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 s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL; 623 struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(mirror->layout); 624 625 nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now); 626 if (!mirror->start_time) 627 mirror->start_time = now; 628 if (mirror->report_interval != 0) 629 report_interval = (s64)mirror->report_interval * 1000LL; 630 else if (layoutstats_timer != 0) 631 report_interval = (s64)layoutstats_timer * 1000LL; 632 if (ktime_to_ms(ktime_sub(now, ffl->last_report_time)) >= 633 report_interval) { 634 ffl->last_report_time = now; 635 return true; 636 } 637 638 return false; 639 } 640 641 static void 642 nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat, 643 __u64 requested) 644 { 645 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat; 646 647 iostat->ops_requested++; 648 iostat->bytes_requested += requested; 649 } 650 651 static void 652 nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat, 653 __u64 requested, 654 __u64 completed, 655 ktime_t time_completed, 656 ktime_t time_started) 657 { 658 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat; 659 ktime_t completion_time = ktime_sub(time_completed, time_started); 660 ktime_t timer; 661 662 iostat->ops_completed++; 663 iostat->bytes_completed += completed; 664 iostat->bytes_not_delivered += requested - completed; 665 666 timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed); 667 iostat->total_busy_time = 668 ktime_add(iostat->total_busy_time, timer); 669 iostat->aggregate_completion_time = 670 ktime_add(iostat->aggregate_completion_time, 671 completion_time); 672 } 673 674 static void 675 nfs4_ff_layout_stat_io_start_read(struct inode *inode, 676 struct nfs4_ff_layout_mirror *mirror, 677 __u64 requested, ktime_t now) 678 { 679 bool report; 680 681 spin_lock(&mirror->lock); 682 report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now); 683 nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested); 684 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags); 685 spin_unlock(&mirror->lock); 686 687 if (report) 688 pnfs_report_layoutstat(inode, GFP_KERNEL); 689 } 690 691 static void 692 nfs4_ff_layout_stat_io_end_read(struct rpc_task *task, 693 struct nfs4_ff_layout_mirror *mirror, 694 __u64 requested, 695 __u64 completed) 696 { 697 spin_lock(&mirror->lock); 698 nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat, 699 requested, completed, 700 ktime_get(), task->tk_start); 701 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags); 702 spin_unlock(&mirror->lock); 703 } 704 705 static void 706 nfs4_ff_layout_stat_io_start_write(struct inode *inode, 707 struct nfs4_ff_layout_mirror *mirror, 708 __u64 requested, ktime_t now) 709 { 710 bool report; 711 712 spin_lock(&mirror->lock); 713 report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now); 714 nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested); 715 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags); 716 spin_unlock(&mirror->lock); 717 718 if (report) 719 pnfs_report_layoutstat(inode, GFP_NOIO); 720 } 721 722 static void 723 nfs4_ff_layout_stat_io_end_write(struct rpc_task *task, 724 struct nfs4_ff_layout_mirror *mirror, 725 __u64 requested, 726 __u64 completed, 727 enum nfs3_stable_how committed) 728 { 729 if (committed == NFS_UNSTABLE) 730 requested = completed = 0; 731 732 spin_lock(&mirror->lock); 733 nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat, 734 requested, completed, ktime_get(), task->tk_start); 735 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags); 736 spin_unlock(&mirror->lock); 737 } 738 739 static int 740 ff_layout_alloc_commit_info(struct pnfs_layout_segment *lseg, 741 struct nfs_commit_info *cinfo, 742 gfp_t gfp_flags) 743 { 744 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg); 745 struct pnfs_commit_bucket *buckets; 746 int size; 747 748 if (cinfo->ds->nbuckets != 0) { 749 /* This assumes there is only one RW lseg per file. 750 * To support multiple lseg per file, we need to 751 * change struct pnfs_commit_bucket to allow dynamic 752 * increasing nbuckets. 753 */ 754 return 0; 755 } 756 757 size = ff_layout_get_lseg_count(fls) * FF_LAYOUT_MIRROR_COUNT(lseg); 758 759 buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket), 760 gfp_flags); 761 if (!buckets) 762 return -ENOMEM; 763 else { 764 int i; 765 766 spin_lock(&cinfo->inode->i_lock); 767 if (cinfo->ds->nbuckets != 0) 768 kfree(buckets); 769 else { 770 cinfo->ds->buckets = buckets; 771 cinfo->ds->nbuckets = size; 772 for (i = 0; i < size; i++) { 773 INIT_LIST_HEAD(&buckets[i].written); 774 INIT_LIST_HEAD(&buckets[i].committing); 775 /* mark direct verifier as unset */ 776 buckets[i].direct_verf.committed = 777 NFS_INVALID_STABLE_HOW; 778 } 779 } 780 spin_unlock(&cinfo->inode->i_lock); 781 return 0; 782 } 783 } 784 785 static struct nfs4_pnfs_ds * 786 ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg, 787 int start_idx, 788 int *best_idx) 789 { 790 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg); 791 struct nfs4_pnfs_ds *ds; 792 bool fail_return = false; 793 int idx; 794 795 /* mirrors are sorted by efficiency */ 796 for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) { 797 if (idx+1 == fls->mirror_array_cnt) 798 fail_return = true; 799 ds = nfs4_ff_layout_prepare_ds(lseg, idx, fail_return); 800 if (ds) { 801 *best_idx = idx; 802 return ds; 803 } 804 } 805 806 return NULL; 807 } 808 809 static void 810 ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio, 811 struct nfs_page *req, 812 bool strict_iomode) 813 { 814 retry_strict: 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 /* If we don't have checking, do get a IOMODE_RW 829 * segment, and the server wants to avoid READs 830 * there, then retry! 831 */ 832 if (pgio->pg_lseg && !strict_iomode && 833 ff_layout_avoid_read_on_rw(pgio->pg_lseg)) { 834 strict_iomode = true; 835 goto retry_strict; 836 } 837 } 838 839 static void 840 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio, 841 struct nfs_page *req) 842 { 843 struct nfs_pgio_mirror *pgm; 844 struct nfs4_ff_layout_mirror *mirror; 845 struct nfs4_pnfs_ds *ds; 846 int ds_idx; 847 848 retry: 849 /* Use full layout for now */ 850 if (!pgio->pg_lseg) 851 ff_layout_pg_get_read(pgio, req, false); 852 else if (ff_layout_avoid_read_on_rw(pgio->pg_lseg)) 853 ff_layout_pg_get_read(pgio, req, true); 854 855 /* If no lseg, fall back to read through mds */ 856 if (pgio->pg_lseg == NULL) 857 goto out_mds; 858 859 ds = ff_layout_choose_best_ds_for_read(pgio->pg_lseg, 0, &ds_idx); 860 if (!ds) { 861 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg)) 862 goto out_mds; 863 pnfs_put_lseg(pgio->pg_lseg); 864 pgio->pg_lseg = NULL; 865 /* Sleep for 1 second before retrying */ 866 ssleep(1); 867 goto retry; 868 } 869 870 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx); 871 872 pgio->pg_mirror_idx = ds_idx; 873 874 /* read always uses only one mirror - idx 0 for pgio layer */ 875 pgm = &pgio->pg_mirrors[0]; 876 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize; 877 878 return; 879 out_mds: 880 pnfs_put_lseg(pgio->pg_lseg); 881 pgio->pg_lseg = NULL; 882 nfs_pageio_reset_read_mds(pgio); 883 } 884 885 static void 886 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio, 887 struct nfs_page *req) 888 { 889 struct nfs4_ff_layout_mirror *mirror; 890 struct nfs_pgio_mirror *pgm; 891 struct nfs_commit_info cinfo; 892 struct nfs4_pnfs_ds *ds; 893 int i; 894 int status; 895 896 retry: 897 if (!pgio->pg_lseg) { 898 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, 899 req->wb_context, 900 0, 901 NFS4_MAX_UINT64, 902 IOMODE_RW, 903 false, 904 GFP_NOFS); 905 if (IS_ERR(pgio->pg_lseg)) { 906 pgio->pg_error = PTR_ERR(pgio->pg_lseg); 907 pgio->pg_lseg = NULL; 908 return; 909 } 910 } 911 /* If no lseg, fall back to write through mds */ 912 if (pgio->pg_lseg == NULL) 913 goto out_mds; 914 915 nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq); 916 status = ff_layout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS); 917 if (status < 0) 918 goto out_mds; 919 920 /* Use a direct mapping of ds_idx to pgio mirror_idx */ 921 if (WARN_ON_ONCE(pgio->pg_mirror_count != 922 FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg))) 923 goto out_mds; 924 925 for (i = 0; i < pgio->pg_mirror_count; i++) { 926 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, i, true); 927 if (!ds) { 928 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg)) 929 goto out_mds; 930 pnfs_put_lseg(pgio->pg_lseg); 931 pgio->pg_lseg = NULL; 932 /* Sleep for 1 second before retrying */ 933 ssleep(1); 934 goto retry; 935 } 936 pgm = &pgio->pg_mirrors[i]; 937 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i); 938 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize; 939 } 940 941 return; 942 943 out_mds: 944 pnfs_put_lseg(pgio->pg_lseg); 945 pgio->pg_lseg = NULL; 946 nfs_pageio_reset_write_mds(pgio); 947 } 948 949 static unsigned int 950 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio, 951 struct nfs_page *req) 952 { 953 if (!pgio->pg_lseg) { 954 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, 955 req->wb_context, 956 0, 957 NFS4_MAX_UINT64, 958 IOMODE_RW, 959 false, 960 GFP_NOFS); 961 if (IS_ERR(pgio->pg_lseg)) { 962 pgio->pg_error = PTR_ERR(pgio->pg_lseg); 963 pgio->pg_lseg = NULL; 964 goto out; 965 } 966 } 967 if (pgio->pg_lseg) 968 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg); 969 970 /* no lseg means that pnfs is not in use, so no mirroring here */ 971 nfs_pageio_reset_write_mds(pgio); 972 out: 973 return 1; 974 } 975 976 static const struct nfs_pageio_ops ff_layout_pg_read_ops = { 977 .pg_init = ff_layout_pg_init_read, 978 .pg_test = pnfs_generic_pg_test, 979 .pg_doio = pnfs_generic_pg_readpages, 980 .pg_cleanup = pnfs_generic_pg_cleanup, 981 }; 982 983 static const struct nfs_pageio_ops ff_layout_pg_write_ops = { 984 .pg_init = ff_layout_pg_init_write, 985 .pg_test = pnfs_generic_pg_test, 986 .pg_doio = pnfs_generic_pg_writepages, 987 .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write, 988 .pg_cleanup = pnfs_generic_pg_cleanup, 989 }; 990 991 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs) 992 { 993 struct rpc_task *task = &hdr->task; 994 995 pnfs_layoutcommit_inode(hdr->inode, false); 996 997 if (retry_pnfs) { 998 dprintk("%s Reset task %5u for i/o through pNFS " 999 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__, 1000 hdr->task.tk_pid, 1001 hdr->inode->i_sb->s_id, 1002 (unsigned long long)NFS_FILEID(hdr->inode), 1003 hdr->args.count, 1004 (unsigned long long)hdr->args.offset); 1005 1006 hdr->completion_ops->reschedule_io(hdr); 1007 return; 1008 } 1009 1010 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { 1011 dprintk("%s Reset task %5u for i/o through MDS " 1012 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__, 1013 hdr->task.tk_pid, 1014 hdr->inode->i_sb->s_id, 1015 (unsigned long long)NFS_FILEID(hdr->inode), 1016 hdr->args.count, 1017 (unsigned long long)hdr->args.offset); 1018 1019 task->tk_status = pnfs_write_done_resend_to_mds(hdr); 1020 } 1021 } 1022 1023 static void ff_layout_reset_read(struct nfs_pgio_header *hdr) 1024 { 1025 struct rpc_task *task = &hdr->task; 1026 1027 pnfs_layoutcommit_inode(hdr->inode, false); 1028 1029 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { 1030 dprintk("%s Reset task %5u for i/o through MDS " 1031 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__, 1032 hdr->task.tk_pid, 1033 hdr->inode->i_sb->s_id, 1034 (unsigned long long)NFS_FILEID(hdr->inode), 1035 hdr->args.count, 1036 (unsigned long long)hdr->args.offset); 1037 1038 task->tk_status = pnfs_read_done_resend_to_mds(hdr); 1039 } 1040 } 1041 1042 static int ff_layout_async_handle_error_v4(struct rpc_task *task, 1043 struct nfs4_state *state, 1044 struct nfs_client *clp, 1045 struct pnfs_layout_segment *lseg, 1046 int idx) 1047 { 1048 struct pnfs_layout_hdr *lo = lseg->pls_layout; 1049 struct inode *inode = lo->plh_inode; 1050 struct nfs_server *mds_server = NFS_SERVER(inode); 1051 1052 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx); 1053 struct nfs_client *mds_client = mds_server->nfs_client; 1054 struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table; 1055 1056 switch (task->tk_status) { 1057 /* MDS state errors */ 1058 case -NFS4ERR_DELEG_REVOKED: 1059 case -NFS4ERR_ADMIN_REVOKED: 1060 case -NFS4ERR_BAD_STATEID: 1061 if (state == NULL) 1062 break; 1063 nfs_remove_bad_delegation(state->inode, NULL); 1064 case -NFS4ERR_OPENMODE: 1065 if (state == NULL) 1066 break; 1067 if (nfs4_schedule_stateid_recovery(mds_server, state) < 0) 1068 goto out_bad_stateid; 1069 goto wait_on_recovery; 1070 case -NFS4ERR_EXPIRED: 1071 if (state != NULL) { 1072 if (nfs4_schedule_stateid_recovery(mds_server, state) < 0) 1073 goto out_bad_stateid; 1074 } 1075 nfs4_schedule_lease_recovery(mds_client); 1076 goto wait_on_recovery; 1077 /* DS session errors */ 1078 case -NFS4ERR_BADSESSION: 1079 case -NFS4ERR_BADSLOT: 1080 case -NFS4ERR_BAD_HIGH_SLOT: 1081 case -NFS4ERR_DEADSESSION: 1082 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION: 1083 case -NFS4ERR_SEQ_FALSE_RETRY: 1084 case -NFS4ERR_SEQ_MISORDERED: 1085 dprintk("%s ERROR %d, Reset session. Exchangeid " 1086 "flags 0x%x\n", __func__, task->tk_status, 1087 clp->cl_exchange_flags); 1088 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status); 1089 break; 1090 case -NFS4ERR_DELAY: 1091 case -NFS4ERR_GRACE: 1092 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX); 1093 break; 1094 case -NFS4ERR_RETRY_UNCACHED_REP: 1095 break; 1096 /* Invalidate Layout errors */ 1097 case -NFS4ERR_PNFS_NO_LAYOUT: 1098 case -ESTALE: /* mapped NFS4ERR_STALE */ 1099 case -EBADHANDLE: /* mapped NFS4ERR_BADHANDLE */ 1100 case -EISDIR: /* mapped NFS4ERR_ISDIR */ 1101 case -NFS4ERR_FHEXPIRED: 1102 case -NFS4ERR_WRONG_TYPE: 1103 dprintk("%s Invalid layout error %d\n", __func__, 1104 task->tk_status); 1105 /* 1106 * Destroy layout so new i/o will get a new layout. 1107 * Layout will not be destroyed until all current lseg 1108 * references are put. Mark layout as invalid to resend failed 1109 * i/o and all i/o waiting on the slot table to the MDS until 1110 * layout is destroyed and a new valid layout is obtained. 1111 */ 1112 pnfs_destroy_layout(NFS_I(inode)); 1113 rpc_wake_up(&tbl->slot_tbl_waitq); 1114 goto reset; 1115 /* RPC connection errors */ 1116 case -ECONNREFUSED: 1117 case -EHOSTDOWN: 1118 case -EHOSTUNREACH: 1119 case -ENETUNREACH: 1120 case -EIO: 1121 case -ETIMEDOUT: 1122 case -EPIPE: 1123 dprintk("%s DS connection error %d\n", __func__, 1124 task->tk_status); 1125 nfs4_delete_deviceid(devid->ld, devid->nfs_client, 1126 &devid->deviceid); 1127 rpc_wake_up(&tbl->slot_tbl_waitq); 1128 /* fall through */ 1129 default: 1130 if (ff_layout_avoid_mds_available_ds(lseg)) 1131 return -NFS4ERR_RESET_TO_PNFS; 1132 reset: 1133 dprintk("%s Retry through MDS. Error %d\n", __func__, 1134 task->tk_status); 1135 return -NFS4ERR_RESET_TO_MDS; 1136 } 1137 out: 1138 task->tk_status = 0; 1139 return -EAGAIN; 1140 out_bad_stateid: 1141 task->tk_status = -EIO; 1142 return 0; 1143 wait_on_recovery: 1144 rpc_sleep_on(&mds_client->cl_rpcwaitq, task, NULL); 1145 if (test_bit(NFS4CLNT_MANAGER_RUNNING, &mds_client->cl_state) == 0) 1146 rpc_wake_up_queued_task(&mds_client->cl_rpcwaitq, task); 1147 goto out; 1148 } 1149 1150 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */ 1151 static int ff_layout_async_handle_error_v3(struct rpc_task *task, 1152 struct pnfs_layout_segment *lseg, 1153 int idx) 1154 { 1155 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx); 1156 1157 switch (task->tk_status) { 1158 /* File access problems. Don't mark the device as unavailable */ 1159 case -EACCES: 1160 case -ESTALE: 1161 case -EISDIR: 1162 case -EBADHANDLE: 1163 case -ELOOP: 1164 case -ENOSPC: 1165 break; 1166 case -EJUKEBOX: 1167 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY); 1168 goto out_retry; 1169 default: 1170 dprintk("%s DS connection error %d\n", __func__, 1171 task->tk_status); 1172 nfs4_delete_deviceid(devid->ld, devid->nfs_client, 1173 &devid->deviceid); 1174 } 1175 /* FIXME: Need to prevent infinite looping here. */ 1176 return -NFS4ERR_RESET_TO_PNFS; 1177 out_retry: 1178 task->tk_status = 0; 1179 rpc_restart_call_prepare(task); 1180 rpc_delay(task, NFS_JUKEBOX_RETRY_TIME); 1181 return -EAGAIN; 1182 } 1183 1184 static int ff_layout_async_handle_error(struct rpc_task *task, 1185 struct nfs4_state *state, 1186 struct nfs_client *clp, 1187 struct pnfs_layout_segment *lseg, 1188 int idx) 1189 { 1190 int vers = clp->cl_nfs_mod->rpc_vers->number; 1191 1192 if (task->tk_status >= 0) 1193 return 0; 1194 1195 /* Handle the case of an invalid layout segment */ 1196 if (!pnfs_is_valid_lseg(lseg)) 1197 return -NFS4ERR_RESET_TO_PNFS; 1198 1199 switch (vers) { 1200 case 3: 1201 return ff_layout_async_handle_error_v3(task, lseg, idx); 1202 case 4: 1203 return ff_layout_async_handle_error_v4(task, state, clp, 1204 lseg, idx); 1205 default: 1206 /* should never happen */ 1207 WARN_ON_ONCE(1); 1208 return 0; 1209 } 1210 } 1211 1212 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg, 1213 int idx, u64 offset, u64 length, 1214 u32 status, int opnum, int error) 1215 { 1216 struct nfs4_ff_layout_mirror *mirror; 1217 int err; 1218 1219 if (status == 0) { 1220 switch (error) { 1221 case -ETIMEDOUT: 1222 case -EPFNOSUPPORT: 1223 case -EPROTONOSUPPORT: 1224 case -EOPNOTSUPP: 1225 case -ECONNREFUSED: 1226 case -ECONNRESET: 1227 case -EHOSTDOWN: 1228 case -EHOSTUNREACH: 1229 case -ENETUNREACH: 1230 case -EADDRINUSE: 1231 case -ENOBUFS: 1232 case -EPIPE: 1233 case -EPERM: 1234 status = NFS4ERR_NXIO; 1235 break; 1236 case -EACCES: 1237 status = NFS4ERR_ACCESS; 1238 break; 1239 default: 1240 return; 1241 } 1242 } 1243 1244 switch (status) { 1245 case NFS4ERR_DELAY: 1246 case NFS4ERR_GRACE: 1247 return; 1248 default: 1249 break; 1250 } 1251 1252 mirror = FF_LAYOUT_COMP(lseg, idx); 1253 err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout), 1254 mirror, offset, length, status, opnum, 1255 GFP_NOIO); 1256 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode, lseg); 1257 dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status); 1258 } 1259 1260 /* NFS_PROTO call done callback routines */ 1261 static int ff_layout_read_done_cb(struct rpc_task *task, 1262 struct nfs_pgio_header *hdr) 1263 { 1264 int err; 1265 1266 trace_nfs4_pnfs_read(hdr, task->tk_status); 1267 if (task->tk_status < 0) 1268 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx, 1269 hdr->args.offset, hdr->args.count, 1270 hdr->res.op_status, OP_READ, 1271 task->tk_status); 1272 err = ff_layout_async_handle_error(task, hdr->args.context->state, 1273 hdr->ds_clp, hdr->lseg, 1274 hdr->pgio_mirror_idx); 1275 1276 switch (err) { 1277 case -NFS4ERR_RESET_TO_PNFS: 1278 if (ff_layout_choose_best_ds_for_read(hdr->lseg, 1279 hdr->pgio_mirror_idx + 1, 1280 &hdr->pgio_mirror_idx)) 1281 goto out_eagain; 1282 ff_layout_read_record_layoutstats_done(task, hdr); 1283 pnfs_read_resend_pnfs(hdr); 1284 return task->tk_status; 1285 case -NFS4ERR_RESET_TO_MDS: 1286 ff_layout_reset_read(hdr); 1287 return task->tk_status; 1288 case -EAGAIN: 1289 goto out_eagain; 1290 } 1291 1292 return 0; 1293 out_eagain: 1294 rpc_restart_call_prepare(task); 1295 return -EAGAIN; 1296 } 1297 1298 static bool 1299 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg) 1300 { 1301 return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT); 1302 } 1303 1304 /* 1305 * We reference the rpc_cred of the first WRITE that triggers the need for 1306 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound. 1307 * rfc5661 is not clear about which credential should be used. 1308 * 1309 * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so 1310 * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751 1311 * we always send layoutcommit after DS writes. 1312 */ 1313 static void 1314 ff_layout_set_layoutcommit(struct inode *inode, 1315 struct pnfs_layout_segment *lseg, 1316 loff_t end_offset) 1317 { 1318 if (!ff_layout_need_layoutcommit(lseg)) 1319 return; 1320 1321 pnfs_set_layoutcommit(inode, lseg, end_offset); 1322 dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino, 1323 (unsigned long long) NFS_I(inode)->layout->plh_lwb); 1324 } 1325 1326 static bool 1327 ff_layout_device_unavailable(struct pnfs_layout_segment *lseg, int idx) 1328 { 1329 /* No mirroring for now */ 1330 struct nfs4_deviceid_node *node = FF_LAYOUT_DEVID_NODE(lseg, idx); 1331 1332 return ff_layout_test_devid_unavailable(node); 1333 } 1334 1335 static void ff_layout_read_record_layoutstats_start(struct rpc_task *task, 1336 struct nfs_pgio_header *hdr) 1337 { 1338 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags)) 1339 return; 1340 nfs4_ff_layout_stat_io_start_read(hdr->inode, 1341 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx), 1342 hdr->args.count, 1343 task->tk_start); 1344 } 1345 1346 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task, 1347 struct nfs_pgio_header *hdr) 1348 { 1349 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags)) 1350 return; 1351 nfs4_ff_layout_stat_io_end_read(task, 1352 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx), 1353 hdr->args.count, 1354 hdr->res.count); 1355 } 1356 1357 static int ff_layout_read_prepare_common(struct rpc_task *task, 1358 struct nfs_pgio_header *hdr) 1359 { 1360 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) { 1361 rpc_exit(task, -EIO); 1362 return -EIO; 1363 } 1364 if (ff_layout_device_unavailable(hdr->lseg, hdr->pgio_mirror_idx)) { 1365 rpc_exit(task, -EHOSTDOWN); 1366 return -EAGAIN; 1367 } 1368 1369 ff_layout_read_record_layoutstats_start(task, hdr); 1370 return 0; 1371 } 1372 1373 /* 1374 * Call ops for the async read/write cases 1375 * In the case of dense layouts, the offset needs to be reset to its 1376 * original value. 1377 */ 1378 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data) 1379 { 1380 struct nfs_pgio_header *hdr = data; 1381 1382 if (ff_layout_read_prepare_common(task, hdr)) 1383 return; 1384 1385 rpc_call_start(task); 1386 } 1387 1388 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data) 1389 { 1390 struct nfs_pgio_header *hdr = data; 1391 1392 if (nfs4_setup_sequence(hdr->ds_clp, 1393 &hdr->args.seq_args, 1394 &hdr->res.seq_res, 1395 task)) 1396 return; 1397 1398 if (ff_layout_read_prepare_common(task, hdr)) 1399 return; 1400 1401 if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context, 1402 hdr->args.lock_context, FMODE_READ) == -EIO) 1403 rpc_exit(task, -EIO); /* lost lock, terminate I/O */ 1404 } 1405 1406 static void ff_layout_read_call_done(struct rpc_task *task, void *data) 1407 { 1408 struct nfs_pgio_header *hdr = data; 1409 1410 dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status); 1411 1412 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) && 1413 task->tk_status == 0) { 1414 nfs4_sequence_done(task, &hdr->res.seq_res); 1415 return; 1416 } 1417 1418 /* Note this may cause RPC to be resent */ 1419 hdr->mds_ops->rpc_call_done(task, hdr); 1420 } 1421 1422 static void ff_layout_read_count_stats(struct rpc_task *task, void *data) 1423 { 1424 struct nfs_pgio_header *hdr = data; 1425 1426 ff_layout_read_record_layoutstats_done(task, hdr); 1427 rpc_count_iostats_metrics(task, 1428 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]); 1429 } 1430 1431 static void ff_layout_read_release(void *data) 1432 { 1433 struct nfs_pgio_header *hdr = data; 1434 1435 ff_layout_read_record_layoutstats_done(&hdr->task, hdr); 1436 pnfs_generic_rw_release(data); 1437 } 1438 1439 1440 static int ff_layout_write_done_cb(struct rpc_task *task, 1441 struct nfs_pgio_header *hdr) 1442 { 1443 loff_t end_offs = 0; 1444 int err; 1445 1446 trace_nfs4_pnfs_write(hdr, task->tk_status); 1447 if (task->tk_status < 0) 1448 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx, 1449 hdr->args.offset, hdr->args.count, 1450 hdr->res.op_status, OP_WRITE, 1451 task->tk_status); 1452 err = ff_layout_async_handle_error(task, hdr->args.context->state, 1453 hdr->ds_clp, hdr->lseg, 1454 hdr->pgio_mirror_idx); 1455 1456 switch (err) { 1457 case -NFS4ERR_RESET_TO_PNFS: 1458 ff_layout_reset_write(hdr, true); 1459 return task->tk_status; 1460 case -NFS4ERR_RESET_TO_MDS: 1461 ff_layout_reset_write(hdr, false); 1462 return task->tk_status; 1463 case -EAGAIN: 1464 return -EAGAIN; 1465 } 1466 1467 if (hdr->res.verf->committed == NFS_FILE_SYNC || 1468 hdr->res.verf->committed == NFS_DATA_SYNC) 1469 end_offs = hdr->mds_offset + (loff_t)hdr->res.count; 1470 1471 /* Note: if the write is unstable, don't set end_offs until commit */ 1472 ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs); 1473 1474 /* zero out fattr since we don't care DS attr at all */ 1475 hdr->fattr.valid = 0; 1476 if (task->tk_status >= 0) 1477 nfs_writeback_update_inode(hdr); 1478 1479 return 0; 1480 } 1481 1482 static int ff_layout_commit_done_cb(struct rpc_task *task, 1483 struct nfs_commit_data *data) 1484 { 1485 int err; 1486 1487 trace_nfs4_pnfs_commit_ds(data, task->tk_status); 1488 if (task->tk_status < 0) 1489 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index, 1490 data->args.offset, data->args.count, 1491 data->res.op_status, OP_COMMIT, 1492 task->tk_status); 1493 err = ff_layout_async_handle_error(task, NULL, data->ds_clp, 1494 data->lseg, data->ds_commit_index); 1495 1496 switch (err) { 1497 case -NFS4ERR_RESET_TO_PNFS: 1498 pnfs_generic_prepare_to_resend_writes(data); 1499 return -EAGAIN; 1500 case -NFS4ERR_RESET_TO_MDS: 1501 pnfs_generic_prepare_to_resend_writes(data); 1502 return -EAGAIN; 1503 case -EAGAIN: 1504 rpc_restart_call_prepare(task); 1505 return -EAGAIN; 1506 } 1507 1508 ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb); 1509 1510 return 0; 1511 } 1512 1513 static void ff_layout_write_record_layoutstats_start(struct rpc_task *task, 1514 struct nfs_pgio_header *hdr) 1515 { 1516 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags)) 1517 return; 1518 nfs4_ff_layout_stat_io_start_write(hdr->inode, 1519 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx), 1520 hdr->args.count, 1521 task->tk_start); 1522 } 1523 1524 static void ff_layout_write_record_layoutstats_done(struct rpc_task *task, 1525 struct nfs_pgio_header *hdr) 1526 { 1527 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags)) 1528 return; 1529 nfs4_ff_layout_stat_io_end_write(task, 1530 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx), 1531 hdr->args.count, hdr->res.count, 1532 hdr->res.verf->committed); 1533 } 1534 1535 static int ff_layout_write_prepare_common(struct rpc_task *task, 1536 struct nfs_pgio_header *hdr) 1537 { 1538 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) { 1539 rpc_exit(task, -EIO); 1540 return -EIO; 1541 } 1542 1543 if (ff_layout_device_unavailable(hdr->lseg, hdr->pgio_mirror_idx)) { 1544 rpc_exit(task, -EHOSTDOWN); 1545 return -EAGAIN; 1546 } 1547 1548 ff_layout_write_record_layoutstats_start(task, hdr); 1549 return 0; 1550 } 1551 1552 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data) 1553 { 1554 struct nfs_pgio_header *hdr = data; 1555 1556 if (ff_layout_write_prepare_common(task, hdr)) 1557 return; 1558 1559 rpc_call_start(task); 1560 } 1561 1562 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data) 1563 { 1564 struct nfs_pgio_header *hdr = data; 1565 1566 if (nfs4_setup_sequence(hdr->ds_clp, 1567 &hdr->args.seq_args, 1568 &hdr->res.seq_res, 1569 task)) 1570 return; 1571 1572 if (ff_layout_write_prepare_common(task, hdr)) 1573 return; 1574 1575 if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context, 1576 hdr->args.lock_context, FMODE_WRITE) == -EIO) 1577 rpc_exit(task, -EIO); /* lost lock, terminate I/O */ 1578 } 1579 1580 static void ff_layout_write_call_done(struct rpc_task *task, void *data) 1581 { 1582 struct nfs_pgio_header *hdr = data; 1583 1584 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) && 1585 task->tk_status == 0) { 1586 nfs4_sequence_done(task, &hdr->res.seq_res); 1587 return; 1588 } 1589 1590 /* Note this may cause RPC to be resent */ 1591 hdr->mds_ops->rpc_call_done(task, hdr); 1592 } 1593 1594 static void ff_layout_write_count_stats(struct rpc_task *task, void *data) 1595 { 1596 struct nfs_pgio_header *hdr = data; 1597 1598 ff_layout_write_record_layoutstats_done(task, hdr); 1599 rpc_count_iostats_metrics(task, 1600 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]); 1601 } 1602 1603 static void ff_layout_write_release(void *data) 1604 { 1605 struct nfs_pgio_header *hdr = data; 1606 1607 ff_layout_write_record_layoutstats_done(&hdr->task, hdr); 1608 pnfs_generic_rw_release(data); 1609 } 1610 1611 static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task, 1612 struct nfs_commit_data *cdata) 1613 { 1614 if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags)) 1615 return; 1616 nfs4_ff_layout_stat_io_start_write(cdata->inode, 1617 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index), 1618 0, task->tk_start); 1619 } 1620 1621 static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task, 1622 struct nfs_commit_data *cdata) 1623 { 1624 struct nfs_page *req; 1625 __u64 count = 0; 1626 1627 if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags)) 1628 return; 1629 1630 if (task->tk_status == 0) { 1631 list_for_each_entry(req, &cdata->pages, wb_list) 1632 count += req->wb_bytes; 1633 } 1634 nfs4_ff_layout_stat_io_end_write(task, 1635 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index), 1636 count, count, NFS_FILE_SYNC); 1637 } 1638 1639 static void ff_layout_commit_prepare_common(struct rpc_task *task, 1640 struct nfs_commit_data *cdata) 1641 { 1642 ff_layout_commit_record_layoutstats_start(task, cdata); 1643 } 1644 1645 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data) 1646 { 1647 ff_layout_commit_prepare_common(task, data); 1648 rpc_call_start(task); 1649 } 1650 1651 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data) 1652 { 1653 struct nfs_commit_data *wdata = data; 1654 1655 if (nfs4_setup_sequence(wdata->ds_clp, 1656 &wdata->args.seq_args, 1657 &wdata->res.seq_res, 1658 task)) 1659 return; 1660 ff_layout_commit_prepare_common(task, data); 1661 } 1662 1663 static void ff_layout_commit_done(struct rpc_task *task, void *data) 1664 { 1665 pnfs_generic_write_commit_done(task, data); 1666 } 1667 1668 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data) 1669 { 1670 struct nfs_commit_data *cdata = data; 1671 1672 ff_layout_commit_record_layoutstats_done(task, cdata); 1673 rpc_count_iostats_metrics(task, 1674 &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]); 1675 } 1676 1677 static void ff_layout_commit_release(void *data) 1678 { 1679 struct nfs_commit_data *cdata = data; 1680 1681 ff_layout_commit_record_layoutstats_done(&cdata->task, cdata); 1682 pnfs_generic_commit_release(data); 1683 } 1684 1685 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = { 1686 .rpc_call_prepare = ff_layout_read_prepare_v3, 1687 .rpc_call_done = ff_layout_read_call_done, 1688 .rpc_count_stats = ff_layout_read_count_stats, 1689 .rpc_release = ff_layout_read_release, 1690 }; 1691 1692 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = { 1693 .rpc_call_prepare = ff_layout_read_prepare_v4, 1694 .rpc_call_done = ff_layout_read_call_done, 1695 .rpc_count_stats = ff_layout_read_count_stats, 1696 .rpc_release = ff_layout_read_release, 1697 }; 1698 1699 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = { 1700 .rpc_call_prepare = ff_layout_write_prepare_v3, 1701 .rpc_call_done = ff_layout_write_call_done, 1702 .rpc_count_stats = ff_layout_write_count_stats, 1703 .rpc_release = ff_layout_write_release, 1704 }; 1705 1706 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = { 1707 .rpc_call_prepare = ff_layout_write_prepare_v4, 1708 .rpc_call_done = ff_layout_write_call_done, 1709 .rpc_count_stats = ff_layout_write_count_stats, 1710 .rpc_release = ff_layout_write_release, 1711 }; 1712 1713 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = { 1714 .rpc_call_prepare = ff_layout_commit_prepare_v3, 1715 .rpc_call_done = ff_layout_commit_done, 1716 .rpc_count_stats = ff_layout_commit_count_stats, 1717 .rpc_release = ff_layout_commit_release, 1718 }; 1719 1720 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = { 1721 .rpc_call_prepare = ff_layout_commit_prepare_v4, 1722 .rpc_call_done = ff_layout_commit_done, 1723 .rpc_count_stats = ff_layout_commit_count_stats, 1724 .rpc_release = ff_layout_commit_release, 1725 }; 1726 1727 static enum pnfs_try_status 1728 ff_layout_read_pagelist(struct nfs_pgio_header *hdr) 1729 { 1730 struct pnfs_layout_segment *lseg = hdr->lseg; 1731 struct nfs4_pnfs_ds *ds; 1732 struct rpc_clnt *ds_clnt; 1733 struct rpc_cred *ds_cred; 1734 loff_t offset = hdr->args.offset; 1735 u32 idx = hdr->pgio_mirror_idx; 1736 int vers; 1737 struct nfs_fh *fh; 1738 1739 dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n", 1740 __func__, hdr->inode->i_ino, 1741 hdr->args.pgbase, (size_t)hdr->args.count, offset); 1742 1743 ds = nfs4_ff_layout_prepare_ds(lseg, idx, false); 1744 if (!ds) 1745 goto out_failed; 1746 1747 ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp, 1748 hdr->inode); 1749 if (IS_ERR(ds_clnt)) 1750 goto out_failed; 1751 1752 ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred); 1753 if (!ds_cred) 1754 goto out_failed; 1755 1756 vers = nfs4_ff_layout_ds_version(lseg, idx); 1757 1758 dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__, 1759 ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count), vers); 1760 1761 hdr->pgio_done_cb = ff_layout_read_done_cb; 1762 atomic_inc(&ds->ds_clp->cl_count); 1763 hdr->ds_clp = ds->ds_clp; 1764 fh = nfs4_ff_layout_select_ds_fh(lseg, idx); 1765 if (fh) 1766 hdr->args.fh = fh; 1767 /* 1768 * Note that if we ever decide to split across DSes, 1769 * then we may need to handle dense-like offsets. 1770 */ 1771 hdr->args.offset = offset; 1772 hdr->mds_offset = offset; 1773 1774 /* Perform an asynchronous read to ds */ 1775 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops, 1776 vers == 3 ? &ff_layout_read_call_ops_v3 : 1777 &ff_layout_read_call_ops_v4, 1778 0, RPC_TASK_SOFTCONN); 1779 put_rpccred(ds_cred); 1780 return PNFS_ATTEMPTED; 1781 1782 out_failed: 1783 if (ff_layout_avoid_mds_available_ds(lseg)) 1784 return PNFS_TRY_AGAIN; 1785 return PNFS_NOT_ATTEMPTED; 1786 } 1787 1788 /* Perform async writes. */ 1789 static enum pnfs_try_status 1790 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync) 1791 { 1792 struct pnfs_layout_segment *lseg = hdr->lseg; 1793 struct nfs4_pnfs_ds *ds; 1794 struct rpc_clnt *ds_clnt; 1795 struct rpc_cred *ds_cred; 1796 loff_t offset = hdr->args.offset; 1797 int vers; 1798 struct nfs_fh *fh; 1799 int idx = hdr->pgio_mirror_idx; 1800 1801 ds = nfs4_ff_layout_prepare_ds(lseg, idx, true); 1802 if (!ds) 1803 return PNFS_NOT_ATTEMPTED; 1804 1805 ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp, 1806 hdr->inode); 1807 if (IS_ERR(ds_clnt)) 1808 return PNFS_NOT_ATTEMPTED; 1809 1810 ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred); 1811 if (!ds_cred) 1812 return PNFS_NOT_ATTEMPTED; 1813 1814 vers = nfs4_ff_layout_ds_version(lseg, idx); 1815 1816 dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n", 1817 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count, 1818 offset, ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count), 1819 vers); 1820 1821 hdr->pgio_done_cb = ff_layout_write_done_cb; 1822 atomic_inc(&ds->ds_clp->cl_count); 1823 hdr->ds_clp = ds->ds_clp; 1824 hdr->ds_commit_idx = idx; 1825 fh = nfs4_ff_layout_select_ds_fh(lseg, idx); 1826 if (fh) 1827 hdr->args.fh = fh; 1828 1829 /* 1830 * Note that if we ever decide to split across DSes, 1831 * then we may need to handle dense-like offsets. 1832 */ 1833 hdr->args.offset = offset; 1834 1835 /* Perform an asynchronous write */ 1836 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops, 1837 vers == 3 ? &ff_layout_write_call_ops_v3 : 1838 &ff_layout_write_call_ops_v4, 1839 sync, RPC_TASK_SOFTCONN); 1840 put_rpccred(ds_cred); 1841 return PNFS_ATTEMPTED; 1842 } 1843 1844 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i) 1845 { 1846 return i; 1847 } 1848 1849 static struct nfs_fh * 1850 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i) 1851 { 1852 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg); 1853 1854 /* FIXME: Assume that there is only one NFS version available 1855 * for the DS. 1856 */ 1857 return &flseg->mirror_array[i]->fh_versions[0]; 1858 } 1859 1860 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how) 1861 { 1862 struct pnfs_layout_segment *lseg = data->lseg; 1863 struct nfs4_pnfs_ds *ds; 1864 struct rpc_clnt *ds_clnt; 1865 struct rpc_cred *ds_cred; 1866 u32 idx; 1867 int vers, ret; 1868 struct nfs_fh *fh; 1869 1870 idx = calc_ds_index_from_commit(lseg, data->ds_commit_index); 1871 ds = nfs4_ff_layout_prepare_ds(lseg, idx, true); 1872 if (!ds) 1873 goto out_err; 1874 1875 ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp, 1876 data->inode); 1877 if (IS_ERR(ds_clnt)) 1878 goto out_err; 1879 1880 ds_cred = ff_layout_get_ds_cred(lseg, idx, data->cred); 1881 if (!ds_cred) 1882 goto out_err; 1883 1884 vers = nfs4_ff_layout_ds_version(lseg, idx); 1885 1886 dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__, 1887 data->inode->i_ino, how, atomic_read(&ds->ds_clp->cl_count), 1888 vers); 1889 data->commit_done_cb = ff_layout_commit_done_cb; 1890 data->cred = ds_cred; 1891 atomic_inc(&ds->ds_clp->cl_count); 1892 data->ds_clp = ds->ds_clp; 1893 fh = select_ds_fh_from_commit(lseg, data->ds_commit_index); 1894 if (fh) 1895 data->args.fh = fh; 1896 1897 ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops, 1898 vers == 3 ? &ff_layout_commit_call_ops_v3 : 1899 &ff_layout_commit_call_ops_v4, 1900 how, RPC_TASK_SOFTCONN); 1901 put_rpccred(ds_cred); 1902 return ret; 1903 out_err: 1904 pnfs_generic_prepare_to_resend_writes(data); 1905 pnfs_generic_commit_release(data); 1906 return -EAGAIN; 1907 } 1908 1909 static int 1910 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages, 1911 int how, struct nfs_commit_info *cinfo) 1912 { 1913 return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo, 1914 ff_layout_initiate_commit); 1915 } 1916 1917 static struct pnfs_ds_commit_info * 1918 ff_layout_get_ds_info(struct inode *inode) 1919 { 1920 struct pnfs_layout_hdr *layout = NFS_I(inode)->layout; 1921 1922 if (layout == NULL) 1923 return NULL; 1924 1925 return &FF_LAYOUT_FROM_HDR(layout)->commit_info; 1926 } 1927 1928 static void 1929 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d) 1930 { 1931 nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds, 1932 id_node)); 1933 } 1934 1935 static int ff_layout_encode_ioerr(struct xdr_stream *xdr, 1936 const struct nfs4_layoutreturn_args *args, 1937 const struct nfs4_flexfile_layoutreturn_args *ff_args) 1938 { 1939 __be32 *start; 1940 1941 start = xdr_reserve_space(xdr, 4); 1942 if (unlikely(!start)) 1943 return -E2BIG; 1944 1945 *start = cpu_to_be32(ff_args->num_errors); 1946 /* This assume we always return _ALL_ layouts */ 1947 return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors); 1948 } 1949 1950 static void 1951 encode_opaque_fixed(struct xdr_stream *xdr, const void *buf, size_t len) 1952 { 1953 WARN_ON_ONCE(xdr_stream_encode_opaque_fixed(xdr, buf, len) < 0); 1954 } 1955 1956 static void 1957 ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr, 1958 const nfs4_stateid *stateid, 1959 const struct nfs42_layoutstat_devinfo *devinfo) 1960 { 1961 __be32 *p; 1962 1963 p = xdr_reserve_space(xdr, 8 + 8); 1964 p = xdr_encode_hyper(p, devinfo->offset); 1965 p = xdr_encode_hyper(p, devinfo->length); 1966 encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE); 1967 p = xdr_reserve_space(xdr, 4*8); 1968 p = xdr_encode_hyper(p, devinfo->read_count); 1969 p = xdr_encode_hyper(p, devinfo->read_bytes); 1970 p = xdr_encode_hyper(p, devinfo->write_count); 1971 p = xdr_encode_hyper(p, devinfo->write_bytes); 1972 encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE); 1973 } 1974 1975 static void 1976 ff_layout_encode_ff_iostat(struct xdr_stream *xdr, 1977 const nfs4_stateid *stateid, 1978 const struct nfs42_layoutstat_devinfo *devinfo) 1979 { 1980 ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo); 1981 ff_layout_encode_ff_layoutupdate(xdr, devinfo, 1982 devinfo->ld_private.data); 1983 } 1984 1985 /* report nothing for now */ 1986 static void ff_layout_encode_iostats_array(struct xdr_stream *xdr, 1987 const struct nfs4_layoutreturn_args *args, 1988 struct nfs4_flexfile_layoutreturn_args *ff_args) 1989 { 1990 __be32 *p; 1991 int i; 1992 1993 p = xdr_reserve_space(xdr, 4); 1994 *p = cpu_to_be32(ff_args->num_dev); 1995 for (i = 0; i < ff_args->num_dev; i++) 1996 ff_layout_encode_ff_iostat(xdr, 1997 &args->layout->plh_stateid, 1998 &ff_args->devinfo[i]); 1999 } 2000 2001 static void 2002 ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo, 2003 unsigned int num_entries) 2004 { 2005 unsigned int i; 2006 2007 for (i = 0; i < num_entries; i++) { 2008 if (!devinfo[i].ld_private.ops) 2009 continue; 2010 if (!devinfo[i].ld_private.ops->free) 2011 continue; 2012 devinfo[i].ld_private.ops->free(&devinfo[i].ld_private); 2013 } 2014 } 2015 2016 static struct nfs4_deviceid_node * 2017 ff_layout_alloc_deviceid_node(struct nfs_server *server, 2018 struct pnfs_device *pdev, gfp_t gfp_flags) 2019 { 2020 struct nfs4_ff_layout_ds *dsaddr; 2021 2022 dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags); 2023 if (!dsaddr) 2024 return NULL; 2025 return &dsaddr->id_node; 2026 } 2027 2028 static void 2029 ff_layout_encode_layoutreturn(struct xdr_stream *xdr, 2030 const void *voidargs, 2031 const struct nfs4_xdr_opaque_data *ff_opaque) 2032 { 2033 const struct nfs4_layoutreturn_args *args = voidargs; 2034 struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data; 2035 struct xdr_buf tmp_buf = { 2036 .head = { 2037 [0] = { 2038 .iov_base = page_address(ff_args->pages[0]), 2039 }, 2040 }, 2041 .buflen = PAGE_SIZE, 2042 }; 2043 struct xdr_stream tmp_xdr; 2044 __be32 *start; 2045 2046 dprintk("%s: Begin\n", __func__); 2047 2048 xdr_init_encode(&tmp_xdr, &tmp_buf, NULL); 2049 2050 ff_layout_encode_ioerr(&tmp_xdr, args, ff_args); 2051 ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args); 2052 2053 start = xdr_reserve_space(xdr, 4); 2054 *start = cpu_to_be32(tmp_buf.len); 2055 xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len); 2056 2057 dprintk("%s: Return\n", __func__); 2058 } 2059 2060 static void 2061 ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args) 2062 { 2063 struct nfs4_flexfile_layoutreturn_args *ff_args; 2064 2065 if (!args->data) 2066 return; 2067 ff_args = args->data; 2068 args->data = NULL; 2069 2070 ff_layout_free_ds_ioerr(&ff_args->errors); 2071 ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev); 2072 2073 put_page(ff_args->pages[0]); 2074 kfree(ff_args); 2075 } 2076 2077 static const struct nfs4_xdr_opaque_ops layoutreturn_ops = { 2078 .encode = ff_layout_encode_layoutreturn, 2079 .free = ff_layout_free_layoutreturn, 2080 }; 2081 2082 static int 2083 ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args) 2084 { 2085 struct nfs4_flexfile_layoutreturn_args *ff_args; 2086 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout); 2087 2088 ff_args = kmalloc(sizeof(*ff_args), GFP_KERNEL); 2089 if (!ff_args) 2090 goto out_nomem; 2091 ff_args->pages[0] = alloc_page(GFP_KERNEL); 2092 if (!ff_args->pages[0]) 2093 goto out_nomem_free; 2094 2095 INIT_LIST_HEAD(&ff_args->errors); 2096 ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout, 2097 &args->range, &ff_args->errors, 2098 FF_LAYOUTRETURN_MAXERR); 2099 2100 spin_lock(&args->inode->i_lock); 2101 ff_args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr, 2102 &ff_args->devinfo[0], ARRAY_SIZE(ff_args->devinfo)); 2103 spin_unlock(&args->inode->i_lock); 2104 2105 args->ld_private->ops = &layoutreturn_ops; 2106 args->ld_private->data = ff_args; 2107 return 0; 2108 out_nomem_free: 2109 kfree(ff_args); 2110 out_nomem: 2111 return -ENOMEM; 2112 } 2113 2114 static int 2115 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen) 2116 { 2117 const struct sockaddr_in *sin = (struct sockaddr_in *)sap; 2118 2119 return snprintf(buf, buflen, "%pI4", &sin->sin_addr); 2120 } 2121 2122 static size_t 2123 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf, 2124 const int buflen) 2125 { 2126 const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap; 2127 const struct in6_addr *addr = &sin6->sin6_addr; 2128 2129 /* 2130 * RFC 4291, Section 2.2.2 2131 * 2132 * Shorthanded ANY address 2133 */ 2134 if (ipv6_addr_any(addr)) 2135 return snprintf(buf, buflen, "::"); 2136 2137 /* 2138 * RFC 4291, Section 2.2.2 2139 * 2140 * Shorthanded loopback address 2141 */ 2142 if (ipv6_addr_loopback(addr)) 2143 return snprintf(buf, buflen, "::1"); 2144 2145 /* 2146 * RFC 4291, Section 2.2.3 2147 * 2148 * Special presentation address format for mapped v4 2149 * addresses. 2150 */ 2151 if (ipv6_addr_v4mapped(addr)) 2152 return snprintf(buf, buflen, "::ffff:%pI4", 2153 &addr->s6_addr32[3]); 2154 2155 /* 2156 * RFC 4291, Section 2.2.1 2157 */ 2158 return snprintf(buf, buflen, "%pI6c", addr); 2159 } 2160 2161 /* Derived from rpc_sockaddr2uaddr */ 2162 static void 2163 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da) 2164 { 2165 struct sockaddr *sap = (struct sockaddr *)&da->da_addr; 2166 char portbuf[RPCBIND_MAXUADDRPLEN]; 2167 char addrbuf[RPCBIND_MAXUADDRLEN]; 2168 char *netid; 2169 unsigned short port; 2170 int len, netid_len; 2171 __be32 *p; 2172 2173 switch (sap->sa_family) { 2174 case AF_INET: 2175 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0) 2176 return; 2177 port = ntohs(((struct sockaddr_in *)sap)->sin_port); 2178 netid = "tcp"; 2179 netid_len = 3; 2180 break; 2181 case AF_INET6: 2182 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0) 2183 return; 2184 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port); 2185 netid = "tcp6"; 2186 netid_len = 4; 2187 break; 2188 default: 2189 /* we only support tcp and tcp6 */ 2190 WARN_ON_ONCE(1); 2191 return; 2192 } 2193 2194 snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff); 2195 len = strlcat(addrbuf, portbuf, sizeof(addrbuf)); 2196 2197 p = xdr_reserve_space(xdr, 4 + netid_len); 2198 xdr_encode_opaque(p, netid, netid_len); 2199 2200 p = xdr_reserve_space(xdr, 4 + len); 2201 xdr_encode_opaque(p, addrbuf, len); 2202 } 2203 2204 static void 2205 ff_layout_encode_nfstime(struct xdr_stream *xdr, 2206 ktime_t t) 2207 { 2208 struct timespec64 ts; 2209 __be32 *p; 2210 2211 p = xdr_reserve_space(xdr, 12); 2212 ts = ktime_to_timespec64(t); 2213 p = xdr_encode_hyper(p, ts.tv_sec); 2214 *p++ = cpu_to_be32(ts.tv_nsec); 2215 } 2216 2217 static void 2218 ff_layout_encode_io_latency(struct xdr_stream *xdr, 2219 struct nfs4_ff_io_stat *stat) 2220 { 2221 __be32 *p; 2222 2223 p = xdr_reserve_space(xdr, 5 * 8); 2224 p = xdr_encode_hyper(p, stat->ops_requested); 2225 p = xdr_encode_hyper(p, stat->bytes_requested); 2226 p = xdr_encode_hyper(p, stat->ops_completed); 2227 p = xdr_encode_hyper(p, stat->bytes_completed); 2228 p = xdr_encode_hyper(p, stat->bytes_not_delivered); 2229 ff_layout_encode_nfstime(xdr, stat->total_busy_time); 2230 ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time); 2231 } 2232 2233 static void 2234 ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr, 2235 const struct nfs42_layoutstat_devinfo *devinfo, 2236 struct nfs4_ff_layout_mirror *mirror) 2237 { 2238 struct nfs4_pnfs_ds_addr *da; 2239 struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds; 2240 struct nfs_fh *fh = &mirror->fh_versions[0]; 2241 __be32 *p; 2242 2243 da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node); 2244 dprintk("%s: DS %s: encoding address %s\n", 2245 __func__, ds->ds_remotestr, da->da_remotestr); 2246 /* netaddr4 */ 2247 ff_layout_encode_netaddr(xdr, da); 2248 /* nfs_fh4 */ 2249 p = xdr_reserve_space(xdr, 4 + fh->size); 2250 xdr_encode_opaque(p, fh->data, fh->size); 2251 /* ff_io_latency4 read */ 2252 spin_lock(&mirror->lock); 2253 ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat); 2254 /* ff_io_latency4 write */ 2255 ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat); 2256 spin_unlock(&mirror->lock); 2257 /* nfstime4 */ 2258 ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time)); 2259 /* bool */ 2260 p = xdr_reserve_space(xdr, 4); 2261 *p = cpu_to_be32(false); 2262 } 2263 2264 static void 2265 ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args, 2266 const struct nfs4_xdr_opaque_data *opaque) 2267 { 2268 struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque, 2269 struct nfs42_layoutstat_devinfo, ld_private); 2270 __be32 *start; 2271 2272 /* layoutupdate length */ 2273 start = xdr_reserve_space(xdr, 4); 2274 ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data); 2275 2276 *start = cpu_to_be32((xdr->p - start - 1) * 4); 2277 } 2278 2279 static void 2280 ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque) 2281 { 2282 struct nfs4_ff_layout_mirror *mirror = opaque->data; 2283 2284 ff_layout_put_mirror(mirror); 2285 } 2286 2287 static const struct nfs4_xdr_opaque_ops layoutstat_ops = { 2288 .encode = ff_layout_encode_layoutstats, 2289 .free = ff_layout_free_layoutstats, 2290 }; 2291 2292 static int 2293 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo, 2294 struct nfs42_layoutstat_devinfo *devinfo, 2295 int dev_limit) 2296 { 2297 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo); 2298 struct nfs4_ff_layout_mirror *mirror; 2299 struct nfs4_deviceid_node *dev; 2300 int i = 0; 2301 2302 list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) { 2303 if (i >= dev_limit) 2304 break; 2305 if (IS_ERR_OR_NULL(mirror->mirror_ds)) 2306 continue; 2307 if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags)) 2308 continue; 2309 /* mirror refcount put in cleanup_layoutstats */ 2310 if (!atomic_inc_not_zero(&mirror->ref)) 2311 continue; 2312 dev = &mirror->mirror_ds->id_node; 2313 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE); 2314 devinfo->offset = 0; 2315 devinfo->length = NFS4_MAX_UINT64; 2316 spin_lock(&mirror->lock); 2317 devinfo->read_count = mirror->read_stat.io_stat.ops_completed; 2318 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed; 2319 devinfo->write_count = mirror->write_stat.io_stat.ops_completed; 2320 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed; 2321 spin_unlock(&mirror->lock); 2322 devinfo->layout_type = LAYOUT_FLEX_FILES; 2323 devinfo->ld_private.ops = &layoutstat_ops; 2324 devinfo->ld_private.data = mirror; 2325 2326 devinfo++; 2327 i++; 2328 } 2329 return i; 2330 } 2331 2332 static int 2333 ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args) 2334 { 2335 struct nfs4_flexfile_layout *ff_layout; 2336 const int dev_count = PNFS_LAYOUTSTATS_MAXDEV; 2337 2338 /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */ 2339 args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO); 2340 if (!args->devinfo) 2341 return -ENOMEM; 2342 2343 spin_lock(&args->inode->i_lock); 2344 ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout); 2345 args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr, 2346 &args->devinfo[0], dev_count); 2347 spin_unlock(&args->inode->i_lock); 2348 if (!args->num_dev) { 2349 kfree(args->devinfo); 2350 args->devinfo = NULL; 2351 return -ENOENT; 2352 } 2353 2354 return 0; 2355 } 2356 2357 static struct pnfs_layoutdriver_type flexfilelayout_type = { 2358 .id = LAYOUT_FLEX_FILES, 2359 .name = "LAYOUT_FLEX_FILES", 2360 .owner = THIS_MODULE, 2361 .alloc_layout_hdr = ff_layout_alloc_layout_hdr, 2362 .free_layout_hdr = ff_layout_free_layout_hdr, 2363 .alloc_lseg = ff_layout_alloc_lseg, 2364 .free_lseg = ff_layout_free_lseg, 2365 .add_lseg = ff_layout_add_lseg, 2366 .pg_read_ops = &ff_layout_pg_read_ops, 2367 .pg_write_ops = &ff_layout_pg_write_ops, 2368 .get_ds_info = ff_layout_get_ds_info, 2369 .free_deviceid_node = ff_layout_free_deviceid_node, 2370 .mark_request_commit = pnfs_layout_mark_request_commit, 2371 .clear_request_commit = pnfs_generic_clear_request_commit, 2372 .scan_commit_lists = pnfs_generic_scan_commit_lists, 2373 .recover_commit_reqs = pnfs_generic_recover_commit_reqs, 2374 .commit_pagelist = ff_layout_commit_pagelist, 2375 .read_pagelist = ff_layout_read_pagelist, 2376 .write_pagelist = ff_layout_write_pagelist, 2377 .alloc_deviceid_node = ff_layout_alloc_deviceid_node, 2378 .prepare_layoutreturn = ff_layout_prepare_layoutreturn, 2379 .sync = pnfs_nfs_generic_sync, 2380 .prepare_layoutstats = ff_layout_prepare_layoutstats, 2381 }; 2382 2383 static int __init nfs4flexfilelayout_init(void) 2384 { 2385 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n", 2386 __func__); 2387 if (!ff_zero_group) { 2388 ff_zero_group = groups_alloc(0); 2389 if (!ff_zero_group) 2390 return -ENOMEM; 2391 } 2392 return pnfs_register_layoutdriver(&flexfilelayout_type); 2393 } 2394 2395 static void __exit nfs4flexfilelayout_exit(void) 2396 { 2397 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n", 2398 __func__); 2399 pnfs_unregister_layoutdriver(&flexfilelayout_type); 2400 if (ff_zero_group) { 2401 put_group_info(ff_zero_group); 2402 ff_zero_group = NULL; 2403 } 2404 } 2405 2406 MODULE_ALIAS("nfs-layouttype4-4"); 2407 2408 MODULE_LICENSE("GPL"); 2409 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver"); 2410 2411 module_init(nfs4flexfilelayout_init); 2412 module_exit(nfs4flexfilelayout_exit); 2413