| 1 | /* |
| 2 | * Device operations for the pnfs nfs4 file layout driver. |
| 3 | * |
| 4 | * Copyright (c) 2002 |
| 5 | * The Regents of the University of Michigan |
| 6 | * All Rights Reserved |
| 7 | * |
| 8 | * Dean Hildebrand <dhildebz@umich.edu> |
| 9 | * Garth Goodson <Garth.Goodson@netapp.com> |
| 10 | * |
| 11 | * Permission is granted to use, copy, create derivative works, and |
| 12 | * redistribute this software and such derivative works for any purpose, |
| 13 | * so long as the name of the University of Michigan is not used in |
| 14 | * any advertising or publicity pertaining to the use or distribution |
| 15 | * of this software without specific, written prior authorization. If |
| 16 | * the above copyright notice or any other identification of the |
| 17 | * University of Michigan is included in any copy of any portion of |
| 18 | * this software, then the disclaimer below must also be included. |
| 19 | * |
| 20 | * This software is provided as is, without representation or warranty |
| 21 | * of any kind either express or implied, including without limitation |
| 22 | * the implied warranties of merchantability, fitness for a particular |
| 23 | * purpose, or noninfringement. The Regents of the University of |
| 24 | * Michigan shall not be liable for any damages, including special, |
| 25 | * indirect, incidental, or consequential damages, with respect to any |
| 26 | * claim arising out of or in connection with the use of the software, |
| 27 | * even if it has been or is hereafter advised of the possibility of |
| 28 | * such damages. |
| 29 | */ |
| 30 | |
| 31 | #include <linux/nfs_fs.h> |
| 32 | #include <linux/vmalloc.h> |
| 33 | #include <linux/module.h> |
| 34 | #include <linux/sunrpc/addr.h> |
| 35 | |
| 36 | #include "internal.h" |
| 37 | #include "nfs4session.h" |
| 38 | #include "nfs4filelayout.h" |
| 39 | |
| 40 | #define NFSDBG_FACILITY NFSDBG_PNFS_LD |
| 41 | |
| 42 | static unsigned int dataserver_timeo = NFS4_DEF_DS_TIMEO; |
| 43 | static unsigned int dataserver_retrans = NFS4_DEF_DS_RETRANS; |
| 44 | |
| 45 | /* |
| 46 | * Data server cache |
| 47 | * |
| 48 | * Data servers can be mapped to different device ids. |
| 49 | * nfs4_pnfs_ds reference counting |
| 50 | * - set to 1 on allocation |
| 51 | * - incremented when a device id maps a data server already in the cache. |
| 52 | * - decremented when deviceid is removed from the cache. |
| 53 | */ |
| 54 | static DEFINE_SPINLOCK(nfs4_ds_cache_lock); |
| 55 | static LIST_HEAD(nfs4_data_server_cache); |
| 56 | |
| 57 | /* Debug routines */ |
| 58 | void |
| 59 | print_ds(struct nfs4_pnfs_ds *ds) |
| 60 | { |
| 61 | if (ds == NULL) { |
| 62 | printk("%s NULL device\n", __func__); |
| 63 | return; |
| 64 | } |
| 65 | printk(" ds %s\n" |
| 66 | " ref count %d\n" |
| 67 | " client %p\n" |
| 68 | " cl_exchange_flags %x\n", |
| 69 | ds->ds_remotestr, |
| 70 | atomic_read(&ds->ds_count), ds->ds_clp, |
| 71 | ds->ds_clp ? ds->ds_clp->cl_exchange_flags : 0); |
| 72 | } |
| 73 | |
| 74 | static bool |
| 75 | same_sockaddr(struct sockaddr *addr1, struct sockaddr *addr2) |
| 76 | { |
| 77 | struct sockaddr_in *a, *b; |
| 78 | struct sockaddr_in6 *a6, *b6; |
| 79 | |
| 80 | if (addr1->sa_family != addr2->sa_family) |
| 81 | return false; |
| 82 | |
| 83 | switch (addr1->sa_family) { |
| 84 | case AF_INET: |
| 85 | a = (struct sockaddr_in *)addr1; |
| 86 | b = (struct sockaddr_in *)addr2; |
| 87 | |
| 88 | if (a->sin_addr.s_addr == b->sin_addr.s_addr && |
| 89 | a->sin_port == b->sin_port) |
| 90 | return true; |
| 91 | break; |
| 92 | |
| 93 | case AF_INET6: |
| 94 | a6 = (struct sockaddr_in6 *)addr1; |
| 95 | b6 = (struct sockaddr_in6 *)addr2; |
| 96 | |
| 97 | /* LINKLOCAL addresses must have matching scope_id */ |
| 98 | if (ipv6_addr_scope(&a6->sin6_addr) == |
| 99 | IPV6_ADDR_SCOPE_LINKLOCAL && |
| 100 | a6->sin6_scope_id != b6->sin6_scope_id) |
| 101 | return false; |
| 102 | |
| 103 | if (ipv6_addr_equal(&a6->sin6_addr, &b6->sin6_addr) && |
| 104 | a6->sin6_port == b6->sin6_port) |
| 105 | return true; |
| 106 | break; |
| 107 | |
| 108 | default: |
| 109 | dprintk("%s: unhandled address family: %u\n", |
| 110 | __func__, addr1->sa_family); |
| 111 | return false; |
| 112 | } |
| 113 | |
| 114 | return false; |
| 115 | } |
| 116 | |
| 117 | static bool |
| 118 | _same_data_server_addrs_locked(const struct list_head *dsaddrs1, |
| 119 | const struct list_head *dsaddrs2) |
| 120 | { |
| 121 | struct nfs4_pnfs_ds_addr *da1, *da2; |
| 122 | |
| 123 | /* step through both lists, comparing as we go */ |
| 124 | for (da1 = list_first_entry(dsaddrs1, typeof(*da1), da_node), |
| 125 | da2 = list_first_entry(dsaddrs2, typeof(*da2), da_node); |
| 126 | da1 != NULL && da2 != NULL; |
| 127 | da1 = list_entry(da1->da_node.next, typeof(*da1), da_node), |
| 128 | da2 = list_entry(da2->da_node.next, typeof(*da2), da_node)) { |
| 129 | if (!same_sockaddr((struct sockaddr *)&da1->da_addr, |
| 130 | (struct sockaddr *)&da2->da_addr)) |
| 131 | return false; |
| 132 | } |
| 133 | if (da1 == NULL && da2 == NULL) |
| 134 | return true; |
| 135 | |
| 136 | return false; |
| 137 | } |
| 138 | |
| 139 | /* |
| 140 | * Lookup DS by addresses. nfs4_ds_cache_lock is held |
| 141 | */ |
| 142 | static struct nfs4_pnfs_ds * |
| 143 | _data_server_lookup_locked(const struct list_head *dsaddrs) |
| 144 | { |
| 145 | struct nfs4_pnfs_ds *ds; |
| 146 | |
| 147 | list_for_each_entry(ds, &nfs4_data_server_cache, ds_node) |
| 148 | if (_same_data_server_addrs_locked(&ds->ds_addrs, dsaddrs)) |
| 149 | return ds; |
| 150 | return NULL; |
| 151 | } |
| 152 | |
| 153 | /* |
| 154 | * Create an rpc connection to the nfs4_pnfs_ds data server |
| 155 | * Currently only supports IPv4 and IPv6 addresses |
| 156 | */ |
| 157 | static int |
| 158 | nfs4_ds_connect(struct nfs_server *mds_srv, struct nfs4_pnfs_ds *ds) |
| 159 | { |
| 160 | struct nfs_client *clp = ERR_PTR(-EIO); |
| 161 | struct nfs4_pnfs_ds_addr *da; |
| 162 | int status = 0; |
| 163 | |
| 164 | dprintk("--> %s DS %s au_flavor %d\n", __func__, ds->ds_remotestr, |
| 165 | mds_srv->nfs_client->cl_rpcclient->cl_auth->au_flavor); |
| 166 | |
| 167 | list_for_each_entry(da, &ds->ds_addrs, da_node) { |
| 168 | dprintk("%s: DS %s: trying address %s\n", |
| 169 | __func__, ds->ds_remotestr, da->da_remotestr); |
| 170 | |
| 171 | clp = nfs4_set_ds_client(mds_srv->nfs_client, |
| 172 | (struct sockaddr *)&da->da_addr, |
| 173 | da->da_addrlen, IPPROTO_TCP, |
| 174 | dataserver_timeo, dataserver_retrans); |
| 175 | if (!IS_ERR(clp)) |
| 176 | break; |
| 177 | } |
| 178 | |
| 179 | if (IS_ERR(clp)) { |
| 180 | status = PTR_ERR(clp); |
| 181 | goto out; |
| 182 | } |
| 183 | |
| 184 | status = nfs4_init_ds_session(clp, mds_srv->nfs_client->cl_lease_time); |
| 185 | if (status) |
| 186 | goto out_put; |
| 187 | |
| 188 | ds->ds_clp = clp; |
| 189 | dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr); |
| 190 | out: |
| 191 | return status; |
| 192 | out_put: |
| 193 | nfs_put_client(clp); |
| 194 | goto out; |
| 195 | } |
| 196 | |
| 197 | static void |
| 198 | destroy_ds(struct nfs4_pnfs_ds *ds) |
| 199 | { |
| 200 | struct nfs4_pnfs_ds_addr *da; |
| 201 | |
| 202 | dprintk("--> %s\n", __func__); |
| 203 | ifdebug(FACILITY) |
| 204 | print_ds(ds); |
| 205 | |
| 206 | if (ds->ds_clp) |
| 207 | nfs_put_client(ds->ds_clp); |
| 208 | |
| 209 | while (!list_empty(&ds->ds_addrs)) { |
| 210 | da = list_first_entry(&ds->ds_addrs, |
| 211 | struct nfs4_pnfs_ds_addr, |
| 212 | da_node); |
| 213 | list_del_init(&da->da_node); |
| 214 | kfree(da->da_remotestr); |
| 215 | kfree(da); |
| 216 | } |
| 217 | |
| 218 | kfree(ds->ds_remotestr); |
| 219 | kfree(ds); |
| 220 | } |
| 221 | |
| 222 | void |
| 223 | nfs4_fl_free_deviceid(struct nfs4_file_layout_dsaddr *dsaddr) |
| 224 | { |
| 225 | struct nfs4_pnfs_ds *ds; |
| 226 | int i; |
| 227 | |
| 228 | nfs4_print_deviceid(&dsaddr->id_node.deviceid); |
| 229 | |
| 230 | for (i = 0; i < dsaddr->ds_num; i++) { |
| 231 | ds = dsaddr->ds_list[i]; |
| 232 | if (ds != NULL) { |
| 233 | if (atomic_dec_and_lock(&ds->ds_count, |
| 234 | &nfs4_ds_cache_lock)) { |
| 235 | list_del_init(&ds->ds_node); |
| 236 | spin_unlock(&nfs4_ds_cache_lock); |
| 237 | destroy_ds(ds); |
| 238 | } |
| 239 | } |
| 240 | } |
| 241 | kfree(dsaddr->stripe_indices); |
| 242 | kfree(dsaddr); |
| 243 | } |
| 244 | |
| 245 | /* |
| 246 | * Create a string with a human readable address and port to avoid |
| 247 | * complicated setup around many dprinks. |
| 248 | */ |
| 249 | static char * |
| 250 | nfs4_pnfs_remotestr(struct list_head *dsaddrs, gfp_t gfp_flags) |
| 251 | { |
| 252 | struct nfs4_pnfs_ds_addr *da; |
| 253 | char *remotestr; |
| 254 | size_t len; |
| 255 | char *p; |
| 256 | |
| 257 | len = 3; /* '{', '}' and eol */ |
| 258 | list_for_each_entry(da, dsaddrs, da_node) { |
| 259 | len += strlen(da->da_remotestr) + 1; /* string plus comma */ |
| 260 | } |
| 261 | |
| 262 | remotestr = kzalloc(len, gfp_flags); |
| 263 | if (!remotestr) |
| 264 | return NULL; |
| 265 | |
| 266 | p = remotestr; |
| 267 | *(p++) = '{'; |
| 268 | len--; |
| 269 | list_for_each_entry(da, dsaddrs, da_node) { |
| 270 | size_t ll = strlen(da->da_remotestr); |
| 271 | |
| 272 | if (ll > len) |
| 273 | goto out_err; |
| 274 | |
| 275 | memcpy(p, da->da_remotestr, ll); |
| 276 | p += ll; |
| 277 | len -= ll; |
| 278 | |
| 279 | if (len < 1) |
| 280 | goto out_err; |
| 281 | (*p++) = ','; |
| 282 | len--; |
| 283 | } |
| 284 | if (len < 2) |
| 285 | goto out_err; |
| 286 | *(p++) = '}'; |
| 287 | *p = '\0'; |
| 288 | return remotestr; |
| 289 | out_err: |
| 290 | kfree(remotestr); |
| 291 | return NULL; |
| 292 | } |
| 293 | |
| 294 | static struct nfs4_pnfs_ds * |
| 295 | nfs4_pnfs_ds_add(struct list_head *dsaddrs, gfp_t gfp_flags) |
| 296 | { |
| 297 | struct nfs4_pnfs_ds *tmp_ds, *ds = NULL; |
| 298 | char *remotestr; |
| 299 | |
| 300 | if (list_empty(dsaddrs)) { |
| 301 | dprintk("%s: no addresses defined\n", __func__); |
| 302 | goto out; |
| 303 | } |
| 304 | |
| 305 | ds = kzalloc(sizeof(*ds), gfp_flags); |
| 306 | if (!ds) |
| 307 | goto out; |
| 308 | |
| 309 | /* this is only used for debugging, so it's ok if its NULL */ |
| 310 | remotestr = nfs4_pnfs_remotestr(dsaddrs, gfp_flags); |
| 311 | |
| 312 | spin_lock(&nfs4_ds_cache_lock); |
| 313 | tmp_ds = _data_server_lookup_locked(dsaddrs); |
| 314 | if (tmp_ds == NULL) { |
| 315 | INIT_LIST_HEAD(&ds->ds_addrs); |
| 316 | list_splice_init(dsaddrs, &ds->ds_addrs); |
| 317 | ds->ds_remotestr = remotestr; |
| 318 | atomic_set(&ds->ds_count, 1); |
| 319 | INIT_LIST_HEAD(&ds->ds_node); |
| 320 | ds->ds_clp = NULL; |
| 321 | list_add(&ds->ds_node, &nfs4_data_server_cache); |
| 322 | dprintk("%s add new data server %s\n", __func__, |
| 323 | ds->ds_remotestr); |
| 324 | } else { |
| 325 | kfree(remotestr); |
| 326 | kfree(ds); |
| 327 | atomic_inc(&tmp_ds->ds_count); |
| 328 | dprintk("%s data server %s found, inc'ed ds_count to %d\n", |
| 329 | __func__, tmp_ds->ds_remotestr, |
| 330 | atomic_read(&tmp_ds->ds_count)); |
| 331 | ds = tmp_ds; |
| 332 | } |
| 333 | spin_unlock(&nfs4_ds_cache_lock); |
| 334 | out: |
| 335 | return ds; |
| 336 | } |
| 337 | |
| 338 | /* |
| 339 | * Currently only supports ipv4, ipv6 and one multi-path address. |
| 340 | */ |
| 341 | static struct nfs4_pnfs_ds_addr * |
| 342 | decode_ds_addr(struct net *net, struct xdr_stream *streamp, gfp_t gfp_flags) |
| 343 | { |
| 344 | struct nfs4_pnfs_ds_addr *da = NULL; |
| 345 | char *buf, *portstr; |
| 346 | __be16 port; |
| 347 | int nlen, rlen; |
| 348 | int tmp[2]; |
| 349 | __be32 *p; |
| 350 | char *netid, *match_netid; |
| 351 | size_t len, match_netid_len; |
| 352 | char *startsep = ""; |
| 353 | char *endsep = ""; |
| 354 | |
| 355 | |
| 356 | /* r_netid */ |
| 357 | p = xdr_inline_decode(streamp, 4); |
| 358 | if (unlikely(!p)) |
| 359 | goto out_err; |
| 360 | nlen = be32_to_cpup(p++); |
| 361 | |
| 362 | p = xdr_inline_decode(streamp, nlen); |
| 363 | if (unlikely(!p)) |
| 364 | goto out_err; |
| 365 | |
| 366 | netid = kmalloc(nlen+1, gfp_flags); |
| 367 | if (unlikely(!netid)) |
| 368 | goto out_err; |
| 369 | |
| 370 | netid[nlen] = '\0'; |
| 371 | memcpy(netid, p, nlen); |
| 372 | |
| 373 | /* r_addr: ip/ip6addr with port in dec octets - see RFC 5665 */ |
| 374 | p = xdr_inline_decode(streamp, 4); |
| 375 | if (unlikely(!p)) |
| 376 | goto out_free_netid; |
| 377 | rlen = be32_to_cpup(p); |
| 378 | |
| 379 | p = xdr_inline_decode(streamp, rlen); |
| 380 | if (unlikely(!p)) |
| 381 | goto out_free_netid; |
| 382 | |
| 383 | /* port is ".ABC.DEF", 8 chars max */ |
| 384 | if (rlen > INET6_ADDRSTRLEN + IPV6_SCOPE_ID_LEN + 8) { |
| 385 | dprintk("%s: Invalid address, length %d\n", __func__, |
| 386 | rlen); |
| 387 | goto out_free_netid; |
| 388 | } |
| 389 | buf = kmalloc(rlen + 1, gfp_flags); |
| 390 | if (!buf) { |
| 391 | dprintk("%s: Not enough memory\n", __func__); |
| 392 | goto out_free_netid; |
| 393 | } |
| 394 | buf[rlen] = '\0'; |
| 395 | memcpy(buf, p, rlen); |
| 396 | |
| 397 | /* replace port '.' with '-' */ |
| 398 | portstr = strrchr(buf, '.'); |
| 399 | if (!portstr) { |
| 400 | dprintk("%s: Failed finding expected dot in port\n", |
| 401 | __func__); |
| 402 | goto out_free_buf; |
| 403 | } |
| 404 | *portstr = '-'; |
| 405 | |
| 406 | /* find '.' between address and port */ |
| 407 | portstr = strrchr(buf, '.'); |
| 408 | if (!portstr) { |
| 409 | dprintk("%s: Failed finding expected dot between address and " |
| 410 | "port\n", __func__); |
| 411 | goto out_free_buf; |
| 412 | } |
| 413 | *portstr = '\0'; |
| 414 | |
| 415 | da = kzalloc(sizeof(*da), gfp_flags); |
| 416 | if (unlikely(!da)) |
| 417 | goto out_free_buf; |
| 418 | |
| 419 | INIT_LIST_HEAD(&da->da_node); |
| 420 | |
| 421 | if (!rpc_pton(net, buf, portstr-buf, (struct sockaddr *)&da->da_addr, |
| 422 | sizeof(da->da_addr))) { |
| 423 | dprintk("%s: error parsing address %s\n", __func__, buf); |
| 424 | goto out_free_da; |
| 425 | } |
| 426 | |
| 427 | portstr++; |
| 428 | sscanf(portstr, "%d-%d", &tmp[0], &tmp[1]); |
| 429 | port = htons((tmp[0] << 8) | (tmp[1])); |
| 430 | |
| 431 | switch (da->da_addr.ss_family) { |
| 432 | case AF_INET: |
| 433 | ((struct sockaddr_in *)&da->da_addr)->sin_port = port; |
| 434 | da->da_addrlen = sizeof(struct sockaddr_in); |
| 435 | match_netid = "tcp"; |
| 436 | match_netid_len = 3; |
| 437 | break; |
| 438 | |
| 439 | case AF_INET6: |
| 440 | ((struct sockaddr_in6 *)&da->da_addr)->sin6_port = port; |
| 441 | da->da_addrlen = sizeof(struct sockaddr_in6); |
| 442 | match_netid = "tcp6"; |
| 443 | match_netid_len = 4; |
| 444 | startsep = "["; |
| 445 | endsep = "]"; |
| 446 | break; |
| 447 | |
| 448 | default: |
| 449 | dprintk("%s: unsupported address family: %u\n", |
| 450 | __func__, da->da_addr.ss_family); |
| 451 | goto out_free_da; |
| 452 | } |
| 453 | |
| 454 | if (nlen != match_netid_len || strncmp(netid, match_netid, nlen)) { |
| 455 | dprintk("%s: ERROR: r_netid \"%s\" != \"%s\"\n", |
| 456 | __func__, netid, match_netid); |
| 457 | goto out_free_da; |
| 458 | } |
| 459 | |
| 460 | /* save human readable address */ |
| 461 | len = strlen(startsep) + strlen(buf) + strlen(endsep) + 7; |
| 462 | da->da_remotestr = kzalloc(len, gfp_flags); |
| 463 | |
| 464 | /* NULL is ok, only used for dprintk */ |
| 465 | if (da->da_remotestr) |
| 466 | snprintf(da->da_remotestr, len, "%s%s%s:%u", startsep, |
| 467 | buf, endsep, ntohs(port)); |
| 468 | |
| 469 | dprintk("%s: Parsed DS addr %s\n", __func__, da->da_remotestr); |
| 470 | kfree(buf); |
| 471 | kfree(netid); |
| 472 | return da; |
| 473 | |
| 474 | out_free_da: |
| 475 | kfree(da); |
| 476 | out_free_buf: |
| 477 | dprintk("%s: Error parsing DS addr: %s\n", __func__, buf); |
| 478 | kfree(buf); |
| 479 | out_free_netid: |
| 480 | kfree(netid); |
| 481 | out_err: |
| 482 | return NULL; |
| 483 | } |
| 484 | |
| 485 | /* Decode opaque device data and return the result */ |
| 486 | static struct nfs4_file_layout_dsaddr* |
| 487 | decode_device(struct inode *ino, struct pnfs_device *pdev, gfp_t gfp_flags) |
| 488 | { |
| 489 | int i; |
| 490 | u32 cnt, num; |
| 491 | u8 *indexp; |
| 492 | __be32 *p; |
| 493 | u8 *stripe_indices; |
| 494 | u8 max_stripe_index; |
| 495 | struct nfs4_file_layout_dsaddr *dsaddr = NULL; |
| 496 | struct xdr_stream stream; |
| 497 | struct xdr_buf buf; |
| 498 | struct page *scratch; |
| 499 | struct list_head dsaddrs; |
| 500 | struct nfs4_pnfs_ds_addr *da; |
| 501 | |
| 502 | /* set up xdr stream */ |
| 503 | scratch = alloc_page(gfp_flags); |
| 504 | if (!scratch) |
| 505 | goto out_err; |
| 506 | |
| 507 | xdr_init_decode_pages(&stream, &buf, pdev->pages, pdev->pglen); |
| 508 | xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE); |
| 509 | |
| 510 | /* Get the stripe count (number of stripe index) */ |
| 511 | p = xdr_inline_decode(&stream, 4); |
| 512 | if (unlikely(!p)) |
| 513 | goto out_err_free_scratch; |
| 514 | |
| 515 | cnt = be32_to_cpup(p); |
| 516 | dprintk("%s stripe count %d\n", __func__, cnt); |
| 517 | if (cnt > NFS4_PNFS_MAX_STRIPE_CNT) { |
| 518 | printk(KERN_WARNING "NFS: %s: stripe count %d greater than " |
| 519 | "supported maximum %d\n", __func__, |
| 520 | cnt, NFS4_PNFS_MAX_STRIPE_CNT); |
| 521 | goto out_err_free_scratch; |
| 522 | } |
| 523 | |
| 524 | /* read stripe indices */ |
| 525 | stripe_indices = kcalloc(cnt, sizeof(u8), gfp_flags); |
| 526 | if (!stripe_indices) |
| 527 | goto out_err_free_scratch; |
| 528 | |
| 529 | p = xdr_inline_decode(&stream, cnt << 2); |
| 530 | if (unlikely(!p)) |
| 531 | goto out_err_free_stripe_indices; |
| 532 | |
| 533 | indexp = &stripe_indices[0]; |
| 534 | max_stripe_index = 0; |
| 535 | for (i = 0; i < cnt; i++) { |
| 536 | *indexp = be32_to_cpup(p++); |
| 537 | max_stripe_index = max(max_stripe_index, *indexp); |
| 538 | indexp++; |
| 539 | } |
| 540 | |
| 541 | /* Check the multipath list count */ |
| 542 | p = xdr_inline_decode(&stream, 4); |
| 543 | if (unlikely(!p)) |
| 544 | goto out_err_free_stripe_indices; |
| 545 | |
| 546 | num = be32_to_cpup(p); |
| 547 | dprintk("%s ds_num %u\n", __func__, num); |
| 548 | if (num > NFS4_PNFS_MAX_MULTI_CNT) { |
| 549 | printk(KERN_WARNING "NFS: %s: multipath count %d greater than " |
| 550 | "supported maximum %d\n", __func__, |
| 551 | num, NFS4_PNFS_MAX_MULTI_CNT); |
| 552 | goto out_err_free_stripe_indices; |
| 553 | } |
| 554 | |
| 555 | /* validate stripe indices are all < num */ |
| 556 | if (max_stripe_index >= num) { |
| 557 | printk(KERN_WARNING "NFS: %s: stripe index %u >= num ds %u\n", |
| 558 | __func__, max_stripe_index, num); |
| 559 | goto out_err_free_stripe_indices; |
| 560 | } |
| 561 | |
| 562 | dsaddr = kzalloc(sizeof(*dsaddr) + |
| 563 | (sizeof(struct nfs4_pnfs_ds *) * (num - 1)), |
| 564 | gfp_flags); |
| 565 | if (!dsaddr) |
| 566 | goto out_err_free_stripe_indices; |
| 567 | |
| 568 | dsaddr->stripe_count = cnt; |
| 569 | dsaddr->stripe_indices = stripe_indices; |
| 570 | stripe_indices = NULL; |
| 571 | dsaddr->ds_num = num; |
| 572 | nfs4_init_deviceid_node(&dsaddr->id_node, |
| 573 | NFS_SERVER(ino)->pnfs_curr_ld, |
| 574 | NFS_SERVER(ino)->nfs_client, |
| 575 | &pdev->dev_id); |
| 576 | |
| 577 | INIT_LIST_HEAD(&dsaddrs); |
| 578 | |
| 579 | for (i = 0; i < dsaddr->ds_num; i++) { |
| 580 | int j; |
| 581 | u32 mp_count; |
| 582 | |
| 583 | p = xdr_inline_decode(&stream, 4); |
| 584 | if (unlikely(!p)) |
| 585 | goto out_err_free_deviceid; |
| 586 | |
| 587 | mp_count = be32_to_cpup(p); /* multipath count */ |
| 588 | for (j = 0; j < mp_count; j++) { |
| 589 | da = decode_ds_addr(NFS_SERVER(ino)->nfs_client->cl_net, |
| 590 | &stream, gfp_flags); |
| 591 | if (da) |
| 592 | list_add_tail(&da->da_node, &dsaddrs); |
| 593 | } |
| 594 | if (list_empty(&dsaddrs)) { |
| 595 | dprintk("%s: no suitable DS addresses found\n", |
| 596 | __func__); |
| 597 | goto out_err_free_deviceid; |
| 598 | } |
| 599 | |
| 600 | dsaddr->ds_list[i] = nfs4_pnfs_ds_add(&dsaddrs, gfp_flags); |
| 601 | if (!dsaddr->ds_list[i]) |
| 602 | goto out_err_drain_dsaddrs; |
| 603 | |
| 604 | /* If DS was already in cache, free ds addrs */ |
| 605 | while (!list_empty(&dsaddrs)) { |
| 606 | da = list_first_entry(&dsaddrs, |
| 607 | struct nfs4_pnfs_ds_addr, |
| 608 | da_node); |
| 609 | list_del_init(&da->da_node); |
| 610 | kfree(da->da_remotestr); |
| 611 | kfree(da); |
| 612 | } |
| 613 | } |
| 614 | |
| 615 | __free_page(scratch); |
| 616 | return dsaddr; |
| 617 | |
| 618 | out_err_drain_dsaddrs: |
| 619 | while (!list_empty(&dsaddrs)) { |
| 620 | da = list_first_entry(&dsaddrs, struct nfs4_pnfs_ds_addr, |
| 621 | da_node); |
| 622 | list_del_init(&da->da_node); |
| 623 | kfree(da->da_remotestr); |
| 624 | kfree(da); |
| 625 | } |
| 626 | out_err_free_deviceid: |
| 627 | nfs4_fl_free_deviceid(dsaddr); |
| 628 | /* stripe_indicies was part of dsaddr */ |
| 629 | goto out_err_free_scratch; |
| 630 | out_err_free_stripe_indices: |
| 631 | kfree(stripe_indices); |
| 632 | out_err_free_scratch: |
| 633 | __free_page(scratch); |
| 634 | out_err: |
| 635 | dprintk("%s ERROR: returning NULL\n", __func__); |
| 636 | return NULL; |
| 637 | } |
| 638 | |
| 639 | /* |
| 640 | * Decode the opaque device specified in 'dev' and add it to the cache of |
| 641 | * available devices. |
| 642 | */ |
| 643 | static struct nfs4_file_layout_dsaddr * |
| 644 | decode_and_add_device(struct inode *inode, struct pnfs_device *dev, gfp_t gfp_flags) |
| 645 | { |
| 646 | struct nfs4_deviceid_node *d; |
| 647 | struct nfs4_file_layout_dsaddr *n, *new; |
| 648 | |
| 649 | new = decode_device(inode, dev, gfp_flags); |
| 650 | if (!new) { |
| 651 | printk(KERN_WARNING "NFS: %s: Could not decode or add device\n", |
| 652 | __func__); |
| 653 | return NULL; |
| 654 | } |
| 655 | |
| 656 | d = nfs4_insert_deviceid_node(&new->id_node); |
| 657 | n = container_of(d, struct nfs4_file_layout_dsaddr, id_node); |
| 658 | if (n != new) { |
| 659 | nfs4_fl_free_deviceid(new); |
| 660 | return n; |
| 661 | } |
| 662 | |
| 663 | return new; |
| 664 | } |
| 665 | |
| 666 | /* |
| 667 | * Retrieve the information for dev_id, add it to the list |
| 668 | * of available devices, and return it. |
| 669 | */ |
| 670 | struct nfs4_file_layout_dsaddr * |
| 671 | filelayout_get_device_info(struct inode *inode, struct nfs4_deviceid *dev_id, gfp_t gfp_flags) |
| 672 | { |
| 673 | struct pnfs_device *pdev = NULL; |
| 674 | u32 max_resp_sz; |
| 675 | int max_pages; |
| 676 | struct page **pages = NULL; |
| 677 | struct nfs4_file_layout_dsaddr *dsaddr = NULL; |
| 678 | int rc, i; |
| 679 | struct nfs_server *server = NFS_SERVER(inode); |
| 680 | |
| 681 | /* |
| 682 | * Use the session max response size as the basis for setting |
| 683 | * GETDEVICEINFO's maxcount |
| 684 | */ |
| 685 | max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz; |
| 686 | max_pages = nfs_page_array_len(0, max_resp_sz); |
| 687 | dprintk("%s inode %p max_resp_sz %u max_pages %d\n", |
| 688 | __func__, inode, max_resp_sz, max_pages); |
| 689 | |
| 690 | pdev = kzalloc(sizeof(struct pnfs_device), gfp_flags); |
| 691 | if (pdev == NULL) |
| 692 | return NULL; |
| 693 | |
| 694 | pages = kzalloc(max_pages * sizeof(struct page *), gfp_flags); |
| 695 | if (pages == NULL) { |
| 696 | kfree(pdev); |
| 697 | return NULL; |
| 698 | } |
| 699 | for (i = 0; i < max_pages; i++) { |
| 700 | pages[i] = alloc_page(gfp_flags); |
| 701 | if (!pages[i]) |
| 702 | goto out_free; |
| 703 | } |
| 704 | |
| 705 | memcpy(&pdev->dev_id, dev_id, sizeof(*dev_id)); |
| 706 | pdev->layout_type = LAYOUT_NFSV4_1_FILES; |
| 707 | pdev->pages = pages; |
| 708 | pdev->pgbase = 0; |
| 709 | pdev->pglen = max_resp_sz; |
| 710 | pdev->mincount = 0; |
| 711 | |
| 712 | rc = nfs4_proc_getdeviceinfo(server, pdev); |
| 713 | dprintk("%s getdevice info returns %d\n", __func__, rc); |
| 714 | if (rc) |
| 715 | goto out_free; |
| 716 | |
| 717 | /* |
| 718 | * Found new device, need to decode it and then add it to the |
| 719 | * list of known devices for this mountpoint. |
| 720 | */ |
| 721 | dsaddr = decode_and_add_device(inode, pdev, gfp_flags); |
| 722 | out_free: |
| 723 | for (i = 0; i < max_pages; i++) |
| 724 | __free_page(pages[i]); |
| 725 | kfree(pages); |
| 726 | kfree(pdev); |
| 727 | dprintk("<-- %s dsaddr %p\n", __func__, dsaddr); |
| 728 | return dsaddr; |
| 729 | } |
| 730 | |
| 731 | void |
| 732 | nfs4_fl_put_deviceid(struct nfs4_file_layout_dsaddr *dsaddr) |
| 733 | { |
| 734 | nfs4_put_deviceid_node(&dsaddr->id_node); |
| 735 | } |
| 736 | |
| 737 | /* |
| 738 | * Want res = (offset - layout->pattern_offset)/ layout->stripe_unit |
| 739 | * Then: ((res + fsi) % dsaddr->stripe_count) |
| 740 | */ |
| 741 | u32 |
| 742 | nfs4_fl_calc_j_index(struct pnfs_layout_segment *lseg, loff_t offset) |
| 743 | { |
| 744 | struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg); |
| 745 | u64 tmp; |
| 746 | |
| 747 | tmp = offset - flseg->pattern_offset; |
| 748 | do_div(tmp, flseg->stripe_unit); |
| 749 | tmp += flseg->first_stripe_index; |
| 750 | return do_div(tmp, flseg->dsaddr->stripe_count); |
| 751 | } |
| 752 | |
| 753 | u32 |
| 754 | nfs4_fl_calc_ds_index(struct pnfs_layout_segment *lseg, u32 j) |
| 755 | { |
| 756 | return FILELAYOUT_LSEG(lseg)->dsaddr->stripe_indices[j]; |
| 757 | } |
| 758 | |
| 759 | struct nfs_fh * |
| 760 | nfs4_fl_select_ds_fh(struct pnfs_layout_segment *lseg, u32 j) |
| 761 | { |
| 762 | struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg); |
| 763 | u32 i; |
| 764 | |
| 765 | if (flseg->stripe_type == STRIPE_SPARSE) { |
| 766 | if (flseg->num_fh == 1) |
| 767 | i = 0; |
| 768 | else if (flseg->num_fh == 0) |
| 769 | /* Use the MDS OPEN fh set in nfs_read_rpcsetup */ |
| 770 | return NULL; |
| 771 | else |
| 772 | i = nfs4_fl_calc_ds_index(lseg, j); |
| 773 | } else |
| 774 | i = j; |
| 775 | return flseg->fh_array[i]; |
| 776 | } |
| 777 | |
| 778 | static void nfs4_wait_ds_connect(struct nfs4_pnfs_ds *ds) |
| 779 | { |
| 780 | might_sleep(); |
| 781 | wait_on_bit(&ds->ds_state, NFS4DS_CONNECTING, |
| 782 | nfs_wait_bit_killable, TASK_KILLABLE); |
| 783 | } |
| 784 | |
| 785 | static void nfs4_clear_ds_conn_bit(struct nfs4_pnfs_ds *ds) |
| 786 | { |
| 787 | smp_mb__before_clear_bit(); |
| 788 | clear_bit(NFS4DS_CONNECTING, &ds->ds_state); |
| 789 | smp_mb__after_clear_bit(); |
| 790 | wake_up_bit(&ds->ds_state, NFS4DS_CONNECTING); |
| 791 | } |
| 792 | |
| 793 | |
| 794 | struct nfs4_pnfs_ds * |
| 795 | nfs4_fl_prepare_ds(struct pnfs_layout_segment *lseg, u32 ds_idx) |
| 796 | { |
| 797 | struct nfs4_file_layout_dsaddr *dsaddr = FILELAYOUT_LSEG(lseg)->dsaddr; |
| 798 | struct nfs4_pnfs_ds *ds = dsaddr->ds_list[ds_idx]; |
| 799 | struct nfs4_deviceid_node *devid = FILELAYOUT_DEVID_NODE(lseg); |
| 800 | struct nfs4_pnfs_ds *ret = ds; |
| 801 | |
| 802 | if (ds == NULL) { |
| 803 | printk(KERN_ERR "NFS: %s: No data server for offset index %d\n", |
| 804 | __func__, ds_idx); |
| 805 | filelayout_mark_devid_invalid(devid); |
| 806 | goto out; |
| 807 | } |
| 808 | if (ds->ds_clp) |
| 809 | goto out_test_devid; |
| 810 | |
| 811 | if (test_and_set_bit(NFS4DS_CONNECTING, &ds->ds_state) == 0) { |
| 812 | struct nfs_server *s = NFS_SERVER(lseg->pls_layout->plh_inode); |
| 813 | int err; |
| 814 | |
| 815 | err = nfs4_ds_connect(s, ds); |
| 816 | if (err) |
| 817 | nfs4_mark_deviceid_unavailable(devid); |
| 818 | nfs4_clear_ds_conn_bit(ds); |
| 819 | } else { |
| 820 | /* Either ds is connected, or ds is NULL */ |
| 821 | nfs4_wait_ds_connect(ds); |
| 822 | } |
| 823 | out_test_devid: |
| 824 | if (ret->ds_clp == NULL || |
| 825 | filelayout_test_devid_unavailable(devid)) |
| 826 | ret = NULL; |
| 827 | out: |
| 828 | return ret; |
| 829 | } |
| 830 | |
| 831 | module_param(dataserver_retrans, uint, 0644); |
| 832 | MODULE_PARM_DESC(dataserver_retrans, "The number of times the NFSv4.1 client " |
| 833 | "retries a request before it attempts further " |
| 834 | " recovery action."); |
| 835 | module_param(dataserver_timeo, uint, 0644); |
| 836 | MODULE_PARM_DESC(dataserver_timeo, "The time (in tenths of a second) the " |
| 837 | "NFSv4.1 client waits for a response from a " |
| 838 | " data server before it retries an NFS request."); |