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 (cpu = 0; cpu < NR_CPUS; cpu++) { 704 struct nfs_iostats *stats; 705 706 if (!cpu_possible(cpu)) 707 continue; 708 709 preempt_disable(); 710 stats = per_cpu_ptr(nfss->io_stats, cpu); 711 712 for (i = 0; i < __NFSIOS_COUNTSMAX; i++) 713 totals.events[i] += stats->events[i]; 714 for (i = 0; i < __NFSIOS_BYTESMAX; i++) 715 totals.bytes[i] += stats->bytes[i]; 716 717 preempt_enable(); 718 } 719 720 seq_printf(m, "\n\tevents:\t"); 721 for (i = 0; i < __NFSIOS_COUNTSMAX; i++) 722 seq_printf(m, "%lu ", totals.events[i]); 723 seq_printf(m, "\n\tbytes:\t"); 724 for (i = 0; i < __NFSIOS_BYTESMAX; i++) 725 seq_printf(m, "%Lu ", totals.bytes[i]); 726 seq_printf(m, "\n"); 727 728 rpc_print_iostats(m, nfss->client); 729 730 return 0; 731 } 732 733 /** 734 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk 735 */ 736 int nfs_sync_mapping(struct address_space *mapping) 737 { 738 int ret; 739 740 if (mapping->nrpages == 0) 741 return 0; 742 unmap_mapping_range(mapping, 0, 0, 0); 743 ret = filemap_write_and_wait(mapping); 744 if (ret != 0) 745 goto out; 746 ret = nfs_wb_all(mapping->host); 747 out: 748 return ret; 749 } 750 751 /* 752 * Invalidate the local caches 753 */ 754 static void nfs_zap_caches_locked(struct inode *inode) 755 { 756 struct nfs_inode *nfsi = NFS_I(inode); 757 int mode = inode->i_mode; 758 759 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 760 761 NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode); 762 NFS_ATTRTIMEO_UPDATE(inode) = jiffies; 763 764 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode))); 765 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) 766 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE; 767 else 768 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE; 769 } 770 771 void nfs_zap_caches(struct inode *inode) 772 { 773 spin_lock(&inode->i_lock); 774 nfs_zap_caches_locked(inode); 775 spin_unlock(&inode->i_lock); 776 } 777 778 static void nfs_zap_acl_cache(struct inode *inode) 779 { 780 void (*clear_acl_cache)(struct inode *); 781 782 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache; 783 if (clear_acl_cache != NULL) 784 clear_acl_cache(inode); 785 spin_lock(&inode->i_lock); 786 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL; 787 spin_unlock(&inode->i_lock); 788 } 789 790 /* 791 * Invalidate, but do not unhash, the inode. 792 * NB: must be called with inode->i_lock held! 793 */ 794 static void nfs_invalidate_inode(struct inode *inode) 795 { 796 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode)); 797 nfs_zap_caches_locked(inode); 798 } 799 800 struct nfs_find_desc { 801 struct nfs_fh *fh; 802 struct nfs_fattr *fattr; 803 }; 804 805 /* 806 * In NFSv3 we can have 64bit inode numbers. In order to support 807 * this, and re-exported directories (also seen in NFSv2) 808 * we are forced to allow 2 different inodes to have the same 809 * i_ino. 810 */ 811 static int 812 nfs_find_actor(struct inode *inode, void *opaque) 813 { 814 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque; 815 struct nfs_fh *fh = desc->fh; 816 struct nfs_fattr *fattr = desc->fattr; 817 818 if (NFS_FILEID(inode) != fattr->fileid) 819 return 0; 820 if (nfs_compare_fh(NFS_FH(inode), fh)) 821 return 0; 822 if (is_bad_inode(inode) || NFS_STALE(inode)) 823 return 0; 824 return 1; 825 } 826 827 static int 828 nfs_init_locked(struct inode *inode, void *opaque) 829 { 830 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque; 831 struct nfs_fattr *fattr = desc->fattr; 832 833 NFS_FILEID(inode) = fattr->fileid; 834 nfs_copy_fh(NFS_FH(inode), desc->fh); 835 return 0; 836 } 837 838 /* Don't use READDIRPLUS on directories that we believe are too large */ 839 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE) 840 841 /* 842 * This is our front-end to iget that looks up inodes by file handle 843 * instead of inode number. 844 */ 845 struct inode * 846 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr) 847 { 848 struct nfs_find_desc desc = { 849 .fh = fh, 850 .fattr = fattr 851 }; 852 struct inode *inode = ERR_PTR(-ENOENT); 853 unsigned long hash; 854 855 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 856 goto out_no_inode; 857 858 if (!fattr->nlink) { 859 printk("NFS: Buggy server - nlink == 0!\n"); 860 goto out_no_inode; 861 } 862 863 hash = nfs_fattr_to_ino_t(fattr); 864 865 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc); 866 if (inode == NULL) { 867 inode = ERR_PTR(-ENOMEM); 868 goto out_no_inode; 869 } 870 871 if (inode->i_state & I_NEW) { 872 struct nfs_inode *nfsi = NFS_I(inode); 873 874 /* We set i_ino for the few things that still rely on it, 875 * such as stat(2) */ 876 inode->i_ino = hash; 877 878 /* We can't support update_atime(), since the server will reset it */ 879 inode->i_flags |= S_NOATIME|S_NOCMTIME; 880 inode->i_mode = fattr->mode; 881 /* Why so? Because we want revalidate for devices/FIFOs, and 882 * that's precisely what we have in nfs_file_inode_operations. 883 */ 884 inode->i_op = NFS_SB(sb)->rpc_ops->file_inode_ops; 885 if (S_ISREG(inode->i_mode)) { 886 inode->i_fop = &nfs_file_operations; 887 inode->i_data.a_ops = &nfs_file_aops; 888 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info; 889 } else if (S_ISDIR(inode->i_mode)) { 890 inode->i_op = NFS_SB(sb)->rpc_ops->dir_inode_ops; 891 inode->i_fop = &nfs_dir_operations; 892 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS) 893 && fattr->size <= NFS_LIMIT_READDIRPLUS) 894 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode)); 895 } else if (S_ISLNK(inode->i_mode)) 896 inode->i_op = &nfs_symlink_inode_operations; 897 else 898 init_special_inode(inode, inode->i_mode, fattr->rdev); 899 900 nfsi->read_cache_jiffies = fattr->time_start; 901 nfsi->last_updated = jiffies; 902 inode->i_atime = fattr->atime; 903 inode->i_mtime = fattr->mtime; 904 inode->i_ctime = fattr->ctime; 905 if (fattr->valid & NFS_ATTR_FATTR_V4) 906 nfsi->change_attr = fattr->change_attr; 907 inode->i_size = nfs_size_to_loff_t(fattr->size); 908 inode->i_nlink = fattr->nlink; 909 inode->i_uid = fattr->uid; 910 inode->i_gid = fattr->gid; 911 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) { 912 /* 913 * report the blocks in 512byte units 914 */ 915 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 916 inode->i_blksize = inode->i_sb->s_blocksize; 917 } else { 918 inode->i_blocks = fattr->du.nfs2.blocks; 919 inode->i_blksize = fattr->du.nfs2.blocksize; 920 } 921 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 922 nfsi->attrtimeo_timestamp = jiffies; 923 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf)); 924 nfsi->cache_access.cred = NULL; 925 926 unlock_new_inode(inode); 927 } else 928 nfs_refresh_inode(inode, fattr); 929 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n", 930 inode->i_sb->s_id, 931 (long long)NFS_FILEID(inode), 932 atomic_read(&inode->i_count)); 933 934 out: 935 return inode; 936 937 out_no_inode: 938 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode)); 939 goto out; 940 } 941 942 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET) 943 944 int 945 nfs_setattr(struct dentry *dentry, struct iattr *attr) 946 { 947 struct inode *inode = dentry->d_inode; 948 struct nfs_fattr fattr; 949 int error; 950 951 nfs_inc_stats(inode, NFSIOS_VFSSETATTR); 952 953 if (attr->ia_valid & ATTR_SIZE) { 954 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode)) 955 attr->ia_valid &= ~ATTR_SIZE; 956 } 957 958 /* Optimization: if the end result is no change, don't RPC */ 959 attr->ia_valid &= NFS_VALID_ATTRS; 960 if (attr->ia_valid == 0) 961 return 0; 962 963 lock_kernel(); 964 nfs_begin_data_update(inode); 965 /* Write all dirty data */ 966 filemap_write_and_wait(inode->i_mapping); 967 nfs_wb_all(inode); 968 /* 969 * Return any delegations if we're going to change ACLs 970 */ 971 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) 972 nfs_inode_return_delegation(inode); 973 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr); 974 if (error == 0) 975 nfs_refresh_inode(inode, &fattr); 976 nfs_end_data_update(inode); 977 unlock_kernel(); 978 return error; 979 } 980 981 /** 982 * nfs_setattr_update_inode - Update inode metadata after a setattr call. 983 * @inode: pointer to struct inode 984 * @attr: pointer to struct iattr 985 * 986 * Note: we do this in the *proc.c in order to ensure that 987 * it works for things like exclusive creates too. 988 */ 989 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr) 990 { 991 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) { 992 if ((attr->ia_valid & ATTR_MODE) != 0) { 993 int mode = attr->ia_mode & S_IALLUGO; 994 mode |= inode->i_mode & ~S_IALLUGO; 995 inode->i_mode = mode; 996 } 997 if ((attr->ia_valid & ATTR_UID) != 0) 998 inode->i_uid = attr->ia_uid; 999 if ((attr->ia_valid & ATTR_GID) != 0) 1000 inode->i_gid = attr->ia_gid; 1001 spin_lock(&inode->i_lock); 1002 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 1003 spin_unlock(&inode->i_lock); 1004 } 1005 if ((attr->ia_valid & ATTR_SIZE) != 0) { 1006 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC); 1007 inode->i_size = attr->ia_size; 1008 vmtruncate(inode, attr->ia_size); 1009 } 1010 } 1011 1012 static int nfs_wait_schedule(void *word) 1013 { 1014 if (signal_pending(current)) 1015 return -ERESTARTSYS; 1016 schedule(); 1017 return 0; 1018 } 1019 1020 /* 1021 * Wait for the inode to get unlocked. 1022 */ 1023 static int nfs_wait_on_inode(struct inode *inode) 1024 { 1025 struct rpc_clnt *clnt = NFS_CLIENT(inode); 1026 struct nfs_inode *nfsi = NFS_I(inode); 1027 sigset_t oldmask; 1028 int error; 1029 1030 rpc_clnt_sigmask(clnt, &oldmask); 1031 error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING, 1032 nfs_wait_schedule, TASK_INTERRUPTIBLE); 1033 rpc_clnt_sigunmask(clnt, &oldmask); 1034 1035 return error; 1036 } 1037 1038 static void nfs_wake_up_inode(struct inode *inode) 1039 { 1040 struct nfs_inode *nfsi = NFS_I(inode); 1041 1042 clear_bit(NFS_INO_REVALIDATING, &nfsi->flags); 1043 smp_mb__after_clear_bit(); 1044 wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING); 1045 } 1046 1047 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat) 1048 { 1049 struct inode *inode = dentry->d_inode; 1050 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME; 1051 int err; 1052 1053 /* Flush out writes to the server in order to update c/mtime */ 1054 nfs_sync_inode_wait(inode, 0, 0, FLUSH_NOCOMMIT); 1055 1056 /* 1057 * We may force a getattr if the user cares about atime. 1058 * 1059 * Note that we only have to check the vfsmount flags here: 1060 * - NFS always sets S_NOATIME by so checking it would give a 1061 * bogus result 1062 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is 1063 * no point in checking those. 1064 */ 1065 if ((mnt->mnt_flags & MNT_NOATIME) || 1066 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) 1067 need_atime = 0; 1068 1069 if (need_atime) 1070 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 1071 else 1072 err = nfs_revalidate_inode(NFS_SERVER(inode), inode); 1073 if (!err) 1074 generic_fillattr(inode, stat); 1075 return err; 1076 } 1077 1078 static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred) 1079 { 1080 struct nfs_open_context *ctx; 1081 1082 ctx = (struct nfs_open_context *)kmalloc(sizeof(*ctx), GFP_KERNEL); 1083 if (ctx != NULL) { 1084 atomic_set(&ctx->count, 1); 1085 ctx->dentry = dget(dentry); 1086 ctx->vfsmnt = mntget(mnt); 1087 ctx->cred = get_rpccred(cred); 1088 ctx->state = NULL; 1089 ctx->lockowner = current->files; 1090 ctx->error = 0; 1091 ctx->dir_cookie = 0; 1092 } 1093 return ctx; 1094 } 1095 1096 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx) 1097 { 1098 if (ctx != NULL) 1099 atomic_inc(&ctx->count); 1100 return ctx; 1101 } 1102 1103 void put_nfs_open_context(struct nfs_open_context *ctx) 1104 { 1105 if (atomic_dec_and_test(&ctx->count)) { 1106 if (!list_empty(&ctx->list)) { 1107 struct inode *inode = ctx->dentry->d_inode; 1108 spin_lock(&inode->i_lock); 1109 list_del(&ctx->list); 1110 spin_unlock(&inode->i_lock); 1111 } 1112 if (ctx->state != NULL) 1113 nfs4_close_state(ctx->state, ctx->mode); 1114 if (ctx->cred != NULL) 1115 put_rpccred(ctx->cred); 1116 dput(ctx->dentry); 1117 mntput(ctx->vfsmnt); 1118 kfree(ctx); 1119 } 1120 } 1121 1122 /* 1123 * Ensure that mmap has a recent RPC credential for use when writing out 1124 * shared pages 1125 */ 1126 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx) 1127 { 1128 struct inode *inode = filp->f_dentry->d_inode; 1129 struct nfs_inode *nfsi = NFS_I(inode); 1130 1131 filp->private_data = get_nfs_open_context(ctx); 1132 spin_lock(&inode->i_lock); 1133 list_add(&ctx->list, &nfsi->open_files); 1134 spin_unlock(&inode->i_lock); 1135 } 1136 1137 /* 1138 * Given an inode, search for an open context with the desired characteristics 1139 */ 1140 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode) 1141 { 1142 struct nfs_inode *nfsi = NFS_I(inode); 1143 struct nfs_open_context *pos, *ctx = NULL; 1144 1145 spin_lock(&inode->i_lock); 1146 list_for_each_entry(pos, &nfsi->open_files, list) { 1147 if (cred != NULL && pos->cred != cred) 1148 continue; 1149 if ((pos->mode & mode) == mode) { 1150 ctx = get_nfs_open_context(pos); 1151 break; 1152 } 1153 } 1154 spin_unlock(&inode->i_lock); 1155 return ctx; 1156 } 1157 1158 static void nfs_file_clear_open_context(struct file *filp) 1159 { 1160 struct inode *inode = filp->f_dentry->d_inode; 1161 struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data; 1162 1163 if (ctx) { 1164 filp->private_data = NULL; 1165 spin_lock(&inode->i_lock); 1166 list_move_tail(&ctx->list, &NFS_I(inode)->open_files); 1167 spin_unlock(&inode->i_lock); 1168 put_nfs_open_context(ctx); 1169 } 1170 } 1171 1172 /* 1173 * These allocate and release file read/write context information. 1174 */ 1175 int nfs_open(struct inode *inode, struct file *filp) 1176 { 1177 struct nfs_open_context *ctx; 1178 struct rpc_cred *cred; 1179 1180 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0); 1181 if (IS_ERR(cred)) 1182 return PTR_ERR(cred); 1183 ctx = alloc_nfs_open_context(filp->f_vfsmnt, filp->f_dentry, cred); 1184 put_rpccred(cred); 1185 if (ctx == NULL) 1186 return -ENOMEM; 1187 ctx->mode = filp->f_mode; 1188 nfs_file_set_open_context(filp, ctx); 1189 put_nfs_open_context(ctx); 1190 return 0; 1191 } 1192 1193 int nfs_release(struct inode *inode, struct file *filp) 1194 { 1195 nfs_file_clear_open_context(filp); 1196 return 0; 1197 } 1198 1199 /* 1200 * This function is called whenever some part of NFS notices that 1201 * the cached attributes have to be refreshed. 1202 */ 1203 int 1204 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) 1205 { 1206 int status = -ESTALE; 1207 struct nfs_fattr fattr; 1208 struct nfs_inode *nfsi = NFS_I(inode); 1209 1210 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n", 1211 inode->i_sb->s_id, (long long)NFS_FILEID(inode)); 1212 1213 lock_kernel(); 1214 if (!inode || is_bad_inode(inode)) 1215 goto out_nowait; 1216 if (NFS_STALE(inode)) 1217 goto out_nowait; 1218 1219 status = nfs_wait_on_inode(inode); 1220 if (status < 0) 1221 goto out; 1222 if (NFS_STALE(inode)) { 1223 status = -ESTALE; 1224 /* Do we trust the cached ESTALE? */ 1225 if (NFS_ATTRTIMEO(inode) != 0) { 1226 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ATIME)) { 1227 /* no */ 1228 } else 1229 goto out; 1230 } 1231 } 1232 1233 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr); 1234 if (status != 0) { 1235 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n", 1236 inode->i_sb->s_id, 1237 (long long)NFS_FILEID(inode), status); 1238 if (status == -ESTALE) { 1239 nfs_zap_caches(inode); 1240 if (!S_ISDIR(inode->i_mode)) 1241 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode)); 1242 } 1243 goto out; 1244 } 1245 1246 spin_lock(&inode->i_lock); 1247 status = nfs_update_inode(inode, &fattr); 1248 if (status) { 1249 spin_unlock(&inode->i_lock); 1250 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n", 1251 inode->i_sb->s_id, 1252 (long long)NFS_FILEID(inode), status); 1253 goto out; 1254 } 1255 spin_unlock(&inode->i_lock); 1256 1257 nfs_revalidate_mapping(inode, inode->i_mapping); 1258 1259 if (nfsi->cache_validity & NFS_INO_INVALID_ACL) 1260 nfs_zap_acl_cache(inode); 1261 1262 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n", 1263 inode->i_sb->s_id, 1264 (long long)NFS_FILEID(inode)); 1265 1266 out: 1267 nfs_wake_up_inode(inode); 1268 1269 out_nowait: 1270 unlock_kernel(); 1271 return status; 1272 } 1273 1274 int nfs_attribute_timeout(struct inode *inode) 1275 { 1276 struct nfs_inode *nfsi = NFS_I(inode); 1277 1278 if (nfs_have_delegation(inode, FMODE_READ)) 1279 return 0; 1280 return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo); 1281 } 1282 1283 /** 1284 * nfs_revalidate_inode - Revalidate the inode attributes 1285 * @server - pointer to nfs_server struct 1286 * @inode - pointer to inode struct 1287 * 1288 * Updates inode attribute information by retrieving the data from the server. 1289 */ 1290 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) 1291 { 1292 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE); 1293 if (!(NFS_I(inode)->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA)) 1294 && !nfs_attribute_timeout(inode)) 1295 return NFS_STALE(inode) ? -ESTALE : 0; 1296 return __nfs_revalidate_inode(server, inode); 1297 } 1298 1299 /** 1300 * nfs_revalidate_mapping - Revalidate the pagecache 1301 * @inode - pointer to host inode 1302 * @mapping - pointer to mapping 1303 */ 1304 void nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping) 1305 { 1306 struct nfs_inode *nfsi = NFS_I(inode); 1307 1308 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) { 1309 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE); 1310 if (S_ISREG(inode->i_mode)) 1311 nfs_sync_mapping(mapping); 1312 invalidate_inode_pages2(mapping); 1313 1314 spin_lock(&inode->i_lock); 1315 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA; 1316 if (S_ISDIR(inode->i_mode)) { 1317 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf)); 1318 /* This ensures we revalidate child dentries */ 1319 nfsi->cache_change_attribute = jiffies; 1320 } 1321 spin_unlock(&inode->i_lock); 1322 1323 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n", 1324 inode->i_sb->s_id, 1325 (long long)NFS_FILEID(inode)); 1326 } 1327 } 1328 1329 /** 1330 * nfs_begin_data_update 1331 * @inode - pointer to inode 1332 * Declare that a set of operations will update file data on the server 1333 */ 1334 void nfs_begin_data_update(struct inode *inode) 1335 { 1336 atomic_inc(&NFS_I(inode)->data_updates); 1337 } 1338 1339 /** 1340 * nfs_end_data_update 1341 * @inode - pointer to inode 1342 * Declare end of the operations that will update file data 1343 * This will mark the inode as immediately needing revalidation 1344 * of its attribute cache. 1345 */ 1346 void nfs_end_data_update(struct inode *inode) 1347 { 1348 struct nfs_inode *nfsi = NFS_I(inode); 1349 1350 if (!nfs_have_delegation(inode, FMODE_READ)) { 1351 /* Directories and symlinks: invalidate page cache */ 1352 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) { 1353 spin_lock(&inode->i_lock); 1354 nfsi->cache_validity |= NFS_INO_INVALID_DATA; 1355 spin_unlock(&inode->i_lock); 1356 } 1357 } 1358 nfsi->cache_change_attribute = jiffies; 1359 atomic_dec(&nfsi->data_updates); 1360 } 1361 1362 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1363 { 1364 struct nfs_inode *nfsi = NFS_I(inode); 1365 1366 if ((fattr->valid & NFS_ATTR_PRE_CHANGE) != 0 1367 && nfsi->change_attr == fattr->pre_change_attr) { 1368 nfsi->change_attr = fattr->change_attr; 1369 nfsi->cache_change_attribute = jiffies; 1370 } 1371 1372 /* If we have atomic WCC data, we may update some attributes */ 1373 if ((fattr->valid & NFS_ATTR_WCC) != 0) { 1374 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) { 1375 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime)); 1376 nfsi->cache_change_attribute = jiffies; 1377 } 1378 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) { 1379 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime)); 1380 nfsi->cache_change_attribute = jiffies; 1381 } 1382 if (inode->i_size == fattr->pre_size && nfsi->npages == 0) { 1383 inode->i_size = fattr->size; 1384 nfsi->cache_change_attribute = jiffies; 1385 } 1386 } 1387 } 1388 1389 /** 1390 * nfs_check_inode_attributes - verify consistency of the inode attribute cache 1391 * @inode - pointer to inode 1392 * @fattr - updated attributes 1393 * 1394 * Verifies the attribute cache. If we have just changed the attributes, 1395 * so that fattr carries weak cache consistency data, then it may 1396 * also update the ctime/mtime/change_attribute. 1397 */ 1398 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr) 1399 { 1400 struct nfs_inode *nfsi = NFS_I(inode); 1401 loff_t cur_size, new_isize; 1402 int data_unstable; 1403 1404 1405 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 1406 return 0; 1407 1408 /* Has the inode gone and changed behind our back? */ 1409 if (nfsi->fileid != fattr->fileid 1410 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) { 1411 return -EIO; 1412 } 1413 1414 /* Are we in the process of updating data on the server? */ 1415 data_unstable = nfs_caches_unstable(inode); 1416 1417 /* Do atomic weak cache consistency updates */ 1418 nfs_wcc_update_inode(inode, fattr); 1419 1420 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0) { 1421 if (nfsi->change_attr == fattr->change_attr) 1422 goto out; 1423 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 1424 if (!data_unstable) 1425 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE; 1426 } 1427 1428 /* Verify a few of the more important attributes */ 1429 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) { 1430 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 1431 if (!data_unstable) 1432 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE; 1433 } 1434 1435 cur_size = i_size_read(inode); 1436 new_isize = nfs_size_to_loff_t(fattr->size); 1437 if (cur_size != new_isize && nfsi->npages == 0) 1438 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE; 1439 1440 /* Have any file permissions changed? */ 1441 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) 1442 || inode->i_uid != fattr->uid 1443 || inode->i_gid != fattr->gid) 1444 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL; 1445 1446 /* Has the link count changed? */ 1447 if (inode->i_nlink != fattr->nlink) 1448 nfsi->cache_validity |= NFS_INO_INVALID_ATTR; 1449 1450 out: 1451 if (!timespec_equal(&inode->i_atime, &fattr->atime)) 1452 nfsi->cache_validity |= NFS_INO_INVALID_ATIME; 1453 1454 nfsi->read_cache_jiffies = fattr->time_start; 1455 return 0; 1456 } 1457 1458 /** 1459 * nfs_refresh_inode - try to update the inode attribute cache 1460 * @inode - pointer to inode 1461 * @fattr - updated attributes 1462 * 1463 * Check that an RPC call that returned attributes has not overlapped with 1464 * other recent updates of the inode metadata, then decide whether it is 1465 * safe to do a full update of the inode attributes, or whether just to 1466 * call nfs_check_inode_attributes. 1467 */ 1468 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr) 1469 { 1470 struct nfs_inode *nfsi = NFS_I(inode); 1471 int status; 1472 1473 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 1474 return 0; 1475 spin_lock(&inode->i_lock); 1476 nfsi->cache_validity &= ~NFS_INO_REVAL_PAGECACHE; 1477 if (time_after(fattr->time_start, nfsi->last_updated)) 1478 status = nfs_update_inode(inode, fattr); 1479 else 1480 status = nfs_check_inode_attributes(inode, fattr); 1481 1482 spin_unlock(&inode->i_lock); 1483 return status; 1484 } 1485 1486 /** 1487 * nfs_post_op_update_inode - try to update the inode attribute cache 1488 * @inode - pointer to inode 1489 * @fattr - updated attributes 1490 * 1491 * After an operation that has changed the inode metadata, mark the 1492 * attribute cache as being invalid, then try to update it. 1493 */ 1494 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1495 { 1496 struct nfs_inode *nfsi = NFS_I(inode); 1497 int status = 0; 1498 1499 spin_lock(&inode->i_lock); 1500 if (unlikely((fattr->valid & NFS_ATTR_FATTR) == 0)) { 1501 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS; 1502 goto out; 1503 } 1504 status = nfs_update_inode(inode, fattr); 1505 out: 1506 spin_unlock(&inode->i_lock); 1507 return status; 1508 } 1509 1510 /* 1511 * Many nfs protocol calls return the new file attributes after 1512 * an operation. Here we update the inode to reflect the state 1513 * of the server's inode. 1514 * 1515 * This is a bit tricky because we have to make sure all dirty pages 1516 * have been sent off to the server before calling invalidate_inode_pages. 1517 * To make sure no other process adds more write requests while we try 1518 * our best to flush them, we make them sleep during the attribute refresh. 1519 * 1520 * A very similar scenario holds for the dir cache. 1521 */ 1522 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr) 1523 { 1524 struct nfs_inode *nfsi = NFS_I(inode); 1525 loff_t cur_isize, new_isize; 1526 unsigned int invalid = 0; 1527 int data_stable; 1528 1529 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n", 1530 __FUNCTION__, inode->i_sb->s_id, inode->i_ino, 1531 atomic_read(&inode->i_count), fattr->valid); 1532 1533 if ((fattr->valid & NFS_ATTR_FATTR) == 0) 1534 return 0; 1535 1536 if (nfsi->fileid != fattr->fileid) 1537 goto out_fileid; 1538 1539 /* 1540 * Make sure the inode's type hasn't changed. 1541 */ 1542 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) 1543 goto out_changed; 1544 1545 /* 1546 * Update the read time so we don't revalidate too often. 1547 */ 1548 nfsi->read_cache_jiffies = fattr->time_start; 1549 nfsi->last_updated = jiffies; 1550 1551 /* Are we racing with known updates of the metadata on the server? */ 1552 data_stable = nfs_verify_change_attribute(inode, fattr->time_start); 1553 if (data_stable) 1554 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME); 1555 1556 /* Do atomic weak cache consistency updates */ 1557 nfs_wcc_update_inode(inode, fattr); 1558 1559 /* Check if our cached file size is stale */ 1560 new_isize = nfs_size_to_loff_t(fattr->size); 1561 cur_isize = i_size_read(inode); 1562 if (new_isize != cur_isize) { 1563 /* Do we perhaps have any outstanding writes? */ 1564 if (nfsi->npages == 0) { 1565 /* No, but did we race with nfs_end_data_update()? */ 1566 if (data_stable) { 1567 inode->i_size = new_isize; 1568 invalid |= NFS_INO_INVALID_DATA; 1569 } 1570 invalid |= NFS_INO_INVALID_ATTR; 1571 } else if (new_isize > cur_isize) { 1572 inode->i_size = new_isize; 1573 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA; 1574 } 1575 nfsi->cache_change_attribute = jiffies; 1576 dprintk("NFS: isize change on server for file %s/%ld\n", 1577 inode->i_sb->s_id, inode->i_ino); 1578 } 1579 1580 /* Check if the mtime agrees */ 1581 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) { 1582 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime)); 1583 dprintk("NFS: mtime change on server for file %s/%ld\n", 1584 inode->i_sb->s_id, inode->i_ino); 1585 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA; 1586 nfsi->cache_change_attribute = jiffies; 1587 } 1588 1589 /* If ctime has changed we should definitely clear access+acl caches */ 1590 if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) { 1591 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 1592 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime)); 1593 nfsi->cache_change_attribute = jiffies; 1594 } 1595 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime)); 1596 1597 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) || 1598 inode->i_uid != fattr->uid || 1599 inode->i_gid != fattr->gid) 1600 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 1601 1602 inode->i_mode = fattr->mode; 1603 inode->i_nlink = fattr->nlink; 1604 inode->i_uid = fattr->uid; 1605 inode->i_gid = fattr->gid; 1606 1607 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) { 1608 /* 1609 * report the blocks in 512byte units 1610 */ 1611 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); 1612 inode->i_blksize = inode->i_sb->s_blocksize; 1613 } else { 1614 inode->i_blocks = fattr->du.nfs2.blocks; 1615 inode->i_blksize = fattr->du.nfs2.blocksize; 1616 } 1617 1618 if ((fattr->valid & NFS_ATTR_FATTR_V4)) { 1619 if (nfsi->change_attr != fattr->change_attr) { 1620 dprintk("NFS: change_attr change on server for file %s/%ld\n", 1621 inode->i_sb->s_id, inode->i_ino); 1622 nfsi->change_attr = fattr->change_attr; 1623 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; 1624 nfsi->cache_change_attribute = jiffies; 1625 } else 1626 invalid &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA); 1627 } 1628 1629 /* Update attrtimeo value if we're out of the unstable period */ 1630 if (invalid & NFS_INO_INVALID_ATTR) { 1631 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); 1632 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); 1633 nfsi->attrtimeo_timestamp = jiffies; 1634 } else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) { 1635 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode)) 1636 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode); 1637 nfsi->attrtimeo_timestamp = jiffies; 1638 } 1639 /* Don't invalidate the data if we were to blame */ 1640 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) 1641 || S_ISLNK(inode->i_mode))) 1642 invalid &= ~NFS_INO_INVALID_DATA; 1643 if (data_stable) 1644 invalid &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME|NFS_INO_REVAL_PAGECACHE); 1645 if (!nfs_have_delegation(inode, FMODE_READ)) 1646 nfsi->cache_validity |= invalid; 1647 1648 return 0; 1649 out_changed: 1650 /* 1651 * Big trouble! The inode has become a different object. 1652 */ 1653 #ifdef NFS_PARANOIA 1654 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n", 1655 __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode); 1656 #endif 1657 out_err: 1658 /* 1659 * No need to worry about unhashing the dentry, as the 1660 * lookup validation will know that the inode is bad. 1661 * (But we fall through to invalidate the caches.) 1662 */ 1663 nfs_invalidate_inode(inode); 1664 return -ESTALE; 1665 1666 out_fileid: 1667 printk(KERN_ERR "NFS: server %s error: fileid changed\n" 1668 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n", 1669 NFS_SERVER(inode)->hostname, inode->i_sb->s_id, 1670 (long long)nfsi->fileid, (long long)fattr->fileid); 1671 goto out_err; 1672 } 1673 1674 /* 1675 * File system information 1676 */ 1677 1678 static int nfs_set_super(struct super_block *s, void *data) 1679 { 1680 s->s_fs_info = data; 1681 return set_anon_super(s, data); 1682 } 1683 1684 static int nfs_compare_super(struct super_block *sb, void *data) 1685 { 1686 struct nfs_server *server = data; 1687 struct nfs_server *old = NFS_SB(sb); 1688 1689 if (old->addr.sin_addr.s_addr != server->addr.sin_addr.s_addr) 1690 return 0; 1691 if (old->addr.sin_port != server->addr.sin_port) 1692 return 0; 1693 return !nfs_compare_fh(&old->fh, &server->fh); 1694 } 1695 1696 static struct super_block *nfs_get_sb(struct file_system_type *fs_type, 1697 int flags, const char *dev_name, void *raw_data) 1698 { 1699 int error; 1700 struct nfs_server *server = NULL; 1701 struct super_block *s; 1702 struct nfs_fh *root; 1703 struct nfs_mount_data *data = raw_data; 1704 1705 s = ERR_PTR(-EINVAL); 1706 if (data == NULL) { 1707 dprintk("%s: missing data argument\n", __FUNCTION__); 1708 goto out_err; 1709 } 1710 if (data->version <= 0 || data->version > NFS_MOUNT_VERSION) { 1711 dprintk("%s: bad mount version\n", __FUNCTION__); 1712 goto out_err; 1713 } 1714 switch (data->version) { 1715 case 1: 1716 data->namlen = 0; 1717 case 2: 1718 data->bsize = 0; 1719 case 3: 1720 if (data->flags & NFS_MOUNT_VER3) { 1721 dprintk("%s: mount structure version %d does not support NFSv3\n", 1722 __FUNCTION__, 1723 data->version); 1724 goto out_err; 1725 } 1726 data->root.size = NFS2_FHSIZE; 1727 memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE); 1728 case 4: 1729 if (data->flags & NFS_MOUNT_SECFLAVOUR) { 1730 dprintk("%s: mount structure version %d does not support strong security\n", 1731 __FUNCTION__, 1732 data->version); 1733 goto out_err; 1734 } 1735 case 5: 1736 memset(data->context, 0, sizeof(data->context)); 1737 } 1738 #ifndef CONFIG_NFS_V3 1739 /* If NFSv3 is not compiled in, return -EPROTONOSUPPORT */ 1740 s = ERR_PTR(-EPROTONOSUPPORT); 1741 if (data->flags & NFS_MOUNT_VER3) { 1742 dprintk("%s: NFSv3 not compiled into kernel\n", __FUNCTION__); 1743 goto out_err; 1744 } 1745 #endif /* CONFIG_NFS_V3 */ 1746 1747 s = ERR_PTR(-ENOMEM); 1748 server = kzalloc(sizeof(struct nfs_server), GFP_KERNEL); 1749 if (!server) 1750 goto out_err; 1751 /* Zero out the NFS state stuff */ 1752 init_nfsv4_state(server); 1753 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL); 1754 1755 root = &server->fh; 1756 if (data->flags & NFS_MOUNT_VER3) 1757 root->size = data->root.size; 1758 else 1759 root->size = NFS2_FHSIZE; 1760 s = ERR_PTR(-EINVAL); 1761 if (root->size > sizeof(root->data)) { 1762 dprintk("%s: invalid root filehandle\n", __FUNCTION__); 1763 goto out_err; 1764 } 1765 memcpy(root->data, data->root.data, root->size); 1766 1767 /* We now require that the mount process passes the remote address */ 1768 memcpy(&server->addr, &data->addr, sizeof(server->addr)); 1769 if (server->addr.sin_addr.s_addr == INADDR_ANY) { 1770 dprintk("%s: mount program didn't pass remote address!\n", 1771 __FUNCTION__); 1772 goto out_err; 1773 } 1774 1775 /* Fire up rpciod if not yet running */ 1776 s = ERR_PTR(rpciod_up()); 1777 if (IS_ERR(s)) { 1778 dprintk("%s: couldn't start rpciod! Error = %ld\n", 1779 __FUNCTION__, PTR_ERR(s)); 1780 goto out_err; 1781 } 1782 1783 s = sget(fs_type, nfs_compare_super, nfs_set_super, server); 1784 if (IS_ERR(s) || s->s_root) 1785 goto out_rpciod_down; 1786 1787 s->s_flags = flags; 1788 1789 error = nfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0); 1790 if (error) { 1791 up_write(&s->s_umount); 1792 deactivate_super(s); 1793 return ERR_PTR(error); 1794 } 1795 s->s_flags |= MS_ACTIVE; 1796 return s; 1797 out_rpciod_down: 1798 rpciod_down(); 1799 out_err: 1800 kfree(server); 1801 return s; 1802 } 1803 1804 static void nfs_kill_super(struct super_block *s) 1805 { 1806 struct nfs_server *server = NFS_SB(s); 1807 1808 kill_anon_super(s); 1809 1810 if (!IS_ERR(server->client)) 1811 rpc_shutdown_client(server->client); 1812 if (!IS_ERR(server->client_sys)) 1813 rpc_shutdown_client(server->client_sys); 1814 if (!IS_ERR(server->client_acl)) 1815 rpc_shutdown_client(server->client_acl); 1816 1817 if (!(server->flags & NFS_MOUNT_NONLM)) 1818 lockd_down(); /* release rpc.lockd */ 1819 1820 rpciod_down(); /* release rpciod */ 1821 1822 nfs_free_iostats(server->io_stats); 1823 kfree(server->hostname); 1824 kfree(server); 1825 } 1826 1827 static struct file_system_type nfs_fs_type = { 1828 .owner = THIS_MODULE, 1829 .name = "nfs", 1830 .get_sb = nfs_get_sb, 1831 .kill_sb = nfs_kill_super, 1832 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA, 1833 }; 1834 1835 #ifdef CONFIG_NFS_V4 1836 1837 static void nfs4_clear_inode(struct inode *); 1838 1839 1840 static struct super_operations nfs4_sops = { 1841 .alloc_inode = nfs_alloc_inode, 1842 .destroy_inode = nfs_destroy_inode, 1843 .write_inode = nfs_write_inode, 1844 .delete_inode = nfs_delete_inode, 1845 .statfs = nfs_statfs, 1846 .clear_inode = nfs4_clear_inode, 1847 .umount_begin = nfs_umount_begin, 1848 .show_options = nfs_show_options, 1849 .show_stats = nfs_show_stats, 1850 }; 1851 1852 /* 1853 * Clean out any remaining NFSv4 state that might be left over due 1854 * to open() calls that passed nfs_atomic_lookup, but failed to call 1855 * nfs_open(). 1856 */ 1857 static void nfs4_clear_inode(struct inode *inode) 1858 { 1859 struct nfs_inode *nfsi = NFS_I(inode); 1860 1861 /* If we are holding a delegation, return it! */ 1862 nfs_inode_return_delegation(inode); 1863 /* First call standard NFS clear_inode() code */ 1864 nfs_clear_inode(inode); 1865 /* Now clear out any remaining state */ 1866 while (!list_empty(&nfsi->open_states)) { 1867 struct nfs4_state *state; 1868 1869 state = list_entry(nfsi->open_states.next, 1870 struct nfs4_state, 1871 inode_states); 1872 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n", 1873 __FUNCTION__, 1874 inode->i_sb->s_id, 1875 (long long)NFS_FILEID(inode), 1876 state); 1877 BUG_ON(atomic_read(&state->count) != 1); 1878 nfs4_close_state(state, state->state); 1879 } 1880 } 1881 1882 1883 static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data, int silent) 1884 { 1885 struct nfs_server *server; 1886 struct nfs4_client *clp = NULL; 1887 struct rpc_xprt *xprt = NULL; 1888 struct rpc_clnt *clnt = NULL; 1889 struct rpc_timeout timeparms; 1890 rpc_authflavor_t authflavour; 1891 int err = -EIO; 1892 1893 sb->s_blocksize_bits = 0; 1894 sb->s_blocksize = 0; 1895 server = NFS_SB(sb); 1896 if (data->rsize != 0) 1897 server->rsize = nfs_block_size(data->rsize, NULL); 1898 if (data->wsize != 0) 1899 server->wsize = nfs_block_size(data->wsize, NULL); 1900 server->flags = data->flags & NFS_MOUNT_FLAGMASK; 1901 server->caps = NFS_CAP_ATOMIC_OPEN; 1902 1903 server->acregmin = data->acregmin*HZ; 1904 server->acregmax = data->acregmax*HZ; 1905 server->acdirmin = data->acdirmin*HZ; 1906 server->acdirmax = data->acdirmax*HZ; 1907 1908 server->rpc_ops = &nfs_v4_clientops; 1909 1910 nfs_init_timeout_values(&timeparms, data->proto, data->timeo, data->retrans); 1911 1912 server->retrans_timeo = timeparms.to_initval; 1913 server->retrans_count = timeparms.to_retries; 1914 1915 clp = nfs4_get_client(&server->addr.sin_addr); 1916 if (!clp) { 1917 dprintk("%s: failed to create NFS4 client.\n", __FUNCTION__); 1918 return -EIO; 1919 } 1920 1921 /* Now create transport and client */ 1922 authflavour = RPC_AUTH_UNIX; 1923 if (data->auth_flavourlen != 0) { 1924 if (data->auth_flavourlen != 1) { 1925 dprintk("%s: Invalid number of RPC auth flavours %d.\n", 1926 __FUNCTION__, data->auth_flavourlen); 1927 err = -EINVAL; 1928 goto out_fail; 1929 } 1930 if (copy_from_user(&authflavour, data->auth_flavours, sizeof(authflavour))) { 1931 err = -EFAULT; 1932 goto out_fail; 1933 } 1934 } 1935 1936 down_write(&clp->cl_sem); 1937 if (IS_ERR(clp->cl_rpcclient)) { 1938 xprt = xprt_create_proto(data->proto, &server->addr, &timeparms); 1939 if (IS_ERR(xprt)) { 1940 up_write(&clp->cl_sem); 1941 err = PTR_ERR(xprt); 1942 dprintk("%s: cannot create RPC transport. Error = %d\n", 1943 __FUNCTION__, err); 1944 goto out_fail; 1945 } 1946 clnt = rpc_create_client(xprt, server->hostname, &nfs_program, 1947 server->rpc_ops->version, authflavour); 1948 if (IS_ERR(clnt)) { 1949 up_write(&clp->cl_sem); 1950 err = PTR_ERR(clnt); 1951 dprintk("%s: cannot create RPC client. Error = %d\n", 1952 __FUNCTION__, err); 1953 goto out_fail; 1954 } 1955 clnt->cl_intr = 1; 1956 clnt->cl_softrtry = 1; 1957 clp->cl_rpcclient = clnt; 1958 memcpy(clp->cl_ipaddr, server->ip_addr, sizeof(clp->cl_ipaddr)); 1959 nfs_idmap_new(clp); 1960 } 1961 list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks); 1962 clnt = rpc_clone_client(clp->cl_rpcclient); 1963 if (!IS_ERR(clnt)) 1964 server->nfs4_state = clp; 1965 up_write(&clp->cl_sem); 1966 clp = NULL; 1967 1968 if (IS_ERR(clnt)) { 1969 err = PTR_ERR(clnt); 1970 dprintk("%s: cannot create RPC client. Error = %d\n", 1971 __FUNCTION__, err); 1972 return err; 1973 } 1974 1975 server->client = clnt; 1976 1977 if (server->nfs4_state->cl_idmap == NULL) { 1978 dprintk("%s: failed to create idmapper.\n", __FUNCTION__); 1979 return -ENOMEM; 1980 } 1981 1982 if (clnt->cl_auth->au_flavor != authflavour) { 1983 struct rpc_auth *auth; 1984 1985 auth = rpcauth_create(authflavour, clnt); 1986 if (IS_ERR(auth)) { 1987 dprintk("%s: couldn't create credcache!\n", __FUNCTION__); 1988 return PTR_ERR(auth); 1989 } 1990 } 1991 1992 sb->s_time_gran = 1; 1993 1994 sb->s_op = &nfs4_sops; 1995 err = nfs_sb_init(sb, authflavour); 1996 if (err == 0) 1997 return 0; 1998 out_fail: 1999 if (clp) 2000 nfs4_put_client(clp); 2001 return err; 2002 } 2003 2004 static int nfs4_compare_super(struct super_block *sb, void *data) 2005 { 2006 struct nfs_server *server = data; 2007 struct nfs_server *old = NFS_SB(sb); 2008 2009 if (strcmp(server->hostname, old->hostname) != 0) 2010 return 0; 2011 if (strcmp(server->mnt_path, old->mnt_path) != 0) 2012 return 0; 2013 return 1; 2014 } 2015 2016 static void * 2017 nfs_copy_user_string(char *dst, struct nfs_string *src, int maxlen) 2018 { 2019 void *p = NULL; 2020 2021 if (!src->len) 2022 return ERR_PTR(-EINVAL); 2023 if (src->len < maxlen) 2024 maxlen = src->len; 2025 if (dst == NULL) { 2026 p = dst = kmalloc(maxlen + 1, GFP_KERNEL); 2027 if (p == NULL) 2028 return ERR_PTR(-ENOMEM); 2029 } 2030 if (copy_from_user(dst, src->data, maxlen)) { 2031 kfree(p); 2032 return ERR_PTR(-EFAULT); 2033 } 2034 dst[maxlen] = '\0'; 2035 return dst; 2036 } 2037 2038 static struct super_block *nfs4_get_sb(struct file_system_type *fs_type, 2039 int flags, const char *dev_name, void *raw_data) 2040 { 2041 int error; 2042 struct nfs_server *server; 2043 struct super_block *s; 2044 struct nfs4_mount_data *data = raw_data; 2045 void *p; 2046 2047 if (data == NULL) { 2048 dprintk("%s: missing data argument\n", __FUNCTION__); 2049 return ERR_PTR(-EINVAL); 2050 } 2051 if (data->version <= 0 || data->version > NFS4_MOUNT_VERSION) { 2052 dprintk("%s: bad mount version\n", __FUNCTION__); 2053 return ERR_PTR(-EINVAL); 2054 } 2055 2056 server = kzalloc(sizeof(struct nfs_server), GFP_KERNEL); 2057 if (!server) 2058 return ERR_PTR(-ENOMEM); 2059 /* Zero out the NFS state stuff */ 2060 init_nfsv4_state(server); 2061 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL); 2062 2063 p = nfs_copy_user_string(NULL, &data->hostname, 256); 2064 if (IS_ERR(p)) 2065 goto out_err; 2066 server->hostname = p; 2067 2068 p = nfs_copy_user_string(NULL, &data->mnt_path, 1024); 2069 if (IS_ERR(p)) 2070 goto out_err; 2071 server->mnt_path = p; 2072 2073 p = nfs_copy_user_string(server->ip_addr, &data->client_addr, 2074 sizeof(server->ip_addr) - 1); 2075 if (IS_ERR(p)) 2076 goto out_err; 2077 2078 /* We now require that the mount process passes the remote address */ 2079 if (data->host_addrlen != sizeof(server->addr)) { 2080 s = ERR_PTR(-EINVAL); 2081 goto out_free; 2082 } 2083 if (copy_from_user(&server->addr, data->host_addr, sizeof(server->addr))) { 2084 s = ERR_PTR(-EFAULT); 2085 goto out_free; 2086 } 2087 if (server->addr.sin_family != AF_INET || 2088 server->addr.sin_addr.s_addr == INADDR_ANY) { 2089 dprintk("%s: mount program didn't pass remote IP address!\n", 2090 __FUNCTION__); 2091 s = ERR_PTR(-EINVAL); 2092 goto out_free; 2093 } 2094 2095 /* Fire up rpciod if not yet running */ 2096 s = ERR_PTR(rpciod_up()); 2097 if (IS_ERR(s)) { 2098 dprintk("%s: couldn't start rpciod! Error = %ld\n", 2099 __FUNCTION__, PTR_ERR(s)); 2100 goto out_free; 2101 } 2102 2103 s = sget(fs_type, nfs4_compare_super, nfs_set_super, server); 2104 2105 if (IS_ERR(s) || s->s_root) 2106 goto out_free; 2107 2108 s->s_flags = flags; 2109 2110 error = nfs4_fill_super(s, data, flags & MS_SILENT ? 1 : 0); 2111 if (error) { 2112 up_write(&s->s_umount); 2113 deactivate_super(s); 2114 return ERR_PTR(error); 2115 } 2116 s->s_flags |= MS_ACTIVE; 2117 return s; 2118 out_err: 2119 s = (struct super_block *)p; 2120 out_free: 2121 kfree(server->mnt_path); 2122 kfree(server->hostname); 2123 kfree(server); 2124 return s; 2125 } 2126 2127 static void nfs4_kill_super(struct super_block *sb) 2128 { 2129 struct nfs_server *server = NFS_SB(sb); 2130 2131 nfs_return_all_delegations(sb); 2132 kill_anon_super(sb); 2133 2134 nfs4_renewd_prepare_shutdown(server); 2135 2136 if (server->client != NULL && !IS_ERR(server->client)) 2137 rpc_shutdown_client(server->client); 2138 2139 destroy_nfsv4_state(server); 2140 2141 rpciod_down(); 2142 2143 nfs_free_iostats(server->io_stats); 2144 kfree(server->hostname); 2145 kfree(server); 2146 } 2147 2148 static struct file_system_type nfs4_fs_type = { 2149 .owner = THIS_MODULE, 2150 .name = "nfs4", 2151 .get_sb = nfs4_get_sb, 2152 .kill_sb = nfs4_kill_super, 2153 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA, 2154 }; 2155 2156 static const int nfs_set_port_min = 0; 2157 static const int nfs_set_port_max = 65535; 2158 static int param_set_port(const char *val, struct kernel_param *kp) 2159 { 2160 char *endp; 2161 int num = simple_strtol(val, &endp, 0); 2162 if (endp == val || *endp || num < nfs_set_port_min || num > nfs_set_port_max) 2163 return -EINVAL; 2164 *((int *)kp->arg) = num; 2165 return 0; 2166 } 2167 2168 module_param_call(callback_tcpport, param_set_port, param_get_int, 2169 &nfs_callback_set_tcpport, 0644); 2170 2171 static int param_set_idmap_timeout(const char *val, struct kernel_param *kp) 2172 { 2173 char *endp; 2174 int num = simple_strtol(val, &endp, 0); 2175 int jif = num * HZ; 2176 if (endp == val || *endp || num < 0 || jif < num) 2177 return -EINVAL; 2178 *((int *)kp->arg) = jif; 2179 return 0; 2180 } 2181 2182 module_param_call(idmap_cache_timeout, param_set_idmap_timeout, param_get_int, 2183 &nfs_idmap_cache_timeout, 0644); 2184 2185 #define nfs4_init_once(nfsi) \ 2186 do { \ 2187 INIT_LIST_HEAD(&(nfsi)->open_states); \ 2188 nfsi->delegation = NULL; \ 2189 nfsi->delegation_state = 0; \ 2190 init_rwsem(&nfsi->rwsem); \ 2191 } while(0) 2192 2193 static inline int register_nfs4fs(void) 2194 { 2195 int ret; 2196 2197 ret = nfs_register_sysctl(); 2198 if (ret != 0) 2199 return ret; 2200 ret = register_filesystem(&nfs4_fs_type); 2201 if (ret != 0) 2202 nfs_unregister_sysctl(); 2203 return ret; 2204 } 2205 2206 static inline void unregister_nfs4fs(void) 2207 { 2208 unregister_filesystem(&nfs4_fs_type); 2209 nfs_unregister_sysctl(); 2210 } 2211 #else 2212 #define nfs4_init_once(nfsi) \ 2213 do { } while (0) 2214 #define register_nfs4fs() (0) 2215 #define unregister_nfs4fs() 2216 #endif 2217 2218 extern int nfs_init_nfspagecache(void); 2219 extern void nfs_destroy_nfspagecache(void); 2220 extern int nfs_init_readpagecache(void); 2221 extern void nfs_destroy_readpagecache(void); 2222 extern int nfs_init_writepagecache(void); 2223 extern void nfs_destroy_writepagecache(void); 2224 #ifdef CONFIG_NFS_DIRECTIO 2225 extern int nfs_init_directcache(void); 2226 extern void nfs_destroy_directcache(void); 2227 #endif 2228 2229 static kmem_cache_t * nfs_inode_cachep; 2230 2231 static struct inode *nfs_alloc_inode(struct super_block *sb) 2232 { 2233 struct nfs_inode *nfsi; 2234 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL); 2235 if (!nfsi) 2236 return NULL; 2237 nfsi->flags = 0UL; 2238 nfsi->cache_validity = 0UL; 2239 nfsi->cache_change_attribute = jiffies; 2240 #ifdef CONFIG_NFS_V3_ACL 2241 nfsi->acl_access = ERR_PTR(-EAGAIN); 2242 nfsi->acl_default = ERR_PTR(-EAGAIN); 2243 #endif 2244 #ifdef CONFIG_NFS_V4 2245 nfsi->nfs4_acl = NULL; 2246 #endif /* CONFIG_NFS_V4 */ 2247 return &nfsi->vfs_inode; 2248 } 2249 2250 static void nfs_destroy_inode(struct inode *inode) 2251 { 2252 kmem_cache_free(nfs_inode_cachep, NFS_I(inode)); 2253 } 2254 2255 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags) 2256 { 2257 struct nfs_inode *nfsi = (struct nfs_inode *) foo; 2258 2259 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) == 2260 SLAB_CTOR_CONSTRUCTOR) { 2261 inode_init_once(&nfsi->vfs_inode); 2262 spin_lock_init(&nfsi->req_lock); 2263 INIT_LIST_HEAD(&nfsi->dirty); 2264 INIT_LIST_HEAD(&nfsi->commit); 2265 INIT_LIST_HEAD(&nfsi->open_files); 2266 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC); 2267 atomic_set(&nfsi->data_updates, 0); 2268 nfsi->ndirty = 0; 2269 nfsi->ncommit = 0; 2270 nfsi->npages = 0; 2271 nfs4_init_once(nfsi); 2272 } 2273 } 2274 2275 static int nfs_init_inodecache(void) 2276 { 2277 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache", 2278 sizeof(struct nfs_inode), 2279 0, (SLAB_RECLAIM_ACCOUNT| 2280 SLAB_MEM_SPREAD), 2281 init_once, NULL); 2282 if (nfs_inode_cachep == NULL) 2283 return -ENOMEM; 2284 2285 return 0; 2286 } 2287 2288 static void nfs_destroy_inodecache(void) 2289 { 2290 if (kmem_cache_destroy(nfs_inode_cachep)) 2291 printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n"); 2292 } 2293 2294 /* 2295 * Initialize NFS 2296 */ 2297 static int __init init_nfs_fs(void) 2298 { 2299 int err; 2300 2301 err = nfs_init_nfspagecache(); 2302 if (err) 2303 goto out4; 2304 2305 err = nfs_init_inodecache(); 2306 if (err) 2307 goto out3; 2308 2309 err = nfs_init_readpagecache(); 2310 if (err) 2311 goto out2; 2312 2313 err = nfs_init_writepagecache(); 2314 if (err) 2315 goto out1; 2316 2317 #ifdef CONFIG_NFS_DIRECTIO 2318 err = nfs_init_directcache(); 2319 if (err) 2320 goto out0; 2321 #endif 2322 2323 #ifdef CONFIG_PROC_FS 2324 rpc_proc_register(&nfs_rpcstat); 2325 #endif 2326 err = register_filesystem(&nfs_fs_type); 2327 if (err) 2328 goto out; 2329 if ((err = register_nfs4fs()) != 0) 2330 goto out; 2331 return 0; 2332 out: 2333 #ifdef CONFIG_PROC_FS 2334 rpc_proc_unregister("nfs"); 2335 #endif 2336 #ifdef CONFIG_NFS_DIRECTIO 2337 nfs_destroy_directcache(); 2338 out0: 2339 #endif 2340 nfs_destroy_writepagecache(); 2341 out1: 2342 nfs_destroy_readpagecache(); 2343 out2: 2344 nfs_destroy_inodecache(); 2345 out3: 2346 nfs_destroy_nfspagecache(); 2347 out4: 2348 return err; 2349 } 2350 2351 static void __exit exit_nfs_fs(void) 2352 { 2353 #ifdef CONFIG_NFS_DIRECTIO 2354 nfs_destroy_directcache(); 2355 #endif 2356 nfs_destroy_writepagecache(); 2357 nfs_destroy_readpagecache(); 2358 nfs_destroy_inodecache(); 2359 nfs_destroy_nfspagecache(); 2360 #ifdef CONFIG_PROC_FS 2361 rpc_proc_unregister("nfs"); 2362 #endif 2363 unregister_filesystem(&nfs_fs_type); 2364 unregister_nfs4fs(); 2365 } 2366 2367 /* Not quite true; I just maintain it */ 2368 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>"); 2369 MODULE_LICENSE("GPL"); 2370 2371 module_init(init_nfs_fs) 2372 module_exit(exit_nfs_fs) 2373