1 /* 2 * Device operations for the pnfs nfs4 file layout driver. 3 * 4 * Copyright (c) 2014, Primary Data, Inc. All rights reserved. 5 * 6 * Tao Peng <bergwolf@primarydata.com> 7 */ 8 9 #include <linux/nfs_fs.h> 10 #include <linux/vmalloc.h> 11 #include <linux/module.h> 12 #include <linux/sunrpc/addr.h> 13 14 #include "../internal.h" 15 #include "../nfs4session.h" 16 #include "flexfilelayout.h" 17 18 #define NFSDBG_FACILITY NFSDBG_PNFS_LD 19 20 static unsigned int dataserver_timeo = NFS_DEF_TCP_RETRANS; 21 static unsigned int dataserver_retrans; 22 23 void nfs4_ff_layout_put_deviceid(struct nfs4_ff_layout_ds *mirror_ds) 24 { 25 if (mirror_ds) 26 nfs4_put_deviceid_node(&mirror_ds->id_node); 27 } 28 29 void nfs4_ff_layout_free_deviceid(struct nfs4_ff_layout_ds *mirror_ds) 30 { 31 nfs4_print_deviceid(&mirror_ds->id_node.deviceid); 32 nfs4_pnfs_ds_put(mirror_ds->ds); 33 kfree_rcu(mirror_ds, id_node.rcu); 34 } 35 36 /* Decode opaque device data and construct new_ds using it */ 37 struct nfs4_ff_layout_ds * 38 nfs4_ff_alloc_deviceid_node(struct nfs_server *server, struct pnfs_device *pdev, 39 gfp_t gfp_flags) 40 { 41 struct xdr_stream stream; 42 struct xdr_buf buf; 43 struct page *scratch; 44 struct list_head dsaddrs; 45 struct nfs4_pnfs_ds_addr *da; 46 struct nfs4_ff_layout_ds *new_ds = NULL; 47 struct nfs4_ff_ds_version *ds_versions = NULL; 48 u32 mp_count; 49 u32 version_count; 50 __be32 *p; 51 int i, ret = -ENOMEM; 52 53 /* set up xdr stream */ 54 scratch = alloc_page(gfp_flags); 55 if (!scratch) 56 goto out_err; 57 58 new_ds = kzalloc(sizeof(struct nfs4_ff_layout_ds), gfp_flags); 59 if (!new_ds) 60 goto out_scratch; 61 62 nfs4_init_deviceid_node(&new_ds->id_node, 63 server, 64 &pdev->dev_id); 65 INIT_LIST_HEAD(&dsaddrs); 66 67 xdr_init_decode_pages(&stream, &buf, pdev->pages, pdev->pglen); 68 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE); 69 70 /* multipath count */ 71 p = xdr_inline_decode(&stream, 4); 72 if (unlikely(!p)) 73 goto out_err_drain_dsaddrs; 74 mp_count = be32_to_cpup(p); 75 dprintk("%s: multipath ds count %d\n", __func__, mp_count); 76 77 for (i = 0; i < mp_count; i++) { 78 /* multipath ds */ 79 da = nfs4_decode_mp_ds_addr(server->nfs_client->cl_net, 80 &stream, gfp_flags); 81 if (da) 82 list_add_tail(&da->da_node, &dsaddrs); 83 } 84 if (list_empty(&dsaddrs)) { 85 dprintk("%s: no suitable DS addresses found\n", 86 __func__); 87 ret = -ENOMEDIUM; 88 goto out_err_drain_dsaddrs; 89 } 90 91 /* version count */ 92 p = xdr_inline_decode(&stream, 4); 93 if (unlikely(!p)) 94 goto out_err_drain_dsaddrs; 95 version_count = be32_to_cpup(p); 96 dprintk("%s: version count %d\n", __func__, version_count); 97 98 ds_versions = kzalloc(version_count * sizeof(struct nfs4_ff_ds_version), 99 gfp_flags); 100 if (!ds_versions) 101 goto out_scratch; 102 103 for (i = 0; i < version_count; i++) { 104 /* 20 = version(4) + minor_version(4) + rsize(4) + wsize(4) + 105 * tightly_coupled(4) */ 106 p = xdr_inline_decode(&stream, 20); 107 if (unlikely(!p)) 108 goto out_err_drain_dsaddrs; 109 ds_versions[i].version = be32_to_cpup(p++); 110 ds_versions[i].minor_version = be32_to_cpup(p++); 111 ds_versions[i].rsize = nfs_block_size(be32_to_cpup(p++), NULL); 112 ds_versions[i].wsize = nfs_block_size(be32_to_cpup(p++), NULL); 113 ds_versions[i].tightly_coupled = be32_to_cpup(p); 114 115 if (ds_versions[i].rsize > NFS_MAX_FILE_IO_SIZE) 116 ds_versions[i].rsize = NFS_MAX_FILE_IO_SIZE; 117 if (ds_versions[i].wsize > NFS_MAX_FILE_IO_SIZE) 118 ds_versions[i].wsize = NFS_MAX_FILE_IO_SIZE; 119 120 if (ds_versions[i].version != 3 || ds_versions[i].minor_version != 0) { 121 dprintk("%s: [%d] unsupported ds version %d-%d\n", __func__, 122 i, ds_versions[i].version, 123 ds_versions[i].minor_version); 124 ret = -EPROTONOSUPPORT; 125 goto out_err_drain_dsaddrs; 126 } 127 128 dprintk("%s: [%d] vers %u minor_ver %u rsize %u wsize %u coupled %d\n", 129 __func__, i, ds_versions[i].version, 130 ds_versions[i].minor_version, 131 ds_versions[i].rsize, 132 ds_versions[i].wsize, 133 ds_versions[i].tightly_coupled); 134 } 135 136 new_ds->ds_versions = ds_versions; 137 new_ds->ds_versions_cnt = version_count; 138 139 new_ds->ds = nfs4_pnfs_ds_add(&dsaddrs, gfp_flags); 140 if (!new_ds->ds) 141 goto out_err_drain_dsaddrs; 142 143 /* If DS was already in cache, free ds addrs */ 144 while (!list_empty(&dsaddrs)) { 145 da = list_first_entry(&dsaddrs, 146 struct nfs4_pnfs_ds_addr, 147 da_node); 148 list_del_init(&da->da_node); 149 kfree(da->da_remotestr); 150 kfree(da); 151 } 152 153 __free_page(scratch); 154 return new_ds; 155 156 out_err_drain_dsaddrs: 157 while (!list_empty(&dsaddrs)) { 158 da = list_first_entry(&dsaddrs, struct nfs4_pnfs_ds_addr, 159 da_node); 160 list_del_init(&da->da_node); 161 kfree(da->da_remotestr); 162 kfree(da); 163 } 164 165 kfree(ds_versions); 166 out_scratch: 167 __free_page(scratch); 168 out_err: 169 kfree(new_ds); 170 171 dprintk("%s ERROR: returning %d\n", __func__, ret); 172 return NULL; 173 } 174 175 static void ff_layout_mark_devid_invalid(struct pnfs_layout_segment *lseg, 176 struct nfs4_deviceid_node *devid) 177 { 178 nfs4_mark_deviceid_unavailable(devid); 179 if (!ff_layout_has_available_ds(lseg)) 180 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode, 181 lseg); 182 } 183 184 static bool ff_layout_mirror_valid(struct pnfs_layout_segment *lseg, 185 struct nfs4_ff_layout_mirror *mirror) 186 { 187 if (mirror == NULL || mirror->mirror_ds == NULL) { 188 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode, 189 lseg); 190 return false; 191 } 192 if (mirror->mirror_ds->ds == NULL) { 193 struct nfs4_deviceid_node *devid; 194 devid = &mirror->mirror_ds->id_node; 195 ff_layout_mark_devid_invalid(lseg, devid); 196 return false; 197 } 198 return true; 199 } 200 201 static u64 202 end_offset(u64 start, u64 len) 203 { 204 u64 end; 205 206 end = start + len; 207 return end >= start ? end : NFS4_MAX_UINT64; 208 } 209 210 static void extend_ds_error(struct nfs4_ff_layout_ds_err *err, 211 u64 offset, u64 length) 212 { 213 u64 end; 214 215 end = max_t(u64, end_offset(err->offset, err->length), 216 end_offset(offset, length)); 217 err->offset = min_t(u64, err->offset, offset); 218 err->length = end - err->offset; 219 } 220 221 static int 222 ff_ds_error_match(const struct nfs4_ff_layout_ds_err *e1, 223 const struct nfs4_ff_layout_ds_err *e2) 224 { 225 int ret; 226 227 if (e1->opnum != e2->opnum) 228 return e1->opnum < e2->opnum ? -1 : 1; 229 if (e1->status != e2->status) 230 return e1->status < e2->status ? -1 : 1; 231 ret = memcmp(e1->stateid.data, e2->stateid.data, 232 sizeof(e1->stateid.data)); 233 if (ret != 0) 234 return ret; 235 ret = memcmp(&e1->deviceid, &e2->deviceid, sizeof(e1->deviceid)); 236 if (ret != 0) 237 return ret; 238 if (end_offset(e1->offset, e1->length) < e2->offset) 239 return -1; 240 if (e1->offset > end_offset(e2->offset, e2->length)) 241 return 1; 242 /* If ranges overlap or are contiguous, they are the same */ 243 return 0; 244 } 245 246 static void 247 ff_layout_add_ds_error_locked(struct nfs4_flexfile_layout *flo, 248 struct nfs4_ff_layout_ds_err *dserr) 249 { 250 struct nfs4_ff_layout_ds_err *err, *tmp; 251 struct list_head *head = &flo->error_list; 252 int match; 253 254 /* Do insertion sort w/ merges */ 255 list_for_each_entry_safe(err, tmp, &flo->error_list, list) { 256 match = ff_ds_error_match(err, dserr); 257 if (match < 0) 258 continue; 259 if (match > 0) { 260 /* Add entry "dserr" _before_ entry "err" */ 261 head = &err->list; 262 break; 263 } 264 /* Entries match, so merge "err" into "dserr" */ 265 extend_ds_error(dserr, err->offset, err->length); 266 list_del(&err->list); 267 kfree(err); 268 } 269 270 list_add_tail(&dserr->list, head); 271 } 272 273 int ff_layout_track_ds_error(struct nfs4_flexfile_layout *flo, 274 struct nfs4_ff_layout_mirror *mirror, u64 offset, 275 u64 length, int status, enum nfs_opnum4 opnum, 276 gfp_t gfp_flags) 277 { 278 struct nfs4_ff_layout_ds_err *dserr; 279 280 if (status == 0) 281 return 0; 282 283 if (mirror->mirror_ds == NULL) 284 return -EINVAL; 285 286 dserr = kmalloc(sizeof(*dserr), gfp_flags); 287 if (!dserr) 288 return -ENOMEM; 289 290 INIT_LIST_HEAD(&dserr->list); 291 dserr->offset = offset; 292 dserr->length = length; 293 dserr->status = status; 294 dserr->opnum = opnum; 295 nfs4_stateid_copy(&dserr->stateid, &mirror->stateid); 296 memcpy(&dserr->deviceid, &mirror->mirror_ds->id_node.deviceid, 297 NFS4_DEVICEID4_SIZE); 298 299 spin_lock(&flo->generic_hdr.plh_inode->i_lock); 300 ff_layout_add_ds_error_locked(flo, dserr); 301 spin_unlock(&flo->generic_hdr.plh_inode->i_lock); 302 303 return 0; 304 } 305 306 static struct rpc_cred * 307 ff_layout_get_mirror_cred(struct nfs4_ff_layout_mirror *mirror, u32 iomode) 308 { 309 struct rpc_cred *cred, __rcu **pcred; 310 311 if (iomode == IOMODE_READ) 312 pcred = &mirror->ro_cred; 313 else 314 pcred = &mirror->rw_cred; 315 316 rcu_read_lock(); 317 do { 318 cred = rcu_dereference(*pcred); 319 if (!cred) 320 break; 321 322 cred = get_rpccred_rcu(cred); 323 } while(!cred); 324 rcu_read_unlock(); 325 return cred; 326 } 327 328 struct nfs_fh * 329 nfs4_ff_layout_select_ds_fh(struct pnfs_layout_segment *lseg, u32 mirror_idx) 330 { 331 struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, mirror_idx); 332 struct nfs_fh *fh = NULL; 333 334 if (!ff_layout_mirror_valid(lseg, mirror)) { 335 pr_err_ratelimited("NFS: %s: No data server for mirror offset index %d\n", 336 __func__, mirror_idx); 337 goto out; 338 } 339 340 /* FIXME: For now assume there is only 1 version available for the DS */ 341 fh = &mirror->fh_versions[0]; 342 out: 343 return fh; 344 } 345 346 /** 347 * nfs4_ff_layout_prepare_ds - prepare a DS connection for an RPC call 348 * @lseg: the layout segment we're operating on 349 * @ds_idx: index of the DS to use 350 * @fail_return: return layout on connect failure? 351 * 352 * Try to prepare a DS connection to accept an RPC call. This involves 353 * selecting a mirror to use and connecting the client to it if it's not 354 * already connected. 355 * 356 * Since we only need a single functioning mirror to satisfy a read, we don't 357 * want to return the layout if there is one. For writes though, any down 358 * mirror should result in a LAYOUTRETURN. @fail_return is how we distinguish 359 * between the two cases. 360 * 361 * Returns a pointer to a connected DS object on success or NULL on failure. 362 */ 363 struct nfs4_pnfs_ds * 364 nfs4_ff_layout_prepare_ds(struct pnfs_layout_segment *lseg, u32 ds_idx, 365 bool fail_return) 366 { 367 struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx); 368 struct nfs4_pnfs_ds *ds = NULL; 369 struct nfs4_deviceid_node *devid; 370 struct inode *ino = lseg->pls_layout->plh_inode; 371 struct nfs_server *s = NFS_SERVER(ino); 372 unsigned int max_payload; 373 374 if (!ff_layout_mirror_valid(lseg, mirror)) { 375 pr_err_ratelimited("NFS: %s: No data server for offset index %d\n", 376 __func__, ds_idx); 377 goto out; 378 } 379 380 devid = &mirror->mirror_ds->id_node; 381 if (ff_layout_test_devid_unavailable(devid)) 382 goto out_fail; 383 384 ds = mirror->mirror_ds->ds; 385 /* matching smp_wmb() in _nfs4_pnfs_v3/4_ds_connect */ 386 smp_rmb(); 387 if (ds->ds_clp) 388 goto out; 389 390 /* FIXME: For now we assume the server sent only one version of NFS 391 * to use for the DS. 392 */ 393 nfs4_pnfs_ds_connect(s, ds, devid, dataserver_timeo, 394 dataserver_retrans, 395 mirror->mirror_ds->ds_versions[0].version, 396 mirror->mirror_ds->ds_versions[0].minor_version, 397 RPC_AUTH_UNIX); 398 399 /* connect success, check rsize/wsize limit */ 400 if (ds->ds_clp) { 401 max_payload = 402 nfs_block_size(rpc_max_payload(ds->ds_clp->cl_rpcclient), 403 NULL); 404 if (mirror->mirror_ds->ds_versions[0].rsize > max_payload) 405 mirror->mirror_ds->ds_versions[0].rsize = max_payload; 406 if (mirror->mirror_ds->ds_versions[0].wsize > max_payload) 407 mirror->mirror_ds->ds_versions[0].wsize = max_payload; 408 goto out; 409 } 410 ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout), 411 mirror, lseg->pls_range.offset, 412 lseg->pls_range.length, NFS4ERR_NXIO, 413 OP_ILLEGAL, GFP_NOIO); 414 out_fail: 415 if (fail_return || !ff_layout_has_available_ds(lseg)) 416 pnfs_error_mark_layout_for_return(ino, lseg); 417 ds = NULL; 418 out: 419 return ds; 420 } 421 422 struct rpc_cred * 423 ff_layout_get_ds_cred(struct pnfs_layout_segment *lseg, u32 ds_idx, 424 struct rpc_cred *mdscred) 425 { 426 struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx); 427 struct rpc_cred *cred; 428 429 if (mirror) { 430 cred = ff_layout_get_mirror_cred(mirror, lseg->pls_range.iomode); 431 if (!cred) 432 cred = get_rpccred(mdscred); 433 } else { 434 cred = get_rpccred(mdscred); 435 } 436 return cred; 437 } 438 439 /** 440 * Find or create a DS rpc client with th MDS server rpc client auth flavor 441 * in the nfs_client cl_ds_clients list. 442 */ 443 struct rpc_clnt * 444 nfs4_ff_find_or_create_ds_client(struct pnfs_layout_segment *lseg, u32 ds_idx, 445 struct nfs_client *ds_clp, struct inode *inode) 446 { 447 struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx); 448 449 switch (mirror->mirror_ds->ds_versions[0].version) { 450 case 3: 451 /* For NFSv3 DS, flavor is set when creating DS connections */ 452 return ds_clp->cl_rpcclient; 453 case 4: 454 return nfs4_find_or_create_ds_client(ds_clp, inode); 455 default: 456 BUG(); 457 } 458 } 459 460 static bool is_range_intersecting(u64 offset1, u64 length1, 461 u64 offset2, u64 length2) 462 { 463 u64 end1 = end_offset(offset1, length1); 464 u64 end2 = end_offset(offset2, length2); 465 466 return (end1 == NFS4_MAX_UINT64 || end1 > offset2) && 467 (end2 == NFS4_MAX_UINT64 || end2 > offset1); 468 } 469 470 /* called with inode i_lock held */ 471 int ff_layout_encode_ds_ioerr(struct nfs4_flexfile_layout *flo, 472 struct xdr_stream *xdr, int *count, 473 const struct pnfs_layout_range *range) 474 { 475 struct nfs4_ff_layout_ds_err *err, *n; 476 __be32 *p; 477 478 list_for_each_entry_safe(err, n, &flo->error_list, list) { 479 if (!is_range_intersecting(err->offset, err->length, 480 range->offset, range->length)) 481 continue; 482 /* offset(8) + length(8) + stateid(NFS4_STATEID_SIZE) 483 * + array length + deviceid(NFS4_DEVICEID4_SIZE) 484 * + status(4) + opnum(4) 485 */ 486 p = xdr_reserve_space(xdr, 487 28 + NFS4_STATEID_SIZE + NFS4_DEVICEID4_SIZE); 488 if (unlikely(!p)) 489 return -ENOBUFS; 490 p = xdr_encode_hyper(p, err->offset); 491 p = xdr_encode_hyper(p, err->length); 492 p = xdr_encode_opaque_fixed(p, &err->stateid, 493 NFS4_STATEID_SIZE); 494 /* Encode 1 error */ 495 *p++ = cpu_to_be32(1); 496 p = xdr_encode_opaque_fixed(p, &err->deviceid, 497 NFS4_DEVICEID4_SIZE); 498 *p++ = cpu_to_be32(err->status); 499 *p++ = cpu_to_be32(err->opnum); 500 *count += 1; 501 list_del(&err->list); 502 dprintk("%s: offset %llu length %llu status %d op %d count %d\n", 503 __func__, err->offset, err->length, err->status, 504 err->opnum, *count); 505 kfree(err); 506 } 507 508 return 0; 509 } 510 511 static bool ff_read_layout_has_available_ds(struct pnfs_layout_segment *lseg) 512 { 513 struct nfs4_ff_layout_mirror *mirror; 514 struct nfs4_deviceid_node *devid; 515 u32 idx; 516 517 for (idx = 0; idx < FF_LAYOUT_MIRROR_COUNT(lseg); idx++) { 518 mirror = FF_LAYOUT_COMP(lseg, idx); 519 if (mirror && mirror->mirror_ds) { 520 devid = &mirror->mirror_ds->id_node; 521 if (!ff_layout_test_devid_unavailable(devid)) 522 return true; 523 } 524 } 525 526 return false; 527 } 528 529 static bool ff_rw_layout_has_available_ds(struct pnfs_layout_segment *lseg) 530 { 531 struct nfs4_ff_layout_mirror *mirror; 532 struct nfs4_deviceid_node *devid; 533 u32 idx; 534 535 for (idx = 0; idx < FF_LAYOUT_MIRROR_COUNT(lseg); idx++) { 536 mirror = FF_LAYOUT_COMP(lseg, idx); 537 if (!mirror || !mirror->mirror_ds) 538 return false; 539 devid = &mirror->mirror_ds->id_node; 540 if (ff_layout_test_devid_unavailable(devid)) 541 return false; 542 } 543 544 return FF_LAYOUT_MIRROR_COUNT(lseg) != 0; 545 } 546 547 bool ff_layout_has_available_ds(struct pnfs_layout_segment *lseg) 548 { 549 if (lseg->pls_range.iomode == IOMODE_READ) 550 return ff_read_layout_has_available_ds(lseg); 551 /* Note: RW layout needs all mirrors available */ 552 return ff_rw_layout_has_available_ds(lseg); 553 } 554 555 bool ff_layout_avoid_mds_available_ds(struct pnfs_layout_segment *lseg) 556 { 557 return ff_layout_no_fallback_to_mds(lseg) || 558 ff_layout_has_available_ds(lseg); 559 } 560 561 bool ff_layout_avoid_read_on_rw(struct pnfs_layout_segment *lseg) 562 { 563 return lseg->pls_range.iomode == IOMODE_RW && 564 ff_layout_no_read_on_rw(lseg); 565 } 566 567 module_param(dataserver_retrans, uint, 0644); 568 MODULE_PARM_DESC(dataserver_retrans, "The number of times the NFSv4.1 client " 569 "retries a request before it attempts further " 570 " recovery action."); 571 module_param(dataserver_timeo, uint, 0644); 572 MODULE_PARM_DESC(dataserver_timeo, "The time (in tenths of a second) the " 573 "NFSv4.1 client waits for a response from a " 574 " data server before it retries an NFS request."); 575