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