1 /* 2 * Module for pnfs flexfile 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/nfs_page.h> 11 #include <linux/module.h> 12 13 #include <linux/sunrpc/metrics.h> 14 15 #include "flexfilelayout.h" 16 #include "../nfs4session.h" 17 #include "../nfs4idmap.h" 18 #include "../internal.h" 19 #include "../delegation.h" 20 #include "../nfs4trace.h" 21 #include "../iostat.h" 22 #include "../nfs.h" 23 #include "../nfs42.h" 24 25 #define NFSDBG_FACILITY NFSDBG_PNFS_LD 26 27 #define FF_LAYOUT_POLL_RETRY_MAX (15*HZ) 28 #define FF_LAYOUTRETURN_MAXERR 20 29 30 31 static struct group_info *ff_zero_group; 32 33 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task, 34 struct nfs_pgio_header *hdr); 35 static int ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo, 36 struct nfs42_layoutstat_devinfo *devinfo, 37 int dev_limit); 38 static void ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr, 39 const struct nfs42_layoutstat_devinfo *devinfo, 40 struct nfs4_ff_layout_mirror *mirror); 41 42 static struct pnfs_layout_hdr * 43 ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags) 44 { 45 struct nfs4_flexfile_layout *ffl; 46 47 ffl = kzalloc(sizeof(*ffl), gfp_flags); 48 if (ffl) { 49 INIT_LIST_HEAD(&ffl->error_list); 50 INIT_LIST_HEAD(&ffl->mirrors); 51 ffl->last_report_time = ktime_get(); 52 return &ffl->generic_hdr; 53 } else 54 return NULL; 55 } 56 57 static void 58 ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo) 59 { 60 struct nfs4_ff_layout_ds_err *err, *n; 61 62 list_for_each_entry_safe(err, n, &FF_LAYOUT_FROM_HDR(lo)->error_list, 63 list) { 64 list_del(&err->list); 65 kfree(err); 66 } 67 kfree(FF_LAYOUT_FROM_HDR(lo)); 68 } 69 70 static int decode_pnfs_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid) 71 { 72 __be32 *p; 73 74 p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE); 75 if (unlikely(p == NULL)) 76 return -ENOBUFS; 77 stateid->type = NFS4_PNFS_DS_STATEID_TYPE; 78 memcpy(stateid->data, p, NFS4_STATEID_SIZE); 79 dprintk("%s: stateid id= [%x%x%x%x]\n", __func__, 80 p[0], p[1], p[2], p[3]); 81 return 0; 82 } 83 84 static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid) 85 { 86 __be32 *p; 87 88 p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE); 89 if (unlikely(!p)) 90 return -ENOBUFS; 91 memcpy(devid, p, NFS4_DEVICEID4_SIZE); 92 nfs4_print_deviceid(devid); 93 return 0; 94 } 95 96 static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh) 97 { 98 __be32 *p; 99 100 p = xdr_inline_decode(xdr, 4); 101 if (unlikely(!p)) 102 return -ENOBUFS; 103 fh->size = be32_to_cpup(p++); 104 if (fh->size > sizeof(struct nfs_fh)) { 105 printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n", 106 fh->size); 107 return -EOVERFLOW; 108 } 109 /* fh.data */ 110 p = xdr_inline_decode(xdr, fh->size); 111 if (unlikely(!p)) 112 return -ENOBUFS; 113 memcpy(&fh->data, p, fh->size); 114 dprintk("%s: fh len %d\n", __func__, fh->size); 115 116 return 0; 117 } 118 119 /* 120 * Currently only stringified uids and gids are accepted. 121 * I.e., kerberos is not supported to the DSes, so no pricipals. 122 * 123 * That means that one common function will suffice, but when 124 * principals are added, this should be split to accomodate 125 * calls to both nfs_map_name_to_uid() and nfs_map_group_to_gid(). 126 */ 127 static int 128 decode_name(struct xdr_stream *xdr, u32 *id) 129 { 130 __be32 *p; 131 int len; 132 133 /* opaque_length(4)*/ 134 p = xdr_inline_decode(xdr, 4); 135 if (unlikely(!p)) 136 return -ENOBUFS; 137 len = be32_to_cpup(p++); 138 if (len < 0) 139 return -EINVAL; 140 141 dprintk("%s: len %u\n", __func__, len); 142 143 /* opaque body */ 144 p = xdr_inline_decode(xdr, len); 145 if (unlikely(!p)) 146 return -ENOBUFS; 147 148 if (!nfs_map_string_to_numeric((char *)p, len, id)) 149 return -EINVAL; 150 151 return 0; 152 } 153 154 static bool ff_mirror_match_fh(const struct nfs4_ff_layout_mirror *m1, 155 const struct nfs4_ff_layout_mirror *m2) 156 { 157 int i, j; 158 159 if (m1->fh_versions_cnt != m2->fh_versions_cnt) 160 return false; 161 for (i = 0; i < m1->fh_versions_cnt; i++) { 162 bool found_fh = false; 163 for (j = 0; j < m2->fh_versions_cnt; j++) { 164 if (nfs_compare_fh(&m1->fh_versions[i], 165 &m2->fh_versions[j]) == 0) { 166 found_fh = true; 167 break; 168 } 169 } 170 if (!found_fh) 171 return false; 172 } 173 return true; 174 } 175 176 static struct nfs4_ff_layout_mirror * 177 ff_layout_add_mirror(struct pnfs_layout_hdr *lo, 178 struct nfs4_ff_layout_mirror *mirror) 179 { 180 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo); 181 struct nfs4_ff_layout_mirror *pos; 182 struct inode *inode = lo->plh_inode; 183 184 spin_lock(&inode->i_lock); 185 list_for_each_entry(pos, &ff_layout->mirrors, mirrors) { 186 if (memcmp(&mirror->devid, &pos->devid, sizeof(pos->devid)) != 0) 187 continue; 188 if (!ff_mirror_match_fh(mirror, pos)) 189 continue; 190 if (refcount_inc_not_zero(&pos->ref)) { 191 spin_unlock(&inode->i_lock); 192 return pos; 193 } 194 } 195 list_add(&mirror->mirrors, &ff_layout->mirrors); 196 mirror->layout = lo; 197 spin_unlock(&inode->i_lock); 198 return mirror; 199 } 200 201 static void 202 ff_layout_remove_mirror(struct nfs4_ff_layout_mirror *mirror) 203 { 204 struct inode *inode; 205 if (mirror->layout == NULL) 206 return; 207 inode = mirror->layout->plh_inode; 208 spin_lock(&inode->i_lock); 209 list_del(&mirror->mirrors); 210 spin_unlock(&inode->i_lock); 211 mirror->layout = NULL; 212 } 213 214 static struct nfs4_ff_layout_mirror *ff_layout_alloc_mirror(gfp_t gfp_flags) 215 { 216 struct nfs4_ff_layout_mirror *mirror; 217 218 mirror = kzalloc(sizeof(*mirror), gfp_flags); 219 if (mirror != NULL) { 220 spin_lock_init(&mirror->lock); 221 refcount_set(&mirror->ref, 1); 222 INIT_LIST_HEAD(&mirror->mirrors); 223 } 224 return mirror; 225 } 226 227 static void ff_layout_free_mirror(struct nfs4_ff_layout_mirror *mirror) 228 { 229 struct rpc_cred *cred; 230 231 ff_layout_remove_mirror(mirror); 232 kfree(mirror->fh_versions); 233 cred = rcu_access_pointer(mirror->ro_cred); 234 if (cred) 235 put_rpccred(cred); 236 cred = rcu_access_pointer(mirror->rw_cred); 237 if (cred) 238 put_rpccred(cred); 239 nfs4_ff_layout_put_deviceid(mirror->mirror_ds); 240 kfree(mirror); 241 } 242 243 static void ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror) 244 { 245 if (mirror != NULL && refcount_dec_and_test(&mirror->ref)) 246 ff_layout_free_mirror(mirror); 247 } 248 249 static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls) 250 { 251 int i; 252 253 if (fls->mirror_array) { 254 for (i = 0; i < fls->mirror_array_cnt; i++) { 255 /* normally mirror_ds is freed in 256 * .free_deviceid_node but we still do it here 257 * for .alloc_lseg error path */ 258 ff_layout_put_mirror(fls->mirror_array[i]); 259 } 260 kfree(fls->mirror_array); 261 fls->mirror_array = NULL; 262 } 263 } 264 265 static int ff_layout_check_layout(struct nfs4_layoutget_res *lgr) 266 { 267 int ret = 0; 268 269 dprintk("--> %s\n", __func__); 270 271 /* FIXME: remove this check when layout segment support is added */ 272 if (lgr->range.offset != 0 || 273 lgr->range.length != NFS4_MAX_UINT64) { 274 dprintk("%s Only whole file layouts supported. Use MDS i/o\n", 275 __func__); 276 ret = -EINVAL; 277 } 278 279 dprintk("--> %s returns %d\n", __func__, ret); 280 return ret; 281 } 282 283 static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls) 284 { 285 if (fls) { 286 ff_layout_free_mirror_array(fls); 287 kfree(fls); 288 } 289 } 290 291 static bool 292 ff_lseg_range_is_after(const struct pnfs_layout_range *l1, 293 const struct pnfs_layout_range *l2) 294 { 295 u64 end1, end2; 296 297 if (l1->iomode != l2->iomode) 298 return l1->iomode != IOMODE_READ; 299 end1 = pnfs_calc_offset_end(l1->offset, l1->length); 300 end2 = pnfs_calc_offset_end(l2->offset, l2->length); 301 if (end1 < l2->offset) 302 return false; 303 if (end2 < l1->offset) 304 return true; 305 return l2->offset <= l1->offset; 306 } 307 308 static bool 309 ff_lseg_merge(struct pnfs_layout_segment *new, 310 struct pnfs_layout_segment *old) 311 { 312 u64 new_end, old_end; 313 314 if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags)) 315 return false; 316 if (new->pls_range.iomode != old->pls_range.iomode) 317 return false; 318 old_end = pnfs_calc_offset_end(old->pls_range.offset, 319 old->pls_range.length); 320 if (old_end < new->pls_range.offset) 321 return false; 322 new_end = pnfs_calc_offset_end(new->pls_range.offset, 323 new->pls_range.length); 324 if (new_end < old->pls_range.offset) 325 return false; 326 327 /* Mergeable: copy info from 'old' to 'new' */ 328 if (new_end < old_end) 329 new_end = old_end; 330 if (new->pls_range.offset < old->pls_range.offset) 331 new->pls_range.offset = old->pls_range.offset; 332 new->pls_range.length = pnfs_calc_offset_length(new->pls_range.offset, 333 new_end); 334 if (test_bit(NFS_LSEG_ROC, &old->pls_flags)) 335 set_bit(NFS_LSEG_ROC, &new->pls_flags); 336 return true; 337 } 338 339 static void 340 ff_layout_add_lseg(struct pnfs_layout_hdr *lo, 341 struct pnfs_layout_segment *lseg, 342 struct list_head *free_me) 343 { 344 pnfs_generic_layout_insert_lseg(lo, lseg, 345 ff_lseg_range_is_after, 346 ff_lseg_merge, 347 free_me); 348 } 349 350 static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls) 351 { 352 int i, j; 353 354 for (i = 0; i < fls->mirror_array_cnt - 1; i++) { 355 for (j = i + 1; j < fls->mirror_array_cnt; j++) 356 if (fls->mirror_array[i]->efficiency < 357 fls->mirror_array[j]->efficiency) 358 swap(fls->mirror_array[i], 359 fls->mirror_array[j]); 360 } 361 } 362 363 static struct pnfs_layout_segment * 364 ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh, 365 struct nfs4_layoutget_res *lgr, 366 gfp_t gfp_flags) 367 { 368 struct pnfs_layout_segment *ret; 369 struct nfs4_ff_layout_segment *fls = NULL; 370 struct xdr_stream stream; 371 struct xdr_buf buf; 372 struct page *scratch; 373 u64 stripe_unit; 374 u32 mirror_array_cnt; 375 __be32 *p; 376 int i, rc; 377 378 dprintk("--> %s\n", __func__); 379 scratch = alloc_page(gfp_flags); 380 if (!scratch) 381 return ERR_PTR(-ENOMEM); 382 383 xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages, 384 lgr->layoutp->len); 385 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE); 386 387 /* stripe unit and mirror_array_cnt */ 388 rc = -EIO; 389 p = xdr_inline_decode(&stream, 8 + 4); 390 if (!p) 391 goto out_err_free; 392 393 p = xdr_decode_hyper(p, &stripe_unit); 394 mirror_array_cnt = be32_to_cpup(p++); 395 dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__, 396 stripe_unit, mirror_array_cnt); 397 398 if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT || 399 mirror_array_cnt == 0) 400 goto out_err_free; 401 402 rc = -ENOMEM; 403 fls = kzalloc(sizeof(*fls), gfp_flags); 404 if (!fls) 405 goto out_err_free; 406 407 fls->mirror_array_cnt = mirror_array_cnt; 408 fls->stripe_unit = stripe_unit; 409 fls->mirror_array = kcalloc(fls->mirror_array_cnt, 410 sizeof(fls->mirror_array[0]), gfp_flags); 411 if (fls->mirror_array == NULL) 412 goto out_err_free; 413 414 for (i = 0; i < fls->mirror_array_cnt; i++) { 415 struct nfs4_ff_layout_mirror *mirror; 416 struct auth_cred acred = { .group_info = ff_zero_group }; 417 struct rpc_cred __rcu *cred; 418 u32 ds_count, fh_count, id; 419 int j; 420 421 rc = -EIO; 422 p = xdr_inline_decode(&stream, 4); 423 if (!p) 424 goto out_err_free; 425 ds_count = be32_to_cpup(p); 426 427 /* FIXME: allow for striping? */ 428 if (ds_count != 1) 429 goto out_err_free; 430 431 fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags); 432 if (fls->mirror_array[i] == NULL) { 433 rc = -ENOMEM; 434 goto out_err_free; 435 } 436 437 fls->mirror_array[i]->ds_count = ds_count; 438 439 /* deviceid */ 440 rc = decode_deviceid(&stream, &fls->mirror_array[i]->devid); 441 if (rc) 442 goto out_err_free; 443 444 /* efficiency */ 445 rc = -EIO; 446 p = xdr_inline_decode(&stream, 4); 447 if (!p) 448 goto out_err_free; 449 fls->mirror_array[i]->efficiency = be32_to_cpup(p); 450 451 /* stateid */ 452 rc = decode_pnfs_stateid(&stream, &fls->mirror_array[i]->stateid); 453 if (rc) 454 goto out_err_free; 455 456 /* fh */ 457 rc = -EIO; 458 p = xdr_inline_decode(&stream, 4); 459 if (!p) 460 goto out_err_free; 461 fh_count = be32_to_cpup(p); 462 463 fls->mirror_array[i]->fh_versions = 464 kcalloc(fh_count, sizeof(struct nfs_fh), 465 gfp_flags); 466 if (fls->mirror_array[i]->fh_versions == NULL) { 467 rc = -ENOMEM; 468 goto out_err_free; 469 } 470 471 for (j = 0; j < fh_count; j++) { 472 rc = decode_nfs_fh(&stream, 473 &fls->mirror_array[i]->fh_versions[j]); 474 if (rc) 475 goto out_err_free; 476 } 477 478 fls->mirror_array[i]->fh_versions_cnt = fh_count; 479 480 /* user */ 481 rc = decode_name(&stream, &id); 482 if (rc) 483 goto out_err_free; 484 485 acred.uid = make_kuid(&init_user_ns, id); 486 487 /* group */ 488 rc = decode_name(&stream, &id); 489 if (rc) 490 goto out_err_free; 491 492 acred.gid = make_kgid(&init_user_ns, id); 493 494 /* find the cred for it */ 495 rcu_assign_pointer(cred, rpc_lookup_generic_cred(&acred, 0, gfp_flags)); 496 if (IS_ERR(cred)) { 497 rc = PTR_ERR(cred); 498 goto out_err_free; 499 } 500 501 if (lgr->range.iomode == IOMODE_READ) 502 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred); 503 else 504 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred); 505 506 mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]); 507 if (mirror != fls->mirror_array[i]) { 508 /* swap cred ptrs so free_mirror will clean up old */ 509 if (lgr->range.iomode == IOMODE_READ) { 510 cred = xchg(&mirror->ro_cred, cred); 511 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred); 512 } else { 513 cred = xchg(&mirror->rw_cred, cred); 514 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred); 515 } 516 ff_layout_free_mirror(fls->mirror_array[i]); 517 fls->mirror_array[i] = mirror; 518 } 519 520 dprintk("%s: iomode %s uid %u gid %u\n", __func__, 521 lgr->range.iomode == IOMODE_READ ? "READ" : "RW", 522 from_kuid(&init_user_ns, acred.uid), 523 from_kgid(&init_user_ns, acred.gid)); 524 } 525 526 p = xdr_inline_decode(&stream, 4); 527 if (!p) 528 goto out_sort_mirrors; 529 fls->flags = be32_to_cpup(p); 530 531 p = xdr_inline_decode(&stream, 4); 532 if (!p) 533 goto out_sort_mirrors; 534 for (i=0; i < fls->mirror_array_cnt; i++) 535 fls->mirror_array[i]->report_interval = be32_to_cpup(p); 536 537 out_sort_mirrors: 538 ff_layout_sort_mirrors(fls); 539 rc = ff_layout_check_layout(lgr); 540 if (rc) 541 goto out_err_free; 542 ret = &fls->generic_hdr; 543 dprintk("<-- %s (success)\n", __func__); 544 out_free_page: 545 __free_page(scratch); 546 return ret; 547 out_err_free: 548 _ff_layout_free_lseg(fls); 549 ret = ERR_PTR(rc); 550 dprintk("<-- %s (%d)\n", __func__, rc); 551 goto out_free_page; 552 } 553 554 static bool ff_layout_has_rw_segments(struct pnfs_layout_hdr *layout) 555 { 556 struct pnfs_layout_segment *lseg; 557 558 list_for_each_entry(lseg, &layout->plh_segs, pls_list) 559 if (lseg->pls_range.iomode == IOMODE_RW) 560 return true; 561 562 return false; 563 } 564 565 static void 566 ff_layout_free_lseg(struct pnfs_layout_segment *lseg) 567 { 568 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg); 569 570 dprintk("--> %s\n", __func__); 571 572 if (lseg->pls_range.iomode == IOMODE_RW) { 573 struct nfs4_flexfile_layout *ffl; 574 struct inode *inode; 575 576 ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout); 577 inode = ffl->generic_hdr.plh_inode; 578 spin_lock(&inode->i_lock); 579 if (!ff_layout_has_rw_segments(lseg->pls_layout)) { 580 ffl->commit_info.nbuckets = 0; 581 kfree(ffl->commit_info.buckets); 582 ffl->commit_info.buckets = NULL; 583 } 584 spin_unlock(&inode->i_lock); 585 } 586 _ff_layout_free_lseg(fls); 587 } 588 589 /* Return 1 until we have multiple lsegs support */ 590 static int 591 ff_layout_get_lseg_count(struct nfs4_ff_layout_segment *fls) 592 { 593 return 1; 594 } 595 596 static void 597 nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now) 598 { 599 /* first IO request? */ 600 if (atomic_inc_return(&timer->n_ops) == 1) { 601 timer->start_time = now; 602 } 603 } 604 605 static ktime_t 606 nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now) 607 { 608 ktime_t start; 609 610 if (atomic_dec_return(&timer->n_ops) < 0) 611 WARN_ON_ONCE(1); 612 613 start = timer->start_time; 614 timer->start_time = now; 615 return ktime_sub(now, start); 616 } 617 618 static bool 619 nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror, 620 struct nfs4_ff_layoutstat *layoutstat, 621 ktime_t now) 622 { 623 s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL; 624 struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(mirror->layout); 625 626 nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now); 627 if (!mirror->start_time) 628 mirror->start_time = now; 629 if (mirror->report_interval != 0) 630 report_interval = (s64)mirror->report_interval * 1000LL; 631 else if (layoutstats_timer != 0) 632 report_interval = (s64)layoutstats_timer * 1000LL; 633 if (ktime_to_ms(ktime_sub(now, ffl->last_report_time)) >= 634 report_interval) { 635 ffl->last_report_time = now; 636 return true; 637 } 638 639 return false; 640 } 641 642 static void 643 nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat, 644 __u64 requested) 645 { 646 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat; 647 648 iostat->ops_requested++; 649 iostat->bytes_requested += requested; 650 } 651 652 static void 653 nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat, 654 __u64 requested, 655 __u64 completed, 656 ktime_t time_completed, 657 ktime_t time_started) 658 { 659 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat; 660 ktime_t completion_time = ktime_sub(time_completed, time_started); 661 ktime_t timer; 662 663 iostat->ops_completed++; 664 iostat->bytes_completed += completed; 665 iostat->bytes_not_delivered += requested - completed; 666 667 timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed); 668 iostat->total_busy_time = 669 ktime_add(iostat->total_busy_time, timer); 670 iostat->aggregate_completion_time = 671 ktime_add(iostat->aggregate_completion_time, 672 completion_time); 673 } 674 675 static void 676 nfs4_ff_layout_stat_io_start_read(struct inode *inode, 677 struct nfs4_ff_layout_mirror *mirror, 678 __u64 requested, ktime_t now) 679 { 680 bool report; 681 682 spin_lock(&mirror->lock); 683 report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now); 684 nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested); 685 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags); 686 spin_unlock(&mirror->lock); 687 688 if (report) 689 pnfs_report_layoutstat(inode, GFP_KERNEL); 690 } 691 692 static void 693 nfs4_ff_layout_stat_io_end_read(struct rpc_task *task, 694 struct nfs4_ff_layout_mirror *mirror, 695 __u64 requested, 696 __u64 completed) 697 { 698 spin_lock(&mirror->lock); 699 nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat, 700 requested, completed, 701 ktime_get(), task->tk_start); 702 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags); 703 spin_unlock(&mirror->lock); 704 } 705 706 static void 707 nfs4_ff_layout_stat_io_start_write(struct inode *inode, 708 struct nfs4_ff_layout_mirror *mirror, 709 __u64 requested, ktime_t now) 710 { 711 bool report; 712 713 spin_lock(&mirror->lock); 714 report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now); 715 nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested); 716 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags); 717 spin_unlock(&mirror->lock); 718 719 if (report) 720 pnfs_report_layoutstat(inode, GFP_NOIO); 721 } 722 723 static void 724 nfs4_ff_layout_stat_io_end_write(struct rpc_task *task, 725 struct nfs4_ff_layout_mirror *mirror, 726 __u64 requested, 727 __u64 completed, 728 enum nfs3_stable_how committed) 729 { 730 if (committed == NFS_UNSTABLE) 731 requested = completed = 0; 732 733 spin_lock(&mirror->lock); 734 nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat, 735 requested, completed, ktime_get(), task->tk_start); 736 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags); 737 spin_unlock(&mirror->lock); 738 } 739 740 static int 741 ff_layout_alloc_commit_info(struct pnfs_layout_segment *lseg, 742 struct nfs_commit_info *cinfo, 743 gfp_t gfp_flags) 744 { 745 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg); 746 struct pnfs_commit_bucket *buckets; 747 int size; 748 749 if (cinfo->ds->nbuckets != 0) { 750 /* This assumes there is only one RW lseg per file. 751 * To support multiple lseg per file, we need to 752 * change struct pnfs_commit_bucket to allow dynamic 753 * increasing nbuckets. 754 */ 755 return 0; 756 } 757 758 size = ff_layout_get_lseg_count(fls) * FF_LAYOUT_MIRROR_COUNT(lseg); 759 760 buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket), 761 gfp_flags); 762 if (!buckets) 763 return -ENOMEM; 764 else { 765 int i; 766 767 spin_lock(&cinfo->inode->i_lock); 768 if (cinfo->ds->nbuckets != 0) 769 kfree(buckets); 770 else { 771 cinfo->ds->buckets = buckets; 772 cinfo->ds->nbuckets = size; 773 for (i = 0; i < size; i++) { 774 INIT_LIST_HEAD(&buckets[i].written); 775 INIT_LIST_HEAD(&buckets[i].committing); 776 /* mark direct verifier as unset */ 777 buckets[i].direct_verf.committed = 778 NFS_INVALID_STABLE_HOW; 779 } 780 } 781 spin_unlock(&cinfo->inode->i_lock); 782 return 0; 783 } 784 } 785 786 static struct nfs4_pnfs_ds * 787 ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg, 788 int start_idx, 789 int *best_idx) 790 { 791 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg); 792 struct nfs4_pnfs_ds *ds; 793 bool fail_return = false; 794 int idx; 795 796 /* mirrors are sorted by efficiency */ 797 for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) { 798 if (idx+1 == fls->mirror_array_cnt) 799 fail_return = true; 800 ds = nfs4_ff_layout_prepare_ds(lseg, idx, fail_return); 801 if (ds) { 802 *best_idx = idx; 803 return ds; 804 } 805 } 806 807 return NULL; 808 } 809 810 static void 811 ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio, 812 struct nfs_page *req, 813 bool strict_iomode) 814 { 815 pnfs_put_lseg(pgio->pg_lseg); 816 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, 817 req->wb_context, 818 0, 819 NFS4_MAX_UINT64, 820 IOMODE_READ, 821 strict_iomode, 822 GFP_KERNEL); 823 if (IS_ERR(pgio->pg_lseg)) { 824 pgio->pg_error = PTR_ERR(pgio->pg_lseg); 825 pgio->pg_lseg = NULL; 826 } 827 } 828 829 static void 830 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio, 831 struct nfs_page *req) 832 { 833 struct nfs_pgio_mirror *pgm; 834 struct nfs4_ff_layout_mirror *mirror; 835 struct nfs4_pnfs_ds *ds; 836 int ds_idx; 837 838 retry: 839 pnfs_generic_pg_check_layout(pgio); 840 /* Use full layout for now */ 841 if (!pgio->pg_lseg) { 842 ff_layout_pg_get_read(pgio, req, false); 843 if (!pgio->pg_lseg) 844 goto out_nolseg; 845 } 846 if (ff_layout_avoid_read_on_rw(pgio->pg_lseg)) { 847 ff_layout_pg_get_read(pgio, req, true); 848 if (!pgio->pg_lseg) 849 goto out_nolseg; 850 } 851 852 ds = ff_layout_choose_best_ds_for_read(pgio->pg_lseg, 0, &ds_idx); 853 if (!ds) { 854 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg)) 855 goto out_mds; 856 pnfs_put_lseg(pgio->pg_lseg); 857 pgio->pg_lseg = NULL; 858 /* Sleep for 1 second before retrying */ 859 ssleep(1); 860 goto retry; 861 } 862 863 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx); 864 865 pgio->pg_mirror_idx = ds_idx; 866 867 /* read always uses only one mirror - idx 0 for pgio layer */ 868 pgm = &pgio->pg_mirrors[0]; 869 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize; 870 871 return; 872 out_nolseg: 873 if (pgio->pg_error < 0) 874 return; 875 out_mds: 876 pnfs_put_lseg(pgio->pg_lseg); 877 pgio->pg_lseg = NULL; 878 nfs_pageio_reset_read_mds(pgio); 879 } 880 881 static void 882 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio, 883 struct nfs_page *req) 884 { 885 struct nfs4_ff_layout_mirror *mirror; 886 struct nfs_pgio_mirror *pgm; 887 struct nfs_commit_info cinfo; 888 struct nfs4_pnfs_ds *ds; 889 int i; 890 int status; 891 892 retry: 893 pnfs_generic_pg_check_layout(pgio); 894 if (!pgio->pg_lseg) { 895 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, 896 req->wb_context, 897 0, 898 NFS4_MAX_UINT64, 899 IOMODE_RW, 900 false, 901 GFP_NOFS); 902 if (IS_ERR(pgio->pg_lseg)) { 903 pgio->pg_error = PTR_ERR(pgio->pg_lseg); 904 pgio->pg_lseg = NULL; 905 return; 906 } 907 } 908 /* If no lseg, fall back to write through mds */ 909 if (pgio->pg_lseg == NULL) 910 goto out_mds; 911 912 nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq); 913 status = ff_layout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS); 914 if (status < 0) 915 goto out_mds; 916 917 /* Use a direct mapping of ds_idx to pgio mirror_idx */ 918 if (WARN_ON_ONCE(pgio->pg_mirror_count != 919 FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg))) 920 goto out_mds; 921 922 for (i = 0; i < pgio->pg_mirror_count; i++) { 923 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, i, true); 924 if (!ds) { 925 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg)) 926 goto out_mds; 927 pnfs_put_lseg(pgio->pg_lseg); 928 pgio->pg_lseg = NULL; 929 /* Sleep for 1 second before retrying */ 930 ssleep(1); 931 goto retry; 932 } 933 pgm = &pgio->pg_mirrors[i]; 934 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i); 935 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize; 936 } 937 938 return; 939 940 out_mds: 941 pnfs_put_lseg(pgio->pg_lseg); 942 pgio->pg_lseg = NULL; 943 nfs_pageio_reset_write_mds(pgio); 944 } 945 946 static unsigned int 947 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio, 948 struct nfs_page *req) 949 { 950 if (!pgio->pg_lseg) { 951 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, 952 req->wb_context, 953 0, 954 NFS4_MAX_UINT64, 955 IOMODE_RW, 956 false, 957 GFP_NOFS); 958 if (IS_ERR(pgio->pg_lseg)) { 959 pgio->pg_error = PTR_ERR(pgio->pg_lseg); 960 pgio->pg_lseg = NULL; 961 goto out; 962 } 963 } 964 if (pgio->pg_lseg) 965 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg); 966 967 /* no lseg means that pnfs is not in use, so no mirroring here */ 968 nfs_pageio_reset_write_mds(pgio); 969 out: 970 return 1; 971 } 972 973 static const struct nfs_pageio_ops ff_layout_pg_read_ops = { 974 .pg_init = ff_layout_pg_init_read, 975 .pg_test = pnfs_generic_pg_test, 976 .pg_doio = pnfs_generic_pg_readpages, 977 .pg_cleanup = pnfs_generic_pg_cleanup, 978 }; 979 980 static const struct nfs_pageio_ops ff_layout_pg_write_ops = { 981 .pg_init = ff_layout_pg_init_write, 982 .pg_test = pnfs_generic_pg_test, 983 .pg_doio = pnfs_generic_pg_writepages, 984 .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write, 985 .pg_cleanup = pnfs_generic_pg_cleanup, 986 }; 987 988 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs) 989 { 990 struct rpc_task *task = &hdr->task; 991 992 pnfs_layoutcommit_inode(hdr->inode, false); 993 994 if (retry_pnfs) { 995 dprintk("%s Reset task %5u for i/o through pNFS " 996 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__, 997 hdr->task.tk_pid, 998 hdr->inode->i_sb->s_id, 999 (unsigned long long)NFS_FILEID(hdr->inode), 1000 hdr->args.count, 1001 (unsigned long long)hdr->args.offset); 1002 1003 hdr->completion_ops->reschedule_io(hdr); 1004 return; 1005 } 1006 1007 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { 1008 dprintk("%s Reset task %5u for i/o through MDS " 1009 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__, 1010 hdr->task.tk_pid, 1011 hdr->inode->i_sb->s_id, 1012 (unsigned long long)NFS_FILEID(hdr->inode), 1013 hdr->args.count, 1014 (unsigned long long)hdr->args.offset); 1015 1016 task->tk_status = pnfs_write_done_resend_to_mds(hdr); 1017 } 1018 } 1019 1020 static void ff_layout_reset_read(struct nfs_pgio_header *hdr) 1021 { 1022 struct rpc_task *task = &hdr->task; 1023 1024 pnfs_layoutcommit_inode(hdr->inode, false); 1025 1026 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { 1027 dprintk("%s Reset task %5u for i/o through MDS " 1028 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__, 1029 hdr->task.tk_pid, 1030 hdr->inode->i_sb->s_id, 1031 (unsigned long long)NFS_FILEID(hdr->inode), 1032 hdr->args.count, 1033 (unsigned long long)hdr->args.offset); 1034 1035 task->tk_status = pnfs_read_done_resend_to_mds(hdr); 1036 } 1037 } 1038 1039 static int ff_layout_async_handle_error_v4(struct rpc_task *task, 1040 struct nfs4_state *state, 1041 struct nfs_client *clp, 1042 struct pnfs_layout_segment *lseg, 1043 int idx) 1044 { 1045 struct pnfs_layout_hdr *lo = lseg->pls_layout; 1046 struct inode *inode = lo->plh_inode; 1047 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx); 1048 struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table; 1049 1050 switch (task->tk_status) { 1051 case -NFS4ERR_BADSESSION: 1052 case -NFS4ERR_BADSLOT: 1053 case -NFS4ERR_BAD_HIGH_SLOT: 1054 case -NFS4ERR_DEADSESSION: 1055 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION: 1056 case -NFS4ERR_SEQ_FALSE_RETRY: 1057 case -NFS4ERR_SEQ_MISORDERED: 1058 dprintk("%s ERROR %d, Reset session. Exchangeid " 1059 "flags 0x%x\n", __func__, task->tk_status, 1060 clp->cl_exchange_flags); 1061 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status); 1062 break; 1063 case -NFS4ERR_DELAY: 1064 case -NFS4ERR_GRACE: 1065 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX); 1066 break; 1067 case -NFS4ERR_RETRY_UNCACHED_REP: 1068 break; 1069 /* Invalidate Layout errors */ 1070 case -NFS4ERR_PNFS_NO_LAYOUT: 1071 case -ESTALE: /* mapped NFS4ERR_STALE */ 1072 case -EBADHANDLE: /* mapped NFS4ERR_BADHANDLE */ 1073 case -EISDIR: /* mapped NFS4ERR_ISDIR */ 1074 case -NFS4ERR_FHEXPIRED: 1075 case -NFS4ERR_WRONG_TYPE: 1076 dprintk("%s Invalid layout error %d\n", __func__, 1077 task->tk_status); 1078 /* 1079 * Destroy layout so new i/o will get a new layout. 1080 * Layout will not be destroyed until all current lseg 1081 * references are put. Mark layout as invalid to resend failed 1082 * i/o and all i/o waiting on the slot table to the MDS until 1083 * layout is destroyed and a new valid layout is obtained. 1084 */ 1085 pnfs_destroy_layout(NFS_I(inode)); 1086 rpc_wake_up(&tbl->slot_tbl_waitq); 1087 goto reset; 1088 /* RPC connection errors */ 1089 case -ECONNREFUSED: 1090 case -EHOSTDOWN: 1091 case -EHOSTUNREACH: 1092 case -ENETUNREACH: 1093 case -EIO: 1094 case -ETIMEDOUT: 1095 case -EPIPE: 1096 dprintk("%s DS connection error %d\n", __func__, 1097 task->tk_status); 1098 nfs4_delete_deviceid(devid->ld, devid->nfs_client, 1099 &devid->deviceid); 1100 rpc_wake_up(&tbl->slot_tbl_waitq); 1101 /* fall through */ 1102 default: 1103 if (ff_layout_avoid_mds_available_ds(lseg)) 1104 return -NFS4ERR_RESET_TO_PNFS; 1105 reset: 1106 dprintk("%s Retry through MDS. Error %d\n", __func__, 1107 task->tk_status); 1108 return -NFS4ERR_RESET_TO_MDS; 1109 } 1110 task->tk_status = 0; 1111 return -EAGAIN; 1112 } 1113 1114 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */ 1115 static int ff_layout_async_handle_error_v3(struct rpc_task *task, 1116 struct pnfs_layout_segment *lseg, 1117 int idx) 1118 { 1119 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx); 1120 1121 switch (task->tk_status) { 1122 /* File access problems. Don't mark the device as unavailable */ 1123 case -EACCES: 1124 case -ESTALE: 1125 case -EISDIR: 1126 case -EBADHANDLE: 1127 case -ELOOP: 1128 case -ENOSPC: 1129 break; 1130 case -EJUKEBOX: 1131 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY); 1132 goto out_retry; 1133 default: 1134 dprintk("%s DS connection error %d\n", __func__, 1135 task->tk_status); 1136 nfs4_delete_deviceid(devid->ld, devid->nfs_client, 1137 &devid->deviceid); 1138 } 1139 /* FIXME: Need to prevent infinite looping here. */ 1140 return -NFS4ERR_RESET_TO_PNFS; 1141 out_retry: 1142 task->tk_status = 0; 1143 rpc_restart_call_prepare(task); 1144 rpc_delay(task, NFS_JUKEBOX_RETRY_TIME); 1145 return -EAGAIN; 1146 } 1147 1148 static int ff_layout_async_handle_error(struct rpc_task *task, 1149 struct nfs4_state *state, 1150 struct nfs_client *clp, 1151 struct pnfs_layout_segment *lseg, 1152 int idx) 1153 { 1154 int vers = clp->cl_nfs_mod->rpc_vers->number; 1155 1156 if (task->tk_status >= 0) 1157 return 0; 1158 1159 /* Handle the case of an invalid layout segment */ 1160 if (!pnfs_is_valid_lseg(lseg)) 1161 return -NFS4ERR_RESET_TO_PNFS; 1162 1163 switch (vers) { 1164 case 3: 1165 return ff_layout_async_handle_error_v3(task, lseg, idx); 1166 case 4: 1167 return ff_layout_async_handle_error_v4(task, state, clp, 1168 lseg, idx); 1169 default: 1170 /* should never happen */ 1171 WARN_ON_ONCE(1); 1172 return 0; 1173 } 1174 } 1175 1176 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg, 1177 int idx, u64 offset, u64 length, 1178 u32 status, int opnum, int error) 1179 { 1180 struct nfs4_ff_layout_mirror *mirror; 1181 int err; 1182 1183 if (status == 0) { 1184 switch (error) { 1185 case -ETIMEDOUT: 1186 case -EPFNOSUPPORT: 1187 case -EPROTONOSUPPORT: 1188 case -EOPNOTSUPP: 1189 case -ECONNREFUSED: 1190 case -ECONNRESET: 1191 case -EHOSTDOWN: 1192 case -EHOSTUNREACH: 1193 case -ENETUNREACH: 1194 case -EADDRINUSE: 1195 case -ENOBUFS: 1196 case -EPIPE: 1197 case -EPERM: 1198 status = NFS4ERR_NXIO; 1199 break; 1200 case -EACCES: 1201 status = NFS4ERR_ACCESS; 1202 break; 1203 default: 1204 return; 1205 } 1206 } 1207 1208 switch (status) { 1209 case NFS4ERR_DELAY: 1210 case NFS4ERR_GRACE: 1211 return; 1212 default: 1213 break; 1214 } 1215 1216 mirror = FF_LAYOUT_COMP(lseg, idx); 1217 err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout), 1218 mirror, offset, length, status, opnum, 1219 GFP_NOIO); 1220 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode, lseg); 1221 dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status); 1222 } 1223 1224 /* NFS_PROTO call done callback routines */ 1225 static int ff_layout_read_done_cb(struct rpc_task *task, 1226 struct nfs_pgio_header *hdr) 1227 { 1228 int err; 1229 1230 trace_nfs4_pnfs_read(hdr, task->tk_status); 1231 if (task->tk_status < 0) 1232 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx, 1233 hdr->args.offset, hdr->args.count, 1234 hdr->res.op_status, OP_READ, 1235 task->tk_status); 1236 err = ff_layout_async_handle_error(task, hdr->args.context->state, 1237 hdr->ds_clp, hdr->lseg, 1238 hdr->pgio_mirror_idx); 1239 1240 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags); 1241 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags); 1242 switch (err) { 1243 case -NFS4ERR_RESET_TO_PNFS: 1244 if (ff_layout_choose_best_ds_for_read(hdr->lseg, 1245 hdr->pgio_mirror_idx + 1, 1246 &hdr->pgio_mirror_idx)) 1247 goto out_eagain; 1248 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags); 1249 return task->tk_status; 1250 case -NFS4ERR_RESET_TO_MDS: 1251 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags); 1252 return task->tk_status; 1253 case -EAGAIN: 1254 goto out_eagain; 1255 } 1256 1257 return 0; 1258 out_eagain: 1259 rpc_restart_call_prepare(task); 1260 return -EAGAIN; 1261 } 1262 1263 static bool 1264 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg) 1265 { 1266 return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT); 1267 } 1268 1269 /* 1270 * We reference the rpc_cred of the first WRITE that triggers the need for 1271 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound. 1272 * rfc5661 is not clear about which credential should be used. 1273 * 1274 * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so 1275 * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751 1276 * we always send layoutcommit after DS writes. 1277 */ 1278 static void 1279 ff_layout_set_layoutcommit(struct inode *inode, 1280 struct pnfs_layout_segment *lseg, 1281 loff_t end_offset) 1282 { 1283 if (!ff_layout_need_layoutcommit(lseg)) 1284 return; 1285 1286 pnfs_set_layoutcommit(inode, lseg, end_offset); 1287 dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino, 1288 (unsigned long long) NFS_I(inode)->layout->plh_lwb); 1289 } 1290 1291 static bool 1292 ff_layout_device_unavailable(struct pnfs_layout_segment *lseg, int idx) 1293 { 1294 /* No mirroring for now */ 1295 struct nfs4_deviceid_node *node = FF_LAYOUT_DEVID_NODE(lseg, idx); 1296 1297 return ff_layout_test_devid_unavailable(node); 1298 } 1299 1300 static void ff_layout_read_record_layoutstats_start(struct rpc_task *task, 1301 struct nfs_pgio_header *hdr) 1302 { 1303 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags)) 1304 return; 1305 nfs4_ff_layout_stat_io_start_read(hdr->inode, 1306 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx), 1307 hdr->args.count, 1308 task->tk_start); 1309 } 1310 1311 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task, 1312 struct nfs_pgio_header *hdr) 1313 { 1314 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags)) 1315 return; 1316 nfs4_ff_layout_stat_io_end_read(task, 1317 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx), 1318 hdr->args.count, 1319 hdr->res.count); 1320 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags); 1321 } 1322 1323 static int ff_layout_read_prepare_common(struct rpc_task *task, 1324 struct nfs_pgio_header *hdr) 1325 { 1326 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) { 1327 rpc_exit(task, -EIO); 1328 return -EIO; 1329 } 1330 if (ff_layout_device_unavailable(hdr->lseg, hdr->pgio_mirror_idx)) { 1331 rpc_exit(task, -EHOSTDOWN); 1332 return -EAGAIN; 1333 } 1334 1335 ff_layout_read_record_layoutstats_start(task, hdr); 1336 return 0; 1337 } 1338 1339 /* 1340 * Call ops for the async read/write cases 1341 * In the case of dense layouts, the offset needs to be reset to its 1342 * original value. 1343 */ 1344 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data) 1345 { 1346 struct nfs_pgio_header *hdr = data; 1347 1348 if (ff_layout_read_prepare_common(task, hdr)) 1349 return; 1350 1351 rpc_call_start(task); 1352 } 1353 1354 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data) 1355 { 1356 struct nfs_pgio_header *hdr = data; 1357 1358 if (nfs4_setup_sequence(hdr->ds_clp, 1359 &hdr->args.seq_args, 1360 &hdr->res.seq_res, 1361 task)) 1362 return; 1363 1364 ff_layout_read_prepare_common(task, hdr); 1365 } 1366 1367 static void ff_layout_read_call_done(struct rpc_task *task, void *data) 1368 { 1369 struct nfs_pgio_header *hdr = data; 1370 1371 dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status); 1372 1373 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) && 1374 task->tk_status == 0) { 1375 nfs4_sequence_done(task, &hdr->res.seq_res); 1376 return; 1377 } 1378 1379 /* Note this may cause RPC to be resent */ 1380 hdr->mds_ops->rpc_call_done(task, hdr); 1381 } 1382 1383 static void ff_layout_read_count_stats(struct rpc_task *task, void *data) 1384 { 1385 struct nfs_pgio_header *hdr = data; 1386 1387 ff_layout_read_record_layoutstats_done(task, hdr); 1388 rpc_count_iostats_metrics(task, 1389 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]); 1390 } 1391 1392 static void ff_layout_read_release(void *data) 1393 { 1394 struct nfs_pgio_header *hdr = data; 1395 1396 ff_layout_read_record_layoutstats_done(&hdr->task, hdr); 1397 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) 1398 pnfs_read_resend_pnfs(hdr); 1399 else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags)) 1400 ff_layout_reset_read(hdr); 1401 pnfs_generic_rw_release(data); 1402 } 1403 1404 1405 static int ff_layout_write_done_cb(struct rpc_task *task, 1406 struct nfs_pgio_header *hdr) 1407 { 1408 loff_t end_offs = 0; 1409 int err; 1410 1411 trace_nfs4_pnfs_write(hdr, task->tk_status); 1412 if (task->tk_status < 0) 1413 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx, 1414 hdr->args.offset, hdr->args.count, 1415 hdr->res.op_status, OP_WRITE, 1416 task->tk_status); 1417 err = ff_layout_async_handle_error(task, hdr->args.context->state, 1418 hdr->ds_clp, hdr->lseg, 1419 hdr->pgio_mirror_idx); 1420 1421 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags); 1422 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags); 1423 switch (err) { 1424 case -NFS4ERR_RESET_TO_PNFS: 1425 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags); 1426 return task->tk_status; 1427 case -NFS4ERR_RESET_TO_MDS: 1428 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags); 1429 return task->tk_status; 1430 case -EAGAIN: 1431 return -EAGAIN; 1432 } 1433 1434 if (hdr->res.verf->committed == NFS_FILE_SYNC || 1435 hdr->res.verf->committed == NFS_DATA_SYNC) 1436 end_offs = hdr->mds_offset + (loff_t)hdr->res.count; 1437 1438 /* Note: if the write is unstable, don't set end_offs until commit */ 1439 ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs); 1440 1441 /* zero out fattr since we don't care DS attr at all */ 1442 hdr->fattr.valid = 0; 1443 if (task->tk_status >= 0) 1444 nfs_writeback_update_inode(hdr); 1445 1446 return 0; 1447 } 1448 1449 static int ff_layout_commit_done_cb(struct rpc_task *task, 1450 struct nfs_commit_data *data) 1451 { 1452 int err; 1453 1454 trace_nfs4_pnfs_commit_ds(data, task->tk_status); 1455 if (task->tk_status < 0) 1456 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index, 1457 data->args.offset, data->args.count, 1458 data->res.op_status, OP_COMMIT, 1459 task->tk_status); 1460 err = ff_layout_async_handle_error(task, NULL, data->ds_clp, 1461 data->lseg, data->ds_commit_index); 1462 1463 switch (err) { 1464 case -NFS4ERR_RESET_TO_PNFS: 1465 pnfs_generic_prepare_to_resend_writes(data); 1466 return -EAGAIN; 1467 case -NFS4ERR_RESET_TO_MDS: 1468 pnfs_generic_prepare_to_resend_writes(data); 1469 return -EAGAIN; 1470 case -EAGAIN: 1471 rpc_restart_call_prepare(task); 1472 return -EAGAIN; 1473 } 1474 1475 ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb); 1476 1477 return 0; 1478 } 1479 1480 static void ff_layout_write_record_layoutstats_start(struct rpc_task *task, 1481 struct nfs_pgio_header *hdr) 1482 { 1483 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags)) 1484 return; 1485 nfs4_ff_layout_stat_io_start_write(hdr->inode, 1486 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx), 1487 hdr->args.count, 1488 task->tk_start); 1489 } 1490 1491 static void ff_layout_write_record_layoutstats_done(struct rpc_task *task, 1492 struct nfs_pgio_header *hdr) 1493 { 1494 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags)) 1495 return; 1496 nfs4_ff_layout_stat_io_end_write(task, 1497 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx), 1498 hdr->args.count, hdr->res.count, 1499 hdr->res.verf->committed); 1500 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags); 1501 } 1502 1503 static int ff_layout_write_prepare_common(struct rpc_task *task, 1504 struct nfs_pgio_header *hdr) 1505 { 1506 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) { 1507 rpc_exit(task, -EIO); 1508 return -EIO; 1509 } 1510 1511 if (ff_layout_device_unavailable(hdr->lseg, hdr->pgio_mirror_idx)) { 1512 rpc_exit(task, -EHOSTDOWN); 1513 return -EAGAIN; 1514 } 1515 1516 ff_layout_write_record_layoutstats_start(task, hdr); 1517 return 0; 1518 } 1519 1520 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data) 1521 { 1522 struct nfs_pgio_header *hdr = data; 1523 1524 if (ff_layout_write_prepare_common(task, hdr)) 1525 return; 1526 1527 rpc_call_start(task); 1528 } 1529 1530 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data) 1531 { 1532 struct nfs_pgio_header *hdr = data; 1533 1534 if (nfs4_setup_sequence(hdr->ds_clp, 1535 &hdr->args.seq_args, 1536 &hdr->res.seq_res, 1537 task)) 1538 return; 1539 1540 ff_layout_write_prepare_common(task, hdr); 1541 } 1542 1543 static void ff_layout_write_call_done(struct rpc_task *task, void *data) 1544 { 1545 struct nfs_pgio_header *hdr = data; 1546 1547 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) && 1548 task->tk_status == 0) { 1549 nfs4_sequence_done(task, &hdr->res.seq_res); 1550 return; 1551 } 1552 1553 /* Note this may cause RPC to be resent */ 1554 hdr->mds_ops->rpc_call_done(task, hdr); 1555 } 1556 1557 static void ff_layout_write_count_stats(struct rpc_task *task, void *data) 1558 { 1559 struct nfs_pgio_header *hdr = data; 1560 1561 ff_layout_write_record_layoutstats_done(task, hdr); 1562 rpc_count_iostats_metrics(task, 1563 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]); 1564 } 1565 1566 static void ff_layout_write_release(void *data) 1567 { 1568 struct nfs_pgio_header *hdr = data; 1569 1570 ff_layout_write_record_layoutstats_done(&hdr->task, hdr); 1571 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) 1572 ff_layout_reset_write(hdr, true); 1573 else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags)) 1574 ff_layout_reset_write(hdr, false); 1575 pnfs_generic_rw_release(data); 1576 } 1577 1578 static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task, 1579 struct nfs_commit_data *cdata) 1580 { 1581 if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags)) 1582 return; 1583 nfs4_ff_layout_stat_io_start_write(cdata->inode, 1584 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index), 1585 0, task->tk_start); 1586 } 1587 1588 static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task, 1589 struct nfs_commit_data *cdata) 1590 { 1591 struct nfs_page *req; 1592 __u64 count = 0; 1593 1594 if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags)) 1595 return; 1596 1597 if (task->tk_status == 0) { 1598 list_for_each_entry(req, &cdata->pages, wb_list) 1599 count += req->wb_bytes; 1600 } 1601 nfs4_ff_layout_stat_io_end_write(task, 1602 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index), 1603 count, count, NFS_FILE_SYNC); 1604 set_bit(NFS_LSEG_LAYOUTRETURN, &cdata->lseg->pls_flags); 1605 } 1606 1607 static void ff_layout_commit_prepare_common(struct rpc_task *task, 1608 struct nfs_commit_data *cdata) 1609 { 1610 ff_layout_commit_record_layoutstats_start(task, cdata); 1611 } 1612 1613 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data) 1614 { 1615 ff_layout_commit_prepare_common(task, data); 1616 rpc_call_start(task); 1617 } 1618 1619 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data) 1620 { 1621 struct nfs_commit_data *wdata = data; 1622 1623 if (nfs4_setup_sequence(wdata->ds_clp, 1624 &wdata->args.seq_args, 1625 &wdata->res.seq_res, 1626 task)) 1627 return; 1628 ff_layout_commit_prepare_common(task, data); 1629 } 1630 1631 static void ff_layout_commit_done(struct rpc_task *task, void *data) 1632 { 1633 pnfs_generic_write_commit_done(task, data); 1634 } 1635 1636 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data) 1637 { 1638 struct nfs_commit_data *cdata = data; 1639 1640 ff_layout_commit_record_layoutstats_done(task, cdata); 1641 rpc_count_iostats_metrics(task, 1642 &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]); 1643 } 1644 1645 static void ff_layout_commit_release(void *data) 1646 { 1647 struct nfs_commit_data *cdata = data; 1648 1649 ff_layout_commit_record_layoutstats_done(&cdata->task, cdata); 1650 pnfs_generic_commit_release(data); 1651 } 1652 1653 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = { 1654 .rpc_call_prepare = ff_layout_read_prepare_v3, 1655 .rpc_call_done = ff_layout_read_call_done, 1656 .rpc_count_stats = ff_layout_read_count_stats, 1657 .rpc_release = ff_layout_read_release, 1658 }; 1659 1660 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = { 1661 .rpc_call_prepare = ff_layout_read_prepare_v4, 1662 .rpc_call_done = ff_layout_read_call_done, 1663 .rpc_count_stats = ff_layout_read_count_stats, 1664 .rpc_release = ff_layout_read_release, 1665 }; 1666 1667 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = { 1668 .rpc_call_prepare = ff_layout_write_prepare_v3, 1669 .rpc_call_done = ff_layout_write_call_done, 1670 .rpc_count_stats = ff_layout_write_count_stats, 1671 .rpc_release = ff_layout_write_release, 1672 }; 1673 1674 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = { 1675 .rpc_call_prepare = ff_layout_write_prepare_v4, 1676 .rpc_call_done = ff_layout_write_call_done, 1677 .rpc_count_stats = ff_layout_write_count_stats, 1678 .rpc_release = ff_layout_write_release, 1679 }; 1680 1681 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = { 1682 .rpc_call_prepare = ff_layout_commit_prepare_v3, 1683 .rpc_call_done = ff_layout_commit_done, 1684 .rpc_count_stats = ff_layout_commit_count_stats, 1685 .rpc_release = ff_layout_commit_release, 1686 }; 1687 1688 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = { 1689 .rpc_call_prepare = ff_layout_commit_prepare_v4, 1690 .rpc_call_done = ff_layout_commit_done, 1691 .rpc_count_stats = ff_layout_commit_count_stats, 1692 .rpc_release = ff_layout_commit_release, 1693 }; 1694 1695 static enum pnfs_try_status 1696 ff_layout_read_pagelist(struct nfs_pgio_header *hdr) 1697 { 1698 struct pnfs_layout_segment *lseg = hdr->lseg; 1699 struct nfs4_pnfs_ds *ds; 1700 struct rpc_clnt *ds_clnt; 1701 struct rpc_cred *ds_cred; 1702 loff_t offset = hdr->args.offset; 1703 u32 idx = hdr->pgio_mirror_idx; 1704 int vers; 1705 struct nfs_fh *fh; 1706 1707 dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n", 1708 __func__, hdr->inode->i_ino, 1709 hdr->args.pgbase, (size_t)hdr->args.count, offset); 1710 1711 ds = nfs4_ff_layout_prepare_ds(lseg, idx, false); 1712 if (!ds) 1713 goto out_failed; 1714 1715 ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp, 1716 hdr->inode); 1717 if (IS_ERR(ds_clnt)) 1718 goto out_failed; 1719 1720 ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred); 1721 if (!ds_cred) 1722 goto out_failed; 1723 1724 vers = nfs4_ff_layout_ds_version(lseg, idx); 1725 1726 dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__, 1727 ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), vers); 1728 1729 hdr->pgio_done_cb = ff_layout_read_done_cb; 1730 refcount_inc(&ds->ds_clp->cl_count); 1731 hdr->ds_clp = ds->ds_clp; 1732 fh = nfs4_ff_layout_select_ds_fh(lseg, idx); 1733 if (fh) 1734 hdr->args.fh = fh; 1735 1736 if (!nfs4_ff_layout_select_ds_stateid(lseg, idx, &hdr->args.stateid)) 1737 goto out_failed; 1738 1739 /* 1740 * Note that if we ever decide to split across DSes, 1741 * then we may need to handle dense-like offsets. 1742 */ 1743 hdr->args.offset = offset; 1744 hdr->mds_offset = offset; 1745 1746 /* Perform an asynchronous read to ds */ 1747 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops, 1748 vers == 3 ? &ff_layout_read_call_ops_v3 : 1749 &ff_layout_read_call_ops_v4, 1750 0, RPC_TASK_SOFTCONN); 1751 put_rpccred(ds_cred); 1752 return PNFS_ATTEMPTED; 1753 1754 out_failed: 1755 if (ff_layout_avoid_mds_available_ds(lseg)) 1756 return PNFS_TRY_AGAIN; 1757 return PNFS_NOT_ATTEMPTED; 1758 } 1759 1760 /* Perform async writes. */ 1761 static enum pnfs_try_status 1762 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync) 1763 { 1764 struct pnfs_layout_segment *lseg = hdr->lseg; 1765 struct nfs4_pnfs_ds *ds; 1766 struct rpc_clnt *ds_clnt; 1767 struct rpc_cred *ds_cred; 1768 loff_t offset = hdr->args.offset; 1769 int vers; 1770 struct nfs_fh *fh; 1771 int idx = hdr->pgio_mirror_idx; 1772 1773 ds = nfs4_ff_layout_prepare_ds(lseg, idx, true); 1774 if (!ds) 1775 goto out_failed; 1776 1777 ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp, 1778 hdr->inode); 1779 if (IS_ERR(ds_clnt)) 1780 goto out_failed; 1781 1782 ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred); 1783 if (!ds_cred) 1784 goto out_failed; 1785 1786 vers = nfs4_ff_layout_ds_version(lseg, idx); 1787 1788 dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n", 1789 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count, 1790 offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), 1791 vers); 1792 1793 hdr->pgio_done_cb = ff_layout_write_done_cb; 1794 refcount_inc(&ds->ds_clp->cl_count); 1795 hdr->ds_clp = ds->ds_clp; 1796 hdr->ds_commit_idx = idx; 1797 fh = nfs4_ff_layout_select_ds_fh(lseg, idx); 1798 if (fh) 1799 hdr->args.fh = fh; 1800 1801 if (!nfs4_ff_layout_select_ds_stateid(lseg, idx, &hdr->args.stateid)) 1802 goto out_failed; 1803 1804 /* 1805 * Note that if we ever decide to split across DSes, 1806 * then we may need to handle dense-like offsets. 1807 */ 1808 hdr->args.offset = offset; 1809 1810 /* Perform an asynchronous write */ 1811 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops, 1812 vers == 3 ? &ff_layout_write_call_ops_v3 : 1813 &ff_layout_write_call_ops_v4, 1814 sync, RPC_TASK_SOFTCONN); 1815 put_rpccred(ds_cred); 1816 return PNFS_ATTEMPTED; 1817 1818 out_failed: 1819 if (ff_layout_avoid_mds_available_ds(lseg)) 1820 return PNFS_TRY_AGAIN; 1821 return PNFS_NOT_ATTEMPTED; 1822 } 1823 1824 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i) 1825 { 1826 return i; 1827 } 1828 1829 static struct nfs_fh * 1830 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i) 1831 { 1832 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg); 1833 1834 /* FIXME: Assume that there is only one NFS version available 1835 * for the DS. 1836 */ 1837 return &flseg->mirror_array[i]->fh_versions[0]; 1838 } 1839 1840 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how) 1841 { 1842 struct pnfs_layout_segment *lseg = data->lseg; 1843 struct nfs4_pnfs_ds *ds; 1844 struct rpc_clnt *ds_clnt; 1845 struct rpc_cred *ds_cred; 1846 u32 idx; 1847 int vers, ret; 1848 struct nfs_fh *fh; 1849 1850 if (!lseg || !(pnfs_is_valid_lseg(lseg) || 1851 test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))) 1852 goto out_err; 1853 1854 idx = calc_ds_index_from_commit(lseg, data->ds_commit_index); 1855 ds = nfs4_ff_layout_prepare_ds(lseg, idx, true); 1856 if (!ds) 1857 goto out_err; 1858 1859 ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp, 1860 data->inode); 1861 if (IS_ERR(ds_clnt)) 1862 goto out_err; 1863 1864 ds_cred = ff_layout_get_ds_cred(lseg, idx, data->cred); 1865 if (!ds_cred) 1866 goto out_err; 1867 1868 vers = nfs4_ff_layout_ds_version(lseg, idx); 1869 1870 dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__, 1871 data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count), 1872 vers); 1873 data->commit_done_cb = ff_layout_commit_done_cb; 1874 data->cred = ds_cred; 1875 refcount_inc(&ds->ds_clp->cl_count); 1876 data->ds_clp = ds->ds_clp; 1877 fh = select_ds_fh_from_commit(lseg, data->ds_commit_index); 1878 if (fh) 1879 data->args.fh = fh; 1880 1881 ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops, 1882 vers == 3 ? &ff_layout_commit_call_ops_v3 : 1883 &ff_layout_commit_call_ops_v4, 1884 how, RPC_TASK_SOFTCONN); 1885 put_rpccred(ds_cred); 1886 return ret; 1887 out_err: 1888 pnfs_generic_prepare_to_resend_writes(data); 1889 pnfs_generic_commit_release(data); 1890 return -EAGAIN; 1891 } 1892 1893 static int 1894 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages, 1895 int how, struct nfs_commit_info *cinfo) 1896 { 1897 return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo, 1898 ff_layout_initiate_commit); 1899 } 1900 1901 static struct pnfs_ds_commit_info * 1902 ff_layout_get_ds_info(struct inode *inode) 1903 { 1904 struct pnfs_layout_hdr *layout = NFS_I(inode)->layout; 1905 1906 if (layout == NULL) 1907 return NULL; 1908 1909 return &FF_LAYOUT_FROM_HDR(layout)->commit_info; 1910 } 1911 1912 static void 1913 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d) 1914 { 1915 nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds, 1916 id_node)); 1917 } 1918 1919 static int ff_layout_encode_ioerr(struct xdr_stream *xdr, 1920 const struct nfs4_layoutreturn_args *args, 1921 const struct nfs4_flexfile_layoutreturn_args *ff_args) 1922 { 1923 __be32 *start; 1924 1925 start = xdr_reserve_space(xdr, 4); 1926 if (unlikely(!start)) 1927 return -E2BIG; 1928 1929 *start = cpu_to_be32(ff_args->num_errors); 1930 /* This assume we always return _ALL_ layouts */ 1931 return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors); 1932 } 1933 1934 static void 1935 encode_opaque_fixed(struct xdr_stream *xdr, const void *buf, size_t len) 1936 { 1937 WARN_ON_ONCE(xdr_stream_encode_opaque_fixed(xdr, buf, len) < 0); 1938 } 1939 1940 static void 1941 ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr, 1942 const nfs4_stateid *stateid, 1943 const struct nfs42_layoutstat_devinfo *devinfo) 1944 { 1945 __be32 *p; 1946 1947 p = xdr_reserve_space(xdr, 8 + 8); 1948 p = xdr_encode_hyper(p, devinfo->offset); 1949 p = xdr_encode_hyper(p, devinfo->length); 1950 encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE); 1951 p = xdr_reserve_space(xdr, 4*8); 1952 p = xdr_encode_hyper(p, devinfo->read_count); 1953 p = xdr_encode_hyper(p, devinfo->read_bytes); 1954 p = xdr_encode_hyper(p, devinfo->write_count); 1955 p = xdr_encode_hyper(p, devinfo->write_bytes); 1956 encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE); 1957 } 1958 1959 static void 1960 ff_layout_encode_ff_iostat(struct xdr_stream *xdr, 1961 const nfs4_stateid *stateid, 1962 const struct nfs42_layoutstat_devinfo *devinfo) 1963 { 1964 ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo); 1965 ff_layout_encode_ff_layoutupdate(xdr, devinfo, 1966 devinfo->ld_private.data); 1967 } 1968 1969 /* report nothing for now */ 1970 static void ff_layout_encode_iostats_array(struct xdr_stream *xdr, 1971 const struct nfs4_layoutreturn_args *args, 1972 struct nfs4_flexfile_layoutreturn_args *ff_args) 1973 { 1974 __be32 *p; 1975 int i; 1976 1977 p = xdr_reserve_space(xdr, 4); 1978 *p = cpu_to_be32(ff_args->num_dev); 1979 for (i = 0; i < ff_args->num_dev; i++) 1980 ff_layout_encode_ff_iostat(xdr, 1981 &args->layout->plh_stateid, 1982 &ff_args->devinfo[i]); 1983 } 1984 1985 static void 1986 ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo, 1987 unsigned int num_entries) 1988 { 1989 unsigned int i; 1990 1991 for (i = 0; i < num_entries; i++) { 1992 if (!devinfo[i].ld_private.ops) 1993 continue; 1994 if (!devinfo[i].ld_private.ops->free) 1995 continue; 1996 devinfo[i].ld_private.ops->free(&devinfo[i].ld_private); 1997 } 1998 } 1999 2000 static struct nfs4_deviceid_node * 2001 ff_layout_alloc_deviceid_node(struct nfs_server *server, 2002 struct pnfs_device *pdev, gfp_t gfp_flags) 2003 { 2004 struct nfs4_ff_layout_ds *dsaddr; 2005 2006 dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags); 2007 if (!dsaddr) 2008 return NULL; 2009 return &dsaddr->id_node; 2010 } 2011 2012 static void 2013 ff_layout_encode_layoutreturn(struct xdr_stream *xdr, 2014 const void *voidargs, 2015 const struct nfs4_xdr_opaque_data *ff_opaque) 2016 { 2017 const struct nfs4_layoutreturn_args *args = voidargs; 2018 struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data; 2019 struct xdr_buf tmp_buf = { 2020 .head = { 2021 [0] = { 2022 .iov_base = page_address(ff_args->pages[0]), 2023 }, 2024 }, 2025 .buflen = PAGE_SIZE, 2026 }; 2027 struct xdr_stream tmp_xdr; 2028 __be32 *start; 2029 2030 dprintk("%s: Begin\n", __func__); 2031 2032 xdr_init_encode(&tmp_xdr, &tmp_buf, NULL); 2033 2034 ff_layout_encode_ioerr(&tmp_xdr, args, ff_args); 2035 ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args); 2036 2037 start = xdr_reserve_space(xdr, 4); 2038 *start = cpu_to_be32(tmp_buf.len); 2039 xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len); 2040 2041 dprintk("%s: Return\n", __func__); 2042 } 2043 2044 static void 2045 ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args) 2046 { 2047 struct nfs4_flexfile_layoutreturn_args *ff_args; 2048 2049 if (!args->data) 2050 return; 2051 ff_args = args->data; 2052 args->data = NULL; 2053 2054 ff_layout_free_ds_ioerr(&ff_args->errors); 2055 ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev); 2056 2057 put_page(ff_args->pages[0]); 2058 kfree(ff_args); 2059 } 2060 2061 static const struct nfs4_xdr_opaque_ops layoutreturn_ops = { 2062 .encode = ff_layout_encode_layoutreturn, 2063 .free = ff_layout_free_layoutreturn, 2064 }; 2065 2066 static int 2067 ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args) 2068 { 2069 struct nfs4_flexfile_layoutreturn_args *ff_args; 2070 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout); 2071 2072 ff_args = kmalloc(sizeof(*ff_args), GFP_KERNEL); 2073 if (!ff_args) 2074 goto out_nomem; 2075 ff_args->pages[0] = alloc_page(GFP_KERNEL); 2076 if (!ff_args->pages[0]) 2077 goto out_nomem_free; 2078 2079 INIT_LIST_HEAD(&ff_args->errors); 2080 ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout, 2081 &args->range, &ff_args->errors, 2082 FF_LAYOUTRETURN_MAXERR); 2083 2084 spin_lock(&args->inode->i_lock); 2085 ff_args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr, 2086 &ff_args->devinfo[0], ARRAY_SIZE(ff_args->devinfo)); 2087 spin_unlock(&args->inode->i_lock); 2088 2089 args->ld_private->ops = &layoutreturn_ops; 2090 args->ld_private->data = ff_args; 2091 return 0; 2092 out_nomem_free: 2093 kfree(ff_args); 2094 out_nomem: 2095 return -ENOMEM; 2096 } 2097 2098 static int 2099 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen) 2100 { 2101 const struct sockaddr_in *sin = (struct sockaddr_in *)sap; 2102 2103 return snprintf(buf, buflen, "%pI4", &sin->sin_addr); 2104 } 2105 2106 static size_t 2107 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf, 2108 const int buflen) 2109 { 2110 const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap; 2111 const struct in6_addr *addr = &sin6->sin6_addr; 2112 2113 /* 2114 * RFC 4291, Section 2.2.2 2115 * 2116 * Shorthanded ANY address 2117 */ 2118 if (ipv6_addr_any(addr)) 2119 return snprintf(buf, buflen, "::"); 2120 2121 /* 2122 * RFC 4291, Section 2.2.2 2123 * 2124 * Shorthanded loopback address 2125 */ 2126 if (ipv6_addr_loopback(addr)) 2127 return snprintf(buf, buflen, "::1"); 2128 2129 /* 2130 * RFC 4291, Section 2.2.3 2131 * 2132 * Special presentation address format for mapped v4 2133 * addresses. 2134 */ 2135 if (ipv6_addr_v4mapped(addr)) 2136 return snprintf(buf, buflen, "::ffff:%pI4", 2137 &addr->s6_addr32[3]); 2138 2139 /* 2140 * RFC 4291, Section 2.2.1 2141 */ 2142 return snprintf(buf, buflen, "%pI6c", addr); 2143 } 2144 2145 /* Derived from rpc_sockaddr2uaddr */ 2146 static void 2147 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da) 2148 { 2149 struct sockaddr *sap = (struct sockaddr *)&da->da_addr; 2150 char portbuf[RPCBIND_MAXUADDRPLEN]; 2151 char addrbuf[RPCBIND_MAXUADDRLEN]; 2152 char *netid; 2153 unsigned short port; 2154 int len, netid_len; 2155 __be32 *p; 2156 2157 switch (sap->sa_family) { 2158 case AF_INET: 2159 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0) 2160 return; 2161 port = ntohs(((struct sockaddr_in *)sap)->sin_port); 2162 netid = "tcp"; 2163 netid_len = 3; 2164 break; 2165 case AF_INET6: 2166 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0) 2167 return; 2168 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port); 2169 netid = "tcp6"; 2170 netid_len = 4; 2171 break; 2172 default: 2173 /* we only support tcp and tcp6 */ 2174 WARN_ON_ONCE(1); 2175 return; 2176 } 2177 2178 snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff); 2179 len = strlcat(addrbuf, portbuf, sizeof(addrbuf)); 2180 2181 p = xdr_reserve_space(xdr, 4 + netid_len); 2182 xdr_encode_opaque(p, netid, netid_len); 2183 2184 p = xdr_reserve_space(xdr, 4 + len); 2185 xdr_encode_opaque(p, addrbuf, len); 2186 } 2187 2188 static void 2189 ff_layout_encode_nfstime(struct xdr_stream *xdr, 2190 ktime_t t) 2191 { 2192 struct timespec64 ts; 2193 __be32 *p; 2194 2195 p = xdr_reserve_space(xdr, 12); 2196 ts = ktime_to_timespec64(t); 2197 p = xdr_encode_hyper(p, ts.tv_sec); 2198 *p++ = cpu_to_be32(ts.tv_nsec); 2199 } 2200 2201 static void 2202 ff_layout_encode_io_latency(struct xdr_stream *xdr, 2203 struct nfs4_ff_io_stat *stat) 2204 { 2205 __be32 *p; 2206 2207 p = xdr_reserve_space(xdr, 5 * 8); 2208 p = xdr_encode_hyper(p, stat->ops_requested); 2209 p = xdr_encode_hyper(p, stat->bytes_requested); 2210 p = xdr_encode_hyper(p, stat->ops_completed); 2211 p = xdr_encode_hyper(p, stat->bytes_completed); 2212 p = xdr_encode_hyper(p, stat->bytes_not_delivered); 2213 ff_layout_encode_nfstime(xdr, stat->total_busy_time); 2214 ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time); 2215 } 2216 2217 static void 2218 ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr, 2219 const struct nfs42_layoutstat_devinfo *devinfo, 2220 struct nfs4_ff_layout_mirror *mirror) 2221 { 2222 struct nfs4_pnfs_ds_addr *da; 2223 struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds; 2224 struct nfs_fh *fh = &mirror->fh_versions[0]; 2225 __be32 *p; 2226 2227 da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node); 2228 dprintk("%s: DS %s: encoding address %s\n", 2229 __func__, ds->ds_remotestr, da->da_remotestr); 2230 /* netaddr4 */ 2231 ff_layout_encode_netaddr(xdr, da); 2232 /* nfs_fh4 */ 2233 p = xdr_reserve_space(xdr, 4 + fh->size); 2234 xdr_encode_opaque(p, fh->data, fh->size); 2235 /* ff_io_latency4 read */ 2236 spin_lock(&mirror->lock); 2237 ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat); 2238 /* ff_io_latency4 write */ 2239 ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat); 2240 spin_unlock(&mirror->lock); 2241 /* nfstime4 */ 2242 ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time)); 2243 /* bool */ 2244 p = xdr_reserve_space(xdr, 4); 2245 *p = cpu_to_be32(false); 2246 } 2247 2248 static void 2249 ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args, 2250 const struct nfs4_xdr_opaque_data *opaque) 2251 { 2252 struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque, 2253 struct nfs42_layoutstat_devinfo, ld_private); 2254 __be32 *start; 2255 2256 /* layoutupdate length */ 2257 start = xdr_reserve_space(xdr, 4); 2258 ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data); 2259 2260 *start = cpu_to_be32((xdr->p - start - 1) * 4); 2261 } 2262 2263 static void 2264 ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque) 2265 { 2266 struct nfs4_ff_layout_mirror *mirror = opaque->data; 2267 2268 ff_layout_put_mirror(mirror); 2269 } 2270 2271 static const struct nfs4_xdr_opaque_ops layoutstat_ops = { 2272 .encode = ff_layout_encode_layoutstats, 2273 .free = ff_layout_free_layoutstats, 2274 }; 2275 2276 static int 2277 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo, 2278 struct nfs42_layoutstat_devinfo *devinfo, 2279 int dev_limit) 2280 { 2281 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo); 2282 struct nfs4_ff_layout_mirror *mirror; 2283 struct nfs4_deviceid_node *dev; 2284 int i = 0; 2285 2286 list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) { 2287 if (i >= dev_limit) 2288 break; 2289 if (IS_ERR_OR_NULL(mirror->mirror_ds)) 2290 continue; 2291 if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags)) 2292 continue; 2293 /* mirror refcount put in cleanup_layoutstats */ 2294 if (!refcount_inc_not_zero(&mirror->ref)) 2295 continue; 2296 dev = &mirror->mirror_ds->id_node; 2297 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE); 2298 devinfo->offset = 0; 2299 devinfo->length = NFS4_MAX_UINT64; 2300 spin_lock(&mirror->lock); 2301 devinfo->read_count = mirror->read_stat.io_stat.ops_completed; 2302 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed; 2303 devinfo->write_count = mirror->write_stat.io_stat.ops_completed; 2304 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed; 2305 spin_unlock(&mirror->lock); 2306 devinfo->layout_type = LAYOUT_FLEX_FILES; 2307 devinfo->ld_private.ops = &layoutstat_ops; 2308 devinfo->ld_private.data = mirror; 2309 2310 devinfo++; 2311 i++; 2312 } 2313 return i; 2314 } 2315 2316 static int 2317 ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args) 2318 { 2319 struct nfs4_flexfile_layout *ff_layout; 2320 const int dev_count = PNFS_LAYOUTSTATS_MAXDEV; 2321 2322 /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */ 2323 args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO); 2324 if (!args->devinfo) 2325 return -ENOMEM; 2326 2327 spin_lock(&args->inode->i_lock); 2328 ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout); 2329 args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr, 2330 &args->devinfo[0], dev_count); 2331 spin_unlock(&args->inode->i_lock); 2332 if (!args->num_dev) { 2333 kfree(args->devinfo); 2334 args->devinfo = NULL; 2335 return -ENOENT; 2336 } 2337 2338 return 0; 2339 } 2340 2341 static int 2342 ff_layout_set_layoutdriver(struct nfs_server *server, 2343 const struct nfs_fh *dummy) 2344 { 2345 #if IS_ENABLED(CONFIG_NFS_V4_2) 2346 server->caps |= NFS_CAP_LAYOUTSTATS; 2347 #endif 2348 return 0; 2349 } 2350 2351 static struct pnfs_layoutdriver_type flexfilelayout_type = { 2352 .id = LAYOUT_FLEX_FILES, 2353 .name = "LAYOUT_FLEX_FILES", 2354 .owner = THIS_MODULE, 2355 .flags = PNFS_LAYOUTGET_ON_OPEN, 2356 .max_layoutget_response = 4096, /* 1 page or so... */ 2357 .set_layoutdriver = ff_layout_set_layoutdriver, 2358 .alloc_layout_hdr = ff_layout_alloc_layout_hdr, 2359 .free_layout_hdr = ff_layout_free_layout_hdr, 2360 .alloc_lseg = ff_layout_alloc_lseg, 2361 .free_lseg = ff_layout_free_lseg, 2362 .add_lseg = ff_layout_add_lseg, 2363 .pg_read_ops = &ff_layout_pg_read_ops, 2364 .pg_write_ops = &ff_layout_pg_write_ops, 2365 .get_ds_info = ff_layout_get_ds_info, 2366 .free_deviceid_node = ff_layout_free_deviceid_node, 2367 .mark_request_commit = pnfs_layout_mark_request_commit, 2368 .clear_request_commit = pnfs_generic_clear_request_commit, 2369 .scan_commit_lists = pnfs_generic_scan_commit_lists, 2370 .recover_commit_reqs = pnfs_generic_recover_commit_reqs, 2371 .commit_pagelist = ff_layout_commit_pagelist, 2372 .read_pagelist = ff_layout_read_pagelist, 2373 .write_pagelist = ff_layout_write_pagelist, 2374 .alloc_deviceid_node = ff_layout_alloc_deviceid_node, 2375 .prepare_layoutreturn = ff_layout_prepare_layoutreturn, 2376 .sync = pnfs_nfs_generic_sync, 2377 .prepare_layoutstats = ff_layout_prepare_layoutstats, 2378 }; 2379 2380 static int __init nfs4flexfilelayout_init(void) 2381 { 2382 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n", 2383 __func__); 2384 if (!ff_zero_group) { 2385 ff_zero_group = groups_alloc(0); 2386 if (!ff_zero_group) 2387 return -ENOMEM; 2388 } 2389 return pnfs_register_layoutdriver(&flexfilelayout_type); 2390 } 2391 2392 static void __exit nfs4flexfilelayout_exit(void) 2393 { 2394 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n", 2395 __func__); 2396 pnfs_unregister_layoutdriver(&flexfilelayout_type); 2397 if (ff_zero_group) { 2398 put_group_info(ff_zero_group); 2399 ff_zero_group = NULL; 2400 } 2401 } 2402 2403 MODULE_ALIAS("nfs-layouttype4-4"); 2404 2405 MODULE_LICENSE("GPL"); 2406 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver"); 2407 2408 module_init(nfs4flexfilelayout_init); 2409 module_exit(nfs4flexfilelayout_exit); 2410