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