1 /* 2 * linux/fs/nfs/inode.c 3 * 4 * Copyright (C) 1992 Rick Sladkey 5 * 6 * nfs inode and superblock handling functions 7 * 8 * Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some 9 * experimental NFS changes. Modularisation taken straight from SYS5 fs. 10 * 11 * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts. 12 * J.S.Peatfield@damtp.cam.ac.uk 13 * 14 */ 15 16 #include <linux/module.h> 17 #include <linux/init.h> 18 #include <linux/sched/signal.h> 19 #include <linux/time.h> 20 #include <linux/kernel.h> 21 #include <linux/mm.h> 22 #include <linux/string.h> 23 #include <linux/stat.h> 24 #include <linux/errno.h> 25 #include <linux/unistd.h> 26 #include <linux/sunrpc/clnt.h> 27 #include <linux/sunrpc/stats.h> 28 #include <linux/sunrpc/metrics.h> 29 #include <linux/nfs_fs.h> 30 #include <linux/nfs_mount.h> 31 #include <linux/nfs4_mount.h> 32 #include <linux/lockd/bind.h> 33 #include <linux/seq_file.h> 34 #include <linux/mount.h> 35 #include <linux/vfs.h> 36 #include <linux/inet.h> 37 #include <linux/nfs_xdr.h> 38 #include <linux/slab.h> 39 #include <linux/compat.h> 40 #include <linux/freezer.h> 41 #include <linux/uaccess.h> 42 #include <linux/iversion.h> 43 44 #include "nfs4_fs.h" 45 #include "callback.h" 46 #include "delegation.h" 47 #include "iostat.h" 48 #include "internal.h" 49 #include "fscache.h" 50 #include "pnfs.h" 51 #include "nfs.h" 52 #include "netns.h" 53 54 #include "nfstrace.h" 55 56 #define NFSDBG_FACILITY NFSDBG_VFS 57 58 #define NFS_64_BIT_INODE_NUMBERS_ENABLED 1 59 60 /* Default is to see 64-bit inode numbers */ 61 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED; 62 63 static void nfs_invalidate_inode(struct inode *); 64 static int nfs_update_inode(struct inode *, struct nfs_fattr *); 65 66 static struct kmem_cache * nfs_inode_cachep; 67 68 static inline unsigned long 69 nfs_fattr_to_ino_t(struct nfs_fattr *fattr) 70 { 71 return nfs_fileid_to_ino_t(fattr->fileid); 72 } 73 74 static int nfs_wait_killable(int mode) 75 { 76 freezable_schedule_unsafe(); 77 if (signal_pending_state(mode, current)) 78 return -ERESTARTSYS; 79 return 0; 80 } 81 82 int nfs_wait_bit_killable(struct wait_bit_key *key, int mode) 83 { 84 return nfs_wait_killable(mode); 85 } 86 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable); 87 88 /** 89 * nfs_compat_user_ino64 - returns the user-visible inode number 90 * @fileid: 64-bit fileid 91 * 92 * This function returns a 32-bit inode number if the boot parameter 93 * nfs.enable_ino64 is zero. 94 */ 95 u64 nfs_compat_user_ino64(u64 fileid) 96 { 97 #ifdef CONFIG_COMPAT 98 compat_ulong_t ino; 99 #else 100 unsigned long ino; 101 #endif 102 103 if (enable_ino64) 104 return fileid; 105 ino = fileid; 106 if (sizeof(ino) < sizeof(fileid)) 107 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8; 108 return ino; 109 } 110 111 int nfs_drop_inode(struct inode *inode) 112 { 113 return NFS_STALE(inode) || generic_drop_inode(inode); 114 } 115 EXPORT_SYMBOL_GPL(nfs_drop_inode); 116 117 void nfs_clear_inode(struct inode *inode) 118 { 119 /* 120 * The following should never happen... 121 */ 122 WARN_ON_ONCE(nfs_have_writebacks(inode)); 123 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files)); 124 nfs_zap_acl_cache(inode); 125 nfs_access_zap_cache(inode); 126 nfs_fscache_clear_inode(inode); 127 } 128 EXPORT_SYMBOL_GPL(nfs_clear_inode); 129 130 void nfs_evict_inode(struct inode *inode) 131 { 132 truncate_inode_pages_final(&inode->i_data); 133 clear_inode(inode); 134 nfs_clear_inode(inode); 135 } 136 137 int nfs_sync_inode(struct inode *inode) 138 { 139 inode_dio_wait(inode); 140 return nfs_wb_all(inode); 141 } 142 EXPORT_SYMBOL_GPL(nfs_sync_inode); 143 144 /** 145 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk 146 */ 147 int nfs_sync_mapping(struct address_space *mapping) 148 { 149 int ret = 0; 150 151 if (mapping->nrpages != 0) { 152 unmap_mapping_range(mapping, 0, 0, 0); 153 ret = nfs_wb_all(mapping->host); 154 } 155 return ret; 156 } 157 158 static int nfs_attribute_timeout(struct inode *inode) 159 { 160 struct nfs_inode *nfsi = NFS_I(inode); 161 162 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo); 163 } 164 165 static bool nfs_check_cache_invalid_delegated(struct inode *inode, unsigned long flags) 166 { 167 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity); 168 169 /* Special case for the pagecache or access cache */ 170 if (flags == NFS_INO_REVAL_PAGECACHE && 171 !(cache_validity & NFS_INO_REVAL_FORCED)) 172 return false; 173 return (cache_validity & flags) != 0; 174 } 175 176 static bool nfs_check_cache_invalid_not_delegated(struct inode *inode, unsigned long flags) 177 { 178 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity); 179 180 if ((cache_validity & flags) != 0) 181 return true; 182 if (nfs_attribute_timeout(inode)) 183 return true; 184 return false; 185 } 186 187 bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags) 188 { 189 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) 190 return nfs_check_cache_invalid_delegated(inode, flags); 191 192 return nfs_check_cache_invalid_not_delegated(inode, flags); 193 } 194 195 static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags) 196 { 197 struct nfs_inode *nfsi = NFS_I(inode); 198 bool have_delegation = NFS_PROTO(inode)->have_delegation(inode, FMODE_READ); 199 200 if (have_delegation) { 201 if (!(flags & NFS_INO_REVAL_FORCED)) 202 flags &= ~NFS_INO_INVALID_OTHER; 203 flags &= ~(NFS_INO_INVALID_CHANGE 204 | NFS_INO_INVALID_SIZE 205 | NFS_INO_REVAL_PAGECACHE); 206 } 207 208 if (inode->i_mapping->nrpages == 0) 209 flags &= ~NFS_INO_INVALID_DATA; 210 nfsi->cache_validity |= flags; 211 if (flags & NFS_INO_INVALID_DATA) 212 nfs_fscache_invalidate(inode); 213 } 214 215 /* 216 * Invalidate the local caches 217 */ 218 static void nfs_zap_caches_locked(struct inode *inode) 219 { 220 struct nfs_inode *nfsi = NFS_I(inode); 221 int mode = inode->i_mode; 222 223 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 224 225 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 226 nfsi->attrtimeo_timestamp = jiffies; 227 228 memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf)); 229 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) { 230 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR 231 | NFS_INO_INVALID_DATA 232 | NFS_INO_INVALID_ACCESS 233 | NFS_INO_INVALID_ACL 234 | NFS_INO_REVAL_PAGECACHE); 235 } else 236 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR 237 | NFS_INO_INVALID_ACCESS 238 | NFS_INO_INVALID_ACL 239 | NFS_INO_REVAL_PAGECACHE); 240 nfs_zap_label_cache_locked(nfsi); 241 } 242 243 void nfs_zap_caches(struct inode *inode) 244 { 245 spin_lock(&inode->i_lock); 246 nfs_zap_caches_locked(inode); 247 spin_unlock(&inode->i_lock); 248 } 249 250 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping) 251 { 252 if (mapping->nrpages != 0) { 253 spin_lock(&inode->i_lock); 254 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA); 255 spin_unlock(&inode->i_lock); 256 } 257 } 258 259 void nfs_zap_acl_cache(struct inode *inode) 260 { 261 void (*clear_acl_cache)(struct inode *); 262 263 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache; 264 if (clear_acl_cache != NULL) 265 clear_acl_cache(inode); 266 spin_lock(&inode->i_lock); 267 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL; 268 spin_unlock(&inode->i_lock); 269 } 270 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache); 271 272 void nfs_invalidate_atime(struct inode *inode) 273 { 274 spin_lock(&inode->i_lock); 275 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME); 276 spin_unlock(&inode->i_lock); 277 } 278 EXPORT_SYMBOL_GPL(nfs_invalidate_atime); 279 280 /* 281 * Invalidate, but do not unhash, the inode. 282 * NB: must be called with inode->i_lock held! 283 */ 284 static void nfs_invalidate_inode(struct inode *inode) 285 { 286 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); 287 nfs_zap_caches_locked(inode); 288 } 289 290 struct nfs_find_desc { 291 struct nfs_fh *fh; 292 struct nfs_fattr *fattr; 293 }; 294 295 /* 296 * In NFSv3 we can have 64bit inode numbers. In order to support 297 * this, and re-exported directories (also seen in NFSv2) 298 * we are forced to allow 2 different inodes to have the same 299 * i_ino. 300 */ 301 static int 302 nfs_find_actor(struct inode *inode, void *opaque) 303 { 304 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque; 305 struct nfs_fh *fh = desc->fh; 306 struct nfs_fattr *fattr = desc->fattr; 307 308 if (NFS_FILEID(inode) != fattr->fileid) 309 return 0; 310 if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode)) 311 return 0; 312 if (nfs_compare_fh(NFS_FH(inode), fh)) 313 return 0; 314 if (is_bad_inode(inode) || NFS_STALE(inode)) 315 return 0; 316 return 1; 317 } 318 319 static int 320 nfs_init_locked(struct inode *inode, void *opaque) 321 { 322 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque; 323 struct nfs_fattr *fattr = desc->fattr; 324 325 set_nfs_fileid(inode, fattr->fileid); 326 inode->i_mode = fattr->mode; 327 nfs_copy_fh(NFS_FH(inode), desc->fh); 328 return 0; 329 } 330 331 #ifdef CONFIG_NFS_V4_SECURITY_LABEL 332 static void nfs_clear_label_invalid(struct inode *inode) 333 { 334 spin_lock(&inode->i_lock); 335 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL; 336 spin_unlock(&inode->i_lock); 337 } 338 339 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr, 340 struct nfs4_label *label) 341 { 342 int error; 343 344 if (label == NULL) 345 return; 346 347 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) { 348 error = security_inode_notifysecctx(inode, label->label, 349 label->len); 350 if (error) 351 printk(KERN_ERR "%s() %s %d " 352 "security_inode_notifysecctx() %d\n", 353 __func__, 354 (char *)label->label, 355 label->len, error); 356 nfs_clear_label_invalid(inode); 357 } 358 } 359 360 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags) 361 { 362 struct nfs4_label *label = NULL; 363 int minor_version = server->nfs_client->cl_minorversion; 364 365 if (minor_version < 2) 366 return label; 367 368 if (!(server->caps & NFS_CAP_SECURITY_LABEL)) 369 return label; 370 371 label = kzalloc(sizeof(struct nfs4_label), flags); 372 if (label == NULL) 373 return ERR_PTR(-ENOMEM); 374 375 label->label = kzalloc(NFS4_MAXLABELLEN, flags); 376 if (label->label == NULL) { 377 kfree(label); 378 return ERR_PTR(-ENOMEM); 379 } 380 label->len = NFS4_MAXLABELLEN; 381 382 return label; 383 } 384 EXPORT_SYMBOL_GPL(nfs4_label_alloc); 385 #else 386 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr, 387 struct nfs4_label *label) 388 { 389 } 390 #endif 391 EXPORT_SYMBOL_GPL(nfs_setsecurity); 392 393 /* Search for inode identified by fh, fileid and i_mode in inode cache. */ 394 struct inode * 395 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh) 396 { 397 struct nfs_find_desc desc = { 398 .fh = fh, 399 .fattr = fattr, 400 }; 401 struct inode *inode; 402 unsigned long hash; 403 404 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) || 405 !(fattr->valid & NFS_ATTR_FATTR_TYPE)) 406 return NULL; 407 408 hash = nfs_fattr_to_ino_t(fattr); 409 inode = ilookup5(sb, hash, nfs_find_actor, &desc); 410 411 dprintk("%s: returning %p\n", __func__, inode); 412 return inode; 413 } 414 415 /* 416 * This is our front-end to iget that looks up inodes by file handle 417 * instead of inode number. 418 */ 419 struct inode * 420 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label) 421 { 422 struct nfs_find_desc desc = { 423 .fh = fh, 424 .fattr = fattr 425 }; 426 struct inode *inode = ERR_PTR(-ENOENT); 427 unsigned long hash; 428 429 nfs_attr_check_mountpoint(sb, fattr); 430 431 if (nfs_attr_use_mounted_on_fileid(fattr)) 432 fattr->fileid = fattr->mounted_on_fileid; 433 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0) 434 goto out_no_inode; 435 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0) 436 goto out_no_inode; 437 438 hash = nfs_fattr_to_ino_t(fattr); 439 440 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc); 441 if (inode == NULL) { 442 inode = ERR_PTR(-ENOMEM); 443 goto out_no_inode; 444 } 445 446 if (inode->i_state & I_NEW) { 447 struct nfs_inode *nfsi = NFS_I(inode); 448 unsigned long now = jiffies; 449 450 /* We set i_ino for the few things that still rely on it, 451 * such as stat(2) */ 452 inode->i_ino = hash; 453 454 /* We can't support update_atime(), since the server will reset it */ 455 inode->i_flags |= S_NOATIME|S_NOCMTIME; 456 inode->i_mode = fattr->mode; 457 nfsi->cache_validity = 0; 458 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0 459 && nfs_server_capable(inode, NFS_CAP_MODE)) 460 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER); 461 /* Why so? Because we want revalidate for devices/FIFOs, and 462 * that's precisely what we have in nfs_file_inode_operations. 463 */ 464 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops; 465 if (S_ISREG(inode->i_mode)) { 466 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops; 467 inode->i_data.a_ops = &nfs_file_aops; 468 } else if (S_ISDIR(inode->i_mode)) { 469 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops; 470 inode->i_fop = &nfs_dir_operations; 471 inode->i_data.a_ops = &nfs_dir_aops; 472 /* Deal with crossing mountpoints */ 473 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT || 474 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) { 475 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) 476 inode->i_op = &nfs_referral_inode_operations; 477 else 478 inode->i_op = &nfs_mountpoint_inode_operations; 479 inode->i_fop = NULL; 480 inode->i_flags |= S_AUTOMOUNT; 481 } 482 } else if (S_ISLNK(inode->i_mode)) { 483 inode->i_op = &nfs_symlink_inode_operations; 484 inode_nohighmem(inode); 485 } else 486 init_special_inode(inode, inode->i_mode, fattr->rdev); 487 488 memset(&inode->i_atime, 0, sizeof(inode->i_atime)); 489 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime)); 490 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime)); 491 inode_set_iversion_raw(inode, 0); 492 inode->i_size = 0; 493 clear_nlink(inode); 494 inode->i_uid = make_kuid(&init_user_ns, -2); 495 inode->i_gid = make_kgid(&init_user_ns, -2); 496 inode->i_blocks = 0; 497 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf)); 498 nfsi->write_io = 0; 499 nfsi->read_io = 0; 500 501 nfsi->read_cache_jiffies = fattr->time_start; 502 nfsi->attr_gencount = fattr->gencount; 503 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 504 inode->i_atime = timespec_to_timespec64(fattr->atime); 505 else if (nfs_server_capable(inode, NFS_CAP_ATIME)) 506 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME); 507 if (fattr->valid & NFS_ATTR_FATTR_MTIME) 508 inode->i_mtime = timespec_to_timespec64(fattr->mtime); 509 else if (nfs_server_capable(inode, NFS_CAP_MTIME)) 510 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME); 511 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 512 inode->i_ctime = timespec_to_timespec64(fattr->ctime); 513 else if (nfs_server_capable(inode, NFS_CAP_CTIME)) 514 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME); 515 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) 516 inode_set_iversion_raw(inode, fattr->change_attr); 517 else 518 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE); 519 if (fattr->valid & NFS_ATTR_FATTR_SIZE) 520 inode->i_size = nfs_size_to_loff_t(fattr->size); 521 else 522 nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE); 523 if (fattr->valid & NFS_ATTR_FATTR_NLINK) 524 set_nlink(inode, fattr->nlink); 525 else if (nfs_server_capable(inode, NFS_CAP_NLINK)) 526 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER); 527 if (fattr->valid & NFS_ATTR_FATTR_OWNER) 528 inode->i_uid = fattr->uid; 529 else if (nfs_server_capable(inode, NFS_CAP_OWNER)) 530 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER); 531 if (fattr->valid & NFS_ATTR_FATTR_GROUP) 532 inode->i_gid = fattr->gid; 533 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP)) 534 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER); 535 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED) 536 inode->i_blocks = fattr->du.nfs2.blocks; 537 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) { 538 /* 539 * report the blocks in 512byte units 540 */ 541 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 542 } 543 544 if (nfsi->cache_validity != 0) 545 nfsi->cache_validity |= NFS_INO_REVAL_FORCED; 546 547 nfs_setsecurity(inode, fattr, label); 548 549 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 550 nfsi->attrtimeo_timestamp = now; 551 nfsi->access_cache = RB_ROOT; 552 553 nfs_fscache_init_inode(inode); 554 555 unlock_new_inode(inode); 556 } else { 557 int err = nfs_refresh_inode(inode, fattr); 558 if (err < 0) { 559 iput(inode); 560 inode = ERR_PTR(err); 561 goto out_no_inode; 562 } 563 } 564 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n", 565 inode->i_sb->s_id, 566 (unsigned long long)NFS_FILEID(inode), 567 nfs_display_fhandle_hash(fh), 568 atomic_read(&inode->i_count)); 569 570 out: 571 return inode; 572 573 out_no_inode: 574 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode)); 575 goto out; 576 } 577 EXPORT_SYMBOL_GPL(nfs_fhget); 578 579 #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) 580 581 int 582 nfs_setattr(struct dentry *dentry, struct iattr *attr) 583 { 584 struct inode *inode = d_inode(dentry); 585 struct nfs_fattr *fattr; 586 int error = 0; 587 588 nfs_inc_stats(inode, NFSIOS_VFSSETATTR); 589 590 /* skip mode change if it's just for clearing setuid/setgid */ 591 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID)) 592 attr->ia_valid &= ~ATTR_MODE; 593 594 if (attr->ia_valid & ATTR_SIZE) { 595 BUG_ON(!S_ISREG(inode->i_mode)); 596 597 error = inode_newsize_ok(inode, attr->ia_size); 598 if (error) 599 return error; 600 601 if (attr->ia_size == i_size_read(inode)) 602 attr->ia_valid &= ~ATTR_SIZE; 603 } 604 605 /* Optimization: if the end result is no change, don't RPC */ 606 attr->ia_valid &= NFS_VALID_ATTRS; 607 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0) 608 return 0; 609 610 trace_nfs_setattr_enter(inode); 611 612 /* Write all dirty data */ 613 if (S_ISREG(inode->i_mode)) 614 nfs_sync_inode(inode); 615 616 fattr = nfs_alloc_fattr(); 617 if (fattr == NULL) { 618 error = -ENOMEM; 619 goto out; 620 } 621 622 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr); 623 if (error == 0) 624 error = nfs_refresh_inode(inode, fattr); 625 nfs_free_fattr(fattr); 626 out: 627 trace_nfs_setattr_exit(inode, error); 628 return error; 629 } 630 EXPORT_SYMBOL_GPL(nfs_setattr); 631 632 /** 633 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall 634 * @inode: inode of the file used 635 * @offset: file offset to start truncating 636 * 637 * This is a copy of the common vmtruncate, but with the locking 638 * corrected to take into account the fact that NFS requires 639 * inode->i_size to be updated under the inode->i_lock. 640 * Note: must be called with inode->i_lock held! 641 */ 642 static int nfs_vmtruncate(struct inode * inode, loff_t offset) 643 { 644 int err; 645 646 err = inode_newsize_ok(inode, offset); 647 if (err) 648 goto out; 649 650 i_size_write(inode, offset); 651 /* Optimisation */ 652 if (offset == 0) 653 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA; 654 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE; 655 656 spin_unlock(&inode->i_lock); 657 truncate_pagecache(inode, offset); 658 spin_lock(&inode->i_lock); 659 out: 660 return err; 661 } 662 663 /** 664 * nfs_setattr_update_inode - Update inode metadata after a setattr call. 665 * @inode: pointer to struct inode 666 * @attr: pointer to struct iattr 667 * @fattr: pointer to struct nfs_fattr 668 * 669 * Note: we do this in the *proc.c in order to ensure that 670 * it works for things like exclusive creates too. 671 */ 672 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr, 673 struct nfs_fattr *fattr) 674 { 675 /* Barrier: bump the attribute generation count. */ 676 nfs_fattr_set_barrier(fattr); 677 678 spin_lock(&inode->i_lock); 679 NFS_I(inode)->attr_gencount = fattr->gencount; 680 if ((attr->ia_valid & ATTR_SIZE) != 0) { 681 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME); 682 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC); 683 nfs_vmtruncate(inode, attr->ia_size); 684 } 685 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) { 686 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME; 687 if ((attr->ia_valid & ATTR_MODE) != 0) { 688 int mode = attr->ia_mode & S_IALLUGO; 689 mode |= inode->i_mode & ~S_IALLUGO; 690 inode->i_mode = mode; 691 } 692 if ((attr->ia_valid & ATTR_UID) != 0) 693 inode->i_uid = attr->ia_uid; 694 if ((attr->ia_valid & ATTR_GID) != 0) 695 inode->i_gid = attr->ia_gid; 696 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 697 inode->i_ctime = timespec_to_timespec64(fattr->ctime); 698 else 699 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE 700 | NFS_INO_INVALID_CTIME); 701 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS 702 | NFS_INO_INVALID_ACL); 703 } 704 if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) { 705 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME 706 | NFS_INO_INVALID_CTIME); 707 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 708 inode->i_atime = timespec_to_timespec64(fattr->atime); 709 else if (attr->ia_valid & ATTR_ATIME_SET) 710 inode->i_atime = attr->ia_atime; 711 else 712 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME); 713 714 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 715 inode->i_ctime = timespec_to_timespec64(fattr->ctime); 716 else 717 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE 718 | NFS_INO_INVALID_CTIME); 719 } 720 if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) { 721 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME 722 | NFS_INO_INVALID_CTIME); 723 if (fattr->valid & NFS_ATTR_FATTR_MTIME) 724 inode->i_mtime = timespec_to_timespec64(fattr->mtime); 725 else if (attr->ia_valid & ATTR_MTIME_SET) 726 inode->i_mtime = attr->ia_mtime; 727 else 728 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME); 729 730 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 731 inode->i_ctime = timespec_to_timespec64(fattr->ctime); 732 else 733 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE 734 | NFS_INO_INVALID_CTIME); 735 } 736 if (fattr->valid) 737 nfs_update_inode(inode, fattr); 738 spin_unlock(&inode->i_lock); 739 } 740 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode); 741 742 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry) 743 { 744 struct dentry *parent; 745 746 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS)) 747 return; 748 parent = dget_parent(dentry); 749 nfs_force_use_readdirplus(d_inode(parent)); 750 dput(parent); 751 } 752 753 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry) 754 { 755 struct dentry *parent; 756 757 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS)) 758 return; 759 parent = dget_parent(dentry); 760 nfs_advise_use_readdirplus(d_inode(parent)); 761 dput(parent); 762 } 763 764 static bool nfs_need_revalidate_inode(struct inode *inode) 765 { 766 if (NFS_I(inode)->cache_validity & 767 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL)) 768 return true; 769 if (nfs_attribute_cache_expired(inode)) 770 return true; 771 return false; 772 } 773 774 int nfs_getattr(const struct path *path, struct kstat *stat, 775 u32 request_mask, unsigned int query_flags) 776 { 777 struct inode *inode = d_inode(path->dentry); 778 struct nfs_server *server = NFS_SERVER(inode); 779 unsigned long cache_validity; 780 int err = 0; 781 bool force_sync = query_flags & AT_STATX_FORCE_SYNC; 782 bool do_update = false; 783 784 trace_nfs_getattr_enter(inode); 785 786 if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) 787 goto out_no_update; 788 789 /* Flush out writes to the server in order to update c/mtime. */ 790 if ((request_mask & (STATX_CTIME|STATX_MTIME)) && 791 S_ISREG(inode->i_mode)) { 792 err = filemap_write_and_wait(inode->i_mapping); 793 if (err) 794 goto out; 795 } 796 797 /* 798 * We may force a getattr if the user cares about atime. 799 * 800 * Note that we only have to check the vfsmount flags here: 801 * - NFS always sets S_NOATIME by so checking it would give a 802 * bogus result 803 * - NFS never sets SB_NOATIME or SB_NODIRATIME so there is 804 * no point in checking those. 805 */ 806 if ((path->mnt->mnt_flags & MNT_NOATIME) || 807 ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) 808 request_mask &= ~STATX_ATIME; 809 810 /* Is the user requesting attributes that might need revalidation? */ 811 if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME| 812 STATX_MTIME|STATX_UID|STATX_GID| 813 STATX_SIZE|STATX_BLOCKS))) 814 goto out_no_revalidate; 815 816 /* Check whether the cached attributes are stale */ 817 do_update |= force_sync || nfs_attribute_cache_expired(inode); 818 cache_validity = READ_ONCE(NFS_I(inode)->cache_validity); 819 do_update |= cache_validity & 820 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL); 821 if (request_mask & STATX_ATIME) 822 do_update |= cache_validity & NFS_INO_INVALID_ATIME; 823 if (request_mask & (STATX_CTIME|STATX_MTIME)) 824 do_update |= cache_validity & NFS_INO_REVAL_PAGECACHE; 825 if (do_update) { 826 /* Update the attribute cache */ 827 if (!(server->flags & NFS_MOUNT_NOAC)) 828 nfs_readdirplus_parent_cache_miss(path->dentry); 829 else 830 nfs_readdirplus_parent_cache_hit(path->dentry); 831 err = __nfs_revalidate_inode(server, inode); 832 if (err) 833 goto out; 834 } else 835 nfs_readdirplus_parent_cache_hit(path->dentry); 836 out_no_revalidate: 837 /* Only return attributes that were revalidated. */ 838 stat->result_mask &= request_mask; 839 out_no_update: 840 generic_fillattr(inode, stat); 841 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode)); 842 if (S_ISDIR(inode->i_mode)) 843 stat->blksize = NFS_SERVER(inode)->dtsize; 844 out: 845 trace_nfs_getattr_exit(inode, err); 846 return err; 847 } 848 EXPORT_SYMBOL_GPL(nfs_getattr); 849 850 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx) 851 { 852 refcount_set(&l_ctx->count, 1); 853 l_ctx->lockowner = current->files; 854 INIT_LIST_HEAD(&l_ctx->list); 855 atomic_set(&l_ctx->io_count, 0); 856 } 857 858 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx) 859 { 860 struct nfs_lock_context *head = &ctx->lock_context; 861 struct nfs_lock_context *pos = head; 862 863 do { 864 if (pos->lockowner != current->files) 865 continue; 866 refcount_inc(&pos->count); 867 return pos; 868 } while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head); 869 return NULL; 870 } 871 872 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx) 873 { 874 struct nfs_lock_context *res, *new = NULL; 875 struct inode *inode = d_inode(ctx->dentry); 876 877 spin_lock(&inode->i_lock); 878 res = __nfs_find_lock_context(ctx); 879 if (res == NULL) { 880 spin_unlock(&inode->i_lock); 881 new = kmalloc(sizeof(*new), GFP_KERNEL); 882 if (new == NULL) 883 return ERR_PTR(-ENOMEM); 884 nfs_init_lock_context(new); 885 spin_lock(&inode->i_lock); 886 res = __nfs_find_lock_context(ctx); 887 if (res == NULL) { 888 list_add_tail(&new->list, &ctx->lock_context.list); 889 new->open_context = ctx; 890 res = new; 891 new = NULL; 892 } 893 } 894 spin_unlock(&inode->i_lock); 895 kfree(new); 896 return res; 897 } 898 EXPORT_SYMBOL_GPL(nfs_get_lock_context); 899 900 void nfs_put_lock_context(struct nfs_lock_context *l_ctx) 901 { 902 struct nfs_open_context *ctx = l_ctx->open_context; 903 struct inode *inode = d_inode(ctx->dentry); 904 905 if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock)) 906 return; 907 list_del(&l_ctx->list); 908 spin_unlock(&inode->i_lock); 909 kfree(l_ctx); 910 } 911 EXPORT_SYMBOL_GPL(nfs_put_lock_context); 912 913 /** 914 * nfs_close_context - Common close_context() routine NFSv2/v3 915 * @ctx: pointer to context 916 * @is_sync: is this a synchronous close 917 * 918 * Ensure that the attributes are up to date if we're mounted 919 * with close-to-open semantics and we have cached data that will 920 * need to be revalidated on open. 921 */ 922 void nfs_close_context(struct nfs_open_context *ctx, int is_sync) 923 { 924 struct nfs_inode *nfsi; 925 struct inode *inode; 926 struct nfs_server *server; 927 928 if (!(ctx->mode & FMODE_WRITE)) 929 return; 930 if (!is_sync) 931 return; 932 inode = d_inode(ctx->dentry); 933 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) 934 return; 935 nfsi = NFS_I(inode); 936 if (inode->i_mapping->nrpages == 0) 937 return; 938 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 939 return; 940 if (!list_empty(&nfsi->open_files)) 941 return; 942 server = NFS_SERVER(inode); 943 if (server->flags & NFS_MOUNT_NOCTO) 944 return; 945 nfs_revalidate_inode(server, inode); 946 } 947 EXPORT_SYMBOL_GPL(nfs_close_context); 948 949 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, 950 fmode_t f_mode, 951 struct file *filp) 952 { 953 struct nfs_open_context *ctx; 954 struct rpc_cred *cred = rpc_lookup_cred(); 955 if (IS_ERR(cred)) 956 return ERR_CAST(cred); 957 958 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); 959 if (!ctx) { 960 put_rpccred(cred); 961 return ERR_PTR(-ENOMEM); 962 } 963 nfs_sb_active(dentry->d_sb); 964 ctx->dentry = dget(dentry); 965 ctx->cred = cred; 966 ctx->state = NULL; 967 ctx->mode = f_mode; 968 ctx->flags = 0; 969 ctx->error = 0; 970 ctx->flock_owner = (fl_owner_t)filp; 971 nfs_init_lock_context(&ctx->lock_context); 972 ctx->lock_context.open_context = ctx; 973 INIT_LIST_HEAD(&ctx->list); 974 ctx->mdsthreshold = NULL; 975 return ctx; 976 } 977 EXPORT_SYMBOL_GPL(alloc_nfs_open_context); 978 979 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx) 980 { 981 if (ctx != NULL) 982 refcount_inc(&ctx->lock_context.count); 983 return ctx; 984 } 985 EXPORT_SYMBOL_GPL(get_nfs_open_context); 986 987 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync) 988 { 989 struct inode *inode = d_inode(ctx->dentry); 990 struct super_block *sb = ctx->dentry->d_sb; 991 992 if (!list_empty(&ctx->list)) { 993 if (!refcount_dec_and_lock(&ctx->lock_context.count, &inode->i_lock)) 994 return; 995 list_del(&ctx->list); 996 spin_unlock(&inode->i_lock); 997 } else if (!refcount_dec_and_test(&ctx->lock_context.count)) 998 return; 999 if (inode != NULL) 1000 NFS_PROTO(inode)->close_context(ctx, is_sync); 1001 if (ctx->cred != NULL) 1002 put_rpccred(ctx->cred); 1003 dput(ctx->dentry); 1004 nfs_sb_deactive(sb); 1005 kfree(ctx->mdsthreshold); 1006 kfree(ctx); 1007 } 1008 1009 void put_nfs_open_context(struct nfs_open_context *ctx) 1010 { 1011 __put_nfs_open_context(ctx, 0); 1012 } 1013 EXPORT_SYMBOL_GPL(put_nfs_open_context); 1014 1015 static void put_nfs_open_context_sync(struct nfs_open_context *ctx) 1016 { 1017 __put_nfs_open_context(ctx, 1); 1018 } 1019 1020 /* 1021 * Ensure that mmap has a recent RPC credential for use when writing out 1022 * shared pages 1023 */ 1024 void nfs_inode_attach_open_context(struct nfs_open_context *ctx) 1025 { 1026 struct inode *inode = d_inode(ctx->dentry); 1027 struct nfs_inode *nfsi = NFS_I(inode); 1028 1029 spin_lock(&inode->i_lock); 1030 if (ctx->mode & FMODE_WRITE) 1031 list_add(&ctx->list, &nfsi->open_files); 1032 else 1033 list_add_tail(&ctx->list, &nfsi->open_files); 1034 spin_unlock(&inode->i_lock); 1035 } 1036 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context); 1037 1038 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx) 1039 { 1040 filp->private_data = get_nfs_open_context(ctx); 1041 if (list_empty(&ctx->list)) 1042 nfs_inode_attach_open_context(ctx); 1043 } 1044 EXPORT_SYMBOL_GPL(nfs_file_set_open_context); 1045 1046 /* 1047 * Given an inode, search for an open context with the desired characteristics 1048 */ 1049 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode) 1050 { 1051 struct nfs_inode *nfsi = NFS_I(inode); 1052 struct nfs_open_context *pos, *ctx = NULL; 1053 1054 spin_lock(&inode->i_lock); 1055 list_for_each_entry(pos, &nfsi->open_files, list) { 1056 if (cred != NULL && pos->cred != cred) 1057 continue; 1058 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode) 1059 continue; 1060 ctx = get_nfs_open_context(pos); 1061 break; 1062 } 1063 spin_unlock(&inode->i_lock); 1064 return ctx; 1065 } 1066 1067 void nfs_file_clear_open_context(struct file *filp) 1068 { 1069 struct nfs_open_context *ctx = nfs_file_open_context(filp); 1070 1071 if (ctx) { 1072 struct inode *inode = d_inode(ctx->dentry); 1073 1074 /* 1075 * We fatal error on write before. Try to writeback 1076 * every page again. 1077 */ 1078 if (ctx->error < 0) 1079 invalidate_inode_pages2(inode->i_mapping); 1080 filp->private_data = NULL; 1081 spin_lock(&inode->i_lock); 1082 list_move_tail(&ctx->list, &NFS_I(inode)->open_files); 1083 spin_unlock(&inode->i_lock); 1084 put_nfs_open_context_sync(ctx); 1085 } 1086 } 1087 1088 /* 1089 * These allocate and release file read/write context information. 1090 */ 1091 int nfs_open(struct inode *inode, struct file *filp) 1092 { 1093 struct nfs_open_context *ctx; 1094 1095 ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp); 1096 if (IS_ERR(ctx)) 1097 return PTR_ERR(ctx); 1098 nfs_file_set_open_context(filp, ctx); 1099 put_nfs_open_context(ctx); 1100 nfs_fscache_open_file(inode, filp); 1101 return 0; 1102 } 1103 1104 /* 1105 * This function is called whenever some part of NFS notices that 1106 * the cached attributes have to be refreshed. 1107 */ 1108 int 1109 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) 1110 { 1111 int status = -ESTALE; 1112 struct nfs4_label *label = NULL; 1113 struct nfs_fattr *fattr = NULL; 1114 struct nfs_inode *nfsi = NFS_I(inode); 1115 1116 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n", 1117 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode)); 1118 1119 trace_nfs_revalidate_inode_enter(inode); 1120 1121 if (is_bad_inode(inode)) 1122 goto out; 1123 if (NFS_STALE(inode)) 1124 goto out; 1125 1126 /* pNFS: Attributes aren't updated until we layoutcommit */ 1127 if (S_ISREG(inode->i_mode)) { 1128 status = pnfs_sync_inode(inode, false); 1129 if (status) 1130 goto out; 1131 } 1132 1133 status = -ENOMEM; 1134 fattr = nfs_alloc_fattr(); 1135 if (fattr == NULL) 1136 goto out; 1137 1138 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE); 1139 1140 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL); 1141 if (IS_ERR(label)) { 1142 status = PTR_ERR(label); 1143 goto out; 1144 } 1145 1146 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, 1147 label, inode); 1148 if (status != 0) { 1149 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n", 1150 inode->i_sb->s_id, 1151 (unsigned long long)NFS_FILEID(inode), status); 1152 if (status == -ESTALE) { 1153 nfs_zap_caches(inode); 1154 if (!S_ISDIR(inode->i_mode)) 1155 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); 1156 } 1157 goto err_out; 1158 } 1159 1160 status = nfs_refresh_inode(inode, fattr); 1161 if (status) { 1162 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n", 1163 inode->i_sb->s_id, 1164 (unsigned long long)NFS_FILEID(inode), status); 1165 goto err_out; 1166 } 1167 1168 if (nfsi->cache_validity & NFS_INO_INVALID_ACL) 1169 nfs_zap_acl_cache(inode); 1170 1171 nfs_setsecurity(inode, fattr, label); 1172 1173 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n", 1174 inode->i_sb->s_id, 1175 (unsigned long long)NFS_FILEID(inode)); 1176 1177 err_out: 1178 nfs4_label_free(label); 1179 out: 1180 nfs_free_fattr(fattr); 1181 trace_nfs_revalidate_inode_exit(inode, status); 1182 return status; 1183 } 1184 1185 int nfs_attribute_cache_expired(struct inode *inode) 1186 { 1187 if (nfs_have_delegated_attributes(inode)) 1188 return 0; 1189 return nfs_attribute_timeout(inode); 1190 } 1191 1192 /** 1193 * nfs_revalidate_inode - Revalidate the inode attributes 1194 * @server - pointer to nfs_server struct 1195 * @inode - pointer to inode struct 1196 * 1197 * Updates inode attribute information by retrieving the data from the server. 1198 */ 1199 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) 1200 { 1201 if (!nfs_need_revalidate_inode(inode)) 1202 return NFS_STALE(inode) ? -ESTALE : 0; 1203 return __nfs_revalidate_inode(server, inode); 1204 } 1205 EXPORT_SYMBOL_GPL(nfs_revalidate_inode); 1206 1207 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping) 1208 { 1209 struct nfs_inode *nfsi = NFS_I(inode); 1210 int ret; 1211 1212 if (mapping->nrpages != 0) { 1213 if (S_ISREG(inode->i_mode)) { 1214 ret = nfs_sync_mapping(mapping); 1215 if (ret < 0) 1216 return ret; 1217 } 1218 ret = invalidate_inode_pages2(mapping); 1219 if (ret < 0) 1220 return ret; 1221 } 1222 if (S_ISDIR(inode->i_mode)) { 1223 spin_lock(&inode->i_lock); 1224 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf)); 1225 spin_unlock(&inode->i_lock); 1226 } 1227 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE); 1228 nfs_fscache_wait_on_invalidate(inode); 1229 1230 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n", 1231 inode->i_sb->s_id, 1232 (unsigned long long)NFS_FILEID(inode)); 1233 return 0; 1234 } 1235 1236 bool nfs_mapping_need_revalidate_inode(struct inode *inode) 1237 { 1238 return nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE) || 1239 NFS_STALE(inode); 1240 } 1241 1242 int nfs_revalidate_mapping_rcu(struct inode *inode) 1243 { 1244 struct nfs_inode *nfsi = NFS_I(inode); 1245 unsigned long *bitlock = &nfsi->flags; 1246 int ret = 0; 1247 1248 if (IS_SWAPFILE(inode)) 1249 goto out; 1250 if (nfs_mapping_need_revalidate_inode(inode)) { 1251 ret = -ECHILD; 1252 goto out; 1253 } 1254 spin_lock(&inode->i_lock); 1255 if (test_bit(NFS_INO_INVALIDATING, bitlock) || 1256 (nfsi->cache_validity & NFS_INO_INVALID_DATA)) 1257 ret = -ECHILD; 1258 spin_unlock(&inode->i_lock); 1259 out: 1260 return ret; 1261 } 1262 1263 /** 1264 * nfs_revalidate_mapping - Revalidate the pagecache 1265 * @inode - pointer to host inode 1266 * @mapping - pointer to mapping 1267 */ 1268 int nfs_revalidate_mapping(struct inode *inode, 1269 struct address_space *mapping) 1270 { 1271 struct nfs_inode *nfsi = NFS_I(inode); 1272 unsigned long *bitlock = &nfsi->flags; 1273 int ret = 0; 1274 1275 /* swapfiles are not supposed to be shared. */ 1276 if (IS_SWAPFILE(inode)) 1277 goto out; 1278 1279 if (nfs_mapping_need_revalidate_inode(inode)) { 1280 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 1281 if (ret < 0) 1282 goto out; 1283 } 1284 1285 /* 1286 * We must clear NFS_INO_INVALID_DATA first to ensure that 1287 * invalidations that come in while we're shooting down the mappings 1288 * are respected. But, that leaves a race window where one revalidator 1289 * can clear the flag, and then another checks it before the mapping 1290 * gets invalidated. Fix that by serializing access to this part of 1291 * the function. 1292 * 1293 * At the same time, we need to allow other tasks to see whether we 1294 * might be in the middle of invalidating the pages, so we only set 1295 * the bit lock here if it looks like we're going to be doing that. 1296 */ 1297 for (;;) { 1298 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING, 1299 nfs_wait_bit_killable, TASK_KILLABLE); 1300 if (ret) 1301 goto out; 1302 spin_lock(&inode->i_lock); 1303 if (test_bit(NFS_INO_INVALIDATING, bitlock)) { 1304 spin_unlock(&inode->i_lock); 1305 continue; 1306 } 1307 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 1308 break; 1309 spin_unlock(&inode->i_lock); 1310 goto out; 1311 } 1312 1313 set_bit(NFS_INO_INVALIDATING, bitlock); 1314 smp_wmb(); 1315 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA; 1316 spin_unlock(&inode->i_lock); 1317 trace_nfs_invalidate_mapping_enter(inode); 1318 ret = nfs_invalidate_mapping(inode, mapping); 1319 trace_nfs_invalidate_mapping_exit(inode, ret); 1320 1321 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock); 1322 smp_mb__after_atomic(); 1323 wake_up_bit(bitlock, NFS_INO_INVALIDATING); 1324 out: 1325 return ret; 1326 } 1327 1328 static bool nfs_file_has_writers(struct nfs_inode *nfsi) 1329 { 1330 struct inode *inode = &nfsi->vfs_inode; 1331 1332 assert_spin_locked(&inode->i_lock); 1333 1334 if (!S_ISREG(inode->i_mode)) 1335 return false; 1336 if (list_empty(&nfsi->open_files)) 1337 return false; 1338 /* Note: This relies on nfsi->open_files being ordered with writers 1339 * being placed at the head of the list. 1340 * See nfs_inode_attach_open_context() 1341 */ 1342 return (list_first_entry(&nfsi->open_files, 1343 struct nfs_open_context, 1344 list)->mode & FMODE_WRITE) == FMODE_WRITE; 1345 } 1346 1347 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi) 1348 { 1349 return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi); 1350 } 1351 1352 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1353 { 1354 struct timespec ts; 1355 1356 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE) 1357 && (fattr->valid & NFS_ATTR_FATTR_CHANGE) 1358 && inode_eq_iversion_raw(inode, fattr->pre_change_attr)) { 1359 inode_set_iversion_raw(inode, fattr->change_attr); 1360 if (S_ISDIR(inode->i_mode)) 1361 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA); 1362 } 1363 /* If we have atomic WCC data, we may update some attributes */ 1364 ts = timespec64_to_timespec(inode->i_ctime); 1365 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME) 1366 && (fattr->valid & NFS_ATTR_FATTR_CTIME) 1367 && timespec_equal(&ts, &fattr->pre_ctime)) { 1368 inode->i_ctime = timespec_to_timespec64(fattr->ctime); 1369 } 1370 1371 ts = timespec64_to_timespec(inode->i_mtime); 1372 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME) 1373 && (fattr->valid & NFS_ATTR_FATTR_MTIME) 1374 && timespec_equal(&ts, &fattr->pre_mtime)) { 1375 inode->i_mtime = timespec_to_timespec64(fattr->mtime); 1376 if (S_ISDIR(inode->i_mode)) 1377 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA); 1378 } 1379 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE) 1380 && (fattr->valid & NFS_ATTR_FATTR_SIZE) 1381 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size) 1382 && !nfs_have_writebacks(inode)) { 1383 i_size_write(inode, nfs_size_to_loff_t(fattr->size)); 1384 } 1385 } 1386 1387 /** 1388 * nfs_check_inode_attributes - verify consistency of the inode attribute cache 1389 * @inode - pointer to inode 1390 * @fattr - updated attributes 1391 * 1392 * Verifies the attribute cache. If we have just changed the attributes, 1393 * so that fattr carries weak cache consistency data, then it may 1394 * also update the ctime/mtime/change_attribute. 1395 */ 1396 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr) 1397 { 1398 struct nfs_inode *nfsi = NFS_I(inode); 1399 loff_t cur_size, new_isize; 1400 unsigned long invalid = 0; 1401 struct timespec ts; 1402 1403 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) 1404 return 0; 1405 1406 /* Has the inode gone and changed behind our back? */ 1407 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) 1408 return -ESTALE; 1409 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) 1410 return -ESTALE; 1411 1412 if (!nfs_file_has_buffered_writers(nfsi)) { 1413 /* Verify a few of the more important attributes */ 1414 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr)) 1415 invalid |= NFS_INO_INVALID_CHANGE 1416 | NFS_INO_REVAL_PAGECACHE; 1417 1418 ts = timespec64_to_timespec(inode->i_mtime); 1419 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&ts, &fattr->mtime)) 1420 invalid |= NFS_INO_INVALID_MTIME; 1421 1422 ts = timespec64_to_timespec(inode->i_ctime); 1423 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec_equal(&ts, &fattr->ctime)) 1424 invalid |= NFS_INO_INVALID_CTIME; 1425 1426 if (fattr->valid & NFS_ATTR_FATTR_SIZE) { 1427 cur_size = i_size_read(inode); 1428 new_isize = nfs_size_to_loff_t(fattr->size); 1429 if (cur_size != new_isize) 1430 invalid |= NFS_INO_INVALID_SIZE 1431 | NFS_INO_REVAL_PAGECACHE; 1432 } 1433 } 1434 1435 /* Have any file permissions changed? */ 1436 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) 1437 invalid |= NFS_INO_INVALID_ACCESS 1438 | NFS_INO_INVALID_ACL 1439 | NFS_INO_INVALID_OTHER; 1440 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid)) 1441 invalid |= NFS_INO_INVALID_ACCESS 1442 | NFS_INO_INVALID_ACL 1443 | NFS_INO_INVALID_OTHER; 1444 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid)) 1445 invalid |= NFS_INO_INVALID_ACCESS 1446 | NFS_INO_INVALID_ACL 1447 | NFS_INO_INVALID_OTHER; 1448 1449 /* Has the link count changed? */ 1450 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink) 1451 invalid |= NFS_INO_INVALID_OTHER; 1452 1453 ts = timespec64_to_timespec(inode->i_atime); 1454 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&ts, &fattr->atime)) 1455 invalid |= NFS_INO_INVALID_ATIME; 1456 1457 if (invalid != 0) 1458 nfs_set_cache_invalid(inode, invalid); 1459 1460 nfsi->read_cache_jiffies = fattr->time_start; 1461 return 0; 1462 } 1463 1464 static atomic_long_t nfs_attr_generation_counter; 1465 1466 static unsigned long nfs_read_attr_generation_counter(void) 1467 { 1468 return atomic_long_read(&nfs_attr_generation_counter); 1469 } 1470 1471 unsigned long nfs_inc_attr_generation_counter(void) 1472 { 1473 return atomic_long_inc_return(&nfs_attr_generation_counter); 1474 } 1475 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter); 1476 1477 void nfs_fattr_init(struct nfs_fattr *fattr) 1478 { 1479 fattr->valid = 0; 1480 fattr->time_start = jiffies; 1481 fattr->gencount = nfs_inc_attr_generation_counter(); 1482 fattr->owner_name = NULL; 1483 fattr->group_name = NULL; 1484 } 1485 EXPORT_SYMBOL_GPL(nfs_fattr_init); 1486 1487 /** 1488 * nfs_fattr_set_barrier 1489 * @fattr: attributes 1490 * 1491 * Used to set a barrier after an attribute was updated. This 1492 * barrier ensures that older attributes from RPC calls that may 1493 * have raced with our update cannot clobber these new values. 1494 * Note that you are still responsible for ensuring that other 1495 * operations which change the attribute on the server do not 1496 * collide. 1497 */ 1498 void nfs_fattr_set_barrier(struct nfs_fattr *fattr) 1499 { 1500 fattr->gencount = nfs_inc_attr_generation_counter(); 1501 } 1502 1503 struct nfs_fattr *nfs_alloc_fattr(void) 1504 { 1505 struct nfs_fattr *fattr; 1506 1507 fattr = kmalloc(sizeof(*fattr), GFP_NOFS); 1508 if (fattr != NULL) 1509 nfs_fattr_init(fattr); 1510 return fattr; 1511 } 1512 EXPORT_SYMBOL_GPL(nfs_alloc_fattr); 1513 1514 struct nfs_fh *nfs_alloc_fhandle(void) 1515 { 1516 struct nfs_fh *fh; 1517 1518 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS); 1519 if (fh != NULL) 1520 fh->size = 0; 1521 return fh; 1522 } 1523 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle); 1524 1525 #ifdef NFS_DEBUG 1526 /* 1527 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle 1528 * in the same way that wireshark does 1529 * 1530 * @fh: file handle 1531 * 1532 * For debugging only. 1533 */ 1534 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh) 1535 { 1536 /* wireshark uses 32-bit AUTODIN crc and does a bitwise 1537 * not on the result */ 1538 return nfs_fhandle_hash(fh); 1539 } 1540 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash); 1541 1542 /* 1543 * _nfs_display_fhandle - display an NFS file handle on the console 1544 * 1545 * @fh: file handle to display 1546 * @caption: display caption 1547 * 1548 * For debugging only. 1549 */ 1550 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption) 1551 { 1552 unsigned short i; 1553 1554 if (fh == NULL || fh->size == 0) { 1555 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh); 1556 return; 1557 } 1558 1559 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n", 1560 caption, fh, fh->size, _nfs_display_fhandle_hash(fh)); 1561 for (i = 0; i < fh->size; i += 16) { 1562 __be32 *pos = (__be32 *)&fh->data[i]; 1563 1564 switch ((fh->size - i - 1) >> 2) { 1565 case 0: 1566 printk(KERN_DEFAULT " %08x\n", 1567 be32_to_cpup(pos)); 1568 break; 1569 case 1: 1570 printk(KERN_DEFAULT " %08x %08x\n", 1571 be32_to_cpup(pos), be32_to_cpup(pos + 1)); 1572 break; 1573 case 2: 1574 printk(KERN_DEFAULT " %08x %08x %08x\n", 1575 be32_to_cpup(pos), be32_to_cpup(pos + 1), 1576 be32_to_cpup(pos + 2)); 1577 break; 1578 default: 1579 printk(KERN_DEFAULT " %08x %08x %08x %08x\n", 1580 be32_to_cpup(pos), be32_to_cpup(pos + 1), 1581 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3)); 1582 } 1583 } 1584 } 1585 EXPORT_SYMBOL_GPL(_nfs_display_fhandle); 1586 #endif 1587 1588 /** 1589 * nfs_inode_attrs_need_update - check if the inode attributes need updating 1590 * @inode - pointer to inode 1591 * @fattr - attributes 1592 * 1593 * Attempt to divine whether or not an RPC call reply carrying stale 1594 * attributes got scheduled after another call carrying updated ones. 1595 * 1596 * To do so, the function first assumes that a more recent ctime means 1597 * that the attributes in fattr are newer, however it also attempt to 1598 * catch the case where ctime either didn't change, or went backwards 1599 * (if someone reset the clock on the server) by looking at whether 1600 * or not this RPC call was started after the inode was last updated. 1601 * Note also the check for wraparound of 'attr_gencount' 1602 * 1603 * The function returns 'true' if it thinks the attributes in 'fattr' are 1604 * more recent than the ones cached in the inode. 1605 * 1606 */ 1607 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr) 1608 { 1609 const struct nfs_inode *nfsi = NFS_I(inode); 1610 1611 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 || 1612 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0); 1613 } 1614 1615 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr) 1616 { 1617 int ret; 1618 1619 trace_nfs_refresh_inode_enter(inode); 1620 1621 if (nfs_inode_attrs_need_update(inode, fattr)) 1622 ret = nfs_update_inode(inode, fattr); 1623 else 1624 ret = nfs_check_inode_attributes(inode, fattr); 1625 1626 trace_nfs_refresh_inode_exit(inode, ret); 1627 return ret; 1628 } 1629 1630 /** 1631 * nfs_refresh_inode - try to update the inode attribute cache 1632 * @inode - pointer to inode 1633 * @fattr - updated attributes 1634 * 1635 * Check that an RPC call that returned attributes has not overlapped with 1636 * other recent updates of the inode metadata, then decide whether it is 1637 * safe to do a full update of the inode attributes, or whether just to 1638 * call nfs_check_inode_attributes. 1639 */ 1640 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr) 1641 { 1642 int status; 1643 1644 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 1645 return 0; 1646 spin_lock(&inode->i_lock); 1647 status = nfs_refresh_inode_locked(inode, fattr); 1648 spin_unlock(&inode->i_lock); 1649 1650 return status; 1651 } 1652 EXPORT_SYMBOL_GPL(nfs_refresh_inode); 1653 1654 static int nfs_post_op_update_inode_locked(struct inode *inode, 1655 struct nfs_fattr *fattr, unsigned int invalid) 1656 { 1657 if (S_ISDIR(inode->i_mode)) 1658 invalid |= NFS_INO_INVALID_DATA; 1659 nfs_set_cache_invalid(inode, invalid); 1660 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 1661 return 0; 1662 return nfs_refresh_inode_locked(inode, fattr); 1663 } 1664 1665 /** 1666 * nfs_post_op_update_inode - try to update the inode attribute cache 1667 * @inode - pointer to inode 1668 * @fattr - updated attributes 1669 * 1670 * After an operation that has changed the inode metadata, mark the 1671 * attribute cache as being invalid, then try to update it. 1672 * 1673 * NB: if the server didn't return any post op attributes, this 1674 * function will force the retrieval of attributes before the next 1675 * NFS request. Thus it should be used only for operations that 1676 * are expected to change one or more attributes, to avoid 1677 * unnecessary NFS requests and trips through nfs_update_inode(). 1678 */ 1679 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1680 { 1681 int status; 1682 1683 spin_lock(&inode->i_lock); 1684 nfs_fattr_set_barrier(fattr); 1685 status = nfs_post_op_update_inode_locked(inode, fattr, 1686 NFS_INO_INVALID_CHANGE 1687 | NFS_INO_INVALID_CTIME 1688 | NFS_INO_REVAL_FORCED); 1689 spin_unlock(&inode->i_lock); 1690 1691 return status; 1692 } 1693 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode); 1694 1695 /** 1696 * nfs_post_op_update_inode_force_wcc_locked - 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. Fake up 1702 * weak cache consistency data, if none exist. 1703 * 1704 * This function is mainly designed to be used by the ->write_done() functions. 1705 */ 1706 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr) 1707 { 1708 int status; 1709 1710 /* Don't do a WCC update if these attributes are already stale */ 1711 if ((fattr->valid & NFS_ATTR_FATTR) == 0 || 1712 !nfs_inode_attrs_need_update(inode, fattr)) { 1713 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE 1714 | NFS_ATTR_FATTR_PRESIZE 1715 | NFS_ATTR_FATTR_PREMTIME 1716 | NFS_ATTR_FATTR_PRECTIME); 1717 goto out_noforce; 1718 } 1719 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && 1720 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) { 1721 fattr->pre_change_attr = inode_peek_iversion_raw(inode); 1722 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE; 1723 } 1724 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 && 1725 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) { 1726 fattr->pre_ctime = timespec64_to_timespec(inode->i_ctime); 1727 fattr->valid |= NFS_ATTR_FATTR_PRECTIME; 1728 } 1729 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 && 1730 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) { 1731 fattr->pre_mtime = timespec64_to_timespec(inode->i_mtime); 1732 fattr->valid |= NFS_ATTR_FATTR_PREMTIME; 1733 } 1734 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 && 1735 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) { 1736 fattr->pre_size = i_size_read(inode); 1737 fattr->valid |= NFS_ATTR_FATTR_PRESIZE; 1738 } 1739 out_noforce: 1740 status = nfs_post_op_update_inode_locked(inode, fattr, 1741 NFS_INO_INVALID_CHANGE 1742 | NFS_INO_INVALID_CTIME 1743 | NFS_INO_INVALID_MTIME); 1744 return status; 1745 } 1746 1747 /** 1748 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache 1749 * @inode - pointer to inode 1750 * @fattr - updated attributes 1751 * 1752 * After an operation that has changed the inode metadata, mark the 1753 * attribute cache as being invalid, then try to update it. Fake up 1754 * weak cache consistency data, if none exist. 1755 * 1756 * This function is mainly designed to be used by the ->write_done() functions. 1757 */ 1758 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr) 1759 { 1760 int status; 1761 1762 spin_lock(&inode->i_lock); 1763 nfs_fattr_set_barrier(fattr); 1764 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 1765 spin_unlock(&inode->i_lock); 1766 return status; 1767 } 1768 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc); 1769 1770 1771 static inline bool nfs_fileid_valid(struct nfs_inode *nfsi, 1772 struct nfs_fattr *fattr) 1773 { 1774 bool ret1 = true, ret2 = true; 1775 1776 if (fattr->valid & NFS_ATTR_FATTR_FILEID) 1777 ret1 = (nfsi->fileid == fattr->fileid); 1778 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) 1779 ret2 = (nfsi->fileid == fattr->mounted_on_fileid); 1780 return ret1 || ret2; 1781 } 1782 1783 /* 1784 * Many nfs protocol calls return the new file attributes after 1785 * an operation. Here we update the inode to reflect the state 1786 * of the server's inode. 1787 * 1788 * This is a bit tricky because we have to make sure all dirty pages 1789 * have been sent off to the server before calling invalidate_inode_pages. 1790 * To make sure no other process adds more write requests while we try 1791 * our best to flush them, we make them sleep during the attribute refresh. 1792 * 1793 * A very similar scenario holds for the dir cache. 1794 */ 1795 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1796 { 1797 struct nfs_server *server; 1798 struct nfs_inode *nfsi = NFS_I(inode); 1799 loff_t cur_isize, new_isize; 1800 unsigned long invalid = 0; 1801 unsigned long now = jiffies; 1802 unsigned long save_cache_validity; 1803 bool have_writers = nfs_file_has_buffered_writers(nfsi); 1804 bool cache_revalidated = true; 1805 bool attr_changed = false; 1806 bool have_delegation; 1807 1808 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n", 1809 __func__, inode->i_sb->s_id, inode->i_ino, 1810 nfs_display_fhandle_hash(NFS_FH(inode)), 1811 atomic_read(&inode->i_count), fattr->valid); 1812 1813 if (!nfs_fileid_valid(nfsi, fattr)) { 1814 printk(KERN_ERR "NFS: server %s error: fileid changed\n" 1815 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n", 1816 NFS_SERVER(inode)->nfs_client->cl_hostname, 1817 inode->i_sb->s_id, (long long)nfsi->fileid, 1818 (long long)fattr->fileid); 1819 goto out_err; 1820 } 1821 1822 /* 1823 * Make sure the inode's type hasn't changed. 1824 */ 1825 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) { 1826 /* 1827 * Big trouble! The inode has become a different object. 1828 */ 1829 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n", 1830 __func__, inode->i_ino, inode->i_mode, fattr->mode); 1831 goto out_err; 1832 } 1833 1834 server = NFS_SERVER(inode); 1835 /* Update the fsid? */ 1836 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) && 1837 !nfs_fsid_equal(&server->fsid, &fattr->fsid) && 1838 !IS_AUTOMOUNT(inode)) 1839 server->fsid = fattr->fsid; 1840 1841 /* Save the delegation state before clearing cache_validity */ 1842 have_delegation = nfs_have_delegated_attributes(inode); 1843 1844 /* 1845 * Update the read time so we don't revalidate too often. 1846 */ 1847 nfsi->read_cache_jiffies = fattr->time_start; 1848 1849 save_cache_validity = nfsi->cache_validity; 1850 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR 1851 | NFS_INO_INVALID_ATIME 1852 | NFS_INO_REVAL_FORCED 1853 | NFS_INO_REVAL_PAGECACHE); 1854 1855 /* Do atomic weak cache consistency updates */ 1856 nfs_wcc_update_inode(inode, fattr); 1857 1858 if (pnfs_layoutcommit_outstanding(inode)) { 1859 nfsi->cache_validity |= save_cache_validity & NFS_INO_INVALID_ATTR; 1860 cache_revalidated = false; 1861 } 1862 1863 /* More cache consistency checks */ 1864 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) { 1865 if (!inode_eq_iversion_raw(inode, fattr->change_attr)) { 1866 /* Could it be a race with writeback? */ 1867 if (!(have_writers || have_delegation)) { 1868 invalid |= NFS_INO_INVALID_DATA 1869 | NFS_INO_INVALID_ACCESS 1870 | NFS_INO_INVALID_ACL; 1871 /* Force revalidate of all attributes */ 1872 save_cache_validity |= NFS_INO_INVALID_CTIME 1873 | NFS_INO_INVALID_MTIME 1874 | NFS_INO_INVALID_SIZE 1875 | NFS_INO_INVALID_OTHER; 1876 if (S_ISDIR(inode->i_mode)) 1877 nfs_force_lookup_revalidate(inode); 1878 dprintk("NFS: change_attr change on server for file %s/%ld\n", 1879 inode->i_sb->s_id, 1880 inode->i_ino); 1881 } 1882 inode_set_iversion_raw(inode, fattr->change_attr); 1883 attr_changed = true; 1884 } 1885 } else { 1886 nfsi->cache_validity |= save_cache_validity & 1887 (NFS_INO_INVALID_CHANGE 1888 | NFS_INO_REVAL_PAGECACHE 1889 | NFS_INO_REVAL_FORCED); 1890 cache_revalidated = false; 1891 } 1892 1893 if (fattr->valid & NFS_ATTR_FATTR_MTIME) { 1894 inode->i_mtime = timespec_to_timespec64(fattr->mtime); 1895 } else if (server->caps & NFS_CAP_MTIME) { 1896 nfsi->cache_validity |= save_cache_validity & 1897 (NFS_INO_INVALID_MTIME 1898 | NFS_INO_REVAL_FORCED); 1899 cache_revalidated = false; 1900 } 1901 1902 if (fattr->valid & NFS_ATTR_FATTR_CTIME) { 1903 inode->i_ctime = timespec_to_timespec64(fattr->ctime); 1904 } else if (server->caps & NFS_CAP_CTIME) { 1905 nfsi->cache_validity |= save_cache_validity & 1906 (NFS_INO_INVALID_CTIME 1907 | NFS_INO_REVAL_FORCED); 1908 cache_revalidated = false; 1909 } 1910 1911 /* Check if our cached file size is stale */ 1912 if (fattr->valid & NFS_ATTR_FATTR_SIZE) { 1913 new_isize = nfs_size_to_loff_t(fattr->size); 1914 cur_isize = i_size_read(inode); 1915 if (new_isize != cur_isize && !have_delegation) { 1916 /* Do we perhaps have any outstanding writes, or has 1917 * the file grown beyond our last write? */ 1918 if (!nfs_have_writebacks(inode) || new_isize > cur_isize) { 1919 i_size_write(inode, new_isize); 1920 if (!have_writers) 1921 invalid |= NFS_INO_INVALID_DATA; 1922 attr_changed = true; 1923 } 1924 dprintk("NFS: isize change on server for file %s/%ld " 1925 "(%Ld to %Ld)\n", 1926 inode->i_sb->s_id, 1927 inode->i_ino, 1928 (long long)cur_isize, 1929 (long long)new_isize); 1930 } 1931 } else { 1932 nfsi->cache_validity |= save_cache_validity & 1933 (NFS_INO_INVALID_SIZE 1934 | NFS_INO_REVAL_PAGECACHE 1935 | NFS_INO_REVAL_FORCED); 1936 cache_revalidated = false; 1937 } 1938 1939 1940 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 1941 inode->i_atime = timespec_to_timespec64(fattr->atime); 1942 else if (server->caps & NFS_CAP_ATIME) { 1943 nfsi->cache_validity |= save_cache_validity & 1944 (NFS_INO_INVALID_ATIME 1945 | NFS_INO_REVAL_FORCED); 1946 cache_revalidated = false; 1947 } 1948 1949 if (fattr->valid & NFS_ATTR_FATTR_MODE) { 1950 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) { 1951 umode_t newmode = inode->i_mode & S_IFMT; 1952 newmode |= fattr->mode & S_IALLUGO; 1953 inode->i_mode = newmode; 1954 invalid |= NFS_INO_INVALID_ACCESS 1955 | NFS_INO_INVALID_ACL; 1956 attr_changed = true; 1957 } 1958 } else if (server->caps & NFS_CAP_MODE) { 1959 nfsi->cache_validity |= save_cache_validity & 1960 (NFS_INO_INVALID_OTHER 1961 | NFS_INO_REVAL_FORCED); 1962 cache_revalidated = false; 1963 } 1964 1965 if (fattr->valid & NFS_ATTR_FATTR_OWNER) { 1966 if (!uid_eq(inode->i_uid, fattr->uid)) { 1967 invalid |= NFS_INO_INVALID_ACCESS 1968 | NFS_INO_INVALID_ACL; 1969 inode->i_uid = fattr->uid; 1970 attr_changed = true; 1971 } 1972 } else if (server->caps & NFS_CAP_OWNER) { 1973 nfsi->cache_validity |= save_cache_validity & 1974 (NFS_INO_INVALID_OTHER 1975 | NFS_INO_REVAL_FORCED); 1976 cache_revalidated = false; 1977 } 1978 1979 if (fattr->valid & NFS_ATTR_FATTR_GROUP) { 1980 if (!gid_eq(inode->i_gid, fattr->gid)) { 1981 invalid |= NFS_INO_INVALID_ACCESS 1982 | NFS_INO_INVALID_ACL; 1983 inode->i_gid = fattr->gid; 1984 attr_changed = true; 1985 } 1986 } else if (server->caps & NFS_CAP_OWNER_GROUP) { 1987 nfsi->cache_validity |= save_cache_validity & 1988 (NFS_INO_INVALID_OTHER 1989 | NFS_INO_REVAL_FORCED); 1990 cache_revalidated = false; 1991 } 1992 1993 if (fattr->valid & NFS_ATTR_FATTR_NLINK) { 1994 if (inode->i_nlink != fattr->nlink) { 1995 if (S_ISDIR(inode->i_mode)) 1996 invalid |= NFS_INO_INVALID_DATA; 1997 set_nlink(inode, fattr->nlink); 1998 attr_changed = true; 1999 } 2000 } else if (server->caps & NFS_CAP_NLINK) { 2001 nfsi->cache_validity |= save_cache_validity & 2002 (NFS_INO_INVALID_OTHER 2003 | NFS_INO_REVAL_FORCED); 2004 cache_revalidated = false; 2005 } 2006 2007 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) { 2008 /* 2009 * report the blocks in 512byte units 2010 */ 2011 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 2012 } else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED) 2013 inode->i_blocks = fattr->du.nfs2.blocks; 2014 else 2015 cache_revalidated = false; 2016 2017 /* Update attrtimeo value if we're out of the unstable period */ 2018 if (attr_changed) { 2019 invalid &= ~NFS_INO_INVALID_ATTR; 2020 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 2021 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 2022 nfsi->attrtimeo_timestamp = now; 2023 /* Set barrier to be more recent than all outstanding updates */ 2024 nfsi->attr_gencount = nfs_inc_attr_generation_counter(); 2025 } else { 2026 if (cache_revalidated) { 2027 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, 2028 nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) { 2029 nfsi->attrtimeo <<= 1; 2030 if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode)) 2031 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode); 2032 } 2033 nfsi->attrtimeo_timestamp = now; 2034 } 2035 /* Set the barrier to be more recent than this fattr */ 2036 if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0) 2037 nfsi->attr_gencount = fattr->gencount; 2038 } 2039 2040 /* Don't invalidate the data if we were to blame */ 2041 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) 2042 || S_ISLNK(inode->i_mode))) 2043 invalid &= ~NFS_INO_INVALID_DATA; 2044 nfs_set_cache_invalid(inode, invalid); 2045 2046 return 0; 2047 out_err: 2048 /* 2049 * No need to worry about unhashing the dentry, as the 2050 * lookup validation will know that the inode is bad. 2051 * (But we fall through to invalidate the caches.) 2052 */ 2053 nfs_invalidate_inode(inode); 2054 return -ESTALE; 2055 } 2056 2057 struct inode *nfs_alloc_inode(struct super_block *sb) 2058 { 2059 struct nfs_inode *nfsi; 2060 nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL); 2061 if (!nfsi) 2062 return NULL; 2063 nfsi->flags = 0UL; 2064 nfsi->cache_validity = 0UL; 2065 #if IS_ENABLED(CONFIG_NFS_V4) 2066 nfsi->nfs4_acl = NULL; 2067 #endif /* CONFIG_NFS_V4 */ 2068 return &nfsi->vfs_inode; 2069 } 2070 EXPORT_SYMBOL_GPL(nfs_alloc_inode); 2071 2072 static void nfs_i_callback(struct rcu_head *head) 2073 { 2074 struct inode *inode = container_of(head, struct inode, i_rcu); 2075 kmem_cache_free(nfs_inode_cachep, NFS_I(inode)); 2076 } 2077 2078 void nfs_destroy_inode(struct inode *inode) 2079 { 2080 call_rcu(&inode->i_rcu, nfs_i_callback); 2081 } 2082 EXPORT_SYMBOL_GPL(nfs_destroy_inode); 2083 2084 static inline void nfs4_init_once(struct nfs_inode *nfsi) 2085 { 2086 #if IS_ENABLED(CONFIG_NFS_V4) 2087 INIT_LIST_HEAD(&nfsi->open_states); 2088 nfsi->delegation = NULL; 2089 init_rwsem(&nfsi->rwsem); 2090 nfsi->layout = NULL; 2091 #endif 2092 } 2093 2094 static void init_once(void *foo) 2095 { 2096 struct nfs_inode *nfsi = (struct nfs_inode *) foo; 2097 2098 inode_init_once(&nfsi->vfs_inode); 2099 INIT_LIST_HEAD(&nfsi->open_files); 2100 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru); 2101 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru); 2102 INIT_LIST_HEAD(&nfsi->commit_info.list); 2103 atomic_long_set(&nfsi->nrequests, 0); 2104 atomic_long_set(&nfsi->commit_info.ncommit, 0); 2105 atomic_set(&nfsi->commit_info.rpcs_out, 0); 2106 init_rwsem(&nfsi->rmdir_sem); 2107 mutex_init(&nfsi->commit_mutex); 2108 nfs4_init_once(nfsi); 2109 } 2110 2111 static int __init nfs_init_inodecache(void) 2112 { 2113 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache", 2114 sizeof(struct nfs_inode), 2115 0, (SLAB_RECLAIM_ACCOUNT| 2116 SLAB_MEM_SPREAD|SLAB_ACCOUNT), 2117 init_once); 2118 if (nfs_inode_cachep == NULL) 2119 return -ENOMEM; 2120 2121 return 0; 2122 } 2123 2124 static void nfs_destroy_inodecache(void) 2125 { 2126 /* 2127 * Make sure all delayed rcu free inodes are flushed before we 2128 * destroy cache. 2129 */ 2130 rcu_barrier(); 2131 kmem_cache_destroy(nfs_inode_cachep); 2132 } 2133 2134 struct workqueue_struct *nfsiod_workqueue; 2135 EXPORT_SYMBOL_GPL(nfsiod_workqueue); 2136 2137 /* 2138 * start up the nfsiod workqueue 2139 */ 2140 static int nfsiod_start(void) 2141 { 2142 struct workqueue_struct *wq; 2143 dprintk("RPC: creating workqueue nfsiod\n"); 2144 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0); 2145 if (wq == NULL) 2146 return -ENOMEM; 2147 nfsiod_workqueue = wq; 2148 return 0; 2149 } 2150 2151 /* 2152 * Destroy the nfsiod workqueue 2153 */ 2154 static void nfsiod_stop(void) 2155 { 2156 struct workqueue_struct *wq; 2157 2158 wq = nfsiod_workqueue; 2159 if (wq == NULL) 2160 return; 2161 nfsiod_workqueue = NULL; 2162 destroy_workqueue(wq); 2163 } 2164 2165 unsigned int nfs_net_id; 2166 EXPORT_SYMBOL_GPL(nfs_net_id); 2167 2168 static int nfs_net_init(struct net *net) 2169 { 2170 nfs_clients_init(net); 2171 return nfs_fs_proc_net_init(net); 2172 } 2173 2174 static void nfs_net_exit(struct net *net) 2175 { 2176 struct nfs_net *nn = net_generic(net, nfs_net_id); 2177 2178 nfs_fs_proc_net_exit(net); 2179 nfs_cleanup_cb_ident_idr(net); 2180 WARN_ON_ONCE(!list_empty(&nn->nfs_client_list)); 2181 WARN_ON_ONCE(!list_empty(&nn->nfs_volume_list)); 2182 } 2183 2184 static struct pernet_operations nfs_net_ops = { 2185 .init = nfs_net_init, 2186 .exit = nfs_net_exit, 2187 .id = &nfs_net_id, 2188 .size = sizeof(struct nfs_net), 2189 }; 2190 2191 /* 2192 * Initialize NFS 2193 */ 2194 static int __init init_nfs_fs(void) 2195 { 2196 int err; 2197 2198 err = register_pernet_subsys(&nfs_net_ops); 2199 if (err < 0) 2200 goto out9; 2201 2202 err = nfs_fscache_register(); 2203 if (err < 0) 2204 goto out8; 2205 2206 err = nfsiod_start(); 2207 if (err) 2208 goto out7; 2209 2210 err = nfs_fs_proc_init(); 2211 if (err) 2212 goto out6; 2213 2214 err = nfs_init_nfspagecache(); 2215 if (err) 2216 goto out5; 2217 2218 err = nfs_init_inodecache(); 2219 if (err) 2220 goto out4; 2221 2222 err = nfs_init_readpagecache(); 2223 if (err) 2224 goto out3; 2225 2226 err = nfs_init_writepagecache(); 2227 if (err) 2228 goto out2; 2229 2230 err = nfs_init_directcache(); 2231 if (err) 2232 goto out1; 2233 2234 rpc_proc_register(&init_net, &nfs_rpcstat); 2235 2236 err = register_nfs_fs(); 2237 if (err) 2238 goto out0; 2239 2240 return 0; 2241 out0: 2242 rpc_proc_unregister(&init_net, "nfs"); 2243 nfs_destroy_directcache(); 2244 out1: 2245 nfs_destroy_writepagecache(); 2246 out2: 2247 nfs_destroy_readpagecache(); 2248 out3: 2249 nfs_destroy_inodecache(); 2250 out4: 2251 nfs_destroy_nfspagecache(); 2252 out5: 2253 nfs_fs_proc_exit(); 2254 out6: 2255 nfsiod_stop(); 2256 out7: 2257 nfs_fscache_unregister(); 2258 out8: 2259 unregister_pernet_subsys(&nfs_net_ops); 2260 out9: 2261 return err; 2262 } 2263 2264 static void __exit exit_nfs_fs(void) 2265 { 2266 nfs_destroy_directcache(); 2267 nfs_destroy_writepagecache(); 2268 nfs_destroy_readpagecache(); 2269 nfs_destroy_inodecache(); 2270 nfs_destroy_nfspagecache(); 2271 nfs_fscache_unregister(); 2272 unregister_pernet_subsys(&nfs_net_ops); 2273 rpc_proc_unregister(&init_net, "nfs"); 2274 unregister_nfs_fs(); 2275 nfs_fs_proc_exit(); 2276 nfsiod_stop(); 2277 } 2278 2279 /* Not quite true; I just maintain it */ 2280 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>"); 2281 MODULE_LICENSE("GPL"); 2282 module_param(enable_ino64, bool, 0644); 2283 2284 module_init(init_nfs_fs) 2285 module_exit(exit_nfs_fs) 2286