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