1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Common NFS I/O operations for the pnfs file based 4 * layout drivers. 5 * 6 * Copyright (c) 2014, Primary Data, Inc. All rights reserved. 7 * 8 * Tom Haynes <loghyr@primarydata.com> 9 */ 10 11 #include <linux/nfs_fs.h> 12 #include <linux/nfs_page.h> 13 #include <linux/sunrpc/addr.h> 14 #include <linux/module.h> 15 16 #include "nfs4session.h" 17 #include "internal.h" 18 #include "pnfs.h" 19 20 #define NFSDBG_FACILITY NFSDBG_PNFS 21 22 void pnfs_generic_rw_release(void *data) 23 { 24 struct nfs_pgio_header *hdr = data; 25 26 nfs_put_client(hdr->ds_clp); 27 hdr->mds_ops->rpc_release(data); 28 } 29 EXPORT_SYMBOL_GPL(pnfs_generic_rw_release); 30 31 /* Fake up some data that will cause nfs_commit_release to retry the writes. */ 32 void pnfs_generic_prepare_to_resend_writes(struct nfs_commit_data *data) 33 { 34 struct nfs_writeverf *verf = data->res.verf; 35 36 data->task.tk_status = 0; 37 memset(&verf->verifier, 0, sizeof(verf->verifier)); 38 verf->committed = NFS_UNSTABLE; 39 } 40 EXPORT_SYMBOL_GPL(pnfs_generic_prepare_to_resend_writes); 41 42 void pnfs_generic_write_commit_done(struct rpc_task *task, void *data) 43 { 44 struct nfs_commit_data *wdata = data; 45 46 /* Note this may cause RPC to be resent */ 47 wdata->mds_ops->rpc_call_done(task, data); 48 } 49 EXPORT_SYMBOL_GPL(pnfs_generic_write_commit_done); 50 51 void pnfs_generic_commit_release(void *calldata) 52 { 53 struct nfs_commit_data *data = calldata; 54 55 data->completion_ops->completion(data); 56 pnfs_put_lseg(data->lseg); 57 nfs_put_client(data->ds_clp); 58 nfs_commitdata_release(data); 59 } 60 EXPORT_SYMBOL_GPL(pnfs_generic_commit_release); 61 62 static struct pnfs_layout_segment * 63 pnfs_free_bucket_lseg(struct pnfs_commit_bucket *bucket) 64 { 65 if (list_empty(&bucket->committing) && list_empty(&bucket->written)) { 66 struct pnfs_layout_segment *freeme = bucket->lseg; 67 bucket->lseg = NULL; 68 return freeme; 69 } 70 return NULL; 71 } 72 73 /* The generic layer is about to remove the req from the commit list. 74 * If this will make the bucket empty, it will need to put the lseg reference. 75 * Note this must be called holding nfsi->commit_mutex 76 */ 77 void 78 pnfs_generic_clear_request_commit(struct nfs_page *req, 79 struct nfs_commit_info *cinfo) 80 { 81 struct pnfs_layout_segment *freeme = NULL; 82 83 if (!test_and_clear_bit(PG_COMMIT_TO_DS, &req->wb_flags)) 84 goto out; 85 cinfo->ds->nwritten--; 86 if (list_is_singular(&req->wb_list)) { 87 struct pnfs_commit_bucket *bucket; 88 89 bucket = list_first_entry(&req->wb_list, 90 struct pnfs_commit_bucket, 91 written); 92 freeme = pnfs_free_bucket_lseg(bucket); 93 } 94 out: 95 nfs_request_remove_commit_list(req, cinfo); 96 pnfs_put_lseg(freeme); 97 } 98 EXPORT_SYMBOL_GPL(pnfs_generic_clear_request_commit); 99 100 struct pnfs_commit_array * 101 pnfs_alloc_commit_array(size_t n, gfp_t gfp_flags) 102 { 103 struct pnfs_commit_array *p; 104 struct pnfs_commit_bucket *b; 105 106 p = kmalloc(struct_size(p, buckets, n), gfp_flags); 107 if (!p) 108 return NULL; 109 p->nbuckets = n; 110 INIT_LIST_HEAD(&p->cinfo_list); 111 INIT_LIST_HEAD(&p->lseg_list); 112 p->lseg = NULL; 113 for (b = &p->buckets[0]; n != 0; b++, n--) { 114 INIT_LIST_HEAD(&b->written); 115 INIT_LIST_HEAD(&b->committing); 116 b->lseg = NULL; 117 b->direct_verf.committed = NFS_INVALID_STABLE_HOW; 118 } 119 return p; 120 } 121 EXPORT_SYMBOL_GPL(pnfs_alloc_commit_array); 122 123 void 124 pnfs_free_commit_array(struct pnfs_commit_array *p) 125 { 126 kfree_rcu(p, rcu); 127 } 128 EXPORT_SYMBOL_GPL(pnfs_free_commit_array); 129 130 static struct pnfs_commit_array * 131 pnfs_find_commit_array_by_lseg(struct pnfs_ds_commit_info *fl_cinfo, 132 struct pnfs_layout_segment *lseg) 133 { 134 struct pnfs_commit_array *array; 135 136 list_for_each_entry_rcu(array, &fl_cinfo->commits, cinfo_list) { 137 if (array->lseg == lseg) 138 return array; 139 } 140 return NULL; 141 } 142 143 struct pnfs_commit_array * 144 pnfs_add_commit_array(struct pnfs_ds_commit_info *fl_cinfo, 145 struct pnfs_commit_array *new, 146 struct pnfs_layout_segment *lseg) 147 { 148 struct pnfs_commit_array *array; 149 150 array = pnfs_find_commit_array_by_lseg(fl_cinfo, lseg); 151 if (array) 152 return array; 153 new->lseg = lseg; 154 refcount_set(&new->refcount, 1); 155 list_add_rcu(&new->cinfo_list, &fl_cinfo->commits); 156 list_add(&new->lseg_list, &lseg->pls_commits); 157 return new; 158 } 159 EXPORT_SYMBOL_GPL(pnfs_add_commit_array); 160 161 static struct pnfs_commit_array * 162 pnfs_lookup_commit_array(struct pnfs_ds_commit_info *fl_cinfo, 163 struct pnfs_layout_segment *lseg) 164 { 165 struct pnfs_commit_array *array; 166 167 rcu_read_lock(); 168 array = pnfs_find_commit_array_by_lseg(fl_cinfo, lseg); 169 if (!array) { 170 rcu_read_unlock(); 171 fl_cinfo->ops->setup_ds_info(fl_cinfo, lseg); 172 rcu_read_lock(); 173 array = pnfs_find_commit_array_by_lseg(fl_cinfo, lseg); 174 } 175 rcu_read_unlock(); 176 return array; 177 } 178 179 static void 180 pnfs_release_commit_array_locked(struct pnfs_commit_array *array) 181 { 182 list_del_rcu(&array->cinfo_list); 183 list_del(&array->lseg_list); 184 pnfs_free_commit_array(array); 185 } 186 187 static void 188 pnfs_put_commit_array_locked(struct pnfs_commit_array *array) 189 { 190 if (refcount_dec_and_test(&array->refcount)) 191 pnfs_release_commit_array_locked(array); 192 } 193 194 static void 195 pnfs_put_commit_array(struct pnfs_commit_array *array, struct inode *inode) 196 { 197 if (refcount_dec_and_lock(&array->refcount, &inode->i_lock)) { 198 pnfs_release_commit_array_locked(array); 199 spin_unlock(&inode->i_lock); 200 } 201 } 202 203 static struct pnfs_commit_array * 204 pnfs_get_commit_array(struct pnfs_commit_array *array) 205 { 206 if (refcount_inc_not_zero(&array->refcount)) 207 return array; 208 return NULL; 209 } 210 211 static void 212 pnfs_remove_and_free_commit_array(struct pnfs_commit_array *array) 213 { 214 array->lseg = NULL; 215 list_del_init(&array->lseg_list); 216 pnfs_put_commit_array_locked(array); 217 } 218 219 void 220 pnfs_generic_ds_cinfo_release_lseg(struct pnfs_ds_commit_info *fl_cinfo, 221 struct pnfs_layout_segment *lseg) 222 { 223 struct pnfs_commit_array *array, *tmp; 224 225 list_for_each_entry_safe(array, tmp, &lseg->pls_commits, lseg_list) 226 pnfs_remove_and_free_commit_array(array); 227 } 228 EXPORT_SYMBOL_GPL(pnfs_generic_ds_cinfo_release_lseg); 229 230 void 231 pnfs_generic_ds_cinfo_destroy(struct pnfs_ds_commit_info *fl_cinfo) 232 { 233 struct pnfs_commit_array *array, *tmp; 234 235 list_for_each_entry_safe(array, tmp, &fl_cinfo->commits, cinfo_list) 236 pnfs_remove_and_free_commit_array(array); 237 } 238 EXPORT_SYMBOL_GPL(pnfs_generic_ds_cinfo_destroy); 239 240 /* 241 * Locks the nfs_page requests for commit and moves them to 242 * @bucket->committing. 243 */ 244 static int 245 pnfs_bucket_scan_ds_commit_list(struct pnfs_commit_bucket *bucket, 246 struct nfs_commit_info *cinfo, 247 int max) 248 { 249 struct list_head *src = &bucket->written; 250 struct list_head *dst = &bucket->committing; 251 int ret; 252 253 lockdep_assert_held(&NFS_I(cinfo->inode)->commit_mutex); 254 ret = nfs_scan_commit_list(src, dst, cinfo, max); 255 if (ret) { 256 cinfo->ds->nwritten -= ret; 257 cinfo->ds->ncommitting += ret; 258 } 259 return ret; 260 } 261 262 static int pnfs_bucket_scan_array(struct nfs_commit_info *cinfo, 263 struct pnfs_commit_bucket *buckets, 264 unsigned int nbuckets, 265 int max) 266 { 267 unsigned int i; 268 int rv = 0, cnt; 269 270 for (i = 0; i < nbuckets && max != 0; i++) { 271 cnt = pnfs_bucket_scan_ds_commit_list(&buckets[i], cinfo, max); 272 rv += cnt; 273 max -= cnt; 274 } 275 return rv; 276 } 277 278 /* Move reqs from written to committing lists, returning count 279 * of number moved. 280 */ 281 int pnfs_generic_scan_commit_lists(struct nfs_commit_info *cinfo, int max) 282 { 283 struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds; 284 struct pnfs_commit_array *array; 285 int rv = 0, cnt; 286 287 rcu_read_lock(); 288 list_for_each_entry_rcu(array, &fl_cinfo->commits, cinfo_list) { 289 if (!array->lseg || !pnfs_get_commit_array(array)) 290 continue; 291 rcu_read_unlock(); 292 cnt = pnfs_bucket_scan_array(cinfo, array->buckets, 293 array->nbuckets, max); 294 rcu_read_lock(); 295 pnfs_put_commit_array(array, cinfo->inode); 296 rv += cnt; 297 max -= cnt; 298 if (!max) 299 break; 300 } 301 rcu_read_unlock(); 302 return rv; 303 } 304 EXPORT_SYMBOL_GPL(pnfs_generic_scan_commit_lists); 305 306 static unsigned int 307 pnfs_bucket_recover_commit_reqs(struct list_head *dst, 308 struct pnfs_commit_bucket *buckets, 309 unsigned int nbuckets, 310 struct nfs_commit_info *cinfo) 311 { 312 struct pnfs_commit_bucket *b; 313 struct pnfs_layout_segment *freeme; 314 unsigned int nwritten, ret = 0; 315 unsigned int i; 316 317 restart: 318 for (i = 0, b = buckets; i < nbuckets; i++, b++) { 319 nwritten = nfs_scan_commit_list(&b->written, dst, cinfo, 0); 320 if (!nwritten) 321 continue; 322 ret += nwritten; 323 freeme = pnfs_free_bucket_lseg(b); 324 if (freeme) { 325 pnfs_put_lseg(freeme); 326 goto restart; 327 } 328 } 329 return ret; 330 } 331 332 /* Pull everything off the committing lists and dump into @dst. */ 333 void pnfs_generic_recover_commit_reqs(struct list_head *dst, 334 struct nfs_commit_info *cinfo) 335 { 336 struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds; 337 struct pnfs_commit_array *array; 338 unsigned int nwritten; 339 340 lockdep_assert_held(&NFS_I(cinfo->inode)->commit_mutex); 341 rcu_read_lock(); 342 list_for_each_entry_rcu(array, &fl_cinfo->commits, cinfo_list) { 343 if (!array->lseg || !pnfs_get_commit_array(array)) 344 continue; 345 rcu_read_unlock(); 346 nwritten = pnfs_bucket_recover_commit_reqs(dst, 347 array->buckets, 348 array->nbuckets, 349 cinfo); 350 rcu_read_lock(); 351 pnfs_put_commit_array(array, cinfo->inode); 352 fl_cinfo->nwritten -= nwritten; 353 } 354 rcu_read_unlock(); 355 } 356 EXPORT_SYMBOL_GPL(pnfs_generic_recover_commit_reqs); 357 358 static struct nfs_page * 359 pnfs_bucket_search_commit_reqs(struct pnfs_commit_bucket *buckets, 360 unsigned int nbuckets, struct page *page) 361 { 362 struct nfs_page *req; 363 struct pnfs_commit_bucket *b; 364 unsigned int i; 365 366 /* Linearly search the commit lists for each bucket until a matching 367 * request is found */ 368 for (i = 0, b = buckets; i < nbuckets; i++, b++) { 369 list_for_each_entry(req, &b->written, wb_list) { 370 if (req->wb_page == page) 371 return req->wb_head; 372 } 373 list_for_each_entry(req, &b->committing, wb_list) { 374 if (req->wb_page == page) 375 return req->wb_head; 376 } 377 } 378 return NULL; 379 } 380 381 /* pnfs_generic_search_commit_reqs - Search lists in @cinfo for the head reqest 382 * for @page 383 * @cinfo - commit info for current inode 384 * @page - page to search for matching head request 385 * 386 * Returns a the head request if one is found, otherwise returns NULL. 387 */ 388 struct nfs_page * 389 pnfs_generic_search_commit_reqs(struct nfs_commit_info *cinfo, struct page *page) 390 { 391 struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds; 392 struct pnfs_commit_array *array; 393 struct nfs_page *req; 394 395 list_for_each_entry(array, &fl_cinfo->commits, cinfo_list) { 396 req = pnfs_bucket_search_commit_reqs(array->buckets, 397 array->nbuckets, page); 398 if (req) 399 return req; 400 } 401 return NULL; 402 } 403 EXPORT_SYMBOL_GPL(pnfs_generic_search_commit_reqs); 404 405 static struct pnfs_layout_segment * 406 pnfs_bucket_get_committing(struct list_head *head, 407 struct pnfs_commit_bucket *bucket, 408 struct nfs_commit_info *cinfo) 409 { 410 struct list_head *pos; 411 412 list_for_each(pos, &bucket->committing) 413 cinfo->ds->ncommitting--; 414 list_splice_init(&bucket->committing, head); 415 return pnfs_free_bucket_lseg(bucket); 416 } 417 418 static struct nfs_commit_data * 419 pnfs_bucket_fetch_commitdata(struct pnfs_commit_bucket *bucket, 420 struct nfs_commit_info *cinfo) 421 { 422 struct nfs_commit_data *data = nfs_commitdata_alloc(false); 423 424 if (!data) 425 return NULL; 426 data->lseg = pnfs_bucket_get_committing(&data->pages, bucket, cinfo); 427 if (!data->lseg) 428 data->lseg = pnfs_get_lseg(bucket->lseg); 429 return data; 430 } 431 432 static void pnfs_generic_retry_commit(struct pnfs_commit_bucket *buckets, 433 unsigned int nbuckets, 434 struct nfs_commit_info *cinfo, 435 unsigned int idx) 436 { 437 struct pnfs_commit_bucket *bucket; 438 struct pnfs_layout_segment *freeme; 439 LIST_HEAD(pages); 440 441 for (bucket = buckets; idx < nbuckets; bucket++, idx++) { 442 if (list_empty(&bucket->committing)) 443 continue; 444 mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 445 freeme = pnfs_bucket_get_committing(&pages, bucket, cinfo); 446 mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 447 nfs_retry_commit(&pages, freeme, cinfo, idx); 448 pnfs_put_lseg(freeme); 449 } 450 } 451 452 static unsigned int 453 pnfs_bucket_alloc_ds_commits(struct list_head *list, 454 struct pnfs_commit_bucket *buckets, 455 unsigned int nbuckets, 456 struct nfs_commit_info *cinfo) 457 { 458 struct pnfs_commit_bucket *bucket; 459 struct nfs_commit_data *data; 460 unsigned int i; 461 unsigned int nreq = 0; 462 463 for (i = 0, bucket = buckets; i < nbuckets; i++, bucket++) { 464 if (list_empty(&bucket->committing)) 465 continue; 466 mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 467 if (!list_empty(&bucket->committing)) { 468 data = pnfs_bucket_fetch_commitdata(bucket, cinfo); 469 if (!data) 470 goto out_error; 471 data->ds_commit_index = i; 472 list_add_tail(&data->list, list); 473 atomic_inc(&cinfo->mds->rpcs_out); 474 nreq++; 475 } 476 mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 477 } 478 return nreq; 479 out_error: 480 mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 481 /* Clean up on error */ 482 pnfs_generic_retry_commit(buckets, nbuckets, cinfo, i); 483 return nreq; 484 } 485 486 static unsigned int 487 pnfs_alloc_ds_commits_list(struct list_head *list, 488 struct pnfs_ds_commit_info *fl_cinfo, 489 struct nfs_commit_info *cinfo) 490 { 491 struct pnfs_commit_array *array; 492 unsigned int ret = 0; 493 494 rcu_read_lock(); 495 list_for_each_entry_rcu(array, &fl_cinfo->commits, cinfo_list) { 496 if (!array->lseg || !pnfs_get_commit_array(array)) 497 continue; 498 rcu_read_unlock(); 499 ret += pnfs_bucket_alloc_ds_commits(list, array->buckets, 500 array->nbuckets, cinfo); 501 rcu_read_lock(); 502 pnfs_put_commit_array(array, cinfo->inode); 503 } 504 return ret; 505 } 506 507 /* This follows nfs_commit_list pretty closely */ 508 int 509 pnfs_generic_commit_pagelist(struct inode *inode, struct list_head *mds_pages, 510 int how, struct nfs_commit_info *cinfo, 511 int (*initiate_commit)(struct nfs_commit_data *data, 512 int how)) 513 { 514 struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds; 515 struct nfs_commit_data *data, *tmp; 516 LIST_HEAD(list); 517 unsigned int nreq = 0; 518 519 if (!list_empty(mds_pages)) { 520 data = nfs_commitdata_alloc(true); 521 data->ds_commit_index = -1; 522 list_splice_init(mds_pages, &data->pages); 523 list_add_tail(&data->list, &list); 524 atomic_inc(&cinfo->mds->rpcs_out); 525 nreq++; 526 } 527 528 nreq += pnfs_alloc_ds_commits_list(&list, fl_cinfo, cinfo); 529 if (nreq == 0) 530 goto out; 531 532 list_for_each_entry_safe(data, tmp, &list, list) { 533 list_del(&data->list); 534 if (data->ds_commit_index < 0) { 535 nfs_init_commit(data, NULL, NULL, cinfo); 536 nfs_initiate_commit(NFS_CLIENT(inode), data, 537 NFS_PROTO(data->inode), 538 data->mds_ops, how, 0); 539 } else { 540 nfs_init_commit(data, NULL, data->lseg, cinfo); 541 initiate_commit(data, how); 542 } 543 } 544 out: 545 return PNFS_ATTEMPTED; 546 } 547 EXPORT_SYMBOL_GPL(pnfs_generic_commit_pagelist); 548 549 /* 550 * Data server cache 551 * 552 * Data servers can be mapped to different device ids. 553 * nfs4_pnfs_ds reference counting 554 * - set to 1 on allocation 555 * - incremented when a device id maps a data server already in the cache. 556 * - decremented when deviceid is removed from the cache. 557 */ 558 static DEFINE_SPINLOCK(nfs4_ds_cache_lock); 559 static LIST_HEAD(nfs4_data_server_cache); 560 561 /* Debug routines */ 562 static void 563 print_ds(struct nfs4_pnfs_ds *ds) 564 { 565 if (ds == NULL) { 566 printk(KERN_WARNING "%s NULL device\n", __func__); 567 return; 568 } 569 printk(KERN_WARNING " ds %s\n" 570 " ref count %d\n" 571 " client %p\n" 572 " cl_exchange_flags %x\n", 573 ds->ds_remotestr, 574 refcount_read(&ds->ds_count), ds->ds_clp, 575 ds->ds_clp ? ds->ds_clp->cl_exchange_flags : 0); 576 } 577 578 static bool 579 same_sockaddr(struct sockaddr *addr1, struct sockaddr *addr2) 580 { 581 struct sockaddr_in *a, *b; 582 struct sockaddr_in6 *a6, *b6; 583 584 if (addr1->sa_family != addr2->sa_family) 585 return false; 586 587 switch (addr1->sa_family) { 588 case AF_INET: 589 a = (struct sockaddr_in *)addr1; 590 b = (struct sockaddr_in *)addr2; 591 592 if (a->sin_addr.s_addr == b->sin_addr.s_addr && 593 a->sin_port == b->sin_port) 594 return true; 595 break; 596 597 case AF_INET6: 598 a6 = (struct sockaddr_in6 *)addr1; 599 b6 = (struct sockaddr_in6 *)addr2; 600 601 /* LINKLOCAL addresses must have matching scope_id */ 602 if (ipv6_addr_src_scope(&a6->sin6_addr) == 603 IPV6_ADDR_SCOPE_LINKLOCAL && 604 a6->sin6_scope_id != b6->sin6_scope_id) 605 return false; 606 607 if (ipv6_addr_equal(&a6->sin6_addr, &b6->sin6_addr) && 608 a6->sin6_port == b6->sin6_port) 609 return true; 610 break; 611 612 default: 613 dprintk("%s: unhandled address family: %u\n", 614 __func__, addr1->sa_family); 615 return false; 616 } 617 618 return false; 619 } 620 621 /* 622 * Checks if 'dsaddrs1' contains a subset of 'dsaddrs2'. If it does, 623 * declare a match. 624 */ 625 static bool 626 _same_data_server_addrs_locked(const struct list_head *dsaddrs1, 627 const struct list_head *dsaddrs2) 628 { 629 struct nfs4_pnfs_ds_addr *da1, *da2; 630 struct sockaddr *sa1, *sa2; 631 bool match = false; 632 633 list_for_each_entry(da1, dsaddrs1, da_node) { 634 sa1 = (struct sockaddr *)&da1->da_addr; 635 match = false; 636 list_for_each_entry(da2, dsaddrs2, da_node) { 637 sa2 = (struct sockaddr *)&da2->da_addr; 638 match = same_sockaddr(sa1, sa2); 639 if (match) 640 break; 641 } 642 if (!match) 643 break; 644 } 645 return match; 646 } 647 648 /* 649 * Lookup DS by addresses. nfs4_ds_cache_lock is held 650 */ 651 static struct nfs4_pnfs_ds * 652 _data_server_lookup_locked(const struct list_head *dsaddrs) 653 { 654 struct nfs4_pnfs_ds *ds; 655 656 list_for_each_entry(ds, &nfs4_data_server_cache, ds_node) 657 if (_same_data_server_addrs_locked(&ds->ds_addrs, dsaddrs)) 658 return ds; 659 return NULL; 660 } 661 662 static void destroy_ds(struct nfs4_pnfs_ds *ds) 663 { 664 struct nfs4_pnfs_ds_addr *da; 665 666 dprintk("--> %s\n", __func__); 667 ifdebug(FACILITY) 668 print_ds(ds); 669 670 nfs_put_client(ds->ds_clp); 671 672 while (!list_empty(&ds->ds_addrs)) { 673 da = list_first_entry(&ds->ds_addrs, 674 struct nfs4_pnfs_ds_addr, 675 da_node); 676 list_del_init(&da->da_node); 677 kfree(da->da_remotestr); 678 kfree(da); 679 } 680 681 kfree(ds->ds_remotestr); 682 kfree(ds); 683 } 684 685 void nfs4_pnfs_ds_put(struct nfs4_pnfs_ds *ds) 686 { 687 if (refcount_dec_and_lock(&ds->ds_count, 688 &nfs4_ds_cache_lock)) { 689 list_del_init(&ds->ds_node); 690 spin_unlock(&nfs4_ds_cache_lock); 691 destroy_ds(ds); 692 } 693 } 694 EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_put); 695 696 /* 697 * Create a string with a human readable address and port to avoid 698 * complicated setup around many dprinks. 699 */ 700 static char * 701 nfs4_pnfs_remotestr(struct list_head *dsaddrs, gfp_t gfp_flags) 702 { 703 struct nfs4_pnfs_ds_addr *da; 704 char *remotestr; 705 size_t len; 706 char *p; 707 708 len = 3; /* '{', '}' and eol */ 709 list_for_each_entry(da, dsaddrs, da_node) { 710 len += strlen(da->da_remotestr) + 1; /* string plus comma */ 711 } 712 713 remotestr = kzalloc(len, gfp_flags); 714 if (!remotestr) 715 return NULL; 716 717 p = remotestr; 718 *(p++) = '{'; 719 len--; 720 list_for_each_entry(da, dsaddrs, da_node) { 721 size_t ll = strlen(da->da_remotestr); 722 723 if (ll > len) 724 goto out_err; 725 726 memcpy(p, da->da_remotestr, ll); 727 p += ll; 728 len -= ll; 729 730 if (len < 1) 731 goto out_err; 732 (*p++) = ','; 733 len--; 734 } 735 if (len < 2) 736 goto out_err; 737 *(p++) = '}'; 738 *p = '\0'; 739 return remotestr; 740 out_err: 741 kfree(remotestr); 742 return NULL; 743 } 744 745 /* 746 * Given a list of multipath struct nfs4_pnfs_ds_addr, add it to ds cache if 747 * uncached and return cached struct nfs4_pnfs_ds. 748 */ 749 struct nfs4_pnfs_ds * 750 nfs4_pnfs_ds_add(struct list_head *dsaddrs, gfp_t gfp_flags) 751 { 752 struct nfs4_pnfs_ds *tmp_ds, *ds = NULL; 753 char *remotestr; 754 755 if (list_empty(dsaddrs)) { 756 dprintk("%s: no addresses defined\n", __func__); 757 goto out; 758 } 759 760 ds = kzalloc(sizeof(*ds), gfp_flags); 761 if (!ds) 762 goto out; 763 764 /* this is only used for debugging, so it's ok if its NULL */ 765 remotestr = nfs4_pnfs_remotestr(dsaddrs, gfp_flags); 766 767 spin_lock(&nfs4_ds_cache_lock); 768 tmp_ds = _data_server_lookup_locked(dsaddrs); 769 if (tmp_ds == NULL) { 770 INIT_LIST_HEAD(&ds->ds_addrs); 771 list_splice_init(dsaddrs, &ds->ds_addrs); 772 ds->ds_remotestr = remotestr; 773 refcount_set(&ds->ds_count, 1); 774 INIT_LIST_HEAD(&ds->ds_node); 775 ds->ds_clp = NULL; 776 list_add(&ds->ds_node, &nfs4_data_server_cache); 777 dprintk("%s add new data server %s\n", __func__, 778 ds->ds_remotestr); 779 } else { 780 kfree(remotestr); 781 kfree(ds); 782 refcount_inc(&tmp_ds->ds_count); 783 dprintk("%s data server %s found, inc'ed ds_count to %d\n", 784 __func__, tmp_ds->ds_remotestr, 785 refcount_read(&tmp_ds->ds_count)); 786 ds = tmp_ds; 787 } 788 spin_unlock(&nfs4_ds_cache_lock); 789 out: 790 return ds; 791 } 792 EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_add); 793 794 static void nfs4_wait_ds_connect(struct nfs4_pnfs_ds *ds) 795 { 796 might_sleep(); 797 wait_on_bit(&ds->ds_state, NFS4DS_CONNECTING, 798 TASK_KILLABLE); 799 } 800 801 static void nfs4_clear_ds_conn_bit(struct nfs4_pnfs_ds *ds) 802 { 803 smp_mb__before_atomic(); 804 clear_bit(NFS4DS_CONNECTING, &ds->ds_state); 805 smp_mb__after_atomic(); 806 wake_up_bit(&ds->ds_state, NFS4DS_CONNECTING); 807 } 808 809 static struct nfs_client *(*get_v3_ds_connect)( 810 struct nfs_server *mds_srv, 811 const struct sockaddr *ds_addr, 812 int ds_addrlen, 813 int ds_proto, 814 unsigned int ds_timeo, 815 unsigned int ds_retrans); 816 817 static bool load_v3_ds_connect(void) 818 { 819 if (!get_v3_ds_connect) { 820 get_v3_ds_connect = symbol_request(nfs3_set_ds_client); 821 WARN_ON_ONCE(!get_v3_ds_connect); 822 } 823 824 return(get_v3_ds_connect != NULL); 825 } 826 827 void nfs4_pnfs_v3_ds_connect_unload(void) 828 { 829 if (get_v3_ds_connect) { 830 symbol_put(nfs3_set_ds_client); 831 get_v3_ds_connect = NULL; 832 } 833 } 834 835 static int _nfs4_pnfs_v3_ds_connect(struct nfs_server *mds_srv, 836 struct nfs4_pnfs_ds *ds, 837 unsigned int timeo, 838 unsigned int retrans) 839 { 840 struct nfs_client *clp = ERR_PTR(-EIO); 841 struct nfs4_pnfs_ds_addr *da; 842 int status = 0; 843 844 dprintk("--> %s DS %s\n", __func__, ds->ds_remotestr); 845 846 if (!load_v3_ds_connect()) 847 goto out; 848 849 list_for_each_entry(da, &ds->ds_addrs, da_node) { 850 dprintk("%s: DS %s: trying address %s\n", 851 __func__, ds->ds_remotestr, da->da_remotestr); 852 853 if (!IS_ERR(clp)) { 854 struct xprt_create xprt_args = { 855 .ident = XPRT_TRANSPORT_TCP, 856 .net = clp->cl_net, 857 .dstaddr = (struct sockaddr *)&da->da_addr, 858 .addrlen = da->da_addrlen, 859 .servername = clp->cl_hostname, 860 }; 861 /* Add this address as an alias */ 862 rpc_clnt_add_xprt(clp->cl_rpcclient, &xprt_args, 863 rpc_clnt_test_and_add_xprt, NULL); 864 continue; 865 } 866 clp = get_v3_ds_connect(mds_srv, 867 (struct sockaddr *)&da->da_addr, 868 da->da_addrlen, IPPROTO_TCP, 869 timeo, retrans); 870 if (IS_ERR(clp)) 871 continue; 872 clp->cl_rpcclient->cl_softerr = 0; 873 clp->cl_rpcclient->cl_softrtry = 0; 874 } 875 876 if (IS_ERR(clp)) { 877 status = PTR_ERR(clp); 878 goto out; 879 } 880 881 smp_wmb(); 882 ds->ds_clp = clp; 883 dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr); 884 out: 885 return status; 886 } 887 888 static int _nfs4_pnfs_v4_ds_connect(struct nfs_server *mds_srv, 889 struct nfs4_pnfs_ds *ds, 890 unsigned int timeo, 891 unsigned int retrans, 892 u32 minor_version) 893 { 894 struct nfs_client *clp = ERR_PTR(-EIO); 895 struct nfs4_pnfs_ds_addr *da; 896 int status = 0; 897 898 dprintk("--> %s DS %s\n", __func__, ds->ds_remotestr); 899 900 list_for_each_entry(da, &ds->ds_addrs, da_node) { 901 dprintk("%s: DS %s: trying address %s\n", 902 __func__, ds->ds_remotestr, da->da_remotestr); 903 904 if (!IS_ERR(clp) && clp->cl_mvops->session_trunk) { 905 struct xprt_create xprt_args = { 906 .ident = XPRT_TRANSPORT_TCP, 907 .net = clp->cl_net, 908 .dstaddr = (struct sockaddr *)&da->da_addr, 909 .addrlen = da->da_addrlen, 910 .servername = clp->cl_hostname, 911 }; 912 struct nfs4_add_xprt_data xprtdata = { 913 .clp = clp, 914 .cred = nfs4_get_clid_cred(clp), 915 }; 916 struct rpc_add_xprt_test rpcdata = { 917 .add_xprt_test = clp->cl_mvops->session_trunk, 918 .data = &xprtdata, 919 }; 920 921 /** 922 * Test this address for session trunking and 923 * add as an alias 924 */ 925 rpc_clnt_add_xprt(clp->cl_rpcclient, &xprt_args, 926 rpc_clnt_setup_test_and_add_xprt, 927 &rpcdata); 928 if (xprtdata.cred) 929 put_cred(xprtdata.cred); 930 } else { 931 clp = nfs4_set_ds_client(mds_srv, 932 (struct sockaddr *)&da->da_addr, 933 da->da_addrlen, IPPROTO_TCP, 934 timeo, retrans, minor_version); 935 if (IS_ERR(clp)) 936 continue; 937 938 status = nfs4_init_ds_session(clp, 939 mds_srv->nfs_client->cl_lease_time); 940 if (status) { 941 nfs_put_client(clp); 942 clp = ERR_PTR(-EIO); 943 continue; 944 } 945 946 } 947 } 948 949 if (IS_ERR(clp)) { 950 status = PTR_ERR(clp); 951 goto out; 952 } 953 954 smp_wmb(); 955 ds->ds_clp = clp; 956 dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr); 957 out: 958 return status; 959 } 960 961 /* 962 * Create an rpc connection to the nfs4_pnfs_ds data server. 963 * Currently only supports IPv4 and IPv6 addresses. 964 * If connection fails, make devid unavailable and return a -errno. 965 */ 966 int nfs4_pnfs_ds_connect(struct nfs_server *mds_srv, struct nfs4_pnfs_ds *ds, 967 struct nfs4_deviceid_node *devid, unsigned int timeo, 968 unsigned int retrans, u32 version, u32 minor_version) 969 { 970 int err; 971 972 again: 973 err = 0; 974 if (test_and_set_bit(NFS4DS_CONNECTING, &ds->ds_state) == 0) { 975 if (version == 3) { 976 err = _nfs4_pnfs_v3_ds_connect(mds_srv, ds, timeo, 977 retrans); 978 } else if (version == 4) { 979 err = _nfs4_pnfs_v4_ds_connect(mds_srv, ds, timeo, 980 retrans, minor_version); 981 } else { 982 dprintk("%s: unsupported DS version %d\n", __func__, 983 version); 984 err = -EPROTONOSUPPORT; 985 } 986 987 nfs4_clear_ds_conn_bit(ds); 988 } else { 989 nfs4_wait_ds_connect(ds); 990 991 /* what was waited on didn't connect AND didn't mark unavail */ 992 if (!ds->ds_clp && !nfs4_test_deviceid_unavailable(devid)) 993 goto again; 994 } 995 996 /* 997 * At this point the ds->ds_clp should be ready, but it might have 998 * hit an error. 999 */ 1000 if (!err) { 1001 if (!ds->ds_clp || !nfs_client_init_is_complete(ds->ds_clp)) { 1002 WARN_ON_ONCE(ds->ds_clp || 1003 !nfs4_test_deviceid_unavailable(devid)); 1004 return -EINVAL; 1005 } 1006 err = nfs_client_init_status(ds->ds_clp); 1007 } 1008 1009 return err; 1010 } 1011 EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_connect); 1012 1013 /* 1014 * Currently only supports ipv4, ipv6 and one multi-path address. 1015 */ 1016 struct nfs4_pnfs_ds_addr * 1017 nfs4_decode_mp_ds_addr(struct net *net, struct xdr_stream *xdr, gfp_t gfp_flags) 1018 { 1019 struct nfs4_pnfs_ds_addr *da = NULL; 1020 char *buf, *portstr; 1021 __be16 port; 1022 int nlen, rlen; 1023 int tmp[2]; 1024 __be32 *p; 1025 char *netid, *match_netid; 1026 size_t len, match_netid_len; 1027 char *startsep = ""; 1028 char *endsep = ""; 1029 1030 1031 /* r_netid */ 1032 p = xdr_inline_decode(xdr, 4); 1033 if (unlikely(!p)) 1034 goto out_err; 1035 nlen = be32_to_cpup(p++); 1036 1037 p = xdr_inline_decode(xdr, nlen); 1038 if (unlikely(!p)) 1039 goto out_err; 1040 1041 netid = kmalloc(nlen+1, gfp_flags); 1042 if (unlikely(!netid)) 1043 goto out_err; 1044 1045 netid[nlen] = '\0'; 1046 memcpy(netid, p, nlen); 1047 1048 /* r_addr: ip/ip6addr with port in dec octets - see RFC 5665 */ 1049 p = xdr_inline_decode(xdr, 4); 1050 if (unlikely(!p)) 1051 goto out_free_netid; 1052 rlen = be32_to_cpup(p); 1053 1054 p = xdr_inline_decode(xdr, rlen); 1055 if (unlikely(!p)) 1056 goto out_free_netid; 1057 1058 /* port is ".ABC.DEF", 8 chars max */ 1059 if (rlen > INET6_ADDRSTRLEN + IPV6_SCOPE_ID_LEN + 8) { 1060 dprintk("%s: Invalid address, length %d\n", __func__, 1061 rlen); 1062 goto out_free_netid; 1063 } 1064 buf = kmalloc(rlen + 1, gfp_flags); 1065 if (!buf) { 1066 dprintk("%s: Not enough memory\n", __func__); 1067 goto out_free_netid; 1068 } 1069 buf[rlen] = '\0'; 1070 memcpy(buf, p, rlen); 1071 1072 /* replace port '.' with '-' */ 1073 portstr = strrchr(buf, '.'); 1074 if (!portstr) { 1075 dprintk("%s: Failed finding expected dot in port\n", 1076 __func__); 1077 goto out_free_buf; 1078 } 1079 *portstr = '-'; 1080 1081 /* find '.' between address and port */ 1082 portstr = strrchr(buf, '.'); 1083 if (!portstr) { 1084 dprintk("%s: Failed finding expected dot between address and " 1085 "port\n", __func__); 1086 goto out_free_buf; 1087 } 1088 *portstr = '\0'; 1089 1090 da = kzalloc(sizeof(*da), gfp_flags); 1091 if (unlikely(!da)) 1092 goto out_free_buf; 1093 1094 INIT_LIST_HEAD(&da->da_node); 1095 1096 if (!rpc_pton(net, buf, portstr-buf, (struct sockaddr *)&da->da_addr, 1097 sizeof(da->da_addr))) { 1098 dprintk("%s: error parsing address %s\n", __func__, buf); 1099 goto out_free_da; 1100 } 1101 1102 portstr++; 1103 sscanf(portstr, "%d-%d", &tmp[0], &tmp[1]); 1104 port = htons((tmp[0] << 8) | (tmp[1])); 1105 1106 switch (da->da_addr.ss_family) { 1107 case AF_INET: 1108 ((struct sockaddr_in *)&da->da_addr)->sin_port = port; 1109 da->da_addrlen = sizeof(struct sockaddr_in); 1110 match_netid = "tcp"; 1111 match_netid_len = 3; 1112 break; 1113 1114 case AF_INET6: 1115 ((struct sockaddr_in6 *)&da->da_addr)->sin6_port = port; 1116 da->da_addrlen = sizeof(struct sockaddr_in6); 1117 match_netid = "tcp6"; 1118 match_netid_len = 4; 1119 startsep = "["; 1120 endsep = "]"; 1121 break; 1122 1123 default: 1124 dprintk("%s: unsupported address family: %u\n", 1125 __func__, da->da_addr.ss_family); 1126 goto out_free_da; 1127 } 1128 1129 if (nlen != match_netid_len || strncmp(netid, match_netid, nlen)) { 1130 dprintk("%s: ERROR: r_netid \"%s\" != \"%s\"\n", 1131 __func__, netid, match_netid); 1132 goto out_free_da; 1133 } 1134 1135 /* save human readable address */ 1136 len = strlen(startsep) + strlen(buf) + strlen(endsep) + 7; 1137 da->da_remotestr = kzalloc(len, gfp_flags); 1138 1139 /* NULL is ok, only used for dprintk */ 1140 if (da->da_remotestr) 1141 snprintf(da->da_remotestr, len, "%s%s%s:%u", startsep, 1142 buf, endsep, ntohs(port)); 1143 1144 dprintk("%s: Parsed DS addr %s\n", __func__, da->da_remotestr); 1145 kfree(buf); 1146 kfree(netid); 1147 return da; 1148 1149 out_free_da: 1150 kfree(da); 1151 out_free_buf: 1152 dprintk("%s: Error parsing DS addr: %s\n", __func__, buf); 1153 kfree(buf); 1154 out_free_netid: 1155 kfree(netid); 1156 out_err: 1157 return NULL; 1158 } 1159 EXPORT_SYMBOL_GPL(nfs4_decode_mp_ds_addr); 1160 1161 void 1162 pnfs_layout_mark_request_commit(struct nfs_page *req, 1163 struct pnfs_layout_segment *lseg, 1164 struct nfs_commit_info *cinfo, 1165 u32 ds_commit_idx) 1166 { 1167 struct list_head *list; 1168 struct pnfs_commit_array *array; 1169 struct pnfs_commit_bucket *bucket; 1170 1171 mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 1172 array = pnfs_lookup_commit_array(cinfo->ds, lseg); 1173 if (!array || !pnfs_is_valid_lseg(lseg)) 1174 goto out_resched; 1175 bucket = &array->buckets[ds_commit_idx]; 1176 list = &bucket->written; 1177 /* Non-empty buckets hold a reference on the lseg. That ref 1178 * is normally transferred to the COMMIT call and released 1179 * there. It could also be released if the last req is pulled 1180 * off due to a rewrite, in which case it will be done in 1181 * pnfs_common_clear_request_commit 1182 */ 1183 if (!bucket->lseg) 1184 bucket->lseg = pnfs_get_lseg(lseg); 1185 set_bit(PG_COMMIT_TO_DS, &req->wb_flags); 1186 cinfo->ds->nwritten++; 1187 1188 nfs_request_add_commit_list_locked(req, list, cinfo); 1189 mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 1190 nfs_mark_page_unstable(req->wb_page, cinfo); 1191 return; 1192 out_resched: 1193 mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 1194 cinfo->completion_ops->resched_write(cinfo, req); 1195 } 1196 EXPORT_SYMBOL_GPL(pnfs_layout_mark_request_commit); 1197 1198 int 1199 pnfs_nfs_generic_sync(struct inode *inode, bool datasync) 1200 { 1201 int ret; 1202 1203 if (!pnfs_layoutcommit_outstanding(inode)) 1204 return 0; 1205 ret = nfs_commit_inode(inode, FLUSH_SYNC); 1206 if (ret < 0) 1207 return ret; 1208 if (datasync) 1209 return 0; 1210 return pnfs_layoutcommit_inode(inode, true); 1211 } 1212 EXPORT_SYMBOL_GPL(pnfs_nfs_generic_sync); 1213 1214