Loading...
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 | /*
* GPL HEADER START
*
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 only,
* as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but
* WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License version 2 for more details (a copy is included
* in the LICENSE file that accompanied this code).
*
* GPL HEADER END
*/
/*
* Copyright (c) 2012, 2015, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
* Lustre is a trademark of Sun Microsystems, Inc.
*
* lnet/lnet/lib-ptl.c
*
* portal & match routines
*
* Author: liang@whamcloud.com
*/
#define DEBUG_SUBSYSTEM S_LNET
#include "../../include/linux/lnet/lib-lnet.h"
/* NB: add /proc interfaces in upcoming patches */
int portal_rotor = LNET_PTL_ROTOR_HASH_RT;
module_param(portal_rotor, int, 0644);
MODULE_PARM_DESC(portal_rotor, "redirect PUTs to different cpu-partitions");
static int
lnet_ptl_match_type(unsigned int index, lnet_process_id_t match_id,
__u64 mbits, __u64 ignore_bits)
{
struct lnet_portal *ptl = the_lnet.ln_portals[index];
int unique;
unique = !ignore_bits &&
match_id.nid != LNET_NID_ANY &&
match_id.pid != LNET_PID_ANY;
LASSERT(!lnet_ptl_is_unique(ptl) || !lnet_ptl_is_wildcard(ptl));
/* prefer to check w/o any lock */
if (likely(lnet_ptl_is_unique(ptl) || lnet_ptl_is_wildcard(ptl)))
goto match;
/* unset, new portal */
lnet_ptl_lock(ptl);
/* check again with lock */
if (unlikely(lnet_ptl_is_unique(ptl) || lnet_ptl_is_wildcard(ptl))) {
lnet_ptl_unlock(ptl);
goto match;
}
/* still not set */
if (unique)
lnet_ptl_setopt(ptl, LNET_PTL_MATCH_UNIQUE);
else
lnet_ptl_setopt(ptl, LNET_PTL_MATCH_WILDCARD);
lnet_ptl_unlock(ptl);
return 1;
match:
if ((lnet_ptl_is_unique(ptl) && !unique) ||
(lnet_ptl_is_wildcard(ptl) && unique))
return 0;
return 1;
}
static void
lnet_ptl_enable_mt(struct lnet_portal *ptl, int cpt)
{
struct lnet_match_table *mtable = ptl->ptl_mtables[cpt];
int i;
/* with hold of both lnet_res_lock(cpt) and lnet_ptl_lock */
LASSERT(lnet_ptl_is_wildcard(ptl));
mtable->mt_enabled = 1;
ptl->ptl_mt_maps[ptl->ptl_mt_nmaps] = cpt;
for (i = ptl->ptl_mt_nmaps - 1; i >= 0; i--) {
LASSERT(ptl->ptl_mt_maps[i] != cpt);
if (ptl->ptl_mt_maps[i] < cpt)
break;
/* swap to order */
ptl->ptl_mt_maps[i + 1] = ptl->ptl_mt_maps[i];
ptl->ptl_mt_maps[i] = cpt;
}
ptl->ptl_mt_nmaps++;
}
static void
lnet_ptl_disable_mt(struct lnet_portal *ptl, int cpt)
{
struct lnet_match_table *mtable = ptl->ptl_mtables[cpt];
int i;
/* with hold of both lnet_res_lock(cpt) and lnet_ptl_lock */
LASSERT(lnet_ptl_is_wildcard(ptl));
if (LNET_CPT_NUMBER == 1)
return; /* never disable the only match-table */
mtable->mt_enabled = 0;
LASSERT(ptl->ptl_mt_nmaps > 0 &&
ptl->ptl_mt_nmaps <= LNET_CPT_NUMBER);
/* remove it from mt_maps */
ptl->ptl_mt_nmaps--;
for (i = 0; i < ptl->ptl_mt_nmaps; i++) {
if (ptl->ptl_mt_maps[i] >= cpt) /* overwrite it */
ptl->ptl_mt_maps[i] = ptl->ptl_mt_maps[i + 1];
}
}
static int
lnet_try_match_md(lnet_libmd_t *md,
struct lnet_match_info *info, struct lnet_msg *msg)
{
/*
* ALWAYS called holding the lnet_res_lock, and can't lnet_res_unlock;
* lnet_match_blocked_msg() relies on this to avoid races
*/
unsigned int offset;
unsigned int mlength;
lnet_me_t *me = md->md_me;
/* MD exhausted */
if (lnet_md_exhausted(md))
return LNET_MATCHMD_NONE | LNET_MATCHMD_EXHAUSTED;
/* mismatched MD op */
if (!(md->md_options & info->mi_opc))
return LNET_MATCHMD_NONE;
/* mismatched ME nid/pid? */
if (me->me_match_id.nid != LNET_NID_ANY &&
me->me_match_id.nid != info->mi_id.nid)
return LNET_MATCHMD_NONE;
if (me->me_match_id.pid != LNET_PID_ANY &&
me->me_match_id.pid != info->mi_id.pid)
return LNET_MATCHMD_NONE;
/* mismatched ME matchbits? */
if ((me->me_match_bits ^ info->mi_mbits) & ~me->me_ignore_bits)
return LNET_MATCHMD_NONE;
/* Hurrah! This _is_ a match; check it out... */
if (!(md->md_options & LNET_MD_MANAGE_REMOTE))
offset = md->md_offset;
else
offset = info->mi_roffset;
if (md->md_options & LNET_MD_MAX_SIZE) {
mlength = md->md_max_size;
LASSERT(md->md_offset + mlength <= md->md_length);
} else {
mlength = md->md_length - offset;
}
if (info->mi_rlength <= mlength) { /* fits in allowed space */
mlength = info->mi_rlength;
} else if (!(md->md_options & LNET_MD_TRUNCATE)) {
/* this packet _really_ is too big */
CERROR("Matching packet from %s, match %llu length %d too big: %d left, %d allowed\n",
libcfs_id2str(info->mi_id), info->mi_mbits,
info->mi_rlength, md->md_length - offset, mlength);
return LNET_MATCHMD_DROP;
}
/* Commit to this ME/MD */
CDEBUG(D_NET, "Incoming %s index %x from %s of length %d/%d into md %#llx [%d] + %d\n",
(info->mi_opc == LNET_MD_OP_PUT) ? "put" : "get",
info->mi_portal, libcfs_id2str(info->mi_id), mlength,
info->mi_rlength, md->md_lh.lh_cookie, md->md_niov, offset);
lnet_msg_attach_md(msg, md, offset, mlength);
md->md_offset = offset + mlength;
if (!lnet_md_exhausted(md))
return LNET_MATCHMD_OK;
/*
* Auto-unlink NOW, so the ME gets unlinked if required.
* We bumped md->md_refcount above so the MD just gets flagged
* for unlink when it is finalized.
*/
if (md->md_flags & LNET_MD_FLAG_AUTO_UNLINK)
lnet_md_unlink(md);
return LNET_MATCHMD_OK | LNET_MATCHMD_EXHAUSTED;
}
static struct lnet_match_table *
lnet_match2mt(struct lnet_portal *ptl, lnet_process_id_t id, __u64 mbits)
{
if (LNET_CPT_NUMBER == 1)
return ptl->ptl_mtables[0]; /* the only one */
/* if it's a unique portal, return match-table hashed by NID */
return lnet_ptl_is_unique(ptl) ?
ptl->ptl_mtables[lnet_cpt_of_nid(id.nid)] : NULL;
}
struct lnet_match_table *
lnet_mt_of_attach(unsigned int index, lnet_process_id_t id,
__u64 mbits, __u64 ignore_bits, lnet_ins_pos_t pos)
{
struct lnet_portal *ptl;
struct lnet_match_table *mtable;
/* NB: called w/o lock */
LASSERT(index < the_lnet.ln_nportals);
if (!lnet_ptl_match_type(index, id, mbits, ignore_bits))
return NULL;
ptl = the_lnet.ln_portals[index];
mtable = lnet_match2mt(ptl, id, mbits);
if (mtable) /* unique portal or only one match-table */
return mtable;
/* it's a wildcard portal */
switch (pos) {
default:
return NULL;
case LNET_INS_BEFORE:
case LNET_INS_AFTER:
/*
* posted by no affinity thread, always hash to specific
* match-table to avoid buffer stealing which is heavy
*/
return ptl->ptl_mtables[ptl->ptl_index % LNET_CPT_NUMBER];
case LNET_INS_LOCAL:
/* posted by cpu-affinity thread */
return ptl->ptl_mtables[lnet_cpt_current()];
}
}
static struct lnet_match_table *
lnet_mt_of_match(struct lnet_match_info *info, struct lnet_msg *msg)
{
struct lnet_match_table *mtable;
struct lnet_portal *ptl;
unsigned int nmaps;
unsigned int rotor;
unsigned int cpt;
bool routed;
/* NB: called w/o lock */
LASSERT(info->mi_portal < the_lnet.ln_nportals);
ptl = the_lnet.ln_portals[info->mi_portal];
LASSERT(lnet_ptl_is_wildcard(ptl) || lnet_ptl_is_unique(ptl));
mtable = lnet_match2mt(ptl, info->mi_id, info->mi_mbits);
if (mtable)
return mtable;
/* it's a wildcard portal */
routed = LNET_NIDNET(msg->msg_hdr.src_nid) !=
LNET_NIDNET(msg->msg_hdr.dest_nid);
if (portal_rotor == LNET_PTL_ROTOR_OFF ||
(portal_rotor != LNET_PTL_ROTOR_ON && !routed)) {
cpt = lnet_cpt_current();
if (ptl->ptl_mtables[cpt]->mt_enabled)
return ptl->ptl_mtables[cpt];
}
rotor = ptl->ptl_rotor++; /* get round-robin factor */
if (portal_rotor == LNET_PTL_ROTOR_HASH_RT && routed)
cpt = lnet_cpt_of_nid(msg->msg_hdr.src_nid);
else
cpt = rotor % LNET_CPT_NUMBER;
if (!ptl->ptl_mtables[cpt]->mt_enabled) {
/* is there any active entry for this portal? */
nmaps = ptl->ptl_mt_nmaps;
/* map to an active mtable to avoid heavy "stealing" */
if (nmaps) {
/*
* NB: there is possibility that ptl_mt_maps is being
* changed because we are not under protection of
* lnet_ptl_lock, but it shouldn't hurt anything
*/
cpt = ptl->ptl_mt_maps[rotor % nmaps];
}
}
return ptl->ptl_mtables[cpt];
}
static int
lnet_mt_test_exhausted(struct lnet_match_table *mtable, int pos)
{
__u64 *bmap;
int i;
if (!lnet_ptl_is_wildcard(the_lnet.ln_portals[mtable->mt_portal]))
return 0;
if (pos < 0) { /* check all bits */
for (i = 0; i < LNET_MT_EXHAUSTED_BMAP; i++) {
if (mtable->mt_exhausted[i] != (__u64)(-1))
return 0;
}
return 1;
}
LASSERT(pos <= LNET_MT_HASH_IGNORE);
/* mtable::mt_mhash[pos] is marked as exhausted or not */
bmap = &mtable->mt_exhausted[pos >> LNET_MT_BITS_U64];
pos &= (1 << LNET_MT_BITS_U64) - 1;
return (*bmap & (1ULL << pos));
}
static void
lnet_mt_set_exhausted(struct lnet_match_table *mtable, int pos, int exhausted)
{
__u64 *bmap;
LASSERT(lnet_ptl_is_wildcard(the_lnet.ln_portals[mtable->mt_portal]));
LASSERT(pos <= LNET_MT_HASH_IGNORE);
/* set mtable::mt_mhash[pos] as exhausted/non-exhausted */
bmap = &mtable->mt_exhausted[pos >> LNET_MT_BITS_U64];
pos &= (1 << LNET_MT_BITS_U64) - 1;
if (!exhausted)
*bmap &= ~(1ULL << pos);
else
*bmap |= 1ULL << pos;
}
struct list_head *
lnet_mt_match_head(struct lnet_match_table *mtable,
lnet_process_id_t id, __u64 mbits)
{
struct lnet_portal *ptl = the_lnet.ln_portals[mtable->mt_portal];
unsigned long hash = mbits;
if (!lnet_ptl_is_wildcard(ptl)) {
hash += id.nid + id.pid;
LASSERT(lnet_ptl_is_unique(ptl));
hash = hash_long(hash, LNET_MT_HASH_BITS);
}
return &mtable->mt_mhash[hash & LNET_MT_HASH_MASK];
}
int
lnet_mt_match_md(struct lnet_match_table *mtable,
struct lnet_match_info *info, struct lnet_msg *msg)
{
struct list_head *head;
lnet_me_t *me;
lnet_me_t *tmp;
int exhausted = 0;
int rc;
/* any ME with ignore bits? */
if (!list_empty(&mtable->mt_mhash[LNET_MT_HASH_IGNORE]))
head = &mtable->mt_mhash[LNET_MT_HASH_IGNORE];
else
head = lnet_mt_match_head(mtable, info->mi_id, info->mi_mbits);
again:
/* NB: only wildcard portal needs to return LNET_MATCHMD_EXHAUSTED */
if (lnet_ptl_is_wildcard(the_lnet.ln_portals[mtable->mt_portal]))
exhausted = LNET_MATCHMD_EXHAUSTED;
list_for_each_entry_safe(me, tmp, head, me_list) {
/* ME attached but MD not attached yet */
if (!me->me_md)
continue;
LASSERT(me == me->me_md->md_me);
rc = lnet_try_match_md(me->me_md, info, msg);
if (!(rc & LNET_MATCHMD_EXHAUSTED))
exhausted = 0; /* mlist is not empty */
if (rc & LNET_MATCHMD_FINISH) {
/*
* don't return EXHAUSTED bit because we don't know
* whether the mlist is empty or not
*/
return rc & ~LNET_MATCHMD_EXHAUSTED;
}
}
if (exhausted == LNET_MATCHMD_EXHAUSTED) { /* @head is exhausted */
lnet_mt_set_exhausted(mtable, head - mtable->mt_mhash, 1);
if (!lnet_mt_test_exhausted(mtable, -1))
exhausted = 0;
}
if (!exhausted && head == &mtable->mt_mhash[LNET_MT_HASH_IGNORE]) {
head = lnet_mt_match_head(mtable, info->mi_id, info->mi_mbits);
goto again; /* re-check MEs w/o ignore-bits */
}
if (info->mi_opc == LNET_MD_OP_GET ||
!lnet_ptl_is_lazy(the_lnet.ln_portals[info->mi_portal]))
return exhausted | LNET_MATCHMD_DROP;
return exhausted | LNET_MATCHMD_NONE;
}
static int
lnet_ptl_match_early(struct lnet_portal *ptl, struct lnet_msg *msg)
{
int rc;
/*
* message arrived before any buffer posting on this portal,
* simply delay or drop this message
*/
if (likely(lnet_ptl_is_wildcard(ptl) || lnet_ptl_is_unique(ptl)))
return 0;
lnet_ptl_lock(ptl);
/* check it again with hold of lock */
if (lnet_ptl_is_wildcard(ptl) || lnet_ptl_is_unique(ptl)) {
lnet_ptl_unlock(ptl);
return 0;
}
if (lnet_ptl_is_lazy(ptl)) {
if (msg->msg_rx_ready_delay) {
msg->msg_rx_delayed = 1;
list_add_tail(&msg->msg_list,
&ptl->ptl_msg_delayed);
}
rc = LNET_MATCHMD_NONE;
} else {
rc = LNET_MATCHMD_DROP;
}
lnet_ptl_unlock(ptl);
return rc;
}
static int
lnet_ptl_match_delay(struct lnet_portal *ptl,
struct lnet_match_info *info, struct lnet_msg *msg)
{
int first = ptl->ptl_mt_maps[0]; /* read w/o lock */
int rc = 0;
int i;
/**
* Steal buffer from other CPTs, and delay msg if nothing to
* steal. This function is more expensive than a regular
* match, but we don't expect it can happen a lot. The return
* code contains one of LNET_MATCHMD_OK, LNET_MATCHMD_DROP, or
* LNET_MATCHMD_NONE.
*/
LASSERT(lnet_ptl_is_wildcard(ptl));
for (i = 0; i < LNET_CPT_NUMBER; i++) {
struct lnet_match_table *mtable;
int cpt;
cpt = (first + i) % LNET_CPT_NUMBER;
mtable = ptl->ptl_mtables[cpt];
if (i && i != LNET_CPT_NUMBER - 1 && !mtable->mt_enabled)
continue;
lnet_res_lock(cpt);
lnet_ptl_lock(ptl);
if (!i) {
/* The first try, add to stealing list. */
list_add_tail(&msg->msg_list,
&ptl->ptl_msg_stealing);
}
if (!list_empty(&msg->msg_list)) {
/* On stealing list. */
rc = lnet_mt_match_md(mtable, info, msg);
if ((rc & LNET_MATCHMD_EXHAUSTED) &&
mtable->mt_enabled)
lnet_ptl_disable_mt(ptl, cpt);
if (rc & LNET_MATCHMD_FINISH) {
/* Match found, remove from stealing list. */
list_del_init(&msg->msg_list);
} else if (i == LNET_CPT_NUMBER - 1 || /* (1) */
!ptl->ptl_mt_nmaps || /* (2) */
(ptl->ptl_mt_nmaps == 1 && /* (3) */
ptl->ptl_mt_maps[0] == cpt)) {
/**
* No match found, and this is either
* (1) the last cpt to check, or
* (2) there is no active cpt, or
* (3) this is the only active cpt.
* There is nothing to steal: delay or
* drop the message.
*/
list_del_init(&msg->msg_list);
if (lnet_ptl_is_lazy(ptl)) {
msg->msg_rx_delayed = 1;
list_add_tail(&msg->msg_list,
&ptl->ptl_msg_delayed);
rc = LNET_MATCHMD_NONE;
} else {
rc = LNET_MATCHMD_DROP;
}
} else {
/* Do another iteration. */
rc = 0;
}
} else {
/**
* No longer on stealing list: another thread
* matched the message in lnet_ptl_attach_md().
* We are now expected to handle the message.
*/
rc = !msg->msg_md ?
LNET_MATCHMD_DROP : LNET_MATCHMD_OK;
}
lnet_ptl_unlock(ptl);
lnet_res_unlock(cpt);
/**
* Note that test (1) above ensures that we always
* exit the loop through this break statement.
*
* LNET_MATCHMD_NONE means msg was added to the
* delayed queue, and we may no longer reference it
* after lnet_ptl_unlock() and lnet_res_unlock().
*/
if (rc & (LNET_MATCHMD_FINISH | LNET_MATCHMD_NONE))
break;
}
return rc;
}
int
lnet_ptl_match_md(struct lnet_match_info *info, struct lnet_msg *msg)
{
struct lnet_match_table *mtable;
struct lnet_portal *ptl;
int rc;
CDEBUG(D_NET, "Request from %s of length %d into portal %d MB=%#llx\n",
libcfs_id2str(info->mi_id), info->mi_rlength, info->mi_portal,
info->mi_mbits);
if (info->mi_portal >= the_lnet.ln_nportals) {
CERROR("Invalid portal %d not in [0-%d]\n",
info->mi_portal, the_lnet.ln_nportals);
return LNET_MATCHMD_DROP;
}
ptl = the_lnet.ln_portals[info->mi_portal];
rc = lnet_ptl_match_early(ptl, msg);
if (rc) /* matched or delayed early message */
return rc;
mtable = lnet_mt_of_match(info, msg);
lnet_res_lock(mtable->mt_cpt);
if (the_lnet.ln_shutdown) {
rc = LNET_MATCHMD_DROP;
goto out1;
}
rc = lnet_mt_match_md(mtable, info, msg);
if ((rc & LNET_MATCHMD_EXHAUSTED) && mtable->mt_enabled) {
lnet_ptl_lock(ptl);
lnet_ptl_disable_mt(ptl, mtable->mt_cpt);
lnet_ptl_unlock(ptl);
}
if (rc & LNET_MATCHMD_FINISH) /* matched or dropping */
goto out1;
if (!msg->msg_rx_ready_delay)
goto out1;
LASSERT(lnet_ptl_is_lazy(ptl));
LASSERT(!msg->msg_rx_delayed);
/* NB: we don't expect "delay" can happen a lot */
if (lnet_ptl_is_unique(ptl) || LNET_CPT_NUMBER == 1) {
lnet_ptl_lock(ptl);
msg->msg_rx_delayed = 1;
list_add_tail(&msg->msg_list, &ptl->ptl_msg_delayed);
lnet_ptl_unlock(ptl);
lnet_res_unlock(mtable->mt_cpt);
rc = LNET_MATCHMD_NONE;
} else {
lnet_res_unlock(mtable->mt_cpt);
rc = lnet_ptl_match_delay(ptl, info, msg);
}
/* LNET_MATCHMD_NONE means msg was added to the delay queue */
if (rc & LNET_MATCHMD_NONE) {
CDEBUG(D_NET,
"Delaying %s from %s ptl %d MB %#llx off %d len %d\n",
info->mi_opc == LNET_MD_OP_PUT ? "PUT" : "GET",
libcfs_id2str(info->mi_id), info->mi_portal,
info->mi_mbits, info->mi_roffset, info->mi_rlength);
}
goto out0;
out1:
lnet_res_unlock(mtable->mt_cpt);
out0:
/* EXHAUSTED bit is only meaningful for internal functions */
return rc & ~LNET_MATCHMD_EXHAUSTED;
}
void
lnet_ptl_detach_md(lnet_me_t *me, lnet_libmd_t *md)
{
LASSERT(me->me_md == md && md->md_me == me);
me->me_md = NULL;
md->md_me = NULL;
}
/* called with lnet_res_lock held */
void
lnet_ptl_attach_md(lnet_me_t *me, lnet_libmd_t *md,
struct list_head *matches, struct list_head *drops)
{
struct lnet_portal *ptl = the_lnet.ln_portals[me->me_portal];
struct lnet_match_table *mtable;
struct list_head *head;
lnet_msg_t *tmp;
lnet_msg_t *msg;
int exhausted = 0;
int cpt;
LASSERT(!md->md_refcount); /* a brand new MD */
me->me_md = md;
md->md_me = me;
cpt = lnet_cpt_of_cookie(md->md_lh.lh_cookie);
mtable = ptl->ptl_mtables[cpt];
if (list_empty(&ptl->ptl_msg_stealing) &&
list_empty(&ptl->ptl_msg_delayed) &&
!lnet_mt_test_exhausted(mtable, me->me_pos))
return;
lnet_ptl_lock(ptl);
head = &ptl->ptl_msg_stealing;
again:
list_for_each_entry_safe(msg, tmp, head, msg_list) {
struct lnet_match_info info;
struct lnet_hdr *hdr;
int rc;
LASSERT(msg->msg_rx_delayed || head == &ptl->ptl_msg_stealing);
hdr = &msg->msg_hdr;
info.mi_id.nid = hdr->src_nid;
info.mi_id.pid = hdr->src_pid;
info.mi_opc = LNET_MD_OP_PUT;
info.mi_portal = hdr->msg.put.ptl_index;
info.mi_rlength = hdr->payload_length;
info.mi_roffset = hdr->msg.put.offset;
info.mi_mbits = hdr->msg.put.match_bits;
rc = lnet_try_match_md(md, &info, msg);
exhausted = (rc & LNET_MATCHMD_EXHAUSTED);
if (rc & LNET_MATCHMD_NONE) {
if (exhausted)
break;
continue;
}
/* Hurrah! This _is_ a match */
LASSERT(rc & LNET_MATCHMD_FINISH);
list_del_init(&msg->msg_list);
if (head == &ptl->ptl_msg_stealing) {
if (exhausted)
break;
/* stealing thread will handle the message */
continue;
}
if (rc & LNET_MATCHMD_OK) {
list_add_tail(&msg->msg_list, matches);
CDEBUG(D_NET, "Resuming delayed PUT from %s portal %d match %llu offset %d length %d.\n",
libcfs_id2str(info.mi_id),
info.mi_portal, info.mi_mbits,
info.mi_roffset, info.mi_rlength);
} else {
list_add_tail(&msg->msg_list, drops);
}
if (exhausted)
break;
}
if (!exhausted && head == &ptl->ptl_msg_stealing) {
head = &ptl->ptl_msg_delayed;
goto again;
}
if (lnet_ptl_is_wildcard(ptl) && !exhausted) {
lnet_mt_set_exhausted(mtable, me->me_pos, 0);
if (!mtable->mt_enabled)
lnet_ptl_enable_mt(ptl, cpt);
}
lnet_ptl_unlock(ptl);
}
static void
lnet_ptl_cleanup(struct lnet_portal *ptl)
{
struct lnet_match_table *mtable;
int i;
if (!ptl->ptl_mtables) /* uninitialized portal */
return;
LASSERT(list_empty(&ptl->ptl_msg_delayed));
LASSERT(list_empty(&ptl->ptl_msg_stealing));
cfs_percpt_for_each(mtable, i, ptl->ptl_mtables) {
struct list_head *mhash;
lnet_me_t *me;
int j;
if (!mtable->mt_mhash) /* uninitialized match-table */
continue;
mhash = mtable->mt_mhash;
/* cleanup ME */
for (j = 0; j < LNET_MT_HASH_SIZE + 1; j++) {
while (!list_empty(&mhash[j])) {
me = list_entry(mhash[j].next,
lnet_me_t, me_list);
CERROR("Active ME %p on exit\n", me);
list_del(&me->me_list);
lnet_me_free(me);
}
}
/* the extra entry is for MEs with ignore bits */
LIBCFS_FREE(mhash, sizeof(*mhash) * (LNET_MT_HASH_SIZE + 1));
}
cfs_percpt_free(ptl->ptl_mtables);
ptl->ptl_mtables = NULL;
}
static int
lnet_ptl_setup(struct lnet_portal *ptl, int index)
{
struct lnet_match_table *mtable;
struct list_head *mhash;
int i;
int j;
ptl->ptl_mtables = cfs_percpt_alloc(lnet_cpt_table(),
sizeof(struct lnet_match_table));
if (!ptl->ptl_mtables) {
CERROR("Failed to create match table for portal %d\n", index);
return -ENOMEM;
}
ptl->ptl_index = index;
INIT_LIST_HEAD(&ptl->ptl_msg_delayed);
INIT_LIST_HEAD(&ptl->ptl_msg_stealing);
spin_lock_init(&ptl->ptl_lock);
cfs_percpt_for_each(mtable, i, ptl->ptl_mtables) {
/* the extra entry is for MEs with ignore bits */
LIBCFS_CPT_ALLOC(mhash, lnet_cpt_table(), i,
sizeof(*mhash) * (LNET_MT_HASH_SIZE + 1));
if (!mhash) {
CERROR("Failed to create match hash for portal %d\n",
index);
goto failed;
}
memset(&mtable->mt_exhausted[0], -1,
sizeof(mtable->mt_exhausted[0]) *
LNET_MT_EXHAUSTED_BMAP);
mtable->mt_mhash = mhash;
for (j = 0; j < LNET_MT_HASH_SIZE + 1; j++)
INIT_LIST_HEAD(&mhash[j]);
mtable->mt_portal = index;
mtable->mt_cpt = i;
}
return 0;
failed:
lnet_ptl_cleanup(ptl);
return -ENOMEM;
}
void
lnet_portals_destroy(void)
{
int i;
if (!the_lnet.ln_portals)
return;
for (i = 0; i < the_lnet.ln_nportals; i++)
lnet_ptl_cleanup(the_lnet.ln_portals[i]);
cfs_array_free(the_lnet.ln_portals);
the_lnet.ln_portals = NULL;
}
int
lnet_portals_create(void)
{
int size;
int i;
size = offsetof(struct lnet_portal, ptl_mt_maps[LNET_CPT_NUMBER]);
the_lnet.ln_nportals = MAX_PORTALS;
the_lnet.ln_portals = cfs_array_alloc(the_lnet.ln_nportals, size);
if (!the_lnet.ln_portals) {
CERROR("Failed to allocate portals table\n");
return -ENOMEM;
}
for (i = 0; i < the_lnet.ln_nportals; i++) {
if (lnet_ptl_setup(the_lnet.ln_portals[i], i)) {
lnet_portals_destroy();
return -ENOMEM;
}
}
return 0;
}
/**
* Turn on the lazy portal attribute. Use with caution!
*
* This portal attribute only affects incoming PUT requests to the portal,
* and is off by default. By default, if there's no matching MD for an
* incoming PUT request, it is simply dropped. With the lazy attribute on,
* such requests are queued indefinitely until either a matching MD is
* posted to the portal or the lazy attribute is turned off.
*
* It would prevent dropped requests, however it should be regarded as the
* last line of defense - i.e. users must keep a close watch on active
* buffers on a lazy portal and once it becomes too low post more buffers as
* soon as possible. This is because delayed requests usually have detrimental
* effects on underlying network connections. A few delayed requests often
* suffice to bring an underlying connection to a complete halt, due to flow
* control mechanisms.
*
* There's also a DOS attack risk. If users don't post match-all MDs on a
* lazy portal, a malicious peer can easily stop a service by sending some
* PUT requests with match bits that won't match any MD. A routed server is
* especially vulnerable since the connections to its neighbor routers are
* shared among all clients.
*
* \param portal Index of the portal to enable the lazy attribute on.
*
* \retval 0 On success.
* \retval -EINVAL If \a portal is not a valid index.
*/
int
LNetSetLazyPortal(int portal)
{
struct lnet_portal *ptl;
if (portal < 0 || portal >= the_lnet.ln_nportals)
return -EINVAL;
CDEBUG(D_NET, "Setting portal %d lazy\n", portal);
ptl = the_lnet.ln_portals[portal];
lnet_res_lock(LNET_LOCK_EX);
lnet_ptl_lock(ptl);
lnet_ptl_setopt(ptl, LNET_PTL_LAZY);
lnet_ptl_unlock(ptl);
lnet_res_unlock(LNET_LOCK_EX);
return 0;
}
EXPORT_SYMBOL(LNetSetLazyPortal);
int
lnet_clear_lazy_portal(struct lnet_ni *ni, int portal, char *reason)
{
struct lnet_portal *ptl;
LIST_HEAD(zombies);
if (portal < 0 || portal >= the_lnet.ln_nportals)
return -EINVAL;
ptl = the_lnet.ln_portals[portal];
lnet_res_lock(LNET_LOCK_EX);
lnet_ptl_lock(ptl);
if (!lnet_ptl_is_lazy(ptl)) {
lnet_ptl_unlock(ptl);
lnet_res_unlock(LNET_LOCK_EX);
return 0;
}
if (ni) {
struct lnet_msg *msg, *tmp;
/* grab all messages which are on the NI passed in */
list_for_each_entry_safe(msg, tmp, &ptl->ptl_msg_delayed,
msg_list) {
if (msg->msg_rxpeer->lp_ni == ni)
list_move(&msg->msg_list, &zombies);
}
} else {
if (the_lnet.ln_shutdown)
CWARN("Active lazy portal %d on exit\n", portal);
else
CDEBUG(D_NET, "clearing portal %d lazy\n", portal);
/* grab all the blocked messages atomically */
list_splice_init(&ptl->ptl_msg_delayed, &zombies);
lnet_ptl_unsetopt(ptl, LNET_PTL_LAZY);
}
lnet_ptl_unlock(ptl);
lnet_res_unlock(LNET_LOCK_EX);
lnet_drop_delayed_msg_list(&zombies, reason);
return 0;
}
/**
* Turn off the lazy portal attribute. Delayed requests on the portal,
* if any, will be all dropped when this function returns.
*
* \param portal Index of the portal to disable the lazy attribute on.
*
* \retval 0 On success.
* \retval -EINVAL If \a portal is not a valid index.
*/
int
LNetClearLazyPortal(int portal)
{
return lnet_clear_lazy_portal(NULL, portal,
"Clearing lazy portal attr");
}
EXPORT_SYMBOL(LNetClearLazyPortal);
|