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 static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags) 199 { 200 struct nfs_inode *nfsi = NFS_I(inode); 201 bool have_delegation = NFS_PROTO(inode)->have_delegation(inode, FMODE_READ); 202 203 if (have_delegation) { 204 if (!(flags & NFS_INO_REVAL_FORCED)) 205 flags &= ~NFS_INO_INVALID_OTHER; 206 flags &= ~(NFS_INO_INVALID_CHANGE 207 | NFS_INO_INVALID_SIZE 208 | NFS_INO_REVAL_PAGECACHE 209 | NFS_INO_INVALID_XATTR); 210 } 211 212 if (inode->i_mapping->nrpages == 0) 213 flags &= ~(NFS_INO_INVALID_DATA|NFS_INO_DATA_INVAL_DEFER); 214 nfsi->cache_validity |= flags; 215 if (flags & NFS_INO_INVALID_DATA) 216 nfs_fscache_invalidate(inode); 217 } 218 219 /* 220 * Invalidate the local caches 221 */ 222 static void nfs_zap_caches_locked(struct inode *inode) 223 { 224 struct nfs_inode *nfsi = NFS_I(inode); 225 int mode = inode->i_mode; 226 227 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 228 229 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 230 nfsi->attrtimeo_timestamp = jiffies; 231 232 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) { 233 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR 234 | NFS_INO_INVALID_DATA 235 | NFS_INO_INVALID_ACCESS 236 | NFS_INO_INVALID_ACL 237 | NFS_INO_INVALID_XATTR 238 | NFS_INO_REVAL_PAGECACHE); 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_INO_REVAL_PAGECACHE); 245 nfs_zap_label_cache_locked(nfsi); 246 } 247 248 void nfs_zap_caches(struct inode *inode) 249 { 250 spin_lock(&inode->i_lock); 251 nfs_zap_caches_locked(inode); 252 spin_unlock(&inode->i_lock); 253 } 254 255 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping) 256 { 257 if (mapping->nrpages != 0) { 258 spin_lock(&inode->i_lock); 259 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA); 260 spin_unlock(&inode->i_lock); 261 } 262 } 263 264 void nfs_zap_acl_cache(struct inode *inode) 265 { 266 void (*clear_acl_cache)(struct inode *); 267 268 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache; 269 if (clear_acl_cache != NULL) 270 clear_acl_cache(inode); 271 spin_lock(&inode->i_lock); 272 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL; 273 spin_unlock(&inode->i_lock); 274 } 275 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache); 276 277 void nfs_invalidate_atime(struct inode *inode) 278 { 279 spin_lock(&inode->i_lock); 280 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME); 281 spin_unlock(&inode->i_lock); 282 } 283 EXPORT_SYMBOL_GPL(nfs_invalidate_atime); 284 285 /* 286 * Invalidate, but do not unhash, the inode. 287 * NB: must be called with inode->i_lock held! 288 */ 289 static void nfs_set_inode_stale_locked(struct inode *inode) 290 { 291 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); 292 nfs_zap_caches_locked(inode); 293 trace_nfs_set_inode_stale(inode); 294 } 295 296 void nfs_set_inode_stale(struct inode *inode) 297 { 298 spin_lock(&inode->i_lock); 299 nfs_set_inode_stale_locked(inode); 300 spin_unlock(&inode->i_lock); 301 } 302 303 struct nfs_find_desc { 304 struct nfs_fh *fh; 305 struct nfs_fattr *fattr; 306 }; 307 308 /* 309 * In NFSv3 we can have 64bit inode numbers. In order to support 310 * this, and re-exported directories (also seen in NFSv2) 311 * we are forced to allow 2 different inodes to have the same 312 * i_ino. 313 */ 314 static int 315 nfs_find_actor(struct inode *inode, void *opaque) 316 { 317 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque; 318 struct nfs_fh *fh = desc->fh; 319 struct nfs_fattr *fattr = desc->fattr; 320 321 if (NFS_FILEID(inode) != fattr->fileid) 322 return 0; 323 if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode)) 324 return 0; 325 if (nfs_compare_fh(NFS_FH(inode), fh)) 326 return 0; 327 if (is_bad_inode(inode) || NFS_STALE(inode)) 328 return 0; 329 return 1; 330 } 331 332 static int 333 nfs_init_locked(struct inode *inode, void *opaque) 334 { 335 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque; 336 struct nfs_fattr *fattr = desc->fattr; 337 338 set_nfs_fileid(inode, fattr->fileid); 339 inode->i_mode = fattr->mode; 340 nfs_copy_fh(NFS_FH(inode), desc->fh); 341 return 0; 342 } 343 344 #ifdef CONFIG_NFS_V4_SECURITY_LABEL 345 static void nfs_clear_label_invalid(struct inode *inode) 346 { 347 spin_lock(&inode->i_lock); 348 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL; 349 spin_unlock(&inode->i_lock); 350 } 351 352 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr, 353 struct nfs4_label *label) 354 { 355 int error; 356 357 if (label == NULL) 358 return; 359 360 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) { 361 error = security_inode_notifysecctx(inode, label->label, 362 label->len); 363 if (error) 364 printk(KERN_ERR "%s() %s %d " 365 "security_inode_notifysecctx() %d\n", 366 __func__, 367 (char *)label->label, 368 label->len, error); 369 nfs_clear_label_invalid(inode); 370 } 371 } 372 373 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags) 374 { 375 struct nfs4_label *label = NULL; 376 int minor_version = server->nfs_client->cl_minorversion; 377 378 if (minor_version < 2) 379 return label; 380 381 if (!(server->caps & NFS_CAP_SECURITY_LABEL)) 382 return label; 383 384 label = kzalloc(sizeof(struct nfs4_label), flags); 385 if (label == NULL) 386 return ERR_PTR(-ENOMEM); 387 388 label->label = kzalloc(NFS4_MAXLABELLEN, flags); 389 if (label->label == NULL) { 390 kfree(label); 391 return ERR_PTR(-ENOMEM); 392 } 393 label->len = NFS4_MAXLABELLEN; 394 395 return label; 396 } 397 EXPORT_SYMBOL_GPL(nfs4_label_alloc); 398 #else 399 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr, 400 struct nfs4_label *label) 401 { 402 } 403 #endif 404 EXPORT_SYMBOL_GPL(nfs_setsecurity); 405 406 /* Search for inode identified by fh, fileid and i_mode in inode cache. */ 407 struct inode * 408 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh) 409 { 410 struct nfs_find_desc desc = { 411 .fh = fh, 412 .fattr = fattr, 413 }; 414 struct inode *inode; 415 unsigned long hash; 416 417 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) || 418 !(fattr->valid & NFS_ATTR_FATTR_TYPE)) 419 return NULL; 420 421 hash = nfs_fattr_to_ino_t(fattr); 422 inode = ilookup5(sb, hash, nfs_find_actor, &desc); 423 424 dprintk("%s: returning %p\n", __func__, inode); 425 return inode; 426 } 427 428 /* 429 * This is our front-end to iget that looks up inodes by file handle 430 * instead of inode number. 431 */ 432 struct inode * 433 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label) 434 { 435 struct nfs_find_desc desc = { 436 .fh = fh, 437 .fattr = fattr 438 }; 439 struct inode *inode = ERR_PTR(-ENOENT); 440 unsigned long hash; 441 442 nfs_attr_check_mountpoint(sb, fattr); 443 444 if (nfs_attr_use_mounted_on_fileid(fattr)) 445 fattr->fileid = fattr->mounted_on_fileid; 446 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0) 447 goto out_no_inode; 448 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0) 449 goto out_no_inode; 450 451 hash = nfs_fattr_to_ino_t(fattr); 452 453 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc); 454 if (inode == NULL) { 455 inode = ERR_PTR(-ENOMEM); 456 goto out_no_inode; 457 } 458 459 if (inode->i_state & I_NEW) { 460 struct nfs_inode *nfsi = NFS_I(inode); 461 unsigned long now = jiffies; 462 463 /* We set i_ino for the few things that still rely on it, 464 * such as stat(2) */ 465 inode->i_ino = hash; 466 467 /* We can't support update_atime(), since the server will reset it */ 468 inode->i_flags |= S_NOATIME|S_NOCMTIME; 469 inode->i_mode = fattr->mode; 470 nfsi->cache_validity = 0; 471 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0 472 && nfs_server_capable(inode, NFS_CAP_MODE)) 473 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER); 474 /* Why so? Because we want revalidate for devices/FIFOs, and 475 * that's precisely what we have in nfs_file_inode_operations. 476 */ 477 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops; 478 if (S_ISREG(inode->i_mode)) { 479 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops; 480 inode->i_data.a_ops = &nfs_file_aops; 481 } else if (S_ISDIR(inode->i_mode)) { 482 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops; 483 inode->i_fop = &nfs_dir_operations; 484 inode->i_data.a_ops = &nfs_dir_aops; 485 /* Deal with crossing mountpoints */ 486 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT || 487 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) { 488 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) 489 inode->i_op = &nfs_referral_inode_operations; 490 else 491 inode->i_op = &nfs_mountpoint_inode_operations; 492 inode->i_fop = NULL; 493 inode->i_flags |= S_AUTOMOUNT; 494 } 495 } else if (S_ISLNK(inode->i_mode)) { 496 inode->i_op = &nfs_symlink_inode_operations; 497 inode_nohighmem(inode); 498 } else 499 init_special_inode(inode, inode->i_mode, fattr->rdev); 500 501 memset(&inode->i_atime, 0, sizeof(inode->i_atime)); 502 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime)); 503 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime)); 504 inode_set_iversion_raw(inode, 0); 505 inode->i_size = 0; 506 clear_nlink(inode); 507 inode->i_uid = make_kuid(&init_user_ns, -2); 508 inode->i_gid = make_kgid(&init_user_ns, -2); 509 inode->i_blocks = 0; 510 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf)); 511 nfsi->write_io = 0; 512 nfsi->read_io = 0; 513 514 nfsi->read_cache_jiffies = fattr->time_start; 515 nfsi->attr_gencount = fattr->gencount; 516 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 517 inode->i_atime = fattr->atime; 518 else if (nfs_server_capable(inode, NFS_CAP_ATIME)) 519 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME); 520 if (fattr->valid & NFS_ATTR_FATTR_MTIME) 521 inode->i_mtime = fattr->mtime; 522 else if (nfs_server_capable(inode, NFS_CAP_MTIME)) 523 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME); 524 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 525 inode->i_ctime = fattr->ctime; 526 else if (nfs_server_capable(inode, NFS_CAP_CTIME)) 527 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME); 528 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) 529 inode_set_iversion_raw(inode, fattr->change_attr); 530 else 531 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE); 532 if (fattr->valid & NFS_ATTR_FATTR_SIZE) 533 inode->i_size = nfs_size_to_loff_t(fattr->size); 534 else 535 nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE); 536 if (fattr->valid & NFS_ATTR_FATTR_NLINK) 537 set_nlink(inode, fattr->nlink); 538 else if (nfs_server_capable(inode, NFS_CAP_NLINK)) 539 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER); 540 if (fattr->valid & NFS_ATTR_FATTR_OWNER) 541 inode->i_uid = fattr->uid; 542 else if (nfs_server_capable(inode, NFS_CAP_OWNER)) 543 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER); 544 if (fattr->valid & NFS_ATTR_FATTR_GROUP) 545 inode->i_gid = fattr->gid; 546 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP)) 547 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER); 548 if (nfs_server_capable(inode, NFS_CAP_XATTR)) 549 nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR); 550 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED) 551 inode->i_blocks = fattr->du.nfs2.blocks; 552 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) { 553 /* 554 * report the blocks in 512byte units 555 */ 556 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 557 } 558 559 if (nfsi->cache_validity != 0) 560 nfsi->cache_validity |= NFS_INO_REVAL_FORCED; 561 562 nfs_setsecurity(inode, fattr, label); 563 564 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 565 nfsi->attrtimeo_timestamp = now; 566 nfsi->access_cache = RB_ROOT; 567 568 nfs_fscache_init_inode(inode); 569 570 unlock_new_inode(inode); 571 } else { 572 int err = nfs_refresh_inode(inode, fattr); 573 if (err < 0) { 574 iput(inode); 575 inode = ERR_PTR(err); 576 goto out_no_inode; 577 } 578 } 579 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n", 580 inode->i_sb->s_id, 581 (unsigned long long)NFS_FILEID(inode), 582 nfs_display_fhandle_hash(fh), 583 atomic_read(&inode->i_count)); 584 585 out: 586 return inode; 587 588 out_no_inode: 589 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode)); 590 goto out; 591 } 592 EXPORT_SYMBOL_GPL(nfs_fhget); 593 594 #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) 595 596 int 597 nfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry, 598 struct iattr *attr) 599 { 600 struct inode *inode = d_inode(dentry); 601 struct nfs_fattr *fattr; 602 int error = 0; 603 604 nfs_inc_stats(inode, NFSIOS_VFSSETATTR); 605 606 /* skip mode change if it's just for clearing setuid/setgid */ 607 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID)) 608 attr->ia_valid &= ~ATTR_MODE; 609 610 if (attr->ia_valid & ATTR_SIZE) { 611 BUG_ON(!S_ISREG(inode->i_mode)); 612 613 error = inode_newsize_ok(inode, attr->ia_size); 614 if (error) 615 return error; 616 617 if (attr->ia_size == i_size_read(inode)) 618 attr->ia_valid &= ~ATTR_SIZE; 619 } 620 621 /* Optimization: if the end result is no change, don't RPC */ 622 attr->ia_valid &= NFS_VALID_ATTRS; 623 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0) 624 return 0; 625 626 trace_nfs_setattr_enter(inode); 627 628 /* Write all dirty data */ 629 if (S_ISREG(inode->i_mode)) 630 nfs_sync_inode(inode); 631 632 fattr = nfs_alloc_fattr(); 633 if (fattr == NULL) { 634 error = -ENOMEM; 635 goto out; 636 } 637 638 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr); 639 if (error == 0) 640 error = nfs_refresh_inode(inode, fattr); 641 nfs_free_fattr(fattr); 642 out: 643 trace_nfs_setattr_exit(inode, error); 644 return error; 645 } 646 EXPORT_SYMBOL_GPL(nfs_setattr); 647 648 /** 649 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall 650 * @inode: inode of the file used 651 * @offset: file offset to start truncating 652 * 653 * This is a copy of the common vmtruncate, but with the locking 654 * corrected to take into account the fact that NFS requires 655 * inode->i_size to be updated under the inode->i_lock. 656 * Note: must be called with inode->i_lock held! 657 */ 658 static int nfs_vmtruncate(struct inode * inode, loff_t offset) 659 { 660 int err; 661 662 err = inode_newsize_ok(inode, offset); 663 if (err) 664 goto out; 665 666 i_size_write(inode, offset); 667 /* Optimisation */ 668 if (offset == 0) 669 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_DATA | 670 NFS_INO_DATA_INVAL_DEFER); 671 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE; 672 673 spin_unlock(&inode->i_lock); 674 truncate_pagecache(inode, offset); 675 spin_lock(&inode->i_lock); 676 out: 677 return err; 678 } 679 680 /** 681 * nfs_setattr_update_inode - Update inode metadata after a setattr call. 682 * @inode: pointer to struct inode 683 * @attr: pointer to struct iattr 684 * @fattr: pointer to struct nfs_fattr 685 * 686 * Note: we do this in the *proc.c in order to ensure that 687 * it works for things like exclusive creates too. 688 */ 689 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr, 690 struct nfs_fattr *fattr) 691 { 692 /* Barrier: bump the attribute generation count. */ 693 nfs_fattr_set_barrier(fattr); 694 695 spin_lock(&inode->i_lock); 696 NFS_I(inode)->attr_gencount = fattr->gencount; 697 if ((attr->ia_valid & ATTR_SIZE) != 0) { 698 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME); 699 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC); 700 nfs_vmtruncate(inode, attr->ia_size); 701 } 702 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) { 703 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME; 704 if ((attr->ia_valid & ATTR_MODE) != 0) { 705 int mode = attr->ia_mode & S_IALLUGO; 706 mode |= inode->i_mode & ~S_IALLUGO; 707 inode->i_mode = mode; 708 } 709 if ((attr->ia_valid & ATTR_UID) != 0) 710 inode->i_uid = attr->ia_uid; 711 if ((attr->ia_valid & ATTR_GID) != 0) 712 inode->i_gid = attr->ia_gid; 713 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 714 inode->i_ctime = fattr->ctime; 715 else 716 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE 717 | NFS_INO_INVALID_CTIME); 718 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS 719 | NFS_INO_INVALID_ACL); 720 } 721 if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) { 722 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME 723 | NFS_INO_INVALID_CTIME); 724 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 725 inode->i_atime = fattr->atime; 726 else if (attr->ia_valid & ATTR_ATIME_SET) 727 inode->i_atime = attr->ia_atime; 728 else 729 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME); 730 731 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 732 inode->i_ctime = fattr->ctime; 733 else 734 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE 735 | NFS_INO_INVALID_CTIME); 736 } 737 if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) { 738 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME 739 | NFS_INO_INVALID_CTIME); 740 if (fattr->valid & NFS_ATTR_FATTR_MTIME) 741 inode->i_mtime = fattr->mtime; 742 else if (attr->ia_valid & ATTR_MTIME_SET) 743 inode->i_mtime = attr->ia_mtime; 744 else 745 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME); 746 747 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 748 inode->i_ctime = fattr->ctime; 749 else 750 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE 751 | NFS_INO_INVALID_CTIME); 752 } 753 if (fattr->valid) 754 nfs_update_inode(inode, fattr); 755 spin_unlock(&inode->i_lock); 756 } 757 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode); 758 759 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry) 760 { 761 struct dentry *parent; 762 763 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS)) 764 return; 765 parent = dget_parent(dentry); 766 nfs_force_use_readdirplus(d_inode(parent)); 767 dput(parent); 768 } 769 770 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry) 771 { 772 struct dentry *parent; 773 774 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS)) 775 return; 776 parent = dget_parent(dentry); 777 nfs_advise_use_readdirplus(d_inode(parent)); 778 dput(parent); 779 } 780 781 static bool nfs_need_revalidate_inode(struct inode *inode) 782 { 783 if (NFS_I(inode)->cache_validity & 784 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL)) 785 return true; 786 if (nfs_attribute_cache_expired(inode)) 787 return true; 788 return false; 789 } 790 791 int nfs_getattr(struct user_namespace *mnt_userns, const struct path *path, 792 struct kstat *stat, u32 request_mask, unsigned int query_flags) 793 { 794 struct inode *inode = d_inode(path->dentry); 795 struct nfs_server *server = NFS_SERVER(inode); 796 unsigned long cache_validity; 797 int err = 0; 798 bool force_sync = query_flags & AT_STATX_FORCE_SYNC; 799 bool do_update = false; 800 801 trace_nfs_getattr_enter(inode); 802 803 if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) { 804 nfs_readdirplus_parent_cache_hit(path->dentry); 805 goto out_no_update; 806 } 807 808 /* Flush out writes to the server in order to update c/mtime. */ 809 if ((request_mask & (STATX_CTIME|STATX_MTIME)) && 810 S_ISREG(inode->i_mode)) { 811 err = filemap_write_and_wait(inode->i_mapping); 812 if (err) 813 goto out; 814 } 815 816 /* 817 * We may force a getattr if the user cares about atime. 818 * 819 * Note that we only have to check the vfsmount flags here: 820 * - NFS always sets S_NOATIME by so checking it would give a 821 * bogus result 822 * - NFS never sets SB_NOATIME or SB_NODIRATIME so there is 823 * no point in checking those. 824 */ 825 if ((path->mnt->mnt_flags & MNT_NOATIME) || 826 ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) 827 request_mask &= ~STATX_ATIME; 828 829 /* Is the user requesting attributes that might need revalidation? */ 830 if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME| 831 STATX_MTIME|STATX_UID|STATX_GID| 832 STATX_SIZE|STATX_BLOCKS))) 833 goto out_no_revalidate; 834 835 /* Check whether the cached attributes are stale */ 836 do_update |= force_sync || nfs_attribute_cache_expired(inode); 837 cache_validity = READ_ONCE(NFS_I(inode)->cache_validity); 838 do_update |= cache_validity & 839 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL); 840 if (request_mask & STATX_ATIME) 841 do_update |= cache_validity & NFS_INO_INVALID_ATIME; 842 if (request_mask & (STATX_CTIME|STATX_MTIME)) 843 do_update |= cache_validity & NFS_INO_REVAL_PAGECACHE; 844 if (request_mask & STATX_BLOCKS) 845 do_update |= cache_validity & NFS_INO_INVALID_BLOCKS; 846 if (do_update) { 847 /* Update the attribute cache */ 848 if (!(server->flags & NFS_MOUNT_NOAC)) 849 nfs_readdirplus_parent_cache_miss(path->dentry); 850 else 851 nfs_readdirplus_parent_cache_hit(path->dentry); 852 err = __nfs_revalidate_inode(server, inode); 853 if (err) 854 goto out; 855 } else 856 nfs_readdirplus_parent_cache_hit(path->dentry); 857 out_no_revalidate: 858 /* Only return attributes that were revalidated. */ 859 stat->result_mask &= request_mask; 860 out_no_update: 861 generic_fillattr(&init_user_ns, inode, stat); 862 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode)); 863 if (S_ISDIR(inode->i_mode)) 864 stat->blksize = NFS_SERVER(inode)->dtsize; 865 out: 866 trace_nfs_getattr_exit(inode, err); 867 return err; 868 } 869 EXPORT_SYMBOL_GPL(nfs_getattr); 870 871 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx) 872 { 873 refcount_set(&l_ctx->count, 1); 874 l_ctx->lockowner = current->files; 875 INIT_LIST_HEAD(&l_ctx->list); 876 atomic_set(&l_ctx->io_count, 0); 877 } 878 879 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx) 880 { 881 struct nfs_lock_context *pos; 882 883 list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) { 884 if (pos->lockowner != current->files) 885 continue; 886 if (refcount_inc_not_zero(&pos->count)) 887 return pos; 888 } 889 return NULL; 890 } 891 892 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx) 893 { 894 struct nfs_lock_context *res, *new = NULL; 895 struct inode *inode = d_inode(ctx->dentry); 896 897 rcu_read_lock(); 898 res = __nfs_find_lock_context(ctx); 899 rcu_read_unlock(); 900 if (res == NULL) { 901 new = kmalloc(sizeof(*new), GFP_KERNEL); 902 if (new == NULL) 903 return ERR_PTR(-ENOMEM); 904 nfs_init_lock_context(new); 905 spin_lock(&inode->i_lock); 906 res = __nfs_find_lock_context(ctx); 907 if (res == NULL) { 908 new->open_context = get_nfs_open_context(ctx); 909 if (new->open_context) { 910 list_add_tail_rcu(&new->list, 911 &ctx->lock_context.list); 912 res = new; 913 new = NULL; 914 } else 915 res = ERR_PTR(-EBADF); 916 } 917 spin_unlock(&inode->i_lock); 918 kfree(new); 919 } 920 return res; 921 } 922 EXPORT_SYMBOL_GPL(nfs_get_lock_context); 923 924 void nfs_put_lock_context(struct nfs_lock_context *l_ctx) 925 { 926 struct nfs_open_context *ctx = l_ctx->open_context; 927 struct inode *inode = d_inode(ctx->dentry); 928 929 if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock)) 930 return; 931 list_del_rcu(&l_ctx->list); 932 spin_unlock(&inode->i_lock); 933 put_nfs_open_context(ctx); 934 kfree_rcu(l_ctx, rcu_head); 935 } 936 EXPORT_SYMBOL_GPL(nfs_put_lock_context); 937 938 /** 939 * nfs_close_context - Common close_context() routine NFSv2/v3 940 * @ctx: pointer to context 941 * @is_sync: is this a synchronous close 942 * 943 * Ensure that the attributes are up to date if we're mounted 944 * with close-to-open semantics and we have cached data that will 945 * need to be revalidated on open. 946 */ 947 void nfs_close_context(struct nfs_open_context *ctx, int is_sync) 948 { 949 struct nfs_inode *nfsi; 950 struct inode *inode; 951 struct nfs_server *server; 952 953 if (!(ctx->mode & FMODE_WRITE)) 954 return; 955 if (!is_sync) 956 return; 957 inode = d_inode(ctx->dentry); 958 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) 959 return; 960 nfsi = NFS_I(inode); 961 if (inode->i_mapping->nrpages == 0) 962 return; 963 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 964 return; 965 if (!list_empty(&nfsi->open_files)) 966 return; 967 server = NFS_SERVER(inode); 968 if (server->flags & NFS_MOUNT_NOCTO) 969 return; 970 nfs_revalidate_inode(server, inode); 971 } 972 EXPORT_SYMBOL_GPL(nfs_close_context); 973 974 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, 975 fmode_t f_mode, 976 struct file *filp) 977 { 978 struct nfs_open_context *ctx; 979 980 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); 981 if (!ctx) 982 return ERR_PTR(-ENOMEM); 983 nfs_sb_active(dentry->d_sb); 984 ctx->dentry = dget(dentry); 985 if (filp) 986 ctx->cred = get_cred(filp->f_cred); 987 else 988 ctx->cred = get_current_cred(); 989 ctx->ll_cred = NULL; 990 ctx->state = NULL; 991 ctx->mode = f_mode; 992 ctx->flags = 0; 993 ctx->error = 0; 994 ctx->flock_owner = (fl_owner_t)filp; 995 nfs_init_lock_context(&ctx->lock_context); 996 ctx->lock_context.open_context = ctx; 997 INIT_LIST_HEAD(&ctx->list); 998 ctx->mdsthreshold = NULL; 999 return ctx; 1000 } 1001 EXPORT_SYMBOL_GPL(alloc_nfs_open_context); 1002 1003 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx) 1004 { 1005 if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count)) 1006 return ctx; 1007 return NULL; 1008 } 1009 EXPORT_SYMBOL_GPL(get_nfs_open_context); 1010 1011 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync) 1012 { 1013 struct inode *inode = d_inode(ctx->dentry); 1014 struct super_block *sb = ctx->dentry->d_sb; 1015 1016 if (!refcount_dec_and_test(&ctx->lock_context.count)) 1017 return; 1018 if (!list_empty(&ctx->list)) { 1019 spin_lock(&inode->i_lock); 1020 list_del_rcu(&ctx->list); 1021 spin_unlock(&inode->i_lock); 1022 } 1023 if (inode != NULL) 1024 NFS_PROTO(inode)->close_context(ctx, is_sync); 1025 put_cred(ctx->cred); 1026 dput(ctx->dentry); 1027 nfs_sb_deactive(sb); 1028 put_rpccred(ctx->ll_cred); 1029 kfree(ctx->mdsthreshold); 1030 kfree_rcu(ctx, rcu_head); 1031 } 1032 1033 void put_nfs_open_context(struct nfs_open_context *ctx) 1034 { 1035 __put_nfs_open_context(ctx, 0); 1036 } 1037 EXPORT_SYMBOL_GPL(put_nfs_open_context); 1038 1039 static void put_nfs_open_context_sync(struct nfs_open_context *ctx) 1040 { 1041 __put_nfs_open_context(ctx, 1); 1042 } 1043 1044 /* 1045 * Ensure that mmap has a recent RPC credential for use when writing out 1046 * shared pages 1047 */ 1048 void nfs_inode_attach_open_context(struct nfs_open_context *ctx) 1049 { 1050 struct inode *inode = d_inode(ctx->dentry); 1051 struct nfs_inode *nfsi = NFS_I(inode); 1052 1053 spin_lock(&inode->i_lock); 1054 if (list_empty(&nfsi->open_files) && 1055 (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER)) 1056 nfsi->cache_validity |= NFS_INO_INVALID_DATA | 1057 NFS_INO_REVAL_FORCED; 1058 list_add_tail_rcu(&ctx->list, &nfsi->open_files); 1059 spin_unlock(&inode->i_lock); 1060 } 1061 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context); 1062 1063 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx) 1064 { 1065 filp->private_data = get_nfs_open_context(ctx); 1066 if (list_empty(&ctx->list)) 1067 nfs_inode_attach_open_context(ctx); 1068 } 1069 EXPORT_SYMBOL_GPL(nfs_file_set_open_context); 1070 1071 /* 1072 * Given an inode, search for an open context with the desired characteristics 1073 */ 1074 struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode) 1075 { 1076 struct nfs_inode *nfsi = NFS_I(inode); 1077 struct nfs_open_context *pos, *ctx = NULL; 1078 1079 rcu_read_lock(); 1080 list_for_each_entry_rcu(pos, &nfsi->open_files, list) { 1081 if (cred != NULL && cred_fscmp(pos->cred, cred) != 0) 1082 continue; 1083 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode) 1084 continue; 1085 ctx = get_nfs_open_context(pos); 1086 if (ctx) 1087 break; 1088 } 1089 rcu_read_unlock(); 1090 return ctx; 1091 } 1092 1093 void nfs_file_clear_open_context(struct file *filp) 1094 { 1095 struct nfs_open_context *ctx = nfs_file_open_context(filp); 1096 1097 if (ctx) { 1098 struct inode *inode = d_inode(ctx->dentry); 1099 1100 /* 1101 * We fatal error on write before. Try to writeback 1102 * every page again. 1103 */ 1104 if (ctx->error < 0) 1105 invalidate_inode_pages2(inode->i_mapping); 1106 filp->private_data = NULL; 1107 put_nfs_open_context_sync(ctx); 1108 } 1109 } 1110 1111 /* 1112 * These allocate and release file read/write context information. 1113 */ 1114 int nfs_open(struct inode *inode, struct file *filp) 1115 { 1116 struct nfs_open_context *ctx; 1117 1118 ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp); 1119 if (IS_ERR(ctx)) 1120 return PTR_ERR(ctx); 1121 nfs_file_set_open_context(filp, ctx); 1122 put_nfs_open_context(ctx); 1123 nfs_fscache_open_file(inode, filp); 1124 return 0; 1125 } 1126 EXPORT_SYMBOL_GPL(nfs_open); 1127 1128 /* 1129 * This function is called whenever some part of NFS notices that 1130 * the cached attributes have to be refreshed. 1131 */ 1132 int 1133 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) 1134 { 1135 int status = -ESTALE; 1136 struct nfs4_label *label = NULL; 1137 struct nfs_fattr *fattr = NULL; 1138 struct nfs_inode *nfsi = NFS_I(inode); 1139 1140 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n", 1141 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode)); 1142 1143 trace_nfs_revalidate_inode_enter(inode); 1144 1145 if (is_bad_inode(inode)) 1146 goto out; 1147 if (NFS_STALE(inode)) 1148 goto out; 1149 1150 /* pNFS: Attributes aren't updated until we layoutcommit */ 1151 if (S_ISREG(inode->i_mode)) { 1152 status = pnfs_sync_inode(inode, false); 1153 if (status) 1154 goto out; 1155 } 1156 1157 status = -ENOMEM; 1158 fattr = nfs_alloc_fattr(); 1159 if (fattr == NULL) 1160 goto out; 1161 1162 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE); 1163 1164 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL); 1165 if (IS_ERR(label)) { 1166 status = PTR_ERR(label); 1167 goto out; 1168 } 1169 1170 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, 1171 label, inode); 1172 if (status != 0) { 1173 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n", 1174 inode->i_sb->s_id, 1175 (unsigned long long)NFS_FILEID(inode), status); 1176 switch (status) { 1177 case -ETIMEDOUT: 1178 /* A soft timeout occurred. Use cached information? */ 1179 if (server->flags & NFS_MOUNT_SOFTREVAL) 1180 status = 0; 1181 break; 1182 case -ESTALE: 1183 if (!S_ISDIR(inode->i_mode)) 1184 nfs_set_inode_stale(inode); 1185 else 1186 nfs_zap_caches(inode); 1187 } 1188 goto err_out; 1189 } 1190 1191 status = nfs_refresh_inode(inode, fattr); 1192 if (status) { 1193 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n", 1194 inode->i_sb->s_id, 1195 (unsigned long long)NFS_FILEID(inode), status); 1196 goto err_out; 1197 } 1198 1199 if (nfsi->cache_validity & NFS_INO_INVALID_ACL) 1200 nfs_zap_acl_cache(inode); 1201 1202 nfs_setsecurity(inode, fattr, label); 1203 1204 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n", 1205 inode->i_sb->s_id, 1206 (unsigned long long)NFS_FILEID(inode)); 1207 1208 err_out: 1209 nfs4_label_free(label); 1210 out: 1211 nfs_free_fattr(fattr); 1212 trace_nfs_revalidate_inode_exit(inode, status); 1213 return status; 1214 } 1215 1216 int nfs_attribute_cache_expired(struct inode *inode) 1217 { 1218 if (nfs_have_delegated_attributes(inode)) 1219 return 0; 1220 return nfs_attribute_timeout(inode); 1221 } 1222 1223 /** 1224 * nfs_revalidate_inode - Revalidate the inode attributes 1225 * @server: pointer to nfs_server struct 1226 * @inode: pointer to inode struct 1227 * 1228 * Updates inode attribute information by retrieving the data from the server. 1229 */ 1230 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) 1231 { 1232 if (!nfs_need_revalidate_inode(inode)) 1233 return NFS_STALE(inode) ? -ESTALE : 0; 1234 return __nfs_revalidate_inode(server, inode); 1235 } 1236 EXPORT_SYMBOL_GPL(nfs_revalidate_inode); 1237 1238 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping) 1239 { 1240 int ret; 1241 1242 if (mapping->nrpages != 0) { 1243 if (S_ISREG(inode->i_mode)) { 1244 ret = nfs_sync_mapping(mapping); 1245 if (ret < 0) 1246 return ret; 1247 } 1248 ret = invalidate_inode_pages2(mapping); 1249 if (ret < 0) 1250 return ret; 1251 } 1252 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE); 1253 nfs_fscache_wait_on_invalidate(inode); 1254 1255 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n", 1256 inode->i_sb->s_id, 1257 (unsigned long long)NFS_FILEID(inode)); 1258 return 0; 1259 } 1260 1261 bool nfs_mapping_need_revalidate_inode(struct inode *inode) 1262 { 1263 return nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE) || 1264 NFS_STALE(inode); 1265 } 1266 1267 int nfs_revalidate_mapping_rcu(struct inode *inode) 1268 { 1269 struct nfs_inode *nfsi = NFS_I(inode); 1270 unsigned long *bitlock = &nfsi->flags; 1271 int ret = 0; 1272 1273 if (IS_SWAPFILE(inode)) 1274 goto out; 1275 if (nfs_mapping_need_revalidate_inode(inode)) { 1276 ret = -ECHILD; 1277 goto out; 1278 } 1279 spin_lock(&inode->i_lock); 1280 if (test_bit(NFS_INO_INVALIDATING, bitlock) || 1281 (nfsi->cache_validity & NFS_INO_INVALID_DATA)) 1282 ret = -ECHILD; 1283 spin_unlock(&inode->i_lock); 1284 out: 1285 return ret; 1286 } 1287 1288 /** 1289 * nfs_revalidate_mapping - Revalidate the pagecache 1290 * @inode: pointer to host inode 1291 * @mapping: pointer to mapping 1292 */ 1293 int nfs_revalidate_mapping(struct inode *inode, 1294 struct address_space *mapping) 1295 { 1296 struct nfs_inode *nfsi = NFS_I(inode); 1297 unsigned long *bitlock = &nfsi->flags; 1298 int ret = 0; 1299 1300 /* swapfiles are not supposed to be shared. */ 1301 if (IS_SWAPFILE(inode)) 1302 goto out; 1303 1304 if (nfs_mapping_need_revalidate_inode(inode)) { 1305 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 1306 if (ret < 0) 1307 goto out; 1308 } 1309 1310 /* 1311 * We must clear NFS_INO_INVALID_DATA first to ensure that 1312 * invalidations that come in while we're shooting down the mappings 1313 * are respected. But, that leaves a race window where one revalidator 1314 * can clear the flag, and then another checks it before the mapping 1315 * gets invalidated. Fix that by serializing access to this part of 1316 * the function. 1317 * 1318 * At the same time, we need to allow other tasks to see whether we 1319 * might be in the middle of invalidating the pages, so we only set 1320 * the bit lock here if it looks like we're going to be doing that. 1321 */ 1322 for (;;) { 1323 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING, 1324 nfs_wait_bit_killable, TASK_KILLABLE); 1325 if (ret) 1326 goto out; 1327 spin_lock(&inode->i_lock); 1328 if (test_bit(NFS_INO_INVALIDATING, bitlock)) { 1329 spin_unlock(&inode->i_lock); 1330 continue; 1331 } 1332 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 1333 break; 1334 spin_unlock(&inode->i_lock); 1335 goto out; 1336 } 1337 1338 set_bit(NFS_INO_INVALIDATING, bitlock); 1339 smp_wmb(); 1340 nfsi->cache_validity &= ~(NFS_INO_INVALID_DATA| 1341 NFS_INO_DATA_INVAL_DEFER); 1342 spin_unlock(&inode->i_lock); 1343 trace_nfs_invalidate_mapping_enter(inode); 1344 ret = nfs_invalidate_mapping(inode, mapping); 1345 trace_nfs_invalidate_mapping_exit(inode, ret); 1346 1347 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock); 1348 smp_mb__after_atomic(); 1349 wake_up_bit(bitlock, NFS_INO_INVALIDATING); 1350 out: 1351 return ret; 1352 } 1353 1354 static bool nfs_file_has_writers(struct nfs_inode *nfsi) 1355 { 1356 struct inode *inode = &nfsi->vfs_inode; 1357 1358 if (!S_ISREG(inode->i_mode)) 1359 return false; 1360 if (list_empty(&nfsi->open_files)) 1361 return false; 1362 return inode_is_open_for_write(inode); 1363 } 1364 1365 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi) 1366 { 1367 return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi); 1368 } 1369 1370 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1371 { 1372 struct timespec64 ts; 1373 1374 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE) 1375 && (fattr->valid & NFS_ATTR_FATTR_CHANGE) 1376 && inode_eq_iversion_raw(inode, fattr->pre_change_attr)) { 1377 inode_set_iversion_raw(inode, fattr->change_attr); 1378 if (S_ISDIR(inode->i_mode)) 1379 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA); 1380 else if (nfs_server_capable(inode, NFS_CAP_XATTR)) 1381 nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR); 1382 } 1383 /* If we have atomic WCC data, we may update some attributes */ 1384 ts = inode->i_ctime; 1385 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME) 1386 && (fattr->valid & NFS_ATTR_FATTR_CTIME) 1387 && timespec64_equal(&ts, &fattr->pre_ctime)) { 1388 inode->i_ctime = fattr->ctime; 1389 } 1390 1391 ts = inode->i_mtime; 1392 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME) 1393 && (fattr->valid & NFS_ATTR_FATTR_MTIME) 1394 && timespec64_equal(&ts, &fattr->pre_mtime)) { 1395 inode->i_mtime = fattr->mtime; 1396 if (S_ISDIR(inode->i_mode)) 1397 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA); 1398 } 1399 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE) 1400 && (fattr->valid & NFS_ATTR_FATTR_SIZE) 1401 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size) 1402 && !nfs_have_writebacks(inode)) { 1403 i_size_write(inode, nfs_size_to_loff_t(fattr->size)); 1404 } 1405 } 1406 1407 /** 1408 * nfs_check_inode_attributes - verify consistency of the inode attribute cache 1409 * @inode: pointer to inode 1410 * @fattr: updated attributes 1411 * 1412 * Verifies the attribute cache. If we have just changed the attributes, 1413 * so that fattr carries weak cache consistency data, then it may 1414 * also update the ctime/mtime/change_attribute. 1415 */ 1416 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr) 1417 { 1418 struct nfs_inode *nfsi = NFS_I(inode); 1419 loff_t cur_size, new_isize; 1420 unsigned long invalid = 0; 1421 struct timespec64 ts; 1422 1423 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) 1424 return 0; 1425 1426 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) { 1427 /* Only a mounted-on-fileid? Just exit */ 1428 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) 1429 return 0; 1430 /* Has the inode gone and changed behind our back? */ 1431 } else if (nfsi->fileid != fattr->fileid) { 1432 /* Is this perhaps the mounted-on fileid? */ 1433 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) && 1434 nfsi->fileid == fattr->mounted_on_fileid) 1435 return 0; 1436 return -ESTALE; 1437 } 1438 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) 1439 return -ESTALE; 1440 1441 1442 if (!nfs_file_has_buffered_writers(nfsi)) { 1443 /* Verify a few of the more important attributes */ 1444 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr)) 1445 invalid |= NFS_INO_INVALID_CHANGE 1446 | NFS_INO_REVAL_PAGECACHE; 1447 1448 ts = inode->i_mtime; 1449 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime)) 1450 invalid |= NFS_INO_INVALID_MTIME; 1451 1452 ts = inode->i_ctime; 1453 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime)) 1454 invalid |= NFS_INO_INVALID_CTIME; 1455 1456 if (fattr->valid & NFS_ATTR_FATTR_SIZE) { 1457 cur_size = i_size_read(inode); 1458 new_isize = nfs_size_to_loff_t(fattr->size); 1459 if (cur_size != new_isize) 1460 invalid |= NFS_INO_INVALID_SIZE 1461 | NFS_INO_REVAL_PAGECACHE; 1462 } 1463 } 1464 1465 /* Have any file permissions changed? */ 1466 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) 1467 invalid |= NFS_INO_INVALID_ACCESS 1468 | NFS_INO_INVALID_ACL 1469 | NFS_INO_INVALID_OTHER; 1470 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid)) 1471 invalid |= NFS_INO_INVALID_ACCESS 1472 | NFS_INO_INVALID_ACL 1473 | NFS_INO_INVALID_OTHER; 1474 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid)) 1475 invalid |= NFS_INO_INVALID_ACCESS 1476 | NFS_INO_INVALID_ACL 1477 | NFS_INO_INVALID_OTHER; 1478 1479 /* Has the link count changed? */ 1480 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink) 1481 invalid |= NFS_INO_INVALID_OTHER; 1482 1483 ts = inode->i_atime; 1484 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime)) 1485 invalid |= NFS_INO_INVALID_ATIME; 1486 1487 if (invalid != 0) 1488 nfs_set_cache_invalid(inode, invalid); 1489 1490 nfsi->read_cache_jiffies = fattr->time_start; 1491 return 0; 1492 } 1493 1494 static atomic_long_t nfs_attr_generation_counter; 1495 1496 static unsigned long nfs_read_attr_generation_counter(void) 1497 { 1498 return atomic_long_read(&nfs_attr_generation_counter); 1499 } 1500 1501 unsigned long nfs_inc_attr_generation_counter(void) 1502 { 1503 return atomic_long_inc_return(&nfs_attr_generation_counter); 1504 } 1505 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter); 1506 1507 void nfs_fattr_init(struct nfs_fattr *fattr) 1508 { 1509 fattr->valid = 0; 1510 fattr->time_start = jiffies; 1511 fattr->gencount = nfs_inc_attr_generation_counter(); 1512 fattr->owner_name = NULL; 1513 fattr->group_name = NULL; 1514 } 1515 EXPORT_SYMBOL_GPL(nfs_fattr_init); 1516 1517 /** 1518 * nfs_fattr_set_barrier 1519 * @fattr: attributes 1520 * 1521 * Used to set a barrier after an attribute was updated. This 1522 * barrier ensures that older attributes from RPC calls that may 1523 * have raced with our update cannot clobber these new values. 1524 * Note that you are still responsible for ensuring that other 1525 * operations which change the attribute on the server do not 1526 * collide. 1527 */ 1528 void nfs_fattr_set_barrier(struct nfs_fattr *fattr) 1529 { 1530 fattr->gencount = nfs_inc_attr_generation_counter(); 1531 } 1532 1533 struct nfs_fattr *nfs_alloc_fattr(void) 1534 { 1535 struct nfs_fattr *fattr; 1536 1537 fattr = kmalloc(sizeof(*fattr), GFP_NOFS); 1538 if (fattr != NULL) 1539 nfs_fattr_init(fattr); 1540 return fattr; 1541 } 1542 EXPORT_SYMBOL_GPL(nfs_alloc_fattr); 1543 1544 struct nfs_fh *nfs_alloc_fhandle(void) 1545 { 1546 struct nfs_fh *fh; 1547 1548 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS); 1549 if (fh != NULL) 1550 fh->size = 0; 1551 return fh; 1552 } 1553 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle); 1554 1555 #ifdef NFS_DEBUG 1556 /* 1557 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle 1558 * in the same way that wireshark does 1559 * 1560 * @fh: file handle 1561 * 1562 * For debugging only. 1563 */ 1564 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh) 1565 { 1566 /* wireshark uses 32-bit AUTODIN crc and does a bitwise 1567 * not on the result */ 1568 return nfs_fhandle_hash(fh); 1569 } 1570 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash); 1571 1572 /* 1573 * _nfs_display_fhandle - display an NFS file handle on the console 1574 * 1575 * @fh: file handle to display 1576 * @caption: display caption 1577 * 1578 * For debugging only. 1579 */ 1580 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption) 1581 { 1582 unsigned short i; 1583 1584 if (fh == NULL || fh->size == 0) { 1585 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh); 1586 return; 1587 } 1588 1589 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n", 1590 caption, fh, fh->size, _nfs_display_fhandle_hash(fh)); 1591 for (i = 0; i < fh->size; i += 16) { 1592 __be32 *pos = (__be32 *)&fh->data[i]; 1593 1594 switch ((fh->size - i - 1) >> 2) { 1595 case 0: 1596 printk(KERN_DEFAULT " %08x\n", 1597 be32_to_cpup(pos)); 1598 break; 1599 case 1: 1600 printk(KERN_DEFAULT " %08x %08x\n", 1601 be32_to_cpup(pos), be32_to_cpup(pos + 1)); 1602 break; 1603 case 2: 1604 printk(KERN_DEFAULT " %08x %08x %08x\n", 1605 be32_to_cpup(pos), be32_to_cpup(pos + 1), 1606 be32_to_cpup(pos + 2)); 1607 break; 1608 default: 1609 printk(KERN_DEFAULT " %08x %08x %08x %08x\n", 1610 be32_to_cpup(pos), be32_to_cpup(pos + 1), 1611 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3)); 1612 } 1613 } 1614 } 1615 EXPORT_SYMBOL_GPL(_nfs_display_fhandle); 1616 #endif 1617 1618 /** 1619 * nfs_inode_attrs_need_update - check if the inode attributes need updating 1620 * @inode: pointer to inode 1621 * @fattr: attributes 1622 * 1623 * Attempt to divine whether or not an RPC call reply carrying stale 1624 * attributes got scheduled after another call carrying updated ones. 1625 * 1626 * To do so, the function first assumes that a more recent ctime means 1627 * that the attributes in fattr are newer, however it also attempt to 1628 * catch the case where ctime either didn't change, or went backwards 1629 * (if someone reset the clock on the server) by looking at whether 1630 * or not this RPC call was started after the inode was last updated. 1631 * Note also the check for wraparound of 'attr_gencount' 1632 * 1633 * The function returns 'true' if it thinks the attributes in 'fattr' are 1634 * more recent than the ones cached in the inode. 1635 * 1636 */ 1637 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr) 1638 { 1639 const struct nfs_inode *nfsi = NFS_I(inode); 1640 1641 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 || 1642 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0); 1643 } 1644 1645 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr) 1646 { 1647 int ret; 1648 1649 trace_nfs_refresh_inode_enter(inode); 1650 1651 if (nfs_inode_attrs_need_update(inode, fattr)) 1652 ret = nfs_update_inode(inode, fattr); 1653 else 1654 ret = nfs_check_inode_attributes(inode, fattr); 1655 1656 trace_nfs_refresh_inode_exit(inode, ret); 1657 return ret; 1658 } 1659 1660 /** 1661 * nfs_refresh_inode - try to update the inode attribute cache 1662 * @inode: pointer to inode 1663 * @fattr: updated attributes 1664 * 1665 * Check that an RPC call that returned attributes has not overlapped with 1666 * other recent updates of the inode metadata, then decide whether it is 1667 * safe to do a full update of the inode attributes, or whether just to 1668 * call nfs_check_inode_attributes. 1669 */ 1670 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr) 1671 { 1672 int status; 1673 1674 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 1675 return 0; 1676 spin_lock(&inode->i_lock); 1677 status = nfs_refresh_inode_locked(inode, fattr); 1678 spin_unlock(&inode->i_lock); 1679 1680 return status; 1681 } 1682 EXPORT_SYMBOL_GPL(nfs_refresh_inode); 1683 1684 static int nfs_post_op_update_inode_locked(struct inode *inode, 1685 struct nfs_fattr *fattr, unsigned int invalid) 1686 { 1687 if (S_ISDIR(inode->i_mode)) 1688 invalid |= NFS_INO_INVALID_DATA; 1689 nfs_set_cache_invalid(inode, invalid); 1690 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 1691 return 0; 1692 return nfs_refresh_inode_locked(inode, fattr); 1693 } 1694 1695 /** 1696 * nfs_post_op_update_inode - try to update the inode attribute cache 1697 * @inode: pointer to inode 1698 * @fattr: updated attributes 1699 * 1700 * After an operation that has changed the inode metadata, mark the 1701 * attribute cache as being invalid, then try to update it. 1702 * 1703 * NB: if the server didn't return any post op attributes, this 1704 * function will force the retrieval of attributes before the next 1705 * NFS request. Thus it should be used only for operations that 1706 * are expected to change one or more attributes, to avoid 1707 * unnecessary NFS requests and trips through nfs_update_inode(). 1708 */ 1709 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1710 { 1711 int status; 1712 1713 spin_lock(&inode->i_lock); 1714 nfs_fattr_set_barrier(fattr); 1715 status = nfs_post_op_update_inode_locked(inode, fattr, 1716 NFS_INO_INVALID_CHANGE 1717 | NFS_INO_INVALID_CTIME 1718 | NFS_INO_REVAL_FORCED); 1719 spin_unlock(&inode->i_lock); 1720 1721 return status; 1722 } 1723 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode); 1724 1725 /** 1726 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache 1727 * @inode: pointer to inode 1728 * @fattr: updated attributes 1729 * 1730 * After an operation that has changed the inode metadata, mark the 1731 * attribute cache as being invalid, then try to update it. Fake up 1732 * weak cache consistency data, if none exist. 1733 * 1734 * This function is mainly designed to be used by the ->write_done() functions. 1735 */ 1736 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr) 1737 { 1738 int status; 1739 1740 /* Don't do a WCC update if these attributes are already stale */ 1741 if ((fattr->valid & NFS_ATTR_FATTR) == 0 || 1742 !nfs_inode_attrs_need_update(inode, fattr)) { 1743 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE 1744 | NFS_ATTR_FATTR_PRESIZE 1745 | NFS_ATTR_FATTR_PREMTIME 1746 | NFS_ATTR_FATTR_PRECTIME); 1747 goto out_noforce; 1748 } 1749 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && 1750 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) { 1751 fattr->pre_change_attr = inode_peek_iversion_raw(inode); 1752 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE; 1753 } 1754 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 && 1755 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) { 1756 fattr->pre_ctime = inode->i_ctime; 1757 fattr->valid |= NFS_ATTR_FATTR_PRECTIME; 1758 } 1759 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 && 1760 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) { 1761 fattr->pre_mtime = inode->i_mtime; 1762 fattr->valid |= NFS_ATTR_FATTR_PREMTIME; 1763 } 1764 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 && 1765 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) { 1766 fattr->pre_size = i_size_read(inode); 1767 fattr->valid |= NFS_ATTR_FATTR_PRESIZE; 1768 } 1769 out_noforce: 1770 status = nfs_post_op_update_inode_locked(inode, fattr, 1771 NFS_INO_INVALID_CHANGE 1772 | NFS_INO_INVALID_CTIME 1773 | NFS_INO_INVALID_MTIME 1774 | NFS_INO_INVALID_BLOCKS); 1775 return status; 1776 } 1777 1778 /** 1779 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache 1780 * @inode: pointer to inode 1781 * @fattr: updated attributes 1782 * 1783 * After an operation that has changed the inode metadata, mark the 1784 * attribute cache as being invalid, then try to update it. Fake up 1785 * weak cache consistency data, if none exist. 1786 * 1787 * This function is mainly designed to be used by the ->write_done() functions. 1788 */ 1789 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr) 1790 { 1791 int status; 1792 1793 spin_lock(&inode->i_lock); 1794 nfs_fattr_set_barrier(fattr); 1795 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 1796 spin_unlock(&inode->i_lock); 1797 return status; 1798 } 1799 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc); 1800 1801 1802 /* 1803 * Many nfs protocol calls return the new file attributes after 1804 * an operation. Here we update the inode to reflect the state 1805 * of the server's inode. 1806 * 1807 * This is a bit tricky because we have to make sure all dirty pages 1808 * have been sent off to the server before calling invalidate_inode_pages. 1809 * To make sure no other process adds more write requests while we try 1810 * our best to flush them, we make them sleep during the attribute refresh. 1811 * 1812 * A very similar scenario holds for the dir cache. 1813 */ 1814 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1815 { 1816 struct nfs_server *server; 1817 struct nfs_inode *nfsi = NFS_I(inode); 1818 loff_t cur_isize, new_isize; 1819 unsigned long invalid = 0; 1820 unsigned long now = jiffies; 1821 unsigned long save_cache_validity; 1822 bool have_writers = nfs_file_has_buffered_writers(nfsi); 1823 bool cache_revalidated = true; 1824 bool attr_changed = false; 1825 bool have_delegation; 1826 1827 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n", 1828 __func__, inode->i_sb->s_id, inode->i_ino, 1829 nfs_display_fhandle_hash(NFS_FH(inode)), 1830 atomic_read(&inode->i_count), fattr->valid); 1831 1832 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) { 1833 /* Only a mounted-on-fileid? Just exit */ 1834 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) 1835 return 0; 1836 /* Has the inode gone and changed behind our back? */ 1837 } else if (nfsi->fileid != fattr->fileid) { 1838 /* Is this perhaps the mounted-on fileid? */ 1839 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) && 1840 nfsi->fileid == fattr->mounted_on_fileid) 1841 return 0; 1842 printk(KERN_ERR "NFS: server %s error: fileid changed\n" 1843 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n", 1844 NFS_SERVER(inode)->nfs_client->cl_hostname, 1845 inode->i_sb->s_id, (long long)nfsi->fileid, 1846 (long long)fattr->fileid); 1847 goto out_err; 1848 } 1849 1850 /* 1851 * Make sure the inode's type hasn't changed. 1852 */ 1853 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) { 1854 /* 1855 * Big trouble! The inode has become a different object. 1856 */ 1857 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n", 1858 __func__, inode->i_ino, inode->i_mode, fattr->mode); 1859 goto out_err; 1860 } 1861 1862 server = NFS_SERVER(inode); 1863 /* Update the fsid? */ 1864 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) && 1865 !nfs_fsid_equal(&server->fsid, &fattr->fsid) && 1866 !IS_AUTOMOUNT(inode)) 1867 server->fsid = fattr->fsid; 1868 1869 /* Save the delegation state before clearing cache_validity */ 1870 have_delegation = nfs_have_delegated_attributes(inode); 1871 1872 /* 1873 * Update the read time so we don't revalidate too often. 1874 */ 1875 nfsi->read_cache_jiffies = fattr->time_start; 1876 1877 save_cache_validity = nfsi->cache_validity; 1878 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR 1879 | NFS_INO_INVALID_ATIME 1880 | NFS_INO_REVAL_FORCED 1881 | NFS_INO_REVAL_PAGECACHE 1882 | NFS_INO_INVALID_BLOCKS); 1883 1884 /* Do atomic weak cache consistency updates */ 1885 nfs_wcc_update_inode(inode, fattr); 1886 1887 if (pnfs_layoutcommit_outstanding(inode)) { 1888 nfsi->cache_validity |= save_cache_validity & NFS_INO_INVALID_ATTR; 1889 cache_revalidated = false; 1890 } 1891 1892 /* More cache consistency checks */ 1893 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) { 1894 if (!inode_eq_iversion_raw(inode, fattr->change_attr)) { 1895 /* Could it be a race with writeback? */ 1896 if (!(have_writers || have_delegation)) { 1897 invalid |= NFS_INO_INVALID_DATA 1898 | NFS_INO_INVALID_ACCESS 1899 | NFS_INO_INVALID_ACL 1900 | NFS_INO_INVALID_XATTR; 1901 /* Force revalidate of all attributes */ 1902 save_cache_validity |= NFS_INO_INVALID_CTIME 1903 | NFS_INO_INVALID_MTIME 1904 | NFS_INO_INVALID_SIZE 1905 | NFS_INO_INVALID_OTHER; 1906 if (S_ISDIR(inode->i_mode)) 1907 nfs_force_lookup_revalidate(inode); 1908 dprintk("NFS: change_attr change on server for file %s/%ld\n", 1909 inode->i_sb->s_id, 1910 inode->i_ino); 1911 } else if (!have_delegation) 1912 nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER; 1913 inode_set_iversion_raw(inode, fattr->change_attr); 1914 attr_changed = true; 1915 } 1916 } else { 1917 nfsi->cache_validity |= save_cache_validity & 1918 (NFS_INO_INVALID_CHANGE 1919 | NFS_INO_REVAL_PAGECACHE 1920 | NFS_INO_REVAL_FORCED); 1921 cache_revalidated = false; 1922 } 1923 1924 if (fattr->valid & NFS_ATTR_FATTR_MTIME) { 1925 inode->i_mtime = fattr->mtime; 1926 } else if (server->caps & NFS_CAP_MTIME) { 1927 nfsi->cache_validity |= save_cache_validity & 1928 (NFS_INO_INVALID_MTIME 1929 | NFS_INO_REVAL_FORCED); 1930 cache_revalidated = false; 1931 } 1932 1933 if (fattr->valid & NFS_ATTR_FATTR_CTIME) { 1934 inode->i_ctime = fattr->ctime; 1935 } else if (server->caps & NFS_CAP_CTIME) { 1936 nfsi->cache_validity |= save_cache_validity & 1937 (NFS_INO_INVALID_CTIME 1938 | NFS_INO_REVAL_FORCED); 1939 cache_revalidated = false; 1940 } 1941 1942 /* Check if our cached file size is stale */ 1943 if (fattr->valid & NFS_ATTR_FATTR_SIZE) { 1944 new_isize = nfs_size_to_loff_t(fattr->size); 1945 cur_isize = i_size_read(inode); 1946 if (new_isize != cur_isize && !have_delegation) { 1947 /* Do we perhaps have any outstanding writes, or has 1948 * the file grown beyond our last write? */ 1949 if (!nfs_have_writebacks(inode) || new_isize > cur_isize) { 1950 i_size_write(inode, new_isize); 1951 if (!have_writers) 1952 invalid |= NFS_INO_INVALID_DATA; 1953 attr_changed = true; 1954 } 1955 dprintk("NFS: isize change on server for file %s/%ld " 1956 "(%Ld to %Ld)\n", 1957 inode->i_sb->s_id, 1958 inode->i_ino, 1959 (long long)cur_isize, 1960 (long long)new_isize); 1961 } 1962 } else { 1963 nfsi->cache_validity |= save_cache_validity & 1964 (NFS_INO_INVALID_SIZE 1965 | NFS_INO_REVAL_PAGECACHE 1966 | NFS_INO_REVAL_FORCED); 1967 cache_revalidated = false; 1968 } 1969 1970 1971 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 1972 inode->i_atime = fattr->atime; 1973 else if (server->caps & NFS_CAP_ATIME) { 1974 nfsi->cache_validity |= save_cache_validity & 1975 (NFS_INO_INVALID_ATIME 1976 | NFS_INO_REVAL_FORCED); 1977 cache_revalidated = false; 1978 } 1979 1980 if (fattr->valid & NFS_ATTR_FATTR_MODE) { 1981 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) { 1982 umode_t newmode = inode->i_mode & S_IFMT; 1983 newmode |= fattr->mode & S_IALLUGO; 1984 inode->i_mode = newmode; 1985 invalid |= NFS_INO_INVALID_ACCESS 1986 | NFS_INO_INVALID_ACL; 1987 attr_changed = true; 1988 } 1989 } else if (server->caps & NFS_CAP_MODE) { 1990 nfsi->cache_validity |= save_cache_validity & 1991 (NFS_INO_INVALID_OTHER 1992 | NFS_INO_REVAL_FORCED); 1993 cache_revalidated = false; 1994 } 1995 1996 if (fattr->valid & NFS_ATTR_FATTR_OWNER) { 1997 if (!uid_eq(inode->i_uid, fattr->uid)) { 1998 invalid |= NFS_INO_INVALID_ACCESS 1999 | NFS_INO_INVALID_ACL; 2000 inode->i_uid = fattr->uid; 2001 attr_changed = true; 2002 } 2003 } else if (server->caps & NFS_CAP_OWNER) { 2004 nfsi->cache_validity |= save_cache_validity & 2005 (NFS_INO_INVALID_OTHER 2006 | NFS_INO_REVAL_FORCED); 2007 cache_revalidated = false; 2008 } 2009 2010 if (fattr->valid & NFS_ATTR_FATTR_GROUP) { 2011 if (!gid_eq(inode->i_gid, fattr->gid)) { 2012 invalid |= NFS_INO_INVALID_ACCESS 2013 | NFS_INO_INVALID_ACL; 2014 inode->i_gid = fattr->gid; 2015 attr_changed = true; 2016 } 2017 } else if (server->caps & NFS_CAP_OWNER_GROUP) { 2018 nfsi->cache_validity |= save_cache_validity & 2019 (NFS_INO_INVALID_OTHER 2020 | NFS_INO_REVAL_FORCED); 2021 cache_revalidated = false; 2022 } 2023 2024 if (fattr->valid & NFS_ATTR_FATTR_NLINK) { 2025 if (inode->i_nlink != fattr->nlink) { 2026 if (S_ISDIR(inode->i_mode)) 2027 invalid |= NFS_INO_INVALID_DATA; 2028 set_nlink(inode, fattr->nlink); 2029 attr_changed = true; 2030 } 2031 } else if (server->caps & NFS_CAP_NLINK) { 2032 nfsi->cache_validity |= save_cache_validity & 2033 (NFS_INO_INVALID_OTHER 2034 | NFS_INO_REVAL_FORCED); 2035 cache_revalidated = false; 2036 } 2037 2038 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) { 2039 /* 2040 * report the blocks in 512byte units 2041 */ 2042 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 2043 } else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED) 2044 inode->i_blocks = fattr->du.nfs2.blocks; 2045 else { 2046 nfsi->cache_validity |= save_cache_validity & 2047 (NFS_INO_INVALID_BLOCKS 2048 | NFS_INO_REVAL_FORCED); 2049 cache_revalidated = false; 2050 } 2051 2052 /* Update attrtimeo value if we're out of the unstable period */ 2053 if (attr_changed) { 2054 invalid &= ~NFS_INO_INVALID_ATTR; 2055 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 2056 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 2057 nfsi->attrtimeo_timestamp = now; 2058 /* Set barrier to be more recent than all outstanding updates */ 2059 nfsi->attr_gencount = nfs_inc_attr_generation_counter(); 2060 } else { 2061 if (cache_revalidated) { 2062 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, 2063 nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) { 2064 nfsi->attrtimeo <<= 1; 2065 if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode)) 2066 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode); 2067 } 2068 nfsi->attrtimeo_timestamp = now; 2069 } 2070 /* Set the barrier to be more recent than this fattr */ 2071 if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0) 2072 nfsi->attr_gencount = fattr->gencount; 2073 } 2074 2075 /* Don't invalidate the data if we were to blame */ 2076 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) 2077 || S_ISLNK(inode->i_mode))) 2078 invalid &= ~NFS_INO_INVALID_DATA; 2079 nfs_set_cache_invalid(inode, invalid); 2080 2081 return 0; 2082 out_err: 2083 /* 2084 * No need to worry about unhashing the dentry, as the 2085 * lookup validation will know that the inode is bad. 2086 * (But we fall through to invalidate the caches.) 2087 */ 2088 nfs_set_inode_stale_locked(inode); 2089 return -ESTALE; 2090 } 2091 2092 struct inode *nfs_alloc_inode(struct super_block *sb) 2093 { 2094 struct nfs_inode *nfsi; 2095 nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL); 2096 if (!nfsi) 2097 return NULL; 2098 nfsi->flags = 0UL; 2099 nfsi->cache_validity = 0UL; 2100 #if IS_ENABLED(CONFIG_NFS_V4) 2101 nfsi->nfs4_acl = NULL; 2102 #endif /* CONFIG_NFS_V4 */ 2103 #ifdef CONFIG_NFS_V4_2 2104 nfsi->xattr_cache = NULL; 2105 #endif 2106 return &nfsi->vfs_inode; 2107 } 2108 EXPORT_SYMBOL_GPL(nfs_alloc_inode); 2109 2110 void nfs_free_inode(struct inode *inode) 2111 { 2112 kmem_cache_free(nfs_inode_cachep, NFS_I(inode)); 2113 } 2114 EXPORT_SYMBOL_GPL(nfs_free_inode); 2115 2116 static inline void nfs4_init_once(struct nfs_inode *nfsi) 2117 { 2118 #if IS_ENABLED(CONFIG_NFS_V4) 2119 INIT_LIST_HEAD(&nfsi->open_states); 2120 nfsi->delegation = NULL; 2121 init_rwsem(&nfsi->rwsem); 2122 nfsi->layout = NULL; 2123 #endif 2124 } 2125 2126 static void init_once(void *foo) 2127 { 2128 struct nfs_inode *nfsi = (struct nfs_inode *) foo; 2129 2130 inode_init_once(&nfsi->vfs_inode); 2131 INIT_LIST_HEAD(&nfsi->open_files); 2132 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru); 2133 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru); 2134 INIT_LIST_HEAD(&nfsi->commit_info.list); 2135 atomic_long_set(&nfsi->nrequests, 0); 2136 atomic_long_set(&nfsi->commit_info.ncommit, 0); 2137 atomic_set(&nfsi->commit_info.rpcs_out, 0); 2138 init_rwsem(&nfsi->rmdir_sem); 2139 mutex_init(&nfsi->commit_mutex); 2140 nfs4_init_once(nfsi); 2141 nfsi->cache_change_attribute = 0; 2142 } 2143 2144 static int __init nfs_init_inodecache(void) 2145 { 2146 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache", 2147 sizeof(struct nfs_inode), 2148 0, (SLAB_RECLAIM_ACCOUNT| 2149 SLAB_MEM_SPREAD|SLAB_ACCOUNT), 2150 init_once); 2151 if (nfs_inode_cachep == NULL) 2152 return -ENOMEM; 2153 2154 return 0; 2155 } 2156 2157 static void nfs_destroy_inodecache(void) 2158 { 2159 /* 2160 * Make sure all delayed rcu free inodes are flushed before we 2161 * destroy cache. 2162 */ 2163 rcu_barrier(); 2164 kmem_cache_destroy(nfs_inode_cachep); 2165 } 2166 2167 struct workqueue_struct *nfsiod_workqueue; 2168 EXPORT_SYMBOL_GPL(nfsiod_workqueue); 2169 2170 /* 2171 * start up the nfsiod workqueue 2172 */ 2173 static int nfsiod_start(void) 2174 { 2175 struct workqueue_struct *wq; 2176 dprintk("RPC: creating workqueue nfsiod\n"); 2177 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0); 2178 if (wq == NULL) 2179 return -ENOMEM; 2180 nfsiod_workqueue = wq; 2181 return 0; 2182 } 2183 2184 /* 2185 * Destroy the nfsiod workqueue 2186 */ 2187 static void nfsiod_stop(void) 2188 { 2189 struct workqueue_struct *wq; 2190 2191 wq = nfsiod_workqueue; 2192 if (wq == NULL) 2193 return; 2194 nfsiod_workqueue = NULL; 2195 destroy_workqueue(wq); 2196 } 2197 2198 unsigned int nfs_net_id; 2199 EXPORT_SYMBOL_GPL(nfs_net_id); 2200 2201 static int nfs_net_init(struct net *net) 2202 { 2203 nfs_clients_init(net); 2204 return nfs_fs_proc_net_init(net); 2205 } 2206 2207 static void nfs_net_exit(struct net *net) 2208 { 2209 nfs_fs_proc_net_exit(net); 2210 nfs_clients_exit(net); 2211 } 2212 2213 static struct pernet_operations nfs_net_ops = { 2214 .init = nfs_net_init, 2215 .exit = nfs_net_exit, 2216 .id = &nfs_net_id, 2217 .size = sizeof(struct nfs_net), 2218 }; 2219 2220 /* 2221 * Initialize NFS 2222 */ 2223 static int __init init_nfs_fs(void) 2224 { 2225 int err; 2226 2227 err = nfs_sysfs_init(); 2228 if (err < 0) 2229 goto out10; 2230 2231 err = register_pernet_subsys(&nfs_net_ops); 2232 if (err < 0) 2233 goto out9; 2234 2235 err = nfs_fscache_register(); 2236 if (err < 0) 2237 goto out8; 2238 2239 err = nfsiod_start(); 2240 if (err) 2241 goto out7; 2242 2243 err = nfs_fs_proc_init(); 2244 if (err) 2245 goto out6; 2246 2247 err = nfs_init_nfspagecache(); 2248 if (err) 2249 goto out5; 2250 2251 err = nfs_init_inodecache(); 2252 if (err) 2253 goto out4; 2254 2255 err = nfs_init_readpagecache(); 2256 if (err) 2257 goto out3; 2258 2259 err = nfs_init_writepagecache(); 2260 if (err) 2261 goto out2; 2262 2263 err = nfs_init_directcache(); 2264 if (err) 2265 goto out1; 2266 2267 rpc_proc_register(&init_net, &nfs_rpcstat); 2268 2269 err = register_nfs_fs(); 2270 if (err) 2271 goto out0; 2272 2273 return 0; 2274 out0: 2275 rpc_proc_unregister(&init_net, "nfs"); 2276 nfs_destroy_directcache(); 2277 out1: 2278 nfs_destroy_writepagecache(); 2279 out2: 2280 nfs_destroy_readpagecache(); 2281 out3: 2282 nfs_destroy_inodecache(); 2283 out4: 2284 nfs_destroy_nfspagecache(); 2285 out5: 2286 nfs_fs_proc_exit(); 2287 out6: 2288 nfsiod_stop(); 2289 out7: 2290 nfs_fscache_unregister(); 2291 out8: 2292 unregister_pernet_subsys(&nfs_net_ops); 2293 out9: 2294 nfs_sysfs_exit(); 2295 out10: 2296 return err; 2297 } 2298 2299 static void __exit exit_nfs_fs(void) 2300 { 2301 nfs_destroy_directcache(); 2302 nfs_destroy_writepagecache(); 2303 nfs_destroy_readpagecache(); 2304 nfs_destroy_inodecache(); 2305 nfs_destroy_nfspagecache(); 2306 nfs_fscache_unregister(); 2307 unregister_pernet_subsys(&nfs_net_ops); 2308 rpc_proc_unregister(&init_net, "nfs"); 2309 unregister_nfs_fs(); 2310 nfs_fs_proc_exit(); 2311 nfsiod_stop(); 2312 nfs_sysfs_exit(); 2313 } 2314 2315 /* Not quite true; I just maintain it */ 2316 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>"); 2317 MODULE_LICENSE("GPL"); 2318 module_param(enable_ino64, bool, 0644); 2319 2320 module_init(init_nfs_fs) 2321 module_exit(exit_nfs_fs) 2322