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/nfs_idmap.h> 36 #include <linux/vfs.h> 37 #include <linux/inet.h> 38 #include <linux/nfs_xdr.h> 39 #include <linux/slab.h> 40 #include <linux/compat.h> 41 #include <linux/freezer.h> 42 #include <linux/crc32.h> 43 44 #include <asm/uaccess.h> 45 46 #include "nfs4_fs.h" 47 #include "callback.h" 48 #include "delegation.h" 49 #include "iostat.h" 50 #include "internal.h" 51 #include "fscache.h" 52 #include "dns_resolve.h" 53 #include "pnfs.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 86 /** 87 * nfs_compat_user_ino64 - returns the user-visible inode number 88 * @fileid: 64-bit fileid 89 * 90 * This function returns a 32-bit inode number if the boot parameter 91 * nfs.enable_ino64 is zero. 92 */ 93 u64 nfs_compat_user_ino64(u64 fileid) 94 { 95 #ifdef CONFIG_COMPAT 96 compat_ulong_t ino; 97 #else 98 unsigned long ino; 99 #endif 100 101 if (enable_ino64) 102 return fileid; 103 ino = fileid; 104 if (sizeof(ino) < sizeof(fileid)) 105 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8; 106 return ino; 107 } 108 109 static void nfs_clear_inode(struct inode *inode) 110 { 111 /* 112 * The following should never happen... 113 */ 114 BUG_ON(nfs_have_writebacks(inode)); 115 BUG_ON(!list_empty(&NFS_I(inode)->open_files)); 116 nfs_zap_acl_cache(inode); 117 nfs_access_zap_cache(inode); 118 nfs_fscache_release_inode_cookie(inode); 119 } 120 121 void nfs_evict_inode(struct inode *inode) 122 { 123 truncate_inode_pages(&inode->i_data, 0); 124 clear_inode(inode); 125 nfs_clear_inode(inode); 126 } 127 128 /** 129 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk 130 */ 131 int nfs_sync_mapping(struct address_space *mapping) 132 { 133 int ret = 0; 134 135 if (mapping->nrpages != 0) { 136 unmap_mapping_range(mapping, 0, 0, 0); 137 ret = nfs_wb_all(mapping->host); 138 } 139 return ret; 140 } 141 142 /* 143 * Invalidate the local caches 144 */ 145 static void nfs_zap_caches_locked(struct inode *inode) 146 { 147 struct nfs_inode *nfsi = NFS_I(inode); 148 int mode = inode->i_mode; 149 150 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 151 152 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 153 nfsi->attrtimeo_timestamp = jiffies; 154 155 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode))); 156 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) 157 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE; 158 else 159 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE; 160 } 161 162 void nfs_zap_caches(struct inode *inode) 163 { 164 spin_lock(&inode->i_lock); 165 nfs_zap_caches_locked(inode); 166 spin_unlock(&inode->i_lock); 167 } 168 169 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping) 170 { 171 if (mapping->nrpages != 0) { 172 spin_lock(&inode->i_lock); 173 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; 174 spin_unlock(&inode->i_lock); 175 } 176 } 177 178 void nfs_zap_acl_cache(struct inode *inode) 179 { 180 void (*clear_acl_cache)(struct inode *); 181 182 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache; 183 if (clear_acl_cache != NULL) 184 clear_acl_cache(inode); 185 spin_lock(&inode->i_lock); 186 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL; 187 spin_unlock(&inode->i_lock); 188 } 189 190 void nfs_invalidate_atime(struct inode *inode) 191 { 192 spin_lock(&inode->i_lock); 193 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME; 194 spin_unlock(&inode->i_lock); 195 } 196 197 /* 198 * Invalidate, but do not unhash, the inode. 199 * NB: must be called with inode->i_lock held! 200 */ 201 static void nfs_invalidate_inode(struct inode *inode) 202 { 203 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); 204 nfs_zap_caches_locked(inode); 205 } 206 207 struct nfs_find_desc { 208 struct nfs_fh *fh; 209 struct nfs_fattr *fattr; 210 }; 211 212 /* 213 * In NFSv3 we can have 64bit inode numbers. In order to support 214 * this, and re-exported directories (also seen in NFSv2) 215 * we are forced to allow 2 different inodes to have the same 216 * i_ino. 217 */ 218 static int 219 nfs_find_actor(struct inode *inode, void *opaque) 220 { 221 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque; 222 struct nfs_fh *fh = desc->fh; 223 struct nfs_fattr *fattr = desc->fattr; 224 225 if (NFS_FILEID(inode) != fattr->fileid) 226 return 0; 227 if (nfs_compare_fh(NFS_FH(inode), fh)) 228 return 0; 229 if (is_bad_inode(inode) || NFS_STALE(inode)) 230 return 0; 231 return 1; 232 } 233 234 static int 235 nfs_init_locked(struct inode *inode, void *opaque) 236 { 237 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque; 238 struct nfs_fattr *fattr = desc->fattr; 239 240 set_nfs_fileid(inode, fattr->fileid); 241 nfs_copy_fh(NFS_FH(inode), desc->fh); 242 return 0; 243 } 244 245 /* 246 * This is our front-end to iget that looks up inodes by file handle 247 * instead of inode number. 248 */ 249 struct inode * 250 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr) 251 { 252 struct nfs_find_desc desc = { 253 .fh = fh, 254 .fattr = fattr 255 }; 256 struct inode *inode = ERR_PTR(-ENOENT); 257 unsigned long hash; 258 259 nfs_attr_check_mountpoint(sb, fattr); 260 261 if (((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0) && 262 !nfs_attr_use_mounted_on_fileid(fattr)) 263 goto out_no_inode; 264 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0) 265 goto out_no_inode; 266 267 hash = nfs_fattr_to_ino_t(fattr); 268 269 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc); 270 if (inode == NULL) { 271 inode = ERR_PTR(-ENOMEM); 272 goto out_no_inode; 273 } 274 275 if (inode->i_state & I_NEW) { 276 struct nfs_inode *nfsi = NFS_I(inode); 277 unsigned long now = jiffies; 278 279 /* We set i_ino for the few things that still rely on it, 280 * such as stat(2) */ 281 inode->i_ino = hash; 282 283 /* We can't support update_atime(), since the server will reset it */ 284 inode->i_flags |= S_NOATIME|S_NOCMTIME; 285 inode->i_mode = fattr->mode; 286 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0 287 && nfs_server_capable(inode, NFS_CAP_MODE)) 288 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 289 /* Why so? Because we want revalidate for devices/FIFOs, and 290 * that's precisely what we have in nfs_file_inode_operations. 291 */ 292 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops; 293 if (S_ISREG(inode->i_mode)) { 294 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops; 295 inode->i_data.a_ops = &nfs_file_aops; 296 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info; 297 } else if (S_ISDIR(inode->i_mode)) { 298 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops; 299 inode->i_fop = &nfs_dir_operations; 300 inode->i_data.a_ops = &nfs_dir_aops; 301 /* Deal with crossing mountpoints */ 302 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT || 303 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) { 304 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) 305 inode->i_op = &nfs_referral_inode_operations; 306 else 307 inode->i_op = &nfs_mountpoint_inode_operations; 308 inode->i_fop = NULL; 309 inode->i_flags |= S_AUTOMOUNT; 310 } 311 } else if (S_ISLNK(inode->i_mode)) 312 inode->i_op = &nfs_symlink_inode_operations; 313 else 314 init_special_inode(inode, inode->i_mode, fattr->rdev); 315 316 memset(&inode->i_atime, 0, sizeof(inode->i_atime)); 317 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime)); 318 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime)); 319 inode->i_version = 0; 320 inode->i_size = 0; 321 clear_nlink(inode); 322 inode->i_uid = -2; 323 inode->i_gid = -2; 324 inode->i_blocks = 0; 325 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf)); 326 nfsi->write_io = 0; 327 nfsi->read_io = 0; 328 329 nfsi->read_cache_jiffies = fattr->time_start; 330 nfsi->attr_gencount = fattr->gencount; 331 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 332 inode->i_atime = fattr->atime; 333 else if (nfs_server_capable(inode, NFS_CAP_ATIME)) 334 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 335 if (fattr->valid & NFS_ATTR_FATTR_MTIME) 336 inode->i_mtime = fattr->mtime; 337 else if (nfs_server_capable(inode, NFS_CAP_MTIME)) 338 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 339 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 340 inode->i_ctime = fattr->ctime; 341 else if (nfs_server_capable(inode, NFS_CAP_CTIME)) 342 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 343 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) 344 inode->i_version = fattr->change_attr; 345 else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR)) 346 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 347 if (fattr->valid & NFS_ATTR_FATTR_SIZE) 348 inode->i_size = nfs_size_to_loff_t(fattr->size); 349 else 350 nfsi->cache_validity |= NFS_INO_INVALID_ATTR 351 | NFS_INO_REVAL_PAGECACHE; 352 if (fattr->valid & NFS_ATTR_FATTR_NLINK) 353 set_nlink(inode, fattr->nlink); 354 else if (nfs_server_capable(inode, NFS_CAP_NLINK)) 355 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 356 if (fattr->valid & NFS_ATTR_FATTR_OWNER) 357 inode->i_uid = fattr->uid; 358 else if (nfs_server_capable(inode, NFS_CAP_OWNER)) 359 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 360 if (fattr->valid & NFS_ATTR_FATTR_GROUP) 361 inode->i_gid = fattr->gid; 362 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP)) 363 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 364 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED) 365 inode->i_blocks = fattr->du.nfs2.blocks; 366 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) { 367 /* 368 * report the blocks in 512byte units 369 */ 370 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 371 } 372 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 373 nfsi->attrtimeo_timestamp = now; 374 nfsi->access_cache = RB_ROOT; 375 376 nfs_fscache_init_inode_cookie(inode); 377 378 unlock_new_inode(inode); 379 } else 380 nfs_refresh_inode(inode, fattr); 381 dprintk("NFS: nfs_fhget(%s/%Ld fh_crc=0x%08x ct=%d)\n", 382 inode->i_sb->s_id, 383 (long long)NFS_FILEID(inode), 384 nfs_display_fhandle_hash(fh), 385 atomic_read(&inode->i_count)); 386 387 out: 388 return inode; 389 390 out_no_inode: 391 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode)); 392 goto out; 393 } 394 395 #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) 396 397 int 398 nfs_setattr(struct dentry *dentry, struct iattr *attr) 399 { 400 struct inode *inode = dentry->d_inode; 401 struct nfs_fattr *fattr; 402 int error = -ENOMEM; 403 404 nfs_inc_stats(inode, NFSIOS_VFSSETATTR); 405 406 /* skip mode change if it's just for clearing setuid/setgid */ 407 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID)) 408 attr->ia_valid &= ~ATTR_MODE; 409 410 if (attr->ia_valid & ATTR_SIZE) { 411 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode)) 412 attr->ia_valid &= ~ATTR_SIZE; 413 } 414 415 /* Optimization: if the end result is no change, don't RPC */ 416 attr->ia_valid &= NFS_VALID_ATTRS; 417 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0) 418 return 0; 419 420 /* Write all dirty data */ 421 if (S_ISREG(inode->i_mode)) { 422 nfs_inode_dio_wait(inode); 423 nfs_wb_all(inode); 424 } 425 426 fattr = nfs_alloc_fattr(); 427 if (fattr == NULL) 428 goto out; 429 /* 430 * Return any delegations if we're going to change ACLs 431 */ 432 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) 433 nfs_inode_return_delegation(inode); 434 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr); 435 if (error == 0) 436 nfs_refresh_inode(inode, fattr); 437 nfs_free_fattr(fattr); 438 out: 439 return error; 440 } 441 442 /** 443 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall 444 * @inode: inode of the file used 445 * @offset: file offset to start truncating 446 * 447 * This is a copy of the common vmtruncate, but with the locking 448 * corrected to take into account the fact that NFS requires 449 * inode->i_size to be updated under the inode->i_lock. 450 */ 451 static int nfs_vmtruncate(struct inode * inode, loff_t offset) 452 { 453 loff_t oldsize; 454 int err; 455 456 err = inode_newsize_ok(inode, offset); 457 if (err) 458 goto out; 459 460 spin_lock(&inode->i_lock); 461 oldsize = inode->i_size; 462 i_size_write(inode, offset); 463 spin_unlock(&inode->i_lock); 464 465 truncate_pagecache(inode, oldsize, offset); 466 out: 467 return err; 468 } 469 470 /** 471 * nfs_setattr_update_inode - Update inode metadata after a setattr call. 472 * @inode: pointer to struct inode 473 * @attr: pointer to struct iattr 474 * 475 * Note: we do this in the *proc.c in order to ensure that 476 * it works for things like exclusive creates too. 477 */ 478 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr) 479 { 480 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) { 481 spin_lock(&inode->i_lock); 482 if ((attr->ia_valid & ATTR_MODE) != 0) { 483 int mode = attr->ia_mode & S_IALLUGO; 484 mode |= inode->i_mode & ~S_IALLUGO; 485 inode->i_mode = mode; 486 } 487 if ((attr->ia_valid & ATTR_UID) != 0) 488 inode->i_uid = attr->ia_uid; 489 if ((attr->ia_valid & ATTR_GID) != 0) 490 inode->i_gid = attr->ia_gid; 491 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 492 spin_unlock(&inode->i_lock); 493 } 494 if ((attr->ia_valid & ATTR_SIZE) != 0) { 495 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC); 496 nfs_vmtruncate(inode, attr->ia_size); 497 } 498 } 499 500 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat) 501 { 502 struct inode *inode = dentry->d_inode; 503 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME; 504 int err; 505 506 /* Flush out writes to the server in order to update c/mtime. */ 507 if (S_ISREG(inode->i_mode)) { 508 nfs_inode_dio_wait(inode); 509 err = filemap_write_and_wait(inode->i_mapping); 510 if (err) 511 goto out; 512 } 513 514 /* 515 * We may force a getattr if the user cares about atime. 516 * 517 * Note that we only have to check the vfsmount flags here: 518 * - NFS always sets S_NOATIME by so checking it would give a 519 * bogus result 520 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is 521 * no point in checking those. 522 */ 523 if ((mnt->mnt_flags & MNT_NOATIME) || 524 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) 525 need_atime = 0; 526 527 if (need_atime) 528 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 529 else 530 err = nfs_revalidate_inode(NFS_SERVER(inode), inode); 531 if (!err) { 532 generic_fillattr(inode, stat); 533 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode)); 534 } 535 out: 536 return err; 537 } 538 539 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx) 540 { 541 atomic_set(&l_ctx->count, 1); 542 l_ctx->lockowner = current->files; 543 l_ctx->pid = current->tgid; 544 INIT_LIST_HEAD(&l_ctx->list); 545 } 546 547 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx) 548 { 549 struct nfs_lock_context *pos; 550 551 list_for_each_entry(pos, &ctx->lock_context.list, list) { 552 if (pos->lockowner != current->files) 553 continue; 554 if (pos->pid != current->tgid) 555 continue; 556 atomic_inc(&pos->count); 557 return pos; 558 } 559 return NULL; 560 } 561 562 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx) 563 { 564 struct nfs_lock_context *res, *new = NULL; 565 struct inode *inode = ctx->dentry->d_inode; 566 567 spin_lock(&inode->i_lock); 568 res = __nfs_find_lock_context(ctx); 569 if (res == NULL) { 570 spin_unlock(&inode->i_lock); 571 new = kmalloc(sizeof(*new), GFP_KERNEL); 572 if (new == NULL) 573 return NULL; 574 nfs_init_lock_context(new); 575 spin_lock(&inode->i_lock); 576 res = __nfs_find_lock_context(ctx); 577 if (res == NULL) { 578 list_add_tail(&new->list, &ctx->lock_context.list); 579 new->open_context = ctx; 580 res = new; 581 new = NULL; 582 } 583 } 584 spin_unlock(&inode->i_lock); 585 kfree(new); 586 return res; 587 } 588 589 void nfs_put_lock_context(struct nfs_lock_context *l_ctx) 590 { 591 struct nfs_open_context *ctx = l_ctx->open_context; 592 struct inode *inode = ctx->dentry->d_inode; 593 594 if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock)) 595 return; 596 list_del(&l_ctx->list); 597 spin_unlock(&inode->i_lock); 598 kfree(l_ctx); 599 } 600 601 /** 602 * nfs_close_context - Common close_context() routine NFSv2/v3 603 * @ctx: pointer to context 604 * @is_sync: is this a synchronous close 605 * 606 * always ensure that the attributes are up to date if we're mounted 607 * with close-to-open semantics 608 */ 609 void nfs_close_context(struct nfs_open_context *ctx, int is_sync) 610 { 611 struct inode *inode; 612 struct nfs_server *server; 613 614 if (!(ctx->mode & FMODE_WRITE)) 615 return; 616 if (!is_sync) 617 return; 618 inode = ctx->dentry->d_inode; 619 if (!list_empty(&NFS_I(inode)->open_files)) 620 return; 621 server = NFS_SERVER(inode); 622 if (server->flags & NFS_MOUNT_NOCTO) 623 return; 624 nfs_revalidate_inode(server, inode); 625 } 626 627 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode) 628 { 629 struct nfs_open_context *ctx; 630 struct rpc_cred *cred = rpc_lookup_cred(); 631 if (IS_ERR(cred)) 632 return ERR_CAST(cred); 633 634 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); 635 if (!ctx) { 636 put_rpccred(cred); 637 return ERR_PTR(-ENOMEM); 638 } 639 nfs_sb_active(dentry->d_sb); 640 ctx->dentry = dget(dentry); 641 ctx->cred = cred; 642 ctx->state = NULL; 643 ctx->mode = f_mode; 644 ctx->flags = 0; 645 ctx->error = 0; 646 nfs_init_lock_context(&ctx->lock_context); 647 ctx->lock_context.open_context = ctx; 648 INIT_LIST_HEAD(&ctx->list); 649 ctx->mdsthreshold = NULL; 650 return ctx; 651 } 652 653 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx) 654 { 655 if (ctx != NULL) 656 atomic_inc(&ctx->lock_context.count); 657 return ctx; 658 } 659 660 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync) 661 { 662 struct inode *inode = ctx->dentry->d_inode; 663 struct super_block *sb = ctx->dentry->d_sb; 664 665 if (!list_empty(&ctx->list)) { 666 if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock)) 667 return; 668 list_del(&ctx->list); 669 spin_unlock(&inode->i_lock); 670 } else if (!atomic_dec_and_test(&ctx->lock_context.count)) 671 return; 672 if (inode != NULL) 673 NFS_PROTO(inode)->close_context(ctx, is_sync); 674 if (ctx->cred != NULL) 675 put_rpccred(ctx->cred); 676 dput(ctx->dentry); 677 nfs_sb_deactive(sb); 678 kfree(ctx->mdsthreshold); 679 kfree(ctx); 680 } 681 682 void put_nfs_open_context(struct nfs_open_context *ctx) 683 { 684 __put_nfs_open_context(ctx, 0); 685 } 686 687 /* 688 * Ensure that mmap has a recent RPC credential for use when writing out 689 * shared pages 690 */ 691 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx) 692 { 693 struct inode *inode = filp->f_path.dentry->d_inode; 694 struct nfs_inode *nfsi = NFS_I(inode); 695 696 filp->private_data = get_nfs_open_context(ctx); 697 spin_lock(&inode->i_lock); 698 list_add(&ctx->list, &nfsi->open_files); 699 spin_unlock(&inode->i_lock); 700 } 701 702 /* 703 * Given an inode, search for an open context with the desired characteristics 704 */ 705 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode) 706 { 707 struct nfs_inode *nfsi = NFS_I(inode); 708 struct nfs_open_context *pos, *ctx = NULL; 709 710 spin_lock(&inode->i_lock); 711 list_for_each_entry(pos, &nfsi->open_files, list) { 712 if (cred != NULL && pos->cred != cred) 713 continue; 714 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode) 715 continue; 716 ctx = get_nfs_open_context(pos); 717 break; 718 } 719 spin_unlock(&inode->i_lock); 720 return ctx; 721 } 722 723 static void nfs_file_clear_open_context(struct file *filp) 724 { 725 struct inode *inode = filp->f_path.dentry->d_inode; 726 struct nfs_open_context *ctx = nfs_file_open_context(filp); 727 728 if (ctx) { 729 filp->private_data = NULL; 730 spin_lock(&inode->i_lock); 731 list_move_tail(&ctx->list, &NFS_I(inode)->open_files); 732 spin_unlock(&inode->i_lock); 733 __put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1); 734 } 735 } 736 737 /* 738 * These allocate and release file read/write context information. 739 */ 740 int nfs_open(struct inode *inode, struct file *filp) 741 { 742 struct nfs_open_context *ctx; 743 744 ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode); 745 if (IS_ERR(ctx)) 746 return PTR_ERR(ctx); 747 nfs_file_set_open_context(filp, ctx); 748 put_nfs_open_context(ctx); 749 nfs_fscache_set_inode_cookie(inode, filp); 750 return 0; 751 } 752 753 int nfs_release(struct inode *inode, struct file *filp) 754 { 755 nfs_file_clear_open_context(filp); 756 return 0; 757 } 758 759 /* 760 * This function is called whenever some part of NFS notices that 761 * the cached attributes have to be refreshed. 762 */ 763 int 764 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) 765 { 766 int status = -ESTALE; 767 struct nfs_fattr *fattr = NULL; 768 struct nfs_inode *nfsi = NFS_I(inode); 769 770 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n", 771 inode->i_sb->s_id, (long long)NFS_FILEID(inode)); 772 773 if (is_bad_inode(inode)) 774 goto out; 775 if (NFS_STALE(inode)) 776 goto out; 777 778 status = -ENOMEM; 779 fattr = nfs_alloc_fattr(); 780 if (fattr == NULL) 781 goto out; 782 783 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE); 784 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr); 785 if (status != 0) { 786 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n", 787 inode->i_sb->s_id, 788 (long long)NFS_FILEID(inode), status); 789 if (status == -ESTALE) { 790 nfs_zap_caches(inode); 791 if (!S_ISDIR(inode->i_mode)) 792 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); 793 } 794 goto out; 795 } 796 797 status = nfs_refresh_inode(inode, fattr); 798 if (status) { 799 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n", 800 inode->i_sb->s_id, 801 (long long)NFS_FILEID(inode), status); 802 goto out; 803 } 804 805 if (nfsi->cache_validity & NFS_INO_INVALID_ACL) 806 nfs_zap_acl_cache(inode); 807 808 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n", 809 inode->i_sb->s_id, 810 (long long)NFS_FILEID(inode)); 811 812 out: 813 nfs_free_fattr(fattr); 814 return status; 815 } 816 817 int nfs_attribute_timeout(struct inode *inode) 818 { 819 struct nfs_inode *nfsi = NFS_I(inode); 820 821 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo); 822 } 823 824 static int nfs_attribute_cache_expired(struct inode *inode) 825 { 826 if (nfs_have_delegated_attributes(inode)) 827 return 0; 828 return nfs_attribute_timeout(inode); 829 } 830 831 /** 832 * nfs_revalidate_inode - Revalidate the inode attributes 833 * @server - pointer to nfs_server struct 834 * @inode - pointer to inode struct 835 * 836 * Updates inode attribute information by retrieving the data from the server. 837 */ 838 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) 839 { 840 if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR) 841 && !nfs_attribute_cache_expired(inode)) 842 return NFS_STALE(inode) ? -ESTALE : 0; 843 return __nfs_revalidate_inode(server, inode); 844 } 845 846 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping) 847 { 848 struct nfs_inode *nfsi = NFS_I(inode); 849 850 if (mapping->nrpages != 0) { 851 int ret = invalidate_inode_pages2(mapping); 852 if (ret < 0) 853 return ret; 854 } 855 spin_lock(&inode->i_lock); 856 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA; 857 if (S_ISDIR(inode->i_mode)) 858 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf)); 859 spin_unlock(&inode->i_lock); 860 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE); 861 nfs_fscache_reset_inode_cookie(inode); 862 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n", 863 inode->i_sb->s_id, (long long)NFS_FILEID(inode)); 864 return 0; 865 } 866 867 static bool nfs_mapping_need_revalidate_inode(struct inode *inode) 868 { 869 if (nfs_have_delegated_attributes(inode)) 870 return false; 871 return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE) 872 || nfs_attribute_timeout(inode) 873 || NFS_STALE(inode); 874 } 875 876 /** 877 * nfs_revalidate_mapping - Revalidate the pagecache 878 * @inode - pointer to host inode 879 * @mapping - pointer to mapping 880 */ 881 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping) 882 { 883 struct nfs_inode *nfsi = NFS_I(inode); 884 int ret = 0; 885 886 if (nfs_mapping_need_revalidate_inode(inode)) { 887 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 888 if (ret < 0) 889 goto out; 890 } 891 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 892 ret = nfs_invalidate_mapping(inode, mapping); 893 out: 894 return ret; 895 } 896 897 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr) 898 { 899 struct nfs_inode *nfsi = NFS_I(inode); 900 unsigned long ret = 0; 901 902 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE) 903 && (fattr->valid & NFS_ATTR_FATTR_CHANGE) 904 && inode->i_version == fattr->pre_change_attr) { 905 inode->i_version = fattr->change_attr; 906 if (S_ISDIR(inode->i_mode)) 907 nfsi->cache_validity |= NFS_INO_INVALID_DATA; 908 ret |= NFS_INO_INVALID_ATTR; 909 } 910 /* If we have atomic WCC data, we may update some attributes */ 911 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME) 912 && (fattr->valid & NFS_ATTR_FATTR_CTIME) 913 && timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) { 914 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime)); 915 ret |= NFS_INO_INVALID_ATTR; 916 } 917 918 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME) 919 && (fattr->valid & NFS_ATTR_FATTR_MTIME) 920 && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) { 921 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime)); 922 if (S_ISDIR(inode->i_mode)) 923 nfsi->cache_validity |= NFS_INO_INVALID_DATA; 924 ret |= NFS_INO_INVALID_ATTR; 925 } 926 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE) 927 && (fattr->valid & NFS_ATTR_FATTR_SIZE) 928 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size) 929 && nfsi->npages == 0) { 930 i_size_write(inode, nfs_size_to_loff_t(fattr->size)); 931 ret |= NFS_INO_INVALID_ATTR; 932 } 933 return ret; 934 } 935 936 /** 937 * nfs_check_inode_attributes - verify consistency of the inode attribute cache 938 * @inode - pointer to inode 939 * @fattr - updated attributes 940 * 941 * Verifies the attribute cache. If we have just changed the attributes, 942 * so that fattr carries weak cache consistency data, then it may 943 * also update the ctime/mtime/change_attribute. 944 */ 945 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr) 946 { 947 struct nfs_inode *nfsi = NFS_I(inode); 948 loff_t cur_size, new_isize; 949 unsigned long invalid = 0; 950 951 952 if (nfs_have_delegated_attributes(inode)) 953 return 0; 954 /* Has the inode gone and changed behind our back? */ 955 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) 956 return -EIO; 957 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) 958 return -EIO; 959 960 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && 961 inode->i_version != fattr->change_attr) 962 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE; 963 964 /* Verify a few of the more important attributes */ 965 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime)) 966 invalid |= NFS_INO_INVALID_ATTR; 967 968 if (fattr->valid & NFS_ATTR_FATTR_SIZE) { 969 cur_size = i_size_read(inode); 970 new_isize = nfs_size_to_loff_t(fattr->size); 971 if (cur_size != new_isize && nfsi->npages == 0) 972 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE; 973 } 974 975 /* Have any file permissions changed? */ 976 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) 977 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL; 978 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && inode->i_uid != fattr->uid) 979 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL; 980 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && inode->i_gid != fattr->gid) 981 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL; 982 983 /* Has the link count changed? */ 984 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink) 985 invalid |= NFS_INO_INVALID_ATTR; 986 987 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime)) 988 invalid |= NFS_INO_INVALID_ATIME; 989 990 if (invalid != 0) 991 nfsi->cache_validity |= invalid; 992 993 nfsi->read_cache_jiffies = fattr->time_start; 994 return 0; 995 } 996 997 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr) 998 { 999 if (!(fattr->valid & NFS_ATTR_FATTR_CTIME)) 1000 return 0; 1001 return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0; 1002 } 1003 1004 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr) 1005 { 1006 if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) 1007 return 0; 1008 return nfs_size_to_loff_t(fattr->size) > i_size_read(inode); 1009 } 1010 1011 static atomic_long_t nfs_attr_generation_counter; 1012 1013 static unsigned long nfs_read_attr_generation_counter(void) 1014 { 1015 return atomic_long_read(&nfs_attr_generation_counter); 1016 } 1017 1018 unsigned long nfs_inc_attr_generation_counter(void) 1019 { 1020 return atomic_long_inc_return(&nfs_attr_generation_counter); 1021 } 1022 1023 void nfs_fattr_init(struct nfs_fattr *fattr) 1024 { 1025 fattr->valid = 0; 1026 fattr->time_start = jiffies; 1027 fattr->gencount = nfs_inc_attr_generation_counter(); 1028 fattr->owner_name = NULL; 1029 fattr->group_name = NULL; 1030 } 1031 1032 struct nfs_fattr *nfs_alloc_fattr(void) 1033 { 1034 struct nfs_fattr *fattr; 1035 1036 fattr = kmalloc(sizeof(*fattr), GFP_NOFS); 1037 if (fattr != NULL) 1038 nfs_fattr_init(fattr); 1039 return fattr; 1040 } 1041 1042 struct nfs_fh *nfs_alloc_fhandle(void) 1043 { 1044 struct nfs_fh *fh; 1045 1046 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS); 1047 if (fh != NULL) 1048 fh->size = 0; 1049 return fh; 1050 } 1051 1052 #ifdef NFS_DEBUG 1053 /* 1054 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle 1055 * in the same way that wireshark does 1056 * 1057 * @fh: file handle 1058 * 1059 * For debugging only. 1060 */ 1061 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh) 1062 { 1063 /* wireshark uses 32-bit AUTODIN crc and does a bitwise 1064 * not on the result */ 1065 return ~crc32(0xFFFFFFFF, &fh->data[0], fh->size); 1066 } 1067 1068 /* 1069 * _nfs_display_fhandle - display an NFS file handle on the console 1070 * 1071 * @fh: file handle to display 1072 * @caption: display caption 1073 * 1074 * For debugging only. 1075 */ 1076 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption) 1077 { 1078 unsigned short i; 1079 1080 if (fh == NULL || fh->size == 0) { 1081 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh); 1082 return; 1083 } 1084 1085 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n", 1086 caption, fh, fh->size, _nfs_display_fhandle_hash(fh)); 1087 for (i = 0; i < fh->size; i += 16) { 1088 __be32 *pos = (__be32 *)&fh->data[i]; 1089 1090 switch ((fh->size - i - 1) >> 2) { 1091 case 0: 1092 printk(KERN_DEFAULT " %08x\n", 1093 be32_to_cpup(pos)); 1094 break; 1095 case 1: 1096 printk(KERN_DEFAULT " %08x %08x\n", 1097 be32_to_cpup(pos), be32_to_cpup(pos + 1)); 1098 break; 1099 case 2: 1100 printk(KERN_DEFAULT " %08x %08x %08x\n", 1101 be32_to_cpup(pos), be32_to_cpup(pos + 1), 1102 be32_to_cpup(pos + 2)); 1103 break; 1104 default: 1105 printk(KERN_DEFAULT " %08x %08x %08x %08x\n", 1106 be32_to_cpup(pos), be32_to_cpup(pos + 1), 1107 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3)); 1108 } 1109 } 1110 } 1111 #endif 1112 1113 /** 1114 * nfs_inode_attrs_need_update - check if the inode attributes need updating 1115 * @inode - pointer to inode 1116 * @fattr - attributes 1117 * 1118 * Attempt to divine whether or not an RPC call reply carrying stale 1119 * attributes got scheduled after another call carrying updated ones. 1120 * 1121 * To do so, the function first assumes that a more recent ctime means 1122 * that the attributes in fattr are newer, however it also attempt to 1123 * catch the case where ctime either didn't change, or went backwards 1124 * (if someone reset the clock on the server) by looking at whether 1125 * or not this RPC call was started after the inode was last updated. 1126 * Note also the check for wraparound of 'attr_gencount' 1127 * 1128 * The function returns 'true' if it thinks the attributes in 'fattr' are 1129 * more recent than the ones cached in the inode. 1130 * 1131 */ 1132 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr) 1133 { 1134 const struct nfs_inode *nfsi = NFS_I(inode); 1135 1136 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 || 1137 nfs_ctime_need_update(inode, fattr) || 1138 nfs_size_need_update(inode, fattr) || 1139 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0); 1140 } 1141 1142 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr) 1143 { 1144 if (nfs_inode_attrs_need_update(inode, fattr)) 1145 return nfs_update_inode(inode, fattr); 1146 return nfs_check_inode_attributes(inode, fattr); 1147 } 1148 1149 /** 1150 * nfs_refresh_inode - try to update the inode attribute cache 1151 * @inode - pointer to inode 1152 * @fattr - updated attributes 1153 * 1154 * Check that an RPC call that returned attributes has not overlapped with 1155 * other recent updates of the inode metadata, then decide whether it is 1156 * safe to do a full update of the inode attributes, or whether just to 1157 * call nfs_check_inode_attributes. 1158 */ 1159 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr) 1160 { 1161 int status; 1162 1163 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 1164 return 0; 1165 spin_lock(&inode->i_lock); 1166 status = nfs_refresh_inode_locked(inode, fattr); 1167 spin_unlock(&inode->i_lock); 1168 1169 return status; 1170 } 1171 1172 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr) 1173 { 1174 struct nfs_inode *nfsi = NFS_I(inode); 1175 1176 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE; 1177 if (S_ISDIR(inode->i_mode)) 1178 nfsi->cache_validity |= NFS_INO_INVALID_DATA; 1179 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 1180 return 0; 1181 return nfs_refresh_inode_locked(inode, fattr); 1182 } 1183 1184 /** 1185 * nfs_post_op_update_inode - try to update the inode attribute cache 1186 * @inode - pointer to inode 1187 * @fattr - updated attributes 1188 * 1189 * After an operation that has changed the inode metadata, mark the 1190 * attribute cache as being invalid, then try to update it. 1191 * 1192 * NB: if the server didn't return any post op attributes, this 1193 * function will force the retrieval of attributes before the next 1194 * NFS request. Thus it should be used only for operations that 1195 * are expected to change one or more attributes, to avoid 1196 * unnecessary NFS requests and trips through nfs_update_inode(). 1197 */ 1198 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1199 { 1200 int status; 1201 1202 spin_lock(&inode->i_lock); 1203 status = nfs_post_op_update_inode_locked(inode, fattr); 1204 spin_unlock(&inode->i_lock); 1205 return status; 1206 } 1207 1208 /** 1209 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache 1210 * @inode - pointer to inode 1211 * @fattr - updated attributes 1212 * 1213 * After an operation that has changed the inode metadata, mark the 1214 * attribute cache as being invalid, then try to update it. Fake up 1215 * weak cache consistency data, if none exist. 1216 * 1217 * This function is mainly designed to be used by the ->write_done() functions. 1218 */ 1219 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr) 1220 { 1221 int status; 1222 1223 spin_lock(&inode->i_lock); 1224 /* Don't do a WCC update if these attributes are already stale */ 1225 if ((fattr->valid & NFS_ATTR_FATTR) == 0 || 1226 !nfs_inode_attrs_need_update(inode, fattr)) { 1227 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE 1228 | NFS_ATTR_FATTR_PRESIZE 1229 | NFS_ATTR_FATTR_PREMTIME 1230 | NFS_ATTR_FATTR_PRECTIME); 1231 goto out_noforce; 1232 } 1233 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && 1234 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) { 1235 fattr->pre_change_attr = inode->i_version; 1236 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE; 1237 } 1238 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 && 1239 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) { 1240 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime)); 1241 fattr->valid |= NFS_ATTR_FATTR_PRECTIME; 1242 } 1243 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 && 1244 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) { 1245 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime)); 1246 fattr->valid |= NFS_ATTR_FATTR_PREMTIME; 1247 } 1248 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 && 1249 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) { 1250 fattr->pre_size = i_size_read(inode); 1251 fattr->valid |= NFS_ATTR_FATTR_PRESIZE; 1252 } 1253 out_noforce: 1254 status = nfs_post_op_update_inode_locked(inode, fattr); 1255 spin_unlock(&inode->i_lock); 1256 return status; 1257 } 1258 1259 /* 1260 * Many nfs protocol calls return the new file attributes after 1261 * an operation. Here we update the inode to reflect the state 1262 * of the server's inode. 1263 * 1264 * This is a bit tricky because we have to make sure all dirty pages 1265 * have been sent off to the server before calling invalidate_inode_pages. 1266 * To make sure no other process adds more write requests while we try 1267 * our best to flush them, we make them sleep during the attribute refresh. 1268 * 1269 * A very similar scenario holds for the dir cache. 1270 */ 1271 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1272 { 1273 struct nfs_server *server; 1274 struct nfs_inode *nfsi = NFS_I(inode); 1275 loff_t cur_isize, new_isize; 1276 unsigned long invalid = 0; 1277 unsigned long now = jiffies; 1278 unsigned long save_cache_validity; 1279 1280 dfprintk(VFS, "NFS: %s(%s/%ld fh_crc=0x%08x ct=%d info=0x%x)\n", 1281 __func__, inode->i_sb->s_id, inode->i_ino, 1282 nfs_display_fhandle_hash(NFS_FH(inode)), 1283 atomic_read(&inode->i_count), fattr->valid); 1284 1285 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) { 1286 printk(KERN_ERR "NFS: server %s error: fileid changed\n" 1287 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n", 1288 NFS_SERVER(inode)->nfs_client->cl_hostname, 1289 inode->i_sb->s_id, (long long)nfsi->fileid, 1290 (long long)fattr->fileid); 1291 goto out_err; 1292 } 1293 1294 /* 1295 * Make sure the inode's type hasn't changed. 1296 */ 1297 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) { 1298 /* 1299 * Big trouble! The inode has become a different object. 1300 */ 1301 printk(KERN_DEBUG "NFS: %s: inode %ld mode changed, %07o to %07o\n", 1302 __func__, inode->i_ino, inode->i_mode, fattr->mode); 1303 goto out_err; 1304 } 1305 1306 server = NFS_SERVER(inode); 1307 /* Update the fsid? */ 1308 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) && 1309 !nfs_fsid_equal(&server->fsid, &fattr->fsid) && 1310 !IS_AUTOMOUNT(inode)) 1311 server->fsid = fattr->fsid; 1312 1313 /* 1314 * Update the read time so we don't revalidate too often. 1315 */ 1316 nfsi->read_cache_jiffies = fattr->time_start; 1317 1318 save_cache_validity = nfsi->cache_validity; 1319 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR 1320 | NFS_INO_INVALID_ATIME 1321 | NFS_INO_REVAL_FORCED 1322 | NFS_INO_REVAL_PAGECACHE); 1323 1324 /* Do atomic weak cache consistency updates */ 1325 invalid |= nfs_wcc_update_inode(inode, fattr); 1326 1327 /* More cache consistency checks */ 1328 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) { 1329 if (inode->i_version != fattr->change_attr) { 1330 dprintk("NFS: change_attr change on server for file %s/%ld\n", 1331 inode->i_sb->s_id, inode->i_ino); 1332 invalid |= NFS_INO_INVALID_ATTR 1333 | NFS_INO_INVALID_DATA 1334 | NFS_INO_INVALID_ACCESS 1335 | NFS_INO_INVALID_ACL 1336 | NFS_INO_REVAL_PAGECACHE; 1337 if (S_ISDIR(inode->i_mode)) 1338 nfs_force_lookup_revalidate(inode); 1339 inode->i_version = fattr->change_attr; 1340 } 1341 } else if (server->caps & NFS_CAP_CHANGE_ATTR) 1342 invalid |= save_cache_validity; 1343 1344 if (fattr->valid & NFS_ATTR_FATTR_MTIME) { 1345 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime)); 1346 } else if (server->caps & NFS_CAP_MTIME) 1347 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1348 | NFS_INO_REVAL_FORCED); 1349 1350 if (fattr->valid & NFS_ATTR_FATTR_CTIME) { 1351 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime)); 1352 } else if (server->caps & NFS_CAP_CTIME) 1353 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1354 | NFS_INO_REVAL_FORCED); 1355 1356 /* Check if our cached file size is stale */ 1357 if (fattr->valid & NFS_ATTR_FATTR_SIZE) { 1358 new_isize = nfs_size_to_loff_t(fattr->size); 1359 cur_isize = i_size_read(inode); 1360 if (new_isize != cur_isize) { 1361 /* Do we perhaps have any outstanding writes, or has 1362 * the file grown beyond our last write? */ 1363 if ((nfsi->npages == 0 && !test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) || 1364 new_isize > cur_isize) { 1365 i_size_write(inode, new_isize); 1366 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA; 1367 } 1368 dprintk("NFS: isize change on server for file %s/%ld " 1369 "(%Ld to %Ld)\n", 1370 inode->i_sb->s_id, 1371 inode->i_ino, 1372 (long long)cur_isize, 1373 (long long)new_isize); 1374 } 1375 } else 1376 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1377 | NFS_INO_REVAL_PAGECACHE 1378 | NFS_INO_REVAL_FORCED); 1379 1380 1381 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 1382 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime)); 1383 else if (server->caps & NFS_CAP_ATIME) 1384 invalid |= save_cache_validity & (NFS_INO_INVALID_ATIME 1385 | NFS_INO_REVAL_FORCED); 1386 1387 if (fattr->valid & NFS_ATTR_FATTR_MODE) { 1388 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) { 1389 umode_t newmode = inode->i_mode & S_IFMT; 1390 newmode |= fattr->mode & S_IALLUGO; 1391 inode->i_mode = newmode; 1392 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 1393 } 1394 } else if (server->caps & NFS_CAP_MODE) 1395 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1396 | NFS_INO_INVALID_ACCESS 1397 | NFS_INO_INVALID_ACL 1398 | NFS_INO_REVAL_FORCED); 1399 1400 if (fattr->valid & NFS_ATTR_FATTR_OWNER) { 1401 if (inode->i_uid != fattr->uid) { 1402 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 1403 inode->i_uid = fattr->uid; 1404 } 1405 } else if (server->caps & NFS_CAP_OWNER) 1406 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1407 | NFS_INO_INVALID_ACCESS 1408 | NFS_INO_INVALID_ACL 1409 | NFS_INO_REVAL_FORCED); 1410 1411 if (fattr->valid & NFS_ATTR_FATTR_GROUP) { 1412 if (inode->i_gid != fattr->gid) { 1413 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 1414 inode->i_gid = fattr->gid; 1415 } 1416 } else if (server->caps & NFS_CAP_OWNER_GROUP) 1417 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1418 | NFS_INO_INVALID_ACCESS 1419 | NFS_INO_INVALID_ACL 1420 | NFS_INO_REVAL_FORCED); 1421 1422 if (fattr->valid & NFS_ATTR_FATTR_NLINK) { 1423 if (inode->i_nlink != fattr->nlink) { 1424 invalid |= NFS_INO_INVALID_ATTR; 1425 if (S_ISDIR(inode->i_mode)) 1426 invalid |= NFS_INO_INVALID_DATA; 1427 set_nlink(inode, fattr->nlink); 1428 } 1429 } else if (server->caps & NFS_CAP_NLINK) 1430 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1431 | NFS_INO_REVAL_FORCED); 1432 1433 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) { 1434 /* 1435 * report the blocks in 512byte units 1436 */ 1437 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 1438 } 1439 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED) 1440 inode->i_blocks = fattr->du.nfs2.blocks; 1441 1442 /* Update attrtimeo value if we're out of the unstable period */ 1443 if (invalid & NFS_INO_INVALID_ATTR) { 1444 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 1445 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 1446 nfsi->attrtimeo_timestamp = now; 1447 nfsi->attr_gencount = nfs_inc_attr_generation_counter(); 1448 } else { 1449 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) { 1450 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode)) 1451 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode); 1452 nfsi->attrtimeo_timestamp = now; 1453 } 1454 } 1455 invalid &= ~NFS_INO_INVALID_ATTR; 1456 /* Don't invalidate the data if we were to blame */ 1457 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) 1458 || S_ISLNK(inode->i_mode))) 1459 invalid &= ~NFS_INO_INVALID_DATA; 1460 if (!nfs_have_delegation(inode, FMODE_READ) || 1461 (save_cache_validity & NFS_INO_REVAL_FORCED)) 1462 nfsi->cache_validity |= invalid; 1463 1464 return 0; 1465 out_err: 1466 /* 1467 * No need to worry about unhashing the dentry, as the 1468 * lookup validation will know that the inode is bad. 1469 * (But we fall through to invalidate the caches.) 1470 */ 1471 nfs_invalidate_inode(inode); 1472 return -ESTALE; 1473 } 1474 1475 1476 #ifdef CONFIG_NFS_V4 1477 1478 /* 1479 * Clean out any remaining NFSv4 state that might be left over due 1480 * to open() calls that passed nfs_atomic_lookup, but failed to call 1481 * nfs_open(). 1482 */ 1483 void nfs4_evict_inode(struct inode *inode) 1484 { 1485 truncate_inode_pages(&inode->i_data, 0); 1486 clear_inode(inode); 1487 pnfs_return_layout(inode); 1488 pnfs_destroy_layout(NFS_I(inode)); 1489 /* If we are holding a delegation, return it! */ 1490 nfs_inode_return_delegation_noreclaim(inode); 1491 /* First call standard NFS clear_inode() code */ 1492 nfs_clear_inode(inode); 1493 } 1494 #endif 1495 1496 struct inode *nfs_alloc_inode(struct super_block *sb) 1497 { 1498 struct nfs_inode *nfsi; 1499 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL); 1500 if (!nfsi) 1501 return NULL; 1502 nfsi->flags = 0UL; 1503 nfsi->cache_validity = 0UL; 1504 #ifdef CONFIG_NFS_V3_ACL 1505 nfsi->acl_access = ERR_PTR(-EAGAIN); 1506 nfsi->acl_default = ERR_PTR(-EAGAIN); 1507 #endif 1508 #ifdef CONFIG_NFS_V4 1509 nfsi->nfs4_acl = NULL; 1510 #endif /* CONFIG_NFS_V4 */ 1511 return &nfsi->vfs_inode; 1512 } 1513 1514 static void nfs_i_callback(struct rcu_head *head) 1515 { 1516 struct inode *inode = container_of(head, struct inode, i_rcu); 1517 kmem_cache_free(nfs_inode_cachep, NFS_I(inode)); 1518 } 1519 1520 void nfs_destroy_inode(struct inode *inode) 1521 { 1522 call_rcu(&inode->i_rcu, nfs_i_callback); 1523 } 1524 1525 static inline void nfs4_init_once(struct nfs_inode *nfsi) 1526 { 1527 #ifdef CONFIG_NFS_V4 1528 INIT_LIST_HEAD(&nfsi->open_states); 1529 nfsi->delegation = NULL; 1530 nfsi->delegation_state = 0; 1531 init_rwsem(&nfsi->rwsem); 1532 nfsi->layout = NULL; 1533 #endif 1534 } 1535 1536 static void init_once(void *foo) 1537 { 1538 struct nfs_inode *nfsi = (struct nfs_inode *) foo; 1539 1540 inode_init_once(&nfsi->vfs_inode); 1541 INIT_LIST_HEAD(&nfsi->open_files); 1542 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru); 1543 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru); 1544 INIT_LIST_HEAD(&nfsi->commit_info.list); 1545 nfsi->npages = 0; 1546 nfsi->commit_info.ncommit = 0; 1547 atomic_set(&nfsi->commit_info.rpcs_out, 0); 1548 atomic_set(&nfsi->silly_count, 1); 1549 INIT_HLIST_HEAD(&nfsi->silly_list); 1550 init_waitqueue_head(&nfsi->waitqueue); 1551 nfs4_init_once(nfsi); 1552 } 1553 1554 static int __init nfs_init_inodecache(void) 1555 { 1556 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache", 1557 sizeof(struct nfs_inode), 1558 0, (SLAB_RECLAIM_ACCOUNT| 1559 SLAB_MEM_SPREAD), 1560 init_once); 1561 if (nfs_inode_cachep == NULL) 1562 return -ENOMEM; 1563 1564 return 0; 1565 } 1566 1567 static void nfs_destroy_inodecache(void) 1568 { 1569 kmem_cache_destroy(nfs_inode_cachep); 1570 } 1571 1572 struct workqueue_struct *nfsiod_workqueue; 1573 1574 /* 1575 * start up the nfsiod workqueue 1576 */ 1577 static int nfsiod_start(void) 1578 { 1579 struct workqueue_struct *wq; 1580 dprintk("RPC: creating workqueue nfsiod\n"); 1581 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0); 1582 if (wq == NULL) 1583 return -ENOMEM; 1584 nfsiod_workqueue = wq; 1585 return 0; 1586 } 1587 1588 /* 1589 * Destroy the nfsiod workqueue 1590 */ 1591 static void nfsiod_stop(void) 1592 { 1593 struct workqueue_struct *wq; 1594 1595 wq = nfsiod_workqueue; 1596 if (wq == NULL) 1597 return; 1598 nfsiod_workqueue = NULL; 1599 destroy_workqueue(wq); 1600 } 1601 1602 int nfs_net_id; 1603 EXPORT_SYMBOL_GPL(nfs_net_id); 1604 1605 static int nfs_net_init(struct net *net) 1606 { 1607 nfs_clients_init(net); 1608 return nfs_dns_resolver_cache_init(net); 1609 } 1610 1611 static void nfs_net_exit(struct net *net) 1612 { 1613 nfs_dns_resolver_cache_destroy(net); 1614 nfs_cleanup_cb_ident_idr(net); 1615 } 1616 1617 static struct pernet_operations nfs_net_ops = { 1618 .init = nfs_net_init, 1619 .exit = nfs_net_exit, 1620 .id = &nfs_net_id, 1621 .size = sizeof(struct nfs_net), 1622 }; 1623 1624 /* 1625 * Initialize NFS 1626 */ 1627 static int __init init_nfs_fs(void) 1628 { 1629 int err; 1630 1631 err = nfs_idmap_init(); 1632 if (err < 0) 1633 goto out10; 1634 1635 err = nfs_dns_resolver_init(); 1636 if (err < 0) 1637 goto out9; 1638 1639 err = register_pernet_subsys(&nfs_net_ops); 1640 if (err < 0) 1641 goto out8; 1642 1643 err = nfs_fscache_register(); 1644 if (err < 0) 1645 goto out7; 1646 1647 err = nfsiod_start(); 1648 if (err) 1649 goto out6; 1650 1651 err = nfs_fs_proc_init(); 1652 if (err) 1653 goto out5; 1654 1655 err = nfs_init_nfspagecache(); 1656 if (err) 1657 goto out4; 1658 1659 err = nfs_init_inodecache(); 1660 if (err) 1661 goto out3; 1662 1663 err = nfs_init_readpagecache(); 1664 if (err) 1665 goto out2; 1666 1667 err = nfs_init_writepagecache(); 1668 if (err) 1669 goto out1; 1670 1671 err = nfs_init_directcache(); 1672 if (err) 1673 goto out0; 1674 1675 #ifdef CONFIG_PROC_FS 1676 rpc_proc_register(&init_net, &nfs_rpcstat); 1677 #endif 1678 if ((err = register_nfs_fs()) != 0) 1679 goto out; 1680 return 0; 1681 out: 1682 #ifdef CONFIG_PROC_FS 1683 rpc_proc_unregister(&init_net, "nfs"); 1684 #endif 1685 nfs_destroy_directcache(); 1686 out0: 1687 nfs_destroy_writepagecache(); 1688 out1: 1689 nfs_destroy_readpagecache(); 1690 out2: 1691 nfs_destroy_inodecache(); 1692 out3: 1693 nfs_destroy_nfspagecache(); 1694 out4: 1695 nfs_fs_proc_exit(); 1696 out5: 1697 nfsiod_stop(); 1698 out6: 1699 nfs_fscache_unregister(); 1700 out7: 1701 unregister_pernet_subsys(&nfs_net_ops); 1702 out8: 1703 nfs_dns_resolver_destroy(); 1704 out9: 1705 nfs_idmap_quit(); 1706 out10: 1707 return err; 1708 } 1709 1710 static void __exit exit_nfs_fs(void) 1711 { 1712 nfs_destroy_directcache(); 1713 nfs_destroy_writepagecache(); 1714 nfs_destroy_readpagecache(); 1715 nfs_destroy_inodecache(); 1716 nfs_destroy_nfspagecache(); 1717 nfs_fscache_unregister(); 1718 unregister_pernet_subsys(&nfs_net_ops); 1719 nfs_dns_resolver_destroy(); 1720 nfs_idmap_quit(); 1721 #ifdef CONFIG_PROC_FS 1722 rpc_proc_unregister(&init_net, "nfs"); 1723 #endif 1724 unregister_nfs_fs(); 1725 nfs_fs_proc_exit(); 1726 nfsiod_stop(); 1727 } 1728 1729 /* Not quite true; I just maintain it */ 1730 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>"); 1731 MODULE_LICENSE("GPL"); 1732 module_param(enable_ino64, bool, 0644); 1733 1734 module_init(init_nfs_fs) 1735 module_exit(exit_nfs_fs) 1736