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/config.h> 17 #include <linux/module.h> 18 #include <linux/init.h> 19 20 #include <linux/time.h> 21 #include <linux/kernel.h> 22 #include <linux/mm.h> 23 #include <linux/string.h> 24 #include <linux/stat.h> 25 #include <linux/errno.h> 26 #include <linux/unistd.h> 27 #include <linux/sunrpc/clnt.h> 28 #include <linux/sunrpc/stats.h> 29 #include <linux/sunrpc/metrics.h> 30 #include <linux/nfs_fs.h> 31 #include <linux/nfs_mount.h> 32 #include <linux/nfs4_mount.h> 33 #include <linux/lockd/bind.h> 34 #include <linux/smp_lock.h> 35 #include <linux/seq_file.h> 36 #include <linux/mount.h> 37 #include <linux/nfs_idmap.h> 38 #include <linux/vfs.h> 39 40 #include <asm/system.h> 41 #include <asm/uaccess.h> 42 43 #include "nfs4_fs.h" 44 #include "callback.h" 45 #include "delegation.h" 46 #include "iostat.h" 47 48 #define NFSDBG_FACILITY NFSDBG_VFS 49 #define NFS_PARANOIA 1 50 51 /* Maximum number of readahead requests 52 * FIXME: this should really be a sysctl so that users may tune it to suit 53 * their needs. People that do NFS over a slow network, might for 54 * instance want to reduce it to something closer to 1 for improved 55 * interactive response. 56 */ 57 #define NFS_MAX_READAHEAD (RPC_DEF_SLOT_TABLE - 1) 58 59 static void nfs_invalidate_inode(struct inode *); 60 static int nfs_update_inode(struct inode *, struct nfs_fattr *); 61 62 static struct inode *nfs_alloc_inode(struct super_block *sb); 63 static void nfs_destroy_inode(struct inode *); 64 static int nfs_write_inode(struct inode *,int); 65 static void nfs_delete_inode(struct inode *); 66 static void nfs_clear_inode(struct inode *); 67 static void nfs_umount_begin(struct super_block *); 68 static int nfs_statfs(struct super_block *, struct kstatfs *); 69 static int nfs_show_options(struct seq_file *, struct vfsmount *); 70 static int nfs_show_stats(struct seq_file *, struct vfsmount *); 71 static void nfs_zap_acl_cache(struct inode *); 72 73 static struct rpc_program nfs_program; 74 75 static struct super_operations nfs_sops = { 76 .alloc_inode = nfs_alloc_inode, 77 .destroy_inode = nfs_destroy_inode, 78 .write_inode = nfs_write_inode, 79 .delete_inode = nfs_delete_inode, 80 .statfs = nfs_statfs, 81 .clear_inode = nfs_clear_inode, 82 .umount_begin = nfs_umount_begin, 83 .show_options = nfs_show_options, 84 .show_stats = nfs_show_stats, 85 }; 86 87 /* 88 * RPC cruft for NFS 89 */ 90 static struct rpc_stat nfs_rpcstat = { 91 .program = &nfs_program 92 }; 93 static struct rpc_version * nfs_version[] = { 94 NULL, 95 NULL, 96 &nfs_version2, 97 #if defined(CONFIG_NFS_V3) 98 &nfs_version3, 99 #elif defined(CONFIG_NFS_V4) 100 NULL, 101 #endif 102 #if defined(CONFIG_NFS_V4) 103 &nfs_version4, 104 #endif 105 }; 106 107 static struct rpc_program nfs_program = { 108 .name = "nfs", 109 .number = NFS_PROGRAM, 110 .nrvers = ARRAY_SIZE(nfs_version), 111 .version = nfs_version, 112 .stats = &nfs_rpcstat, 113 .pipe_dir_name = "/nfs", 114 }; 115 116 #ifdef CONFIG_NFS_V3_ACL 117 static struct rpc_stat nfsacl_rpcstat = { &nfsacl_program }; 118 static struct rpc_version * nfsacl_version[] = { 119 [3] = &nfsacl_version3, 120 }; 121 122 struct rpc_program nfsacl_program = { 123 .name = "nfsacl", 124 .number = NFS_ACL_PROGRAM, 125 .nrvers = ARRAY_SIZE(nfsacl_version), 126 .version = nfsacl_version, 127 .stats = &nfsacl_rpcstat, 128 }; 129 #endif /* CONFIG_NFS_V3_ACL */ 130 131 static inline unsigned long 132 nfs_fattr_to_ino_t(struct nfs_fattr *fattr) 133 { 134 return nfs_fileid_to_ino_t(fattr->fileid); 135 } 136 137 static int 138 nfs_write_inode(struct inode *inode, int sync) 139 { 140 int flags = sync ? FLUSH_SYNC : 0; 141 int ret; 142 143 ret = nfs_commit_inode(inode, flags); 144 if (ret < 0) 145 return ret; 146 return 0; 147 } 148 149 static void 150 nfs_delete_inode(struct inode * inode) 151 { 152 dprintk("NFS: delete_inode(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino); 153 154 truncate_inode_pages(&inode->i_data, 0); 155 156 nfs_wb_all(inode); 157 /* 158 * The following should never happen... 159 */ 160 if (nfs_have_writebacks(inode)) { 161 printk(KERN_ERR "nfs_delete_inode: inode %ld has pending RPC requests\n", inode->i_ino); 162 } 163 164 clear_inode(inode); 165 } 166 167 static void 168 nfs_clear_inode(struct inode *inode) 169 { 170 struct nfs_inode *nfsi = NFS_I(inode); 171 struct rpc_cred *cred; 172 173 nfs_wb_all(inode); 174 BUG_ON (!list_empty(&nfsi->open_files)); 175 nfs_zap_acl_cache(inode); 176 cred = nfsi->cache_access.cred; 177 if (cred) 178 put_rpccred(cred); 179 BUG_ON(atomic_read(&nfsi->data_updates) != 0); 180 } 181 182 void 183 nfs_umount_begin(struct super_block *sb) 184 { 185 struct rpc_clnt *rpc = NFS_SB(sb)->client; 186 187 /* -EIO all pending I/O */ 188 if (!IS_ERR(rpc)) 189 rpc_killall_tasks(rpc); 190 rpc = NFS_SB(sb)->client_acl; 191 if (!IS_ERR(rpc)) 192 rpc_killall_tasks(rpc); 193 } 194 195 196 static inline unsigned long 197 nfs_block_bits(unsigned long bsize, unsigned char *nrbitsp) 198 { 199 /* make sure blocksize is a power of two */ 200 if ((bsize & (bsize - 1)) || nrbitsp) { 201 unsigned char nrbits; 202 203 for (nrbits = 31; nrbits && !(bsize & (1 << nrbits)); nrbits--) 204 ; 205 bsize = 1 << nrbits; 206 if (nrbitsp) 207 *nrbitsp = nrbits; 208 } 209 210 return bsize; 211 } 212 213 /* 214 * Calculate the number of 512byte blocks used. 215 */ 216 static inline unsigned long 217 nfs_calc_block_size(u64 tsize) 218 { 219 loff_t used = (tsize + 511) >> 9; 220 return (used > ULONG_MAX) ? ULONG_MAX : used; 221 } 222 223 /* 224 * Compute and set NFS server blocksize 225 */ 226 static inline unsigned long 227 nfs_block_size(unsigned long bsize, unsigned char *nrbitsp) 228 { 229 if (bsize < NFS_MIN_FILE_IO_SIZE) 230 bsize = NFS_DEF_FILE_IO_SIZE; 231 else if (bsize >= NFS_MAX_FILE_IO_SIZE) 232 bsize = NFS_MAX_FILE_IO_SIZE; 233 234 return nfs_block_bits(bsize, nrbitsp); 235 } 236 237 /* 238 * Obtain the root inode of the file system. 239 */ 240 static struct inode * 241 nfs_get_root(struct super_block *sb, struct nfs_fh *rootfh, struct nfs_fsinfo *fsinfo) 242 { 243 struct nfs_server *server = NFS_SB(sb); 244 int error; 245 246 error = server->rpc_ops->getroot(server, rootfh, fsinfo); 247 if (error < 0) { 248 dprintk("nfs_get_root: getattr error = %d\n", -error); 249 return ERR_PTR(error); 250 } 251 252 return nfs_fhget(sb, rootfh, fsinfo->fattr); 253 } 254 255 /* 256 * Do NFS version-independent mount processing, and sanity checking 257 */ 258 static int 259 nfs_sb_init(struct super_block *sb, rpc_authflavor_t authflavor) 260 { 261 struct nfs_server *server; 262 struct inode *root_inode; 263 struct nfs_fattr fattr; 264 struct nfs_fsinfo fsinfo = { 265 .fattr = &fattr, 266 }; 267 struct nfs_pathconf pathinfo = { 268 .fattr = &fattr, 269 }; 270 int no_root_error = 0; 271 unsigned long max_rpc_payload; 272 273 /* We probably want something more informative here */ 274 snprintf(sb->s_id, sizeof(sb->s_id), "%x:%x", MAJOR(sb->s_dev), MINOR(sb->s_dev)); 275 276 server = NFS_SB(sb); 277 278 sb->s_magic = NFS_SUPER_MAGIC; 279 280 server->io_stats = nfs_alloc_iostats(); 281 if (server->io_stats == NULL) 282 return -ENOMEM; 283 284 root_inode = nfs_get_root(sb, &server->fh, &fsinfo); 285 /* Did getting the root inode fail? */ 286 if (IS_ERR(root_inode)) { 287 no_root_error = PTR_ERR(root_inode); 288 goto out_no_root; 289 } 290 sb->s_root = d_alloc_root(root_inode); 291 if (!sb->s_root) { 292 no_root_error = -ENOMEM; 293 goto out_no_root; 294 } 295 sb->s_root->d_op = server->rpc_ops->dentry_ops; 296 297 /* mount time stamp, in seconds */ 298 server->mount_time = jiffies; 299 300 /* Get some general file system info */ 301 if (server->namelen == 0 && 302 server->rpc_ops->pathconf(server, &server->fh, &pathinfo) >= 0) 303 server->namelen = pathinfo.max_namelen; 304 /* Work out a lot of parameters */ 305 if (server->rsize == 0) 306 server->rsize = nfs_block_size(fsinfo.rtpref, NULL); 307 if (server->wsize == 0) 308 server->wsize = nfs_block_size(fsinfo.wtpref, NULL); 309 310 if (fsinfo.rtmax >= 512 && server->rsize > fsinfo.rtmax) 311 server->rsize = nfs_block_size(fsinfo.rtmax, NULL); 312 if (fsinfo.wtmax >= 512 && server->wsize > fsinfo.wtmax) 313 server->wsize = nfs_block_size(fsinfo.wtmax, NULL); 314 315 max_rpc_payload = nfs_block_size(rpc_max_payload(server->client), NULL); 316 if (server->rsize > max_rpc_payload) 317 server->rsize = max_rpc_payload; 318 if (server->rsize > NFS_MAX_FILE_IO_SIZE) 319 server->rsize = NFS_MAX_FILE_IO_SIZE; 320 server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 321 322 if (server->wsize > max_rpc_payload) 323 server->wsize = max_rpc_payload; 324 if (server->wsize > NFS_MAX_FILE_IO_SIZE) 325 server->wsize = NFS_MAX_FILE_IO_SIZE; 326 server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 327 328 if (sb->s_blocksize == 0) 329 sb->s_blocksize = nfs_block_bits(server->wsize, 330 &sb->s_blocksize_bits); 331 server->wtmult = nfs_block_bits(fsinfo.wtmult, NULL); 332 333 server->dtsize = nfs_block_size(fsinfo.dtpref, NULL); 334 if (server->dtsize > PAGE_CACHE_SIZE) 335 server->dtsize = PAGE_CACHE_SIZE; 336 if (server->dtsize > server->rsize) 337 server->dtsize = server->rsize; 338 339 if (server->flags & NFS_MOUNT_NOAC) { 340 server->acregmin = server->acregmax = 0; 341 server->acdirmin = server->acdirmax = 0; 342 sb->s_flags |= MS_SYNCHRONOUS; 343 } 344 server->backing_dev_info.ra_pages = server->rpages * NFS_MAX_READAHEAD; 345 346 sb->s_maxbytes = fsinfo.maxfilesize; 347 if (sb->s_maxbytes > MAX_LFS_FILESIZE) 348 sb->s_maxbytes = MAX_LFS_FILESIZE; 349 350 server->client->cl_intr = (server->flags & NFS_MOUNT_INTR) ? 1 : 0; 351 server->client->cl_softrtry = (server->flags & NFS_MOUNT_SOFT) ? 1 : 0; 352 353 /* We're airborne Set socket buffersize */ 354 rpc_setbufsize(server->client, server->wsize + 100, server->rsize + 100); 355 return 0; 356 /* Yargs. It didn't work out. */ 357 out_no_root: 358 dprintk("nfs_sb_init: get root inode failed: errno %d\n", -no_root_error); 359 if (!IS_ERR(root_inode)) 360 iput(root_inode); 361 return no_root_error; 362 } 363 364 static void nfs_init_timeout_values(struct rpc_timeout *to, int proto, unsigned int timeo, unsigned int retrans) 365 { 366 to->to_initval = timeo * HZ / 10; 367 to->to_retries = retrans; 368 if (!to->to_retries) 369 to->to_retries = 2; 370 371 switch (proto) { 372 case IPPROTO_TCP: 373 if (!to->to_initval) 374 to->to_initval = 60 * HZ; 375 if (to->to_initval > NFS_MAX_TCP_TIMEOUT) 376 to->to_initval = NFS_MAX_TCP_TIMEOUT; 377 to->to_increment = to->to_initval; 378 to->to_maxval = to->to_initval + (to->to_increment * to->to_retries); 379 to->to_exponential = 0; 380 break; 381 case IPPROTO_UDP: 382 default: 383 if (!to->to_initval) 384 to->to_initval = 11 * HZ / 10; 385 if (to->to_initval > NFS_MAX_UDP_TIMEOUT) 386 to->to_initval = NFS_MAX_UDP_TIMEOUT; 387 to->to_maxval = NFS_MAX_UDP_TIMEOUT; 388 to->to_exponential = 1; 389 break; 390 } 391 } 392 393 /* 394 * Create an RPC client handle. 395 */ 396 static struct rpc_clnt * 397 nfs_create_client(struct nfs_server *server, const struct nfs_mount_data *data) 398 { 399 struct rpc_timeout timeparms; 400 struct rpc_xprt *xprt = NULL; 401 struct rpc_clnt *clnt = NULL; 402 int proto = (data->flags & NFS_MOUNT_TCP) ? IPPROTO_TCP : IPPROTO_UDP; 403 404 nfs_init_timeout_values(&timeparms, proto, data->timeo, data->retrans); 405 406 server->retrans_timeo = timeparms.to_initval; 407 server->retrans_count = timeparms.to_retries; 408 409 /* create transport and client */ 410 xprt = xprt_create_proto(proto, &server->addr, &timeparms); 411 if (IS_ERR(xprt)) { 412 dprintk("%s: cannot create RPC transport. Error = %ld\n", 413 __FUNCTION__, PTR_ERR(xprt)); 414 return (struct rpc_clnt *)xprt; 415 } 416 clnt = rpc_create_client(xprt, server->hostname, &nfs_program, 417 server->rpc_ops->version, data->pseudoflavor); 418 if (IS_ERR(clnt)) { 419 dprintk("%s: cannot create RPC client. Error = %ld\n", 420 __FUNCTION__, PTR_ERR(xprt)); 421 goto out_fail; 422 } 423 424 clnt->cl_intr = 1; 425 clnt->cl_softrtry = 1; 426 427 return clnt; 428 429 out_fail: 430 return clnt; 431 } 432 433 /* 434 * The way this works is that the mount process passes a structure 435 * in the data argument which contains the server's IP address 436 * and the root file handle obtained from the server's mount 437 * daemon. We stash these away in the private superblock fields. 438 */ 439 static int 440 nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data, int silent) 441 { 442 struct nfs_server *server; 443 rpc_authflavor_t authflavor; 444 445 server = NFS_SB(sb); 446 sb->s_blocksize_bits = 0; 447 sb->s_blocksize = 0; 448 if (data->bsize) 449 sb->s_blocksize = nfs_block_size(data->bsize, &sb->s_blocksize_bits); 450 if (data->rsize) 451 server->rsize = nfs_block_size(data->rsize, NULL); 452 if (data->wsize) 453 server->wsize = nfs_block_size(data->wsize, NULL); 454 server->flags = data->flags & NFS_MOUNT_FLAGMASK; 455 456 server->acregmin = data->acregmin*HZ; 457 server->acregmax = data->acregmax*HZ; 458 server->acdirmin = data->acdirmin*HZ; 459 server->acdirmax = data->acdirmax*HZ; 460 461 /* Start lockd here, before we might error out */ 462 if (!(server->flags & NFS_MOUNT_NONLM)) 463 lockd_up(); 464 465 server->namelen = data->namlen; 466 server->hostname = kmalloc(strlen(data->hostname) + 1, GFP_KERNEL); 467 if (!server->hostname) 468 return -ENOMEM; 469 strcpy(server->hostname, data->hostname); 470 471 /* Check NFS protocol revision and initialize RPC op vector 472 * and file handle pool. */ 473 #ifdef CONFIG_NFS_V3 474 if (server->flags & NFS_MOUNT_VER3) { 475 server->rpc_ops = &nfs_v3_clientops; 476 server->caps |= NFS_CAP_READDIRPLUS; 477 } else { 478 server->rpc_ops = &nfs_v2_clientops; 479 } 480 #else 481 server->rpc_ops = &nfs_v2_clientops; 482 #endif 483 484 /* Fill in pseudoflavor for mount version < 5 */ 485 if (!(data->flags & NFS_MOUNT_SECFLAVOUR)) 486 data->pseudoflavor = RPC_AUTH_UNIX; 487 authflavor = data->pseudoflavor; /* save for sb_init() */ 488 /* XXX maybe we want to add a server->pseudoflavor field */ 489 490 /* Create RPC client handles */ 491 server->client = nfs_create_client(server, data); 492 if (IS_ERR(server->client)) 493 return PTR_ERR(server->client); 494 /* RFC 2623, sec 2.3.2 */ 495 if (authflavor != RPC_AUTH_UNIX) { 496 struct rpc_auth *auth; 497 498 server->client_sys = rpc_clone_client(server->client); 499 if (IS_ERR(server->client_sys)) 500 return PTR_ERR(server->client_sys); 501 auth = rpcauth_create(RPC_AUTH_UNIX, server->client_sys); 502 if (IS_ERR(auth)) 503 return PTR_ERR(auth); 504 } else { 505 atomic_inc(&server->client->cl_count); 506 server->client_sys = server->client; 507 } 508 if (server->flags & NFS_MOUNT_VER3) { 509 #ifdef CONFIG_NFS_V3_ACL 510 if (!(server->flags & NFS_MOUNT_NOACL)) { 511 server->client_acl = rpc_bind_new_program(server->client, &nfsacl_program, 3); 512 /* No errors! Assume that Sun nfsacls are supported */ 513 if (!IS_ERR(server->client_acl)) 514 server->caps |= NFS_CAP_ACLS; 515 } 516 #else 517 server->flags &= ~NFS_MOUNT_NOACL; 518 #endif /* CONFIG_NFS_V3_ACL */ 519 /* 520 * The VFS shouldn't apply the umask to mode bits. We will 521 * do so ourselves when necessary. 522 */ 523 sb->s_flags |= MS_POSIXACL; 524 if (server->namelen == 0 || server->namelen > NFS3_MAXNAMLEN) 525 server->namelen = NFS3_MAXNAMLEN; 526 sb->s_time_gran = 1; 527 } else { 528 if (server->namelen == 0 || server->namelen > NFS2_MAXNAMLEN) 529 server->namelen = NFS2_MAXNAMLEN; 530 } 531 532 sb->s_op = &nfs_sops; 533 return nfs_sb_init(sb, authflavor); 534 } 535 536 static int 537 nfs_statfs(struct super_block *sb, struct kstatfs *buf) 538 { 539 struct nfs_server *server = NFS_SB(sb); 540 unsigned char blockbits; 541 unsigned long blockres; 542 struct nfs_fh *rootfh = NFS_FH(sb->s_root->d_inode); 543 struct nfs_fattr fattr; 544 struct nfs_fsstat res = { 545 .fattr = &fattr, 546 }; 547 int error; 548 549 lock_kernel(); 550 551 error = server->rpc_ops->statfs(server, rootfh, &res); 552 buf->f_type = NFS_SUPER_MAGIC; 553 if (error < 0) 554 goto out_err; 555 556 /* 557 * Current versions of glibc do not correctly handle the 558 * case where f_frsize != f_bsize. Eventually we want to 559 * report the value of wtmult in this field. 560 */ 561 buf->f_frsize = sb->s_blocksize; 562 563 /* 564 * On most *nix systems, f_blocks, f_bfree, and f_bavail 565 * are reported in units of f_frsize. Linux hasn't had 566 * an f_frsize field in its statfs struct until recently, 567 * thus historically Linux's sys_statfs reports these 568 * fields in units of f_bsize. 569 */ 570 buf->f_bsize = sb->s_blocksize; 571 blockbits = sb->s_blocksize_bits; 572 blockres = (1 << blockbits) - 1; 573 buf->f_blocks = (res.tbytes + blockres) >> blockbits; 574 buf->f_bfree = (res.fbytes + blockres) >> blockbits; 575 buf->f_bavail = (res.abytes + blockres) >> blockbits; 576 577 buf->f_files = res.tfiles; 578 buf->f_ffree = res.afiles; 579 580 buf->f_namelen = server->namelen; 581 out: 582 unlock_kernel(); 583 return 0; 584 585 out_err: 586 dprintk("%s: statfs error = %d\n", __FUNCTION__, -error); 587 buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1; 588 goto out; 589 590 } 591 592 static void nfs_show_mount_options(struct seq_file *m, struct nfs_server *nfss, int showdefaults) 593 { 594 static struct proc_nfs_info { 595 int flag; 596 char *str; 597 char *nostr; 598 } nfs_info[] = { 599 { NFS_MOUNT_SOFT, ",soft", ",hard" }, 600 { NFS_MOUNT_INTR, ",intr", "" }, 601 { NFS_MOUNT_NOCTO, ",nocto", "" }, 602 { NFS_MOUNT_NOAC, ",noac", "" }, 603 { NFS_MOUNT_NONLM, ",nolock", "" }, 604 { NFS_MOUNT_NOACL, ",noacl", "" }, 605 { 0, NULL, NULL } 606 }; 607 struct proc_nfs_info *nfs_infop; 608 char buf[12]; 609 char *proto; 610 611 seq_printf(m, ",vers=%d", nfss->rpc_ops->version); 612 seq_printf(m, ",rsize=%d", nfss->rsize); 613 seq_printf(m, ",wsize=%d", nfss->wsize); 614 if (nfss->acregmin != 3*HZ || showdefaults) 615 seq_printf(m, ",acregmin=%d", nfss->acregmin/HZ); 616 if (nfss->acregmax != 60*HZ || showdefaults) 617 seq_printf(m, ",acregmax=%d", nfss->acregmax/HZ); 618 if (nfss->acdirmin != 30*HZ || showdefaults) 619 seq_printf(m, ",acdirmin=%d", nfss->acdirmin/HZ); 620 if (nfss->acdirmax != 60*HZ || showdefaults) 621 seq_printf(m, ",acdirmax=%d", nfss->acdirmax/HZ); 622 for (nfs_infop = nfs_info; nfs_infop->flag; nfs_infop++) { 623 if (nfss->flags & nfs_infop->flag) 624 seq_puts(m, nfs_infop->str); 625 else 626 seq_puts(m, nfs_infop->nostr); 627 } 628 switch (nfss->client->cl_xprt->prot) { 629 case IPPROTO_TCP: 630 proto = "tcp"; 631 break; 632 case IPPROTO_UDP: 633 proto = "udp"; 634 break; 635 default: 636 snprintf(buf, sizeof(buf), "%u", nfss->client->cl_xprt->prot); 637 proto = buf; 638 } 639 seq_printf(m, ",proto=%s", proto); 640 seq_printf(m, ",timeo=%lu", 10U * nfss->retrans_timeo / HZ); 641 seq_printf(m, ",retrans=%u", nfss->retrans_count); 642 } 643 644 static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt) 645 { 646 struct nfs_server *nfss = NFS_SB(mnt->mnt_sb); 647 648 nfs_show_mount_options(m, nfss, 0); 649 650 seq_puts(m, ",addr="); 651 seq_escape(m, nfss->hostname, " \t\n\\"); 652 653 return 0; 654 } 655 656 static int nfs_show_stats(struct seq_file *m, struct vfsmount *mnt) 657 { 658 int i, cpu; 659 struct nfs_server *nfss = NFS_SB(mnt->mnt_sb); 660 struct rpc_auth *auth = nfss->client->cl_auth; 661 struct nfs_iostats totals = { }; 662 663 seq_printf(m, "statvers=%s", NFS_IOSTAT_VERS); 664 665 /* 666 * Display all mount option settings 667 */ 668 seq_printf(m, "\n\topts:\t"); 669 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? "ro" : "rw"); 670 seq_puts(m, mnt->mnt_sb->s_flags & MS_SYNCHRONOUS ? ",sync" : ""); 671 seq_puts(m, mnt->mnt_sb->s_flags & MS_NOATIME ? ",noatime" : ""); 672 seq_puts(m, mnt->mnt_sb->s_flags & MS_NODIRATIME ? ",nodiratime" : ""); 673 nfs_show_mount_options(m, nfss, 1); 674 675 seq_printf(m, "\n\tage:\t%lu", (jiffies - nfss->mount_time) / HZ); 676 677 seq_printf(m, "\n\tcaps:\t"); 678 seq_printf(m, "caps=0x%x", nfss->caps); 679 seq_printf(m, ",wtmult=%d", nfss->wtmult); 680 seq_printf(m, ",dtsize=%d", nfss->dtsize); 681 seq_printf(m, ",bsize=%d", nfss->bsize); 682 seq_printf(m, ",namelen=%d", nfss->namelen); 683 684 #ifdef CONFIG_NFS_V4 685 if (nfss->rpc_ops->version == 4) { 686 seq_printf(m, "\n\tnfsv4:\t"); 687 seq_printf(m, "bm0=0x%x", nfss->attr_bitmask[0]); 688 seq_printf(m, ",bm1=0x%x", nfss->attr_bitmask[1]); 689 seq_printf(m, ",acl=0x%x", nfss->acl_bitmask); 690 } 691 #endif 692 693 /* 694 * Display security flavor in effect for this mount 695 */ 696 seq_printf(m, "\n\tsec:\tflavor=%d", auth->au_ops->au_flavor); 697 if (auth->au_flavor) 698 seq_printf(m, ",pseudoflavor=%d", auth->au_flavor); 699 700 /* 701 * Display superblock I/O counters 702 */ 703 for_each_possible_cpu(cpu) { 704 struct nfs_iostats *stats; 705 706 preempt_disable(); 707 stats = per_cpu_ptr(nfss->io_stats, cpu); 708 709 for (i = 0; i < __NFSIOS_COUNTSMAX; i++) 710 totals.events[i] += stats->events[i]; 711 for (i = 0; i < __NFSIOS_BYTESMAX; i++) 712 totals.bytes[i] += stats->bytes[i]; 713 714 preempt_enable(); 715 } 716 717 seq_printf(m, "\n\tevents:\t"); 718 for (i = 0; i < __NFSIOS_COUNTSMAX; i++) 719 seq_printf(m, "%lu ", totals.events[i]); 720 seq_printf(m, "\n\tbytes:\t"); 721 for (i = 0; i < __NFSIOS_BYTESMAX; i++) 722 seq_printf(m, "%Lu ", totals.bytes[i]); 723 seq_printf(m, "\n"); 724 725 rpc_print_iostats(m, nfss->client); 726 727 return 0; 728 } 729 730 /** 731 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk 732 */ 733 int nfs_sync_mapping(struct address_space *mapping) 734 { 735 int ret; 736 737 if (mapping->nrpages == 0) 738 return 0; 739 unmap_mapping_range(mapping, 0, 0, 0); 740 ret = filemap_write_and_wait(mapping); 741 if (ret != 0) 742 goto out; 743 ret = nfs_wb_all(mapping->host); 744 out: 745 return ret; 746 } 747 748 /* 749 * Invalidate the local caches 750 */ 751 static void nfs_zap_caches_locked(struct inode *inode) 752 { 753 struct nfs_inode *nfsi = NFS_I(inode); 754 int mode = inode->i_mode; 755 756 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 757 758 NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode); 759 NFS_ATTRTIMEO_UPDATE(inode) = jiffies; 760 761 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode))); 762 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) 763 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE; 764 else 765 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE; 766 } 767 768 void nfs_zap_caches(struct inode *inode) 769 { 770 spin_lock(&inode->i_lock); 771 nfs_zap_caches_locked(inode); 772 spin_unlock(&inode->i_lock); 773 } 774 775 static void nfs_zap_acl_cache(struct inode *inode) 776 { 777 void (*clear_acl_cache)(struct inode *); 778 779 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache; 780 if (clear_acl_cache != NULL) 781 clear_acl_cache(inode); 782 spin_lock(&inode->i_lock); 783 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL; 784 spin_unlock(&inode->i_lock); 785 } 786 787 /* 788 * Invalidate, but do not unhash, the inode. 789 * NB: must be called with inode->i_lock held! 790 */ 791 static void nfs_invalidate_inode(struct inode *inode) 792 { 793 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode)); 794 nfs_zap_caches_locked(inode); 795 } 796 797 struct nfs_find_desc { 798 struct nfs_fh *fh; 799 struct nfs_fattr *fattr; 800 }; 801 802 /* 803 * In NFSv3 we can have 64bit inode numbers. In order to support 804 * this, and re-exported directories (also seen in NFSv2) 805 * we are forced to allow 2 different inodes to have the same 806 * i_ino. 807 */ 808 static int 809 nfs_find_actor(struct inode *inode, void *opaque) 810 { 811 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque; 812 struct nfs_fh *fh = desc->fh; 813 struct nfs_fattr *fattr = desc->fattr; 814 815 if (NFS_FILEID(inode) != fattr->fileid) 816 return 0; 817 if (nfs_compare_fh(NFS_FH(inode), fh)) 818 return 0; 819 if (is_bad_inode(inode) || NFS_STALE(inode)) 820 return 0; 821 return 1; 822 } 823 824 static int 825 nfs_init_locked(struct inode *inode, void *opaque) 826 { 827 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque; 828 struct nfs_fattr *fattr = desc->fattr; 829 830 NFS_FILEID(inode) = fattr->fileid; 831 nfs_copy_fh(NFS_FH(inode), desc->fh); 832 return 0; 833 } 834 835 /* Don't use READDIRPLUS on directories that we believe are too large */ 836 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE) 837 838 /* 839 * This is our front-end to iget that looks up inodes by file handle 840 * instead of inode number. 841 */ 842 struct inode * 843 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr) 844 { 845 struct nfs_find_desc desc = { 846 .fh = fh, 847 .fattr = fattr 848 }; 849 struct inode *inode = ERR_PTR(-ENOENT); 850 unsigned long hash; 851 852 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 853 goto out_no_inode; 854 855 if (!fattr->nlink) { 856 printk("NFS: Buggy server - nlink == 0!\n"); 857 goto out_no_inode; 858 } 859 860 hash = nfs_fattr_to_ino_t(fattr); 861 862 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc); 863 if (inode == NULL) { 864 inode = ERR_PTR(-ENOMEM); 865 goto out_no_inode; 866 } 867 868 if (inode->i_state & I_NEW) { 869 struct nfs_inode *nfsi = NFS_I(inode); 870 871 /* We set i_ino for the few things that still rely on it, 872 * such as stat(2) */ 873 inode->i_ino = hash; 874 875 /* We can't support update_atime(), since the server will reset it */ 876 inode->i_flags |= S_NOATIME|S_NOCMTIME; 877 inode->i_mode = fattr->mode; 878 /* Why so? Because we want revalidate for devices/FIFOs, and 879 * that's precisely what we have in nfs_file_inode_operations. 880 */ 881 inode->i_op = NFS_SB(sb)->rpc_ops->file_inode_ops; 882 if (S_ISREG(inode->i_mode)) { 883 inode->i_fop = &nfs_file_operations; 884 inode->i_data.a_ops = &nfs_file_aops; 885 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info; 886 } else if (S_ISDIR(inode->i_mode)) { 887 inode->i_op = NFS_SB(sb)->rpc_ops->dir_inode_ops; 888 inode->i_fop = &nfs_dir_operations; 889 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS) 890 && fattr->size <= NFS_LIMIT_READDIRPLUS) 891 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode)); 892 } else if (S_ISLNK(inode->i_mode)) 893 inode->i_op = &nfs_symlink_inode_operations; 894 else 895 init_special_inode(inode, inode->i_mode, fattr->rdev); 896 897 nfsi->read_cache_jiffies = fattr->time_start; 898 nfsi->last_updated = jiffies; 899 inode->i_atime = fattr->atime; 900 inode->i_mtime = fattr->mtime; 901 inode->i_ctime = fattr->ctime; 902 if (fattr->valid & NFS_ATTR_FATTR_V4) 903 nfsi->change_attr = fattr->change_attr; 904 inode->i_size = nfs_size_to_loff_t(fattr->size); 905 inode->i_nlink = fattr->nlink; 906 inode->i_uid = fattr->uid; 907 inode->i_gid = fattr->gid; 908 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) { 909 /* 910 * report the blocks in 512byte units 911 */ 912 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 913 inode->i_blksize = inode->i_sb->s_blocksize; 914 } else { 915 inode->i_blocks = fattr->du.nfs2.blocks; 916 inode->i_blksize = fattr->du.nfs2.blocksize; 917 } 918 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 919 nfsi->attrtimeo_timestamp = jiffies; 920 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf)); 921 nfsi->cache_access.cred = NULL; 922 923 unlock_new_inode(inode); 924 } else 925 nfs_refresh_inode(inode, fattr); 926 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n", 927 inode->i_sb->s_id, 928 (long long)NFS_FILEID(inode), 929 atomic_read(&inode->i_count)); 930 931 out: 932 return inode; 933 934 out_no_inode: 935 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode)); 936 goto out; 937 } 938 939 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET) 940 941 int 942 nfs_setattr(struct dentry *dentry, struct iattr *attr) 943 { 944 struct inode *inode = dentry->d_inode; 945 struct nfs_fattr fattr; 946 int error; 947 948 nfs_inc_stats(inode, NFSIOS_VFSSETATTR); 949 950 if (attr->ia_valid & ATTR_SIZE) { 951 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode)) 952 attr->ia_valid &= ~ATTR_SIZE; 953 } 954 955 /* Optimization: if the end result is no change, don't RPC */ 956 attr->ia_valid &= NFS_VALID_ATTRS; 957 if (attr->ia_valid == 0) 958 return 0; 959 960 lock_kernel(); 961 nfs_begin_data_update(inode); 962 /* Write all dirty data */ 963 filemap_write_and_wait(inode->i_mapping); 964 nfs_wb_all(inode); 965 /* 966 * Return any delegations if we're going to change ACLs 967 */ 968 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) 969 nfs_inode_return_delegation(inode); 970 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr); 971 if (error == 0) 972 nfs_refresh_inode(inode, &fattr); 973 nfs_end_data_update(inode); 974 unlock_kernel(); 975 return error; 976 } 977 978 /** 979 * nfs_setattr_update_inode - Update inode metadata after a setattr call. 980 * @inode: pointer to struct inode 981 * @attr: pointer to struct iattr 982 * 983 * Note: we do this in the *proc.c in order to ensure that 984 * it works for things like exclusive creates too. 985 */ 986 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr) 987 { 988 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) { 989 if ((attr->ia_valid & ATTR_MODE) != 0) { 990 int mode = attr->ia_mode & S_IALLUGO; 991 mode |= inode->i_mode & ~S_IALLUGO; 992 inode->i_mode = mode; 993 } 994 if ((attr->ia_valid & ATTR_UID) != 0) 995 inode->i_uid = attr->ia_uid; 996 if ((attr->ia_valid & ATTR_GID) != 0) 997 inode->i_gid = attr->ia_gid; 998 spin_lock(&inode->i_lock); 999 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 1000 spin_unlock(&inode->i_lock); 1001 } 1002 if ((attr->ia_valid & ATTR_SIZE) != 0) { 1003 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC); 1004 inode->i_size = attr->ia_size; 1005 vmtruncate(inode, attr->ia_size); 1006 } 1007 } 1008 1009 static int nfs_wait_schedule(void *word) 1010 { 1011 if (signal_pending(current)) 1012 return -ERESTARTSYS; 1013 schedule(); 1014 return 0; 1015 } 1016 1017 /* 1018 * Wait for the inode to get unlocked. 1019 */ 1020 static int nfs_wait_on_inode(struct inode *inode) 1021 { 1022 struct rpc_clnt *clnt = NFS_CLIENT(inode); 1023 struct nfs_inode *nfsi = NFS_I(inode); 1024 sigset_t oldmask; 1025 int error; 1026 1027 rpc_clnt_sigmask(clnt, &oldmask); 1028 error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING, 1029 nfs_wait_schedule, TASK_INTERRUPTIBLE); 1030 rpc_clnt_sigunmask(clnt, &oldmask); 1031 1032 return error; 1033 } 1034 1035 static void nfs_wake_up_inode(struct inode *inode) 1036 { 1037 struct nfs_inode *nfsi = NFS_I(inode); 1038 1039 clear_bit(NFS_INO_REVALIDATING, &nfsi->flags); 1040 smp_mb__after_clear_bit(); 1041 wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING); 1042 } 1043 1044 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat) 1045 { 1046 struct inode *inode = dentry->d_inode; 1047 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME; 1048 int err; 1049 1050 /* Flush out writes to the server in order to update c/mtime */ 1051 nfs_sync_inode_wait(inode, 0, 0, FLUSH_NOCOMMIT); 1052 1053 /* 1054 * We may force a getattr if the user cares about atime. 1055 * 1056 * Note that we only have to check the vfsmount flags here: 1057 * - NFS always sets S_NOATIME by so checking it would give a 1058 * bogus result 1059 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is 1060 * no point in checking those. 1061 */ 1062 if ((mnt->mnt_flags & MNT_NOATIME) || 1063 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) 1064 need_atime = 0; 1065 1066 if (need_atime) 1067 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 1068 else 1069 err = nfs_revalidate_inode(NFS_SERVER(inode), inode); 1070 if (!err) 1071 generic_fillattr(inode, stat); 1072 return err; 1073 } 1074 1075 static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred) 1076 { 1077 struct nfs_open_context *ctx; 1078 1079 ctx = (struct nfs_open_context *)kmalloc(sizeof(*ctx), GFP_KERNEL); 1080 if (ctx != NULL) { 1081 atomic_set(&ctx->count, 1); 1082 ctx->dentry = dget(dentry); 1083 ctx->vfsmnt = mntget(mnt); 1084 ctx->cred = get_rpccred(cred); 1085 ctx->state = NULL; 1086 ctx->lockowner = current->files; 1087 ctx->error = 0; 1088 ctx->dir_cookie = 0; 1089 } 1090 return ctx; 1091 } 1092 1093 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx) 1094 { 1095 if (ctx != NULL) 1096 atomic_inc(&ctx->count); 1097 return ctx; 1098 } 1099 1100 void put_nfs_open_context(struct nfs_open_context *ctx) 1101 { 1102 if (atomic_dec_and_test(&ctx->count)) { 1103 if (!list_empty(&ctx->list)) { 1104 struct inode *inode = ctx->dentry->d_inode; 1105 spin_lock(&inode->i_lock); 1106 list_del(&ctx->list); 1107 spin_unlock(&inode->i_lock); 1108 } 1109 if (ctx->state != NULL) 1110 nfs4_close_state(ctx->state, ctx->mode); 1111 if (ctx->cred != NULL) 1112 put_rpccred(ctx->cred); 1113 dput(ctx->dentry); 1114 mntput(ctx->vfsmnt); 1115 kfree(ctx); 1116 } 1117 } 1118 1119 /* 1120 * Ensure that mmap has a recent RPC credential for use when writing out 1121 * shared pages 1122 */ 1123 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx) 1124 { 1125 struct inode *inode = filp->f_dentry->d_inode; 1126 struct nfs_inode *nfsi = NFS_I(inode); 1127 1128 filp->private_data = get_nfs_open_context(ctx); 1129 spin_lock(&inode->i_lock); 1130 list_add(&ctx->list, &nfsi->open_files); 1131 spin_unlock(&inode->i_lock); 1132 } 1133 1134 /* 1135 * Given an inode, search for an open context with the desired characteristics 1136 */ 1137 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode) 1138 { 1139 struct nfs_inode *nfsi = NFS_I(inode); 1140 struct nfs_open_context *pos, *ctx = NULL; 1141 1142 spin_lock(&inode->i_lock); 1143 list_for_each_entry(pos, &nfsi->open_files, list) { 1144 if (cred != NULL && pos->cred != cred) 1145 continue; 1146 if ((pos->mode & mode) == mode) { 1147 ctx = get_nfs_open_context(pos); 1148 break; 1149 } 1150 } 1151 spin_unlock(&inode->i_lock); 1152 return ctx; 1153 } 1154 1155 static void nfs_file_clear_open_context(struct file *filp) 1156 { 1157 struct inode *inode = filp->f_dentry->d_inode; 1158 struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data; 1159 1160 if (ctx) { 1161 filp->private_data = NULL; 1162 spin_lock(&inode->i_lock); 1163 list_move_tail(&ctx->list, &NFS_I(inode)->open_files); 1164 spin_unlock(&inode->i_lock); 1165 put_nfs_open_context(ctx); 1166 } 1167 } 1168 1169 /* 1170 * These allocate and release file read/write context information. 1171 */ 1172 int nfs_open(struct inode *inode, struct file *filp) 1173 { 1174 struct nfs_open_context *ctx; 1175 struct rpc_cred *cred; 1176 1177 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0); 1178 if (IS_ERR(cred)) 1179 return PTR_ERR(cred); 1180 ctx = alloc_nfs_open_context(filp->f_vfsmnt, filp->f_dentry, cred); 1181 put_rpccred(cred); 1182 if (ctx == NULL) 1183 return -ENOMEM; 1184 ctx->mode = filp->f_mode; 1185 nfs_file_set_open_context(filp, ctx); 1186 put_nfs_open_context(ctx); 1187 return 0; 1188 } 1189 1190 int nfs_release(struct inode *inode, struct file *filp) 1191 { 1192 nfs_file_clear_open_context(filp); 1193 return 0; 1194 } 1195 1196 /* 1197 * This function is called whenever some part of NFS notices that 1198 * the cached attributes have to be refreshed. 1199 */ 1200 int 1201 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) 1202 { 1203 int status = -ESTALE; 1204 struct nfs_fattr fattr; 1205 struct nfs_inode *nfsi = NFS_I(inode); 1206 1207 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n", 1208 inode->i_sb->s_id, (long long)NFS_FILEID(inode)); 1209 1210 lock_kernel(); 1211 if (!inode || is_bad_inode(inode)) 1212 goto out_nowait; 1213 if (NFS_STALE(inode)) 1214 goto out_nowait; 1215 1216 status = nfs_wait_on_inode(inode); 1217 if (status < 0) 1218 goto out; 1219 if (NFS_STALE(inode)) { 1220 status = -ESTALE; 1221 /* Do we trust the cached ESTALE? */ 1222 if (NFS_ATTRTIMEO(inode) != 0) { 1223 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ATIME)) { 1224 /* no */ 1225 } else 1226 goto out; 1227 } 1228 } 1229 1230 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr); 1231 if (status != 0) { 1232 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n", 1233 inode->i_sb->s_id, 1234 (long long)NFS_FILEID(inode), status); 1235 if (status == -ESTALE) { 1236 nfs_zap_caches(inode); 1237 if (!S_ISDIR(inode->i_mode)) 1238 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode)); 1239 } 1240 goto out; 1241 } 1242 1243 spin_lock(&inode->i_lock); 1244 status = nfs_update_inode(inode, &fattr); 1245 if (status) { 1246 spin_unlock(&inode->i_lock); 1247 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n", 1248 inode->i_sb->s_id, 1249 (long long)NFS_FILEID(inode), status); 1250 goto out; 1251 } 1252 spin_unlock(&inode->i_lock); 1253 1254 nfs_revalidate_mapping(inode, inode->i_mapping); 1255 1256 if (nfsi->cache_validity & NFS_INO_INVALID_ACL) 1257 nfs_zap_acl_cache(inode); 1258 1259 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n", 1260 inode->i_sb->s_id, 1261 (long long)NFS_FILEID(inode)); 1262 1263 out: 1264 nfs_wake_up_inode(inode); 1265 1266 out_nowait: 1267 unlock_kernel(); 1268 return status; 1269 } 1270 1271 int nfs_attribute_timeout(struct inode *inode) 1272 { 1273 struct nfs_inode *nfsi = NFS_I(inode); 1274 1275 if (nfs_have_delegation(inode, FMODE_READ)) 1276 return 0; 1277 return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo); 1278 } 1279 1280 /** 1281 * nfs_revalidate_inode - Revalidate the inode attributes 1282 * @server - pointer to nfs_server struct 1283 * @inode - pointer to inode struct 1284 * 1285 * Updates inode attribute information by retrieving the data from the server. 1286 */ 1287 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) 1288 { 1289 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE); 1290 if (!(NFS_I(inode)->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA)) 1291 && !nfs_attribute_timeout(inode)) 1292 return NFS_STALE(inode) ? -ESTALE : 0; 1293 return __nfs_revalidate_inode(server, inode); 1294 } 1295 1296 /** 1297 * nfs_revalidate_mapping - Revalidate the pagecache 1298 * @inode - pointer to host inode 1299 * @mapping - pointer to mapping 1300 */ 1301 void nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping) 1302 { 1303 struct nfs_inode *nfsi = NFS_I(inode); 1304 1305 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) { 1306 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE); 1307 if (S_ISREG(inode->i_mode)) 1308 nfs_sync_mapping(mapping); 1309 invalidate_inode_pages2(mapping); 1310 1311 spin_lock(&inode->i_lock); 1312 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA; 1313 if (S_ISDIR(inode->i_mode)) { 1314 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf)); 1315 /* This ensures we revalidate child dentries */ 1316 nfsi->cache_change_attribute = jiffies; 1317 } 1318 spin_unlock(&inode->i_lock); 1319 1320 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n", 1321 inode->i_sb->s_id, 1322 (long long)NFS_FILEID(inode)); 1323 } 1324 } 1325 1326 /** 1327 * nfs_begin_data_update 1328 * @inode - pointer to inode 1329 * Declare that a set of operations will update file data on the server 1330 */ 1331 void nfs_begin_data_update(struct inode *inode) 1332 { 1333 atomic_inc(&NFS_I(inode)->data_updates); 1334 } 1335 1336 /** 1337 * nfs_end_data_update 1338 * @inode - pointer to inode 1339 * Declare end of the operations that will update file data 1340 * This will mark the inode as immediately needing revalidation 1341 * of its attribute cache. 1342 */ 1343 void nfs_end_data_update(struct inode *inode) 1344 { 1345 struct nfs_inode *nfsi = NFS_I(inode); 1346 1347 if (!nfs_have_delegation(inode, FMODE_READ)) { 1348 /* Directories and symlinks: invalidate page cache */ 1349 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) { 1350 spin_lock(&inode->i_lock); 1351 nfsi->cache_validity |= NFS_INO_INVALID_DATA; 1352 spin_unlock(&inode->i_lock); 1353 } 1354 } 1355 nfsi->cache_change_attribute = jiffies; 1356 atomic_dec(&nfsi->data_updates); 1357 } 1358 1359 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1360 { 1361 struct nfs_inode *nfsi = NFS_I(inode); 1362 1363 if ((fattr->valid & NFS_ATTR_PRE_CHANGE) != 0 1364 && nfsi->change_attr == fattr->pre_change_attr) { 1365 nfsi->change_attr = fattr->change_attr; 1366 nfsi->cache_change_attribute = jiffies; 1367 } 1368 1369 /* If we have atomic WCC data, we may update some attributes */ 1370 if ((fattr->valid & NFS_ATTR_WCC) != 0) { 1371 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) { 1372 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime)); 1373 nfsi->cache_change_attribute = jiffies; 1374 } 1375 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) { 1376 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime)); 1377 nfsi->cache_change_attribute = jiffies; 1378 } 1379 if (inode->i_size == fattr->pre_size && nfsi->npages == 0) { 1380 inode->i_size = fattr->size; 1381 nfsi->cache_change_attribute = jiffies; 1382 } 1383 } 1384 } 1385 1386 /** 1387 * nfs_check_inode_attributes - verify consistency of the inode attribute cache 1388 * @inode - pointer to inode 1389 * @fattr - updated attributes 1390 * 1391 * Verifies the attribute cache. If we have just changed the attributes, 1392 * so that fattr carries weak cache consistency data, then it may 1393 * also update the ctime/mtime/change_attribute. 1394 */ 1395 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr) 1396 { 1397 struct nfs_inode *nfsi = NFS_I(inode); 1398 loff_t cur_size, new_isize; 1399 int data_unstable; 1400 1401 1402 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 1403 return 0; 1404 1405 /* Has the inode gone and changed behind our back? */ 1406 if (nfsi->fileid != fattr->fileid 1407 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) { 1408 return -EIO; 1409 } 1410 1411 /* Are we in the process of updating data on the server? */ 1412 data_unstable = nfs_caches_unstable(inode); 1413 1414 /* Do atomic weak cache consistency updates */ 1415 nfs_wcc_update_inode(inode, fattr); 1416 1417 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0) { 1418 if (nfsi->change_attr == fattr->change_attr) 1419 goto out; 1420 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 1421 if (!data_unstable) 1422 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE; 1423 } 1424 1425 /* Verify a few of the more important attributes */ 1426 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) { 1427 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 1428 if (!data_unstable) 1429 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE; 1430 } 1431 1432 cur_size = i_size_read(inode); 1433 new_isize = nfs_size_to_loff_t(fattr->size); 1434 if (cur_size != new_isize && nfsi->npages == 0) 1435 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE; 1436 1437 /* Have any file permissions changed? */ 1438 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) 1439 || inode->i_uid != fattr->uid 1440 || inode->i_gid != fattr->gid) 1441 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL; 1442 1443 /* Has the link count changed? */ 1444 if (inode->i_nlink != fattr->nlink) 1445 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 1446 1447 out: 1448 if (!timespec_equal(&inode->i_atime, &fattr->atime)) 1449 nfsi->cache_validity |= NFS_INO_INVALID_ATIME; 1450 1451 nfsi->read_cache_jiffies = fattr->time_start; 1452 return 0; 1453 } 1454 1455 /** 1456 * nfs_refresh_inode - try to update the inode attribute cache 1457 * @inode - pointer to inode 1458 * @fattr - updated attributes 1459 * 1460 * Check that an RPC call that returned attributes has not overlapped with 1461 * other recent updates of the inode metadata, then decide whether it is 1462 * safe to do a full update of the inode attributes, or whether just to 1463 * call nfs_check_inode_attributes. 1464 */ 1465 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr) 1466 { 1467 struct nfs_inode *nfsi = NFS_I(inode); 1468 int status; 1469 1470 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 1471 return 0; 1472 spin_lock(&inode->i_lock); 1473 nfsi->cache_validity &= ~NFS_INO_REVAL_PAGECACHE; 1474 if (time_after(fattr->time_start, nfsi->last_updated)) 1475 status = nfs_update_inode(inode, fattr); 1476 else 1477 status = nfs_check_inode_attributes(inode, fattr); 1478 1479 spin_unlock(&inode->i_lock); 1480 return status; 1481 } 1482 1483 /** 1484 * nfs_post_op_update_inode - try to update the inode attribute cache 1485 * @inode - pointer to inode 1486 * @fattr - updated attributes 1487 * 1488 * After an operation that has changed the inode metadata, mark the 1489 * attribute cache as being invalid, then try to update it. 1490 */ 1491 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1492 { 1493 struct nfs_inode *nfsi = NFS_I(inode); 1494 int status = 0; 1495 1496 spin_lock(&inode->i_lock); 1497 if (unlikely((fattr->valid & NFS_ATTR_FATTR) == 0)) { 1498 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS; 1499 goto out; 1500 } 1501 status = nfs_update_inode(inode, fattr); 1502 out: 1503 spin_unlock(&inode->i_lock); 1504 return status; 1505 } 1506 1507 /* 1508 * Many nfs protocol calls return the new file attributes after 1509 * an operation. Here we update the inode to reflect the state 1510 * of the server's inode. 1511 * 1512 * This is a bit tricky because we have to make sure all dirty pages 1513 * have been sent off to the server before calling invalidate_inode_pages. 1514 * To make sure no other process adds more write requests while we try 1515 * our best to flush them, we make them sleep during the attribute refresh. 1516 * 1517 * A very similar scenario holds for the dir cache. 1518 */ 1519 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1520 { 1521 struct nfs_inode *nfsi = NFS_I(inode); 1522 loff_t cur_isize, new_isize; 1523 unsigned int invalid = 0; 1524 int data_stable; 1525 1526 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n", 1527 __FUNCTION__, inode->i_sb->s_id, inode->i_ino, 1528 atomic_read(&inode->i_count), fattr->valid); 1529 1530 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 1531 return 0; 1532 1533 if (nfsi->fileid != fattr->fileid) 1534 goto out_fileid; 1535 1536 /* 1537 * Make sure the inode's type hasn't changed. 1538 */ 1539 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) 1540 goto out_changed; 1541 1542 /* 1543 * Update the read time so we don't revalidate too often. 1544 */ 1545 nfsi->read_cache_jiffies = fattr->time_start; 1546 nfsi->last_updated = jiffies; 1547 1548 /* Are we racing with known updates of the metadata on the server? */ 1549 data_stable = nfs_verify_change_attribute(inode, fattr->time_start); 1550 if (data_stable) 1551 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME); 1552 1553 /* Do atomic weak cache consistency updates */ 1554 nfs_wcc_update_inode(inode, fattr); 1555 1556 /* Check if our cached file size is stale */ 1557 new_isize = nfs_size_to_loff_t(fattr->size); 1558 cur_isize = i_size_read(inode); 1559 if (new_isize != cur_isize) { 1560 /* Do we perhaps have any outstanding writes? */ 1561 if (nfsi->npages == 0) { 1562 /* No, but did we race with nfs_end_data_update()? */ 1563 if (data_stable) { 1564 inode->i_size = new_isize; 1565 invalid |= NFS_INO_INVALID_DATA; 1566 } 1567 invalid |= NFS_INO_INVALID_ATTR; 1568 } else if (new_isize > cur_isize) { 1569 inode->i_size = new_isize; 1570 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA; 1571 } 1572 nfsi->cache_change_attribute = jiffies; 1573 dprintk("NFS: isize change on server for file %s/%ld\n", 1574 inode->i_sb->s_id, inode->i_ino); 1575 } 1576 1577 /* Check if the mtime agrees */ 1578 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) { 1579 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime)); 1580 dprintk("NFS: mtime change on server for file %s/%ld\n", 1581 inode->i_sb->s_id, inode->i_ino); 1582 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA; 1583 nfsi->cache_change_attribute = jiffies; 1584 } 1585 1586 /* If ctime has changed we should definitely clear access+acl caches */ 1587 if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) { 1588 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 1589 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime)); 1590 nfsi->cache_change_attribute = jiffies; 1591 } 1592 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime)); 1593 1594 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) || 1595 inode->i_uid != fattr->uid || 1596 inode->i_gid != fattr->gid) 1597 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 1598 1599 inode->i_mode = fattr->mode; 1600 inode->i_nlink = fattr->nlink; 1601 inode->i_uid = fattr->uid; 1602 inode->i_gid = fattr->gid; 1603 1604 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) { 1605 /* 1606 * report the blocks in 512byte units 1607 */ 1608 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 1609 inode->i_blksize = inode->i_sb->s_blocksize; 1610 } else { 1611 inode->i_blocks = fattr->du.nfs2.blocks; 1612 inode->i_blksize = fattr->du.nfs2.blocksize; 1613 } 1614 1615 if ((fattr->valid & NFS_ATTR_FATTR_V4)) { 1616 if (nfsi->change_attr != fattr->change_attr) { 1617 dprintk("NFS: change_attr change on server for file %s/%ld\n", 1618 inode->i_sb->s_id, inode->i_ino); 1619 nfsi->change_attr = fattr->change_attr; 1620 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 1621 nfsi->cache_change_attribute = jiffies; 1622 } else 1623 invalid &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA); 1624 } 1625 1626 /* Update attrtimeo value if we're out of the unstable period */ 1627 if (invalid & NFS_INO_INVALID_ATTR) { 1628 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 1629 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 1630 nfsi->attrtimeo_timestamp = jiffies; 1631 } else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) { 1632 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode)) 1633 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode); 1634 nfsi->attrtimeo_timestamp = jiffies; 1635 } 1636 /* Don't invalidate the data if we were to blame */ 1637 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) 1638 || S_ISLNK(inode->i_mode))) 1639 invalid &= ~NFS_INO_INVALID_DATA; 1640 if (data_stable) 1641 invalid &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME|NFS_INO_REVAL_PAGECACHE); 1642 if (!nfs_have_delegation(inode, FMODE_READ)) 1643 nfsi->cache_validity |= invalid; 1644 1645 return 0; 1646 out_changed: 1647 /* 1648 * Big trouble! The inode has become a different object. 1649 */ 1650 #ifdef NFS_PARANOIA 1651 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n", 1652 __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode); 1653 #endif 1654 out_err: 1655 /* 1656 * No need to worry about unhashing the dentry, as the 1657 * lookup validation will know that the inode is bad. 1658 * (But we fall through to invalidate the caches.) 1659 */ 1660 nfs_invalidate_inode(inode); 1661 return -ESTALE; 1662 1663 out_fileid: 1664 printk(KERN_ERR "NFS: server %s error: fileid changed\n" 1665 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n", 1666 NFS_SERVER(inode)->hostname, inode->i_sb->s_id, 1667 (long long)nfsi->fileid, (long long)fattr->fileid); 1668 goto out_err; 1669 } 1670 1671 /* 1672 * File system information 1673 */ 1674 1675 static int nfs_set_super(struct super_block *s, void *data) 1676 { 1677 s->s_fs_info = data; 1678 return set_anon_super(s, data); 1679 } 1680 1681 static int nfs_compare_super(struct super_block *sb, void *data) 1682 { 1683 struct nfs_server *server = data; 1684 struct nfs_server *old = NFS_SB(sb); 1685 1686 if (old->addr.sin_addr.s_addr != server->addr.sin_addr.s_addr) 1687 return 0; 1688 if (old->addr.sin_port != server->addr.sin_port) 1689 return 0; 1690 return !nfs_compare_fh(&old->fh, &server->fh); 1691 } 1692 1693 static struct super_block *nfs_get_sb(struct file_system_type *fs_type, 1694 int flags, const char *dev_name, void *raw_data) 1695 { 1696 int error; 1697 struct nfs_server *server = NULL; 1698 struct super_block *s; 1699 struct nfs_fh *root; 1700 struct nfs_mount_data *data = raw_data; 1701 1702 s = ERR_PTR(-EINVAL); 1703 if (data == NULL) { 1704 dprintk("%s: missing data argument\n", __FUNCTION__); 1705 goto out_err; 1706 } 1707 if (data->version <= 0 || data->version > NFS_MOUNT_VERSION) { 1708 dprintk("%s: bad mount version\n", __FUNCTION__); 1709 goto out_err; 1710 } 1711 switch (data->version) { 1712 case 1: 1713 data->namlen = 0; 1714 case 2: 1715 data->bsize = 0; 1716 case 3: 1717 if (data->flags & NFS_MOUNT_VER3) { 1718 dprintk("%s: mount structure version %d does not support NFSv3\n", 1719 __FUNCTION__, 1720 data->version); 1721 goto out_err; 1722 } 1723 data->root.size = NFS2_FHSIZE; 1724 memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE); 1725 case 4: 1726 if (data->flags & NFS_MOUNT_SECFLAVOUR) { 1727 dprintk("%s: mount structure version %d does not support strong security\n", 1728 __FUNCTION__, 1729 data->version); 1730 goto out_err; 1731 } 1732 case 5: 1733 memset(data->context, 0, sizeof(data->context)); 1734 } 1735 #ifndef CONFIG_NFS_V3 1736 /* If NFSv3 is not compiled in, return -EPROTONOSUPPORT */ 1737 s = ERR_PTR(-EPROTONOSUPPORT); 1738 if (data->flags & NFS_MOUNT_VER3) { 1739 dprintk("%s: NFSv3 not compiled into kernel\n", __FUNCTION__); 1740 goto out_err; 1741 } 1742 #endif /* CONFIG_NFS_V3 */ 1743 1744 s = ERR_PTR(-ENOMEM); 1745 server = kzalloc(sizeof(struct nfs_server), GFP_KERNEL); 1746 if (!server) 1747 goto out_err; 1748 /* Zero out the NFS state stuff */ 1749 init_nfsv4_state(server); 1750 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL); 1751 1752 root = &server->fh; 1753 if (data->flags & NFS_MOUNT_VER3) 1754 root->size = data->root.size; 1755 else 1756 root->size = NFS2_FHSIZE; 1757 s = ERR_PTR(-EINVAL); 1758 if (root->size > sizeof(root->data)) { 1759 dprintk("%s: invalid root filehandle\n", __FUNCTION__); 1760 goto out_err; 1761 } 1762 memcpy(root->data, data->root.data, root->size); 1763 1764 /* We now require that the mount process passes the remote address */ 1765 memcpy(&server->addr, &data->addr, sizeof(server->addr)); 1766 if (server->addr.sin_addr.s_addr == INADDR_ANY) { 1767 dprintk("%s: mount program didn't pass remote address!\n", 1768 __FUNCTION__); 1769 goto out_err; 1770 } 1771 1772 /* Fire up rpciod if not yet running */ 1773 s = ERR_PTR(rpciod_up()); 1774 if (IS_ERR(s)) { 1775 dprintk("%s: couldn't start rpciod! Error = %ld\n", 1776 __FUNCTION__, PTR_ERR(s)); 1777 goto out_err; 1778 } 1779 1780 s = sget(fs_type, nfs_compare_super, nfs_set_super, server); 1781 if (IS_ERR(s) || s->s_root) 1782 goto out_rpciod_down; 1783 1784 s->s_flags = flags; 1785 1786 error = nfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0); 1787 if (error) { 1788 up_write(&s->s_umount); 1789 deactivate_super(s); 1790 return ERR_PTR(error); 1791 } 1792 s->s_flags |= MS_ACTIVE; 1793 return s; 1794 out_rpciod_down: 1795 rpciod_down(); 1796 out_err: 1797 kfree(server); 1798 return s; 1799 } 1800 1801 static void nfs_kill_super(struct super_block *s) 1802 { 1803 struct nfs_server *server = NFS_SB(s); 1804 1805 kill_anon_super(s); 1806 1807 if (!IS_ERR(server->client)) 1808 rpc_shutdown_client(server->client); 1809 if (!IS_ERR(server->client_sys)) 1810 rpc_shutdown_client(server->client_sys); 1811 if (!IS_ERR(server->client_acl)) 1812 rpc_shutdown_client(server->client_acl); 1813 1814 if (!(server->flags & NFS_MOUNT_NONLM)) 1815 lockd_down(); /* release rpc.lockd */ 1816 1817 rpciod_down(); /* release rpciod */ 1818 1819 nfs_free_iostats(server->io_stats); 1820 kfree(server->hostname); 1821 kfree(server); 1822 } 1823 1824 static struct file_system_type nfs_fs_type = { 1825 .owner = THIS_MODULE, 1826 .name = "nfs", 1827 .get_sb = nfs_get_sb, 1828 .kill_sb = nfs_kill_super, 1829 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA, 1830 }; 1831 1832 #ifdef CONFIG_NFS_V4 1833 1834 static void nfs4_clear_inode(struct inode *); 1835 1836 1837 static struct super_operations nfs4_sops = { 1838 .alloc_inode = nfs_alloc_inode, 1839 .destroy_inode = nfs_destroy_inode, 1840 .write_inode = nfs_write_inode, 1841 .delete_inode = nfs_delete_inode, 1842 .statfs = nfs_statfs, 1843 .clear_inode = nfs4_clear_inode, 1844 .umount_begin = nfs_umount_begin, 1845 .show_options = nfs_show_options, 1846 .show_stats = nfs_show_stats, 1847 }; 1848 1849 /* 1850 * Clean out any remaining NFSv4 state that might be left over due 1851 * to open() calls that passed nfs_atomic_lookup, but failed to call 1852 * nfs_open(). 1853 */ 1854 static void nfs4_clear_inode(struct inode *inode) 1855 { 1856 struct nfs_inode *nfsi = NFS_I(inode); 1857 1858 /* If we are holding a delegation, return it! */ 1859 nfs_inode_return_delegation(inode); 1860 /* First call standard NFS clear_inode() code */ 1861 nfs_clear_inode(inode); 1862 /* Now clear out any remaining state */ 1863 while (!list_empty(&nfsi->open_states)) { 1864 struct nfs4_state *state; 1865 1866 state = list_entry(nfsi->open_states.next, 1867 struct nfs4_state, 1868 inode_states); 1869 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n", 1870 __FUNCTION__, 1871 inode->i_sb->s_id, 1872 (long long)NFS_FILEID(inode), 1873 state); 1874 BUG_ON(atomic_read(&state->count) != 1); 1875 nfs4_close_state(state, state->state); 1876 } 1877 } 1878 1879 1880 static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data, int silent) 1881 { 1882 struct nfs_server *server; 1883 struct nfs4_client *clp = NULL; 1884 struct rpc_xprt *xprt = NULL; 1885 struct rpc_clnt *clnt = NULL; 1886 struct rpc_timeout timeparms; 1887 rpc_authflavor_t authflavour; 1888 int err = -EIO; 1889 1890 sb->s_blocksize_bits = 0; 1891 sb->s_blocksize = 0; 1892 server = NFS_SB(sb); 1893 if (data->rsize != 0) 1894 server->rsize = nfs_block_size(data->rsize, NULL); 1895 if (data->wsize != 0) 1896 server->wsize = nfs_block_size(data->wsize, NULL); 1897 server->flags = data->flags & NFS_MOUNT_FLAGMASK; 1898 server->caps = NFS_CAP_ATOMIC_OPEN; 1899 1900 server->acregmin = data->acregmin*HZ; 1901 server->acregmax = data->acregmax*HZ; 1902 server->acdirmin = data->acdirmin*HZ; 1903 server->acdirmax = data->acdirmax*HZ; 1904 1905 server->rpc_ops = &nfs_v4_clientops; 1906 1907 nfs_init_timeout_values(&timeparms, data->proto, data->timeo, data->retrans); 1908 1909 server->retrans_timeo = timeparms.to_initval; 1910 server->retrans_count = timeparms.to_retries; 1911 1912 clp = nfs4_get_client(&server->addr.sin_addr); 1913 if (!clp) { 1914 dprintk("%s: failed to create NFS4 client.\n", __FUNCTION__); 1915 return -EIO; 1916 } 1917 1918 /* Now create transport and client */ 1919 authflavour = RPC_AUTH_UNIX; 1920 if (data->auth_flavourlen != 0) { 1921 if (data->auth_flavourlen != 1) { 1922 dprintk("%s: Invalid number of RPC auth flavours %d.\n", 1923 __FUNCTION__, data->auth_flavourlen); 1924 err = -EINVAL; 1925 goto out_fail; 1926 } 1927 if (copy_from_user(&authflavour, data->auth_flavours, sizeof(authflavour))) { 1928 err = -EFAULT; 1929 goto out_fail; 1930 } 1931 } 1932 1933 down_write(&clp->cl_sem); 1934 if (IS_ERR(clp->cl_rpcclient)) { 1935 xprt = xprt_create_proto(data->proto, &server->addr, &timeparms); 1936 if (IS_ERR(xprt)) { 1937 up_write(&clp->cl_sem); 1938 err = PTR_ERR(xprt); 1939 dprintk("%s: cannot create RPC transport. Error = %d\n", 1940 __FUNCTION__, err); 1941 goto out_fail; 1942 } 1943 clnt = rpc_create_client(xprt, server->hostname, &nfs_program, 1944 server->rpc_ops->version, authflavour); 1945 if (IS_ERR(clnt)) { 1946 up_write(&clp->cl_sem); 1947 err = PTR_ERR(clnt); 1948 dprintk("%s: cannot create RPC client. Error = %d\n", 1949 __FUNCTION__, err); 1950 goto out_fail; 1951 } 1952 clnt->cl_intr = 1; 1953 clnt->cl_softrtry = 1; 1954 clp->cl_rpcclient = clnt; 1955 memcpy(clp->cl_ipaddr, server->ip_addr, sizeof(clp->cl_ipaddr)); 1956 nfs_idmap_new(clp); 1957 } 1958 list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks); 1959 clnt = rpc_clone_client(clp->cl_rpcclient); 1960 if (!IS_ERR(clnt)) 1961 server->nfs4_state = clp; 1962 up_write(&clp->cl_sem); 1963 clp = NULL; 1964 1965 if (IS_ERR(clnt)) { 1966 err = PTR_ERR(clnt); 1967 dprintk("%s: cannot create RPC client. Error = %d\n", 1968 __FUNCTION__, err); 1969 return err; 1970 } 1971 1972 server->client = clnt; 1973 1974 if (server->nfs4_state->cl_idmap == NULL) { 1975 dprintk("%s: failed to create idmapper.\n", __FUNCTION__); 1976 return -ENOMEM; 1977 } 1978 1979 if (clnt->cl_auth->au_flavor != authflavour) { 1980 struct rpc_auth *auth; 1981 1982 auth = rpcauth_create(authflavour, clnt); 1983 if (IS_ERR(auth)) { 1984 dprintk("%s: couldn't create credcache!\n", __FUNCTION__); 1985 return PTR_ERR(auth); 1986 } 1987 } 1988 1989 sb->s_time_gran = 1; 1990 1991 sb->s_op = &nfs4_sops; 1992 err = nfs_sb_init(sb, authflavour); 1993 if (err == 0) 1994 return 0; 1995 out_fail: 1996 if (clp) 1997 nfs4_put_client(clp); 1998 return err; 1999 } 2000 2001 static int nfs4_compare_super(struct super_block *sb, void *data) 2002 { 2003 struct nfs_server *server = data; 2004 struct nfs_server *old = NFS_SB(sb); 2005 2006 if (strcmp(server->hostname, old->hostname) != 0) 2007 return 0; 2008 if (strcmp(server->mnt_path, old->mnt_path) != 0) 2009 return 0; 2010 return 1; 2011 } 2012 2013 static void * 2014 nfs_copy_user_string(char *dst, struct nfs_string *src, int maxlen) 2015 { 2016 void *p = NULL; 2017 2018 if (!src->len) 2019 return ERR_PTR(-EINVAL); 2020 if (src->len < maxlen) 2021 maxlen = src->len; 2022 if (dst == NULL) { 2023 p = dst = kmalloc(maxlen + 1, GFP_KERNEL); 2024 if (p == NULL) 2025 return ERR_PTR(-ENOMEM); 2026 } 2027 if (copy_from_user(dst, src->data, maxlen)) { 2028 kfree(p); 2029 return ERR_PTR(-EFAULT); 2030 } 2031 dst[maxlen] = '\0'; 2032 return dst; 2033 } 2034 2035 static struct super_block *nfs4_get_sb(struct file_system_type *fs_type, 2036 int flags, const char *dev_name, void *raw_data) 2037 { 2038 int error; 2039 struct nfs_server *server; 2040 struct super_block *s; 2041 struct nfs4_mount_data *data = raw_data; 2042 void *p; 2043 2044 if (data == NULL) { 2045 dprintk("%s: missing data argument\n", __FUNCTION__); 2046 return ERR_PTR(-EINVAL); 2047 } 2048 if (data->version <= 0 || data->version > NFS4_MOUNT_VERSION) { 2049 dprintk("%s: bad mount version\n", __FUNCTION__); 2050 return ERR_PTR(-EINVAL); 2051 } 2052 2053 server = kzalloc(sizeof(struct nfs_server), GFP_KERNEL); 2054 if (!server) 2055 return ERR_PTR(-ENOMEM); 2056 /* Zero out the NFS state stuff */ 2057 init_nfsv4_state(server); 2058 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL); 2059 2060 p = nfs_copy_user_string(NULL, &data->hostname, 256); 2061 if (IS_ERR(p)) 2062 goto out_err; 2063 server->hostname = p; 2064 2065 p = nfs_copy_user_string(NULL, &data->mnt_path, 1024); 2066 if (IS_ERR(p)) 2067 goto out_err; 2068 server->mnt_path = p; 2069 2070 p = nfs_copy_user_string(server->ip_addr, &data->client_addr, 2071 sizeof(server->ip_addr) - 1); 2072 if (IS_ERR(p)) 2073 goto out_err; 2074 2075 /* We now require that the mount process passes the remote address */ 2076 if (data->host_addrlen != sizeof(server->addr)) { 2077 s = ERR_PTR(-EINVAL); 2078 goto out_free; 2079 } 2080 if (copy_from_user(&server->addr, data->host_addr, sizeof(server->addr))) { 2081 s = ERR_PTR(-EFAULT); 2082 goto out_free; 2083 } 2084 if (server->addr.sin_family != AF_INET || 2085 server->addr.sin_addr.s_addr == INADDR_ANY) { 2086 dprintk("%s: mount program didn't pass remote IP address!\n", 2087 __FUNCTION__); 2088 s = ERR_PTR(-EINVAL); 2089 goto out_free; 2090 } 2091 2092 /* Fire up rpciod if not yet running */ 2093 s = ERR_PTR(rpciod_up()); 2094 if (IS_ERR(s)) { 2095 dprintk("%s: couldn't start rpciod! Error = %ld\n", 2096 __FUNCTION__, PTR_ERR(s)); 2097 goto out_free; 2098 } 2099 2100 s = sget(fs_type, nfs4_compare_super, nfs_set_super, server); 2101 2102 if (IS_ERR(s) || s->s_root) 2103 goto out_free; 2104 2105 s->s_flags = flags; 2106 2107 error = nfs4_fill_super(s, data, flags & MS_SILENT ? 1 : 0); 2108 if (error) { 2109 up_write(&s->s_umount); 2110 deactivate_super(s); 2111 return ERR_PTR(error); 2112 } 2113 s->s_flags |= MS_ACTIVE; 2114 return s; 2115 out_err: 2116 s = (struct super_block *)p; 2117 out_free: 2118 kfree(server->mnt_path); 2119 kfree(server->hostname); 2120 kfree(server); 2121 return s; 2122 } 2123 2124 static void nfs4_kill_super(struct super_block *sb) 2125 { 2126 struct nfs_server *server = NFS_SB(sb); 2127 2128 nfs_return_all_delegations(sb); 2129 kill_anon_super(sb); 2130 2131 nfs4_renewd_prepare_shutdown(server); 2132 2133 if (server->client != NULL && !IS_ERR(server->client)) 2134 rpc_shutdown_client(server->client); 2135 2136 destroy_nfsv4_state(server); 2137 2138 rpciod_down(); 2139 2140 nfs_free_iostats(server->io_stats); 2141 kfree(server->hostname); 2142 kfree(server); 2143 } 2144 2145 static struct file_system_type nfs4_fs_type = { 2146 .owner = THIS_MODULE, 2147 .name = "nfs4", 2148 .get_sb = nfs4_get_sb, 2149 .kill_sb = nfs4_kill_super, 2150 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA, 2151 }; 2152 2153 static const int nfs_set_port_min = 0; 2154 static const int nfs_set_port_max = 65535; 2155 static int param_set_port(const char *val, struct kernel_param *kp) 2156 { 2157 char *endp; 2158 int num = simple_strtol(val, &endp, 0); 2159 if (endp == val || *endp || num < nfs_set_port_min || num > nfs_set_port_max) 2160 return -EINVAL; 2161 *((int *)kp->arg) = num; 2162 return 0; 2163 } 2164 2165 module_param_call(callback_tcpport, param_set_port, param_get_int, 2166 &nfs_callback_set_tcpport, 0644); 2167 2168 static int param_set_idmap_timeout(const char *val, struct kernel_param *kp) 2169 { 2170 char *endp; 2171 int num = simple_strtol(val, &endp, 0); 2172 int jif = num * HZ; 2173 if (endp == val || *endp || num < 0 || jif < num) 2174 return -EINVAL; 2175 *((int *)kp->arg) = jif; 2176 return 0; 2177 } 2178 2179 module_param_call(idmap_cache_timeout, param_set_idmap_timeout, param_get_int, 2180 &nfs_idmap_cache_timeout, 0644); 2181 2182 #define nfs4_init_once(nfsi) \ 2183 do { \ 2184 INIT_LIST_HEAD(&(nfsi)->open_states); \ 2185 nfsi->delegation = NULL; \ 2186 nfsi->delegation_state = 0; \ 2187 init_rwsem(&nfsi->rwsem); \ 2188 } while(0) 2189 2190 static inline int register_nfs4fs(void) 2191 { 2192 int ret; 2193 2194 ret = nfs_register_sysctl(); 2195 if (ret != 0) 2196 return ret; 2197 ret = register_filesystem(&nfs4_fs_type); 2198 if (ret != 0) 2199 nfs_unregister_sysctl(); 2200 return ret; 2201 } 2202 2203 static inline void unregister_nfs4fs(void) 2204 { 2205 unregister_filesystem(&nfs4_fs_type); 2206 nfs_unregister_sysctl(); 2207 } 2208 #else 2209 #define nfs4_init_once(nfsi) \ 2210 do { } while (0) 2211 #define register_nfs4fs() (0) 2212 #define unregister_nfs4fs() 2213 #endif 2214 2215 extern int nfs_init_nfspagecache(void); 2216 extern void nfs_destroy_nfspagecache(void); 2217 extern int nfs_init_readpagecache(void); 2218 extern void nfs_destroy_readpagecache(void); 2219 extern int nfs_init_writepagecache(void); 2220 extern void nfs_destroy_writepagecache(void); 2221 #ifdef CONFIG_NFS_DIRECTIO 2222 extern int nfs_init_directcache(void); 2223 extern void nfs_destroy_directcache(void); 2224 #endif 2225 2226 static kmem_cache_t * nfs_inode_cachep; 2227 2228 static struct inode *nfs_alloc_inode(struct super_block *sb) 2229 { 2230 struct nfs_inode *nfsi; 2231 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL); 2232 if (!nfsi) 2233 return NULL; 2234 nfsi->flags = 0UL; 2235 nfsi->cache_validity = 0UL; 2236 nfsi->cache_change_attribute = jiffies; 2237 #ifdef CONFIG_NFS_V3_ACL 2238 nfsi->acl_access = ERR_PTR(-EAGAIN); 2239 nfsi->acl_default = ERR_PTR(-EAGAIN); 2240 #endif 2241 #ifdef CONFIG_NFS_V4 2242 nfsi->nfs4_acl = NULL; 2243 #endif /* CONFIG_NFS_V4 */ 2244 return &nfsi->vfs_inode; 2245 } 2246 2247 static void nfs_destroy_inode(struct inode *inode) 2248 { 2249 kmem_cache_free(nfs_inode_cachep, NFS_I(inode)); 2250 } 2251 2252 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags) 2253 { 2254 struct nfs_inode *nfsi = (struct nfs_inode *) foo; 2255 2256 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) == 2257 SLAB_CTOR_CONSTRUCTOR) { 2258 inode_init_once(&nfsi->vfs_inode); 2259 spin_lock_init(&nfsi->req_lock); 2260 INIT_LIST_HEAD(&nfsi->dirty); 2261 INIT_LIST_HEAD(&nfsi->commit); 2262 INIT_LIST_HEAD(&nfsi->open_files); 2263 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC); 2264 atomic_set(&nfsi->data_updates, 0); 2265 nfsi->ndirty = 0; 2266 nfsi->ncommit = 0; 2267 nfsi->npages = 0; 2268 nfs4_init_once(nfsi); 2269 } 2270 } 2271 2272 static int nfs_init_inodecache(void) 2273 { 2274 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache", 2275 sizeof(struct nfs_inode), 2276 0, (SLAB_RECLAIM_ACCOUNT| 2277 SLAB_MEM_SPREAD), 2278 init_once, NULL); 2279 if (nfs_inode_cachep == NULL) 2280 return -ENOMEM; 2281 2282 return 0; 2283 } 2284 2285 static void nfs_destroy_inodecache(void) 2286 { 2287 if (kmem_cache_destroy(nfs_inode_cachep)) 2288 printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n"); 2289 } 2290 2291 /* 2292 * Initialize NFS 2293 */ 2294 static int __init init_nfs_fs(void) 2295 { 2296 int err; 2297 2298 err = nfs_init_nfspagecache(); 2299 if (err) 2300 goto out4; 2301 2302 err = nfs_init_inodecache(); 2303 if (err) 2304 goto out3; 2305 2306 err = nfs_init_readpagecache(); 2307 if (err) 2308 goto out2; 2309 2310 err = nfs_init_writepagecache(); 2311 if (err) 2312 goto out1; 2313 2314 #ifdef CONFIG_NFS_DIRECTIO 2315 err = nfs_init_directcache(); 2316 if (err) 2317 goto out0; 2318 #endif 2319 2320 #ifdef CONFIG_PROC_FS 2321 rpc_proc_register(&nfs_rpcstat); 2322 #endif 2323 err = register_filesystem(&nfs_fs_type); 2324 if (err) 2325 goto out; 2326 if ((err = register_nfs4fs()) != 0) 2327 goto out; 2328 return 0; 2329 out: 2330 #ifdef CONFIG_PROC_FS 2331 rpc_proc_unregister("nfs"); 2332 #endif 2333 #ifdef CONFIG_NFS_DIRECTIO 2334 nfs_destroy_directcache(); 2335 out0: 2336 #endif 2337 nfs_destroy_writepagecache(); 2338 out1: 2339 nfs_destroy_readpagecache(); 2340 out2: 2341 nfs_destroy_inodecache(); 2342 out3: 2343 nfs_destroy_nfspagecache(); 2344 out4: 2345 return err; 2346 } 2347 2348 static void __exit exit_nfs_fs(void) 2349 { 2350 #ifdef CONFIG_NFS_DIRECTIO 2351 nfs_destroy_directcache(); 2352 #endif 2353 nfs_destroy_writepagecache(); 2354 nfs_destroy_readpagecache(); 2355 nfs_destroy_inodecache(); 2356 nfs_destroy_nfspagecache(); 2357 #ifdef CONFIG_PROC_FS 2358 rpc_proc_unregister("nfs"); 2359 #endif 2360 unregister_filesystem(&nfs_fs_type); 2361 unregister_nfs4fs(); 2362 } 2363 2364 /* Not quite true; I just maintain it */ 2365 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>"); 2366 MODULE_LICENSE("GPL"); 2367 2368 module_init(init_nfs_fs) 2369 module_exit(exit_nfs_fs) 2370