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 (atomic_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 atomic_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 && atomic_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 p = xdr_inline_decode(&stream, 4); 458 if (!p) 459 goto out_err_free; 460 fh_count = be32_to_cpup(p); 461 462 fls->mirror_array[i]->fh_versions = 463 kzalloc(fh_count * sizeof(struct nfs_fh), 464 gfp_flags); 465 if (fls->mirror_array[i]->fh_versions == NULL) { 466 rc = -ENOMEM; 467 goto out_err_free; 468 } 469 470 for (j = 0; j < fh_count; j++) { 471 rc = decode_nfs_fh(&stream, 472 &fls->mirror_array[i]->fh_versions[j]); 473 if (rc) 474 goto out_err_free; 475 } 476 477 fls->mirror_array[i]->fh_versions_cnt = fh_count; 478 479 /* user */ 480 rc = decode_name(&stream, &id); 481 if (rc) 482 goto out_err_free; 483 484 acred.uid = make_kuid(&init_user_ns, id); 485 486 /* group */ 487 rc = decode_name(&stream, &id); 488 if (rc) 489 goto out_err_free; 490 491 acred.gid = make_kgid(&init_user_ns, id); 492 493 /* find the cred for it */ 494 rcu_assign_pointer(cred, rpc_lookup_generic_cred(&acred, 0, gfp_flags)); 495 if (IS_ERR(cred)) { 496 rc = PTR_ERR(cred); 497 goto out_err_free; 498 } 499 500 if (lgr->range.iomode == IOMODE_READ) 501 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred); 502 else 503 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred); 504 505 mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]); 506 if (mirror != fls->mirror_array[i]) { 507 /* swap cred ptrs so free_mirror will clean up old */ 508 if (lgr->range.iomode == IOMODE_READ) { 509 cred = xchg(&mirror->ro_cred, cred); 510 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred); 511 } else { 512 cred = xchg(&mirror->rw_cred, cred); 513 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred); 514 } 515 ff_layout_free_mirror(fls->mirror_array[i]); 516 fls->mirror_array[i] = mirror; 517 } 518 519 dprintk("%s: iomode %s uid %u gid %u\n", __func__, 520 lgr->range.iomode == IOMODE_READ ? "READ" : "RW", 521 from_kuid(&init_user_ns, acred.uid), 522 from_kgid(&init_user_ns, acred.gid)); 523 } 524 525 p = xdr_inline_decode(&stream, 4); 526 if (!p) 527 goto out_sort_mirrors; 528 fls->flags = be32_to_cpup(p); 529 530 p = xdr_inline_decode(&stream, 4); 531 if (!p) 532 goto out_sort_mirrors; 533 for (i=0; i < fls->mirror_array_cnt; i++) 534 fls->mirror_array[i]->report_interval = be32_to_cpup(p); 535 536 out_sort_mirrors: 537 ff_layout_sort_mirrors(fls); 538 rc = ff_layout_check_layout(lgr); 539 if (rc) 540 goto out_err_free; 541 ret = &fls->generic_hdr; 542 dprintk("<-- %s (success)\n", __func__); 543 out_free_page: 544 __free_page(scratch); 545 return ret; 546 out_err_free: 547 _ff_layout_free_lseg(fls); 548 ret = ERR_PTR(rc); 549 dprintk("<-- %s (%d)\n", __func__, rc); 550 goto out_free_page; 551 } 552 553 static bool ff_layout_has_rw_segments(struct pnfs_layout_hdr *layout) 554 { 555 struct pnfs_layout_segment *lseg; 556 557 list_for_each_entry(lseg, &layout->plh_segs, pls_list) 558 if (lseg->pls_range.iomode == IOMODE_RW) 559 return true; 560 561 return false; 562 } 563 564 static void 565 ff_layout_free_lseg(struct pnfs_layout_segment *lseg) 566 { 567 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg); 568 569 dprintk("--> %s\n", __func__); 570 571 if (lseg->pls_range.iomode == IOMODE_RW) { 572 struct nfs4_flexfile_layout *ffl; 573 struct inode *inode; 574 575 ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout); 576 inode = ffl->generic_hdr.plh_inode; 577 spin_lock(&inode->i_lock); 578 if (!ff_layout_has_rw_segments(lseg->pls_layout)) { 579 ffl->commit_info.nbuckets = 0; 580 kfree(ffl->commit_info.buckets); 581 ffl->commit_info.buckets = NULL; 582 } 583 spin_unlock(&inode->i_lock); 584 } 585 _ff_layout_free_lseg(fls); 586 } 587 588 /* Return 1 until we have multiple lsegs support */ 589 static int 590 ff_layout_get_lseg_count(struct nfs4_ff_layout_segment *fls) 591 { 592 return 1; 593 } 594 595 static void 596 nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now) 597 { 598 /* first IO request? */ 599 if (atomic_inc_return(&timer->n_ops) == 1) { 600 timer->start_time = now; 601 } 602 } 603 604 static ktime_t 605 nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now) 606 { 607 ktime_t start; 608 609 if (atomic_dec_return(&timer->n_ops) < 0) 610 WARN_ON_ONCE(1); 611 612 start = timer->start_time; 613 timer->start_time = now; 614 return ktime_sub(now, start); 615 } 616 617 static bool 618 nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror, 619 struct nfs4_ff_layoutstat *layoutstat, 620 ktime_t now) 621 { 622 static const ktime_t notime = {0}; 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 (ktime_equal(mirror->start_time, notime)) 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 retry_strict: 816 pnfs_put_lseg(pgio->pg_lseg); 817 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, 818 req->wb_context, 819 0, 820 NFS4_MAX_UINT64, 821 IOMODE_READ, 822 strict_iomode, 823 GFP_KERNEL); 824 if (IS_ERR(pgio->pg_lseg)) { 825 pgio->pg_error = PTR_ERR(pgio->pg_lseg); 826 pgio->pg_lseg = NULL; 827 } 828 829 /* If we don't have checking, do get a IOMODE_RW 830 * segment, and the server wants to avoid READs 831 * there, then retry! 832 */ 833 if (pgio->pg_lseg && !strict_iomode && 834 ff_layout_avoid_read_on_rw(pgio->pg_lseg)) { 835 strict_iomode = true; 836 goto retry_strict; 837 } 838 } 839 840 static void 841 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio, 842 struct nfs_page *req) 843 { 844 struct nfs_pgio_mirror *pgm; 845 struct nfs4_ff_layout_mirror *mirror; 846 struct nfs4_pnfs_ds *ds; 847 int ds_idx; 848 849 retry: 850 /* Use full layout for now */ 851 if (!pgio->pg_lseg) 852 ff_layout_pg_get_read(pgio, req, false); 853 else if (ff_layout_avoid_read_on_rw(pgio->pg_lseg)) 854 ff_layout_pg_get_read(pgio, req, true); 855 856 /* If no lseg, fall back to read through mds */ 857 if (pgio->pg_lseg == NULL) 858 goto out_mds; 859 860 ds = ff_layout_choose_best_ds_for_read(pgio->pg_lseg, 0, &ds_idx); 861 if (!ds) { 862 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg)) 863 goto out_mds; 864 pnfs_put_lseg(pgio->pg_lseg); 865 pgio->pg_lseg = NULL; 866 /* Sleep for 1 second before retrying */ 867 ssleep(1); 868 goto retry; 869 } 870 871 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx); 872 873 pgio->pg_mirror_idx = ds_idx; 874 875 /* read always uses only one mirror - idx 0 for pgio layer */ 876 pgm = &pgio->pg_mirrors[0]; 877 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize; 878 879 return; 880 out_mds: 881 pnfs_put_lseg(pgio->pg_lseg); 882 pgio->pg_lseg = NULL; 883 nfs_pageio_reset_read_mds(pgio); 884 } 885 886 static void 887 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio, 888 struct nfs_page *req) 889 { 890 struct nfs4_ff_layout_mirror *mirror; 891 struct nfs_pgio_mirror *pgm; 892 struct nfs_commit_info cinfo; 893 struct nfs4_pnfs_ds *ds; 894 int i; 895 int status; 896 897 retry: 898 if (!pgio->pg_lseg) { 899 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, 900 req->wb_context, 901 0, 902 NFS4_MAX_UINT64, 903 IOMODE_RW, 904 false, 905 GFP_NOFS); 906 if (IS_ERR(pgio->pg_lseg)) { 907 pgio->pg_error = PTR_ERR(pgio->pg_lseg); 908 pgio->pg_lseg = NULL; 909 return; 910 } 911 } 912 /* If no lseg, fall back to write through mds */ 913 if (pgio->pg_lseg == NULL) 914 goto out_mds; 915 916 nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq); 917 status = ff_layout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS); 918 if (status < 0) 919 goto out_mds; 920 921 /* Use a direct mapping of ds_idx to pgio mirror_idx */ 922 if (WARN_ON_ONCE(pgio->pg_mirror_count != 923 FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg))) 924 goto out_mds; 925 926 for (i = 0; i < pgio->pg_mirror_count; i++) { 927 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, i, true); 928 if (!ds) { 929 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg)) 930 goto out_mds; 931 pnfs_put_lseg(pgio->pg_lseg); 932 pgio->pg_lseg = NULL; 933 /* Sleep for 1 second before retrying */ 934 ssleep(1); 935 goto retry; 936 } 937 pgm = &pgio->pg_mirrors[i]; 938 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i); 939 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize; 940 } 941 942 return; 943 944 out_mds: 945 pnfs_put_lseg(pgio->pg_lseg); 946 pgio->pg_lseg = NULL; 947 nfs_pageio_reset_write_mds(pgio); 948 } 949 950 static unsigned int 951 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio, 952 struct nfs_page *req) 953 { 954 if (!pgio->pg_lseg) { 955 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, 956 req->wb_context, 957 0, 958 NFS4_MAX_UINT64, 959 IOMODE_RW, 960 false, 961 GFP_NOFS); 962 if (IS_ERR(pgio->pg_lseg)) { 963 pgio->pg_error = PTR_ERR(pgio->pg_lseg); 964 pgio->pg_lseg = NULL; 965 goto out; 966 } 967 } 968 if (pgio->pg_lseg) 969 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg); 970 971 /* no lseg means that pnfs is not in use, so no mirroring here */ 972 nfs_pageio_reset_write_mds(pgio); 973 out: 974 return 1; 975 } 976 977 static const struct nfs_pageio_ops ff_layout_pg_read_ops = { 978 .pg_init = ff_layout_pg_init_read, 979 .pg_test = pnfs_generic_pg_test, 980 .pg_doio = pnfs_generic_pg_readpages, 981 .pg_cleanup = pnfs_generic_pg_cleanup, 982 }; 983 984 static const struct nfs_pageio_ops ff_layout_pg_write_ops = { 985 .pg_init = ff_layout_pg_init_write, 986 .pg_test = pnfs_generic_pg_test, 987 .pg_doio = pnfs_generic_pg_writepages, 988 .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write, 989 .pg_cleanup = pnfs_generic_pg_cleanup, 990 }; 991 992 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs) 993 { 994 struct rpc_task *task = &hdr->task; 995 996 pnfs_layoutcommit_inode(hdr->inode, false); 997 998 if (retry_pnfs) { 999 dprintk("%s Reset task %5u for i/o through pNFS " 1000 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__, 1001 hdr->task.tk_pid, 1002 hdr->inode->i_sb->s_id, 1003 (unsigned long long)NFS_FILEID(hdr->inode), 1004 hdr->args.count, 1005 (unsigned long long)hdr->args.offset); 1006 1007 hdr->completion_ops->reschedule_io(hdr); 1008 return; 1009 } 1010 1011 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { 1012 dprintk("%s Reset task %5u for i/o through MDS " 1013 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__, 1014 hdr->task.tk_pid, 1015 hdr->inode->i_sb->s_id, 1016 (unsigned long long)NFS_FILEID(hdr->inode), 1017 hdr->args.count, 1018 (unsigned long long)hdr->args.offset); 1019 1020 task->tk_status = pnfs_write_done_resend_to_mds(hdr); 1021 } 1022 } 1023 1024 static void ff_layout_reset_read(struct nfs_pgio_header *hdr) 1025 { 1026 struct rpc_task *task = &hdr->task; 1027 1028 pnfs_layoutcommit_inode(hdr->inode, false); 1029 1030 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { 1031 dprintk("%s Reset task %5u for i/o through MDS " 1032 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__, 1033 hdr->task.tk_pid, 1034 hdr->inode->i_sb->s_id, 1035 (unsigned long long)NFS_FILEID(hdr->inode), 1036 hdr->args.count, 1037 (unsigned long long)hdr->args.offset); 1038 1039 task->tk_status = pnfs_read_done_resend_to_mds(hdr); 1040 } 1041 } 1042 1043 static int ff_layout_async_handle_error_v4(struct rpc_task *task, 1044 struct nfs4_state *state, 1045 struct nfs_client *clp, 1046 struct pnfs_layout_segment *lseg, 1047 int idx) 1048 { 1049 struct pnfs_layout_hdr *lo = lseg->pls_layout; 1050 struct inode *inode = lo->plh_inode; 1051 struct nfs_server *mds_server = NFS_SERVER(inode); 1052 1053 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx); 1054 struct nfs_client *mds_client = mds_server->nfs_client; 1055 struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table; 1056 1057 if (task->tk_status >= 0) 1058 return 0; 1059 1060 switch (task->tk_status) { 1061 /* MDS state errors */ 1062 case -NFS4ERR_DELEG_REVOKED: 1063 case -NFS4ERR_ADMIN_REVOKED: 1064 case -NFS4ERR_BAD_STATEID: 1065 if (state == NULL) 1066 break; 1067 nfs_remove_bad_delegation(state->inode, NULL); 1068 case -NFS4ERR_OPENMODE: 1069 if (state == NULL) 1070 break; 1071 if (nfs4_schedule_stateid_recovery(mds_server, state) < 0) 1072 goto out_bad_stateid; 1073 goto wait_on_recovery; 1074 case -NFS4ERR_EXPIRED: 1075 if (state != NULL) { 1076 if (nfs4_schedule_stateid_recovery(mds_server, state) < 0) 1077 goto out_bad_stateid; 1078 } 1079 nfs4_schedule_lease_recovery(mds_client); 1080 goto wait_on_recovery; 1081 /* DS session errors */ 1082 case -NFS4ERR_BADSESSION: 1083 case -NFS4ERR_BADSLOT: 1084 case -NFS4ERR_BAD_HIGH_SLOT: 1085 case -NFS4ERR_DEADSESSION: 1086 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION: 1087 case -NFS4ERR_SEQ_FALSE_RETRY: 1088 case -NFS4ERR_SEQ_MISORDERED: 1089 dprintk("%s ERROR %d, Reset session. Exchangeid " 1090 "flags 0x%x\n", __func__, task->tk_status, 1091 clp->cl_exchange_flags); 1092 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status); 1093 break; 1094 case -NFS4ERR_DELAY: 1095 case -NFS4ERR_GRACE: 1096 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX); 1097 break; 1098 case -NFS4ERR_RETRY_UNCACHED_REP: 1099 break; 1100 /* Invalidate Layout errors */ 1101 case -NFS4ERR_PNFS_NO_LAYOUT: 1102 case -ESTALE: /* mapped NFS4ERR_STALE */ 1103 case -EBADHANDLE: /* mapped NFS4ERR_BADHANDLE */ 1104 case -EISDIR: /* mapped NFS4ERR_ISDIR */ 1105 case -NFS4ERR_FHEXPIRED: 1106 case -NFS4ERR_WRONG_TYPE: 1107 dprintk("%s Invalid layout error %d\n", __func__, 1108 task->tk_status); 1109 /* 1110 * Destroy layout so new i/o will get a new layout. 1111 * Layout will not be destroyed until all current lseg 1112 * references are put. Mark layout as invalid to resend failed 1113 * i/o and all i/o waiting on the slot table to the MDS until 1114 * layout is destroyed and a new valid layout is obtained. 1115 */ 1116 pnfs_destroy_layout(NFS_I(inode)); 1117 rpc_wake_up(&tbl->slot_tbl_waitq); 1118 goto reset; 1119 /* RPC connection errors */ 1120 case -ECONNREFUSED: 1121 case -EHOSTDOWN: 1122 case -EHOSTUNREACH: 1123 case -ENETUNREACH: 1124 case -EIO: 1125 case -ETIMEDOUT: 1126 case -EPIPE: 1127 dprintk("%s DS connection error %d\n", __func__, 1128 task->tk_status); 1129 nfs4_mark_deviceid_unavailable(devid); 1130 rpc_wake_up(&tbl->slot_tbl_waitq); 1131 /* fall through */ 1132 default: 1133 if (ff_layout_avoid_mds_available_ds(lseg)) 1134 return -NFS4ERR_RESET_TO_PNFS; 1135 reset: 1136 dprintk("%s Retry through MDS. Error %d\n", __func__, 1137 task->tk_status); 1138 return -NFS4ERR_RESET_TO_MDS; 1139 } 1140 out: 1141 task->tk_status = 0; 1142 return -EAGAIN; 1143 out_bad_stateid: 1144 task->tk_status = -EIO; 1145 return 0; 1146 wait_on_recovery: 1147 rpc_sleep_on(&mds_client->cl_rpcwaitq, task, NULL); 1148 if (test_bit(NFS4CLNT_MANAGER_RUNNING, &mds_client->cl_state) == 0) 1149 rpc_wake_up_queued_task(&mds_client->cl_rpcwaitq, task); 1150 goto out; 1151 } 1152 1153 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */ 1154 static int ff_layout_async_handle_error_v3(struct rpc_task *task, 1155 struct pnfs_layout_segment *lseg, 1156 int idx) 1157 { 1158 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx); 1159 1160 if (task->tk_status >= 0) 1161 return 0; 1162 1163 switch (task->tk_status) { 1164 /* File access problems. Don't mark the device as unavailable */ 1165 case -EACCES: 1166 case -ESTALE: 1167 case -EISDIR: 1168 case -EBADHANDLE: 1169 case -ELOOP: 1170 case -ENOSPC: 1171 break; 1172 case -EJUKEBOX: 1173 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY); 1174 goto out_retry; 1175 default: 1176 dprintk("%s DS connection error %d\n", __func__, 1177 task->tk_status); 1178 nfs4_mark_deviceid_unavailable(devid); 1179 } 1180 /* FIXME: Need to prevent infinite looping here. */ 1181 return -NFS4ERR_RESET_TO_PNFS; 1182 out_retry: 1183 task->tk_status = 0; 1184 rpc_restart_call_prepare(task); 1185 rpc_delay(task, NFS_JUKEBOX_RETRY_TIME); 1186 return -EAGAIN; 1187 } 1188 1189 static int ff_layout_async_handle_error(struct rpc_task *task, 1190 struct nfs4_state *state, 1191 struct nfs_client *clp, 1192 struct pnfs_layout_segment *lseg, 1193 int idx) 1194 { 1195 int vers = clp->cl_nfs_mod->rpc_vers->number; 1196 1197 switch (vers) { 1198 case 3: 1199 return ff_layout_async_handle_error_v3(task, lseg, idx); 1200 case 4: 1201 return ff_layout_async_handle_error_v4(task, state, clp, 1202 lseg, idx); 1203 default: 1204 /* should never happen */ 1205 WARN_ON_ONCE(1); 1206 return 0; 1207 } 1208 } 1209 1210 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg, 1211 int idx, u64 offset, u64 length, 1212 u32 status, int opnum, int error) 1213 { 1214 struct nfs4_ff_layout_mirror *mirror; 1215 int err; 1216 1217 if (status == 0) { 1218 switch (error) { 1219 case -ETIMEDOUT: 1220 case -EPFNOSUPPORT: 1221 case -EPROTONOSUPPORT: 1222 case -EOPNOTSUPP: 1223 case -ECONNREFUSED: 1224 case -ECONNRESET: 1225 case -EHOSTDOWN: 1226 case -EHOSTUNREACH: 1227 case -ENETUNREACH: 1228 case -EADDRINUSE: 1229 case -ENOBUFS: 1230 case -EPIPE: 1231 case -EPERM: 1232 status = NFS4ERR_NXIO; 1233 break; 1234 case -EACCES: 1235 status = NFS4ERR_ACCESS; 1236 break; 1237 default: 1238 return; 1239 } 1240 } 1241 1242 switch (status) { 1243 case NFS4ERR_DELAY: 1244 case NFS4ERR_GRACE: 1245 return; 1246 default: 1247 break; 1248 } 1249 1250 mirror = FF_LAYOUT_COMP(lseg, idx); 1251 err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout), 1252 mirror, offset, length, status, opnum, 1253 GFP_NOIO); 1254 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode, lseg); 1255 dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status); 1256 } 1257 1258 /* NFS_PROTO call done callback routines */ 1259 static int ff_layout_read_done_cb(struct rpc_task *task, 1260 struct nfs_pgio_header *hdr) 1261 { 1262 int err; 1263 1264 trace_nfs4_pnfs_read(hdr, task->tk_status); 1265 if (task->tk_status < 0) 1266 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx, 1267 hdr->args.offset, hdr->args.count, 1268 hdr->res.op_status, OP_READ, 1269 task->tk_status); 1270 err = ff_layout_async_handle_error(task, hdr->args.context->state, 1271 hdr->ds_clp, hdr->lseg, 1272 hdr->pgio_mirror_idx); 1273 1274 switch (err) { 1275 case -NFS4ERR_RESET_TO_PNFS: 1276 if (ff_layout_choose_best_ds_for_read(hdr->lseg, 1277 hdr->pgio_mirror_idx + 1, 1278 &hdr->pgio_mirror_idx)) 1279 goto out_eagain; 1280 ff_layout_read_record_layoutstats_done(task, hdr); 1281 pnfs_read_resend_pnfs(hdr); 1282 return task->tk_status; 1283 case -NFS4ERR_RESET_TO_MDS: 1284 ff_layout_reset_read(hdr); 1285 return task->tk_status; 1286 case -EAGAIN: 1287 goto out_eagain; 1288 } 1289 1290 return 0; 1291 out_eagain: 1292 rpc_restart_call_prepare(task); 1293 return -EAGAIN; 1294 } 1295 1296 static bool 1297 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg) 1298 { 1299 return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT); 1300 } 1301 1302 /* 1303 * We reference the rpc_cred of the first WRITE that triggers the need for 1304 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound. 1305 * rfc5661 is not clear about which credential should be used. 1306 * 1307 * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so 1308 * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751 1309 * we always send layoutcommit after DS writes. 1310 */ 1311 static void 1312 ff_layout_set_layoutcommit(struct inode *inode, 1313 struct pnfs_layout_segment *lseg, 1314 loff_t end_offset) 1315 { 1316 if (!ff_layout_need_layoutcommit(lseg)) 1317 return; 1318 1319 pnfs_set_layoutcommit(inode, lseg, end_offset); 1320 dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino, 1321 (unsigned long long) NFS_I(inode)->layout->plh_lwb); 1322 } 1323 1324 static bool 1325 ff_layout_device_unavailable(struct pnfs_layout_segment *lseg, int idx) 1326 { 1327 /* No mirroring for now */ 1328 struct nfs4_deviceid_node *node = FF_LAYOUT_DEVID_NODE(lseg, idx); 1329 1330 return ff_layout_test_devid_unavailable(node); 1331 } 1332 1333 static void ff_layout_read_record_layoutstats_start(struct rpc_task *task, 1334 struct nfs_pgio_header *hdr) 1335 { 1336 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags)) 1337 return; 1338 nfs4_ff_layout_stat_io_start_read(hdr->inode, 1339 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx), 1340 hdr->args.count, 1341 task->tk_start); 1342 } 1343 1344 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task, 1345 struct nfs_pgio_header *hdr) 1346 { 1347 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags)) 1348 return; 1349 nfs4_ff_layout_stat_io_end_read(task, 1350 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx), 1351 hdr->args.count, 1352 hdr->res.count); 1353 } 1354 1355 static int ff_layout_read_prepare_common(struct rpc_task *task, 1356 struct nfs_pgio_header *hdr) 1357 { 1358 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) { 1359 rpc_exit(task, -EIO); 1360 return -EIO; 1361 } 1362 if (ff_layout_device_unavailable(hdr->lseg, hdr->pgio_mirror_idx)) { 1363 rpc_exit(task, -EHOSTDOWN); 1364 return -EAGAIN; 1365 } 1366 1367 ff_layout_read_record_layoutstats_start(task, hdr); 1368 return 0; 1369 } 1370 1371 /* 1372 * Call ops for the async read/write cases 1373 * In the case of dense layouts, the offset needs to be reset to its 1374 * original value. 1375 */ 1376 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data) 1377 { 1378 struct nfs_pgio_header *hdr = data; 1379 1380 if (ff_layout_read_prepare_common(task, hdr)) 1381 return; 1382 1383 rpc_call_start(task); 1384 } 1385 1386 static int ff_layout_setup_sequence(struct nfs_client *ds_clp, 1387 struct nfs4_sequence_args *args, 1388 struct nfs4_sequence_res *res, 1389 struct rpc_task *task) 1390 { 1391 if (ds_clp->cl_session) 1392 return nfs41_setup_sequence(ds_clp->cl_session, 1393 args, 1394 res, 1395 task); 1396 return nfs40_setup_sequence(ds_clp->cl_slot_tbl, 1397 args, 1398 res, 1399 task); 1400 } 1401 1402 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data) 1403 { 1404 struct nfs_pgio_header *hdr = data; 1405 1406 if (ff_layout_setup_sequence(hdr->ds_clp, 1407 &hdr->args.seq_args, 1408 &hdr->res.seq_res, 1409 task)) 1410 return; 1411 1412 if (ff_layout_read_prepare_common(task, hdr)) 1413 return; 1414 1415 if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context, 1416 hdr->args.lock_context, FMODE_READ) == -EIO) 1417 rpc_exit(task, -EIO); /* lost lock, terminate I/O */ 1418 } 1419 1420 static void ff_layout_read_call_done(struct rpc_task *task, void *data) 1421 { 1422 struct nfs_pgio_header *hdr = data; 1423 1424 dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status); 1425 1426 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) && 1427 task->tk_status == 0) { 1428 nfs4_sequence_done(task, &hdr->res.seq_res); 1429 return; 1430 } 1431 1432 /* Note this may cause RPC to be resent */ 1433 hdr->mds_ops->rpc_call_done(task, hdr); 1434 } 1435 1436 static void ff_layout_read_count_stats(struct rpc_task *task, void *data) 1437 { 1438 struct nfs_pgio_header *hdr = data; 1439 1440 ff_layout_read_record_layoutstats_done(task, hdr); 1441 rpc_count_iostats_metrics(task, 1442 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]); 1443 } 1444 1445 static void ff_layout_read_release(void *data) 1446 { 1447 struct nfs_pgio_header *hdr = data; 1448 1449 ff_layout_read_record_layoutstats_done(&hdr->task, hdr); 1450 pnfs_generic_rw_release(data); 1451 } 1452 1453 1454 static int ff_layout_write_done_cb(struct rpc_task *task, 1455 struct nfs_pgio_header *hdr) 1456 { 1457 loff_t end_offs = 0; 1458 int err; 1459 1460 trace_nfs4_pnfs_write(hdr, task->tk_status); 1461 if (task->tk_status < 0) 1462 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx, 1463 hdr->args.offset, hdr->args.count, 1464 hdr->res.op_status, OP_WRITE, 1465 task->tk_status); 1466 err = ff_layout_async_handle_error(task, hdr->args.context->state, 1467 hdr->ds_clp, hdr->lseg, 1468 hdr->pgio_mirror_idx); 1469 1470 switch (err) { 1471 case -NFS4ERR_RESET_TO_PNFS: 1472 ff_layout_reset_write(hdr, true); 1473 return task->tk_status; 1474 case -NFS4ERR_RESET_TO_MDS: 1475 ff_layout_reset_write(hdr, false); 1476 return task->tk_status; 1477 case -EAGAIN: 1478 return -EAGAIN; 1479 } 1480 1481 if (hdr->res.verf->committed == NFS_FILE_SYNC || 1482 hdr->res.verf->committed == NFS_DATA_SYNC) 1483 end_offs = hdr->mds_offset + (loff_t)hdr->res.count; 1484 1485 /* Note: if the write is unstable, don't set end_offs until commit */ 1486 ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs); 1487 1488 /* zero out fattr since we don't care DS attr at all */ 1489 hdr->fattr.valid = 0; 1490 if (task->tk_status >= 0) 1491 nfs_writeback_update_inode(hdr); 1492 1493 return 0; 1494 } 1495 1496 static int ff_layout_commit_done_cb(struct rpc_task *task, 1497 struct nfs_commit_data *data) 1498 { 1499 int err; 1500 1501 trace_nfs4_pnfs_commit_ds(data, task->tk_status); 1502 if (task->tk_status < 0) 1503 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index, 1504 data->args.offset, data->args.count, 1505 data->res.op_status, OP_COMMIT, 1506 task->tk_status); 1507 err = ff_layout_async_handle_error(task, NULL, data->ds_clp, 1508 data->lseg, data->ds_commit_index); 1509 1510 switch (err) { 1511 case -NFS4ERR_RESET_TO_PNFS: 1512 pnfs_generic_prepare_to_resend_writes(data); 1513 return -EAGAIN; 1514 case -NFS4ERR_RESET_TO_MDS: 1515 pnfs_generic_prepare_to_resend_writes(data); 1516 return -EAGAIN; 1517 case -EAGAIN: 1518 rpc_restart_call_prepare(task); 1519 return -EAGAIN; 1520 } 1521 1522 ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb); 1523 1524 return 0; 1525 } 1526 1527 static void ff_layout_write_record_layoutstats_start(struct rpc_task *task, 1528 struct nfs_pgio_header *hdr) 1529 { 1530 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags)) 1531 return; 1532 nfs4_ff_layout_stat_io_start_write(hdr->inode, 1533 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx), 1534 hdr->args.count, 1535 task->tk_start); 1536 } 1537 1538 static void ff_layout_write_record_layoutstats_done(struct rpc_task *task, 1539 struct nfs_pgio_header *hdr) 1540 { 1541 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags)) 1542 return; 1543 nfs4_ff_layout_stat_io_end_write(task, 1544 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx), 1545 hdr->args.count, hdr->res.count, 1546 hdr->res.verf->committed); 1547 } 1548 1549 static int ff_layout_write_prepare_common(struct rpc_task *task, 1550 struct nfs_pgio_header *hdr) 1551 { 1552 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) { 1553 rpc_exit(task, -EIO); 1554 return -EIO; 1555 } 1556 1557 if (ff_layout_device_unavailable(hdr->lseg, hdr->pgio_mirror_idx)) { 1558 rpc_exit(task, -EHOSTDOWN); 1559 return -EAGAIN; 1560 } 1561 1562 ff_layout_write_record_layoutstats_start(task, hdr); 1563 return 0; 1564 } 1565 1566 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data) 1567 { 1568 struct nfs_pgio_header *hdr = data; 1569 1570 if (ff_layout_write_prepare_common(task, hdr)) 1571 return; 1572 1573 rpc_call_start(task); 1574 } 1575 1576 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data) 1577 { 1578 struct nfs_pgio_header *hdr = data; 1579 1580 if (ff_layout_setup_sequence(hdr->ds_clp, 1581 &hdr->args.seq_args, 1582 &hdr->res.seq_res, 1583 task)) 1584 return; 1585 1586 if (ff_layout_write_prepare_common(task, hdr)) 1587 return; 1588 1589 if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context, 1590 hdr->args.lock_context, FMODE_WRITE) == -EIO) 1591 rpc_exit(task, -EIO); /* lost lock, terminate I/O */ 1592 } 1593 1594 static void ff_layout_write_call_done(struct rpc_task *task, void *data) 1595 { 1596 struct nfs_pgio_header *hdr = data; 1597 1598 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) && 1599 task->tk_status == 0) { 1600 nfs4_sequence_done(task, &hdr->res.seq_res); 1601 return; 1602 } 1603 1604 /* Note this may cause RPC to be resent */ 1605 hdr->mds_ops->rpc_call_done(task, hdr); 1606 } 1607 1608 static void ff_layout_write_count_stats(struct rpc_task *task, void *data) 1609 { 1610 struct nfs_pgio_header *hdr = data; 1611 1612 ff_layout_write_record_layoutstats_done(task, hdr); 1613 rpc_count_iostats_metrics(task, 1614 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]); 1615 } 1616 1617 static void ff_layout_write_release(void *data) 1618 { 1619 struct nfs_pgio_header *hdr = data; 1620 1621 ff_layout_write_record_layoutstats_done(&hdr->task, hdr); 1622 pnfs_generic_rw_release(data); 1623 } 1624 1625 static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task, 1626 struct nfs_commit_data *cdata) 1627 { 1628 if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags)) 1629 return; 1630 nfs4_ff_layout_stat_io_start_write(cdata->inode, 1631 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index), 1632 0, task->tk_start); 1633 } 1634 1635 static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task, 1636 struct nfs_commit_data *cdata) 1637 { 1638 struct nfs_page *req; 1639 __u64 count = 0; 1640 1641 if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags)) 1642 return; 1643 1644 if (task->tk_status == 0) { 1645 list_for_each_entry(req, &cdata->pages, wb_list) 1646 count += req->wb_bytes; 1647 } 1648 nfs4_ff_layout_stat_io_end_write(task, 1649 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index), 1650 count, count, NFS_FILE_SYNC); 1651 } 1652 1653 static void ff_layout_commit_prepare_common(struct rpc_task *task, 1654 struct nfs_commit_data *cdata) 1655 { 1656 ff_layout_commit_record_layoutstats_start(task, cdata); 1657 } 1658 1659 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data) 1660 { 1661 ff_layout_commit_prepare_common(task, data); 1662 rpc_call_start(task); 1663 } 1664 1665 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data) 1666 { 1667 struct nfs_commit_data *wdata = data; 1668 1669 if (ff_layout_setup_sequence(wdata->ds_clp, 1670 &wdata->args.seq_args, 1671 &wdata->res.seq_res, 1672 task)) 1673 return; 1674 ff_layout_commit_prepare_common(task, data); 1675 } 1676 1677 static void ff_layout_commit_done(struct rpc_task *task, void *data) 1678 { 1679 pnfs_generic_write_commit_done(task, data); 1680 } 1681 1682 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data) 1683 { 1684 struct nfs_commit_data *cdata = data; 1685 1686 ff_layout_commit_record_layoutstats_done(task, cdata); 1687 rpc_count_iostats_metrics(task, 1688 &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]); 1689 } 1690 1691 static void ff_layout_commit_release(void *data) 1692 { 1693 struct nfs_commit_data *cdata = data; 1694 1695 ff_layout_commit_record_layoutstats_done(&cdata->task, cdata); 1696 pnfs_generic_commit_release(data); 1697 } 1698 1699 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = { 1700 .rpc_call_prepare = ff_layout_read_prepare_v3, 1701 .rpc_call_done = ff_layout_read_call_done, 1702 .rpc_count_stats = ff_layout_read_count_stats, 1703 .rpc_release = ff_layout_read_release, 1704 }; 1705 1706 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = { 1707 .rpc_call_prepare = ff_layout_read_prepare_v4, 1708 .rpc_call_done = ff_layout_read_call_done, 1709 .rpc_count_stats = ff_layout_read_count_stats, 1710 .rpc_release = ff_layout_read_release, 1711 }; 1712 1713 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = { 1714 .rpc_call_prepare = ff_layout_write_prepare_v3, 1715 .rpc_call_done = ff_layout_write_call_done, 1716 .rpc_count_stats = ff_layout_write_count_stats, 1717 .rpc_release = ff_layout_write_release, 1718 }; 1719 1720 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = { 1721 .rpc_call_prepare = ff_layout_write_prepare_v4, 1722 .rpc_call_done = ff_layout_write_call_done, 1723 .rpc_count_stats = ff_layout_write_count_stats, 1724 .rpc_release = ff_layout_write_release, 1725 }; 1726 1727 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = { 1728 .rpc_call_prepare = ff_layout_commit_prepare_v3, 1729 .rpc_call_done = ff_layout_commit_done, 1730 .rpc_count_stats = ff_layout_commit_count_stats, 1731 .rpc_release = ff_layout_commit_release, 1732 }; 1733 1734 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = { 1735 .rpc_call_prepare = ff_layout_commit_prepare_v4, 1736 .rpc_call_done = ff_layout_commit_done, 1737 .rpc_count_stats = ff_layout_commit_count_stats, 1738 .rpc_release = ff_layout_commit_release, 1739 }; 1740 1741 static enum pnfs_try_status 1742 ff_layout_read_pagelist(struct nfs_pgio_header *hdr) 1743 { 1744 struct pnfs_layout_segment *lseg = hdr->lseg; 1745 struct nfs4_pnfs_ds *ds; 1746 struct rpc_clnt *ds_clnt; 1747 struct rpc_cred *ds_cred; 1748 loff_t offset = hdr->args.offset; 1749 u32 idx = hdr->pgio_mirror_idx; 1750 int vers; 1751 struct nfs_fh *fh; 1752 1753 dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n", 1754 __func__, hdr->inode->i_ino, 1755 hdr->args.pgbase, (size_t)hdr->args.count, offset); 1756 1757 ds = nfs4_ff_layout_prepare_ds(lseg, idx, false); 1758 if (!ds) 1759 goto out_failed; 1760 1761 ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp, 1762 hdr->inode); 1763 if (IS_ERR(ds_clnt)) 1764 goto out_failed; 1765 1766 ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred); 1767 if (!ds_cred) 1768 goto out_failed; 1769 1770 vers = nfs4_ff_layout_ds_version(lseg, idx); 1771 1772 dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__, 1773 ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count), vers); 1774 1775 hdr->pgio_done_cb = ff_layout_read_done_cb; 1776 atomic_inc(&ds->ds_clp->cl_count); 1777 hdr->ds_clp = ds->ds_clp; 1778 fh = nfs4_ff_layout_select_ds_fh(lseg, idx); 1779 if (fh) 1780 hdr->args.fh = fh; 1781 /* 1782 * Note that if we ever decide to split across DSes, 1783 * then we may need to handle dense-like offsets. 1784 */ 1785 hdr->args.offset = offset; 1786 hdr->mds_offset = offset; 1787 1788 /* Perform an asynchronous read to ds */ 1789 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops, 1790 vers == 3 ? &ff_layout_read_call_ops_v3 : 1791 &ff_layout_read_call_ops_v4, 1792 0, RPC_TASK_SOFTCONN); 1793 put_rpccred(ds_cred); 1794 return PNFS_ATTEMPTED; 1795 1796 out_failed: 1797 if (ff_layout_avoid_mds_available_ds(lseg)) 1798 return PNFS_TRY_AGAIN; 1799 return PNFS_NOT_ATTEMPTED; 1800 } 1801 1802 /* Perform async writes. */ 1803 static enum pnfs_try_status 1804 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync) 1805 { 1806 struct pnfs_layout_segment *lseg = hdr->lseg; 1807 struct nfs4_pnfs_ds *ds; 1808 struct rpc_clnt *ds_clnt; 1809 struct rpc_cred *ds_cred; 1810 loff_t offset = hdr->args.offset; 1811 int vers; 1812 struct nfs_fh *fh; 1813 int idx = hdr->pgio_mirror_idx; 1814 1815 ds = nfs4_ff_layout_prepare_ds(lseg, idx, true); 1816 if (!ds) 1817 return PNFS_NOT_ATTEMPTED; 1818 1819 ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp, 1820 hdr->inode); 1821 if (IS_ERR(ds_clnt)) 1822 return PNFS_NOT_ATTEMPTED; 1823 1824 ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred); 1825 if (!ds_cred) 1826 return PNFS_NOT_ATTEMPTED; 1827 1828 vers = nfs4_ff_layout_ds_version(lseg, idx); 1829 1830 dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s cl_count %d vers %d\n", 1831 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count, 1832 offset, ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count), 1833 vers); 1834 1835 hdr->pgio_done_cb = ff_layout_write_done_cb; 1836 atomic_inc(&ds->ds_clp->cl_count); 1837 hdr->ds_clp = ds->ds_clp; 1838 hdr->ds_commit_idx = idx; 1839 fh = nfs4_ff_layout_select_ds_fh(lseg, idx); 1840 if (fh) 1841 hdr->args.fh = fh; 1842 1843 /* 1844 * Note that if we ever decide to split across DSes, 1845 * then we may need to handle dense-like offsets. 1846 */ 1847 hdr->args.offset = offset; 1848 1849 /* Perform an asynchronous write */ 1850 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops, 1851 vers == 3 ? &ff_layout_write_call_ops_v3 : 1852 &ff_layout_write_call_ops_v4, 1853 sync, RPC_TASK_SOFTCONN); 1854 put_rpccred(ds_cred); 1855 return PNFS_ATTEMPTED; 1856 } 1857 1858 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i) 1859 { 1860 return i; 1861 } 1862 1863 static struct nfs_fh * 1864 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i) 1865 { 1866 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg); 1867 1868 /* FIXME: Assume that there is only one NFS version available 1869 * for the DS. 1870 */ 1871 return &flseg->mirror_array[i]->fh_versions[0]; 1872 } 1873 1874 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how) 1875 { 1876 struct pnfs_layout_segment *lseg = data->lseg; 1877 struct nfs4_pnfs_ds *ds; 1878 struct rpc_clnt *ds_clnt; 1879 struct rpc_cred *ds_cred; 1880 u32 idx; 1881 int vers, ret; 1882 struct nfs_fh *fh; 1883 1884 idx = calc_ds_index_from_commit(lseg, data->ds_commit_index); 1885 ds = nfs4_ff_layout_prepare_ds(lseg, idx, true); 1886 if (!ds) 1887 goto out_err; 1888 1889 ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp, 1890 data->inode); 1891 if (IS_ERR(ds_clnt)) 1892 goto out_err; 1893 1894 ds_cred = ff_layout_get_ds_cred(lseg, idx, data->cred); 1895 if (!ds_cred) 1896 goto out_err; 1897 1898 vers = nfs4_ff_layout_ds_version(lseg, idx); 1899 1900 dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__, 1901 data->inode->i_ino, how, atomic_read(&ds->ds_clp->cl_count), 1902 vers); 1903 data->commit_done_cb = ff_layout_commit_done_cb; 1904 data->cred = ds_cred; 1905 atomic_inc(&ds->ds_clp->cl_count); 1906 data->ds_clp = ds->ds_clp; 1907 fh = select_ds_fh_from_commit(lseg, data->ds_commit_index); 1908 if (fh) 1909 data->args.fh = fh; 1910 1911 ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops, 1912 vers == 3 ? &ff_layout_commit_call_ops_v3 : 1913 &ff_layout_commit_call_ops_v4, 1914 how, RPC_TASK_SOFTCONN); 1915 put_rpccred(ds_cred); 1916 return ret; 1917 out_err: 1918 pnfs_generic_prepare_to_resend_writes(data); 1919 pnfs_generic_commit_release(data); 1920 return -EAGAIN; 1921 } 1922 1923 static int 1924 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages, 1925 int how, struct nfs_commit_info *cinfo) 1926 { 1927 return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo, 1928 ff_layout_initiate_commit); 1929 } 1930 1931 static struct pnfs_ds_commit_info * 1932 ff_layout_get_ds_info(struct inode *inode) 1933 { 1934 struct pnfs_layout_hdr *layout = NFS_I(inode)->layout; 1935 1936 if (layout == NULL) 1937 return NULL; 1938 1939 return &FF_LAYOUT_FROM_HDR(layout)->commit_info; 1940 } 1941 1942 static void 1943 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d) 1944 { 1945 nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds, 1946 id_node)); 1947 } 1948 1949 static int ff_layout_encode_ioerr(struct xdr_stream *xdr, 1950 const struct nfs4_layoutreturn_args *args, 1951 const struct nfs4_flexfile_layoutreturn_args *ff_args) 1952 { 1953 __be32 *start; 1954 1955 start = xdr_reserve_space(xdr, 4); 1956 if (unlikely(!start)) 1957 return -E2BIG; 1958 1959 *start = cpu_to_be32(ff_args->num_errors); 1960 /* This assume we always return _ALL_ layouts */ 1961 return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors); 1962 } 1963 1964 static void 1965 encode_opaque_fixed(struct xdr_stream *xdr, const void *buf, size_t len) 1966 { 1967 __be32 *p; 1968 1969 p = xdr_reserve_space(xdr, len); 1970 xdr_encode_opaque_fixed(p, buf, len); 1971 } 1972 1973 static void 1974 ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr, 1975 const nfs4_stateid *stateid, 1976 const struct nfs42_layoutstat_devinfo *devinfo) 1977 { 1978 __be32 *p; 1979 1980 p = xdr_reserve_space(xdr, 8 + 8); 1981 p = xdr_encode_hyper(p, devinfo->offset); 1982 p = xdr_encode_hyper(p, devinfo->length); 1983 encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE); 1984 p = xdr_reserve_space(xdr, 4*8); 1985 p = xdr_encode_hyper(p, devinfo->read_count); 1986 p = xdr_encode_hyper(p, devinfo->read_bytes); 1987 p = xdr_encode_hyper(p, devinfo->write_count); 1988 p = xdr_encode_hyper(p, devinfo->write_bytes); 1989 encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE); 1990 } 1991 1992 static void 1993 ff_layout_encode_ff_iostat(struct xdr_stream *xdr, 1994 const nfs4_stateid *stateid, 1995 const struct nfs42_layoutstat_devinfo *devinfo) 1996 { 1997 ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo); 1998 ff_layout_encode_ff_layoutupdate(xdr, devinfo, 1999 devinfo->ld_private.data); 2000 } 2001 2002 /* report nothing for now */ 2003 static void ff_layout_encode_iostats_array(struct xdr_stream *xdr, 2004 const struct nfs4_layoutreturn_args *args, 2005 struct nfs4_flexfile_layoutreturn_args *ff_args) 2006 { 2007 __be32 *p; 2008 int i; 2009 2010 p = xdr_reserve_space(xdr, 4); 2011 *p = cpu_to_be32(ff_args->num_dev); 2012 for (i = 0; i < ff_args->num_dev; i++) 2013 ff_layout_encode_ff_iostat(xdr, 2014 &args->layout->plh_stateid, 2015 &ff_args->devinfo[i]); 2016 } 2017 2018 static void 2019 ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo, 2020 unsigned int num_entries) 2021 { 2022 unsigned int i; 2023 2024 for (i = 0; i < num_entries; i++) { 2025 if (!devinfo[i].ld_private.ops) 2026 continue; 2027 if (!devinfo[i].ld_private.ops->free) 2028 continue; 2029 devinfo[i].ld_private.ops->free(&devinfo[i].ld_private); 2030 } 2031 } 2032 2033 static struct nfs4_deviceid_node * 2034 ff_layout_alloc_deviceid_node(struct nfs_server *server, 2035 struct pnfs_device *pdev, gfp_t gfp_flags) 2036 { 2037 struct nfs4_ff_layout_ds *dsaddr; 2038 2039 dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags); 2040 if (!dsaddr) 2041 return NULL; 2042 return &dsaddr->id_node; 2043 } 2044 2045 static void 2046 ff_layout_encode_layoutreturn(struct xdr_stream *xdr, 2047 const void *voidargs, 2048 const struct nfs4_xdr_opaque_data *ff_opaque) 2049 { 2050 const struct nfs4_layoutreturn_args *args = voidargs; 2051 struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data; 2052 struct xdr_buf tmp_buf = { 2053 .head = { 2054 [0] = { 2055 .iov_base = page_address(ff_args->pages[0]), 2056 }, 2057 }, 2058 .buflen = PAGE_SIZE, 2059 }; 2060 struct xdr_stream tmp_xdr; 2061 __be32 *start; 2062 2063 dprintk("%s: Begin\n", __func__); 2064 2065 xdr_init_encode(&tmp_xdr, &tmp_buf, NULL); 2066 2067 ff_layout_encode_ioerr(&tmp_xdr, args, ff_args); 2068 ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args); 2069 2070 start = xdr_reserve_space(xdr, 4); 2071 *start = cpu_to_be32(tmp_buf.len); 2072 xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len); 2073 2074 dprintk("%s: Return\n", __func__); 2075 } 2076 2077 static void 2078 ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args) 2079 { 2080 struct nfs4_flexfile_layoutreturn_args *ff_args; 2081 2082 if (!args->data) 2083 return; 2084 ff_args = args->data; 2085 args->data = NULL; 2086 2087 ff_layout_free_ds_ioerr(&ff_args->errors); 2088 ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev); 2089 2090 put_page(ff_args->pages[0]); 2091 kfree(ff_args); 2092 } 2093 2094 const struct nfs4_xdr_opaque_ops layoutreturn_ops = { 2095 .encode = ff_layout_encode_layoutreturn, 2096 .free = ff_layout_free_layoutreturn, 2097 }; 2098 2099 static int 2100 ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args) 2101 { 2102 struct nfs4_flexfile_layoutreturn_args *ff_args; 2103 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout); 2104 2105 ff_args = kmalloc(sizeof(*ff_args), GFP_KERNEL); 2106 if (!ff_args) 2107 goto out_nomem; 2108 ff_args->pages[0] = alloc_page(GFP_KERNEL); 2109 if (!ff_args->pages[0]) 2110 goto out_nomem_free; 2111 2112 INIT_LIST_HEAD(&ff_args->errors); 2113 ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout, 2114 &args->range, &ff_args->errors, 2115 FF_LAYOUTRETURN_MAXERR); 2116 2117 spin_lock(&args->inode->i_lock); 2118 ff_args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr, 2119 &ff_args->devinfo[0], ARRAY_SIZE(ff_args->devinfo)); 2120 spin_unlock(&args->inode->i_lock); 2121 2122 args->ld_private->ops = &layoutreturn_ops; 2123 args->ld_private->data = ff_args; 2124 return 0; 2125 out_nomem_free: 2126 kfree(ff_args); 2127 out_nomem: 2128 return -ENOMEM; 2129 } 2130 2131 static int 2132 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen) 2133 { 2134 const struct sockaddr_in *sin = (struct sockaddr_in *)sap; 2135 2136 return snprintf(buf, buflen, "%pI4", &sin->sin_addr); 2137 } 2138 2139 static size_t 2140 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf, 2141 const int buflen) 2142 { 2143 const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap; 2144 const struct in6_addr *addr = &sin6->sin6_addr; 2145 2146 /* 2147 * RFC 4291, Section 2.2.2 2148 * 2149 * Shorthanded ANY address 2150 */ 2151 if (ipv6_addr_any(addr)) 2152 return snprintf(buf, buflen, "::"); 2153 2154 /* 2155 * RFC 4291, Section 2.2.2 2156 * 2157 * Shorthanded loopback address 2158 */ 2159 if (ipv6_addr_loopback(addr)) 2160 return snprintf(buf, buflen, "::1"); 2161 2162 /* 2163 * RFC 4291, Section 2.2.3 2164 * 2165 * Special presentation address format for mapped v4 2166 * addresses. 2167 */ 2168 if (ipv6_addr_v4mapped(addr)) 2169 return snprintf(buf, buflen, "::ffff:%pI4", 2170 &addr->s6_addr32[3]); 2171 2172 /* 2173 * RFC 4291, Section 2.2.1 2174 */ 2175 return snprintf(buf, buflen, "%pI6c", addr); 2176 } 2177 2178 /* Derived from rpc_sockaddr2uaddr */ 2179 static void 2180 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da) 2181 { 2182 struct sockaddr *sap = (struct sockaddr *)&da->da_addr; 2183 char portbuf[RPCBIND_MAXUADDRPLEN]; 2184 char addrbuf[RPCBIND_MAXUADDRLEN]; 2185 char *netid; 2186 unsigned short port; 2187 int len, netid_len; 2188 __be32 *p; 2189 2190 switch (sap->sa_family) { 2191 case AF_INET: 2192 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0) 2193 return; 2194 port = ntohs(((struct sockaddr_in *)sap)->sin_port); 2195 netid = "tcp"; 2196 netid_len = 3; 2197 break; 2198 case AF_INET6: 2199 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0) 2200 return; 2201 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port); 2202 netid = "tcp6"; 2203 netid_len = 4; 2204 break; 2205 default: 2206 /* we only support tcp and tcp6 */ 2207 WARN_ON_ONCE(1); 2208 return; 2209 } 2210 2211 snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff); 2212 len = strlcat(addrbuf, portbuf, sizeof(addrbuf)); 2213 2214 p = xdr_reserve_space(xdr, 4 + netid_len); 2215 xdr_encode_opaque(p, netid, netid_len); 2216 2217 p = xdr_reserve_space(xdr, 4 + len); 2218 xdr_encode_opaque(p, addrbuf, len); 2219 } 2220 2221 static void 2222 ff_layout_encode_nfstime(struct xdr_stream *xdr, 2223 ktime_t t) 2224 { 2225 struct timespec64 ts; 2226 __be32 *p; 2227 2228 p = xdr_reserve_space(xdr, 12); 2229 ts = ktime_to_timespec64(t); 2230 p = xdr_encode_hyper(p, ts.tv_sec); 2231 *p++ = cpu_to_be32(ts.tv_nsec); 2232 } 2233 2234 static void 2235 ff_layout_encode_io_latency(struct xdr_stream *xdr, 2236 struct nfs4_ff_io_stat *stat) 2237 { 2238 __be32 *p; 2239 2240 p = xdr_reserve_space(xdr, 5 * 8); 2241 p = xdr_encode_hyper(p, stat->ops_requested); 2242 p = xdr_encode_hyper(p, stat->bytes_requested); 2243 p = xdr_encode_hyper(p, stat->ops_completed); 2244 p = xdr_encode_hyper(p, stat->bytes_completed); 2245 p = xdr_encode_hyper(p, stat->bytes_not_delivered); 2246 ff_layout_encode_nfstime(xdr, stat->total_busy_time); 2247 ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time); 2248 } 2249 2250 static void 2251 ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr, 2252 const struct nfs42_layoutstat_devinfo *devinfo, 2253 struct nfs4_ff_layout_mirror *mirror) 2254 { 2255 struct nfs4_pnfs_ds_addr *da; 2256 struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds; 2257 struct nfs_fh *fh = &mirror->fh_versions[0]; 2258 __be32 *p; 2259 2260 da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node); 2261 dprintk("%s: DS %s: encoding address %s\n", 2262 __func__, ds->ds_remotestr, da->da_remotestr); 2263 /* netaddr4 */ 2264 ff_layout_encode_netaddr(xdr, da); 2265 /* nfs_fh4 */ 2266 p = xdr_reserve_space(xdr, 4 + fh->size); 2267 xdr_encode_opaque(p, fh->data, fh->size); 2268 /* ff_io_latency4 read */ 2269 spin_lock(&mirror->lock); 2270 ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat); 2271 /* ff_io_latency4 write */ 2272 ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat); 2273 spin_unlock(&mirror->lock); 2274 /* nfstime4 */ 2275 ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time)); 2276 /* bool */ 2277 p = xdr_reserve_space(xdr, 4); 2278 *p = cpu_to_be32(false); 2279 } 2280 2281 static void 2282 ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args, 2283 const struct nfs4_xdr_opaque_data *opaque) 2284 { 2285 struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque, 2286 struct nfs42_layoutstat_devinfo, ld_private); 2287 __be32 *start; 2288 2289 /* layoutupdate length */ 2290 start = xdr_reserve_space(xdr, 4); 2291 ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data); 2292 2293 *start = cpu_to_be32((xdr->p - start - 1) * 4); 2294 } 2295 2296 static void 2297 ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque) 2298 { 2299 struct nfs4_ff_layout_mirror *mirror = opaque->data; 2300 2301 ff_layout_put_mirror(mirror); 2302 } 2303 2304 static const struct nfs4_xdr_opaque_ops layoutstat_ops = { 2305 .encode = ff_layout_encode_layoutstats, 2306 .free = ff_layout_free_layoutstats, 2307 }; 2308 2309 static int 2310 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo, 2311 struct nfs42_layoutstat_devinfo *devinfo, 2312 int dev_limit) 2313 { 2314 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo); 2315 struct nfs4_ff_layout_mirror *mirror; 2316 struct nfs4_deviceid_node *dev; 2317 int i = 0; 2318 2319 list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) { 2320 if (i >= dev_limit) 2321 break; 2322 if (IS_ERR_OR_NULL(mirror->mirror_ds)) 2323 continue; 2324 if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags)) 2325 continue; 2326 /* mirror refcount put in cleanup_layoutstats */ 2327 if (!atomic_inc_not_zero(&mirror->ref)) 2328 continue; 2329 dev = &mirror->mirror_ds->id_node; 2330 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE); 2331 devinfo->offset = 0; 2332 devinfo->length = NFS4_MAX_UINT64; 2333 spin_lock(&mirror->lock); 2334 devinfo->read_count = mirror->read_stat.io_stat.ops_completed; 2335 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed; 2336 devinfo->write_count = mirror->write_stat.io_stat.ops_completed; 2337 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed; 2338 spin_unlock(&mirror->lock); 2339 devinfo->layout_type = LAYOUT_FLEX_FILES; 2340 devinfo->ld_private.ops = &layoutstat_ops; 2341 devinfo->ld_private.data = mirror; 2342 2343 devinfo++; 2344 i++; 2345 } 2346 return i; 2347 } 2348 2349 static int 2350 ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args) 2351 { 2352 struct nfs4_flexfile_layout *ff_layout; 2353 const int dev_count = PNFS_LAYOUTSTATS_MAXDEV; 2354 2355 /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */ 2356 args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO); 2357 if (!args->devinfo) 2358 return -ENOMEM; 2359 2360 spin_lock(&args->inode->i_lock); 2361 ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout); 2362 args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr, 2363 &args->devinfo[0], dev_count); 2364 spin_unlock(&args->inode->i_lock); 2365 if (!args->num_dev) { 2366 kfree(args->devinfo); 2367 args->devinfo = NULL; 2368 return -ENOENT; 2369 } 2370 2371 return 0; 2372 } 2373 2374 static struct pnfs_layoutdriver_type flexfilelayout_type = { 2375 .id = LAYOUT_FLEX_FILES, 2376 .name = "LAYOUT_FLEX_FILES", 2377 .owner = THIS_MODULE, 2378 .alloc_layout_hdr = ff_layout_alloc_layout_hdr, 2379 .free_layout_hdr = ff_layout_free_layout_hdr, 2380 .alloc_lseg = ff_layout_alloc_lseg, 2381 .free_lseg = ff_layout_free_lseg, 2382 .add_lseg = ff_layout_add_lseg, 2383 .pg_read_ops = &ff_layout_pg_read_ops, 2384 .pg_write_ops = &ff_layout_pg_write_ops, 2385 .get_ds_info = ff_layout_get_ds_info, 2386 .free_deviceid_node = ff_layout_free_deviceid_node, 2387 .mark_request_commit = pnfs_layout_mark_request_commit, 2388 .clear_request_commit = pnfs_generic_clear_request_commit, 2389 .scan_commit_lists = pnfs_generic_scan_commit_lists, 2390 .recover_commit_reqs = pnfs_generic_recover_commit_reqs, 2391 .commit_pagelist = ff_layout_commit_pagelist, 2392 .read_pagelist = ff_layout_read_pagelist, 2393 .write_pagelist = ff_layout_write_pagelist, 2394 .alloc_deviceid_node = ff_layout_alloc_deviceid_node, 2395 .prepare_layoutreturn = ff_layout_prepare_layoutreturn, 2396 .sync = pnfs_nfs_generic_sync, 2397 .prepare_layoutstats = ff_layout_prepare_layoutstats, 2398 }; 2399 2400 static int __init nfs4flexfilelayout_init(void) 2401 { 2402 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n", 2403 __func__); 2404 if (!ff_zero_group) { 2405 ff_zero_group = groups_alloc(0); 2406 if (!ff_zero_group) 2407 return -ENOMEM; 2408 } 2409 return pnfs_register_layoutdriver(&flexfilelayout_type); 2410 } 2411 2412 static void __exit nfs4flexfilelayout_exit(void) 2413 { 2414 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n", 2415 __func__); 2416 pnfs_unregister_layoutdriver(&flexfilelayout_type); 2417 if (ff_zero_group) { 2418 put_group_info(ff_zero_group); 2419 ff_zero_group = NULL; 2420 } 2421 } 2422 2423 MODULE_ALIAS("nfs-layouttype4-4"); 2424 2425 MODULE_LICENSE("GPL"); 2426 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver"); 2427 2428 module_init(nfs4flexfilelayout_init); 2429 module_exit(nfs4flexfilelayout_exit); 2430