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 = -2; 336 inode->i_gid = -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 if (is_sync) 698 nfs_sb_deactive(sb); 699 else 700 nfs_sb_deactive_async(sb); 701 kfree(ctx->mdsthreshold); 702 kfree(ctx); 703 } 704 705 void put_nfs_open_context(struct nfs_open_context *ctx) 706 { 707 __put_nfs_open_context(ctx, 0); 708 } 709 EXPORT_SYMBOL_GPL(put_nfs_open_context); 710 711 /* 712 * Ensure that mmap has a recent RPC credential for use when writing out 713 * shared pages 714 */ 715 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx) 716 { 717 struct inode *inode = filp->f_path.dentry->d_inode; 718 struct nfs_inode *nfsi = NFS_I(inode); 719 720 filp->private_data = get_nfs_open_context(ctx); 721 spin_lock(&inode->i_lock); 722 list_add(&ctx->list, &nfsi->open_files); 723 spin_unlock(&inode->i_lock); 724 } 725 EXPORT_SYMBOL_GPL(nfs_file_set_open_context); 726 727 /* 728 * Given an inode, search for an open context with the desired characteristics 729 */ 730 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode) 731 { 732 struct nfs_inode *nfsi = NFS_I(inode); 733 struct nfs_open_context *pos, *ctx = NULL; 734 735 spin_lock(&inode->i_lock); 736 list_for_each_entry(pos, &nfsi->open_files, list) { 737 if (cred != NULL && pos->cred != cred) 738 continue; 739 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode) 740 continue; 741 ctx = get_nfs_open_context(pos); 742 break; 743 } 744 spin_unlock(&inode->i_lock); 745 return ctx; 746 } 747 748 static void nfs_file_clear_open_context(struct file *filp) 749 { 750 struct inode *inode = filp->f_path.dentry->d_inode; 751 struct nfs_open_context *ctx = nfs_file_open_context(filp); 752 753 if (ctx) { 754 filp->private_data = NULL; 755 spin_lock(&inode->i_lock); 756 list_move_tail(&ctx->list, &NFS_I(inode)->open_files); 757 spin_unlock(&inode->i_lock); 758 __put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1); 759 } 760 } 761 762 /* 763 * These allocate and release file read/write context information. 764 */ 765 int nfs_open(struct inode *inode, struct file *filp) 766 { 767 struct nfs_open_context *ctx; 768 769 ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode); 770 if (IS_ERR(ctx)) 771 return PTR_ERR(ctx); 772 nfs_file_set_open_context(filp, ctx); 773 put_nfs_open_context(ctx); 774 nfs_fscache_set_inode_cookie(inode, filp); 775 return 0; 776 } 777 778 int nfs_release(struct inode *inode, struct file *filp) 779 { 780 nfs_file_clear_open_context(filp); 781 return 0; 782 } 783 784 /* 785 * This function is called whenever some part of NFS notices that 786 * the cached attributes have to be refreshed. 787 */ 788 int 789 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) 790 { 791 int status = -ESTALE; 792 struct nfs_fattr *fattr = NULL; 793 struct nfs_inode *nfsi = NFS_I(inode); 794 795 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n", 796 inode->i_sb->s_id, (long long)NFS_FILEID(inode)); 797 798 if (is_bad_inode(inode)) 799 goto out; 800 if (NFS_STALE(inode)) 801 goto out; 802 803 status = -ENOMEM; 804 fattr = nfs_alloc_fattr(); 805 if (fattr == NULL) 806 goto out; 807 808 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE); 809 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr); 810 if (status != 0) { 811 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n", 812 inode->i_sb->s_id, 813 (long long)NFS_FILEID(inode), status); 814 if (status == -ESTALE) { 815 nfs_zap_caches(inode); 816 if (!S_ISDIR(inode->i_mode)) 817 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); 818 } 819 goto out; 820 } 821 822 status = nfs_refresh_inode(inode, fattr); 823 if (status) { 824 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n", 825 inode->i_sb->s_id, 826 (long long)NFS_FILEID(inode), status); 827 goto out; 828 } 829 830 if (nfsi->cache_validity & NFS_INO_INVALID_ACL) 831 nfs_zap_acl_cache(inode); 832 833 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n", 834 inode->i_sb->s_id, 835 (long long)NFS_FILEID(inode)); 836 837 out: 838 nfs_free_fattr(fattr); 839 return status; 840 } 841 842 int nfs_attribute_timeout(struct inode *inode) 843 { 844 struct nfs_inode *nfsi = NFS_I(inode); 845 846 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo); 847 } 848 849 static int nfs_attribute_cache_expired(struct inode *inode) 850 { 851 if (nfs_have_delegated_attributes(inode)) 852 return 0; 853 return nfs_attribute_timeout(inode); 854 } 855 856 /** 857 * nfs_revalidate_inode - Revalidate the inode attributes 858 * @server - pointer to nfs_server struct 859 * @inode - pointer to inode struct 860 * 861 * Updates inode attribute information by retrieving the data from the server. 862 */ 863 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) 864 { 865 if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR) 866 && !nfs_attribute_cache_expired(inode)) 867 return NFS_STALE(inode) ? -ESTALE : 0; 868 return __nfs_revalidate_inode(server, inode); 869 } 870 EXPORT_SYMBOL_GPL(nfs_revalidate_inode); 871 872 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping) 873 { 874 struct nfs_inode *nfsi = NFS_I(inode); 875 876 if (mapping->nrpages != 0) { 877 int ret = invalidate_inode_pages2(mapping); 878 if (ret < 0) 879 return ret; 880 } 881 spin_lock(&inode->i_lock); 882 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA; 883 if (S_ISDIR(inode->i_mode)) 884 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf)); 885 spin_unlock(&inode->i_lock); 886 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE); 887 nfs_fscache_wait_on_invalidate(inode); 888 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n", 889 inode->i_sb->s_id, (long long)NFS_FILEID(inode)); 890 return 0; 891 } 892 893 static bool nfs_mapping_need_revalidate_inode(struct inode *inode) 894 { 895 if (nfs_have_delegated_attributes(inode)) 896 return false; 897 return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE) 898 || nfs_attribute_timeout(inode) 899 || NFS_STALE(inode); 900 } 901 902 /** 903 * nfs_revalidate_mapping - Revalidate the pagecache 904 * @inode - pointer to host inode 905 * @mapping - pointer to mapping 906 */ 907 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping) 908 { 909 struct nfs_inode *nfsi = NFS_I(inode); 910 int ret = 0; 911 912 /* swapfiles are not supposed to be shared. */ 913 if (IS_SWAPFILE(inode)) 914 goto out; 915 916 if (nfs_mapping_need_revalidate_inode(inode)) { 917 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 918 if (ret < 0) 919 goto out; 920 } 921 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 922 ret = nfs_invalidate_mapping(inode, mapping); 923 out: 924 return ret; 925 } 926 927 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr) 928 { 929 struct nfs_inode *nfsi = NFS_I(inode); 930 unsigned long ret = 0; 931 932 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE) 933 && (fattr->valid & NFS_ATTR_FATTR_CHANGE) 934 && inode->i_version == fattr->pre_change_attr) { 935 inode->i_version = fattr->change_attr; 936 if (S_ISDIR(inode->i_mode)) 937 nfsi->cache_validity |= NFS_INO_INVALID_DATA; 938 ret |= NFS_INO_INVALID_ATTR; 939 } 940 /* If we have atomic WCC data, we may update some attributes */ 941 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME) 942 && (fattr->valid & NFS_ATTR_FATTR_CTIME) 943 && timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) { 944 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime)); 945 ret |= NFS_INO_INVALID_ATTR; 946 } 947 948 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME) 949 && (fattr->valid & NFS_ATTR_FATTR_MTIME) 950 && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) { 951 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime)); 952 if (S_ISDIR(inode->i_mode)) 953 nfsi->cache_validity |= NFS_INO_INVALID_DATA; 954 ret |= NFS_INO_INVALID_ATTR; 955 } 956 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE) 957 && (fattr->valid & NFS_ATTR_FATTR_SIZE) 958 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size) 959 && nfsi->npages == 0) { 960 i_size_write(inode, nfs_size_to_loff_t(fattr->size)); 961 ret |= NFS_INO_INVALID_ATTR; 962 } 963 964 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 965 nfs_fscache_invalidate(inode); 966 967 return ret; 968 } 969 970 /** 971 * nfs_check_inode_attributes - verify consistency of the inode attribute cache 972 * @inode - pointer to inode 973 * @fattr - updated attributes 974 * 975 * Verifies the attribute cache. If we have just changed the attributes, 976 * so that fattr carries weak cache consistency data, then it may 977 * also update the ctime/mtime/change_attribute. 978 */ 979 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr) 980 { 981 struct nfs_inode *nfsi = NFS_I(inode); 982 loff_t cur_size, new_isize; 983 unsigned long invalid = 0; 984 985 986 if (nfs_have_delegated_attributes(inode)) 987 return 0; 988 /* Has the inode gone and changed behind our back? */ 989 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) 990 return -EIO; 991 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) 992 return -EIO; 993 994 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && 995 inode->i_version != fattr->change_attr) 996 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE; 997 998 /* Verify a few of the more important attributes */ 999 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime)) 1000 invalid |= NFS_INO_INVALID_ATTR; 1001 1002 if (fattr->valid & NFS_ATTR_FATTR_SIZE) { 1003 cur_size = i_size_read(inode); 1004 new_isize = nfs_size_to_loff_t(fattr->size); 1005 if (cur_size != new_isize && nfsi->npages == 0) 1006 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE; 1007 } 1008 1009 /* Have any file permissions changed? */ 1010 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) 1011 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL; 1012 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && inode->i_uid != fattr->uid) 1013 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL; 1014 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && inode->i_gid != fattr->gid) 1015 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL; 1016 1017 /* Has the link count changed? */ 1018 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink) 1019 invalid |= NFS_INO_INVALID_ATTR; 1020 1021 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime)) 1022 invalid |= NFS_INO_INVALID_ATIME; 1023 1024 if (invalid != 0) 1025 nfsi->cache_validity |= invalid; 1026 1027 nfsi->read_cache_jiffies = fattr->time_start; 1028 return 0; 1029 } 1030 1031 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr) 1032 { 1033 if (!(fattr->valid & NFS_ATTR_FATTR_CTIME)) 1034 return 0; 1035 return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0; 1036 } 1037 1038 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr) 1039 { 1040 if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) 1041 return 0; 1042 return nfs_size_to_loff_t(fattr->size) > i_size_read(inode); 1043 } 1044 1045 static atomic_long_t nfs_attr_generation_counter; 1046 1047 static unsigned long nfs_read_attr_generation_counter(void) 1048 { 1049 return atomic_long_read(&nfs_attr_generation_counter); 1050 } 1051 1052 unsigned long nfs_inc_attr_generation_counter(void) 1053 { 1054 return atomic_long_inc_return(&nfs_attr_generation_counter); 1055 } 1056 1057 void nfs_fattr_init(struct nfs_fattr *fattr) 1058 { 1059 fattr->valid = 0; 1060 fattr->time_start = jiffies; 1061 fattr->gencount = nfs_inc_attr_generation_counter(); 1062 fattr->owner_name = NULL; 1063 fattr->group_name = NULL; 1064 } 1065 EXPORT_SYMBOL_GPL(nfs_fattr_init); 1066 1067 struct nfs_fattr *nfs_alloc_fattr(void) 1068 { 1069 struct nfs_fattr *fattr; 1070 1071 fattr = kmalloc(sizeof(*fattr), GFP_NOFS); 1072 if (fattr != NULL) 1073 nfs_fattr_init(fattr); 1074 return fattr; 1075 } 1076 EXPORT_SYMBOL_GPL(nfs_alloc_fattr); 1077 1078 struct nfs_fh *nfs_alloc_fhandle(void) 1079 { 1080 struct nfs_fh *fh; 1081 1082 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS); 1083 if (fh != NULL) 1084 fh->size = 0; 1085 return fh; 1086 } 1087 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle); 1088 1089 #ifdef NFS_DEBUG 1090 /* 1091 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle 1092 * in the same way that wireshark does 1093 * 1094 * @fh: file handle 1095 * 1096 * For debugging only. 1097 */ 1098 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh) 1099 { 1100 /* wireshark uses 32-bit AUTODIN crc and does a bitwise 1101 * not on the result */ 1102 return ~crc32(0xFFFFFFFF, &fh->data[0], fh->size); 1103 } 1104 1105 /* 1106 * _nfs_display_fhandle - display an NFS file handle on the console 1107 * 1108 * @fh: file handle to display 1109 * @caption: display caption 1110 * 1111 * For debugging only. 1112 */ 1113 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption) 1114 { 1115 unsigned short i; 1116 1117 if (fh == NULL || fh->size == 0) { 1118 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh); 1119 return; 1120 } 1121 1122 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n", 1123 caption, fh, fh->size, _nfs_display_fhandle_hash(fh)); 1124 for (i = 0; i < fh->size; i += 16) { 1125 __be32 *pos = (__be32 *)&fh->data[i]; 1126 1127 switch ((fh->size - i - 1) >> 2) { 1128 case 0: 1129 printk(KERN_DEFAULT " %08x\n", 1130 be32_to_cpup(pos)); 1131 break; 1132 case 1: 1133 printk(KERN_DEFAULT " %08x %08x\n", 1134 be32_to_cpup(pos), be32_to_cpup(pos + 1)); 1135 break; 1136 case 2: 1137 printk(KERN_DEFAULT " %08x %08x %08x\n", 1138 be32_to_cpup(pos), be32_to_cpup(pos + 1), 1139 be32_to_cpup(pos + 2)); 1140 break; 1141 default: 1142 printk(KERN_DEFAULT " %08x %08x %08x %08x\n", 1143 be32_to_cpup(pos), be32_to_cpup(pos + 1), 1144 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3)); 1145 } 1146 } 1147 } 1148 #endif 1149 1150 /** 1151 * nfs_inode_attrs_need_update - check if the inode attributes need updating 1152 * @inode - pointer to inode 1153 * @fattr - attributes 1154 * 1155 * Attempt to divine whether or not an RPC call reply carrying stale 1156 * attributes got scheduled after another call carrying updated ones. 1157 * 1158 * To do so, the function first assumes that a more recent ctime means 1159 * that the attributes in fattr are newer, however it also attempt to 1160 * catch the case where ctime either didn't change, or went backwards 1161 * (if someone reset the clock on the server) by looking at whether 1162 * or not this RPC call was started after the inode was last updated. 1163 * Note also the check for wraparound of 'attr_gencount' 1164 * 1165 * The function returns 'true' if it thinks the attributes in 'fattr' are 1166 * more recent than the ones cached in the inode. 1167 * 1168 */ 1169 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr) 1170 { 1171 const struct nfs_inode *nfsi = NFS_I(inode); 1172 1173 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 || 1174 nfs_ctime_need_update(inode, fattr) || 1175 nfs_size_need_update(inode, fattr) || 1176 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0); 1177 } 1178 1179 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr) 1180 { 1181 if (nfs_inode_attrs_need_update(inode, fattr)) 1182 return nfs_update_inode(inode, fattr); 1183 return nfs_check_inode_attributes(inode, fattr); 1184 } 1185 1186 /** 1187 * nfs_refresh_inode - try to update the inode attribute cache 1188 * @inode - pointer to inode 1189 * @fattr - updated attributes 1190 * 1191 * Check that an RPC call that returned attributes has not overlapped with 1192 * other recent updates of the inode metadata, then decide whether it is 1193 * safe to do a full update of the inode attributes, or whether just to 1194 * call nfs_check_inode_attributes. 1195 */ 1196 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr) 1197 { 1198 int status; 1199 1200 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 1201 return 0; 1202 spin_lock(&inode->i_lock); 1203 status = nfs_refresh_inode_locked(inode, fattr); 1204 spin_unlock(&inode->i_lock); 1205 1206 return status; 1207 } 1208 EXPORT_SYMBOL_GPL(nfs_refresh_inode); 1209 1210 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr) 1211 { 1212 struct nfs_inode *nfsi = NFS_I(inode); 1213 1214 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE; 1215 if (S_ISDIR(inode->i_mode)) { 1216 nfsi->cache_validity |= NFS_INO_INVALID_DATA; 1217 nfs_fscache_invalidate(inode); 1218 } 1219 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 1220 return 0; 1221 return nfs_refresh_inode_locked(inode, fattr); 1222 } 1223 1224 /** 1225 * nfs_post_op_update_inode - try to update the inode attribute cache 1226 * @inode - pointer to inode 1227 * @fattr - updated attributes 1228 * 1229 * After an operation that has changed the inode metadata, mark the 1230 * attribute cache as being invalid, then try to update it. 1231 * 1232 * NB: if the server didn't return any post op attributes, this 1233 * function will force the retrieval of attributes before the next 1234 * NFS request. Thus it should be used only for operations that 1235 * are expected to change one or more attributes, to avoid 1236 * unnecessary NFS requests and trips through nfs_update_inode(). 1237 */ 1238 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1239 { 1240 int status; 1241 1242 spin_lock(&inode->i_lock); 1243 status = nfs_post_op_update_inode_locked(inode, fattr); 1244 spin_unlock(&inode->i_lock); 1245 return status; 1246 } 1247 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode); 1248 1249 /** 1250 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache 1251 * @inode - pointer to inode 1252 * @fattr - updated attributes 1253 * 1254 * After an operation that has changed the inode metadata, mark the 1255 * attribute cache as being invalid, then try to update it. Fake up 1256 * weak cache consistency data, if none exist. 1257 * 1258 * This function is mainly designed to be used by the ->write_done() functions. 1259 */ 1260 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr) 1261 { 1262 int status; 1263 1264 spin_lock(&inode->i_lock); 1265 /* Don't do a WCC update if these attributes are already stale */ 1266 if ((fattr->valid & NFS_ATTR_FATTR) == 0 || 1267 !nfs_inode_attrs_need_update(inode, fattr)) { 1268 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE 1269 | NFS_ATTR_FATTR_PRESIZE 1270 | NFS_ATTR_FATTR_PREMTIME 1271 | NFS_ATTR_FATTR_PRECTIME); 1272 goto out_noforce; 1273 } 1274 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && 1275 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) { 1276 fattr->pre_change_attr = inode->i_version; 1277 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE; 1278 } 1279 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 && 1280 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) { 1281 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime)); 1282 fattr->valid |= NFS_ATTR_FATTR_PRECTIME; 1283 } 1284 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 && 1285 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) { 1286 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime)); 1287 fattr->valid |= NFS_ATTR_FATTR_PREMTIME; 1288 } 1289 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 && 1290 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) { 1291 fattr->pre_size = i_size_read(inode); 1292 fattr->valid |= NFS_ATTR_FATTR_PRESIZE; 1293 } 1294 out_noforce: 1295 status = nfs_post_op_update_inode_locked(inode, fattr); 1296 spin_unlock(&inode->i_lock); 1297 return status; 1298 } 1299 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc); 1300 1301 /* 1302 * Many nfs protocol calls return the new file attributes after 1303 * an operation. Here we update the inode to reflect the state 1304 * of the server's inode. 1305 * 1306 * This is a bit tricky because we have to make sure all dirty pages 1307 * have been sent off to the server before calling invalidate_inode_pages. 1308 * To make sure no other process adds more write requests while we try 1309 * our best to flush them, we make them sleep during the attribute refresh. 1310 * 1311 * A very similar scenario holds for the dir cache. 1312 */ 1313 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1314 { 1315 struct nfs_server *server; 1316 struct nfs_inode *nfsi = NFS_I(inode); 1317 loff_t cur_isize, new_isize; 1318 unsigned long invalid = 0; 1319 unsigned long now = jiffies; 1320 unsigned long save_cache_validity; 1321 1322 dfprintk(VFS, "NFS: %s(%s/%ld fh_crc=0x%08x ct=%d info=0x%x)\n", 1323 __func__, inode->i_sb->s_id, inode->i_ino, 1324 nfs_display_fhandle_hash(NFS_FH(inode)), 1325 atomic_read(&inode->i_count), fattr->valid); 1326 1327 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) { 1328 printk(KERN_ERR "NFS: server %s error: fileid changed\n" 1329 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n", 1330 NFS_SERVER(inode)->nfs_client->cl_hostname, 1331 inode->i_sb->s_id, (long long)nfsi->fileid, 1332 (long long)fattr->fileid); 1333 goto out_err; 1334 } 1335 1336 /* 1337 * Make sure the inode's type hasn't changed. 1338 */ 1339 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) { 1340 /* 1341 * Big trouble! The inode has become a different object. 1342 */ 1343 printk(KERN_DEBUG "NFS: %s: inode %ld mode changed, %07o to %07o\n", 1344 __func__, inode->i_ino, inode->i_mode, fattr->mode); 1345 goto out_err; 1346 } 1347 1348 server = NFS_SERVER(inode); 1349 /* Update the fsid? */ 1350 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) && 1351 !nfs_fsid_equal(&server->fsid, &fattr->fsid) && 1352 !IS_AUTOMOUNT(inode)) 1353 server->fsid = fattr->fsid; 1354 1355 /* 1356 * Update the read time so we don't revalidate too often. 1357 */ 1358 nfsi->read_cache_jiffies = fattr->time_start; 1359 1360 save_cache_validity = nfsi->cache_validity; 1361 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR 1362 | NFS_INO_INVALID_ATIME 1363 | NFS_INO_REVAL_FORCED 1364 | NFS_INO_REVAL_PAGECACHE); 1365 1366 /* Do atomic weak cache consistency updates */ 1367 invalid |= nfs_wcc_update_inode(inode, fattr); 1368 1369 /* More cache consistency checks */ 1370 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) { 1371 if (inode->i_version != fattr->change_attr) { 1372 dprintk("NFS: change_attr change on server for file %s/%ld\n", 1373 inode->i_sb->s_id, inode->i_ino); 1374 invalid |= NFS_INO_INVALID_ATTR 1375 | NFS_INO_INVALID_DATA 1376 | NFS_INO_INVALID_ACCESS 1377 | NFS_INO_INVALID_ACL 1378 | NFS_INO_REVAL_PAGECACHE; 1379 if (S_ISDIR(inode->i_mode)) 1380 nfs_force_lookup_revalidate(inode); 1381 inode->i_version = fattr->change_attr; 1382 } 1383 } else if (server->caps & NFS_CAP_CHANGE_ATTR) 1384 invalid |= save_cache_validity; 1385 1386 if (fattr->valid & NFS_ATTR_FATTR_MTIME) { 1387 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime)); 1388 } else if (server->caps & NFS_CAP_MTIME) 1389 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1390 | NFS_INO_REVAL_FORCED); 1391 1392 if (fattr->valid & NFS_ATTR_FATTR_CTIME) { 1393 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime)); 1394 } else if (server->caps & NFS_CAP_CTIME) 1395 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1396 | NFS_INO_REVAL_FORCED); 1397 1398 /* Check if our cached file size is stale */ 1399 if (fattr->valid & NFS_ATTR_FATTR_SIZE) { 1400 new_isize = nfs_size_to_loff_t(fattr->size); 1401 cur_isize = i_size_read(inode); 1402 if (new_isize != cur_isize) { 1403 /* Do we perhaps have any outstanding writes, or has 1404 * the file grown beyond our last write? */ 1405 if ((nfsi->npages == 0 && !test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) || 1406 new_isize > cur_isize) { 1407 i_size_write(inode, new_isize); 1408 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA; 1409 } 1410 dprintk("NFS: isize change on server for file %s/%ld " 1411 "(%Ld to %Ld)\n", 1412 inode->i_sb->s_id, 1413 inode->i_ino, 1414 (long long)cur_isize, 1415 (long long)new_isize); 1416 } 1417 } else 1418 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1419 | NFS_INO_REVAL_PAGECACHE 1420 | NFS_INO_REVAL_FORCED); 1421 1422 1423 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 1424 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime)); 1425 else if (server->caps & NFS_CAP_ATIME) 1426 invalid |= save_cache_validity & (NFS_INO_INVALID_ATIME 1427 | NFS_INO_REVAL_FORCED); 1428 1429 if (fattr->valid & NFS_ATTR_FATTR_MODE) { 1430 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) { 1431 umode_t newmode = inode->i_mode & S_IFMT; 1432 newmode |= fattr->mode & S_IALLUGO; 1433 inode->i_mode = newmode; 1434 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 1435 } 1436 } else if (server->caps & NFS_CAP_MODE) 1437 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1438 | NFS_INO_INVALID_ACCESS 1439 | NFS_INO_INVALID_ACL 1440 | NFS_INO_REVAL_FORCED); 1441 1442 if (fattr->valid & NFS_ATTR_FATTR_OWNER) { 1443 if (inode->i_uid != fattr->uid) { 1444 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 1445 inode->i_uid = fattr->uid; 1446 } 1447 } else if (server->caps & NFS_CAP_OWNER) 1448 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1449 | NFS_INO_INVALID_ACCESS 1450 | NFS_INO_INVALID_ACL 1451 | NFS_INO_REVAL_FORCED); 1452 1453 if (fattr->valid & NFS_ATTR_FATTR_GROUP) { 1454 if (inode->i_gid != fattr->gid) { 1455 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 1456 inode->i_gid = fattr->gid; 1457 } 1458 } else if (server->caps & NFS_CAP_OWNER_GROUP) 1459 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1460 | NFS_INO_INVALID_ACCESS 1461 | NFS_INO_INVALID_ACL 1462 | NFS_INO_REVAL_FORCED); 1463 1464 if (fattr->valid & NFS_ATTR_FATTR_NLINK) { 1465 if (inode->i_nlink != fattr->nlink) { 1466 invalid |= NFS_INO_INVALID_ATTR; 1467 if (S_ISDIR(inode->i_mode)) 1468 invalid |= NFS_INO_INVALID_DATA; 1469 set_nlink(inode, fattr->nlink); 1470 } 1471 } else if (server->caps & NFS_CAP_NLINK) 1472 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1473 | NFS_INO_REVAL_FORCED); 1474 1475 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) { 1476 /* 1477 * report the blocks in 512byte units 1478 */ 1479 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 1480 } 1481 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED) 1482 inode->i_blocks = fattr->du.nfs2.blocks; 1483 1484 /* Update attrtimeo value if we're out of the unstable period */ 1485 if (invalid & NFS_INO_INVALID_ATTR) { 1486 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 1487 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 1488 nfsi->attrtimeo_timestamp = now; 1489 nfsi->attr_gencount = nfs_inc_attr_generation_counter(); 1490 } else { 1491 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) { 1492 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode)) 1493 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode); 1494 nfsi->attrtimeo_timestamp = now; 1495 } 1496 } 1497 invalid &= ~NFS_INO_INVALID_ATTR; 1498 /* Don't invalidate the data if we were to blame */ 1499 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) 1500 || S_ISLNK(inode->i_mode))) 1501 invalid &= ~NFS_INO_INVALID_DATA; 1502 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) || 1503 (save_cache_validity & NFS_INO_REVAL_FORCED)) 1504 nfsi->cache_validity |= invalid; 1505 1506 if (invalid & NFS_INO_INVALID_DATA) 1507 nfs_fscache_invalidate(inode); 1508 1509 return 0; 1510 out_err: 1511 /* 1512 * No need to worry about unhashing the dentry, as the 1513 * lookup validation will know that the inode is bad. 1514 * (But we fall through to invalidate the caches.) 1515 */ 1516 nfs_invalidate_inode(inode); 1517 return -ESTALE; 1518 } 1519 1520 struct inode *nfs_alloc_inode(struct super_block *sb) 1521 { 1522 struct nfs_inode *nfsi; 1523 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL); 1524 if (!nfsi) 1525 return NULL; 1526 nfsi->flags = 0UL; 1527 nfsi->cache_validity = 0UL; 1528 #ifdef CONFIG_NFS_V3_ACL 1529 nfsi->acl_access = ERR_PTR(-EAGAIN); 1530 nfsi->acl_default = ERR_PTR(-EAGAIN); 1531 #endif 1532 #if IS_ENABLED(CONFIG_NFS_V4) 1533 nfsi->nfs4_acl = NULL; 1534 #endif /* CONFIG_NFS_V4 */ 1535 return &nfsi->vfs_inode; 1536 } 1537 EXPORT_SYMBOL_GPL(nfs_alloc_inode); 1538 1539 static void nfs_i_callback(struct rcu_head *head) 1540 { 1541 struct inode *inode = container_of(head, struct inode, i_rcu); 1542 kmem_cache_free(nfs_inode_cachep, NFS_I(inode)); 1543 } 1544 1545 void nfs_destroy_inode(struct inode *inode) 1546 { 1547 call_rcu(&inode->i_rcu, nfs_i_callback); 1548 } 1549 EXPORT_SYMBOL_GPL(nfs_destroy_inode); 1550 1551 static inline void nfs4_init_once(struct nfs_inode *nfsi) 1552 { 1553 #if IS_ENABLED(CONFIG_NFS_V4) 1554 INIT_LIST_HEAD(&nfsi->open_states); 1555 nfsi->delegation = NULL; 1556 nfsi->delegation_state = 0; 1557 init_rwsem(&nfsi->rwsem); 1558 nfsi->layout = NULL; 1559 #endif 1560 } 1561 1562 static void init_once(void *foo) 1563 { 1564 struct nfs_inode *nfsi = (struct nfs_inode *) foo; 1565 1566 inode_init_once(&nfsi->vfs_inode); 1567 INIT_LIST_HEAD(&nfsi->open_files); 1568 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru); 1569 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru); 1570 INIT_LIST_HEAD(&nfsi->commit_info.list); 1571 nfsi->npages = 0; 1572 nfsi->commit_info.ncommit = 0; 1573 atomic_set(&nfsi->commit_info.rpcs_out, 0); 1574 atomic_set(&nfsi->silly_count, 1); 1575 INIT_HLIST_HEAD(&nfsi->silly_list); 1576 init_waitqueue_head(&nfsi->waitqueue); 1577 nfs4_init_once(nfsi); 1578 } 1579 1580 static int __init nfs_init_inodecache(void) 1581 { 1582 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache", 1583 sizeof(struct nfs_inode), 1584 0, (SLAB_RECLAIM_ACCOUNT| 1585 SLAB_MEM_SPREAD), 1586 init_once); 1587 if (nfs_inode_cachep == NULL) 1588 return -ENOMEM; 1589 1590 return 0; 1591 } 1592 1593 static void nfs_destroy_inodecache(void) 1594 { 1595 /* 1596 * Make sure all delayed rcu free inodes are flushed before we 1597 * destroy cache. 1598 */ 1599 rcu_barrier(); 1600 kmem_cache_destroy(nfs_inode_cachep); 1601 } 1602 1603 struct workqueue_struct *nfsiod_workqueue; 1604 EXPORT_SYMBOL_GPL(nfsiod_workqueue); 1605 1606 /* 1607 * start up the nfsiod workqueue 1608 */ 1609 static int nfsiod_start(void) 1610 { 1611 struct workqueue_struct *wq; 1612 dprintk("RPC: creating workqueue nfsiod\n"); 1613 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0); 1614 if (wq == NULL) 1615 return -ENOMEM; 1616 nfsiod_workqueue = wq; 1617 return 0; 1618 } 1619 1620 /* 1621 * Destroy the nfsiod workqueue 1622 */ 1623 static void nfsiod_stop(void) 1624 { 1625 struct workqueue_struct *wq; 1626 1627 wq = nfsiod_workqueue; 1628 if (wq == NULL) 1629 return; 1630 nfsiod_workqueue = NULL; 1631 destroy_workqueue(wq); 1632 } 1633 1634 int nfs_net_id; 1635 EXPORT_SYMBOL_GPL(nfs_net_id); 1636 1637 static int nfs_net_init(struct net *net) 1638 { 1639 nfs_clients_init(net); 1640 return nfs_dns_resolver_cache_init(net); 1641 } 1642 1643 static void nfs_net_exit(struct net *net) 1644 { 1645 nfs_dns_resolver_cache_destroy(net); 1646 nfs_cleanup_cb_ident_idr(net); 1647 } 1648 1649 static struct pernet_operations nfs_net_ops = { 1650 .init = nfs_net_init, 1651 .exit = nfs_net_exit, 1652 .id = &nfs_net_id, 1653 .size = sizeof(struct nfs_net), 1654 }; 1655 1656 /* 1657 * Initialize NFS 1658 */ 1659 static int __init init_nfs_fs(void) 1660 { 1661 int err; 1662 1663 err = nfs_dns_resolver_init(); 1664 if (err < 0) 1665 goto out10;; 1666 1667 err = register_pernet_subsys(&nfs_net_ops); 1668 if (err < 0) 1669 goto out9; 1670 1671 err = nfs_fscache_register(); 1672 if (err < 0) 1673 goto out8; 1674 1675 err = nfsiod_start(); 1676 if (err) 1677 goto out7; 1678 1679 err = nfs_fs_proc_init(); 1680 if (err) 1681 goto out6; 1682 1683 err = nfs_init_nfspagecache(); 1684 if (err) 1685 goto out5; 1686 1687 err = nfs_init_inodecache(); 1688 if (err) 1689 goto out4; 1690 1691 err = nfs_init_readpagecache(); 1692 if (err) 1693 goto out3; 1694 1695 err = nfs_init_writepagecache(); 1696 if (err) 1697 goto out2; 1698 1699 err = nfs_init_directcache(); 1700 if (err) 1701 goto out1; 1702 1703 #ifdef CONFIG_PROC_FS 1704 rpc_proc_register(&init_net, &nfs_rpcstat); 1705 #endif 1706 if ((err = register_nfs_fs()) != 0) 1707 goto out0; 1708 1709 return 0; 1710 out0: 1711 #ifdef CONFIG_PROC_FS 1712 rpc_proc_unregister(&init_net, "nfs"); 1713 #endif 1714 nfs_destroy_directcache(); 1715 out1: 1716 nfs_destroy_writepagecache(); 1717 out2: 1718 nfs_destroy_readpagecache(); 1719 out3: 1720 nfs_destroy_inodecache(); 1721 out4: 1722 nfs_destroy_nfspagecache(); 1723 out5: 1724 nfs_fs_proc_exit(); 1725 out6: 1726 nfsiod_stop(); 1727 out7: 1728 nfs_fscache_unregister(); 1729 out8: 1730 unregister_pernet_subsys(&nfs_net_ops); 1731 out9: 1732 nfs_dns_resolver_destroy(); 1733 out10: 1734 return err; 1735 } 1736 1737 static void __exit exit_nfs_fs(void) 1738 { 1739 nfs_destroy_directcache(); 1740 nfs_destroy_writepagecache(); 1741 nfs_destroy_readpagecache(); 1742 nfs_destroy_inodecache(); 1743 nfs_destroy_nfspagecache(); 1744 nfs_fscache_unregister(); 1745 unregister_pernet_subsys(&nfs_net_ops); 1746 nfs_dns_resolver_destroy(); 1747 #ifdef CONFIG_PROC_FS 1748 rpc_proc_unregister(&init_net, "nfs"); 1749 #endif 1750 unregister_nfs_fs(); 1751 nfs_fs_proc_exit(); 1752 nfsiod_stop(); 1753 } 1754 1755 /* Not quite true; I just maintain it */ 1756 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>"); 1757 MODULE_LICENSE("GPL"); 1758 module_param(enable_ino64, bool, 0644); 1759 1760 module_init(init_nfs_fs) 1761 module_exit(exit_nfs_fs) 1762