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.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/crc32.h> 42 43 #include <asm/uaccess.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 "dns_resolve.h" 52 #include "pnfs.h" 53 #include "nfs.h" 54 #include "netns.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 /** 75 * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks 76 * @word: long word containing the bit lock 77 */ 78 int nfs_wait_bit_killable(void *word) 79 { 80 if (fatal_signal_pending(current)) 81 return -ERESTARTSYS; 82 freezable_schedule(); 83 return 0; 84 } 85 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable); 86 87 /** 88 * nfs_compat_user_ino64 - returns the user-visible inode number 89 * @fileid: 64-bit fileid 90 * 91 * This function returns a 32-bit inode number if the boot parameter 92 * nfs.enable_ino64 is zero. 93 */ 94 u64 nfs_compat_user_ino64(u64 fileid) 95 { 96 #ifdef CONFIG_COMPAT 97 compat_ulong_t ino; 98 #else 99 unsigned long ino; 100 #endif 101 102 if (enable_ino64) 103 return fileid; 104 ino = fileid; 105 if (sizeof(ino) < sizeof(fileid)) 106 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8; 107 return ino; 108 } 109 110 int nfs_drop_inode(struct inode *inode) 111 { 112 return NFS_STALE(inode) || generic_drop_inode(inode); 113 } 114 EXPORT_SYMBOL_GPL(nfs_drop_inode); 115 116 void nfs_clear_inode(struct inode *inode) 117 { 118 /* 119 * The following should never happen... 120 */ 121 WARN_ON_ONCE(nfs_have_writebacks(inode)); 122 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files)); 123 nfs_zap_acl_cache(inode); 124 nfs_access_zap_cache(inode); 125 nfs_fscache_release_inode_cookie(inode); 126 } 127 EXPORT_SYMBOL_GPL(nfs_clear_inode); 128 129 void nfs_evict_inode(struct inode *inode) 130 { 131 truncate_inode_pages(&inode->i_data, 0); 132 clear_inode(inode); 133 nfs_clear_inode(inode); 134 } 135 136 /** 137 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk 138 */ 139 int nfs_sync_mapping(struct address_space *mapping) 140 { 141 int ret = 0; 142 143 if (mapping->nrpages != 0) { 144 unmap_mapping_range(mapping, 0, 0, 0); 145 ret = nfs_wb_all(mapping->host); 146 } 147 return ret; 148 } 149 150 /* 151 * Invalidate the local caches 152 */ 153 static void nfs_zap_caches_locked(struct inode *inode) 154 { 155 struct nfs_inode *nfsi = NFS_I(inode); 156 int mode = inode->i_mode; 157 158 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 159 160 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 161 nfsi->attrtimeo_timestamp = jiffies; 162 163 memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf)); 164 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) { 165 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE; 166 nfs_fscache_invalidate(inode); 167 } else { 168 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE; 169 } 170 } 171 172 void nfs_zap_caches(struct inode *inode) 173 { 174 spin_lock(&inode->i_lock); 175 nfs_zap_caches_locked(inode); 176 spin_unlock(&inode->i_lock); 177 } 178 179 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping) 180 { 181 if (mapping->nrpages != 0) { 182 spin_lock(&inode->i_lock); 183 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; 184 nfs_fscache_invalidate(inode); 185 spin_unlock(&inode->i_lock); 186 } 187 } 188 189 void nfs_zap_acl_cache(struct inode *inode) 190 { 191 void (*clear_acl_cache)(struct inode *); 192 193 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache; 194 if (clear_acl_cache != NULL) 195 clear_acl_cache(inode); 196 spin_lock(&inode->i_lock); 197 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL; 198 spin_unlock(&inode->i_lock); 199 } 200 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache); 201 202 void nfs_invalidate_atime(struct inode *inode) 203 { 204 spin_lock(&inode->i_lock); 205 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME; 206 spin_unlock(&inode->i_lock); 207 } 208 EXPORT_SYMBOL_GPL(nfs_invalidate_atime); 209 210 /* 211 * Invalidate, but do not unhash, the inode. 212 * NB: must be called with inode->i_lock held! 213 */ 214 static void nfs_invalidate_inode(struct inode *inode) 215 { 216 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); 217 nfs_zap_caches_locked(inode); 218 } 219 220 struct nfs_find_desc { 221 struct nfs_fh *fh; 222 struct nfs_fattr *fattr; 223 }; 224 225 /* 226 * In NFSv3 we can have 64bit inode numbers. In order to support 227 * this, and re-exported directories (also seen in NFSv2) 228 * we are forced to allow 2 different inodes to have the same 229 * i_ino. 230 */ 231 static int 232 nfs_find_actor(struct inode *inode, void *opaque) 233 { 234 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque; 235 struct nfs_fh *fh = desc->fh; 236 struct nfs_fattr *fattr = desc->fattr; 237 238 if (NFS_FILEID(inode) != fattr->fileid) 239 return 0; 240 if (nfs_compare_fh(NFS_FH(inode), fh)) 241 return 0; 242 if (is_bad_inode(inode) || NFS_STALE(inode)) 243 return 0; 244 return 1; 245 } 246 247 static int 248 nfs_init_locked(struct inode *inode, void *opaque) 249 { 250 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque; 251 struct nfs_fattr *fattr = desc->fattr; 252 253 set_nfs_fileid(inode, fattr->fileid); 254 nfs_copy_fh(NFS_FH(inode), desc->fh); 255 return 0; 256 } 257 258 /* 259 * This is our front-end to iget that looks up inodes by file handle 260 * instead of inode number. 261 */ 262 struct inode * 263 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr) 264 { 265 struct nfs_find_desc desc = { 266 .fh = fh, 267 .fattr = fattr 268 }; 269 struct inode *inode = ERR_PTR(-ENOENT); 270 unsigned long hash; 271 272 nfs_attr_check_mountpoint(sb, fattr); 273 274 if (((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0) && 275 !nfs_attr_use_mounted_on_fileid(fattr)) 276 goto out_no_inode; 277 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0) 278 goto out_no_inode; 279 280 hash = nfs_fattr_to_ino_t(fattr); 281 282 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc); 283 if (inode == NULL) { 284 inode = ERR_PTR(-ENOMEM); 285 goto out_no_inode; 286 } 287 288 if (inode->i_state & I_NEW) { 289 struct nfs_inode *nfsi = NFS_I(inode); 290 unsigned long now = jiffies; 291 292 /* We set i_ino for the few things that still rely on it, 293 * such as stat(2) */ 294 inode->i_ino = hash; 295 296 /* We can't support update_atime(), since the server will reset it */ 297 inode->i_flags |= S_NOATIME|S_NOCMTIME; 298 inode->i_mode = fattr->mode; 299 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0 300 && nfs_server_capable(inode, NFS_CAP_MODE)) 301 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 302 /* Why so? Because we want revalidate for devices/FIFOs, and 303 * that's precisely what we have in nfs_file_inode_operations. 304 */ 305 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops; 306 if (S_ISREG(inode->i_mode)) { 307 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops; 308 inode->i_data.a_ops = &nfs_file_aops; 309 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info; 310 } else if (S_ISDIR(inode->i_mode)) { 311 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops; 312 inode->i_fop = &nfs_dir_operations; 313 inode->i_data.a_ops = &nfs_dir_aops; 314 /* Deal with crossing mountpoints */ 315 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT || 316 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) { 317 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) 318 inode->i_op = &nfs_referral_inode_operations; 319 else 320 inode->i_op = &nfs_mountpoint_inode_operations; 321 inode->i_fop = NULL; 322 inode->i_flags |= S_AUTOMOUNT; 323 } 324 } else if (S_ISLNK(inode->i_mode)) 325 inode->i_op = &nfs_symlink_inode_operations; 326 else 327 init_special_inode(inode, inode->i_mode, fattr->rdev); 328 329 memset(&inode->i_atime, 0, sizeof(inode->i_atime)); 330 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime)); 331 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime)); 332 inode->i_version = 0; 333 inode->i_size = 0; 334 clear_nlink(inode); 335 inode->i_uid = make_kuid(&init_user_ns, -2); 336 inode->i_gid = make_kgid(&init_user_ns, -2); 337 inode->i_blocks = 0; 338 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf)); 339 nfsi->write_io = 0; 340 nfsi->read_io = 0; 341 342 nfsi->read_cache_jiffies = fattr->time_start; 343 nfsi->attr_gencount = fattr->gencount; 344 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 345 inode->i_atime = fattr->atime; 346 else if (nfs_server_capable(inode, NFS_CAP_ATIME)) 347 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 348 if (fattr->valid & NFS_ATTR_FATTR_MTIME) 349 inode->i_mtime = fattr->mtime; 350 else if (nfs_server_capable(inode, NFS_CAP_MTIME)) 351 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 352 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 353 inode->i_ctime = fattr->ctime; 354 else if (nfs_server_capable(inode, NFS_CAP_CTIME)) 355 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 356 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) 357 inode->i_version = fattr->change_attr; 358 else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR)) 359 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 360 if (fattr->valid & NFS_ATTR_FATTR_SIZE) 361 inode->i_size = nfs_size_to_loff_t(fattr->size); 362 else 363 nfsi->cache_validity |= NFS_INO_INVALID_ATTR 364 | NFS_INO_REVAL_PAGECACHE; 365 if (fattr->valid & NFS_ATTR_FATTR_NLINK) 366 set_nlink(inode, fattr->nlink); 367 else if (nfs_server_capable(inode, NFS_CAP_NLINK)) 368 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 369 if (fattr->valid & NFS_ATTR_FATTR_OWNER) 370 inode->i_uid = fattr->uid; 371 else if (nfs_server_capable(inode, NFS_CAP_OWNER)) 372 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 373 if (fattr->valid & NFS_ATTR_FATTR_GROUP) 374 inode->i_gid = fattr->gid; 375 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP)) 376 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 377 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED) 378 inode->i_blocks = fattr->du.nfs2.blocks; 379 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) { 380 /* 381 * report the blocks in 512byte units 382 */ 383 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 384 } 385 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 386 nfsi->attrtimeo_timestamp = now; 387 nfsi->access_cache = RB_ROOT; 388 389 nfs_fscache_init_inode_cookie(inode); 390 391 unlock_new_inode(inode); 392 } else 393 nfs_refresh_inode(inode, fattr); 394 dprintk("NFS: nfs_fhget(%s/%Ld fh_crc=0x%08x ct=%d)\n", 395 inode->i_sb->s_id, 396 (long long)NFS_FILEID(inode), 397 nfs_display_fhandle_hash(fh), 398 atomic_read(&inode->i_count)); 399 400 out: 401 return inode; 402 403 out_no_inode: 404 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode)); 405 goto out; 406 } 407 EXPORT_SYMBOL_GPL(nfs_fhget); 408 409 #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) 410 411 int 412 nfs_setattr(struct dentry *dentry, struct iattr *attr) 413 { 414 struct inode *inode = dentry->d_inode; 415 struct nfs_fattr *fattr; 416 int error = -ENOMEM; 417 418 nfs_inc_stats(inode, NFSIOS_VFSSETATTR); 419 420 /* skip mode change if it's just for clearing setuid/setgid */ 421 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID)) 422 attr->ia_valid &= ~ATTR_MODE; 423 424 if (attr->ia_valid & ATTR_SIZE) { 425 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode)) 426 attr->ia_valid &= ~ATTR_SIZE; 427 } 428 429 /* Optimization: if the end result is no change, don't RPC */ 430 attr->ia_valid &= NFS_VALID_ATTRS; 431 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0) 432 return 0; 433 434 /* Write all dirty data */ 435 if (S_ISREG(inode->i_mode)) { 436 nfs_inode_dio_wait(inode); 437 nfs_wb_all(inode); 438 } 439 440 fattr = nfs_alloc_fattr(); 441 if (fattr == NULL) 442 goto out; 443 /* 444 * Return any delegations if we're going to change ACLs 445 */ 446 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) 447 NFS_PROTO(inode)->return_delegation(inode); 448 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr); 449 if (error == 0) 450 nfs_refresh_inode(inode, fattr); 451 nfs_free_fattr(fattr); 452 out: 453 return error; 454 } 455 EXPORT_SYMBOL_GPL(nfs_setattr); 456 457 /** 458 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall 459 * @inode: inode of the file used 460 * @offset: file offset to start truncating 461 * 462 * This is a copy of the common vmtruncate, but with the locking 463 * corrected to take into account the fact that NFS requires 464 * inode->i_size to be updated under the inode->i_lock. 465 */ 466 static int nfs_vmtruncate(struct inode * inode, loff_t offset) 467 { 468 loff_t oldsize; 469 int err; 470 471 err = inode_newsize_ok(inode, offset); 472 if (err) 473 goto out; 474 475 spin_lock(&inode->i_lock); 476 oldsize = inode->i_size; 477 i_size_write(inode, offset); 478 spin_unlock(&inode->i_lock); 479 480 truncate_pagecache(inode, oldsize, offset); 481 out: 482 return err; 483 } 484 485 /** 486 * nfs_setattr_update_inode - Update inode metadata after a setattr call. 487 * @inode: pointer to struct inode 488 * @attr: pointer to struct iattr 489 * 490 * Note: we do this in the *proc.c in order to ensure that 491 * it works for things like exclusive creates too. 492 */ 493 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr) 494 { 495 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) { 496 spin_lock(&inode->i_lock); 497 if ((attr->ia_valid & ATTR_MODE) != 0) { 498 int mode = attr->ia_mode & S_IALLUGO; 499 mode |= inode->i_mode & ~S_IALLUGO; 500 inode->i_mode = mode; 501 } 502 if ((attr->ia_valid & ATTR_UID) != 0) 503 inode->i_uid = attr->ia_uid; 504 if ((attr->ia_valid & ATTR_GID) != 0) 505 inode->i_gid = attr->ia_gid; 506 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 507 spin_unlock(&inode->i_lock); 508 } 509 if ((attr->ia_valid & ATTR_SIZE) != 0) { 510 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC); 511 nfs_vmtruncate(inode, attr->ia_size); 512 } 513 } 514 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode); 515 516 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat) 517 { 518 struct inode *inode = dentry->d_inode; 519 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME; 520 int err; 521 522 /* Flush out writes to the server in order to update c/mtime. */ 523 if (S_ISREG(inode->i_mode)) { 524 nfs_inode_dio_wait(inode); 525 err = filemap_write_and_wait(inode->i_mapping); 526 if (err) 527 goto out; 528 } 529 530 /* 531 * We may force a getattr if the user cares about atime. 532 * 533 * Note that we only have to check the vfsmount flags here: 534 * - NFS always sets S_NOATIME by so checking it would give a 535 * bogus result 536 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is 537 * no point in checking those. 538 */ 539 if ((mnt->mnt_flags & MNT_NOATIME) || 540 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) 541 need_atime = 0; 542 543 if (need_atime) 544 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 545 else 546 err = nfs_revalidate_inode(NFS_SERVER(inode), inode); 547 if (!err) { 548 generic_fillattr(inode, stat); 549 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode)); 550 } 551 out: 552 return err; 553 } 554 EXPORT_SYMBOL_GPL(nfs_getattr); 555 556 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx) 557 { 558 atomic_set(&l_ctx->count, 1); 559 l_ctx->lockowner.l_owner = current->files; 560 l_ctx->lockowner.l_pid = current->tgid; 561 INIT_LIST_HEAD(&l_ctx->list); 562 } 563 564 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx) 565 { 566 struct nfs_lock_context *pos; 567 568 list_for_each_entry(pos, &ctx->lock_context.list, list) { 569 if (pos->lockowner.l_owner != current->files) 570 continue; 571 if (pos->lockowner.l_pid != current->tgid) 572 continue; 573 atomic_inc(&pos->count); 574 return pos; 575 } 576 return NULL; 577 } 578 579 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx) 580 { 581 struct nfs_lock_context *res, *new = NULL; 582 struct inode *inode = ctx->dentry->d_inode; 583 584 spin_lock(&inode->i_lock); 585 res = __nfs_find_lock_context(ctx); 586 if (res == NULL) { 587 spin_unlock(&inode->i_lock); 588 new = kmalloc(sizeof(*new), GFP_KERNEL); 589 if (new == NULL) 590 return ERR_PTR(-ENOMEM); 591 nfs_init_lock_context(new); 592 spin_lock(&inode->i_lock); 593 res = __nfs_find_lock_context(ctx); 594 if (res == NULL) { 595 list_add_tail(&new->list, &ctx->lock_context.list); 596 new->open_context = ctx; 597 res = new; 598 new = NULL; 599 } 600 } 601 spin_unlock(&inode->i_lock); 602 kfree(new); 603 return res; 604 } 605 606 void nfs_put_lock_context(struct nfs_lock_context *l_ctx) 607 { 608 struct nfs_open_context *ctx = l_ctx->open_context; 609 struct inode *inode = ctx->dentry->d_inode; 610 611 if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock)) 612 return; 613 list_del(&l_ctx->list); 614 spin_unlock(&inode->i_lock); 615 kfree(l_ctx); 616 } 617 618 /** 619 * nfs_close_context - Common close_context() routine NFSv2/v3 620 * @ctx: pointer to context 621 * @is_sync: is this a synchronous close 622 * 623 * always ensure that the attributes are up to date if we're mounted 624 * with close-to-open semantics 625 */ 626 void nfs_close_context(struct nfs_open_context *ctx, int is_sync) 627 { 628 struct inode *inode; 629 struct nfs_server *server; 630 631 if (!(ctx->mode & FMODE_WRITE)) 632 return; 633 if (!is_sync) 634 return; 635 inode = ctx->dentry->d_inode; 636 if (!list_empty(&NFS_I(inode)->open_files)) 637 return; 638 server = NFS_SERVER(inode); 639 if (server->flags & NFS_MOUNT_NOCTO) 640 return; 641 nfs_revalidate_inode(server, inode); 642 } 643 EXPORT_SYMBOL_GPL(nfs_close_context); 644 645 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode) 646 { 647 struct nfs_open_context *ctx; 648 struct rpc_cred *cred = rpc_lookup_cred(); 649 if (IS_ERR(cred)) 650 return ERR_CAST(cred); 651 652 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); 653 if (!ctx) { 654 put_rpccred(cred); 655 return ERR_PTR(-ENOMEM); 656 } 657 nfs_sb_active(dentry->d_sb); 658 ctx->dentry = dget(dentry); 659 ctx->cred = cred; 660 ctx->state = NULL; 661 ctx->mode = f_mode; 662 ctx->flags = 0; 663 ctx->error = 0; 664 nfs_init_lock_context(&ctx->lock_context); 665 ctx->lock_context.open_context = ctx; 666 INIT_LIST_HEAD(&ctx->list); 667 ctx->mdsthreshold = NULL; 668 return ctx; 669 } 670 EXPORT_SYMBOL_GPL(alloc_nfs_open_context); 671 672 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx) 673 { 674 if (ctx != NULL) 675 atomic_inc(&ctx->lock_context.count); 676 return ctx; 677 } 678 EXPORT_SYMBOL_GPL(get_nfs_open_context); 679 680 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync) 681 { 682 struct inode *inode = ctx->dentry->d_inode; 683 struct super_block *sb = ctx->dentry->d_sb; 684 685 if (!list_empty(&ctx->list)) { 686 if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock)) 687 return; 688 list_del(&ctx->list); 689 spin_unlock(&inode->i_lock); 690 } else if (!atomic_dec_and_test(&ctx->lock_context.count)) 691 return; 692 if (inode != NULL) 693 NFS_PROTO(inode)->close_context(ctx, is_sync); 694 if (ctx->cred != NULL) 695 put_rpccred(ctx->cred); 696 dput(ctx->dentry); 697 nfs_sb_deactive(sb); 698 kfree(ctx->mdsthreshold); 699 kfree(ctx); 700 } 701 702 void put_nfs_open_context(struct nfs_open_context *ctx) 703 { 704 __put_nfs_open_context(ctx, 0); 705 } 706 EXPORT_SYMBOL_GPL(put_nfs_open_context); 707 708 /* 709 * Ensure that mmap has a recent RPC credential for use when writing out 710 * shared pages 711 */ 712 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx) 713 { 714 struct inode *inode = file_inode(filp); 715 struct nfs_inode *nfsi = NFS_I(inode); 716 717 filp->private_data = get_nfs_open_context(ctx); 718 spin_lock(&inode->i_lock); 719 list_add(&ctx->list, &nfsi->open_files); 720 spin_unlock(&inode->i_lock); 721 } 722 EXPORT_SYMBOL_GPL(nfs_file_set_open_context); 723 724 /* 725 * Given an inode, search for an open context with the desired characteristics 726 */ 727 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode) 728 { 729 struct nfs_inode *nfsi = NFS_I(inode); 730 struct nfs_open_context *pos, *ctx = NULL; 731 732 spin_lock(&inode->i_lock); 733 list_for_each_entry(pos, &nfsi->open_files, list) { 734 if (cred != NULL && pos->cred != cred) 735 continue; 736 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode) 737 continue; 738 ctx = get_nfs_open_context(pos); 739 break; 740 } 741 spin_unlock(&inode->i_lock); 742 return ctx; 743 } 744 745 static void nfs_file_clear_open_context(struct file *filp) 746 { 747 struct inode *inode = file_inode(filp); 748 struct nfs_open_context *ctx = nfs_file_open_context(filp); 749 750 if (ctx) { 751 filp->private_data = NULL; 752 spin_lock(&inode->i_lock); 753 list_move_tail(&ctx->list, &NFS_I(inode)->open_files); 754 spin_unlock(&inode->i_lock); 755 __put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1); 756 } 757 } 758 759 /* 760 * These allocate and release file read/write context information. 761 */ 762 int nfs_open(struct inode *inode, struct file *filp) 763 { 764 struct nfs_open_context *ctx; 765 766 ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode); 767 if (IS_ERR(ctx)) 768 return PTR_ERR(ctx); 769 nfs_file_set_open_context(filp, ctx); 770 put_nfs_open_context(ctx); 771 nfs_fscache_set_inode_cookie(inode, filp); 772 return 0; 773 } 774 775 int nfs_release(struct inode *inode, struct file *filp) 776 { 777 nfs_file_clear_open_context(filp); 778 return 0; 779 } 780 781 /* 782 * This function is called whenever some part of NFS notices that 783 * the cached attributes have to be refreshed. 784 */ 785 int 786 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) 787 { 788 int status = -ESTALE; 789 struct nfs_fattr *fattr = NULL; 790 struct nfs_inode *nfsi = NFS_I(inode); 791 792 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n", 793 inode->i_sb->s_id, (long long)NFS_FILEID(inode)); 794 795 if (is_bad_inode(inode)) 796 goto out; 797 if (NFS_STALE(inode)) 798 goto out; 799 800 status = -ENOMEM; 801 fattr = nfs_alloc_fattr(); 802 if (fattr == NULL) 803 goto out; 804 805 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE); 806 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr); 807 if (status != 0) { 808 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n", 809 inode->i_sb->s_id, 810 (long long)NFS_FILEID(inode), status); 811 if (status == -ESTALE) { 812 nfs_zap_caches(inode); 813 if (!S_ISDIR(inode->i_mode)) 814 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); 815 } 816 goto out; 817 } 818 819 status = nfs_refresh_inode(inode, fattr); 820 if (status) { 821 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n", 822 inode->i_sb->s_id, 823 (long long)NFS_FILEID(inode), status); 824 goto out; 825 } 826 827 if (nfsi->cache_validity & NFS_INO_INVALID_ACL) 828 nfs_zap_acl_cache(inode); 829 830 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n", 831 inode->i_sb->s_id, 832 (long long)NFS_FILEID(inode)); 833 834 out: 835 nfs_free_fattr(fattr); 836 return status; 837 } 838 839 int nfs_attribute_timeout(struct inode *inode) 840 { 841 struct nfs_inode *nfsi = NFS_I(inode); 842 843 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo); 844 } 845 846 static int nfs_attribute_cache_expired(struct inode *inode) 847 { 848 if (nfs_have_delegated_attributes(inode)) 849 return 0; 850 return nfs_attribute_timeout(inode); 851 } 852 853 /** 854 * nfs_revalidate_inode - Revalidate the inode attributes 855 * @server - pointer to nfs_server struct 856 * @inode - pointer to inode struct 857 * 858 * Updates inode attribute information by retrieving the data from the server. 859 */ 860 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) 861 { 862 if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR) 863 && !nfs_attribute_cache_expired(inode)) 864 return NFS_STALE(inode) ? -ESTALE : 0; 865 return __nfs_revalidate_inode(server, inode); 866 } 867 EXPORT_SYMBOL_GPL(nfs_revalidate_inode); 868 869 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping) 870 { 871 struct nfs_inode *nfsi = NFS_I(inode); 872 873 if (mapping->nrpages != 0) { 874 int ret = invalidate_inode_pages2(mapping); 875 if (ret < 0) 876 return ret; 877 } 878 spin_lock(&inode->i_lock); 879 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA; 880 if (S_ISDIR(inode->i_mode)) 881 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf)); 882 spin_unlock(&inode->i_lock); 883 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE); 884 nfs_fscache_wait_on_invalidate(inode); 885 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n", 886 inode->i_sb->s_id, (long long)NFS_FILEID(inode)); 887 return 0; 888 } 889 890 static bool nfs_mapping_need_revalidate_inode(struct inode *inode) 891 { 892 if (nfs_have_delegated_attributes(inode)) 893 return false; 894 return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE) 895 || nfs_attribute_timeout(inode) 896 || NFS_STALE(inode); 897 } 898 899 /** 900 * nfs_revalidate_mapping - Revalidate the pagecache 901 * @inode - pointer to host inode 902 * @mapping - pointer to mapping 903 */ 904 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping) 905 { 906 struct nfs_inode *nfsi = NFS_I(inode); 907 int ret = 0; 908 909 /* swapfiles are not supposed to be shared. */ 910 if (IS_SWAPFILE(inode)) 911 goto out; 912 913 if (nfs_mapping_need_revalidate_inode(inode)) { 914 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 915 if (ret < 0) 916 goto out; 917 } 918 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 919 ret = nfs_invalidate_mapping(inode, mapping); 920 out: 921 return ret; 922 } 923 924 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr) 925 { 926 struct nfs_inode *nfsi = NFS_I(inode); 927 unsigned long ret = 0; 928 929 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE) 930 && (fattr->valid & NFS_ATTR_FATTR_CHANGE) 931 && inode->i_version == fattr->pre_change_attr) { 932 inode->i_version = fattr->change_attr; 933 if (S_ISDIR(inode->i_mode)) 934 nfsi->cache_validity |= NFS_INO_INVALID_DATA; 935 ret |= NFS_INO_INVALID_ATTR; 936 } 937 /* If we have atomic WCC data, we may update some attributes */ 938 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME) 939 && (fattr->valid & NFS_ATTR_FATTR_CTIME) 940 && timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) { 941 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime)); 942 ret |= NFS_INO_INVALID_ATTR; 943 } 944 945 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME) 946 && (fattr->valid & NFS_ATTR_FATTR_MTIME) 947 && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) { 948 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime)); 949 if (S_ISDIR(inode->i_mode)) 950 nfsi->cache_validity |= NFS_INO_INVALID_DATA; 951 ret |= NFS_INO_INVALID_ATTR; 952 } 953 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE) 954 && (fattr->valid & NFS_ATTR_FATTR_SIZE) 955 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size) 956 && nfsi->npages == 0) { 957 i_size_write(inode, nfs_size_to_loff_t(fattr->size)); 958 ret |= NFS_INO_INVALID_ATTR; 959 } 960 961 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 962 nfs_fscache_invalidate(inode); 963 964 return ret; 965 } 966 967 /** 968 * nfs_check_inode_attributes - verify consistency of the inode attribute cache 969 * @inode - pointer to inode 970 * @fattr - updated attributes 971 * 972 * Verifies the attribute cache. If we have just changed the attributes, 973 * so that fattr carries weak cache consistency data, then it may 974 * also update the ctime/mtime/change_attribute. 975 */ 976 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr) 977 { 978 struct nfs_inode *nfsi = NFS_I(inode); 979 loff_t cur_size, new_isize; 980 unsigned long invalid = 0; 981 982 983 if (nfs_have_delegated_attributes(inode)) 984 return 0; 985 /* Has the inode gone and changed behind our back? */ 986 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) 987 return -EIO; 988 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) 989 return -EIO; 990 991 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && 992 inode->i_version != fattr->change_attr) 993 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE; 994 995 /* Verify a few of the more important attributes */ 996 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime)) 997 invalid |= NFS_INO_INVALID_ATTR; 998 999 if (fattr->valid & NFS_ATTR_FATTR_SIZE) { 1000 cur_size = i_size_read(inode); 1001 new_isize = nfs_size_to_loff_t(fattr->size); 1002 if (cur_size != new_isize && nfsi->npages == 0) 1003 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE; 1004 } 1005 1006 /* Have any file permissions changed? */ 1007 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) 1008 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL; 1009 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid)) 1010 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL; 1011 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid)) 1012 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL; 1013 1014 /* Has the link count changed? */ 1015 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink) 1016 invalid |= NFS_INO_INVALID_ATTR; 1017 1018 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime)) 1019 invalid |= NFS_INO_INVALID_ATIME; 1020 1021 if (invalid != 0) 1022 nfsi->cache_validity |= invalid; 1023 1024 nfsi->read_cache_jiffies = fattr->time_start; 1025 return 0; 1026 } 1027 1028 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr) 1029 { 1030 if (!(fattr->valid & NFS_ATTR_FATTR_CTIME)) 1031 return 0; 1032 return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0; 1033 } 1034 1035 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr) 1036 { 1037 if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) 1038 return 0; 1039 return nfs_size_to_loff_t(fattr->size) > i_size_read(inode); 1040 } 1041 1042 static atomic_long_t nfs_attr_generation_counter; 1043 1044 static unsigned long nfs_read_attr_generation_counter(void) 1045 { 1046 return atomic_long_read(&nfs_attr_generation_counter); 1047 } 1048 1049 unsigned long nfs_inc_attr_generation_counter(void) 1050 { 1051 return atomic_long_inc_return(&nfs_attr_generation_counter); 1052 } 1053 1054 void nfs_fattr_init(struct nfs_fattr *fattr) 1055 { 1056 fattr->valid = 0; 1057 fattr->time_start = jiffies; 1058 fattr->gencount = nfs_inc_attr_generation_counter(); 1059 fattr->owner_name = NULL; 1060 fattr->group_name = NULL; 1061 } 1062 EXPORT_SYMBOL_GPL(nfs_fattr_init); 1063 1064 struct nfs_fattr *nfs_alloc_fattr(void) 1065 { 1066 struct nfs_fattr *fattr; 1067 1068 fattr = kmalloc(sizeof(*fattr), GFP_NOFS); 1069 if (fattr != NULL) 1070 nfs_fattr_init(fattr); 1071 return fattr; 1072 } 1073 EXPORT_SYMBOL_GPL(nfs_alloc_fattr); 1074 1075 struct nfs_fh *nfs_alloc_fhandle(void) 1076 { 1077 struct nfs_fh *fh; 1078 1079 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS); 1080 if (fh != NULL) 1081 fh->size = 0; 1082 return fh; 1083 } 1084 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle); 1085 1086 #ifdef NFS_DEBUG 1087 /* 1088 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle 1089 * in the same way that wireshark does 1090 * 1091 * @fh: file handle 1092 * 1093 * For debugging only. 1094 */ 1095 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh) 1096 { 1097 /* wireshark uses 32-bit AUTODIN crc and does a bitwise 1098 * not on the result */ 1099 return ~crc32(0xFFFFFFFF, &fh->data[0], fh->size); 1100 } 1101 1102 /* 1103 * _nfs_display_fhandle - display an NFS file handle on the console 1104 * 1105 * @fh: file handle to display 1106 * @caption: display caption 1107 * 1108 * For debugging only. 1109 */ 1110 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption) 1111 { 1112 unsigned short i; 1113 1114 if (fh == NULL || fh->size == 0) { 1115 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh); 1116 return; 1117 } 1118 1119 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n", 1120 caption, fh, fh->size, _nfs_display_fhandle_hash(fh)); 1121 for (i = 0; i < fh->size; i += 16) { 1122 __be32 *pos = (__be32 *)&fh->data[i]; 1123 1124 switch ((fh->size - i - 1) >> 2) { 1125 case 0: 1126 printk(KERN_DEFAULT " %08x\n", 1127 be32_to_cpup(pos)); 1128 break; 1129 case 1: 1130 printk(KERN_DEFAULT " %08x %08x\n", 1131 be32_to_cpup(pos), be32_to_cpup(pos + 1)); 1132 break; 1133 case 2: 1134 printk(KERN_DEFAULT " %08x %08x %08x\n", 1135 be32_to_cpup(pos), be32_to_cpup(pos + 1), 1136 be32_to_cpup(pos + 2)); 1137 break; 1138 default: 1139 printk(KERN_DEFAULT " %08x %08x %08x %08x\n", 1140 be32_to_cpup(pos), be32_to_cpup(pos + 1), 1141 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3)); 1142 } 1143 } 1144 } 1145 #endif 1146 1147 /** 1148 * nfs_inode_attrs_need_update - check if the inode attributes need updating 1149 * @inode - pointer to inode 1150 * @fattr - attributes 1151 * 1152 * Attempt to divine whether or not an RPC call reply carrying stale 1153 * attributes got scheduled after another call carrying updated ones. 1154 * 1155 * To do so, the function first assumes that a more recent ctime means 1156 * that the attributes in fattr are newer, however it also attempt to 1157 * catch the case where ctime either didn't change, or went backwards 1158 * (if someone reset the clock on the server) by looking at whether 1159 * or not this RPC call was started after the inode was last updated. 1160 * Note also the check for wraparound of 'attr_gencount' 1161 * 1162 * The function returns 'true' if it thinks the attributes in 'fattr' are 1163 * more recent than the ones cached in the inode. 1164 * 1165 */ 1166 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr) 1167 { 1168 const struct nfs_inode *nfsi = NFS_I(inode); 1169 1170 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 || 1171 nfs_ctime_need_update(inode, fattr) || 1172 nfs_size_need_update(inode, fattr) || 1173 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0); 1174 } 1175 1176 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr) 1177 { 1178 if (nfs_inode_attrs_need_update(inode, fattr)) 1179 return nfs_update_inode(inode, fattr); 1180 return nfs_check_inode_attributes(inode, fattr); 1181 } 1182 1183 /** 1184 * nfs_refresh_inode - try to update the inode attribute cache 1185 * @inode - pointer to inode 1186 * @fattr - updated attributes 1187 * 1188 * Check that an RPC call that returned attributes has not overlapped with 1189 * other recent updates of the inode metadata, then decide whether it is 1190 * safe to do a full update of the inode attributes, or whether just to 1191 * call nfs_check_inode_attributes. 1192 */ 1193 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr) 1194 { 1195 int status; 1196 1197 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 1198 return 0; 1199 spin_lock(&inode->i_lock); 1200 status = nfs_refresh_inode_locked(inode, fattr); 1201 spin_unlock(&inode->i_lock); 1202 1203 return status; 1204 } 1205 EXPORT_SYMBOL_GPL(nfs_refresh_inode); 1206 1207 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr) 1208 { 1209 struct nfs_inode *nfsi = NFS_I(inode); 1210 1211 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE; 1212 if (S_ISDIR(inode->i_mode)) { 1213 nfsi->cache_validity |= NFS_INO_INVALID_DATA; 1214 nfs_fscache_invalidate(inode); 1215 } 1216 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 1217 return 0; 1218 return nfs_refresh_inode_locked(inode, fattr); 1219 } 1220 1221 /** 1222 * nfs_post_op_update_inode - try to update the inode attribute cache 1223 * @inode - pointer to inode 1224 * @fattr - updated attributes 1225 * 1226 * After an operation that has changed the inode metadata, mark the 1227 * attribute cache as being invalid, then try to update it. 1228 * 1229 * NB: if the server didn't return any post op attributes, this 1230 * function will force the retrieval of attributes before the next 1231 * NFS request. Thus it should be used only for operations that 1232 * are expected to change one or more attributes, to avoid 1233 * unnecessary NFS requests and trips through nfs_update_inode(). 1234 */ 1235 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1236 { 1237 int status; 1238 1239 spin_lock(&inode->i_lock); 1240 status = nfs_post_op_update_inode_locked(inode, fattr); 1241 spin_unlock(&inode->i_lock); 1242 return status; 1243 } 1244 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode); 1245 1246 /** 1247 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache 1248 * @inode - pointer to inode 1249 * @fattr - updated attributes 1250 * 1251 * After an operation that has changed the inode metadata, mark the 1252 * attribute cache as being invalid, then try to update it. Fake up 1253 * weak cache consistency data, if none exist. 1254 * 1255 * This function is mainly designed to be used by the ->write_done() functions. 1256 */ 1257 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr) 1258 { 1259 int status; 1260 1261 spin_lock(&inode->i_lock); 1262 /* Don't do a WCC update if these attributes are already stale */ 1263 if ((fattr->valid & NFS_ATTR_FATTR) == 0 || 1264 !nfs_inode_attrs_need_update(inode, fattr)) { 1265 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE 1266 | NFS_ATTR_FATTR_PRESIZE 1267 | NFS_ATTR_FATTR_PREMTIME 1268 | NFS_ATTR_FATTR_PRECTIME); 1269 goto out_noforce; 1270 } 1271 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && 1272 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) { 1273 fattr->pre_change_attr = inode->i_version; 1274 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE; 1275 } 1276 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 && 1277 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) { 1278 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime)); 1279 fattr->valid |= NFS_ATTR_FATTR_PRECTIME; 1280 } 1281 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 && 1282 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) { 1283 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime)); 1284 fattr->valid |= NFS_ATTR_FATTR_PREMTIME; 1285 } 1286 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 && 1287 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) { 1288 fattr->pre_size = i_size_read(inode); 1289 fattr->valid |= NFS_ATTR_FATTR_PRESIZE; 1290 } 1291 out_noforce: 1292 status = nfs_post_op_update_inode_locked(inode, fattr); 1293 spin_unlock(&inode->i_lock); 1294 return status; 1295 } 1296 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc); 1297 1298 /* 1299 * Many nfs protocol calls return the new file attributes after 1300 * an operation. Here we update the inode to reflect the state 1301 * of the server's inode. 1302 * 1303 * This is a bit tricky because we have to make sure all dirty pages 1304 * have been sent off to the server before calling invalidate_inode_pages. 1305 * To make sure no other process adds more write requests while we try 1306 * our best to flush them, we make them sleep during the attribute refresh. 1307 * 1308 * A very similar scenario holds for the dir cache. 1309 */ 1310 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1311 { 1312 struct nfs_server *server; 1313 struct nfs_inode *nfsi = NFS_I(inode); 1314 loff_t cur_isize, new_isize; 1315 unsigned long invalid = 0; 1316 unsigned long now = jiffies; 1317 unsigned long save_cache_validity; 1318 1319 dfprintk(VFS, "NFS: %s(%s/%ld fh_crc=0x%08x ct=%d info=0x%x)\n", 1320 __func__, inode->i_sb->s_id, inode->i_ino, 1321 nfs_display_fhandle_hash(NFS_FH(inode)), 1322 atomic_read(&inode->i_count), fattr->valid); 1323 1324 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) { 1325 printk(KERN_ERR "NFS: server %s error: fileid changed\n" 1326 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n", 1327 NFS_SERVER(inode)->nfs_client->cl_hostname, 1328 inode->i_sb->s_id, (long long)nfsi->fileid, 1329 (long long)fattr->fileid); 1330 goto out_err; 1331 } 1332 1333 /* 1334 * Make sure the inode's type hasn't changed. 1335 */ 1336 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) { 1337 /* 1338 * Big trouble! The inode has become a different object. 1339 */ 1340 printk(KERN_DEBUG "NFS: %s: inode %ld mode changed, %07o to %07o\n", 1341 __func__, inode->i_ino, inode->i_mode, fattr->mode); 1342 goto out_err; 1343 } 1344 1345 server = NFS_SERVER(inode); 1346 /* Update the fsid? */ 1347 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) && 1348 !nfs_fsid_equal(&server->fsid, &fattr->fsid) && 1349 !IS_AUTOMOUNT(inode)) 1350 server->fsid = fattr->fsid; 1351 1352 /* 1353 * Update the read time so we don't revalidate too often. 1354 */ 1355 nfsi->read_cache_jiffies = fattr->time_start; 1356 1357 save_cache_validity = nfsi->cache_validity; 1358 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR 1359 | NFS_INO_INVALID_ATIME 1360 | NFS_INO_REVAL_FORCED 1361 | NFS_INO_REVAL_PAGECACHE); 1362 1363 /* Do atomic weak cache consistency updates */ 1364 invalid |= nfs_wcc_update_inode(inode, fattr); 1365 1366 /* More cache consistency checks */ 1367 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) { 1368 if (inode->i_version != fattr->change_attr) { 1369 dprintk("NFS: change_attr change on server for file %s/%ld\n", 1370 inode->i_sb->s_id, inode->i_ino); 1371 invalid |= NFS_INO_INVALID_ATTR 1372 | NFS_INO_INVALID_DATA 1373 | NFS_INO_INVALID_ACCESS 1374 | NFS_INO_INVALID_ACL 1375 | NFS_INO_REVAL_PAGECACHE; 1376 if (S_ISDIR(inode->i_mode)) 1377 nfs_force_lookup_revalidate(inode); 1378 inode->i_version = fattr->change_attr; 1379 } 1380 } else if (server->caps & NFS_CAP_CHANGE_ATTR) 1381 invalid |= save_cache_validity; 1382 1383 if (fattr->valid & NFS_ATTR_FATTR_MTIME) { 1384 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime)); 1385 } else if (server->caps & NFS_CAP_MTIME) 1386 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1387 | NFS_INO_REVAL_FORCED); 1388 1389 if (fattr->valid & NFS_ATTR_FATTR_CTIME) { 1390 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime)); 1391 } else if (server->caps & NFS_CAP_CTIME) 1392 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1393 | NFS_INO_REVAL_FORCED); 1394 1395 /* Check if our cached file size is stale */ 1396 if (fattr->valid & NFS_ATTR_FATTR_SIZE) { 1397 new_isize = nfs_size_to_loff_t(fattr->size); 1398 cur_isize = i_size_read(inode); 1399 if (new_isize != cur_isize) { 1400 /* Do we perhaps have any outstanding writes, or has 1401 * the file grown beyond our last write? */ 1402 if ((nfsi->npages == 0 && !test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) || 1403 new_isize > cur_isize) { 1404 i_size_write(inode, new_isize); 1405 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA; 1406 } 1407 dprintk("NFS: isize change on server for file %s/%ld " 1408 "(%Ld to %Ld)\n", 1409 inode->i_sb->s_id, 1410 inode->i_ino, 1411 (long long)cur_isize, 1412 (long long)new_isize); 1413 } 1414 } else 1415 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1416 | NFS_INO_REVAL_PAGECACHE 1417 | NFS_INO_REVAL_FORCED); 1418 1419 1420 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 1421 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime)); 1422 else if (server->caps & NFS_CAP_ATIME) 1423 invalid |= save_cache_validity & (NFS_INO_INVALID_ATIME 1424 | NFS_INO_REVAL_FORCED); 1425 1426 if (fattr->valid & NFS_ATTR_FATTR_MODE) { 1427 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) { 1428 umode_t newmode = inode->i_mode & S_IFMT; 1429 newmode |= fattr->mode & S_IALLUGO; 1430 inode->i_mode = newmode; 1431 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 1432 } 1433 } else if (server->caps & NFS_CAP_MODE) 1434 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1435 | NFS_INO_INVALID_ACCESS 1436 | NFS_INO_INVALID_ACL 1437 | NFS_INO_REVAL_FORCED); 1438 1439 if (fattr->valid & NFS_ATTR_FATTR_OWNER) { 1440 if (!uid_eq(inode->i_uid, fattr->uid)) { 1441 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 1442 inode->i_uid = fattr->uid; 1443 } 1444 } else if (server->caps & NFS_CAP_OWNER) 1445 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1446 | NFS_INO_INVALID_ACCESS 1447 | NFS_INO_INVALID_ACL 1448 | NFS_INO_REVAL_FORCED); 1449 1450 if (fattr->valid & NFS_ATTR_FATTR_GROUP) { 1451 if (!gid_eq(inode->i_gid, fattr->gid)) { 1452 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 1453 inode->i_gid = fattr->gid; 1454 } 1455 } else if (server->caps & NFS_CAP_OWNER_GROUP) 1456 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1457 | NFS_INO_INVALID_ACCESS 1458 | NFS_INO_INVALID_ACL 1459 | NFS_INO_REVAL_FORCED); 1460 1461 if (fattr->valid & NFS_ATTR_FATTR_NLINK) { 1462 if (inode->i_nlink != fattr->nlink) { 1463 invalid |= NFS_INO_INVALID_ATTR; 1464 if (S_ISDIR(inode->i_mode)) 1465 invalid |= NFS_INO_INVALID_DATA; 1466 set_nlink(inode, fattr->nlink); 1467 } 1468 } else if (server->caps & NFS_CAP_NLINK) 1469 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1470 | NFS_INO_REVAL_FORCED); 1471 1472 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) { 1473 /* 1474 * report the blocks in 512byte units 1475 */ 1476 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 1477 } 1478 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED) 1479 inode->i_blocks = fattr->du.nfs2.blocks; 1480 1481 /* Update attrtimeo value if we're out of the unstable period */ 1482 if (invalid & NFS_INO_INVALID_ATTR) { 1483 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 1484 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 1485 nfsi->attrtimeo_timestamp = now; 1486 nfsi->attr_gencount = nfs_inc_attr_generation_counter(); 1487 } else { 1488 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) { 1489 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode)) 1490 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode); 1491 nfsi->attrtimeo_timestamp = now; 1492 } 1493 } 1494 invalid &= ~NFS_INO_INVALID_ATTR; 1495 /* Don't invalidate the data if we were to blame */ 1496 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) 1497 || S_ISLNK(inode->i_mode))) 1498 invalid &= ~NFS_INO_INVALID_DATA; 1499 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) || 1500 (save_cache_validity & NFS_INO_REVAL_FORCED)) 1501 nfsi->cache_validity |= invalid; 1502 1503 if (invalid & NFS_INO_INVALID_DATA) 1504 nfs_fscache_invalidate(inode); 1505 1506 return 0; 1507 out_err: 1508 /* 1509 * No need to worry about unhashing the dentry, as the 1510 * lookup validation will know that the inode is bad. 1511 * (But we fall through to invalidate the caches.) 1512 */ 1513 nfs_invalidate_inode(inode); 1514 return -ESTALE; 1515 } 1516 1517 struct inode *nfs_alloc_inode(struct super_block *sb) 1518 { 1519 struct nfs_inode *nfsi; 1520 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL); 1521 if (!nfsi) 1522 return NULL; 1523 nfsi->flags = 0UL; 1524 nfsi->cache_validity = 0UL; 1525 #ifdef CONFIG_NFS_V3_ACL 1526 nfsi->acl_access = ERR_PTR(-EAGAIN); 1527 nfsi->acl_default = ERR_PTR(-EAGAIN); 1528 #endif 1529 #if IS_ENABLED(CONFIG_NFS_V4) 1530 nfsi->nfs4_acl = NULL; 1531 #endif /* CONFIG_NFS_V4 */ 1532 return &nfsi->vfs_inode; 1533 } 1534 EXPORT_SYMBOL_GPL(nfs_alloc_inode); 1535 1536 static void nfs_i_callback(struct rcu_head *head) 1537 { 1538 struct inode *inode = container_of(head, struct inode, i_rcu); 1539 kmem_cache_free(nfs_inode_cachep, NFS_I(inode)); 1540 } 1541 1542 void nfs_destroy_inode(struct inode *inode) 1543 { 1544 call_rcu(&inode->i_rcu, nfs_i_callback); 1545 } 1546 EXPORT_SYMBOL_GPL(nfs_destroy_inode); 1547 1548 static inline void nfs4_init_once(struct nfs_inode *nfsi) 1549 { 1550 #if IS_ENABLED(CONFIG_NFS_V4) 1551 INIT_LIST_HEAD(&nfsi->open_states); 1552 nfsi->delegation = NULL; 1553 nfsi->delegation_state = 0; 1554 init_rwsem(&nfsi->rwsem); 1555 nfsi->layout = NULL; 1556 #endif 1557 } 1558 1559 static void init_once(void *foo) 1560 { 1561 struct nfs_inode *nfsi = (struct nfs_inode *) foo; 1562 1563 inode_init_once(&nfsi->vfs_inode); 1564 INIT_LIST_HEAD(&nfsi->open_files); 1565 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru); 1566 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru); 1567 INIT_LIST_HEAD(&nfsi->commit_info.list); 1568 nfsi->npages = 0; 1569 nfsi->commit_info.ncommit = 0; 1570 atomic_set(&nfsi->commit_info.rpcs_out, 0); 1571 atomic_set(&nfsi->silly_count, 1); 1572 INIT_HLIST_HEAD(&nfsi->silly_list); 1573 init_waitqueue_head(&nfsi->waitqueue); 1574 nfs4_init_once(nfsi); 1575 } 1576 1577 static int __init nfs_init_inodecache(void) 1578 { 1579 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache", 1580 sizeof(struct nfs_inode), 1581 0, (SLAB_RECLAIM_ACCOUNT| 1582 SLAB_MEM_SPREAD), 1583 init_once); 1584 if (nfs_inode_cachep == NULL) 1585 return -ENOMEM; 1586 1587 return 0; 1588 } 1589 1590 static void nfs_destroy_inodecache(void) 1591 { 1592 /* 1593 * Make sure all delayed rcu free inodes are flushed before we 1594 * destroy cache. 1595 */ 1596 rcu_barrier(); 1597 kmem_cache_destroy(nfs_inode_cachep); 1598 } 1599 1600 struct workqueue_struct *nfsiod_workqueue; 1601 EXPORT_SYMBOL_GPL(nfsiod_workqueue); 1602 1603 /* 1604 * start up the nfsiod workqueue 1605 */ 1606 static int nfsiod_start(void) 1607 { 1608 struct workqueue_struct *wq; 1609 dprintk("RPC: creating workqueue nfsiod\n"); 1610 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0); 1611 if (wq == NULL) 1612 return -ENOMEM; 1613 nfsiod_workqueue = wq; 1614 return 0; 1615 } 1616 1617 /* 1618 * Destroy the nfsiod workqueue 1619 */ 1620 static void nfsiod_stop(void) 1621 { 1622 struct workqueue_struct *wq; 1623 1624 wq = nfsiod_workqueue; 1625 if (wq == NULL) 1626 return; 1627 nfsiod_workqueue = NULL; 1628 destroy_workqueue(wq); 1629 } 1630 1631 int nfs_net_id; 1632 EXPORT_SYMBOL_GPL(nfs_net_id); 1633 1634 static int nfs_net_init(struct net *net) 1635 { 1636 nfs_clients_init(net); 1637 return nfs_dns_resolver_cache_init(net); 1638 } 1639 1640 static void nfs_net_exit(struct net *net) 1641 { 1642 nfs_dns_resolver_cache_destroy(net); 1643 nfs_cleanup_cb_ident_idr(net); 1644 } 1645 1646 static struct pernet_operations nfs_net_ops = { 1647 .init = nfs_net_init, 1648 .exit = nfs_net_exit, 1649 .id = &nfs_net_id, 1650 .size = sizeof(struct nfs_net), 1651 }; 1652 1653 /* 1654 * Initialize NFS 1655 */ 1656 static int __init init_nfs_fs(void) 1657 { 1658 int err; 1659 1660 err = nfs_dns_resolver_init(); 1661 if (err < 0) 1662 goto out10;; 1663 1664 err = register_pernet_subsys(&nfs_net_ops); 1665 if (err < 0) 1666 goto out9; 1667 1668 err = nfs_fscache_register(); 1669 if (err < 0) 1670 goto out8; 1671 1672 err = nfsiod_start(); 1673 if (err) 1674 goto out7; 1675 1676 err = nfs_fs_proc_init(); 1677 if (err) 1678 goto out6; 1679 1680 err = nfs_init_nfspagecache(); 1681 if (err) 1682 goto out5; 1683 1684 err = nfs_init_inodecache(); 1685 if (err) 1686 goto out4; 1687 1688 err = nfs_init_readpagecache(); 1689 if (err) 1690 goto out3; 1691 1692 err = nfs_init_writepagecache(); 1693 if (err) 1694 goto out2; 1695 1696 err = nfs_init_directcache(); 1697 if (err) 1698 goto out1; 1699 1700 #ifdef CONFIG_PROC_FS 1701 rpc_proc_register(&init_net, &nfs_rpcstat); 1702 #endif 1703 if ((err = register_nfs_fs()) != 0) 1704 goto out0; 1705 1706 return 0; 1707 out0: 1708 #ifdef CONFIG_PROC_FS 1709 rpc_proc_unregister(&init_net, "nfs"); 1710 #endif 1711 nfs_destroy_directcache(); 1712 out1: 1713 nfs_destroy_writepagecache(); 1714 out2: 1715 nfs_destroy_readpagecache(); 1716 out3: 1717 nfs_destroy_inodecache(); 1718 out4: 1719 nfs_destroy_nfspagecache(); 1720 out5: 1721 nfs_fs_proc_exit(); 1722 out6: 1723 nfsiod_stop(); 1724 out7: 1725 nfs_fscache_unregister(); 1726 out8: 1727 unregister_pernet_subsys(&nfs_net_ops); 1728 out9: 1729 nfs_dns_resolver_destroy(); 1730 out10: 1731 return err; 1732 } 1733 1734 static void __exit exit_nfs_fs(void) 1735 { 1736 nfs_destroy_directcache(); 1737 nfs_destroy_writepagecache(); 1738 nfs_destroy_readpagecache(); 1739 nfs_destroy_inodecache(); 1740 nfs_destroy_nfspagecache(); 1741 nfs_fscache_unregister(); 1742 unregister_pernet_subsys(&nfs_net_ops); 1743 nfs_dns_resolver_destroy(); 1744 #ifdef CONFIG_PROC_FS 1745 rpc_proc_unregister(&init_net, "nfs"); 1746 #endif 1747 unregister_nfs_fs(); 1748 nfs_fs_proc_exit(); 1749 nfsiod_stop(); 1750 } 1751 1752 /* Not quite true; I just maintain it */ 1753 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>"); 1754 MODULE_LICENSE("GPL"); 1755 module_param(enable_ino64, bool, 0644); 1756 1757 module_init(init_nfs_fs) 1758 module_exit(exit_nfs_fs) 1759