1 /* 2 * linux/fs/nfs/inode.c 3 * 4 * Copyright (C) 1992 Rick Sladkey 5 * 6 * nfs inode and superblock handling functions 7 * 8 * Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some 9 * experimental NFS changes. Modularisation taken straight from SYS5 fs. 10 * 11 * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts. 12 * J.S.Peatfield@damtp.cam.ac.uk 13 * 14 */ 15 16 #include <linux/module.h> 17 #include <linux/init.h> 18 #include <linux/sched.h> 19 #include <linux/time.h> 20 #include <linux/kernel.h> 21 #include <linux/mm.h> 22 #include <linux/string.h> 23 #include <linux/stat.h> 24 #include <linux/errno.h> 25 #include <linux/unistd.h> 26 #include <linux/sunrpc/clnt.h> 27 #include <linux/sunrpc/stats.h> 28 #include <linux/sunrpc/metrics.h> 29 #include <linux/nfs_fs.h> 30 #include <linux/nfs_mount.h> 31 #include <linux/nfs4_mount.h> 32 #include <linux/lockd/bind.h> 33 #include <linux/seq_file.h> 34 #include <linux/mount.h> 35 #include <linux/vfs.h> 36 #include <linux/inet.h> 37 #include <linux/nfs_xdr.h> 38 #include <linux/slab.h> 39 #include <linux/compat.h> 40 #include <linux/freezer.h> 41 #include <linux/crc32.h> 42 43 #include <asm/uaccess.h> 44 45 #include "nfs4_fs.h" 46 #include "callback.h" 47 #include "delegation.h" 48 #include "iostat.h" 49 #include "internal.h" 50 #include "fscache.h" 51 #include "dns_resolve.h" 52 #include "pnfs.h" 53 #include "nfs.h" 54 #include "netns.h" 55 56 #define NFSDBG_FACILITY NFSDBG_VFS 57 58 #define NFS_64_BIT_INODE_NUMBERS_ENABLED 1 59 60 /* Default is to see 64-bit inode numbers */ 61 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED; 62 63 static void nfs_invalidate_inode(struct inode *); 64 static int nfs_update_inode(struct inode *, struct nfs_fattr *); 65 66 static struct kmem_cache * nfs_inode_cachep; 67 68 static inline unsigned long 69 nfs_fattr_to_ino_t(struct nfs_fattr *fattr) 70 { 71 return nfs_fileid_to_ino_t(fattr->fileid); 72 } 73 74 /** 75 * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks 76 * @word: long word containing the bit lock 77 */ 78 int nfs_wait_bit_killable(void *word) 79 { 80 if (fatal_signal_pending(current)) 81 return -ERESTARTSYS; 82 freezable_schedule(); 83 return 0; 84 } 85 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable); 86 87 /** 88 * nfs_compat_user_ino64 - returns the user-visible inode number 89 * @fileid: 64-bit fileid 90 * 91 * This function returns a 32-bit inode number if the boot parameter 92 * nfs.enable_ino64 is zero. 93 */ 94 u64 nfs_compat_user_ino64(u64 fileid) 95 { 96 #ifdef CONFIG_COMPAT 97 compat_ulong_t ino; 98 #else 99 unsigned long ino; 100 #endif 101 102 if (enable_ino64) 103 return fileid; 104 ino = fileid; 105 if (sizeof(ino) < sizeof(fileid)) 106 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8; 107 return ino; 108 } 109 110 int nfs_drop_inode(struct inode *inode) 111 { 112 return NFS_STALE(inode) || generic_drop_inode(inode); 113 } 114 EXPORT_SYMBOL_GPL(nfs_drop_inode); 115 116 void nfs_clear_inode(struct inode *inode) 117 { 118 /* 119 * The following should never happen... 120 */ 121 WARN_ON_ONCE(nfs_have_writebacks(inode)); 122 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files)); 123 nfs_zap_acl_cache(inode); 124 nfs_access_zap_cache(inode); 125 nfs_fscache_release_inode_cookie(inode); 126 } 127 EXPORT_SYMBOL_GPL(nfs_clear_inode); 128 129 void nfs_evict_inode(struct inode *inode) 130 { 131 truncate_inode_pages(&inode->i_data, 0); 132 clear_inode(inode); 133 nfs_clear_inode(inode); 134 } 135 136 /** 137 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk 138 */ 139 int nfs_sync_mapping(struct address_space *mapping) 140 { 141 int ret = 0; 142 143 if (mapping->nrpages != 0) { 144 unmap_mapping_range(mapping, 0, 0, 0); 145 ret = nfs_wb_all(mapping->host); 146 } 147 return ret; 148 } 149 150 /* 151 * Invalidate the local caches 152 */ 153 static void nfs_zap_caches_locked(struct inode *inode) 154 { 155 struct nfs_inode *nfsi = NFS_I(inode); 156 int mode = inode->i_mode; 157 158 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 159 160 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 161 nfsi->attrtimeo_timestamp = jiffies; 162 163 memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf)); 164 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) 165 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE; 166 else 167 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE; 168 } 169 170 void nfs_zap_caches(struct inode *inode) 171 { 172 spin_lock(&inode->i_lock); 173 nfs_zap_caches_locked(inode); 174 spin_unlock(&inode->i_lock); 175 } 176 177 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping) 178 { 179 if (mapping->nrpages != 0) { 180 spin_lock(&inode->i_lock); 181 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; 182 spin_unlock(&inode->i_lock); 183 } 184 } 185 186 void nfs_zap_acl_cache(struct inode *inode) 187 { 188 void (*clear_acl_cache)(struct inode *); 189 190 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache; 191 if (clear_acl_cache != NULL) 192 clear_acl_cache(inode); 193 spin_lock(&inode->i_lock); 194 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL; 195 spin_unlock(&inode->i_lock); 196 } 197 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache); 198 199 void nfs_invalidate_atime(struct inode *inode) 200 { 201 spin_lock(&inode->i_lock); 202 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME; 203 spin_unlock(&inode->i_lock); 204 } 205 EXPORT_SYMBOL_GPL(nfs_invalidate_atime); 206 207 /* 208 * Invalidate, but do not unhash, the inode. 209 * NB: must be called with inode->i_lock held! 210 */ 211 static void nfs_invalidate_inode(struct inode *inode) 212 { 213 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); 214 nfs_zap_caches_locked(inode); 215 } 216 217 struct nfs_find_desc { 218 struct nfs_fh *fh; 219 struct nfs_fattr *fattr; 220 }; 221 222 /* 223 * In NFSv3 we can have 64bit inode numbers. In order to support 224 * this, and re-exported directories (also seen in NFSv2) 225 * we are forced to allow 2 different inodes to have the same 226 * i_ino. 227 */ 228 static int 229 nfs_find_actor(struct inode *inode, void *opaque) 230 { 231 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque; 232 struct nfs_fh *fh = desc->fh; 233 struct nfs_fattr *fattr = desc->fattr; 234 235 if (NFS_FILEID(inode) != fattr->fileid) 236 return 0; 237 if (nfs_compare_fh(NFS_FH(inode), fh)) 238 return 0; 239 if (is_bad_inode(inode) || NFS_STALE(inode)) 240 return 0; 241 return 1; 242 } 243 244 static int 245 nfs_init_locked(struct inode *inode, void *opaque) 246 { 247 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque; 248 struct nfs_fattr *fattr = desc->fattr; 249 250 set_nfs_fileid(inode, fattr->fileid); 251 nfs_copy_fh(NFS_FH(inode), desc->fh); 252 return 0; 253 } 254 255 /* 256 * This is our front-end to iget that looks up inodes by file handle 257 * instead of inode number. 258 */ 259 struct inode * 260 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr) 261 { 262 struct nfs_find_desc desc = { 263 .fh = fh, 264 .fattr = fattr 265 }; 266 struct inode *inode = ERR_PTR(-ENOENT); 267 unsigned long hash; 268 269 nfs_attr_check_mountpoint(sb, fattr); 270 271 if (((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0) && 272 !nfs_attr_use_mounted_on_fileid(fattr)) 273 goto out_no_inode; 274 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0) 275 goto out_no_inode; 276 277 hash = nfs_fattr_to_ino_t(fattr); 278 279 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc); 280 if (inode == NULL) { 281 inode = ERR_PTR(-ENOMEM); 282 goto out_no_inode; 283 } 284 285 if (inode->i_state & I_NEW) { 286 struct nfs_inode *nfsi = NFS_I(inode); 287 unsigned long now = jiffies; 288 289 /* We set i_ino for the few things that still rely on it, 290 * such as stat(2) */ 291 inode->i_ino = hash; 292 293 /* We can't support update_atime(), since the server will reset it */ 294 inode->i_flags |= S_NOATIME|S_NOCMTIME; 295 inode->i_mode = fattr->mode; 296 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0 297 && nfs_server_capable(inode, NFS_CAP_MODE)) 298 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 299 /* Why so? Because we want revalidate for devices/FIFOs, and 300 * that's precisely what we have in nfs_file_inode_operations. 301 */ 302 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops; 303 if (S_ISREG(inode->i_mode)) { 304 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops; 305 inode->i_data.a_ops = &nfs_file_aops; 306 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info; 307 } else if (S_ISDIR(inode->i_mode)) { 308 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops; 309 inode->i_fop = &nfs_dir_operations; 310 inode->i_data.a_ops = &nfs_dir_aops; 311 /* Deal with crossing mountpoints */ 312 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT || 313 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) { 314 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) 315 inode->i_op = &nfs_referral_inode_operations; 316 else 317 inode->i_op = &nfs_mountpoint_inode_operations; 318 inode->i_fop = NULL; 319 inode->i_flags |= S_AUTOMOUNT; 320 } 321 } else if (S_ISLNK(inode->i_mode)) 322 inode->i_op = &nfs_symlink_inode_operations; 323 else 324 init_special_inode(inode, inode->i_mode, fattr->rdev); 325 326 memset(&inode->i_atime, 0, sizeof(inode->i_atime)); 327 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime)); 328 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime)); 329 inode->i_version = 0; 330 inode->i_size = 0; 331 clear_nlink(inode); 332 inode->i_uid = -2; 333 inode->i_gid = -2; 334 inode->i_blocks = 0; 335 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf)); 336 nfsi->write_io = 0; 337 nfsi->read_io = 0; 338 339 nfsi->read_cache_jiffies = fattr->time_start; 340 nfsi->attr_gencount = fattr->gencount; 341 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 342 inode->i_atime = fattr->atime; 343 else if (nfs_server_capable(inode, NFS_CAP_ATIME)) 344 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 345 if (fattr->valid & NFS_ATTR_FATTR_MTIME) 346 inode->i_mtime = fattr->mtime; 347 else if (nfs_server_capable(inode, NFS_CAP_MTIME)) 348 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 349 if (fattr->valid & NFS_ATTR_FATTR_CTIME) 350 inode->i_ctime = fattr->ctime; 351 else if (nfs_server_capable(inode, NFS_CAP_CTIME)) 352 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 353 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) 354 inode->i_version = fattr->change_attr; 355 else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR)) 356 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 357 if (fattr->valid & NFS_ATTR_FATTR_SIZE) 358 inode->i_size = nfs_size_to_loff_t(fattr->size); 359 else 360 nfsi->cache_validity |= NFS_INO_INVALID_ATTR 361 | NFS_INO_REVAL_PAGECACHE; 362 if (fattr->valid & NFS_ATTR_FATTR_NLINK) 363 set_nlink(inode, fattr->nlink); 364 else if (nfs_server_capable(inode, NFS_CAP_NLINK)) 365 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 366 if (fattr->valid & NFS_ATTR_FATTR_OWNER) 367 inode->i_uid = fattr->uid; 368 else if (nfs_server_capable(inode, NFS_CAP_OWNER)) 369 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 370 if (fattr->valid & NFS_ATTR_FATTR_GROUP) 371 inode->i_gid = fattr->gid; 372 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP)) 373 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 374 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED) 375 inode->i_blocks = fattr->du.nfs2.blocks; 376 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) { 377 /* 378 * report the blocks in 512byte units 379 */ 380 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 381 } 382 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 383 nfsi->attrtimeo_timestamp = now; 384 nfsi->access_cache = RB_ROOT; 385 386 nfs_fscache_init_inode_cookie(inode); 387 388 unlock_new_inode(inode); 389 } else 390 nfs_refresh_inode(inode, fattr); 391 dprintk("NFS: nfs_fhget(%s/%Ld fh_crc=0x%08x ct=%d)\n", 392 inode->i_sb->s_id, 393 (long long)NFS_FILEID(inode), 394 nfs_display_fhandle_hash(fh), 395 atomic_read(&inode->i_count)); 396 397 out: 398 return inode; 399 400 out_no_inode: 401 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode)); 402 goto out; 403 } 404 EXPORT_SYMBOL_GPL(nfs_fhget); 405 406 #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) 407 408 int 409 nfs_setattr(struct dentry *dentry, struct iattr *attr) 410 { 411 struct inode *inode = dentry->d_inode; 412 struct nfs_fattr *fattr; 413 int error = -ENOMEM; 414 415 nfs_inc_stats(inode, NFSIOS_VFSSETATTR); 416 417 /* skip mode change if it's just for clearing setuid/setgid */ 418 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID)) 419 attr->ia_valid &= ~ATTR_MODE; 420 421 if (attr->ia_valid & ATTR_SIZE) { 422 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode)) 423 attr->ia_valid &= ~ATTR_SIZE; 424 } 425 426 /* Optimization: if the end result is no change, don't RPC */ 427 attr->ia_valid &= NFS_VALID_ATTRS; 428 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0) 429 return 0; 430 431 /* Write all dirty data */ 432 if (S_ISREG(inode->i_mode)) { 433 nfs_inode_dio_wait(inode); 434 nfs_wb_all(inode); 435 } 436 437 fattr = nfs_alloc_fattr(); 438 if (fattr == NULL) 439 goto out; 440 /* 441 * Return any delegations if we're going to change ACLs 442 */ 443 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) 444 NFS_PROTO(inode)->return_delegation(inode); 445 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr); 446 if (error == 0) 447 nfs_refresh_inode(inode, fattr); 448 nfs_free_fattr(fattr); 449 out: 450 return error; 451 } 452 EXPORT_SYMBOL_GPL(nfs_setattr); 453 454 /** 455 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall 456 * @inode: inode of the file used 457 * @offset: file offset to start truncating 458 * 459 * This is a copy of the common vmtruncate, but with the locking 460 * corrected to take into account the fact that NFS requires 461 * inode->i_size to be updated under the inode->i_lock. 462 */ 463 static int nfs_vmtruncate(struct inode * inode, loff_t offset) 464 { 465 loff_t oldsize; 466 int err; 467 468 err = inode_newsize_ok(inode, offset); 469 if (err) 470 goto out; 471 472 spin_lock(&inode->i_lock); 473 oldsize = inode->i_size; 474 i_size_write(inode, offset); 475 spin_unlock(&inode->i_lock); 476 477 truncate_pagecache(inode, oldsize, offset); 478 out: 479 return err; 480 } 481 482 /** 483 * nfs_setattr_update_inode - Update inode metadata after a setattr call. 484 * @inode: pointer to struct inode 485 * @attr: pointer to struct iattr 486 * 487 * Note: we do this in the *proc.c in order to ensure that 488 * it works for things like exclusive creates too. 489 */ 490 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr) 491 { 492 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) { 493 spin_lock(&inode->i_lock); 494 if ((attr->ia_valid & ATTR_MODE) != 0) { 495 int mode = attr->ia_mode & S_IALLUGO; 496 mode |= inode->i_mode & ~S_IALLUGO; 497 inode->i_mode = mode; 498 } 499 if ((attr->ia_valid & ATTR_UID) != 0) 500 inode->i_uid = attr->ia_uid; 501 if ((attr->ia_valid & ATTR_GID) != 0) 502 inode->i_gid = attr->ia_gid; 503 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 504 spin_unlock(&inode->i_lock); 505 } 506 if ((attr->ia_valid & ATTR_SIZE) != 0) { 507 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC); 508 nfs_vmtruncate(inode, attr->ia_size); 509 } 510 } 511 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode); 512 513 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat) 514 { 515 struct inode *inode = dentry->d_inode; 516 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME; 517 int err; 518 519 /* Flush out writes to the server in order to update c/mtime. */ 520 if (S_ISREG(inode->i_mode)) { 521 nfs_inode_dio_wait(inode); 522 err = filemap_write_and_wait(inode->i_mapping); 523 if (err) 524 goto out; 525 } 526 527 /* 528 * We may force a getattr if the user cares about atime. 529 * 530 * Note that we only have to check the vfsmount flags here: 531 * - NFS always sets S_NOATIME by so checking it would give a 532 * bogus result 533 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is 534 * no point in checking those. 535 */ 536 if ((mnt->mnt_flags & MNT_NOATIME) || 537 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) 538 need_atime = 0; 539 540 if (need_atime) 541 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 542 else 543 err = nfs_revalidate_inode(NFS_SERVER(inode), inode); 544 if (!err) { 545 generic_fillattr(inode, stat); 546 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode)); 547 } 548 out: 549 return err; 550 } 551 EXPORT_SYMBOL_GPL(nfs_getattr); 552 553 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx) 554 { 555 atomic_set(&l_ctx->count, 1); 556 l_ctx->lockowner.l_owner = current->files; 557 l_ctx->lockowner.l_pid = current->tgid; 558 INIT_LIST_HEAD(&l_ctx->list); 559 } 560 561 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx) 562 { 563 struct nfs_lock_context *pos; 564 565 list_for_each_entry(pos, &ctx->lock_context.list, list) { 566 if (pos->lockowner.l_owner != current->files) 567 continue; 568 if (pos->lockowner.l_pid != current->tgid) 569 continue; 570 atomic_inc(&pos->count); 571 return pos; 572 } 573 return NULL; 574 } 575 576 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx) 577 { 578 struct nfs_lock_context *res, *new = NULL; 579 struct inode *inode = ctx->dentry->d_inode; 580 581 spin_lock(&inode->i_lock); 582 res = __nfs_find_lock_context(ctx); 583 if (res == NULL) { 584 spin_unlock(&inode->i_lock); 585 new = kmalloc(sizeof(*new), GFP_KERNEL); 586 if (new == NULL) 587 return ERR_PTR(-ENOMEM); 588 nfs_init_lock_context(new); 589 spin_lock(&inode->i_lock); 590 res = __nfs_find_lock_context(ctx); 591 if (res == NULL) { 592 list_add_tail(&new->list, &ctx->lock_context.list); 593 new->open_context = ctx; 594 res = new; 595 new = NULL; 596 } 597 } 598 spin_unlock(&inode->i_lock); 599 kfree(new); 600 return res; 601 } 602 603 void nfs_put_lock_context(struct nfs_lock_context *l_ctx) 604 { 605 struct nfs_open_context *ctx = l_ctx->open_context; 606 struct inode *inode = ctx->dentry->d_inode; 607 608 if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock)) 609 return; 610 list_del(&l_ctx->list); 611 spin_unlock(&inode->i_lock); 612 kfree(l_ctx); 613 } 614 615 /** 616 * nfs_close_context - Common close_context() routine NFSv2/v3 617 * @ctx: pointer to context 618 * @is_sync: is this a synchronous close 619 * 620 * always ensure that the attributes are up to date if we're mounted 621 * with close-to-open semantics 622 */ 623 void nfs_close_context(struct nfs_open_context *ctx, int is_sync) 624 { 625 struct inode *inode; 626 struct nfs_server *server; 627 628 if (!(ctx->mode & FMODE_WRITE)) 629 return; 630 if (!is_sync) 631 return; 632 inode = ctx->dentry->d_inode; 633 if (!list_empty(&NFS_I(inode)->open_files)) 634 return; 635 server = NFS_SERVER(inode); 636 if (server->flags & NFS_MOUNT_NOCTO) 637 return; 638 nfs_revalidate_inode(server, inode); 639 } 640 EXPORT_SYMBOL_GPL(nfs_close_context); 641 642 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode) 643 { 644 struct nfs_open_context *ctx; 645 struct rpc_cred *cred = rpc_lookup_cred(); 646 if (IS_ERR(cred)) 647 return ERR_CAST(cred); 648 649 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); 650 if (!ctx) { 651 put_rpccred(cred); 652 return ERR_PTR(-ENOMEM); 653 } 654 nfs_sb_active(dentry->d_sb); 655 ctx->dentry = dget(dentry); 656 ctx->cred = cred; 657 ctx->state = NULL; 658 ctx->mode = f_mode; 659 ctx->flags = 0; 660 ctx->error = 0; 661 nfs_init_lock_context(&ctx->lock_context); 662 ctx->lock_context.open_context = ctx; 663 INIT_LIST_HEAD(&ctx->list); 664 ctx->mdsthreshold = NULL; 665 return ctx; 666 } 667 EXPORT_SYMBOL_GPL(alloc_nfs_open_context); 668 669 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx) 670 { 671 if (ctx != NULL) 672 atomic_inc(&ctx->lock_context.count); 673 return ctx; 674 } 675 EXPORT_SYMBOL_GPL(get_nfs_open_context); 676 677 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync) 678 { 679 struct inode *inode = ctx->dentry->d_inode; 680 struct super_block *sb = ctx->dentry->d_sb; 681 682 if (!list_empty(&ctx->list)) { 683 if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock)) 684 return; 685 list_del(&ctx->list); 686 spin_unlock(&inode->i_lock); 687 } else if (!atomic_dec_and_test(&ctx->lock_context.count)) 688 return; 689 if (inode != NULL) 690 NFS_PROTO(inode)->close_context(ctx, is_sync); 691 if (ctx->cred != NULL) 692 put_rpccred(ctx->cred); 693 dput(ctx->dentry); 694 if (is_sync) 695 nfs_sb_deactive(sb); 696 else 697 nfs_sb_deactive_async(sb); 698 kfree(ctx->mdsthreshold); 699 kfree(ctx); 700 } 701 702 void put_nfs_open_context(struct nfs_open_context *ctx) 703 { 704 __put_nfs_open_context(ctx, 0); 705 } 706 EXPORT_SYMBOL_GPL(put_nfs_open_context); 707 708 /* 709 * Ensure that mmap has a recent RPC credential for use when writing out 710 * shared pages 711 */ 712 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx) 713 { 714 struct inode *inode = filp->f_path.dentry->d_inode; 715 struct nfs_inode *nfsi = NFS_I(inode); 716 717 filp->private_data = get_nfs_open_context(ctx); 718 spin_lock(&inode->i_lock); 719 list_add(&ctx->list, &nfsi->open_files); 720 spin_unlock(&inode->i_lock); 721 } 722 EXPORT_SYMBOL_GPL(nfs_file_set_open_context); 723 724 /* 725 * Given an inode, search for an open context with the desired characteristics 726 */ 727 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode) 728 { 729 struct nfs_inode *nfsi = NFS_I(inode); 730 struct nfs_open_context *pos, *ctx = NULL; 731 732 spin_lock(&inode->i_lock); 733 list_for_each_entry(pos, &nfsi->open_files, list) { 734 if (cred != NULL && pos->cred != cred) 735 continue; 736 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode) 737 continue; 738 ctx = get_nfs_open_context(pos); 739 break; 740 } 741 spin_unlock(&inode->i_lock); 742 return ctx; 743 } 744 745 static void nfs_file_clear_open_context(struct file *filp) 746 { 747 struct inode *inode = filp->f_path.dentry->d_inode; 748 struct nfs_open_context *ctx = nfs_file_open_context(filp); 749 750 if (ctx) { 751 filp->private_data = NULL; 752 spin_lock(&inode->i_lock); 753 list_move_tail(&ctx->list, &NFS_I(inode)->open_files); 754 spin_unlock(&inode->i_lock); 755 __put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1); 756 } 757 } 758 759 /* 760 * These allocate and release file read/write context information. 761 */ 762 int nfs_open(struct inode *inode, struct file *filp) 763 { 764 struct nfs_open_context *ctx; 765 766 ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode); 767 if (IS_ERR(ctx)) 768 return PTR_ERR(ctx); 769 nfs_file_set_open_context(filp, ctx); 770 put_nfs_open_context(ctx); 771 nfs_fscache_set_inode_cookie(inode, filp); 772 return 0; 773 } 774 775 int nfs_release(struct inode *inode, struct file *filp) 776 { 777 nfs_file_clear_open_context(filp); 778 return 0; 779 } 780 781 /* 782 * This function is called whenever some part of NFS notices that 783 * the cached attributes have to be refreshed. 784 */ 785 int 786 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) 787 { 788 int status = -ESTALE; 789 struct nfs_fattr *fattr = NULL; 790 struct nfs_inode *nfsi = NFS_I(inode); 791 792 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n", 793 inode->i_sb->s_id, (long long)NFS_FILEID(inode)); 794 795 if (is_bad_inode(inode)) 796 goto out; 797 if (NFS_STALE(inode)) 798 goto out; 799 800 status = -ENOMEM; 801 fattr = nfs_alloc_fattr(); 802 if (fattr == NULL) 803 goto out; 804 805 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE); 806 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr); 807 if (status != 0) { 808 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n", 809 inode->i_sb->s_id, 810 (long long)NFS_FILEID(inode), status); 811 if (status == -ESTALE) { 812 nfs_zap_caches(inode); 813 if (!S_ISDIR(inode->i_mode)) 814 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); 815 } 816 goto out; 817 } 818 819 status = nfs_refresh_inode(inode, fattr); 820 if (status) { 821 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n", 822 inode->i_sb->s_id, 823 (long long)NFS_FILEID(inode), status); 824 goto out; 825 } 826 827 if (nfsi->cache_validity & NFS_INO_INVALID_ACL) 828 nfs_zap_acl_cache(inode); 829 830 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n", 831 inode->i_sb->s_id, 832 (long long)NFS_FILEID(inode)); 833 834 out: 835 nfs_free_fattr(fattr); 836 return status; 837 } 838 839 int nfs_attribute_timeout(struct inode *inode) 840 { 841 struct nfs_inode *nfsi = NFS_I(inode); 842 843 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo); 844 } 845 846 static int nfs_attribute_cache_expired(struct inode *inode) 847 { 848 if (nfs_have_delegated_attributes(inode)) 849 return 0; 850 return nfs_attribute_timeout(inode); 851 } 852 853 /** 854 * nfs_revalidate_inode - Revalidate the inode attributes 855 * @server - pointer to nfs_server struct 856 * @inode - pointer to inode struct 857 * 858 * Updates inode attribute information by retrieving the data from the server. 859 */ 860 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) 861 { 862 if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR) 863 && !nfs_attribute_cache_expired(inode)) 864 return NFS_STALE(inode) ? -ESTALE : 0; 865 return __nfs_revalidate_inode(server, inode); 866 } 867 EXPORT_SYMBOL_GPL(nfs_revalidate_inode); 868 869 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping) 870 { 871 struct nfs_inode *nfsi = NFS_I(inode); 872 873 if (mapping->nrpages != 0) { 874 int ret = invalidate_inode_pages2(mapping); 875 if (ret < 0) 876 return ret; 877 } 878 spin_lock(&inode->i_lock); 879 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA; 880 if (S_ISDIR(inode->i_mode)) 881 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf)); 882 spin_unlock(&inode->i_lock); 883 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE); 884 nfs_fscache_reset_inode_cookie(inode); 885 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n", 886 inode->i_sb->s_id, (long long)NFS_FILEID(inode)); 887 return 0; 888 } 889 890 static bool nfs_mapping_need_revalidate_inode(struct inode *inode) 891 { 892 if (nfs_have_delegated_attributes(inode)) 893 return false; 894 return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE) 895 || nfs_attribute_timeout(inode) 896 || NFS_STALE(inode); 897 } 898 899 /** 900 * nfs_revalidate_mapping - Revalidate the pagecache 901 * @inode - pointer to host inode 902 * @mapping - pointer to mapping 903 */ 904 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping) 905 { 906 struct nfs_inode *nfsi = NFS_I(inode); 907 int ret = 0; 908 909 /* swapfiles are not supposed to be shared. */ 910 if (IS_SWAPFILE(inode)) 911 goto out; 912 913 if (nfs_mapping_need_revalidate_inode(inode)) { 914 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 915 if (ret < 0) 916 goto out; 917 } 918 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 919 ret = nfs_invalidate_mapping(inode, mapping); 920 out: 921 return ret; 922 } 923 924 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr) 925 { 926 struct nfs_inode *nfsi = NFS_I(inode); 927 unsigned long ret = 0; 928 929 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE) 930 && (fattr->valid & NFS_ATTR_FATTR_CHANGE) 931 && inode->i_version == fattr->pre_change_attr) { 932 inode->i_version = fattr->change_attr; 933 if (S_ISDIR(inode->i_mode)) 934 nfsi->cache_validity |= NFS_INO_INVALID_DATA; 935 ret |= NFS_INO_INVALID_ATTR; 936 } 937 /* If we have atomic WCC data, we may update some attributes */ 938 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME) 939 && (fattr->valid & NFS_ATTR_FATTR_CTIME) 940 && timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) { 941 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime)); 942 ret |= NFS_INO_INVALID_ATTR; 943 } 944 945 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME) 946 && (fattr->valid & NFS_ATTR_FATTR_MTIME) 947 && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) { 948 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime)); 949 if (S_ISDIR(inode->i_mode)) 950 nfsi->cache_validity |= NFS_INO_INVALID_DATA; 951 ret |= NFS_INO_INVALID_ATTR; 952 } 953 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE) 954 && (fattr->valid & NFS_ATTR_FATTR_SIZE) 955 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size) 956 && nfsi->npages == 0) { 957 i_size_write(inode, nfs_size_to_loff_t(fattr->size)); 958 ret |= NFS_INO_INVALID_ATTR; 959 } 960 return ret; 961 } 962 963 /** 964 * nfs_check_inode_attributes - verify consistency of the inode attribute cache 965 * @inode - pointer to inode 966 * @fattr - updated attributes 967 * 968 * Verifies the attribute cache. If we have just changed the attributes, 969 * so that fattr carries weak cache consistency data, then it may 970 * also update the ctime/mtime/change_attribute. 971 */ 972 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr) 973 { 974 struct nfs_inode *nfsi = NFS_I(inode); 975 loff_t cur_size, new_isize; 976 unsigned long invalid = 0; 977 978 979 if (nfs_have_delegated_attributes(inode)) 980 return 0; 981 /* Has the inode gone and changed behind our back? */ 982 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) 983 return -EIO; 984 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) 985 return -EIO; 986 987 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && 988 inode->i_version != fattr->change_attr) 989 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE; 990 991 /* Verify a few of the more important attributes */ 992 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime)) 993 invalid |= NFS_INO_INVALID_ATTR; 994 995 if (fattr->valid & NFS_ATTR_FATTR_SIZE) { 996 cur_size = i_size_read(inode); 997 new_isize = nfs_size_to_loff_t(fattr->size); 998 if (cur_size != new_isize && nfsi->npages == 0) 999 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE; 1000 } 1001 1002 /* Have any file permissions changed? */ 1003 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) 1004 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL; 1005 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && inode->i_uid != fattr->uid) 1006 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL; 1007 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && inode->i_gid != fattr->gid) 1008 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL; 1009 1010 /* Has the link count changed? */ 1011 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink) 1012 invalid |= NFS_INO_INVALID_ATTR; 1013 1014 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime)) 1015 invalid |= NFS_INO_INVALID_ATIME; 1016 1017 if (invalid != 0) 1018 nfsi->cache_validity |= invalid; 1019 1020 nfsi->read_cache_jiffies = fattr->time_start; 1021 return 0; 1022 } 1023 1024 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr) 1025 { 1026 if (!(fattr->valid & NFS_ATTR_FATTR_CTIME)) 1027 return 0; 1028 return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0; 1029 } 1030 1031 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr) 1032 { 1033 if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) 1034 return 0; 1035 return nfs_size_to_loff_t(fattr->size) > i_size_read(inode); 1036 } 1037 1038 static atomic_long_t nfs_attr_generation_counter; 1039 1040 static unsigned long nfs_read_attr_generation_counter(void) 1041 { 1042 return atomic_long_read(&nfs_attr_generation_counter); 1043 } 1044 1045 unsigned long nfs_inc_attr_generation_counter(void) 1046 { 1047 return atomic_long_inc_return(&nfs_attr_generation_counter); 1048 } 1049 1050 void nfs_fattr_init(struct nfs_fattr *fattr) 1051 { 1052 fattr->valid = 0; 1053 fattr->time_start = jiffies; 1054 fattr->gencount = nfs_inc_attr_generation_counter(); 1055 fattr->owner_name = NULL; 1056 fattr->group_name = NULL; 1057 } 1058 EXPORT_SYMBOL_GPL(nfs_fattr_init); 1059 1060 struct nfs_fattr *nfs_alloc_fattr(void) 1061 { 1062 struct nfs_fattr *fattr; 1063 1064 fattr = kmalloc(sizeof(*fattr), GFP_NOFS); 1065 if (fattr != NULL) 1066 nfs_fattr_init(fattr); 1067 return fattr; 1068 } 1069 EXPORT_SYMBOL_GPL(nfs_alloc_fattr); 1070 1071 struct nfs_fh *nfs_alloc_fhandle(void) 1072 { 1073 struct nfs_fh *fh; 1074 1075 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS); 1076 if (fh != NULL) 1077 fh->size = 0; 1078 return fh; 1079 } 1080 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle); 1081 1082 #ifdef NFS_DEBUG 1083 /* 1084 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle 1085 * in the same way that wireshark does 1086 * 1087 * @fh: file handle 1088 * 1089 * For debugging only. 1090 */ 1091 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh) 1092 { 1093 /* wireshark uses 32-bit AUTODIN crc and does a bitwise 1094 * not on the result */ 1095 return ~crc32(0xFFFFFFFF, &fh->data[0], fh->size); 1096 } 1097 1098 /* 1099 * _nfs_display_fhandle - display an NFS file handle on the console 1100 * 1101 * @fh: file handle to display 1102 * @caption: display caption 1103 * 1104 * For debugging only. 1105 */ 1106 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption) 1107 { 1108 unsigned short i; 1109 1110 if (fh == NULL || fh->size == 0) { 1111 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh); 1112 return; 1113 } 1114 1115 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n", 1116 caption, fh, fh->size, _nfs_display_fhandle_hash(fh)); 1117 for (i = 0; i < fh->size; i += 16) { 1118 __be32 *pos = (__be32 *)&fh->data[i]; 1119 1120 switch ((fh->size - i - 1) >> 2) { 1121 case 0: 1122 printk(KERN_DEFAULT " %08x\n", 1123 be32_to_cpup(pos)); 1124 break; 1125 case 1: 1126 printk(KERN_DEFAULT " %08x %08x\n", 1127 be32_to_cpup(pos), be32_to_cpup(pos + 1)); 1128 break; 1129 case 2: 1130 printk(KERN_DEFAULT " %08x %08x %08x\n", 1131 be32_to_cpup(pos), be32_to_cpup(pos + 1), 1132 be32_to_cpup(pos + 2)); 1133 break; 1134 default: 1135 printk(KERN_DEFAULT " %08x %08x %08x %08x\n", 1136 be32_to_cpup(pos), be32_to_cpup(pos + 1), 1137 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3)); 1138 } 1139 } 1140 } 1141 #endif 1142 1143 /** 1144 * nfs_inode_attrs_need_update - check if the inode attributes need updating 1145 * @inode - pointer to inode 1146 * @fattr - attributes 1147 * 1148 * Attempt to divine whether or not an RPC call reply carrying stale 1149 * attributes got scheduled after another call carrying updated ones. 1150 * 1151 * To do so, the function first assumes that a more recent ctime means 1152 * that the attributes in fattr are newer, however it also attempt to 1153 * catch the case where ctime either didn't change, or went backwards 1154 * (if someone reset the clock on the server) by looking at whether 1155 * or not this RPC call was started after the inode was last updated. 1156 * Note also the check for wraparound of 'attr_gencount' 1157 * 1158 * The function returns 'true' if it thinks the attributes in 'fattr' are 1159 * more recent than the ones cached in the inode. 1160 * 1161 */ 1162 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr) 1163 { 1164 const struct nfs_inode *nfsi = NFS_I(inode); 1165 1166 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 || 1167 nfs_ctime_need_update(inode, fattr) || 1168 nfs_size_need_update(inode, fattr) || 1169 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0); 1170 } 1171 1172 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr) 1173 { 1174 if (nfs_inode_attrs_need_update(inode, fattr)) 1175 return nfs_update_inode(inode, fattr); 1176 return nfs_check_inode_attributes(inode, fattr); 1177 } 1178 1179 /** 1180 * nfs_refresh_inode - try to update the inode attribute cache 1181 * @inode - pointer to inode 1182 * @fattr - updated attributes 1183 * 1184 * Check that an RPC call that returned attributes has not overlapped with 1185 * other recent updates of the inode metadata, then decide whether it is 1186 * safe to do a full update of the inode attributes, or whether just to 1187 * call nfs_check_inode_attributes. 1188 */ 1189 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr) 1190 { 1191 int status; 1192 1193 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 1194 return 0; 1195 spin_lock(&inode->i_lock); 1196 status = nfs_refresh_inode_locked(inode, fattr); 1197 spin_unlock(&inode->i_lock); 1198 1199 return status; 1200 } 1201 EXPORT_SYMBOL_GPL(nfs_refresh_inode); 1202 1203 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr) 1204 { 1205 struct nfs_inode *nfsi = NFS_I(inode); 1206 1207 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE; 1208 if (S_ISDIR(inode->i_mode)) 1209 nfsi->cache_validity |= NFS_INO_INVALID_DATA; 1210 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 1211 return 0; 1212 return nfs_refresh_inode_locked(inode, fattr); 1213 } 1214 1215 /** 1216 * nfs_post_op_update_inode - try to update the inode attribute cache 1217 * @inode - pointer to inode 1218 * @fattr - updated attributes 1219 * 1220 * After an operation that has changed the inode metadata, mark the 1221 * attribute cache as being invalid, then try to update it. 1222 * 1223 * NB: if the server didn't return any post op attributes, this 1224 * function will force the retrieval of attributes before the next 1225 * NFS request. Thus it should be used only for operations that 1226 * are expected to change one or more attributes, to avoid 1227 * unnecessary NFS requests and trips through nfs_update_inode(). 1228 */ 1229 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1230 { 1231 int status; 1232 1233 spin_lock(&inode->i_lock); 1234 status = nfs_post_op_update_inode_locked(inode, fattr); 1235 spin_unlock(&inode->i_lock); 1236 return status; 1237 } 1238 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode); 1239 1240 /** 1241 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache 1242 * @inode - pointer to inode 1243 * @fattr - updated attributes 1244 * 1245 * After an operation that has changed the inode metadata, mark the 1246 * attribute cache as being invalid, then try to update it. Fake up 1247 * weak cache consistency data, if none exist. 1248 * 1249 * This function is mainly designed to be used by the ->write_done() functions. 1250 */ 1251 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr) 1252 { 1253 int status; 1254 1255 spin_lock(&inode->i_lock); 1256 /* Don't do a WCC update if these attributes are already stale */ 1257 if ((fattr->valid & NFS_ATTR_FATTR) == 0 || 1258 !nfs_inode_attrs_need_update(inode, fattr)) { 1259 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE 1260 | NFS_ATTR_FATTR_PRESIZE 1261 | NFS_ATTR_FATTR_PREMTIME 1262 | NFS_ATTR_FATTR_PRECTIME); 1263 goto out_noforce; 1264 } 1265 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && 1266 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) { 1267 fattr->pre_change_attr = inode->i_version; 1268 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE; 1269 } 1270 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 && 1271 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) { 1272 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime)); 1273 fattr->valid |= NFS_ATTR_FATTR_PRECTIME; 1274 } 1275 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 && 1276 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) { 1277 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime)); 1278 fattr->valid |= NFS_ATTR_FATTR_PREMTIME; 1279 } 1280 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 && 1281 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) { 1282 fattr->pre_size = i_size_read(inode); 1283 fattr->valid |= NFS_ATTR_FATTR_PRESIZE; 1284 } 1285 out_noforce: 1286 status = nfs_post_op_update_inode_locked(inode, fattr); 1287 spin_unlock(&inode->i_lock); 1288 return status; 1289 } 1290 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc); 1291 1292 /* 1293 * Many nfs protocol calls return the new file attributes after 1294 * an operation. Here we update the inode to reflect the state 1295 * of the server's inode. 1296 * 1297 * This is a bit tricky because we have to make sure all dirty pages 1298 * have been sent off to the server before calling invalidate_inode_pages. 1299 * To make sure no other process adds more write requests while we try 1300 * our best to flush them, we make them sleep during the attribute refresh. 1301 * 1302 * A very similar scenario holds for the dir cache. 1303 */ 1304 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1305 { 1306 struct nfs_server *server; 1307 struct nfs_inode *nfsi = NFS_I(inode); 1308 loff_t cur_isize, new_isize; 1309 unsigned long invalid = 0; 1310 unsigned long now = jiffies; 1311 unsigned long save_cache_validity; 1312 1313 dfprintk(VFS, "NFS: %s(%s/%ld fh_crc=0x%08x ct=%d info=0x%x)\n", 1314 __func__, inode->i_sb->s_id, inode->i_ino, 1315 nfs_display_fhandle_hash(NFS_FH(inode)), 1316 atomic_read(&inode->i_count), fattr->valid); 1317 1318 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) { 1319 printk(KERN_ERR "NFS: server %s error: fileid changed\n" 1320 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n", 1321 NFS_SERVER(inode)->nfs_client->cl_hostname, 1322 inode->i_sb->s_id, (long long)nfsi->fileid, 1323 (long long)fattr->fileid); 1324 goto out_err; 1325 } 1326 1327 /* 1328 * Make sure the inode's type hasn't changed. 1329 */ 1330 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) { 1331 /* 1332 * Big trouble! The inode has become a different object. 1333 */ 1334 printk(KERN_DEBUG "NFS: %s: inode %ld mode changed, %07o to %07o\n", 1335 __func__, inode->i_ino, inode->i_mode, fattr->mode); 1336 goto out_err; 1337 } 1338 1339 server = NFS_SERVER(inode); 1340 /* Update the fsid? */ 1341 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) && 1342 !nfs_fsid_equal(&server->fsid, &fattr->fsid) && 1343 !IS_AUTOMOUNT(inode)) 1344 server->fsid = fattr->fsid; 1345 1346 /* 1347 * Update the read time so we don't revalidate too often. 1348 */ 1349 nfsi->read_cache_jiffies = fattr->time_start; 1350 1351 save_cache_validity = nfsi->cache_validity; 1352 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR 1353 | NFS_INO_INVALID_ATIME 1354 | NFS_INO_REVAL_FORCED 1355 | NFS_INO_REVAL_PAGECACHE); 1356 1357 /* Do atomic weak cache consistency updates */ 1358 invalid |= nfs_wcc_update_inode(inode, fattr); 1359 1360 /* More cache consistency checks */ 1361 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) { 1362 if (inode->i_version != fattr->change_attr) { 1363 dprintk("NFS: change_attr change on server for file %s/%ld\n", 1364 inode->i_sb->s_id, inode->i_ino); 1365 invalid |= NFS_INO_INVALID_ATTR 1366 | NFS_INO_INVALID_DATA 1367 | NFS_INO_INVALID_ACCESS 1368 | NFS_INO_INVALID_ACL 1369 | NFS_INO_REVAL_PAGECACHE; 1370 if (S_ISDIR(inode->i_mode)) 1371 nfs_force_lookup_revalidate(inode); 1372 inode->i_version = fattr->change_attr; 1373 } 1374 } else if (server->caps & NFS_CAP_CHANGE_ATTR) 1375 invalid |= save_cache_validity; 1376 1377 if (fattr->valid & NFS_ATTR_FATTR_MTIME) { 1378 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime)); 1379 } else if (server->caps & NFS_CAP_MTIME) 1380 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1381 | NFS_INO_REVAL_FORCED); 1382 1383 if (fattr->valid & NFS_ATTR_FATTR_CTIME) { 1384 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime)); 1385 } else if (server->caps & NFS_CAP_CTIME) 1386 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1387 | NFS_INO_REVAL_FORCED); 1388 1389 /* Check if our cached file size is stale */ 1390 if (fattr->valid & NFS_ATTR_FATTR_SIZE) { 1391 new_isize = nfs_size_to_loff_t(fattr->size); 1392 cur_isize = i_size_read(inode); 1393 if (new_isize != cur_isize) { 1394 /* Do we perhaps have any outstanding writes, or has 1395 * the file grown beyond our last write? */ 1396 if ((nfsi->npages == 0 && !test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) || 1397 new_isize > cur_isize) { 1398 i_size_write(inode, new_isize); 1399 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA; 1400 } 1401 dprintk("NFS: isize change on server for file %s/%ld " 1402 "(%Ld to %Ld)\n", 1403 inode->i_sb->s_id, 1404 inode->i_ino, 1405 (long long)cur_isize, 1406 (long long)new_isize); 1407 } 1408 } else 1409 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1410 | NFS_INO_REVAL_PAGECACHE 1411 | NFS_INO_REVAL_FORCED); 1412 1413 1414 if (fattr->valid & NFS_ATTR_FATTR_ATIME) 1415 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime)); 1416 else if (server->caps & NFS_CAP_ATIME) 1417 invalid |= save_cache_validity & (NFS_INO_INVALID_ATIME 1418 | NFS_INO_REVAL_FORCED); 1419 1420 if (fattr->valid & NFS_ATTR_FATTR_MODE) { 1421 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) { 1422 umode_t newmode = inode->i_mode & S_IFMT; 1423 newmode |= fattr->mode & S_IALLUGO; 1424 inode->i_mode = newmode; 1425 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 1426 } 1427 } else if (server->caps & NFS_CAP_MODE) 1428 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1429 | NFS_INO_INVALID_ACCESS 1430 | NFS_INO_INVALID_ACL 1431 | NFS_INO_REVAL_FORCED); 1432 1433 if (fattr->valid & NFS_ATTR_FATTR_OWNER) { 1434 if (inode->i_uid != fattr->uid) { 1435 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 1436 inode->i_uid = fattr->uid; 1437 } 1438 } else if (server->caps & NFS_CAP_OWNER) 1439 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1440 | NFS_INO_INVALID_ACCESS 1441 | NFS_INO_INVALID_ACL 1442 | NFS_INO_REVAL_FORCED); 1443 1444 if (fattr->valid & NFS_ATTR_FATTR_GROUP) { 1445 if (inode->i_gid != fattr->gid) { 1446 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 1447 inode->i_gid = fattr->gid; 1448 } 1449 } else if (server->caps & NFS_CAP_OWNER_GROUP) 1450 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1451 | NFS_INO_INVALID_ACCESS 1452 | NFS_INO_INVALID_ACL 1453 | NFS_INO_REVAL_FORCED); 1454 1455 if (fattr->valid & NFS_ATTR_FATTR_NLINK) { 1456 if (inode->i_nlink != fattr->nlink) { 1457 invalid |= NFS_INO_INVALID_ATTR; 1458 if (S_ISDIR(inode->i_mode)) 1459 invalid |= NFS_INO_INVALID_DATA; 1460 set_nlink(inode, fattr->nlink); 1461 } 1462 } else if (server->caps & NFS_CAP_NLINK) 1463 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR 1464 | NFS_INO_REVAL_FORCED); 1465 1466 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) { 1467 /* 1468 * report the blocks in 512byte units 1469 */ 1470 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 1471 } 1472 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED) 1473 inode->i_blocks = fattr->du.nfs2.blocks; 1474 1475 /* Update attrtimeo value if we're out of the unstable period */ 1476 if (invalid & NFS_INO_INVALID_ATTR) { 1477 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 1478 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 1479 nfsi->attrtimeo_timestamp = now; 1480 nfsi->attr_gencount = nfs_inc_attr_generation_counter(); 1481 } else { 1482 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) { 1483 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode)) 1484 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode); 1485 nfsi->attrtimeo_timestamp = now; 1486 } 1487 } 1488 invalid &= ~NFS_INO_INVALID_ATTR; 1489 /* Don't invalidate the data if we were to blame */ 1490 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) 1491 || S_ISLNK(inode->i_mode))) 1492 invalid &= ~NFS_INO_INVALID_DATA; 1493 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) || 1494 (save_cache_validity & NFS_INO_REVAL_FORCED)) 1495 nfsi->cache_validity |= invalid; 1496 1497 return 0; 1498 out_err: 1499 /* 1500 * No need to worry about unhashing the dentry, as the 1501 * lookup validation will know that the inode is bad. 1502 * (But we fall through to invalidate the caches.) 1503 */ 1504 nfs_invalidate_inode(inode); 1505 return -ESTALE; 1506 } 1507 1508 struct inode *nfs_alloc_inode(struct super_block *sb) 1509 { 1510 struct nfs_inode *nfsi; 1511 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL); 1512 if (!nfsi) 1513 return NULL; 1514 nfsi->flags = 0UL; 1515 nfsi->cache_validity = 0UL; 1516 #ifdef CONFIG_NFS_V3_ACL 1517 nfsi->acl_access = ERR_PTR(-EAGAIN); 1518 nfsi->acl_default = ERR_PTR(-EAGAIN); 1519 #endif 1520 #if IS_ENABLED(CONFIG_NFS_V4) 1521 nfsi->nfs4_acl = NULL; 1522 #endif /* CONFIG_NFS_V4 */ 1523 return &nfsi->vfs_inode; 1524 } 1525 EXPORT_SYMBOL_GPL(nfs_alloc_inode); 1526 1527 static void nfs_i_callback(struct rcu_head *head) 1528 { 1529 struct inode *inode = container_of(head, struct inode, i_rcu); 1530 kmem_cache_free(nfs_inode_cachep, NFS_I(inode)); 1531 } 1532 1533 void nfs_destroy_inode(struct inode *inode) 1534 { 1535 call_rcu(&inode->i_rcu, nfs_i_callback); 1536 } 1537 EXPORT_SYMBOL_GPL(nfs_destroy_inode); 1538 1539 static inline void nfs4_init_once(struct nfs_inode *nfsi) 1540 { 1541 #if IS_ENABLED(CONFIG_NFS_V4) 1542 INIT_LIST_HEAD(&nfsi->open_states); 1543 nfsi->delegation = NULL; 1544 nfsi->delegation_state = 0; 1545 init_rwsem(&nfsi->rwsem); 1546 nfsi->layout = NULL; 1547 #endif 1548 } 1549 1550 static void init_once(void *foo) 1551 { 1552 struct nfs_inode *nfsi = (struct nfs_inode *) foo; 1553 1554 inode_init_once(&nfsi->vfs_inode); 1555 INIT_LIST_HEAD(&nfsi->open_files); 1556 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru); 1557 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru); 1558 INIT_LIST_HEAD(&nfsi->commit_info.list); 1559 nfsi->npages = 0; 1560 nfsi->commit_info.ncommit = 0; 1561 atomic_set(&nfsi->commit_info.rpcs_out, 0); 1562 atomic_set(&nfsi->silly_count, 1); 1563 INIT_HLIST_HEAD(&nfsi->silly_list); 1564 init_waitqueue_head(&nfsi->waitqueue); 1565 nfs4_init_once(nfsi); 1566 } 1567 1568 static int __init nfs_init_inodecache(void) 1569 { 1570 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache", 1571 sizeof(struct nfs_inode), 1572 0, (SLAB_RECLAIM_ACCOUNT| 1573 SLAB_MEM_SPREAD), 1574 init_once); 1575 if (nfs_inode_cachep == NULL) 1576 return -ENOMEM; 1577 1578 return 0; 1579 } 1580 1581 static void nfs_destroy_inodecache(void) 1582 { 1583 /* 1584 * Make sure all delayed rcu free inodes are flushed before we 1585 * destroy cache. 1586 */ 1587 rcu_barrier(); 1588 kmem_cache_destroy(nfs_inode_cachep); 1589 } 1590 1591 struct workqueue_struct *nfsiod_workqueue; 1592 EXPORT_SYMBOL_GPL(nfsiod_workqueue); 1593 1594 /* 1595 * start up the nfsiod workqueue 1596 */ 1597 static int nfsiod_start(void) 1598 { 1599 struct workqueue_struct *wq; 1600 dprintk("RPC: creating workqueue nfsiod\n"); 1601 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0); 1602 if (wq == NULL) 1603 return -ENOMEM; 1604 nfsiod_workqueue = wq; 1605 return 0; 1606 } 1607 1608 /* 1609 * Destroy the nfsiod workqueue 1610 */ 1611 static void nfsiod_stop(void) 1612 { 1613 struct workqueue_struct *wq; 1614 1615 wq = nfsiod_workqueue; 1616 if (wq == NULL) 1617 return; 1618 nfsiod_workqueue = NULL; 1619 destroy_workqueue(wq); 1620 } 1621 1622 int nfs_net_id; 1623 EXPORT_SYMBOL_GPL(nfs_net_id); 1624 1625 static int nfs_net_init(struct net *net) 1626 { 1627 nfs_clients_init(net); 1628 return nfs_dns_resolver_cache_init(net); 1629 } 1630 1631 static void nfs_net_exit(struct net *net) 1632 { 1633 nfs_dns_resolver_cache_destroy(net); 1634 nfs_cleanup_cb_ident_idr(net); 1635 } 1636 1637 static struct pernet_operations nfs_net_ops = { 1638 .init = nfs_net_init, 1639 .exit = nfs_net_exit, 1640 .id = &nfs_net_id, 1641 .size = sizeof(struct nfs_net), 1642 }; 1643 1644 /* 1645 * Initialize NFS 1646 */ 1647 static int __init init_nfs_fs(void) 1648 { 1649 int err; 1650 1651 err = nfs_dns_resolver_init(); 1652 if (err < 0) 1653 goto out10;; 1654 1655 err = register_pernet_subsys(&nfs_net_ops); 1656 if (err < 0) 1657 goto out9; 1658 1659 err = nfs_fscache_register(); 1660 if (err < 0) 1661 goto out8; 1662 1663 err = nfsiod_start(); 1664 if (err) 1665 goto out7; 1666 1667 err = nfs_fs_proc_init(); 1668 if (err) 1669 goto out6; 1670 1671 err = nfs_init_nfspagecache(); 1672 if (err) 1673 goto out5; 1674 1675 err = nfs_init_inodecache(); 1676 if (err) 1677 goto out4; 1678 1679 err = nfs_init_readpagecache(); 1680 if (err) 1681 goto out3; 1682 1683 err = nfs_init_writepagecache(); 1684 if (err) 1685 goto out2; 1686 1687 err = nfs_init_directcache(); 1688 if (err) 1689 goto out1; 1690 1691 #ifdef CONFIG_PROC_FS 1692 rpc_proc_register(&init_net, &nfs_rpcstat); 1693 #endif 1694 if ((err = register_nfs_fs()) != 0) 1695 goto out0; 1696 1697 return 0; 1698 out0: 1699 #ifdef CONFIG_PROC_FS 1700 rpc_proc_unregister(&init_net, "nfs"); 1701 #endif 1702 nfs_destroy_directcache(); 1703 out1: 1704 nfs_destroy_writepagecache(); 1705 out2: 1706 nfs_destroy_readpagecache(); 1707 out3: 1708 nfs_destroy_inodecache(); 1709 out4: 1710 nfs_destroy_nfspagecache(); 1711 out5: 1712 nfs_fs_proc_exit(); 1713 out6: 1714 nfsiod_stop(); 1715 out7: 1716 nfs_fscache_unregister(); 1717 out8: 1718 unregister_pernet_subsys(&nfs_net_ops); 1719 out9: 1720 nfs_dns_resolver_destroy(); 1721 out10: 1722 return err; 1723 } 1724 1725 static void __exit exit_nfs_fs(void) 1726 { 1727 nfs_destroy_directcache(); 1728 nfs_destroy_writepagecache(); 1729 nfs_destroy_readpagecache(); 1730 nfs_destroy_inodecache(); 1731 nfs_destroy_nfspagecache(); 1732 nfs_fscache_unregister(); 1733 unregister_pernet_subsys(&nfs_net_ops); 1734 nfs_dns_resolver_destroy(); 1735 #ifdef CONFIG_PROC_FS 1736 rpc_proc_unregister(&init_net, "nfs"); 1737 #endif 1738 unregister_nfs_fs(); 1739 nfs_fs_proc_exit(); 1740 nfsiod_stop(); 1741 } 1742 1743 /* Not quite true; I just maintain it */ 1744 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>"); 1745 MODULE_LICENSE("GPL"); 1746 module_param(enable_ino64, bool, 0644); 1747 1748 module_init(init_nfs_fs) 1749 module_exit(exit_nfs_fs) 1750