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.Cox@linux.org>, 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 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/smp_lock.h> 34 #include <linux/seq_file.h> 35 #include <linux/mount.h> 36 #include <linux/nfs_idmap.h> 37 #include <linux/vfs.h> 38 #include <linux/inet.h> 39 #include <linux/nfs_xdr.h> 40 41 #include <asm/system.h> 42 #include <asm/uaccess.h> 43 44 #include "nfs4_fs.h" 45 #include "callback.h" 46 #include "delegation.h" 47 #include "iostat.h" 48 #include "internal.h" 49 50 #define NFSDBG_FACILITY NFSDBG_VFS 51 #define NFS_PARANOIA 1 52 53 static void nfs_invalidate_inode(struct inode *); 54 static int nfs_update_inode(struct inode *, struct nfs_fattr *); 55 56 static void nfs_zap_acl_cache(struct inode *); 57 58 static kmem_cache_t * nfs_inode_cachep; 59 60 static inline unsigned long 61 nfs_fattr_to_ino_t(struct nfs_fattr *fattr) 62 { 63 return nfs_fileid_to_ino_t(fattr->fileid); 64 } 65 66 int nfs_write_inode(struct inode *inode, int sync) 67 { 68 int flags = sync ? FLUSH_SYNC : 0; 69 int ret; 70 71 ret = nfs_commit_inode(inode, flags); 72 if (ret < 0) 73 return ret; 74 return 0; 75 } 76 77 void nfs_clear_inode(struct inode *inode) 78 { 79 struct nfs_inode *nfsi = NFS_I(inode); 80 struct rpc_cred *cred; 81 82 /* 83 * The following should never happen... 84 */ 85 BUG_ON(nfs_have_writebacks(inode)); 86 BUG_ON (!list_empty(&nfsi->open_files)); 87 nfs_zap_acl_cache(inode); 88 cred = nfsi->cache_access.cred; 89 if (cred) 90 put_rpccred(cred); 91 BUG_ON(atomic_read(&nfsi->data_updates) != 0); 92 } 93 94 /** 95 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk 96 */ 97 int nfs_sync_mapping(struct address_space *mapping) 98 { 99 int ret; 100 101 if (mapping->nrpages == 0) 102 return 0; 103 unmap_mapping_range(mapping, 0, 0, 0); 104 ret = filemap_write_and_wait(mapping); 105 if (ret != 0) 106 goto out; 107 ret = nfs_wb_all(mapping->host); 108 out: 109 return ret; 110 } 111 112 /* 113 * Invalidate the local caches 114 */ 115 static void nfs_zap_caches_locked(struct inode *inode) 116 { 117 struct nfs_inode *nfsi = NFS_I(inode); 118 int mode = inode->i_mode; 119 120 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 121 122 NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode); 123 NFS_ATTRTIMEO_UPDATE(inode) = jiffies; 124 125 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode))); 126 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) 127 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE; 128 else 129 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE; 130 } 131 132 void nfs_zap_caches(struct inode *inode) 133 { 134 spin_lock(&inode->i_lock); 135 nfs_zap_caches_locked(inode); 136 spin_unlock(&inode->i_lock); 137 } 138 139 static void nfs_zap_acl_cache(struct inode *inode) 140 { 141 void (*clear_acl_cache)(struct inode *); 142 143 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache; 144 if (clear_acl_cache != NULL) 145 clear_acl_cache(inode); 146 spin_lock(&inode->i_lock); 147 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL; 148 spin_unlock(&inode->i_lock); 149 } 150 151 /* 152 * Invalidate, but do not unhash, the inode. 153 * NB: must be called with inode->i_lock held! 154 */ 155 static void nfs_invalidate_inode(struct inode *inode) 156 { 157 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode)); 158 nfs_zap_caches_locked(inode); 159 } 160 161 struct nfs_find_desc { 162 struct nfs_fh *fh; 163 struct nfs_fattr *fattr; 164 }; 165 166 /* 167 * In NFSv3 we can have 64bit inode numbers. In order to support 168 * this, and re-exported directories (also seen in NFSv2) 169 * we are forced to allow 2 different inodes to have the same 170 * i_ino. 171 */ 172 static int 173 nfs_find_actor(struct inode *inode, void *opaque) 174 { 175 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque; 176 struct nfs_fh *fh = desc->fh; 177 struct nfs_fattr *fattr = desc->fattr; 178 179 if (NFS_FILEID(inode) != fattr->fileid) 180 return 0; 181 if (nfs_compare_fh(NFS_FH(inode), fh)) 182 return 0; 183 if (is_bad_inode(inode) || NFS_STALE(inode)) 184 return 0; 185 return 1; 186 } 187 188 static int 189 nfs_init_locked(struct inode *inode, void *opaque) 190 { 191 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque; 192 struct nfs_fattr *fattr = desc->fattr; 193 194 NFS_FILEID(inode) = fattr->fileid; 195 nfs_copy_fh(NFS_FH(inode), desc->fh); 196 return 0; 197 } 198 199 /* Don't use READDIRPLUS on directories that we believe are too large */ 200 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE) 201 202 /* 203 * This is our front-end to iget that looks up inodes by file handle 204 * instead of inode number. 205 */ 206 struct inode * 207 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr) 208 { 209 struct nfs_find_desc desc = { 210 .fh = fh, 211 .fattr = fattr 212 }; 213 struct inode *inode = ERR_PTR(-ENOENT); 214 unsigned long hash; 215 216 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 217 goto out_no_inode; 218 219 if (!fattr->nlink) { 220 printk("NFS: Buggy server - nlink == 0!\n"); 221 goto out_no_inode; 222 } 223 224 hash = nfs_fattr_to_ino_t(fattr); 225 226 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc); 227 if (inode == NULL) { 228 inode = ERR_PTR(-ENOMEM); 229 goto out_no_inode; 230 } 231 232 if (inode->i_state & I_NEW) { 233 struct nfs_inode *nfsi = NFS_I(inode); 234 235 /* We set i_ino for the few things that still rely on it, 236 * such as stat(2) */ 237 inode->i_ino = hash; 238 239 /* We can't support update_atime(), since the server will reset it */ 240 inode->i_flags |= S_NOATIME|S_NOCMTIME; 241 inode->i_mode = fattr->mode; 242 /* Why so? Because we want revalidate for devices/FIFOs, and 243 * that's precisely what we have in nfs_file_inode_operations. 244 */ 245 inode->i_op = NFS_SB(sb)->rpc_ops->file_inode_ops; 246 if (S_ISREG(inode->i_mode)) { 247 inode->i_fop = &nfs_file_operations; 248 inode->i_data.a_ops = &nfs_file_aops; 249 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info; 250 } else if (S_ISDIR(inode->i_mode)) { 251 inode->i_op = NFS_SB(sb)->rpc_ops->dir_inode_ops; 252 inode->i_fop = &nfs_dir_operations; 253 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS) 254 && fattr->size <= NFS_LIMIT_READDIRPLUS) 255 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode)); 256 /* Deal with crossing mountpoints */ 257 if (!nfs_fsid_equal(&NFS_SB(sb)->fsid, &fattr->fsid)) { 258 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) 259 inode->i_op = &nfs_referral_inode_operations; 260 else 261 inode->i_op = &nfs_mountpoint_inode_operations; 262 inode->i_fop = NULL; 263 } 264 } else if (S_ISLNK(inode->i_mode)) 265 inode->i_op = &nfs_symlink_inode_operations; 266 else 267 init_special_inode(inode, inode->i_mode, fattr->rdev); 268 269 nfsi->read_cache_jiffies = fattr->time_start; 270 nfsi->last_updated = jiffies; 271 inode->i_atime = fattr->atime; 272 inode->i_mtime = fattr->mtime; 273 inode->i_ctime = fattr->ctime; 274 if (fattr->valid & NFS_ATTR_FATTR_V4) 275 nfsi->change_attr = fattr->change_attr; 276 inode->i_size = nfs_size_to_loff_t(fattr->size); 277 inode->i_nlink = fattr->nlink; 278 inode->i_uid = fattr->uid; 279 inode->i_gid = fattr->gid; 280 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) { 281 /* 282 * report the blocks in 512byte units 283 */ 284 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 285 inode->i_blksize = inode->i_sb->s_blocksize; 286 } else { 287 inode->i_blocks = fattr->du.nfs2.blocks; 288 inode->i_blksize = fattr->du.nfs2.blocksize; 289 } 290 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 291 nfsi->attrtimeo_timestamp = jiffies; 292 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf)); 293 nfsi->cache_access.cred = NULL; 294 295 unlock_new_inode(inode); 296 } else 297 nfs_refresh_inode(inode, fattr); 298 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n", 299 inode->i_sb->s_id, 300 (long long)NFS_FILEID(inode), 301 atomic_read(&inode->i_count)); 302 303 out: 304 return inode; 305 306 out_no_inode: 307 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode)); 308 goto out; 309 } 310 311 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET) 312 313 int 314 nfs_setattr(struct dentry *dentry, struct iattr *attr) 315 { 316 struct inode *inode = dentry->d_inode; 317 struct nfs_fattr fattr; 318 int error; 319 320 nfs_inc_stats(inode, NFSIOS_VFSSETATTR); 321 322 if (attr->ia_valid & ATTR_SIZE) { 323 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode)) 324 attr->ia_valid &= ~ATTR_SIZE; 325 } 326 327 /* Optimization: if the end result is no change, don't RPC */ 328 attr->ia_valid &= NFS_VALID_ATTRS; 329 if (attr->ia_valid == 0) 330 return 0; 331 332 lock_kernel(); 333 nfs_begin_data_update(inode); 334 /* Write all dirty data */ 335 filemap_write_and_wait(inode->i_mapping); 336 nfs_wb_all(inode); 337 /* 338 * Return any delegations if we're going to change ACLs 339 */ 340 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) 341 nfs_inode_return_delegation(inode); 342 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr); 343 if (error == 0) 344 nfs_refresh_inode(inode, &fattr); 345 nfs_end_data_update(inode); 346 unlock_kernel(); 347 return error; 348 } 349 350 /** 351 * nfs_setattr_update_inode - Update inode metadata after a setattr call. 352 * @inode: pointer to struct inode 353 * @attr: pointer to struct iattr 354 * 355 * Note: we do this in the *proc.c in order to ensure that 356 * it works for things like exclusive creates too. 357 */ 358 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr) 359 { 360 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) { 361 if ((attr->ia_valid & ATTR_MODE) != 0) { 362 int mode = attr->ia_mode & S_IALLUGO; 363 mode |= inode->i_mode & ~S_IALLUGO; 364 inode->i_mode = mode; 365 } 366 if ((attr->ia_valid & ATTR_UID) != 0) 367 inode->i_uid = attr->ia_uid; 368 if ((attr->ia_valid & ATTR_GID) != 0) 369 inode->i_gid = attr->ia_gid; 370 spin_lock(&inode->i_lock); 371 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 372 spin_unlock(&inode->i_lock); 373 } 374 if ((attr->ia_valid & ATTR_SIZE) != 0) { 375 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC); 376 inode->i_size = attr->ia_size; 377 vmtruncate(inode, attr->ia_size); 378 } 379 } 380 381 static int nfs_wait_schedule(void *word) 382 { 383 if (signal_pending(current)) 384 return -ERESTARTSYS; 385 schedule(); 386 return 0; 387 } 388 389 /* 390 * Wait for the inode to get unlocked. 391 */ 392 static int nfs_wait_on_inode(struct inode *inode) 393 { 394 struct rpc_clnt *clnt = NFS_CLIENT(inode); 395 struct nfs_inode *nfsi = NFS_I(inode); 396 sigset_t oldmask; 397 int error; 398 399 rpc_clnt_sigmask(clnt, &oldmask); 400 error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING, 401 nfs_wait_schedule, TASK_INTERRUPTIBLE); 402 rpc_clnt_sigunmask(clnt, &oldmask); 403 404 return error; 405 } 406 407 static void nfs_wake_up_inode(struct inode *inode) 408 { 409 struct nfs_inode *nfsi = NFS_I(inode); 410 411 clear_bit(NFS_INO_REVALIDATING, &nfsi->flags); 412 smp_mb__after_clear_bit(); 413 wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING); 414 } 415 416 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat) 417 { 418 struct inode *inode = dentry->d_inode; 419 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME; 420 int err; 421 422 /* Flush out writes to the server in order to update c/mtime */ 423 nfs_sync_inode_wait(inode, 0, 0, FLUSH_NOCOMMIT); 424 425 /* 426 * We may force a getattr if the user cares about atime. 427 * 428 * Note that we only have to check the vfsmount flags here: 429 * - NFS always sets S_NOATIME by so checking it would give a 430 * bogus result 431 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is 432 * no point in checking those. 433 */ 434 if ((mnt->mnt_flags & MNT_NOATIME) || 435 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) 436 need_atime = 0; 437 438 if (need_atime) 439 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 440 else 441 err = nfs_revalidate_inode(NFS_SERVER(inode), inode); 442 if (!err) 443 generic_fillattr(inode, stat); 444 return err; 445 } 446 447 static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred) 448 { 449 struct nfs_open_context *ctx; 450 451 ctx = (struct nfs_open_context *)kmalloc(sizeof(*ctx), GFP_KERNEL); 452 if (ctx != NULL) { 453 atomic_set(&ctx->count, 1); 454 ctx->dentry = dget(dentry); 455 ctx->vfsmnt = mntget(mnt); 456 ctx->cred = get_rpccred(cred); 457 ctx->state = NULL; 458 ctx->lockowner = current->files; 459 ctx->error = 0; 460 ctx->dir_cookie = 0; 461 } 462 return ctx; 463 } 464 465 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx) 466 { 467 if (ctx != NULL) 468 atomic_inc(&ctx->count); 469 return ctx; 470 } 471 472 void put_nfs_open_context(struct nfs_open_context *ctx) 473 { 474 if (atomic_dec_and_test(&ctx->count)) { 475 if (!list_empty(&ctx->list)) { 476 struct inode *inode = ctx->dentry->d_inode; 477 spin_lock(&inode->i_lock); 478 list_del(&ctx->list); 479 spin_unlock(&inode->i_lock); 480 } 481 if (ctx->state != NULL) 482 nfs4_close_state(ctx->state, ctx->mode); 483 if (ctx->cred != NULL) 484 put_rpccred(ctx->cred); 485 dput(ctx->dentry); 486 mntput(ctx->vfsmnt); 487 kfree(ctx); 488 } 489 } 490 491 /* 492 * Ensure that mmap has a recent RPC credential for use when writing out 493 * shared pages 494 */ 495 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx) 496 { 497 struct inode *inode = filp->f_dentry->d_inode; 498 struct nfs_inode *nfsi = NFS_I(inode); 499 500 filp->private_data = get_nfs_open_context(ctx); 501 spin_lock(&inode->i_lock); 502 list_add(&ctx->list, &nfsi->open_files); 503 spin_unlock(&inode->i_lock); 504 } 505 506 /* 507 * Given an inode, search for an open context with the desired characteristics 508 */ 509 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode) 510 { 511 struct nfs_inode *nfsi = NFS_I(inode); 512 struct nfs_open_context *pos, *ctx = NULL; 513 514 spin_lock(&inode->i_lock); 515 list_for_each_entry(pos, &nfsi->open_files, list) { 516 if (cred != NULL && pos->cred != cred) 517 continue; 518 if ((pos->mode & mode) == mode) { 519 ctx = get_nfs_open_context(pos); 520 break; 521 } 522 } 523 spin_unlock(&inode->i_lock); 524 return ctx; 525 } 526 527 static void nfs_file_clear_open_context(struct file *filp) 528 { 529 struct inode *inode = filp->f_dentry->d_inode; 530 struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data; 531 532 if (ctx) { 533 filp->private_data = NULL; 534 spin_lock(&inode->i_lock); 535 list_move_tail(&ctx->list, &NFS_I(inode)->open_files); 536 spin_unlock(&inode->i_lock); 537 put_nfs_open_context(ctx); 538 } 539 } 540 541 /* 542 * These allocate and release file read/write context information. 543 */ 544 int nfs_open(struct inode *inode, struct file *filp) 545 { 546 struct nfs_open_context *ctx; 547 struct rpc_cred *cred; 548 549 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0); 550 if (IS_ERR(cred)) 551 return PTR_ERR(cred); 552 ctx = alloc_nfs_open_context(filp->f_vfsmnt, filp->f_dentry, cred); 553 put_rpccred(cred); 554 if (ctx == NULL) 555 return -ENOMEM; 556 ctx->mode = filp->f_mode; 557 nfs_file_set_open_context(filp, ctx); 558 put_nfs_open_context(ctx); 559 return 0; 560 } 561 562 int nfs_release(struct inode *inode, struct file *filp) 563 { 564 nfs_file_clear_open_context(filp); 565 return 0; 566 } 567 568 /* 569 * This function is called whenever some part of NFS notices that 570 * the cached attributes have to be refreshed. 571 */ 572 int 573 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) 574 { 575 int status = -ESTALE; 576 struct nfs_fattr fattr; 577 struct nfs_inode *nfsi = NFS_I(inode); 578 579 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n", 580 inode->i_sb->s_id, (long long)NFS_FILEID(inode)); 581 582 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE); 583 lock_kernel(); 584 if (!inode || is_bad_inode(inode)) 585 goto out_nowait; 586 if (NFS_STALE(inode)) 587 goto out_nowait; 588 589 status = nfs_wait_on_inode(inode); 590 if (status < 0) 591 goto out; 592 if (NFS_STALE(inode)) { 593 status = -ESTALE; 594 /* Do we trust the cached ESTALE? */ 595 if (NFS_ATTRTIMEO(inode) != 0) { 596 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME)) { 597 /* no */ 598 } else 599 goto out; 600 } 601 } 602 603 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr); 604 if (status != 0) { 605 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n", 606 inode->i_sb->s_id, 607 (long long)NFS_FILEID(inode), status); 608 if (status == -ESTALE) { 609 nfs_zap_caches(inode); 610 if (!S_ISDIR(inode->i_mode)) 611 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode)); 612 } 613 goto out; 614 } 615 616 spin_lock(&inode->i_lock); 617 status = nfs_update_inode(inode, &fattr); 618 if (status) { 619 spin_unlock(&inode->i_lock); 620 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n", 621 inode->i_sb->s_id, 622 (long long)NFS_FILEID(inode), status); 623 goto out; 624 } 625 spin_unlock(&inode->i_lock); 626 627 if (nfsi->cache_validity & NFS_INO_INVALID_ACL) 628 nfs_zap_acl_cache(inode); 629 630 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n", 631 inode->i_sb->s_id, 632 (long long)NFS_FILEID(inode)); 633 634 out: 635 nfs_wake_up_inode(inode); 636 637 out_nowait: 638 unlock_kernel(); 639 return status; 640 } 641 642 int nfs_attribute_timeout(struct inode *inode) 643 { 644 struct nfs_inode *nfsi = NFS_I(inode); 645 646 if (nfs_have_delegation(inode, FMODE_READ)) 647 return 0; 648 return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo); 649 } 650 651 /** 652 * nfs_revalidate_inode - Revalidate the inode attributes 653 * @server - pointer to nfs_server struct 654 * @inode - pointer to inode struct 655 * 656 * Updates inode attribute information by retrieving the data from the server. 657 */ 658 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) 659 { 660 if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR) 661 && !nfs_attribute_timeout(inode)) 662 return NFS_STALE(inode) ? -ESTALE : 0; 663 return __nfs_revalidate_inode(server, inode); 664 } 665 666 /** 667 * nfs_revalidate_mapping - Revalidate the pagecache 668 * @inode - pointer to host inode 669 * @mapping - pointer to mapping 670 */ 671 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping) 672 { 673 struct nfs_inode *nfsi = NFS_I(inode); 674 int ret = 0; 675 676 if (NFS_STALE(inode)) 677 ret = -ESTALE; 678 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) 679 || nfs_attribute_timeout(inode)) 680 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 681 682 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) { 683 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE); 684 if (S_ISREG(inode->i_mode)) 685 nfs_sync_mapping(mapping); 686 invalidate_inode_pages2(mapping); 687 688 spin_lock(&inode->i_lock); 689 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA; 690 if (S_ISDIR(inode->i_mode)) { 691 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf)); 692 /* This ensures we revalidate child dentries */ 693 nfsi->cache_change_attribute = jiffies; 694 } 695 spin_unlock(&inode->i_lock); 696 697 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n", 698 inode->i_sb->s_id, 699 (long long)NFS_FILEID(inode)); 700 } 701 return ret; 702 } 703 704 /** 705 * nfs_begin_data_update 706 * @inode - pointer to inode 707 * Declare that a set of operations will update file data on the server 708 */ 709 void nfs_begin_data_update(struct inode *inode) 710 { 711 atomic_inc(&NFS_I(inode)->data_updates); 712 } 713 714 /** 715 * nfs_end_data_update 716 * @inode - pointer to inode 717 * Declare end of the operations that will update file data 718 * This will mark the inode as immediately needing revalidation 719 * of its attribute cache. 720 */ 721 void nfs_end_data_update(struct inode *inode) 722 { 723 struct nfs_inode *nfsi = NFS_I(inode); 724 725 if (!nfs_have_delegation(inode, FMODE_READ)) { 726 /* Directories and symlinks: invalidate page cache */ 727 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) { 728 spin_lock(&inode->i_lock); 729 nfsi->cache_validity |= NFS_INO_INVALID_DATA; 730 spin_unlock(&inode->i_lock); 731 } 732 } 733 nfsi->cache_change_attribute = jiffies; 734 atomic_dec(&nfsi->data_updates); 735 } 736 737 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr) 738 { 739 struct nfs_inode *nfsi = NFS_I(inode); 740 741 /* If we have atomic WCC data, we may update some attributes */ 742 if ((fattr->valid & NFS_ATTR_WCC) != 0) { 743 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) { 744 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime)); 745 nfsi->cache_change_attribute = jiffies; 746 } 747 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) { 748 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime)); 749 nfsi->cache_change_attribute = jiffies; 750 } 751 if (inode->i_size == fattr->pre_size && nfsi->npages == 0) { 752 inode->i_size = fattr->size; 753 nfsi->cache_change_attribute = jiffies; 754 } 755 } 756 } 757 758 /** 759 * nfs_check_inode_attributes - verify consistency of the inode attribute cache 760 * @inode - pointer to inode 761 * @fattr - updated attributes 762 * 763 * Verifies the attribute cache. If we have just changed the attributes, 764 * so that fattr carries weak cache consistency data, then it may 765 * also update the ctime/mtime/change_attribute. 766 */ 767 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr) 768 { 769 struct nfs_inode *nfsi = NFS_I(inode); 770 loff_t cur_size, new_isize; 771 int data_unstable; 772 773 774 /* Has the inode gone and changed behind our back? */ 775 if (nfsi->fileid != fattr->fileid 776 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) { 777 return -EIO; 778 } 779 780 /* Are we in the process of updating data on the server? */ 781 data_unstable = nfs_caches_unstable(inode); 782 783 /* Do atomic weak cache consistency updates */ 784 nfs_wcc_update_inode(inode, fattr); 785 786 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 && 787 nfsi->change_attr != fattr->change_attr) 788 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE; 789 790 /* Verify a few of the more important attributes */ 791 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) 792 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE; 793 794 cur_size = i_size_read(inode); 795 new_isize = nfs_size_to_loff_t(fattr->size); 796 if (cur_size != new_isize && nfsi->npages == 0) 797 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE; 798 799 /* Have any file permissions changed? */ 800 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) 801 || inode->i_uid != fattr->uid 802 || inode->i_gid != fattr->gid) 803 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL; 804 805 /* Has the link count changed? */ 806 if (inode->i_nlink != fattr->nlink) 807 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 808 809 if (!timespec_equal(&inode->i_atime, &fattr->atime)) 810 nfsi->cache_validity |= NFS_INO_INVALID_ATIME; 811 812 nfsi->read_cache_jiffies = fattr->time_start; 813 return 0; 814 } 815 816 /** 817 * nfs_refresh_inode - try to update the inode attribute cache 818 * @inode - pointer to inode 819 * @fattr - updated attributes 820 * 821 * Check that an RPC call that returned attributes has not overlapped with 822 * other recent updates of the inode metadata, then decide whether it is 823 * safe to do a full update of the inode attributes, or whether just to 824 * call nfs_check_inode_attributes. 825 */ 826 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr) 827 { 828 struct nfs_inode *nfsi = NFS_I(inode); 829 int status; 830 831 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 832 return 0; 833 spin_lock(&inode->i_lock); 834 if (time_after(fattr->time_start, nfsi->last_updated)) 835 status = nfs_update_inode(inode, fattr); 836 else 837 status = nfs_check_inode_attributes(inode, fattr); 838 839 spin_unlock(&inode->i_lock); 840 return status; 841 } 842 843 /** 844 * nfs_post_op_update_inode - try to update the inode attribute cache 845 * @inode - pointer to inode 846 * @fattr - updated attributes 847 * 848 * After an operation that has changed the inode metadata, mark the 849 * attribute cache as being invalid, then try to update it. 850 */ 851 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr) 852 { 853 struct nfs_inode *nfsi = NFS_I(inode); 854 int status = 0; 855 856 spin_lock(&inode->i_lock); 857 if (unlikely((fattr->valid & NFS_ATTR_FATTR) == 0)) { 858 nfsi->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE; 859 goto out; 860 } 861 status = nfs_update_inode(inode, fattr); 862 out: 863 spin_unlock(&inode->i_lock); 864 return status; 865 } 866 867 /* 868 * Many nfs protocol calls return the new file attributes after 869 * an operation. Here we update the inode to reflect the state 870 * of the server's inode. 871 * 872 * This is a bit tricky because we have to make sure all dirty pages 873 * have been sent off to the server before calling invalidate_inode_pages. 874 * To make sure no other process adds more write requests while we try 875 * our best to flush them, we make them sleep during the attribute refresh. 876 * 877 * A very similar scenario holds for the dir cache. 878 */ 879 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr) 880 { 881 struct nfs_server *server; 882 struct nfs_inode *nfsi = NFS_I(inode); 883 loff_t cur_isize, new_isize; 884 unsigned int invalid = 0; 885 int data_stable; 886 887 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n", 888 __FUNCTION__, inode->i_sb->s_id, inode->i_ino, 889 atomic_read(&inode->i_count), fattr->valid); 890 891 if (nfsi->fileid != fattr->fileid) 892 goto out_fileid; 893 894 /* 895 * Make sure the inode's type hasn't changed. 896 */ 897 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) 898 goto out_changed; 899 900 server = NFS_SERVER(inode); 901 /* Update the fsid if and only if this is the root directory */ 902 if (inode == inode->i_sb->s_root->d_inode 903 && !nfs_fsid_equal(&server->fsid, &fattr->fsid)) 904 server->fsid = fattr->fsid; 905 906 /* 907 * Update the read time so we don't revalidate too often. 908 */ 909 nfsi->read_cache_jiffies = fattr->time_start; 910 nfsi->last_updated = jiffies; 911 912 /* Are we racing with known updates of the metadata on the server? */ 913 data_stable = nfs_verify_change_attribute(inode, fattr->time_start); 914 if (data_stable) 915 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_ATIME); 916 917 /* Do atomic weak cache consistency updates */ 918 nfs_wcc_update_inode(inode, fattr); 919 920 /* Check if our cached file size is stale */ 921 new_isize = nfs_size_to_loff_t(fattr->size); 922 cur_isize = i_size_read(inode); 923 if (new_isize != cur_isize) { 924 /* Do we perhaps have any outstanding writes? */ 925 if (nfsi->npages == 0) { 926 /* No, but did we race with nfs_end_data_update()? */ 927 if (data_stable) { 928 inode->i_size = new_isize; 929 invalid |= NFS_INO_INVALID_DATA; 930 } 931 invalid |= NFS_INO_INVALID_ATTR; 932 } else if (new_isize > cur_isize) { 933 inode->i_size = new_isize; 934 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA; 935 } 936 nfsi->cache_change_attribute = jiffies; 937 dprintk("NFS: isize change on server for file %s/%ld\n", 938 inode->i_sb->s_id, inode->i_ino); 939 } 940 941 /* Check if the mtime agrees */ 942 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) { 943 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime)); 944 dprintk("NFS: mtime change on server for file %s/%ld\n", 945 inode->i_sb->s_id, inode->i_ino); 946 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA; 947 nfsi->cache_change_attribute = jiffies; 948 } 949 950 /* If ctime has changed we should definitely clear access+acl caches */ 951 if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) { 952 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 953 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime)); 954 nfsi->cache_change_attribute = jiffies; 955 } 956 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime)); 957 958 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) || 959 inode->i_uid != fattr->uid || 960 inode->i_gid != fattr->gid) 961 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 962 963 inode->i_mode = fattr->mode; 964 inode->i_nlink = fattr->nlink; 965 inode->i_uid = fattr->uid; 966 inode->i_gid = fattr->gid; 967 968 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) { 969 /* 970 * report the blocks in 512byte units 971 */ 972 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 973 inode->i_blksize = inode->i_sb->s_blocksize; 974 } else { 975 inode->i_blocks = fattr->du.nfs2.blocks; 976 inode->i_blksize = fattr->du.nfs2.blocksize; 977 } 978 979 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 && 980 nfsi->change_attr != fattr->change_attr) { 981 dprintk("NFS: change_attr change on server for file %s/%ld\n", 982 inode->i_sb->s_id, inode->i_ino); 983 nfsi->change_attr = fattr->change_attr; 984 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 985 nfsi->cache_change_attribute = jiffies; 986 } 987 988 /* Update attrtimeo value if we're out of the unstable period */ 989 if (invalid & NFS_INO_INVALID_ATTR) { 990 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 991 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 992 nfsi->attrtimeo_timestamp = jiffies; 993 } else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) { 994 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode)) 995 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode); 996 nfsi->attrtimeo_timestamp = jiffies; 997 } 998 /* Don't invalidate the data if we were to blame */ 999 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) 1000 || S_ISLNK(inode->i_mode))) 1001 invalid &= ~NFS_INO_INVALID_DATA; 1002 if (data_stable) 1003 invalid &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME|NFS_INO_REVAL_PAGECACHE); 1004 if (!nfs_have_delegation(inode, FMODE_READ)) 1005 nfsi->cache_validity |= invalid; 1006 1007 return 0; 1008 out_changed: 1009 /* 1010 * Big trouble! The inode has become a different object. 1011 */ 1012 #ifdef NFS_PARANOIA 1013 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n", 1014 __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode); 1015 #endif 1016 out_err: 1017 /* 1018 * No need to worry about unhashing the dentry, as the 1019 * lookup validation will know that the inode is bad. 1020 * (But we fall through to invalidate the caches.) 1021 */ 1022 nfs_invalidate_inode(inode); 1023 return -ESTALE; 1024 1025 out_fileid: 1026 printk(KERN_ERR "NFS: server %s error: fileid changed\n" 1027 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n", 1028 NFS_SERVER(inode)->hostname, inode->i_sb->s_id, 1029 (long long)nfsi->fileid, (long long)fattr->fileid); 1030 goto out_err; 1031 } 1032 1033 1034 #ifdef CONFIG_NFS_V4 1035 1036 /* 1037 * Clean out any remaining NFSv4 state that might be left over due 1038 * to open() calls that passed nfs_atomic_lookup, but failed to call 1039 * nfs_open(). 1040 */ 1041 void nfs4_clear_inode(struct inode *inode) 1042 { 1043 struct nfs_inode *nfsi = NFS_I(inode); 1044 1045 /* If we are holding a delegation, return it! */ 1046 nfs_inode_return_delegation(inode); 1047 /* First call standard NFS clear_inode() code */ 1048 nfs_clear_inode(inode); 1049 /* Now clear out any remaining state */ 1050 while (!list_empty(&nfsi->open_states)) { 1051 struct nfs4_state *state; 1052 1053 state = list_entry(nfsi->open_states.next, 1054 struct nfs4_state, 1055 inode_states); 1056 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n", 1057 __FUNCTION__, 1058 inode->i_sb->s_id, 1059 (long long)NFS_FILEID(inode), 1060 state); 1061 BUG_ON(atomic_read(&state->count) != 1); 1062 nfs4_close_state(state, state->state); 1063 } 1064 } 1065 #endif 1066 1067 struct inode *nfs_alloc_inode(struct super_block *sb) 1068 { 1069 struct nfs_inode *nfsi; 1070 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL); 1071 if (!nfsi) 1072 return NULL; 1073 nfsi->flags = 0UL; 1074 nfsi->cache_validity = 0UL; 1075 nfsi->cache_change_attribute = jiffies; 1076 #ifdef CONFIG_NFS_V3_ACL 1077 nfsi->acl_access = ERR_PTR(-EAGAIN); 1078 nfsi->acl_default = ERR_PTR(-EAGAIN); 1079 #endif 1080 #ifdef CONFIG_NFS_V4 1081 nfsi->nfs4_acl = NULL; 1082 #endif /* CONFIG_NFS_V4 */ 1083 return &nfsi->vfs_inode; 1084 } 1085 1086 void nfs_destroy_inode(struct inode *inode) 1087 { 1088 kmem_cache_free(nfs_inode_cachep, NFS_I(inode)); 1089 } 1090 1091 static inline void nfs4_init_once(struct nfs_inode *nfsi) 1092 { 1093 #ifdef CONFIG_NFS_V4 1094 INIT_LIST_HEAD(&nfsi->open_states); 1095 nfsi->delegation = NULL; 1096 nfsi->delegation_state = 0; 1097 init_rwsem(&nfsi->rwsem); 1098 #endif 1099 } 1100 1101 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags) 1102 { 1103 struct nfs_inode *nfsi = (struct nfs_inode *) foo; 1104 1105 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) == 1106 SLAB_CTOR_CONSTRUCTOR) { 1107 inode_init_once(&nfsi->vfs_inode); 1108 spin_lock_init(&nfsi->req_lock); 1109 INIT_LIST_HEAD(&nfsi->dirty); 1110 INIT_LIST_HEAD(&nfsi->commit); 1111 INIT_LIST_HEAD(&nfsi->open_files); 1112 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC); 1113 atomic_set(&nfsi->data_updates, 0); 1114 nfsi->ndirty = 0; 1115 nfsi->ncommit = 0; 1116 nfsi->npages = 0; 1117 nfs4_init_once(nfsi); 1118 } 1119 } 1120 1121 static int __init nfs_init_inodecache(void) 1122 { 1123 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache", 1124 sizeof(struct nfs_inode), 1125 0, (SLAB_RECLAIM_ACCOUNT| 1126 SLAB_MEM_SPREAD), 1127 init_once, NULL); 1128 if (nfs_inode_cachep == NULL) 1129 return -ENOMEM; 1130 1131 return 0; 1132 } 1133 1134 static void nfs_destroy_inodecache(void) 1135 { 1136 if (kmem_cache_destroy(nfs_inode_cachep)) 1137 printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n"); 1138 } 1139 1140 /* 1141 * Initialize NFS 1142 */ 1143 static int __init init_nfs_fs(void) 1144 { 1145 int err; 1146 1147 err = nfs_init_nfspagecache(); 1148 if (err) 1149 goto out4; 1150 1151 err = nfs_init_inodecache(); 1152 if (err) 1153 goto out3; 1154 1155 err = nfs_init_readpagecache(); 1156 if (err) 1157 goto out2; 1158 1159 err = nfs_init_writepagecache(); 1160 if (err) 1161 goto out1; 1162 1163 err = nfs_init_directcache(); 1164 if (err) 1165 goto out0; 1166 1167 #ifdef CONFIG_PROC_FS 1168 rpc_proc_register(&nfs_rpcstat); 1169 #endif 1170 if ((err = register_nfs_fs()) != 0) 1171 goto out; 1172 return 0; 1173 out: 1174 #ifdef CONFIG_PROC_FS 1175 rpc_proc_unregister("nfs"); 1176 #endif 1177 nfs_destroy_directcache(); 1178 out0: 1179 nfs_destroy_writepagecache(); 1180 out1: 1181 nfs_destroy_readpagecache(); 1182 out2: 1183 nfs_destroy_inodecache(); 1184 out3: 1185 nfs_destroy_nfspagecache(); 1186 out4: 1187 return err; 1188 } 1189 1190 static void __exit exit_nfs_fs(void) 1191 { 1192 nfs_destroy_directcache(); 1193 nfs_destroy_writepagecache(); 1194 nfs_destroy_readpagecache(); 1195 nfs_destroy_inodecache(); 1196 nfs_destroy_nfspagecache(); 1197 #ifdef CONFIG_PROC_FS 1198 rpc_proc_unregister("nfs"); 1199 #endif 1200 unregister_nfs_fs(); 1201 } 1202 1203 /* Not quite true; I just maintain it */ 1204 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>"); 1205 MODULE_LICENSE("GPL"); 1206 1207 module_init(init_nfs_fs) 1208 module_exit(exit_nfs_fs) 1209