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 = {0};
 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         tcd = cfs_trace_get_tcd();
 265
 266         /* cfs_trace_get_tcd() grabs a lock, which disables preemption and
 267          * pins us to a particular CPU.  This avoids an smp_processor_id()
 268          * warning on Linux when debugging is enabled. */
 269         cfs_set_ptldebug_header(&header, subsys, mask, line, CDEBUG_STACK());
 270
 271         if (tcd == NULL)                /* arch may not log in IRQ context */
 272                 goto console;
 273
 274         if (tcd->tcd_cur_pages == 0)
 275                 header.ph_flags |= PH_FLAG_FIRST_RECORD;
 276
 277         if (tcd->tcd_shutting_down) {
 278                 cfs_trace_put_tcd(tcd);
 279                 tcd = NULL;
 280                 goto console;
 281         }
 282
 283         depth = __current_nesting_level();
 284         known_size = strlen(file) + 1 + depth;
 285         if (fn)
 286                 known_size += strlen(fn) + 1;
 287
 288         if (libcfs_debug_binary)
 289                 known_size += sizeof(header);
 290
 291         /*/
 292          * '2' used because vsnprintf return real size required for output
 293          * _without_ terminating NULL.
 294          * if needed is to small for this format.
 295          */
 296         for (i = 0; i < 2; i++) {
 297                 tage = cfs_trace_get_tage(tcd, needed + known_size + 1);
 298                 if (tage == NULL) {
 299                         if (needed + known_size > CFS_PAGE_SIZE)
 300                                 mask |= D_ERROR;
 301
 302                         cfs_trace_put_tcd(tcd);
 303                         tcd = NULL;
 304                         goto console;
 305                 }
 306
 307                 string_buf = (char *)cfs_page_address(tage->page) +
 308                                         tage->used + known_size;
 309
 310                 max_nob = CFS_PAGE_SIZE - tage->used - known_size;
 311                 if (max_nob <= 0) {
 312                         printk(CFS_KERN_EMERG "negative max_nob: %i\n",
 313                                max_nob);
 314                         mask |= D_ERROR;
 315                         cfs_trace_put_tcd(tcd);
 316                         tcd = NULL;
 317                         goto console;
 318                 }
 319
 320                 needed = 0;
 321                 if (format1) {
 322                         va_copy(ap, args);
 323                         needed = vsnprintf(string_buf, max_nob, format1, ap);
 324                         va_end(ap);
 325                 }
 326
 327                 if (format2) {
 328                         remain = max_nob - needed;
 329                         if (remain < 0)
 330                                 remain = 0;
 331
 332                         va_start(ap, format2);
 333                         needed += vsnprintf(string_buf + needed, remain,
 334                                             format2, ap);
 335                         va_end(ap);
 336                 }
 337
 338                 if (needed < max_nob) /* well. printing ok.. */
 339                         break;
 340         }
 341
 342         if (*(string_buf+needed-1) != '\n')
 343                 printk(CFS_KERN_INFO "format at %s:%d:%s doesn't end in newline\n",
 344                        file, line, fn);
 345
 346         header.ph_len = known_size + needed;
 347         debug_buf = (char *)cfs_page_address(tage->page) + tage->used;
 348
 349         if (libcfs_debug_binary) {
 350                 memcpy(debug_buf, &header, sizeof(header));
 351                 tage->used += sizeof(header);
 352                 debug_buf += sizeof(header);
 353         }
 354
 355         /* indent message according to the nesting level */
 356         while (depth-- > 0) {
 357                 *(debug_buf++) = '.';
 358                 ++ tage->used;
 359         }
 360
 361         strcpy(debug_buf, file);
 362         tage->used += strlen(file) + 1;
 363         debug_buf += strlen(file) + 1;
 364
 365         if (fn) {
 366                 strcpy(debug_buf, fn);
 367                 tage->used += strlen(fn) + 1;
 368                 debug_buf += strlen(fn) + 1;
 369         }
 370
 371         __LASSERT(debug_buf == string_buf);
 372
 373         tage->used += needed;
 374         __LASSERT (tage->used <= CFS_PAGE_SIZE);
 375
 376 console:
 377         if ((mask & libcfs_printk) == 0) {
 378                 /* no console output requested */
 379                 if (tcd != NULL)
 380                         cfs_trace_put_tcd(tcd);
 381                 return 1;
 382         }
 383
 384         if (cdls != NULL) {
 385                 if (libcfs_console_ratelimit &&
 386                     cdls->cdls_next != 0 &&     /* not first time ever */
 387                     !cfs_time_after(cfs_time_current(), cdls->cdls_next)) {
 388                         /* skipping a console message */
 389                         cdls->cdls_count++;
 390                         if (tcd != NULL)
 391                                 cfs_trace_put_tcd(tcd);
 392                         return 1;
 393                 }
 394
 395                 if (cfs_time_after(cfs_time_current(), cdls->cdls_next +
 396                                                        libcfs_console_max_delay
 397                                                        + cfs_time_seconds(10))) {
 398                         /* last timeout was a long time ago */
 399                         cdls->cdls_delay /= libcfs_console_backoff * 4;
 400                 } else {
 401                         cdls->cdls_delay *= libcfs_console_backoff;
 402
 403                         if (cdls->cdls_delay < libcfs_console_min_delay)
 404                                 cdls->cdls_delay = libcfs_console_min_delay;
 405                         else if (cdls->cdls_delay > libcfs_console_max_delay)
 406                                 cdls->cdls_delay = libcfs_console_max_delay;
 407                 }
 408
 409                 /* ensure cdls_next is never zero after it's been seen */
 410                 cdls->cdls_next = (cfs_time_current() + cdls->cdls_delay) | 1;
 411         }
 412
 413         if (tcd != NULL) {
 414                 cfs_print_to_console(&header, mask, string_buf, needed, file,
 415                                      fn);
 416                 cfs_trace_put_tcd(tcd);
 417         } else {
 418                 string_buf = cfs_trace_get_console_buffer();
 419
 420                 needed = 0;
 421                 if (format1 != NULL) {
 422                         va_copy(ap, args);
 423                         needed = vsnprintf(string_buf,
 424                                            CFS_TRACE_CONSOLE_BUFFER_SIZE,
 425                                            format1, ap);
 426                         va_end(ap);
 427                 }
 428                 if (format2 != NULL) {
 429                         remain = CFS_TRACE_CONSOLE_BUFFER_SIZE - needed;
 430                         if (remain > 0) {
 431                                 va_start(ap, format2);
 432                                 needed += vsnprintf(string_buf+needed, remain, format2, ap);
 433                                 va_end(ap);
 434                         }
 435                 }
 436                 cfs_print_to_console(&header, mask,
 437                                      string_buf, needed, file, fn);
 438
 439                 cfs_trace_put_console_buffer(string_buf);
 440         }
 441
 442         if (cdls != NULL && cdls->cdls_count != 0) {
 443                 string_buf = cfs_trace_get_console_buffer();
 444
 445                 needed = snprintf(string_buf, CFS_TRACE_CONSOLE_BUFFER_SIZE,
 446                          "Skipped %d previous similar message%s\n",
 447                          cdls->cdls_count, (cdls->cdls_count > 1) ? "s" : "");
 448
 449                 cfs_print_to_console(&header, mask,
 450                                  string_buf, needed, file, fn);
 451
 452                 cfs_trace_put_console_buffer(string_buf);
 453                 cdls->cdls_count = 0;
 454         }
 455
 456         return 0;
 457 }
 458 EXPORT_SYMBOL(libcfs_debug_vmsg2);
 459
 460 void
 461 libcfs_assertion_failed(const char *expr, const char *file,
 462                         const char *func, const int line)
 463 {
 464         libcfs_debug_msg(NULL, 0, D_EMERG, file, func, line,
 465                          "ASSERTION(%s) failed\n", expr);
 466         /* cfs_enter_debugger(); */
 467         lbug_with_loc(file, func, line);
 468 }
 469 EXPORT_SYMBOL(libcfs_assertion_failed);
 470
 471 void
 472 cfs_trace_assertion_failed(const char *str,
 473                            const char *fn, const char *file, int line)
 474 {
 475         struct ptldebug_header hdr;
 476
 477         libcfs_panic_in_progress = 1;
 478         libcfs_catastrophe = 1;
 479         cfs_mb();
 480
 481         cfs_set_ptldebug_header(&hdr, DEBUG_SUBSYSTEM, D_EMERG, line,
 482                                 CDEBUG_STACK());
 483
 484         cfs_print_to_console(&hdr, D_EMERG, str, strlen(str), file, fn);
 485
 486         LIBCFS_PANIC("Lustre debug assertion failure\n");
 487
 488         /* not reached */
 489 }
 490
 491 static void
 492 panic_collect_pages(struct page_collection *pc)
 493 {
 494         /* Do the collect_pages job on a single CPU: assumes that all other
 495          * CPUs have been stopped during a panic.  If this isn't true for some
 496          * arch, this will have to be implemented separately in each arch.  */
 497         int                        i;
 498         int                        j;
 499         struct cfs_trace_cpu_data *tcd;
 500
 501         CFS_INIT_LIST_HEAD(&pc->pc_pages);
 502
 503         cfs_tcd_for_each(tcd, i, j) {
 504                 cfs_list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
 505                 tcd->tcd_cur_pages = 0;
 506
 507                 if (pc->pc_want_daemon_pages) {
 508                         cfs_list_splice_init(&tcd->tcd_daemon_pages,
 509                                              &pc->pc_pages);
 510                         tcd->tcd_cur_daemon_pages = 0;
 511                 }
 512         }
 513 }
 514
 515 static void collect_pages_on_all_cpus(struct page_collection *pc)
 516 {
 517         struct cfs_trace_cpu_data *tcd;
 518         int i, cpu;
 519
 520         cfs_spin_lock(&pc->pc_lock);
 521         cfs_for_each_possible_cpu(cpu) {
 522                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
 523                         cfs_list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
 524                         tcd->tcd_cur_pages = 0;
 525                         if (pc->pc_want_daemon_pages) {
 526                                 cfs_list_splice_init(&tcd->tcd_daemon_pages,
 527                                                      &pc->pc_pages);
 528                                 tcd->tcd_cur_daemon_pages = 0;
 529                         }
 530                 }
 531         }
 532         cfs_spin_unlock(&pc->pc_lock);
 533 }
 534
 535 static void collect_pages(struct page_collection *pc)
 536 {
 537         CFS_INIT_LIST_HEAD(&pc->pc_pages);
 538
 539         if (libcfs_panic_in_progress)
 540                 panic_collect_pages(pc);
 541         else
 542                 collect_pages_on_all_cpus(pc);
 543 }
 544
 545 static void put_pages_back_on_all_cpus(struct page_collection *pc)
 546 {
 547         struct cfs_trace_cpu_data *tcd;
 548         cfs_list_t *cur_head;
 549         struct cfs_trace_page *tage;
 550         struct cfs_trace_page *tmp;
 551         int i, cpu;
 552
 553         cfs_spin_lock(&pc->pc_lock);
 554         cfs_for_each_possible_cpu(cpu) {
 555                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
 556                         cur_head = tcd->tcd_pages.next;
 557
 558                         cfs_list_for_each_entry_safe_typed(tage, tmp,
 559                                                            &pc->pc_pages,
 560                                                            struct cfs_trace_page,
 561                                                            linkage) {
 562
 563                                 __LASSERT_TAGE_INVARIANT(tage);
 564
 565                                 if (tage->cpu != cpu || tage->type != i)
 566                                         continue;
 567
 568                                 cfs_tage_to_tail(tage, cur_head);
 569                                 tcd->tcd_cur_pages++;
 570                         }
 571                 }
 572         }
 573         cfs_spin_unlock(&pc->pc_lock);
 574 }
 575
 576 static void put_pages_back(struct page_collection *pc)
 577 {
 578         if (!libcfs_panic_in_progress)
 579                 put_pages_back_on_all_cpus(pc);
 580 }
 581
 582 /* Add pages to a per-cpu debug daemon ringbuffer.  This buffer makes sure that
 583  * we have a good amount of data at all times for dumping during an LBUG, even
 584  * if we have been steadily writing (and otherwise discarding) pages via the
 585  * debug daemon. */
 586 static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
 587                                          struct cfs_trace_cpu_data *tcd)
 588 {
 589         struct cfs_trace_page *tage;
 590         struct cfs_trace_page *tmp;
 591
 592         cfs_spin_lock(&pc->pc_lock);
 593         cfs_list_for_each_entry_safe_typed(tage, tmp, &pc->pc_pages,
 594                                            struct cfs_trace_page, linkage) {
 595
 596                 __LASSERT_TAGE_INVARIANT(tage);
 597
 598                 if (tage->cpu != tcd->tcd_cpu || tage->type != tcd->tcd_type)
 599                         continue;
 600
 601                 cfs_tage_to_tail(tage, &tcd->tcd_daemon_pages);
 602                 tcd->tcd_cur_daemon_pages++;
 603
 604                 if (tcd->tcd_cur_daemon_pages > tcd->tcd_max_pages) {
 605                         struct cfs_trace_page *victim;
 606
 607                         __LASSERT(!cfs_list_empty(&tcd->tcd_daemon_pages));
 608                         victim = cfs_tage_from_list(tcd->tcd_daemon_pages.next);
 609
 610                         __LASSERT_TAGE_INVARIANT(victim);
 611
 612                         cfs_list_del(&victim->linkage);
 613                         cfs_tage_free(victim);
 614                         tcd->tcd_cur_daemon_pages--;
 615                 }
 616         }
 617         cfs_spin_unlock(&pc->pc_lock);
 618 }
 619
 620 static void put_pages_on_daemon_list(struct page_collection *pc)
 621 {
 622         struct cfs_trace_cpu_data *tcd;
 623         int i, cpu;
 624
 625         cfs_for_each_possible_cpu(cpu) {
 626                 cfs_tcd_for_each_type_lock(tcd, i, cpu)
 627                         put_pages_on_tcd_daemon_list(pc, tcd);
 628         }
 629 }
 630
 631 void cfs_trace_debug_print(void)
 632 {
 633         struct page_collection pc;
 634         struct cfs_trace_page *tage;
 635         struct cfs_trace_page *tmp;
 636
 637         cfs_spin_lock_init(&pc.pc_lock);
 638
 639         pc.pc_want_daemon_pages = 1;
 640         collect_pages(&pc);
 641         cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
 642                                            struct cfs_trace_page, linkage) {
 643                 char *p, *file, *fn;
 644                 cfs_page_t *page;
 645
 646                 __LASSERT_TAGE_INVARIANT(tage);
 647
 648                 page = tage->page;
 649                 p = cfs_page_address(page);
 650                 while (p < ((char *)cfs_page_address(page) + tage->used)) {
 651                         struct ptldebug_header *hdr;
 652                         int len;
 653                         hdr = (void *)p;
 654                         p += sizeof(*hdr);
 655                         file = p;
 656                         p += strlen(file) + 1;
 657                         fn = p;
 658                         p += strlen(fn) + 1;
 659                         len = hdr->ph_len - (int)(p - (char *)hdr);
 660
 661                         cfs_print_to_console(hdr, D_EMERG, p, len, file, fn);
 662
 663                         p += len;
 664                 }
 665
 666                 cfs_list_del(&tage->linkage);
 667                 cfs_tage_free(tage);
 668         }
 669 }
 670
 671 int cfs_tracefile_dump_all_pages(char *filename)
 672 {
 673         struct page_collection pc;
 674         cfs_file_t *filp;
 675         struct cfs_trace_page *tage;
 676         struct cfs_trace_page *tmp;
 677         int rc;
 678
 679         CFS_DECL_MMSPACE;
 680
 681         cfs_tracefile_write_lock();
 682
 683         filp = cfs_filp_open(filename,
 684                              O_CREAT|O_EXCL|O_WRONLY|O_LARGEFILE, 0600, &rc);
 685         if (!filp) {
 686                 if (rc != -EEXIST)
 687                         printk(CFS_KERN_ERR
 688                                "LustreError: can't open %s for dump: rc %d\n",
 689                                filename, rc);
 690                 goto out;
 691         }
 692
 693         cfs_spin_lock_init(&pc.pc_lock);
 694         pc.pc_want_daemon_pages = 1;
 695         collect_pages(&pc);
 696         if (cfs_list_empty(&pc.pc_pages)) {
 697                 rc = 0;
 698                 goto close;
 699         }
 700
 701         /* ok, for now, just write the pages.  in the future we'll be building
 702          * iobufs with the pages and calling generic_direct_IO */
 703         CFS_MMSPACE_OPEN;
 704         cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
 705                                            struct cfs_trace_page, linkage) {
 706
 707                 __LASSERT_TAGE_INVARIANT(tage);
 708
 709                 rc = cfs_filp_write(filp, cfs_page_address(tage->page),
 710                                     tage->used, cfs_filp_poff(filp));
 711                 if (rc != (int)tage->used) {
 712                         printk(CFS_KERN_WARNING "wanted to write %u but wrote "
 713                                "%d\n", tage->used, rc);
 714                         put_pages_back(&pc);
 715                         __LASSERT(cfs_list_empty(&pc.pc_pages));
 716                         break;
 717                 }
 718                 cfs_list_del(&tage->linkage);
 719                 cfs_tage_free(tage);
 720         }
 721         CFS_MMSPACE_CLOSE;
 722         rc = cfs_filp_fsync(filp);
 723         if (rc)
 724                 printk(CFS_KERN_ERR "sync returns %d\n", rc);
 725  close:
 726         cfs_filp_close(filp);
 727  out:
 728         cfs_tracefile_write_unlock();
 729         return rc;
 730 }
 731
 732 void cfs_trace_flush_pages(void)
 733 {
 734         struct page_collection pc;
 735         struct cfs_trace_page *tage;
 736         struct cfs_trace_page *tmp;
 737
 738         cfs_spin_lock_init(&pc.pc_lock);
 739
 740         pc.pc_want_daemon_pages = 1;
 741         collect_pages(&pc);
 742         cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
 743                                            struct cfs_trace_page, linkage) {
 744
 745                 __LASSERT_TAGE_INVARIANT(tage);
 746
 747                 cfs_list_del(&tage->linkage);
 748                 cfs_tage_free(tage);
 749         }
 750 }
 751
 752 int cfs_trace_copyin_string(char *knl_buffer, int knl_buffer_nob,
 753                             const char *usr_buffer, int usr_buffer_nob)
 754 {
 755         int    nob;
 756
 757         if (usr_buffer_nob > knl_buffer_nob)
 758                 return -EOVERFLOW;
 759
 760         if (cfs_copy_from_user((void *)knl_buffer,
 761                            (void *)usr_buffer, usr_buffer_nob))
 762                 return -EFAULT;
 763
 764         nob = strnlen(knl_buffer, usr_buffer_nob);
 765         while (nob-- >= 0)                      /* strip trailing whitespace */
 766                 if (!isspace(knl_buffer[nob]))
 767                         break;
 768
 769         if (nob < 0)                            /* empty string */
 770                 return -EINVAL;
 771
 772         if (nob == knl_buffer_nob)              /* no space to terminate */
 773                 return -EOVERFLOW;
 774
 775         knl_buffer[nob + 1] = 0;                /* terminate */
 776         return 0;
 777 }
 778
 779 int cfs_trace_copyout_string(char *usr_buffer, int usr_buffer_nob,
 780                              const char *knl_buffer, char *append)
 781 {
 782         /* NB if 'append' != NULL, it's a single character to append to the
 783          * copied out string - usually "\n", for /proc entries and "" (i.e. a
 784          * terminating zero byte) for sysctl entries */
 785         int   nob = strlen(knl_buffer);
 786
 787         if (nob > usr_buffer_nob)
 788                 nob = usr_buffer_nob;
 789
 790         if (cfs_copy_to_user(usr_buffer, knl_buffer, nob))
 791                 return -EFAULT;
 792
 793         if (append != NULL && nob < usr_buffer_nob) {
 794                 if (cfs_copy_to_user(usr_buffer + nob, append, 1))
 795                         return -EFAULT;
 796
 797                 nob++;
 798         }
 799
 800         return nob;
 801 }
 802 EXPORT_SYMBOL(cfs_trace_copyout_string);
 803
 804 int cfs_trace_allocate_string_buffer(char **str, int nob)
 805 {
 806         if (nob > 2 * CFS_PAGE_SIZE)            /* string must be "sensible" */
 807                 return -EINVAL;
 808
 809         *str = cfs_alloc(nob, CFS_ALLOC_STD | CFS_ALLOC_ZERO);
 810         if (*str == NULL)
 811                 return -ENOMEM;
 812
 813         return 0;
 814 }
 815
 816 void cfs_trace_free_string_buffer(char *str, int nob)
 817 {
 818         cfs_free(str);
 819 }
 820
 821 int cfs_trace_dump_debug_buffer_usrstr(void *usr_str, int usr_str_nob)
 822 {
 823         char         *str;
 824         int           rc;
 825
 826         rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
 827         if (rc != 0)
 828                 return rc;
 829
 830         rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
 831                                      usr_str, usr_str_nob);
 832         if (rc != 0)
 833                 goto out;
 834
 835 #if !defined(__WINNT__)
 836         if (str[0] != '/') {
 837                 rc = -EINVAL;
 838                 goto out;
 839         }
 840 #endif
 841         rc = cfs_tracefile_dump_all_pages(str);
 842 out:
 843         cfs_trace_free_string_buffer(str, usr_str_nob + 1);
 844         return rc;
 845 }
 846
 847 int cfs_trace_daemon_command(char *str)
 848 {
 849         int       rc = 0;
 850
 851         cfs_tracefile_write_lock();
 852
 853         if (strcmp(str, "stop") == 0) {
 854                 cfs_tracefile_write_unlock();
 855                 cfs_trace_stop_thread();
 856                 cfs_tracefile_write_lock();
 857                 memset(cfs_tracefile, 0, sizeof(cfs_tracefile));
 858
 859         } else if (strncmp(str, "size=", 5) == 0) {
 860                 cfs_tracefile_size = simple_strtoul(str + 5, NULL, 0);
 861                 if (cfs_tracefile_size < 10 || cfs_tracefile_size > 20480)
 862                         cfs_tracefile_size = CFS_TRACEFILE_SIZE;
 863                 else
 864                         cfs_tracefile_size <<= 20;
 865
 866         } else if (strlen(str) >= sizeof(cfs_tracefile)) {
 867                 rc = -ENAMETOOLONG;
 868 #ifndef __WINNT__
 869         } else if (str[0] != '/') {
 870                 rc = -EINVAL;
 871 #endif
 872         } else {
 873                 strcpy(cfs_tracefile, str);
 874
 875                 printk(CFS_KERN_INFO
 876                        "Lustre: debug daemon will attempt to start writing "
 877                        "to %s (%lukB max)\n", cfs_tracefile,
 878                        (long)(cfs_tracefile_size >> 10));
 879
 880                 cfs_trace_start_thread();
 881         }
 882
 883         cfs_tracefile_write_unlock();
 884         return rc;
 885 }
 886
 887 int cfs_trace_daemon_command_usrstr(void *usr_str, int usr_str_nob)
 888 {
 889         char *str;
 890         int   rc;
 891
 892         rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
 893         if (rc != 0)
 894                 return rc;
 895
 896         rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
 897                                  usr_str, usr_str_nob);
 898         if (rc == 0)
 899                 rc = cfs_trace_daemon_command(str);
 900
 901         cfs_trace_free_string_buffer(str, usr_str_nob + 1);
 902         return rc;
 903 }
 904
 905 int cfs_trace_set_debug_mb(int mb)
 906 {
 907         int i;
 908         int j;
 909         int pages;
 910         int limit = cfs_trace_max_debug_mb();
 911         struct cfs_trace_cpu_data *tcd;
 912
 913         if (mb < cfs_num_possible_cpus()) {
 914                 printk(KERN_ERR "Cannot set debug_mb to %d, the value should be >= %d\n",
 915                        mb, cfs_num_possible_cpus());
 916                 return -EINVAL;
 917         }
 918
 919         if (mb > limit) {
 920                 printk(CFS_KERN_ERR "Lustre: Refusing to set debug buffer size "
 921                        "to %dMB - limit is %d\n", mb, limit);
 922                 return -EINVAL;
 923         }
 924
 925         mb /= cfs_num_possible_cpus();
 926         pages = mb << (20 - CFS_PAGE_SHIFT);
 927
 928         cfs_tracefile_write_lock();
 929
 930         cfs_tcd_for_each(tcd, i, j)
 931                 tcd->tcd_max_pages = (pages * tcd->tcd_pages_factor) / 100;
 932
 933         cfs_tracefile_write_unlock();
 934
 935         return 0;
 936 }
 937
 938 int cfs_trace_set_debug_mb_usrstr(void *usr_str, int usr_str_nob)
 939 {
 940         char     str[32];
 941         int      rc;
 942
 943         rc = cfs_trace_copyin_string(str, sizeof(str), usr_str, usr_str_nob);
 944         if (rc < 0)
 945                 return rc;
 946
 947         return cfs_trace_set_debug_mb(simple_strtoul(str, NULL, 0));
 948 }
 949
 950 int cfs_trace_get_debug_mb(void)
 951 {
 952         int i;
 953         int j;
 954         struct cfs_trace_cpu_data *tcd;
 955         int total_pages = 0;
 956
 957         cfs_tracefile_read_lock();
 958
 959         cfs_tcd_for_each(tcd, i, j)
 960                 total_pages += tcd->tcd_max_pages;
 961
 962         cfs_tracefile_read_unlock();
 963
 964         return (total_pages >> (20 - CFS_PAGE_SHIFT)) + 1;
 965 }
 966
 967 static int tracefiled(void *arg)
 968 {
 969         struct page_collection pc;
 970         struct tracefiled_ctl *tctl = arg;
 971         struct cfs_trace_page *tage;
 972         struct cfs_trace_page *tmp;
 973         cfs_file_t *filp;
 974         int last_loop = 0;
 975         int rc;
 976
 977         CFS_DECL_MMSPACE;
 978
 979         /* we're started late enough that we pick up init's fs context */
 980         /* this is so broken in uml?  what on earth is going on? */
 981         cfs_daemonize("ktracefiled");
 982
 983         cfs_spin_lock_init(&pc.pc_lock);
 984         cfs_complete(&tctl->tctl_start);
 985
 986         while (1) {
 987                 cfs_waitlink_t __wait;
 988
 989                 pc.pc_want_daemon_pages = 0;
 990                 collect_pages(&pc);
 991                 if (cfs_list_empty(&pc.pc_pages))
 992                         goto end_loop;
 993
 994                 filp = NULL;
 995                 cfs_tracefile_read_lock();
 996                 if (cfs_tracefile[0] != 0) {
 997                         filp = cfs_filp_open(cfs_tracefile,
 998                                              O_CREAT | O_RDWR | O_LARGEFILE,
 999                                              0600, &rc);
1000                         if (!(filp))
1001                                 printk(CFS_KERN_WARNING "couldn't open %s: "
1002                                        "%d\n", cfs_tracefile, rc);
1003                 }
1004                 cfs_tracefile_read_unlock();
1005                 if (filp == NULL) {
1006                         put_pages_on_daemon_list(&pc);
1007                         __LASSERT(cfs_list_empty(&pc.pc_pages));
1008                         goto end_loop;
1009                 }
1010
1011                 CFS_MMSPACE_OPEN;
1012
1013                 cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
1014                                                    struct cfs_trace_page,
1015                                                    linkage) {
1016                         static loff_t f_pos;
1017
1018                         __LASSERT_TAGE_INVARIANT(tage);
1019
1020                         if (f_pos >= (off_t)cfs_tracefile_size)
1021                                 f_pos = 0;
1022                         else if (f_pos > (off_t)cfs_filp_size(filp))
1023                                 f_pos = cfs_filp_size(filp);
1024
1025                         rc = cfs_filp_write(filp, cfs_page_address(tage->page),
1026                                             tage->used, &f_pos);
1027                         if (rc != (int)tage->used) {
1028                                 printk(CFS_KERN_WARNING "wanted to write %u "
1029                                        "but wrote %d\n", tage->used, rc);
1030                                 put_pages_back(&pc);
1031                                 __LASSERT(cfs_list_empty(&pc.pc_pages));
1032                         }
1033                 }
1034                 CFS_MMSPACE_CLOSE;
1035
1036                 cfs_filp_close(filp);
1037                 put_pages_on_daemon_list(&pc);
1038                 if (!cfs_list_empty(&pc.pc_pages)) {
1039                         int i;
1040
1041                         printk(CFS_KERN_ALERT "Lustre: trace pages aren't "
1042                                " empty\n");
1043                         printk(CFS_KERN_ERR "total cpus(%d): ",
1044                                cfs_num_possible_cpus());
1045                         for (i = 0; i < cfs_num_possible_cpus(); i++)
1046                                 if (cfs_cpu_online(i))
1047                                         printk(CFS_KERN_ERR "%d(on) ", i);
1048                                 else
1049                                         printk(CFS_KERN_ERR "%d(off) ", i);
1050                         printk(CFS_KERN_ERR "\n");
1051
1052                         i = 0;
1053                         cfs_list_for_each_entry_safe(tage, tmp, &pc.pc_pages,
1054                                                      linkage)
1055                                 printk(CFS_KERN_ERR "page %d belongs to cpu "
1056                                        "%d\n", ++i, tage->cpu);
1057                         printk(CFS_KERN_ERR "There are %d pages unwritten\n",
1058                                i);
1059                 }
1060                 __LASSERT(cfs_list_empty(&pc.pc_pages));
1061 end_loop:
1062                 if (cfs_atomic_read(&tctl->tctl_shutdown)) {
1063                         if (last_loop == 0) {
1064                                 last_loop = 1;
1065                                 continue;
1066                         } else {
1067                                 break;
1068                         }
1069                 }
1070                 cfs_waitlink_init(&__wait);
1071                 cfs_waitq_add(&tctl->tctl_waitq, &__wait);
1072                 cfs_set_current_state(CFS_TASK_INTERRUPTIBLE);
1073                 cfs_waitq_timedwait(&__wait, CFS_TASK_INTERRUPTIBLE,
1074                                     cfs_time_seconds(1));
1075                 cfs_waitq_del(&tctl->tctl_waitq, &__wait);
1076         }
1077         cfs_complete(&tctl->tctl_stop);
1078         return 0;
1079 }
1080
1081 int cfs_trace_start_thread(void)
1082 {
1083         struct tracefiled_ctl *tctl = &trace_tctl;
1084         int rc = 0;
1085
1086         cfs_mutex_down(&cfs_trace_thread_sem);
1087         if (thread_running)
1088                 goto out;
1089
1090         cfs_init_completion(&tctl->tctl_start);
1091         cfs_init_completion(&tctl->tctl_stop);
1092         cfs_waitq_init(&tctl->tctl_waitq);
1093         cfs_atomic_set(&tctl->tctl_shutdown, 0);
1094
1095         if (cfs_kernel_thread(tracefiled, tctl, 0) < 0) {
1096                 rc = -ECHILD;
1097                 goto out;
1098         }
1099
1100         cfs_wait_for_completion(&tctl->tctl_start);
1101         thread_running = 1;
1102 out:
1103         cfs_mutex_up(&cfs_trace_thread_sem);
1104         return rc;
1105 }
1106
1107 void cfs_trace_stop_thread(void)
1108 {
1109         struct tracefiled_ctl *tctl = &trace_tctl;
1110
1111         cfs_mutex_down(&cfs_trace_thread_sem);
1112         if (thread_running) {
1113                 printk(CFS_KERN_INFO
1114                        "Lustre: shutting down debug daemon thread...\n");
1115                 cfs_atomic_set(&tctl->tctl_shutdown, 1);
1116                 cfs_wait_for_completion(&tctl->tctl_stop);
1117                 thread_running = 0;
1118         }
1119         cfs_mutex_up(&cfs_trace_thread_sem);
1120 }
1121
1122 int cfs_tracefile_init(int max_pages)
1123 {
1124         struct cfs_trace_cpu_data *tcd;
1125         int                    i;
1126         int                    j;
1127         int                    rc;
1128         int                    factor;
1129
1130         rc = cfs_tracefile_init_arch();
1131         if (rc != 0)
1132                 return rc;
1133
1134         cfs_tcd_for_each(tcd, i, j) {
1135                 /* tcd_pages_factor is initialized int tracefile_init_arch. */
1136                 factor = tcd->tcd_pages_factor;
1137                 CFS_INIT_LIST_HEAD(&tcd->tcd_pages);
1138                 CFS_INIT_LIST_HEAD(&tcd->tcd_stock_pages);
1139                 CFS_INIT_LIST_HEAD(&tcd->tcd_daemon_pages);
1140                 tcd->tcd_cur_pages = 0;
1141                 tcd->tcd_cur_stock_pages = 0;
1142                 tcd->tcd_cur_daemon_pages = 0;
1143                 tcd->tcd_max_pages = (max_pages * factor) / 100;
1144                 LASSERT(tcd->tcd_max_pages > 0);
1145                 tcd->tcd_shutting_down = 0;
1146         }
1147
1148         return 0;
1149 }
1150
1151 static void trace_cleanup_on_all_cpus(void)
1152 {
1153         struct cfs_trace_cpu_data *tcd;
1154         struct cfs_trace_page *tage;
1155         struct cfs_trace_page *tmp;
1156         int i, cpu;
1157
1158         cfs_for_each_possible_cpu(cpu) {
1159                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
1160                         tcd->tcd_shutting_down = 1;
1161
1162                         cfs_list_for_each_entry_safe_typed(tage, tmp,
1163                                                            &tcd->tcd_pages,
1164                                                            struct cfs_trace_page,
1165                                                            linkage) {
1166                                 __LASSERT_TAGE_INVARIANT(tage);
1167
1168                                 cfs_list_del(&tage->linkage);
1169                                 cfs_tage_free(tage);
1170                         }
1171
1172                         tcd->tcd_cur_pages = 0;
1173                 }
1174         }
1175 }
1176
1177 static void cfs_trace_cleanup(void)
1178 {
1179         struct page_collection pc;
1180
1181         CFS_INIT_LIST_HEAD(&pc.pc_pages);
1182         cfs_spin_lock_init(&pc.pc_lock);
1183
1184         trace_cleanup_on_all_cpus();
1185
1186         cfs_tracefile_fini_arch();
1187 }
1188
1189 void cfs_tracefile_exit(void)
1190 {
1191         cfs_trace_stop_thread();
1192         cfs_trace_cleanup();
1193 }