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi); 1394 } 1395 1396 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1397 { 1398 struct timespec64 ts; 1399 1400 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE) 1401 && (fattr->valid & NFS_ATTR_FATTR_CHANGE) 1402 && inode_eq_iversion_raw(inode, fattr->pre_change_attr)) { 1403 inode_set_iversion_raw(inode, fattr->change_attr); 1404 if (S_ISDIR(inode->i_mode)) 1405 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA); 1406 else if (nfs_server_capable(inode, NFS_CAP_XATTR)) 1407 nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR); 1408 } 1409 /* If we have atomic WCC data, we may update some attributes */ 1410 ts = inode->i_ctime; 1411 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME) 1412 && (fattr->valid & NFS_ATTR_FATTR_CTIME) 1413 && timespec64_equal(&ts, &fattr->pre_ctime)) { 1414 inode->i_ctime = fattr->ctime; 1415 } 1416 1417 ts = inode->i_mtime; 1418 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME) 1419 && (fattr->valid & NFS_ATTR_FATTR_MTIME) 1420 && timespec64_equal(&ts, &fattr->pre_mtime)) { 1421 inode->i_mtime = fattr->mtime; 1422 if (S_ISDIR(inode->i_mode)) 1423 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA); 1424 } 1425 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE) 1426 && (fattr->valid & NFS_ATTR_FATTR_SIZE) 1427 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size) 1428 && !nfs_have_writebacks(inode)) { 1429 i_size_write(inode, nfs_size_to_loff_t(fattr->size)); 1430 } 1431 } 1432 1433 /** 1434 * nfs_check_inode_attributes - verify consistency of the inode attribute cache 1435 * @inode: pointer to inode 1436 * @fattr: updated attributes 1437 * 1438 * Verifies the attribute cache. If we have just changed the attributes, 1439 * so that fattr carries weak cache consistency data, then it may 1440 * also update the ctime/mtime/change_attribute. 1441 */ 1442 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr) 1443 { 1444 struct nfs_inode *nfsi = NFS_I(inode); 1445 loff_t cur_size, new_isize; 1446 unsigned long invalid = 0; 1447 struct timespec64 ts; 1448 1449 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) 1450 return 0; 1451 1452 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) { 1453 /* Only a mounted-on-fileid? Just exit */ 1454 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) 1455 return 0; 1456 /* Has the inode gone and changed behind our back? */ 1457 } else if (nfsi->fileid != fattr->fileid) { 1458 /* Is this perhaps the mounted-on fileid? */ 1459 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) && 1460 nfsi->fileid == fattr->mounted_on_fileid) 1461 return 0; 1462 return -ESTALE; 1463 } 1464 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) 1465 return -ESTALE; 1466 1467 1468 if (!nfs_file_has_buffered_writers(nfsi)) { 1469 /* Verify a few of the more important attributes */ 1470 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr)) 1471 invalid |= NFS_INO_INVALID_CHANGE 1472 | NFS_INO_REVAL_PAGECACHE; 1473 1474 ts = inode->i_mtime; 1475 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime)) 1476 invalid |= NFS_INO_INVALID_MTIME; 1477 1478 ts = inode->i_ctime; 1479 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime)) 1480 invalid |= NFS_INO_INVALID_CTIME; 1481 1482 if (fattr->valid & NFS_ATTR_FATTR_SIZE) { 1483 cur_size = i_size_read(inode); 1484 new_isize = nfs_size_to_loff_t(fattr->size); 1485 if (cur_size != new_isize) 1486 invalid |= NFS_INO_INVALID_SIZE 1487 | NFS_INO_REVAL_PAGECACHE; 1488 } 1489 } 1490 1491 /* Have any file permissions changed? */ 1492 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) 1493 invalid |= NFS_INO_INVALID_ACCESS 1494 | NFS_INO_INVALID_ACL 1495 | NFS_INO_INVALID_OTHER; 1496 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid)) 1497 invalid |= NFS_INO_INVALID_ACCESS 1498 | NFS_INO_INVALID_ACL 1499 | NFS_INO_INVALID_OTHER; 1500 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid)) 1501 invalid |= NFS_INO_INVALID_ACCESS 1502 | NFS_INO_INVALID_ACL 1503 | NFS_INO_INVALID_OTHER; 1504 1505 /* Has the link count changed? */ 1506 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink) 1507 invalid |= NFS_INO_INVALID_OTHER; 1508 1509 ts = inode->i_atime; 1510 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime)) 1511 invalid |= NFS_INO_INVALID_ATIME; 1512 1513 if (invalid != 0) 1514 nfs_set_cache_invalid(inode, invalid); 1515 1516 nfsi->read_cache_jiffies = fattr->time_start; 1517 return 0; 1518 } 1519 1520 static atomic_long_t nfs_attr_generation_counter; 1521 1522 static unsigned long nfs_read_attr_generation_counter(void) 1523 { 1524 return atomic_long_read(&nfs_attr_generation_counter); 1525 } 1526 1527 unsigned long nfs_inc_attr_generation_counter(void) 1528 { 1529 return atomic_long_inc_return(&nfs_attr_generation_counter); 1530 } 1531 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter); 1532 1533 void nfs_fattr_init(struct nfs_fattr *fattr) 1534 { 1535 fattr->valid = 0; 1536 fattr->time_start = jiffies; 1537 fattr->gencount = nfs_inc_attr_generation_counter(); 1538 fattr->owner_name = NULL; 1539 fattr->group_name = NULL; 1540 } 1541 EXPORT_SYMBOL_GPL(nfs_fattr_init); 1542 1543 /** 1544 * nfs_fattr_set_barrier 1545 * @fattr: attributes 1546 * 1547 * Used to set a barrier after an attribute was updated. This 1548 * barrier ensures that older attributes from RPC calls that may 1549 * have raced with our update cannot clobber these new values. 1550 * Note that you are still responsible for ensuring that other 1551 * operations which change the attribute on the server do not 1552 * collide. 1553 */ 1554 void nfs_fattr_set_barrier(struct nfs_fattr *fattr) 1555 { 1556 fattr->gencount = nfs_inc_attr_generation_counter(); 1557 } 1558 1559 struct nfs_fattr *nfs_alloc_fattr(void) 1560 { 1561 struct nfs_fattr *fattr; 1562 1563 fattr = kmalloc(sizeof(*fattr), GFP_NOFS); 1564 if (fattr != NULL) 1565 nfs_fattr_init(fattr); 1566 return fattr; 1567 } 1568 EXPORT_SYMBOL_GPL(nfs_alloc_fattr); 1569 1570 struct nfs_fh *nfs_alloc_fhandle(void) 1571 { 1572 struct nfs_fh *fh; 1573 1574 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS); 1575 if (fh != NULL) 1576 fh->size = 0; 1577 return fh; 1578 } 1579 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle); 1580 1581 #ifdef NFS_DEBUG 1582 /* 1583 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle 1584 * in the same way that wireshark does 1585 * 1586 * @fh: file handle 1587 * 1588 * For debugging only. 1589 */ 1590 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh) 1591 { 1592 /* wireshark uses 32-bit AUTODIN crc and does a bitwise 1593 * not on the result */ 1594 return nfs_fhandle_hash(fh); 1595 } 1596 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash); 1597 1598 /* 1599 * _nfs_display_fhandle - display an NFS file handle on the console 1600 * 1601 * @fh: file handle to display 1602 * @caption: display caption 1603 * 1604 * For debugging only. 1605 */ 1606 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption) 1607 { 1608 unsigned short i; 1609 1610 if (fh == NULL || fh->size == 0) { 1611 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh); 1612 return; 1613 } 1614 1615 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n", 1616 caption, fh, fh->size, _nfs_display_fhandle_hash(fh)); 1617 for (i = 0; i < fh->size; i += 16) { 1618 __be32 *pos = (__be32 *)&fh->data[i]; 1619 1620 switch ((fh->size - i - 1) >> 2) { 1621 case 0: 1622 printk(KERN_DEFAULT " %08x\n", 1623 be32_to_cpup(pos)); 1624 break; 1625 case 1: 1626 printk(KERN_DEFAULT " %08x %08x\n", 1627 be32_to_cpup(pos), be32_to_cpup(pos + 1)); 1628 break; 1629 case 2: 1630 printk(KERN_DEFAULT " %08x %08x %08x\n", 1631 be32_to_cpup(pos), be32_to_cpup(pos + 1), 1632 be32_to_cpup(pos + 2)); 1633 break; 1634 default: 1635 printk(KERN_DEFAULT " %08x %08x %08x %08x\n", 1636 be32_to_cpup(pos), be32_to_cpup(pos + 1), 1637 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3)); 1638 } 1639 } 1640 } 1641 EXPORT_SYMBOL_GPL(_nfs_display_fhandle); 1642 #endif 1643 1644 /** 1645 * nfs_inode_attrs_need_update - check if the inode attributes need updating 1646 * @inode: pointer to inode 1647 * @fattr: attributes 1648 * 1649 * Attempt to divine whether or not an RPC call reply carrying stale 1650 * attributes got scheduled after another call carrying updated ones. 1651 * 1652 * To do so, the function first assumes that a more recent ctime means 1653 * that the attributes in fattr are newer, however it also attempt to 1654 * catch the case where ctime either didn't change, or went backwards 1655 * (if someone reset the clock on the server) by looking at whether 1656 * or not this RPC call was started after the inode was last updated. 1657 * Note also the check for wraparound of 'attr_gencount' 1658 * 1659 * The function returns 'true' if it thinks the attributes in 'fattr' are 1660 * more recent than the ones cached in the inode. 1661 * 1662 */ 1663 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr) 1664 { 1665 const struct nfs_inode *nfsi = NFS_I(inode); 1666 1667 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 || 1668 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0); 1669 } 1670 1671 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr) 1672 { 1673 int ret; 1674 1675 trace_nfs_refresh_inode_enter(inode); 1676 1677 if (nfs_inode_attrs_need_update(inode, fattr)) 1678 ret = nfs_update_inode(inode, fattr); 1679 else 1680 ret = nfs_check_inode_attributes(inode, fattr); 1681 1682 trace_nfs_refresh_inode_exit(inode, ret); 1683 return ret; 1684 } 1685 1686 /** 1687 * nfs_refresh_inode - try to update the inode attribute cache 1688 * @inode: pointer to inode 1689 * @fattr: updated attributes 1690 * 1691 * Check that an RPC call that returned attributes has not overlapped with 1692 * other recent updates of the inode metadata, then decide whether it is 1693 * safe to do a full update of the inode attributes, or whether just to 1694 * call nfs_check_inode_attributes. 1695 */ 1696 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr) 1697 { 1698 int status; 1699 1700 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 1701 return 0; 1702 spin_lock(&inode->i_lock); 1703 status = nfs_refresh_inode_locked(inode, fattr); 1704 spin_unlock(&inode->i_lock); 1705 1706 return status; 1707 } 1708 EXPORT_SYMBOL_GPL(nfs_refresh_inode); 1709 1710 static int nfs_post_op_update_inode_locked(struct inode *inode, 1711 struct nfs_fattr *fattr, unsigned int invalid) 1712 { 1713 if (S_ISDIR(inode->i_mode)) 1714 invalid |= NFS_INO_INVALID_DATA; 1715 nfs_set_cache_invalid(inode, invalid); 1716 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 1717 return 0; 1718 return nfs_refresh_inode_locked(inode, fattr); 1719 } 1720 1721 /** 1722 * nfs_post_op_update_inode - try to update the inode attribute cache 1723 * @inode: pointer to inode 1724 * @fattr: updated attributes 1725 * 1726 * After an operation that has changed the inode metadata, mark the 1727 * attribute cache as being invalid, then try to update it. 1728 * 1729 * NB: if the server didn't return any post op attributes, this 1730 * function will force the retrieval of attributes before the next 1731 * NFS request. Thus it should be used only for operations that 1732 * are expected to change one or more attributes, to avoid 1733 * unnecessary NFS requests and trips through nfs_update_inode(). 1734 */ 1735 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1736 { 1737 int status; 1738 1739 spin_lock(&inode->i_lock); 1740 nfs_fattr_set_barrier(fattr); 1741 status = nfs_post_op_update_inode_locked(inode, fattr, 1742 NFS_INO_INVALID_CHANGE 1743 | NFS_INO_INVALID_CTIME 1744 | NFS_INO_REVAL_FORCED); 1745 spin_unlock(&inode->i_lock); 1746 1747 return status; 1748 } 1749 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode); 1750 1751 /** 1752 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache 1753 * @inode: pointer to inode 1754 * @fattr: updated attributes 1755 * 1756 * After an operation that has changed the inode metadata, mark the 1757 * attribute cache as being invalid, then try to update it. Fake up 1758 * weak cache consistency data, if none exist. 1759 * 1760 * This function is mainly designed to be used by the ->write_done() functions. 1761 */ 1762 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr) 1763 { 1764 int status; 1765 1766 /* Don't do a WCC update if these attributes are already stale */ 1767 if ((fattr->valid & NFS_ATTR_FATTR) == 0 || 1768 !nfs_inode_attrs_need_update(inode, fattr)) { 1769 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE 1770 | NFS_ATTR_FATTR_PRESIZE 1771 | NFS_ATTR_FATTR_PREMTIME 1772 | NFS_ATTR_FATTR_PRECTIME); 1773 goto out_noforce; 1774 } 1775 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && 1776 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) { 1777 fattr->pre_change_attr = inode_peek_iversion_raw(inode); 1778 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE; 1779 } 1780 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 && 1781 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) { 1782 fattr->pre_ctime = inode->i_ctime; 1783 fattr->valid |= NFS_ATTR_FATTR_PRECTIME; 1784 } 1785 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 && 1786 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) { 1787 fattr->pre_mtime = inode->i_mtime; 1788 fattr->valid |= NFS_ATTR_FATTR_PREMTIME; 1789 } 1790 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 && 1791 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) { 1792 fattr->pre_size = i_size_read(inode); 1793 fattr->valid |= NFS_ATTR_FATTR_PRESIZE; 1794 } 1795 out_noforce: 1796 status = nfs_post_op_update_inode_locked(inode, fattr, 1797 NFS_INO_INVALID_CHANGE 1798 | NFS_INO_INVALID_CTIME 1799 | NFS_INO_INVALID_MTIME 1800 | NFS_INO_INVALID_BLOCKS); 1801 return status; 1802 } 1803 1804 /** 1805 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache 1806 * @inode: pointer to inode 1807 * @fattr: updated attributes 1808 * 1809 * After an operation that has changed the inode metadata, mark the 1810 * attribute cache as being invalid, then try to update it. Fake up 1811 * weak cache consistency data, if none exist. 1812 * 1813 * This function is mainly designed to be used by the ->write_done() functions. 1814 */ 1815 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr) 1816 { 1817 int status; 1818 1819 spin_lock(&inode->i_lock); 1820 nfs_fattr_set_barrier(fattr); 1821 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 1822 spin_unlock(&inode->i_lock); 1823 return status; 1824 } 1825 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc); 1826 1827 1828 /* 1829 * Many nfs protocol calls return the new file attributes after 1830 * an operation. Here we update the inode to reflect the state 1831 * of the server's inode. 1832 * 1833 * This is a bit tricky because we have to make sure all dirty pages 1834 * have been sent off to the server before calling invalidate_inode_pages. 1835 * To make sure no other process adds more write requests while we try 1836 * our best to flush them, we make them sleep during the attribute refresh. 1837 * 1838 * A very similar scenario holds for the dir cache. 1839 */ 1840 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1841 { 1842 struct nfs_server *server; 1843 struct nfs_inode *nfsi = NFS_I(inode); 1844 loff_t cur_isize, new_isize; 1845 unsigned long invalid = 0; 1846 unsigned long now = jiffies; 1847 unsigned long save_cache_validity; 1848 bool have_writers = nfs_file_has_buffered_writers(nfsi); 1849 bool cache_revalidated = true; 1850 bool attr_changed = false; 1851 bool have_delegation; 1852 1853 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n", 1854 __func__, inode->i_sb->s_id, inode->i_ino, 1855 nfs_display_fhandle_hash(NFS_FH(inode)), 1856 atomic_read(&inode->i_count), fattr->valid); 1857 1858 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) { 1859 /* Only a mounted-on-fileid? Just exit */ 1860 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) 1861 return 0; 1862 /* Has the inode gone and changed behind our back? */ 1863 } else if (nfsi->fileid != fattr->fileid) { 1864 /* Is this perhaps the mounted-on fileid? */ 1865 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) && 1866 nfsi->fileid == fattr->mounted_on_fileid) 1867 return 0; 1868 printk(KERN_ERR "NFS: server %s error: fileid changed\n" 1869 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n", 1870 NFS_SERVER(inode)->nfs_client->cl_hostname, 1871 inode->i_sb->s_id, (long long)nfsi->fileid, 1872 (long long)fattr->fileid); 1873 goto out_err; 1874 } 1875 1876 /* 1877 * Make sure the inode's type hasn't changed. 1878 */ 1879 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) { 1880 /* 1881 * Big trouble! The inode has become a different object. 1882 */ 1883 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n", 1884 __func__, inode->i_ino, inode->i_mode, fattr->mode); 1885 goto out_err; 1886 } 1887 1888 server = NFS_SERVER(inode); 1889 /* Update the fsid? */ 1890 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) && 1891 !nfs_fsid_equal(&server->fsid, &fattr->fsid) && 1892 !IS_AUTOMOUNT(inode)) 1893 server->fsid = fattr->fsid; 1894 1895 /* Save the delegation state before clearing cache_validity */ 1896 have_delegation = nfs_have_delegated_attributes(inode); 1897 1898 /* 1899 * Update the read time so we don't revalidate too often. 1900 */ 1901 nfsi->read_cache_jiffies = fattr->time_start; 1902 1903 save_cache_validity = nfsi->cache_validity; 1904 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR 1905 | NFS_INO_INVALID_ATIME 1906 | NFS_INO_REVAL_FORCED 1907 | NFS_INO_REVAL_PAGECACHE 1908 | NFS_INO_INVALID_BLOCKS); 1909 1910 /* Do atomic weak cache consistency updates */ 1911 nfs_wcc_update_inode(inode, fattr); 1912 1913 if (pnfs_layoutcommit_outstanding(inode)) { 1914 nfsi->cache_validity |= save_cache_validity & NFS_INO_INVALID_ATTR; 1915 cache_revalidated = false; 1916 } 1917 1918 /* More cache consistency checks */ 1919 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) { 1920 if (!inode_eq_iversion_raw(inode, fattr->change_attr)) { 1921 /* Could it be a race with writeback? */ 1922 if (!(have_writers || have_delegation)) { 1923 invalid |= NFS_INO_INVALID_DATA 1924 | NFS_INO_INVALID_ACCESS 1925 | NFS_INO_INVALID_ACL 1926 | NFS_INO_INVALID_XATTR; 1927 /* Force revalidate of all attributes */ 1928 save_cache_validity |= NFS_INO_INVALID_CTIME 1929 | NFS_INO_INVALID_MTIME 1930 | NFS_INO_INVALID_SIZE 1931 | NFS_INO_INVALID_OTHER; 1932 if (S_ISDIR(inode->i_mode)) 1933 nfs_force_lookup_revalidate(inode); 1934 dprintk("NFS: change_attr change on server for file %s/%ld\n", 1935 inode->i_sb->s_id, 1936 inode->i_ino); 1937 } else if (!have_delegation) 1938 nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER; 1939 inode_set_iversion_raw(inode, fattr->change_attr); 1940 attr_changed = true; 1941 } 1942 } else { 1943 nfsi->cache_validity |= save_cache_validity & 1944 (NFS_INO_INVALID_CHANGE 1945 | NFS_INO_REVAL_PAGECACHE 1946 | NFS_INO_REVAL_FORCED); 1947 cache_revalidated = false; 1948 } 1949 1950 if (fattr->valid & NFS_ATTR_FATTR_MTIME) { 1951 inode->i_mtime = fattr->mtime; 1952 } else if (server->caps & NFS_CAP_MTIME) { 1953 nfsi->cache_validity |= save_cache_validity & 1954 (NFS_INO_INVALID_MTIME 1955 | NFS_INO_REVAL_FORCED); 1956 cache_revalidated = false; 1957 } 1958 1959 if (fattr->valid & NFS_ATTR_FATTR_CTIME) { 1960 inode->i_ctime = fattr->ctime; 1961 } else if (server->caps & NFS_CAP_CTIME) { 1962 nfsi->cache_validity |= save_cache_validity & 1963 (NFS_INO_INVALID_CTIME 1964 | NFS_INO_REVAL_FORCED); 1965 cache_revalidated = false; 1966 } 1967 1968 /* Check if our cached file size is stale */ 1969 if (fattr->valid & NFS_ATTR_FATTR_SIZE) { 1970 new_isize = nfs_size_to_loff_t(fattr->size); 1971 cur_isize = i_size_read(inode); 1972 if (new_isize != cur_isize && !have_delegation) { 1973 /* Do we perhaps have any outstanding writes, or has 1974 * the file grown beyond our last write? */ 1975 if (!nfs_have_writebacks(inode) || new_isize > cur_isize) { 1976 i_size_write(inode, new_isize); 1977 if (!have_writers) 1978 invalid |= NFS_INO_INVALID_DATA; 1979 attr_changed = true; 1980 } 1981 dprintk("NFS: isize change on server for file %s/%ld " 1982 "(%Ld to %Ld)\n", 1983 inode->i_sb->s_id, 1984 inode->i_ino, 1985 (long long)cur_isize, 1986 (long long)new_isize); 1987 } 1988 } else { 1989 nfsi->cache_validity |= save_cache_validity & 1990 (NFS_INO_INVALID_SIZE 1991 | NFS_INO_REVAL_PAGECACHE 1992 | NFS_INO_REVAL_FORCED); 1993 cache_revalidated = false; 1994 } 1995 1996 1997 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 1998 inode->i_atime = fattr->atime; 1999 else if (server->caps & NFS_CAP_ATIME) { 2000 nfsi->cache_validity |= save_cache_validity & 2001 (NFS_INO_INVALID_ATIME 2002 | NFS_INO_REVAL_FORCED); 2003 cache_revalidated = false; 2004 } 2005 2006 if (fattr->valid & NFS_ATTR_FATTR_MODE) { 2007 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) { 2008 umode_t newmode = inode->i_mode & S_IFMT; 2009 newmode |= fattr->mode & S_IALLUGO; 2010 inode->i_mode = newmode; 2011 invalid |= NFS_INO_INVALID_ACCESS 2012 | NFS_INO_INVALID_ACL; 2013 attr_changed = true; 2014 } 2015 } else if (server->caps & NFS_CAP_MODE) { 2016 nfsi->cache_validity |= save_cache_validity & 2017 (NFS_INO_INVALID_OTHER 2018 | NFS_INO_REVAL_FORCED); 2019 cache_revalidated = false; 2020 } 2021 2022 if (fattr->valid & NFS_ATTR_FATTR_OWNER) { 2023 if (!uid_eq(inode->i_uid, fattr->uid)) { 2024 invalid |= NFS_INO_INVALID_ACCESS 2025 | NFS_INO_INVALID_ACL; 2026 inode->i_uid = fattr->uid; 2027 attr_changed = true; 2028 } 2029 } else if (server->caps & NFS_CAP_OWNER) { 2030 nfsi->cache_validity |= save_cache_validity & 2031 (NFS_INO_INVALID_OTHER 2032 | NFS_INO_REVAL_FORCED); 2033 cache_revalidated = false; 2034 } 2035 2036 if (fattr->valid & NFS_ATTR_FATTR_GROUP) { 2037 if (!gid_eq(inode->i_gid, fattr->gid)) { 2038 invalid |= NFS_INO_INVALID_ACCESS 2039 | NFS_INO_INVALID_ACL; 2040 inode->i_gid = fattr->gid; 2041 attr_changed = true; 2042 } 2043 } else if (server->caps & NFS_CAP_OWNER_GROUP) { 2044 nfsi->cache_validity |= save_cache_validity & 2045 (NFS_INO_INVALID_OTHER 2046 | NFS_INO_REVAL_FORCED); 2047 cache_revalidated = false; 2048 } 2049 2050 if (fattr->valid & NFS_ATTR_FATTR_NLINK) { 2051 if (inode->i_nlink != fattr->nlink) { 2052 if (S_ISDIR(inode->i_mode)) 2053 invalid |= NFS_INO_INVALID_DATA; 2054 set_nlink(inode, fattr->nlink); 2055 attr_changed = true; 2056 } 2057 } else if (server->caps & NFS_CAP_NLINK) { 2058 nfsi->cache_validity |= save_cache_validity & 2059 (NFS_INO_INVALID_OTHER 2060 | NFS_INO_REVAL_FORCED); 2061 cache_revalidated = false; 2062 } 2063 2064 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) { 2065 /* 2066 * report the blocks in 512byte units 2067 */ 2068 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 2069 } else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED) 2070 inode->i_blocks = fattr->du.nfs2.blocks; 2071 else { 2072 nfsi->cache_validity |= save_cache_validity & 2073 (NFS_INO_INVALID_BLOCKS 2074 | NFS_INO_REVAL_FORCED); 2075 cache_revalidated = false; 2076 } 2077 2078 /* Update attrtimeo value if we're out of the unstable period */ 2079 if (attr_changed) { 2080 invalid &= ~NFS_INO_INVALID_ATTR; 2081 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 2082 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 2083 nfsi->attrtimeo_timestamp = now; 2084 /* Set barrier to be more recent than all outstanding updates */ 2085 nfsi->attr_gencount = nfs_inc_attr_generation_counter(); 2086 } else { 2087 if (cache_revalidated) { 2088 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, 2089 nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) { 2090 nfsi->attrtimeo <<= 1; 2091 if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode)) 2092 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode); 2093 } 2094 nfsi->attrtimeo_timestamp = now; 2095 } 2096 /* Set the barrier to be more recent than this fattr */ 2097 if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0) 2098 nfsi->attr_gencount = fattr->gencount; 2099 } 2100 2101 /* Don't invalidate the data if we were to blame */ 2102 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) 2103 || S_ISLNK(inode->i_mode))) 2104 invalid &= ~NFS_INO_INVALID_DATA; 2105 nfs_set_cache_invalid(inode, invalid); 2106 2107 return 0; 2108 out_err: 2109 /* 2110 * No need to worry about unhashing the dentry, as the 2111 * lookup validation will know that the inode is bad. 2112 * (But we fall through to invalidate the caches.) 2113 */ 2114 nfs_set_inode_stale_locked(inode); 2115 return -ESTALE; 2116 } 2117 2118 struct inode *nfs_alloc_inode(struct super_block *sb) 2119 { 2120 struct nfs_inode *nfsi; 2121 nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL); 2122 if (!nfsi) 2123 return NULL; 2124 nfsi->flags = 0UL; 2125 nfsi->cache_validity = 0UL; 2126 #if IS_ENABLED(CONFIG_NFS_V4) 2127 nfsi->nfs4_acl = NULL; 2128 #endif /* CONFIG_NFS_V4 */ 2129 #ifdef CONFIG_NFS_V4_2 2130 nfsi->xattr_cache = NULL; 2131 #endif 2132 return &nfsi->vfs_inode; 2133 } 2134 EXPORT_SYMBOL_GPL(nfs_alloc_inode); 2135 2136 void nfs_free_inode(struct inode *inode) 2137 { 2138 kmem_cache_free(nfs_inode_cachep, NFS_I(inode)); 2139 } 2140 EXPORT_SYMBOL_GPL(nfs_free_inode); 2141 2142 static inline void nfs4_init_once(struct nfs_inode *nfsi) 2143 { 2144 #if IS_ENABLED(CONFIG_NFS_V4) 2145 INIT_LIST_HEAD(&nfsi->open_states); 2146 nfsi->delegation = NULL; 2147 init_rwsem(&nfsi->rwsem); 2148 nfsi->layout = NULL; 2149 #endif 2150 } 2151 2152 static void init_once(void *foo) 2153 { 2154 struct nfs_inode *nfsi = (struct nfs_inode *) foo; 2155 2156 inode_init_once(&nfsi->vfs_inode); 2157 INIT_LIST_HEAD(&nfsi->open_files); 2158 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru); 2159 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru); 2160 INIT_LIST_HEAD(&nfsi->commit_info.list); 2161 atomic_long_set(&nfsi->nrequests, 0); 2162 atomic_long_set(&nfsi->commit_info.ncommit, 0); 2163 atomic_set(&nfsi->commit_info.rpcs_out, 0); 2164 init_rwsem(&nfsi->rmdir_sem); 2165 mutex_init(&nfsi->commit_mutex); 2166 nfs4_init_once(nfsi); 2167 nfsi->cache_change_attribute = 0; 2168 } 2169 2170 static int __init nfs_init_inodecache(void) 2171 { 2172 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache", 2173 sizeof(struct nfs_inode), 2174 0, (SLAB_RECLAIM_ACCOUNT| 2175 SLAB_MEM_SPREAD|SLAB_ACCOUNT), 2176 init_once); 2177 if (nfs_inode_cachep == NULL) 2178 return -ENOMEM; 2179 2180 return 0; 2181 } 2182 2183 static void nfs_destroy_inodecache(void) 2184 { 2185 /* 2186 * Make sure all delayed rcu free inodes are flushed before we 2187 * destroy cache. 2188 */ 2189 rcu_barrier(); 2190 kmem_cache_destroy(nfs_inode_cachep); 2191 } 2192 2193 struct workqueue_struct *nfsiod_workqueue; 2194 EXPORT_SYMBOL_GPL(nfsiod_workqueue); 2195 2196 /* 2197 * start up the nfsiod workqueue 2198 */ 2199 static int nfsiod_start(void) 2200 { 2201 struct workqueue_struct *wq; 2202 dprintk("RPC: creating workqueue nfsiod\n"); 2203 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0); 2204 if (wq == NULL) 2205 return -ENOMEM; 2206 nfsiod_workqueue = wq; 2207 return 0; 2208 } 2209 2210 /* 2211 * Destroy the nfsiod workqueue 2212 */ 2213 static void nfsiod_stop(void) 2214 { 2215 struct workqueue_struct *wq; 2216 2217 wq = nfsiod_workqueue; 2218 if (wq == NULL) 2219 return; 2220 nfsiod_workqueue = NULL; 2221 destroy_workqueue(wq); 2222 } 2223 2224 unsigned int nfs_net_id; 2225 EXPORT_SYMBOL_GPL(nfs_net_id); 2226 2227 static int nfs_net_init(struct net *net) 2228 { 2229 nfs_clients_init(net); 2230 return nfs_fs_proc_net_init(net); 2231 } 2232 2233 static void nfs_net_exit(struct net *net) 2234 { 2235 nfs_fs_proc_net_exit(net); 2236 nfs_clients_exit(net); 2237 } 2238 2239 static struct pernet_operations nfs_net_ops = { 2240 .init = nfs_net_init, 2241 .exit = nfs_net_exit, 2242 .id = &nfs_net_id, 2243 .size = sizeof(struct nfs_net), 2244 }; 2245 2246 /* 2247 * Initialize NFS 2248 */ 2249 static int __init init_nfs_fs(void) 2250 { 2251 int err; 2252 2253 err = nfs_sysfs_init(); 2254 if (err < 0) 2255 goto out10; 2256 2257 err = register_pernet_subsys(&nfs_net_ops); 2258 if (err < 0) 2259 goto out9; 2260 2261 err = nfs_fscache_register(); 2262 if (err < 0) 2263 goto out8; 2264 2265 err = nfsiod_start(); 2266 if (err) 2267 goto out7; 2268 2269 err = nfs_fs_proc_init(); 2270 if (err) 2271 goto out6; 2272 2273 err = nfs_init_nfspagecache(); 2274 if (err) 2275 goto out5; 2276 2277 err = nfs_init_inodecache(); 2278 if (err) 2279 goto out4; 2280 2281 err = nfs_init_readpagecache(); 2282 if (err) 2283 goto out3; 2284 2285 err = nfs_init_writepagecache(); 2286 if (err) 2287 goto out2; 2288 2289 err = nfs_init_directcache(); 2290 if (err) 2291 goto out1; 2292 2293 rpc_proc_register(&init_net, &nfs_rpcstat); 2294 2295 err = register_nfs_fs(); 2296 if (err) 2297 goto out0; 2298 2299 return 0; 2300 out0: 2301 rpc_proc_unregister(&init_net, "nfs"); 2302 nfs_destroy_directcache(); 2303 out1: 2304 nfs_destroy_writepagecache(); 2305 out2: 2306 nfs_destroy_readpagecache(); 2307 out3: 2308 nfs_destroy_inodecache(); 2309 out4: 2310 nfs_destroy_nfspagecache(); 2311 out5: 2312 nfs_fs_proc_exit(); 2313 out6: 2314 nfsiod_stop(); 2315 out7: 2316 nfs_fscache_unregister(); 2317 out8: 2318 unregister_pernet_subsys(&nfs_net_ops); 2319 out9: 2320 nfs_sysfs_exit(); 2321 out10: 2322 return err; 2323 } 2324 2325 static void __exit exit_nfs_fs(void) 2326 { 2327 nfs_destroy_directcache(); 2328 nfs_destroy_writepagecache(); 2329 nfs_destroy_readpagecache(); 2330 nfs_destroy_inodecache(); 2331 nfs_destroy_nfspagecache(); 2332 nfs_fscache_unregister(); 2333 unregister_pernet_subsys(&nfs_net_ops); 2334 rpc_proc_unregister(&init_net, "nfs"); 2335 unregister_nfs_fs(); 2336 nfs_fs_proc_exit(); 2337 nfsiod_stop(); 2338 nfs_sysfs_exit(); 2339 } 2340 2341 /* Not quite true; I just maintain it */ 2342 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>"); 2343 MODULE_LICENSE("GPL"); 2344 module_param(enable_ino64, bool, 0644); 2345 2346 module_init(init_nfs_fs) 2347 module_exit(exit_nfs_fs) 2348