libcfs/libcfs/tracefile.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * libcfs/libcfs/tracefile.c
  37  *
  38  * Author: Zach Brown <zab@clusterfs.com>
  39  * Author: Phil Schwan <phil@clusterfs.com>
  40  */
  41
  42
  43 #define DEBUG_SUBSYSTEM S_LNET
  44 #define LUSTRE_TRACEFILE_PRIVATE
  45 #include "tracefile.h"
  46
  47 #include <libcfs/libcfs.h>
  48
  49 /* XXX move things up to the top, comment */
  50 union cfs_trace_data_union (*cfs_trace_data[TCD_MAX_TYPES])[CFS_NR_CPUS] __cacheline_aligned;
  51
  52 char cfs_tracefile[TRACEFILE_NAME_SIZE];
  53 long long cfs_tracefile_size = CFS_TRACEFILE_SIZE;
  54 static struct tracefiled_ctl trace_tctl;
  55 cfs_semaphore_t cfs_trace_thread_sem;
  56 static int thread_running = 0;
  57
  58 cfs_atomic_t cfs_tage_allocated = CFS_ATOMIC_INIT(0);
  59
  60 static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
  61                                          struct cfs_trace_cpu_data *tcd);
  62
  63 static inline struct cfs_trace_page *
  64 cfs_tage_from_list(cfs_list_t *list)
  65 {
  66         return cfs_list_entry(list, struct cfs_trace_page, linkage);
  67 }
  68
  69 static struct cfs_trace_page *cfs_tage_alloc(int gfp)
  70 {
  71         cfs_page_t            *page;
  72         struct cfs_trace_page *tage;
  73
  74         /* My caller is trying to free memory */
  75         if (!cfs_in_interrupt() && cfs_memory_pressure_get())
  76                 return NULL;
  77
  78         /*
  79          * Don't spam console with allocation failures: they will be reported
  80          * by upper layer anyway.
  81          */
  82         gfp |= CFS_ALLOC_NOWARN;
  83         page = cfs_alloc_page(gfp);
  84         if (page == NULL)
  85                 return NULL;
  86
  87         tage = cfs_alloc(sizeof(*tage), gfp);
  88         if (tage == NULL) {
  89                 cfs_free_page(page);
  90                 return NULL;
  91         }
  92
  93         tage->page = page;
  94         cfs_atomic_inc(&cfs_tage_allocated);
  95         return tage;
  96 }
  97
  98 static void cfs_tage_free(struct cfs_trace_page *tage)
  99 {
 100         __LASSERT(tage != NULL);
 101         __LASSERT(tage->page != NULL);
 102
 103         cfs_free_page(tage->page);
 104         cfs_free(tage);
 105         cfs_atomic_dec(&cfs_tage_allocated);
 106 }
 107
 108 static void cfs_tage_to_tail(struct cfs_trace_page *tage,
 109                              cfs_list_t *queue)
 110 {
 111         __LASSERT(tage != NULL);
 112         __LASSERT(queue != NULL);
 113
 114         cfs_list_move_tail(&tage->linkage, queue);
 115 }
 116
 117 int cfs_trace_refill_stock(struct cfs_trace_cpu_data *tcd, int gfp,
 118                            cfs_list_t *stock)
 119 {
 120         int i;
 121
 122         /*
 123          * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
 124          * from here: this will lead to infinite recursion.
 125          */
 126
 127         for (i = 0; i + tcd->tcd_cur_stock_pages < TCD_STOCK_PAGES ; ++ i) {
 128                 struct cfs_trace_page *tage;
 129
 130                 tage = cfs_tage_alloc(gfp);
 131                 if (tage == NULL)
 132                         break;
 133                 cfs_list_add_tail(&tage->linkage, stock);
 134         }
 135         return i;
 136 }
 137
 138 /* return a page that has 'len' bytes left at the end */
 139 static struct cfs_trace_page *
 140 cfs_trace_get_tage_try(struct cfs_trace_cpu_data *tcd, unsigned long len)
 141 {
 142         struct cfs_trace_page *tage;
 143
 144         if (tcd->tcd_cur_pages > 0) {
 145                 __LASSERT(!cfs_list_empty(&tcd->tcd_pages));
 146                 tage = cfs_tage_from_list(tcd->tcd_pages.prev);
 147                 if (tage->used + len <= CFS_PAGE_SIZE)
 148                         return tage;
 149         }
 150
 151         if (tcd->tcd_cur_pages < tcd->tcd_max_pages) {
 152                 if (tcd->tcd_cur_stock_pages > 0) {
 153                         tage = cfs_tage_from_list(tcd->tcd_stock_pages.prev);
 154                         -- tcd->tcd_cur_stock_pages;
 155                         cfs_list_del_init(&tage->linkage);
 156                 } else {
 157                         tage = cfs_tage_alloc(CFS_ALLOC_ATOMIC);
 158                         if (tage == NULL) {
 159                                 if (printk_ratelimit())
 160                                         printk(CFS_KERN_WARNING
 161                                                "cannot allocate a tage (%ld)\n",
 162                                        tcd->tcd_cur_pages);
 163                                 return NULL;
 164                         }
 165                 }
 166
 167                 tage->used = 0;
 168                 tage->cpu = cfs_smp_processor_id();
 169                 tage->type = tcd->tcd_type;
 170                 cfs_list_add_tail(&tage->linkage, &tcd->tcd_pages);
 171                 tcd->tcd_cur_pages++;
 172
 173                 if (tcd->tcd_cur_pages > 8 && thread_running) {
 174                         struct tracefiled_ctl *tctl = &trace_tctl;
 175                         /*
 176                          * wake up tracefiled to process some pages.
 177                          */
 178                         cfs_waitq_signal(&tctl->tctl_waitq);
 179                 }
 180                 return tage;
 181         }
 182         return NULL;
 183 }
 184
 185 static void cfs_tcd_shrink(struct cfs_trace_cpu_data *tcd)
 186 {
 187         int pgcount = tcd->tcd_cur_pages / 10;
 188         struct page_collection pc;
 189         struct cfs_trace_page *tage;
 190         struct cfs_trace_page *tmp;
 191
 192         /*
 193          * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
 194          * from here: this will lead to infinite recursion.
 195          */
 196
 197         if (printk_ratelimit())
 198                 printk(CFS_KERN_WARNING "debug daemon buffer overflowed; "
 199                        "discarding 10%% of pages (%d of %ld)\n",
 200                        pgcount + 1, tcd->tcd_cur_pages);
 201
 202         CFS_INIT_LIST_HEAD(&pc.pc_pages);
 203         cfs_spin_lock_init(&pc.pc_lock);
 204
 205         cfs_list_for_each_entry_safe_typed(tage, tmp, &tcd->tcd_pages,
 206                                            struct cfs_trace_page, linkage) {
 207                 if (pgcount-- == 0)
 208                         break;
 209
 210                 cfs_list_move_tail(&tage->linkage, &pc.pc_pages);
 211                 tcd->tcd_cur_pages--;
 212         }
 213         put_pages_on_tcd_daemon_list(&pc, tcd);
 214 }
 215
 216 /* return a page that has 'len' bytes left at the end */
 217 static struct cfs_trace_page *cfs_trace_get_tage(struct cfs_trace_cpu_data *tcd,
 218                                                  unsigned long len)
 219 {
 220         struct cfs_trace_page *tage;
 221
 222         /*
 223          * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
 224          * from here: this will lead to infinite recursion.
 225          */
 226
 227         if (len > CFS_PAGE_SIZE) {
 228                 printk(CFS_KERN_ERR
 229                        "cowardly refusing to write %lu bytes in a page\n", len);
 230                 return NULL;
 231         }
 232
 233         tage = cfs_trace_get_tage_try(tcd, len);
 234         if (tage != NULL)
 235                 return tage;
 236         if (thread_running)
 237                 cfs_tcd_shrink(tcd);
 238         if (tcd->tcd_cur_pages > 0) {
 239                 tage = cfs_tage_from_list(tcd->tcd_pages.next);
 240                 tage->used = 0;
 241                 cfs_tage_to_tail(tage, &tcd->tcd_pages);
 242         }
 243         return tage;
 244 }
 245
 246 int libcfs_debug_vmsg2(cfs_debug_limit_state_t *cdls, int subsys, int mask,
 247                        const char *file, const char *fn, const int line,
 248                        const char *format1, va_list args,
 249                        const char *format2, ...)
 250 {
 251         struct cfs_trace_cpu_data *tcd = NULL;
 252         struct ptldebug_header     header = {0};
 253         struct cfs_trace_page     *tage;
 254         /* string_buf is used only if tcd != NULL, and is always set then */
 255         char                      *string_buf = NULL;
 256         char                      *debug_buf;
 257         int                        known_size;
 258         int                        needed = 85; /* average message length */
 259         int                        max_nob;
 260         va_list                    ap;
 261         int                        depth;
 262         int                        i;
 263         int                        remain;
 264
 265         if (strchr(file, '/'))
 266                 file = strrchr(file, '/') + 1;
 267
 268         tcd = cfs_trace_get_tcd();
 269
 270         /* cfs_trace_get_tcd() grabs a lock, which disables preemption and
 271          * pins us to a particular CPU.  This avoids an smp_processor_id()
 272          * warning on Linux when debugging is enabled. */
 273         cfs_set_ptldebug_header(&header, subsys, mask, line, CDEBUG_STACK());
 274
 275         if (tcd == NULL)                /* arch may not log in IRQ context */
 276                 goto console;
 277
 278         if (tcd->tcd_cur_pages == 0)
 279                 header.ph_flags |= PH_FLAG_FIRST_RECORD;
 280
 281         if (tcd->tcd_shutting_down) {
 282                 cfs_trace_put_tcd(tcd);
 283                 tcd = NULL;
 284                 goto console;
 285         }
 286
 287         depth = __current_nesting_level();
 288         known_size = strlen(file) + 1 + depth;
 289         if (fn)
 290                 known_size += strlen(fn) + 1;
 291
 292         if (libcfs_debug_binary)
 293                 known_size += sizeof(header);
 294
 295         /*/
 296          * '2' used because vsnprintf return real size required for output
 297          * _without_ terminating NULL.
 298          * if needed is to small for this format.
 299          */
 300         for (i = 0; i < 2; i++) {
 301                 tage = cfs_trace_get_tage(tcd, needed + known_size + 1);
 302                 if (tage == NULL) {
 303                         if (needed + known_size > CFS_PAGE_SIZE)
 304                                 mask |= D_ERROR;
 305
 306                         cfs_trace_put_tcd(tcd);
 307                         tcd = NULL;
 308                         goto console;
 309                 }
 310
 311                 string_buf = (char *)cfs_page_address(tage->page) +
 312                                         tage->used + known_size;
 313
 314                 max_nob = CFS_PAGE_SIZE - tage->used - known_size;
 315                 if (max_nob <= 0) {
 316                         printk(CFS_KERN_EMERG "negative max_nob: %d\n",
 317                                max_nob);
 318                         mask |= D_ERROR;
 319                         cfs_trace_put_tcd(tcd);
 320                         tcd = NULL;
 321                         goto console;
 322                 }
 323
 324                 needed = 0;
 325                 if (format1) {
 326                         va_copy(ap, args);
 327                         needed = vsnprintf(string_buf, max_nob, format1, ap);
 328                         va_end(ap);
 329                 }
 330
 331                 if (format2) {
 332                         remain = max_nob - needed;
 333                         if (remain < 0)
 334                                 remain = 0;
 335
 336                         va_start(ap, format2);
 337                         needed += vsnprintf(string_buf + needed, remain,
 338                                             format2, ap);
 339                         va_end(ap);
 340                 }
 341
 342                 if (needed < max_nob) /* well. printing ok.. */
 343                         break;
 344         }
 345
 346         if (*(string_buf+needed-1) != '\n')
 347                 printk(CFS_KERN_INFO "format at %s:%d:%s doesn't end in newline\n",
 348                        file, line, fn);
 349
 350         header.ph_len = known_size + needed;
 351         debug_buf = (char *)cfs_page_address(tage->page) + tage->used;
 352
 353         if (libcfs_debug_binary) {
 354                 memcpy(debug_buf, &header, sizeof(header));
 355                 tage->used += sizeof(header);
 356                 debug_buf += sizeof(header);
 357         }
 358
 359         /* indent message according to the nesting level */
 360         while (depth-- > 0) {
 361                 *(debug_buf++) = '.';
 362                 ++ tage->used;
 363         }
 364
 365         strcpy(debug_buf, file);
 366         tage->used += strlen(file) + 1;
 367         debug_buf += strlen(file) + 1;
 368
 369         if (fn) {
 370                 strcpy(debug_buf, fn);
 371                 tage->used += strlen(fn) + 1;
 372                 debug_buf += strlen(fn) + 1;
 373         }
 374
 375         __LASSERT(debug_buf == string_buf);
 376
 377         tage->used += needed;
 378         __LASSERT (tage->used <= CFS_PAGE_SIZE);
 379
 380 console:
 381         if ((mask & libcfs_printk) == 0) {
 382                 /* no console output requested */
 383                 if (tcd != NULL)
 384                         cfs_trace_put_tcd(tcd);
 385                 return 1;
 386         }
 387
 388         if (cdls != NULL) {
 389                 if (libcfs_console_ratelimit &&
 390                     cdls->cdls_next != 0 &&     /* not first time ever */
 391                     !cfs_time_after(cfs_time_current(), cdls->cdls_next)) {
 392                         /* skipping a console message */
 393                         cdls->cdls_count++;
 394                         if (tcd != NULL)
 395                                 cfs_trace_put_tcd(tcd);
 396                         return 1;
 397                 }
 398
 399                 if (cfs_time_after(cfs_time_current(), cdls->cdls_next +
 400                                                        libcfs_console_max_delay
 401                                                        + cfs_time_seconds(10))) {
 402                         /* last timeout was a long time ago */
 403                         cdls->cdls_delay /= libcfs_console_backoff * 4;
 404                 } else {
 405                         cdls->cdls_delay *= libcfs_console_backoff;
 406
 407                         if (cdls->cdls_delay < libcfs_console_min_delay)
 408                                 cdls->cdls_delay = libcfs_console_min_delay;
 409                         else if (cdls->cdls_delay > libcfs_console_max_delay)
 410                                 cdls->cdls_delay = libcfs_console_max_delay;
 411                 }
 412
 413                 /* ensure cdls_next is never zero after it's been seen */
 414                 cdls->cdls_next = (cfs_time_current() + cdls->cdls_delay) | 1;
 415         }
 416
 417         if (tcd != NULL) {
 418                 cfs_print_to_console(&header, mask, string_buf, needed, file,
 419                                      fn);
 420                 cfs_trace_put_tcd(tcd);
 421         } else {
 422                 string_buf = cfs_trace_get_console_buffer();
 423
 424                 needed = 0;
 425                 if (format1 != NULL) {
 426                         va_copy(ap, args);
 427                         needed = vsnprintf(string_buf,
 428                                            CFS_TRACE_CONSOLE_BUFFER_SIZE,
 429                                            format1, ap);
 430                         va_end(ap);
 431                 }
 432                 if (format2 != NULL) {
 433                         remain = CFS_TRACE_CONSOLE_BUFFER_SIZE - needed;
 434                         if (remain > 0) {
 435                                 va_start(ap, format2);
 436                                 needed += vsnprintf(string_buf+needed, remain, format2, ap);
 437                                 va_end(ap);
 438                         }
 439                 }
 440                 cfs_print_to_console(&header, mask,
 441                                      string_buf, needed, file, fn);
 442
 443                 cfs_trace_put_console_buffer(string_buf);
 444         }
 445
 446         if (cdls != NULL && cdls->cdls_count != 0) {
 447                 string_buf = cfs_trace_get_console_buffer();
 448
 449                 needed = snprintf(string_buf, CFS_TRACE_CONSOLE_BUFFER_SIZE,
 450                          "Skipped %d previous similar message%s\n",
 451                          cdls->cdls_count, (cdls->cdls_count > 1) ? "s" : "");
 452
 453                 cfs_print_to_console(&header, mask,
 454                                  string_buf, needed, file, fn);
 455
 456                 cfs_trace_put_console_buffer(string_buf);
 457                 cdls->cdls_count = 0;
 458         }
 459
 460         return 0;
 461 }
 462 EXPORT_SYMBOL(libcfs_debug_vmsg2);
 463
 464 void
 465 libcfs_assertion_failed(const char *expr, const char *file,
 466                         const char *func, const int line)
 467 {
 468         libcfs_debug_msg(NULL, 0, D_EMERG, file, func, line,
 469                          "ASSERTION(%s) failed\n", expr);
 470         /* cfs_enter_debugger(); */
 471         lbug_with_loc(file, func, line);
 472 }
 473 EXPORT_SYMBOL(libcfs_assertion_failed);
 474
 475 void
 476 cfs_trace_assertion_failed(const char *str,
 477                            const char *fn, const char *file, int line)
 478 {
 479         struct ptldebug_header hdr;
 480
 481         libcfs_panic_in_progress = 1;
 482         libcfs_catastrophe = 1;
 483         cfs_mb();
 484
 485         cfs_set_ptldebug_header(&hdr, DEBUG_SUBSYSTEM, D_EMERG, line,
 486                                 CDEBUG_STACK());
 487
 488         cfs_print_to_console(&hdr, D_EMERG, str, strlen(str), file, fn);
 489
 490         LIBCFS_PANIC("Lustre debug assertion failure\n");
 491
 492         /* not reached */
 493 }
 494
 495 static void
 496 panic_collect_pages(struct page_collection *pc)
 497 {
 498         /* Do the collect_pages job on a single CPU: assumes that all other
 499          * CPUs have been stopped during a panic.  If this isn't true for some
 500          * arch, this will have to be implemented separately in each arch.  */
 501         int                        i;
 502         int                        j;
 503         struct cfs_trace_cpu_data *tcd;
 504
 505         CFS_INIT_LIST_HEAD(&pc->pc_pages);
 506
 507         cfs_tcd_for_each(tcd, i, j) {
 508                 cfs_list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
 509                 tcd->tcd_cur_pages = 0;
 510
 511                 if (pc->pc_want_daemon_pages) {
 512                         cfs_list_splice_init(&tcd->tcd_daemon_pages,
 513                                              &pc->pc_pages);
 514                         tcd->tcd_cur_daemon_pages = 0;
 515                 }
 516         }
 517 }
 518
 519 static void collect_pages_on_all_cpus(struct page_collection *pc)
 520 {
 521         struct cfs_trace_cpu_data *tcd;
 522         int i, cpu;
 523
 524         cfs_spin_lock(&pc->pc_lock);
 525         cfs_for_each_possible_cpu(cpu) {
 526                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
 527                         cfs_list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
 528                         tcd->tcd_cur_pages = 0;
 529                         if (pc->pc_want_daemon_pages) {
 530                                 cfs_list_splice_init(&tcd->tcd_daemon_pages,
 531                                                      &pc->pc_pages);
 532                                 tcd->tcd_cur_daemon_pages = 0;
 533                         }
 534                 }
 535         }
 536         cfs_spin_unlock(&pc->pc_lock);
 537 }
 538
 539 static void collect_pages(struct page_collection *pc)
 540 {
 541         CFS_INIT_LIST_HEAD(&pc->pc_pages);
 542
 543         if (libcfs_panic_in_progress)
 544                 panic_collect_pages(pc);
 545         else
 546                 collect_pages_on_all_cpus(pc);
 547 }
 548
 549 static void put_pages_back_on_all_cpus(struct page_collection *pc)
 550 {
 551         struct cfs_trace_cpu_data *tcd;
 552         cfs_list_t *cur_head;
 553         struct cfs_trace_page *tage;
 554         struct cfs_trace_page *tmp;
 555         int i, cpu;
 556
 557         cfs_spin_lock(&pc->pc_lock);
 558         cfs_for_each_possible_cpu(cpu) {
 559                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
 560                         cur_head = tcd->tcd_pages.next;
 561
 562                         cfs_list_for_each_entry_safe_typed(tage, tmp,
 563                                                            &pc->pc_pages,
 564                                                            struct cfs_trace_page,
 565                                                            linkage) {
 566
 567                                 __LASSERT_TAGE_INVARIANT(tage);
 568
 569                                 if (tage->cpu != cpu || tage->type != i)
 570                                         continue;
 571
 572                                 cfs_tage_to_tail(tage, cur_head);
 573                                 tcd->tcd_cur_pages++;
 574                         }
 575                 }
 576         }
 577         cfs_spin_unlock(&pc->pc_lock);
 578 }
 579
 580 static void put_pages_back(struct page_collection *pc)
 581 {
 582         if (!libcfs_panic_in_progress)
 583                 put_pages_back_on_all_cpus(pc);
 584 }
 585
 586 /* Add pages to a per-cpu debug daemon ringbuffer.  This buffer makes sure that
 587  * we have a good amount of data at all times for dumping during an LBUG, even
 588  * if we have been steadily writing (and otherwise discarding) pages via the
 589  * debug daemon. */
 590 static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
 591                                          struct cfs_trace_cpu_data *tcd)
 592 {
 593         struct cfs_trace_page *tage;
 594         struct cfs_trace_page *tmp;
 595
 596         cfs_spin_lock(&pc->pc_lock);
 597         cfs_list_for_each_entry_safe_typed(tage, tmp, &pc->pc_pages,
 598                                            struct cfs_trace_page, linkage) {
 599
 600                 __LASSERT_TAGE_INVARIANT(tage);
 601
 602                 if (tage->cpu != tcd->tcd_cpu || tage->type != tcd->tcd_type)
 603                         continue;
 604
 605                 cfs_tage_to_tail(tage, &tcd->tcd_daemon_pages);
 606                 tcd->tcd_cur_daemon_pages++;
 607
 608                 if (tcd->tcd_cur_daemon_pages > tcd->tcd_max_pages) {
 609                         struct cfs_trace_page *victim;
 610
 611                         __LASSERT(!cfs_list_empty(&tcd->tcd_daemon_pages));
 612                         victim = cfs_tage_from_list(tcd->tcd_daemon_pages.next);
 613
 614                         __LASSERT_TAGE_INVARIANT(victim);
 615
 616                         cfs_list_del(&victim->linkage);
 617                         cfs_tage_free(victim);
 618                         tcd->tcd_cur_daemon_pages--;
 619                 }
 620         }
 621         cfs_spin_unlock(&pc->pc_lock);
 622 }
 623
 624 static void put_pages_on_daemon_list(struct page_collection *pc)
 625 {
 626         struct cfs_trace_cpu_data *tcd;
 627         int i, cpu;
 628
 629         cfs_for_each_possible_cpu(cpu) {
 630                 cfs_tcd_for_each_type_lock(tcd, i, cpu)
 631                         put_pages_on_tcd_daemon_list(pc, tcd);
 632         }
 633 }
 634
 635 void cfs_trace_debug_print(void)
 636 {
 637         struct page_collection pc;
 638         struct cfs_trace_page *tage;
 639         struct cfs_trace_page *tmp;
 640
 641         cfs_spin_lock_init(&pc.pc_lock);
 642
 643         pc.pc_want_daemon_pages = 1;
 644         collect_pages(&pc);
 645         cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
 646                                            struct cfs_trace_page, linkage) {
 647                 char *p, *file, *fn;
 648                 cfs_page_t *page;
 649
 650                 __LASSERT_TAGE_INVARIANT(tage);
 651
 652                 page = tage->page;
 653                 p = cfs_page_address(page);
 654                 while (p < ((char *)cfs_page_address(page) + tage->used)) {
 655                         struct ptldebug_header *hdr;
 656                         int len;
 657                         hdr = (void *)p;
 658                         p += sizeof(*hdr);
 659                         file = p;
 660                         p += strlen(file) + 1;
 661                         fn = p;
 662                         p += strlen(fn) + 1;
 663                         len = hdr->ph_len - (int)(p - (char *)hdr);
 664
 665                         cfs_print_to_console(hdr, D_EMERG, p, len, file, fn);
 666
 667                         p += len;
 668                 }
 669
 670                 cfs_list_del(&tage->linkage);
 671                 cfs_tage_free(tage);
 672         }
 673 }
 674
 675 int cfs_tracefile_dump_all_pages(char *filename)
 676 {
 677         struct page_collection pc;
 678         cfs_file_t *filp;
 679         struct cfs_trace_page *tage;
 680         struct cfs_trace_page *tmp;
 681         int rc;
 682
 683         CFS_DECL_MMSPACE;
 684
 685         cfs_tracefile_write_lock();
 686
 687         filp = cfs_filp_open(filename,
 688                              O_CREAT|O_EXCL|O_WRONLY|O_LARGEFILE, 0600, &rc);
 689         if (!filp) {
 690                 if (rc != -EEXIST)
 691                         printk(CFS_KERN_ERR
 692                                "LustreError: can't open %s for dump: rc %d\n",
 693                                filename, rc);
 694                 goto out;
 695         }
 696
 697         cfs_spin_lock_init(&pc.pc_lock);
 698         pc.pc_want_daemon_pages = 1;
 699         collect_pages(&pc);
 700         if (cfs_list_empty(&pc.pc_pages)) {
 701                 rc = 0;
 702                 goto close;
 703         }
 704
 705         /* ok, for now, just write the pages.  in the future we'll be building
 706          * iobufs with the pages and calling generic_direct_IO */
 707         CFS_MMSPACE_OPEN;
 708         cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
 709                                            struct cfs_trace_page, linkage) {
 710
 711                 __LASSERT_TAGE_INVARIANT(tage);
 712
 713                 rc = cfs_filp_write(filp, cfs_page_address(tage->page),
 714                                     tage->used, cfs_filp_poff(filp));
 715                 if (rc != (int)tage->used) {
 716                         printk(CFS_KERN_WARNING "wanted to write %u but wrote "
 717                                "%d\n", tage->used, rc);
 718                         put_pages_back(&pc);
 719                         __LASSERT(cfs_list_empty(&pc.pc_pages));
 720                         break;
 721                 }
 722                 cfs_list_del(&tage->linkage);
 723                 cfs_tage_free(tage);
 724         }
 725         CFS_MMSPACE_CLOSE;
 726         rc = cfs_filp_fsync(filp);
 727         if (rc)
 728                 printk(CFS_KERN_ERR "sync returns %d\n", rc);
 729  close:
 730         cfs_filp_close(filp);
 731  out:
 732         cfs_tracefile_write_unlock();
 733         return rc;
 734 }
 735
 736 void cfs_trace_flush_pages(void)
 737 {
 738         struct page_collection pc;
 739         struct cfs_trace_page *tage;
 740         struct cfs_trace_page *tmp;
 741
 742         cfs_spin_lock_init(&pc.pc_lock);
 743
 744         pc.pc_want_daemon_pages = 1;
 745         collect_pages(&pc);
 746         cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
 747                                            struct cfs_trace_page, linkage) {
 748
 749                 __LASSERT_TAGE_INVARIANT(tage);
 750
 751                 cfs_list_del(&tage->linkage);
 752                 cfs_tage_free(tage);
 753         }
 754 }
 755
 756 int cfs_trace_copyin_string(char *knl_buffer, int knl_buffer_nob,
 757                             const char *usr_buffer, int usr_buffer_nob)
 758 {
 759         int    nob;
 760
 761         if (usr_buffer_nob > knl_buffer_nob)
 762                 return -EOVERFLOW;
 763
 764         if (cfs_copy_from_user((void *)knl_buffer,
 765                            (void *)usr_buffer, usr_buffer_nob))
 766                 return -EFAULT;
 767
 768         nob = strnlen(knl_buffer, usr_buffer_nob);
 769         while (nob-- >= 0)                      /* strip trailing whitespace */
 770                 if (!isspace(knl_buffer[nob]))
 771                         break;
 772
 773         if (nob < 0)                            /* empty string */
 774                 return -EINVAL;
 775
 776         if (nob == knl_buffer_nob)              /* no space to terminate */
 777                 return -EOVERFLOW;
 778
 779         knl_buffer[nob + 1] = 0;                /* terminate */
 780         return 0;
 781 }
 782
 783 int cfs_trace_copyout_string(char *usr_buffer, int usr_buffer_nob,
 784                              const char *knl_buffer, char *append)
 785 {
 786         /* NB if 'append' != NULL, it's a single character to append to the
 787          * copied out string - usually "\n", for /proc entries and "" (i.e. a
 788          * terminating zero byte) for sysctl entries */
 789         int   nob = strlen(knl_buffer);
 790
 791         if (nob > usr_buffer_nob)
 792                 nob = usr_buffer_nob;
 793
 794         if (cfs_copy_to_user(usr_buffer, knl_buffer, nob))
 795                 return -EFAULT;
 796
 797         if (append != NULL && nob < usr_buffer_nob) {
 798                 if (cfs_copy_to_user(usr_buffer + nob, append, 1))
 799                         return -EFAULT;
 800
 801                 nob++;
 802         }
 803
 804         return nob;
 805 }
 806 EXPORT_SYMBOL(cfs_trace_copyout_string);
 807
 808 int cfs_trace_allocate_string_buffer(char **str, int nob)
 809 {
 810         if (nob > 2 * CFS_PAGE_SIZE)            /* string must be "sensible" */
 811                 return -EINVAL;
 812
 813         *str = cfs_alloc(nob, CFS_ALLOC_STD | CFS_ALLOC_ZERO);
 814         if (*str == NULL)
 815                 return -ENOMEM;
 816
 817         return 0;
 818 }
 819
 820 void cfs_trace_free_string_buffer(char *str, int nob)
 821 {
 822         cfs_free(str);
 823 }
 824
 825 int cfs_trace_dump_debug_buffer_usrstr(void *usr_str, int usr_str_nob)
 826 {
 827         char         *str;
 828         int           rc;
 829
 830         rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
 831         if (rc != 0)
 832                 return rc;
 833
 834         rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
 835                                      usr_str, usr_str_nob);
 836         if (rc != 0)
 837                 goto out;
 838
 839 #if !defined(__WINNT__)
 840         if (str[0] != '/') {
 841                 rc = -EINVAL;
 842                 goto out;
 843         }
 844 #endif
 845         rc = cfs_tracefile_dump_all_pages(str);
 846 out:
 847         cfs_trace_free_string_buffer(str, usr_str_nob + 1);
 848         return rc;
 849 }
 850
 851 int cfs_trace_daemon_command(char *str)
 852 {
 853         int       rc = 0;
 854
 855         cfs_tracefile_write_lock();
 856
 857         if (strcmp(str, "stop") == 0) {
 858                 cfs_tracefile_write_unlock();
 859                 cfs_trace_stop_thread();
 860                 cfs_tracefile_write_lock();
 861                 memset(cfs_tracefile, 0, sizeof(cfs_tracefile));
 862
 863         } else if (strncmp(str, "size=", 5) == 0) {
 864                 cfs_tracefile_size = simple_strtoul(str + 5, NULL, 0);
 865                 if (cfs_tracefile_size < 10 || cfs_tracefile_size > 20480)
 866                         cfs_tracefile_size = CFS_TRACEFILE_SIZE;
 867                 else
 868                         cfs_tracefile_size <<= 20;
 869
 870         } else if (strlen(str) >= sizeof(cfs_tracefile)) {
 871                 rc = -ENAMETOOLONG;
 872 #ifndef __WINNT__
 873         } else if (str[0] != '/') {
 874                 rc = -EINVAL;
 875 #endif
 876         } else {
 877                 strcpy(cfs_tracefile, str);
 878
 879                 printk(CFS_KERN_INFO
 880                        "Lustre: debug daemon will attempt to start writing "
 881                        "to %s (%lukB max)\n", cfs_tracefile,
 882                        (long)(cfs_tracefile_size >> 10));
 883
 884                 cfs_trace_start_thread();
 885         }
 886
 887         cfs_tracefile_write_unlock();
 888         return rc;
 889 }
 890
 891 int cfs_trace_daemon_command_usrstr(void *usr_str, int usr_str_nob)
 892 {
 893         char *str;
 894         int   rc;
 895
 896         rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
 897         if (rc != 0)
 898                 return rc;
 899
 900         rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
 901                                  usr_str, usr_str_nob);
 902         if (rc == 0)
 903                 rc = cfs_trace_daemon_command(str);
 904
 905         cfs_trace_free_string_buffer(str, usr_str_nob + 1);
 906         return rc;
 907 }
 908
 909 int cfs_trace_set_debug_mb(int mb)
 910 {
 911         int i;
 912         int j;
 913         int pages;
 914         int limit = cfs_trace_max_debug_mb();
 915         struct cfs_trace_cpu_data *tcd;
 916
 917         if (mb < cfs_num_possible_cpus()) {
 918                 printk(KERN_ERR "Cannot set debug_mb to %d, the value should be >= %d\n",
 919                        mb, cfs_num_possible_cpus());
 920                 return -EINVAL;
 921         }
 922
 923         if (mb > limit) {
 924                 printk(CFS_KERN_ERR "Lustre: Refusing to set debug buffer size "
 925                        "to %dMB - limit is %d\n", mb, limit);
 926                 return -EINVAL;
 927         }
 928
 929         mb /= cfs_num_possible_cpus();
 930         pages = mb << (20 - CFS_PAGE_SHIFT);
 931
 932         cfs_tracefile_write_lock();
 933
 934         cfs_tcd_for_each(tcd, i, j)
 935                 tcd->tcd_max_pages = (pages * tcd->tcd_pages_factor) / 100;
 936
 937         cfs_tracefile_write_unlock();
 938
 939         return 0;
 940 }
 941
 942 int cfs_trace_set_debug_mb_usrstr(void *usr_str, int usr_str_nob)
 943 {
 944         char     str[32];
 945         int      rc;
 946
 947         rc = cfs_trace_copyin_string(str, sizeof(str), usr_str, usr_str_nob);
 948         if (rc < 0)
 949                 return rc;
 950
 951         return cfs_trace_set_debug_mb(simple_strtoul(str, NULL, 0));
 952 }
 953
 954 int cfs_trace_get_debug_mb(void)
 955 {
 956         int i;
 957         int j;
 958         struct cfs_trace_cpu_data *tcd;
 959         int total_pages = 0;
 960
 961         cfs_tracefile_read_lock();
 962
 963         cfs_tcd_for_each(tcd, i, j)
 964                 total_pages += tcd->tcd_max_pages;
 965
 966         cfs_tracefile_read_unlock();
 967
 968         return (total_pages >> (20 - CFS_PAGE_SHIFT)) + 1;
 969 }
 970
 971 static int tracefiled(void *arg)
 972 {
 973         struct page_collection pc;
 974         struct tracefiled_ctl *tctl = arg;
 975         struct cfs_trace_page *tage;
 976         struct cfs_trace_page *tmp;
 977         cfs_file_t *filp;
 978         int last_loop = 0;
 979         int rc;
 980
 981         CFS_DECL_MMSPACE;
 982
 983         /* we're started late enough that we pick up init's fs context */
 984         /* this is so broken in uml?  what on earth is going on? */
 985         cfs_daemonize("ktracefiled");
 986
 987         cfs_spin_lock_init(&pc.pc_lock);
 988         cfs_complete(&tctl->tctl_start);
 989
 990         while (1) {
 991                 cfs_waitlink_t __wait;
 992
 993                 pc.pc_want_daemon_pages = 0;
 994                 collect_pages(&pc);
 995                 if (cfs_list_empty(&pc.pc_pages))
 996                         goto end_loop;
 997
 998                 filp = NULL;
 999                 cfs_tracefile_read_lock();
1000                 if (cfs_tracefile[0] != 0) {
1001                         filp = cfs_filp_open(cfs_tracefile,
1002                                              O_CREAT | O_RDWR | O_LARGEFILE,
1003                                              0600, &rc);
1004                         if (!(filp))
1005                                 printk(CFS_KERN_WARNING "couldn't open %s: "
1006                                        "%d\n", cfs_tracefile, rc);
1007                 }
1008                 cfs_tracefile_read_unlock();
1009                 if (filp == NULL) {
1010                         put_pages_on_daemon_list(&pc);
1011                         __LASSERT(cfs_list_empty(&pc.pc_pages));
1012                         goto end_loop;
1013                 }
1014
1015                 CFS_MMSPACE_OPEN;
1016
1017                 cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
1018                                                    struct cfs_trace_page,
1019                                                    linkage) {
1020                         static loff_t f_pos;
1021
1022                         __LASSERT_TAGE_INVARIANT(tage);
1023
1024                         if (f_pos >= (off_t)cfs_tracefile_size)
1025                                 f_pos = 0;
1026                         else if (f_pos > (off_t)cfs_filp_size(filp))
1027                                 f_pos = cfs_filp_size(filp);
1028
1029                         rc = cfs_filp_write(filp, cfs_page_address(tage->page),
1030                                             tage->used, &f_pos);
1031                         if (rc != (int)tage->used) {
1032                                 printk(CFS_KERN_WARNING "wanted to write %u "
1033                                        "but wrote %d\n", tage->used, rc);
1034                                 put_pages_back(&pc);
1035                                 __LASSERT(cfs_list_empty(&pc.pc_pages));
1036                         }
1037                 }
1038                 CFS_MMSPACE_CLOSE;
1039
1040                 cfs_filp_close(filp);
1041                 put_pages_on_daemon_list(&pc);
1042                 if (!cfs_list_empty(&pc.pc_pages)) {
1043                         int i;
1044
1045                         printk(CFS_KERN_ALERT "Lustre: trace pages aren't "
1046                                " empty\n");
1047                         printk(CFS_KERN_ERR "total cpus(%d): ",
1048                                cfs_num_possible_cpus());
1049                         for (i = 0; i < cfs_num_possible_cpus(); i++)
1050                                 if (cfs_cpu_online(i))
1051                                         printk(CFS_KERN_ERR "%d(on) ", i);
1052                                 else
1053                                         printk(CFS_KERN_ERR "%d(off) ", i);
1054                         printk(CFS_KERN_ERR "\n");
1055
1056                         i = 0;
1057                         cfs_list_for_each_entry_safe(tage, tmp, &pc.pc_pages,
1058                                                      linkage)
1059                                 printk(CFS_KERN_ERR "page %d belongs to cpu "
1060                                        "%d\n", ++i, tage->cpu);
1061                         printk(CFS_KERN_ERR "There are %d pages unwritten\n",
1062                                i);
1063                 }
1064                 __LASSERT(cfs_list_empty(&pc.pc_pages));
1065 end_loop:
1066                 if (cfs_atomic_read(&tctl->tctl_shutdown)) {
1067                         if (last_loop == 0) {
1068                                 last_loop = 1;
1069                                 continue;
1070                         } else {
1071                                 break;
1072                         }
1073                 }
1074                 cfs_waitlink_init(&__wait);
1075                 cfs_waitq_add(&tctl->tctl_waitq, &__wait);
1076                 cfs_set_current_state(CFS_TASK_INTERRUPTIBLE);
1077                 cfs_waitq_timedwait(&__wait, CFS_TASK_INTERRUPTIBLE,
1078                                     cfs_time_seconds(1));
1079                 cfs_waitq_del(&tctl->tctl_waitq, &__wait);
1080         }
1081         cfs_complete(&tctl->tctl_stop);
1082         return 0;
1083 }
1084
1085 int cfs_trace_start_thread(void)
1086 {
1087         struct tracefiled_ctl *tctl = &trace_tctl;
1088         int rc = 0;
1089
1090         cfs_mutex_down(&cfs_trace_thread_sem);
1091         if (thread_running)
1092                 goto out;
1093
1094         cfs_init_completion(&tctl->tctl_start);
1095         cfs_init_completion(&tctl->tctl_stop);
1096         cfs_waitq_init(&tctl->tctl_waitq);
1097         cfs_atomic_set(&tctl->tctl_shutdown, 0);
1098
1099         if (cfs_create_thread(tracefiled, tctl, 0) < 0) {
1100                 rc = -ECHILD;
1101                 goto out;
1102         }
1103
1104         cfs_wait_for_completion(&tctl->tctl_start);
1105         thread_running = 1;
1106 out:
1107         cfs_mutex_up(&cfs_trace_thread_sem);
1108         return rc;
1109 }
1110
1111 void cfs_trace_stop_thread(void)
1112 {
1113         struct tracefiled_ctl *tctl = &trace_tctl;
1114
1115         cfs_mutex_down(&cfs_trace_thread_sem);
1116         if (thread_running) {
1117                 printk(CFS_KERN_INFO
1118                        "Lustre: shutting down debug daemon thread...\n");
1119                 cfs_atomic_set(&tctl->tctl_shutdown, 1);
1120                 cfs_wait_for_completion(&tctl->tctl_stop);
1121                 thread_running = 0;
1122         }
1123         cfs_mutex_up(&cfs_trace_thread_sem);
1124 }
1125
1126 int cfs_tracefile_init(int max_pages)
1127 {
1128         struct cfs_trace_cpu_data *tcd;
1129         int                    i;
1130         int                    j;
1131         int                    rc;
1132         int                    factor;
1133
1134         rc = cfs_tracefile_init_arch();
1135         if (rc != 0)
1136                 return rc;
1137
1138         cfs_tcd_for_each(tcd, i, j) {
1139                 /* tcd_pages_factor is initialized int tracefile_init_arch. */
1140                 factor = tcd->tcd_pages_factor;
1141                 CFS_INIT_LIST_HEAD(&tcd->tcd_pages);
1142                 CFS_INIT_LIST_HEAD(&tcd->tcd_stock_pages);
1143                 CFS_INIT_LIST_HEAD(&tcd->tcd_daemon_pages);
1144                 tcd->tcd_cur_pages = 0;
1145                 tcd->tcd_cur_stock_pages = 0;
1146                 tcd->tcd_cur_daemon_pages = 0;
1147                 tcd->tcd_max_pages = (max_pages * factor) / 100;
1148                 LASSERT(tcd->tcd_max_pages > 0);
1149                 tcd->tcd_shutting_down = 0;
1150         }
1151
1152         return 0;
1153 }
1154
1155 static void trace_cleanup_on_all_cpus(void)
1156 {
1157         struct cfs_trace_cpu_data *tcd;
1158         struct cfs_trace_page *tage;
1159         struct cfs_trace_page *tmp;
1160         int i, cpu;
1161
1162         cfs_for_each_possible_cpu(cpu) {
1163                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
1164                         tcd->tcd_shutting_down = 1;
1165
1166                         cfs_list_for_each_entry_safe_typed(tage, tmp,
1167                                                            &tcd->tcd_pages,
1168                                                            struct cfs_trace_page,
1169                                                            linkage) {
1170                                 __LASSERT_TAGE_INVARIANT(tage);
1171
1172                                 cfs_list_del(&tage->linkage);
1173                                 cfs_tage_free(tage);
1174                         }
1175
1176                         tcd->tcd_cur_pages = 0;
1177                 }
1178         }
1179 }
1180
1181 static void cfs_trace_cleanup(void)
1182 {
1183         struct page_collection pc;
1184
1185         CFS_INIT_LIST_HEAD(&pc.pc_pages);
1186         cfs_spin_lock_init(&pc.pc_lock);
1187
1188         trace_cleanup_on_all_cpus();
1189
1190         cfs_tracefile_fini_arch();
1191 }
1192
1193 void cfs_tracefile_exit(void)
1194 {
1195         cfs_trace_stop_thread();
1196         cfs_trace_cleanup();
1197 }