libcfs/libcfs/tracefile.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2012, Intel Corporation.
  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 struct mutex cfs_trace_thread_mutex;
  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         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_msg(struct libcfs_debug_msg_data *msgdata,
 247                      const char *format, ...)
 248 {
 249         va_list args;
 250         int     rc;
 251
 252         va_start(args, format);
 253         rc = libcfs_debug_vmsg2(msgdata, format, args, NULL);
 254         va_end(args);
 255
 256         return rc;
 257 }
 258 EXPORT_SYMBOL(libcfs_debug_msg);
 259
 260 int libcfs_debug_vmsg2(struct libcfs_debug_msg_data *msgdata,
 261                        const char *format1, va_list args,
 262                        const char *format2, ...)
 263 {
 264         struct cfs_trace_cpu_data *tcd = NULL;
 265         struct ptldebug_header     header = {0};
 266         struct cfs_trace_page     *tage;
 267         /* string_buf is used only if tcd != NULL, and is always set then */
 268         char                      *string_buf = NULL;
 269         char                      *debug_buf;
 270         int                        known_size;
 271         int                        needed = 85; /* average message length */
 272         int                        max_nob;
 273         va_list                    ap;
 274         int                        depth;
 275         int                        i;
 276         int                        remain;
 277         int                        mask = msgdata->msg_mask;
 278         char                      *file = (char *)msgdata->msg_file;
 279         cfs_debug_limit_state_t   *cdls = msgdata->msg_cdls;
 280
 281         if (strchr(file, '/'))
 282                 file = strrchr(file, '/') + 1;
 283
 284         tcd = cfs_trace_get_tcd();
 285
 286         /* cfs_trace_get_tcd() grabs a lock, which disables preemption and
 287          * pins us to a particular CPU.  This avoids an smp_processor_id()
 288          * warning on Linux when debugging is enabled. */
 289         cfs_set_ptldebug_header(&header, msgdata, CDEBUG_STACK());
 290
 291         if (tcd == NULL)                /* arch may not log in IRQ context */
 292                 goto console;
 293
 294         if (tcd->tcd_cur_pages == 0)
 295                 header.ph_flags |= PH_FLAG_FIRST_RECORD;
 296
 297         if (tcd->tcd_shutting_down) {
 298                 cfs_trace_put_tcd(tcd);
 299                 tcd = NULL;
 300                 goto console;
 301         }
 302
 303         depth = __current_nesting_level();
 304         known_size = strlen(file) + 1 + depth;
 305         if (msgdata->msg_fn)
 306                 known_size += strlen(msgdata->msg_fn) + 1;
 307
 308         if (libcfs_debug_binary)
 309                 known_size += sizeof(header);
 310
 311         /*/
 312          * '2' used because vsnprintf return real size required for output
 313          * _without_ terminating NULL.
 314          * if needed is to small for this format.
 315          */
 316         for (i = 0; i < 2; i++) {
 317                 tage = cfs_trace_get_tage(tcd, needed + known_size + 1);
 318                 if (tage == NULL) {
 319                         if (needed + known_size > CFS_PAGE_SIZE)
 320                                 mask |= D_ERROR;
 321
 322                         cfs_trace_put_tcd(tcd);
 323                         tcd = NULL;
 324                         goto console;
 325                 }
 326
 327                 string_buf = (char *)cfs_page_address(tage->page) +
 328                                         tage->used + known_size;
 329
 330                 max_nob = CFS_PAGE_SIZE - tage->used - known_size;
 331                 if (max_nob <= 0) {
 332                         printk(CFS_KERN_EMERG "negative max_nob: %d\n",
 333                                max_nob);
 334                         mask |= D_ERROR;
 335                         cfs_trace_put_tcd(tcd);
 336                         tcd = NULL;
 337                         goto console;
 338                 }
 339
 340                 needed = 0;
 341                 if (format1) {
 342                         va_copy(ap, args);
 343                         needed = vsnprintf(string_buf, max_nob, format1, ap);
 344                         va_end(ap);
 345                 }
 346
 347                 if (format2) {
 348                         remain = max_nob - needed;
 349                         if (remain < 0)
 350                                 remain = 0;
 351
 352                         va_start(ap, format2);
 353                         needed += vsnprintf(string_buf + needed, remain,
 354                                             format2, ap);
 355                         va_end(ap);
 356                 }
 357
 358                 if (needed < max_nob) /* well. printing ok.. */
 359                         break;
 360         }
 361
 362         if (*(string_buf+needed-1) != '\n')
 363                 printk(CFS_KERN_INFO "format at %s:%d:%s doesn't end in "
 364                        "newline\n", file, msgdata->msg_line, msgdata->msg_fn);
 365
 366         header.ph_len = known_size + needed;
 367         debug_buf = (char *)cfs_page_address(tage->page) + tage->used;
 368
 369         if (libcfs_debug_binary) {
 370                 memcpy(debug_buf, &header, sizeof(header));
 371                 tage->used += sizeof(header);
 372                 debug_buf += sizeof(header);
 373         }
 374
 375         /* indent message according to the nesting level */
 376         while (depth-- > 0) {
 377                 *(debug_buf++) = '.';
 378                 ++ tage->used;
 379         }
 380
 381         strcpy(debug_buf, file);
 382         tage->used += strlen(file) + 1;
 383         debug_buf += strlen(file) + 1;
 384
 385         if (msgdata->msg_fn) {
 386                 strcpy(debug_buf, msgdata->msg_fn);
 387                 tage->used += strlen(msgdata->msg_fn) + 1;
 388                 debug_buf += strlen(msgdata->msg_fn) + 1;
 389         }
 390
 391         __LASSERT(debug_buf == string_buf);
 392
 393         tage->used += needed;
 394         __LASSERT (tage->used <= CFS_PAGE_SIZE);
 395
 396 console:
 397         if ((mask & libcfs_printk) == 0) {
 398                 /* no console output requested */
 399                 if (tcd != NULL)
 400                         cfs_trace_put_tcd(tcd);
 401                 return 1;
 402         }
 403
 404         if (cdls != NULL) {
 405                 if (libcfs_console_ratelimit &&
 406                     cdls->cdls_next != 0 &&     /* not first time ever */
 407                     !cfs_time_after(cfs_time_current(), cdls->cdls_next)) {
 408                         /* skipping a console message */
 409                         cdls->cdls_count++;
 410                         if (tcd != NULL)
 411                                 cfs_trace_put_tcd(tcd);
 412                         return 1;
 413                 }
 414
 415                 if (cfs_time_after(cfs_time_current(), cdls->cdls_next +
 416                                                        libcfs_console_max_delay
 417                                                        + cfs_time_seconds(10))) {
 418                         /* last timeout was a long time ago */
 419                         cdls->cdls_delay /= libcfs_console_backoff * 4;
 420                 } else {
 421                         cdls->cdls_delay *= libcfs_console_backoff;
 422
 423                         if (cdls->cdls_delay < libcfs_console_min_delay)
 424                                 cdls->cdls_delay = libcfs_console_min_delay;
 425                         else if (cdls->cdls_delay > libcfs_console_max_delay)
 426                                 cdls->cdls_delay = libcfs_console_max_delay;
 427                 }
 428
 429                 /* ensure cdls_next is never zero after it's been seen */
 430                 cdls->cdls_next = (cfs_time_current() + cdls->cdls_delay) | 1;
 431         }
 432
 433         if (tcd != NULL) {
 434                 cfs_print_to_console(&header, mask, string_buf, needed, file,
 435                                      msgdata->msg_fn);
 436                 cfs_trace_put_tcd(tcd);
 437         } else {
 438                 string_buf = cfs_trace_get_console_buffer();
 439
 440                 needed = 0;
 441                 if (format1 != NULL) {
 442                         va_copy(ap, args);
 443                         needed = vsnprintf(string_buf,
 444                                            CFS_TRACE_CONSOLE_BUFFER_SIZE,
 445                                            format1, ap);
 446                         va_end(ap);
 447                 }
 448                 if (format2 != NULL) {
 449                         remain = CFS_TRACE_CONSOLE_BUFFER_SIZE - needed;
 450                         if (remain > 0) {
 451                                 va_start(ap, format2);
 452                                 needed += vsnprintf(string_buf+needed, remain,
 453                                                     format2, ap);
 454                                 va_end(ap);
 455                         }
 456                 }
 457                 cfs_print_to_console(&header, mask,
 458                                      string_buf, needed, file, msgdata->msg_fn);
 459
 460                 cfs_trace_put_console_buffer(string_buf);
 461         }
 462
 463         if (cdls != NULL && cdls->cdls_count != 0) {
 464                 string_buf = cfs_trace_get_console_buffer();
 465
 466                 needed = snprintf(string_buf, CFS_TRACE_CONSOLE_BUFFER_SIZE,
 467                                   "Skipped %d previous similar message%s\n",
 468                                   cdls->cdls_count,
 469                                   (cdls->cdls_count > 1) ? "s" : "");
 470
 471                 cfs_print_to_console(&header, mask,
 472                                      string_buf, needed, file, msgdata->msg_fn);
 473
 474                 cfs_trace_put_console_buffer(string_buf);
 475                 cdls->cdls_count = 0;
 476         }
 477
 478         return 0;
 479 }
 480 EXPORT_SYMBOL(libcfs_debug_vmsg2);
 481
 482 void
 483 cfs_trace_assertion_failed(const char *str,
 484                            struct libcfs_debug_msg_data *msgdata)
 485 {
 486         struct ptldebug_header hdr;
 487
 488         libcfs_panic_in_progress = 1;
 489         libcfs_catastrophe = 1;
 490         cfs_mb();
 491
 492         cfs_set_ptldebug_header(&hdr, msgdata, CDEBUG_STACK());
 493
 494         cfs_print_to_console(&hdr, D_EMERG, str, strlen(str),
 495                              msgdata->msg_file, msgdata->msg_fn);
 496
 497         LIBCFS_PANIC("Lustre debug assertion failure\n");
 498
 499         /* not reached */
 500 }
 501
 502 static void
 503 panic_collect_pages(struct page_collection *pc)
 504 {
 505         /* Do the collect_pages job on a single CPU: assumes that all other
 506          * CPUs have been stopped during a panic.  If this isn't true for some
 507          * arch, this will have to be implemented separately in each arch.  */
 508         int                        i;
 509         int                        j;
 510         struct cfs_trace_cpu_data *tcd;
 511
 512         CFS_INIT_LIST_HEAD(&pc->pc_pages);
 513
 514         cfs_tcd_for_each(tcd, i, j) {
 515                 cfs_list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
 516                 tcd->tcd_cur_pages = 0;
 517
 518                 if (pc->pc_want_daemon_pages) {
 519                         cfs_list_splice_init(&tcd->tcd_daemon_pages,
 520                                              &pc->pc_pages);
 521                         tcd->tcd_cur_daemon_pages = 0;
 522                 }
 523         }
 524 }
 525
 526 static void collect_pages_on_all_cpus(struct page_collection *pc)
 527 {
 528         struct cfs_trace_cpu_data *tcd;
 529         int i, cpu;
 530
 531         spin_lock(&pc->pc_lock);
 532         cfs_for_each_possible_cpu(cpu) {
 533                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
 534                         cfs_list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
 535                         tcd->tcd_cur_pages = 0;
 536                         if (pc->pc_want_daemon_pages) {
 537                                 cfs_list_splice_init(&tcd->tcd_daemon_pages,
 538                                                      &pc->pc_pages);
 539                                 tcd->tcd_cur_daemon_pages = 0;
 540                         }
 541                 }
 542         }
 543         spin_unlock(&pc->pc_lock);
 544 }
 545
 546 static void collect_pages(struct page_collection *pc)
 547 {
 548         CFS_INIT_LIST_HEAD(&pc->pc_pages);
 549
 550         if (libcfs_panic_in_progress)
 551                 panic_collect_pages(pc);
 552         else
 553                 collect_pages_on_all_cpus(pc);
 554 }
 555
 556 static void put_pages_back_on_all_cpus(struct page_collection *pc)
 557 {
 558         struct cfs_trace_cpu_data *tcd;
 559         cfs_list_t *cur_head;
 560         struct cfs_trace_page *tage;
 561         struct cfs_trace_page *tmp;
 562         int i, cpu;
 563
 564         spin_lock(&pc->pc_lock);
 565         cfs_for_each_possible_cpu(cpu) {
 566                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
 567                         cur_head = tcd->tcd_pages.next;
 568
 569                         cfs_list_for_each_entry_safe_typed(tage, tmp,
 570                                                            &pc->pc_pages,
 571                                                            struct cfs_trace_page,
 572                                                            linkage) {
 573
 574                                 __LASSERT_TAGE_INVARIANT(tage);
 575
 576                                 if (tage->cpu != cpu || tage->type != i)
 577                                         continue;
 578
 579                                 cfs_tage_to_tail(tage, cur_head);
 580                                 tcd->tcd_cur_pages++;
 581                         }
 582                 }
 583         }
 584         spin_unlock(&pc->pc_lock);
 585 }
 586
 587 static void put_pages_back(struct page_collection *pc)
 588 {
 589         if (!libcfs_panic_in_progress)
 590                 put_pages_back_on_all_cpus(pc);
 591 }
 592
 593 /* Add pages to a per-cpu debug daemon ringbuffer.  This buffer makes sure that
 594  * we have a good amount of data at all times for dumping during an LBUG, even
 595  * if we have been steadily writing (and otherwise discarding) pages via the
 596  * debug daemon. */
 597 static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
 598                                          struct cfs_trace_cpu_data *tcd)
 599 {
 600         struct cfs_trace_page *tage;
 601         struct cfs_trace_page *tmp;
 602
 603         spin_lock(&pc->pc_lock);
 604         cfs_list_for_each_entry_safe_typed(tage, tmp, &pc->pc_pages,
 605                                            struct cfs_trace_page, linkage) {
 606
 607                 __LASSERT_TAGE_INVARIANT(tage);
 608
 609                 if (tage->cpu != tcd->tcd_cpu || tage->type != tcd->tcd_type)
 610                         continue;
 611
 612                 cfs_tage_to_tail(tage, &tcd->tcd_daemon_pages);
 613                 tcd->tcd_cur_daemon_pages++;
 614
 615                 if (tcd->tcd_cur_daemon_pages > tcd->tcd_max_pages) {
 616                         struct cfs_trace_page *victim;
 617
 618                         __LASSERT(!cfs_list_empty(&tcd->tcd_daemon_pages));
 619                         victim = cfs_tage_from_list(tcd->tcd_daemon_pages.next);
 620
 621                         __LASSERT_TAGE_INVARIANT(victim);
 622
 623                         cfs_list_del(&victim->linkage);
 624                         cfs_tage_free(victim);
 625                         tcd->tcd_cur_daemon_pages--;
 626                 }
 627         }
 628         spin_unlock(&pc->pc_lock);
 629 }
 630
 631 static void put_pages_on_daemon_list(struct page_collection *pc)
 632 {
 633         struct cfs_trace_cpu_data *tcd;
 634         int i, cpu;
 635
 636         cfs_for_each_possible_cpu(cpu) {
 637                 cfs_tcd_for_each_type_lock(tcd, i, cpu)
 638                         put_pages_on_tcd_daemon_list(pc, tcd);
 639         }
 640 }
 641
 642 void cfs_trace_debug_print(void)
 643 {
 644         struct page_collection pc;
 645         struct cfs_trace_page *tage;
 646         struct cfs_trace_page *tmp;
 647
 648         spin_lock_init(&pc.pc_lock);
 649
 650         pc.pc_want_daemon_pages = 1;
 651         collect_pages(&pc);
 652         cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
 653                                            struct cfs_trace_page, linkage) {
 654                 char *p, *file, *fn;
 655                 cfs_page_t *page;
 656
 657                 __LASSERT_TAGE_INVARIANT(tage);
 658
 659                 page = tage->page;
 660                 p = cfs_page_address(page);
 661                 while (p < ((char *)cfs_page_address(page) + tage->used)) {
 662                         struct ptldebug_header *hdr;
 663                         int len;
 664                         hdr = (void *)p;
 665                         p += sizeof(*hdr);
 666                         file = p;
 667                         p += strlen(file) + 1;
 668                         fn = p;
 669                         p += strlen(fn) + 1;
 670                         len = hdr->ph_len - (int)(p - (char *)hdr);
 671
 672                         cfs_print_to_console(hdr, D_EMERG, p, len, file, fn);
 673
 674                         p += len;
 675                 }
 676
 677                 cfs_list_del(&tage->linkage);
 678                 cfs_tage_free(tage);
 679         }
 680 }
 681
 682 int cfs_tracefile_dump_all_pages(char *filename)
 683 {
 684         struct page_collection  pc;
 685         struct file             *filp;
 686         struct cfs_trace_page   *tage;
 687         struct cfs_trace_page   *tmp;
 688         int rc;
 689
 690         CFS_DECL_MMSPACE;
 691
 692         cfs_tracefile_write_lock();
 693
 694         filp = filp_open(filename, O_CREAT|O_EXCL|O_WRONLY|O_LARGEFILE, 0600);
 695         if (IS_ERR(filp)) {
 696                 rc = PTR_ERR(filp);
 697                 filp = NULL;
 698                 printk(KERN_ERR "LustreError: can't open %s for dump: rc %d\n",
 699                       filename, rc);
 700                 goto out;
 701         }
 702
 703         spin_lock_init(&pc.pc_lock);
 704         pc.pc_want_daemon_pages = 1;
 705         collect_pages(&pc);
 706         if (cfs_list_empty(&pc.pc_pages)) {
 707                 rc = 0;
 708                 goto close;
 709         }
 710
 711         /* ok, for now, just write the pages.  in the future we'll be building
 712          * iobufs with the pages and calling generic_direct_IO */
 713         CFS_MMSPACE_OPEN;
 714         cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
 715                                            struct cfs_trace_page, linkage) {
 716
 717                 __LASSERT_TAGE_INVARIANT(tage);
 718
 719                 rc = filp_write(filp, cfs_page_address(tage->page),
 720                                 tage->used, filp_poff(filp));
 721                 if (rc != (int)tage->used) {
 722                         printk(CFS_KERN_WARNING "wanted to write %u but wrote "
 723                                "%d\n", tage->used, rc);
 724                         put_pages_back(&pc);
 725                         __LASSERT(cfs_list_empty(&pc.pc_pages));
 726                         break;
 727                 }
 728                 cfs_list_del(&tage->linkage);
 729                 cfs_tage_free(tage);
 730         }
 731         CFS_MMSPACE_CLOSE;
 732         rc = filp_fsync(filp);
 733         if (rc)
 734                 printk(CFS_KERN_ERR "sync returns %d\n", rc);
 735 close:
 736         filp_close(filp, NULL);
 737 out:
 738         cfs_tracefile_write_unlock();
 739         return rc;
 740 }
 741
 742 void cfs_trace_flush_pages(void)
 743 {
 744         struct page_collection pc;
 745         struct cfs_trace_page *tage;
 746         struct cfs_trace_page *tmp;
 747
 748         spin_lock_init(&pc.pc_lock);
 749
 750         pc.pc_want_daemon_pages = 1;
 751         collect_pages(&pc);
 752         cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
 753                                            struct cfs_trace_page, linkage) {
 754
 755                 __LASSERT_TAGE_INVARIANT(tage);
 756
 757                 cfs_list_del(&tage->linkage);
 758                 cfs_tage_free(tage);
 759         }
 760 }
 761
 762 int cfs_trace_copyin_string(char *knl_buffer, int knl_buffer_nob,
 763                             const char *usr_buffer, int usr_buffer_nob)
 764 {
 765         int    nob;
 766
 767         if (usr_buffer_nob > knl_buffer_nob)
 768                 return -EOVERFLOW;
 769
 770         if (cfs_copy_from_user((void *)knl_buffer,
 771                            (void *)usr_buffer, usr_buffer_nob))
 772                 return -EFAULT;
 773
 774         nob = strnlen(knl_buffer, usr_buffer_nob);
 775         while (nob-- >= 0)                      /* strip trailing whitespace */
 776                 if (!isspace(knl_buffer[nob]))
 777                         break;
 778
 779         if (nob < 0)                            /* empty string */
 780                 return -EINVAL;
 781
 782         if (nob == knl_buffer_nob)              /* no space to terminate */
 783                 return -EOVERFLOW;
 784
 785         knl_buffer[nob + 1] = 0;                /* terminate */
 786         return 0;
 787 }
 788 EXPORT_SYMBOL(cfs_trace_copyin_string);
 789
 790 int cfs_trace_copyout_string(char *usr_buffer, int usr_buffer_nob,
 791                              const char *knl_buffer, char *append)
 792 {
 793         /* NB if 'append' != NULL, it's a single character to append to the
 794          * copied out string - usually "\n", for /proc entries and "" (i.e. a
 795          * terminating zero byte) for sysctl entries */
 796         int   nob = strlen(knl_buffer);
 797
 798         if (nob > usr_buffer_nob)
 799                 nob = usr_buffer_nob;
 800
 801         if (cfs_copy_to_user(usr_buffer, knl_buffer, nob))
 802                 return -EFAULT;
 803
 804         if (append != NULL && nob < usr_buffer_nob) {
 805                 if (cfs_copy_to_user(usr_buffer + nob, append, 1))
 806                         return -EFAULT;
 807
 808                 nob++;
 809         }
 810
 811         return nob;
 812 }
 813 EXPORT_SYMBOL(cfs_trace_copyout_string);
 814
 815 int cfs_trace_allocate_string_buffer(char **str, int nob)
 816 {
 817         if (nob > 2 * CFS_PAGE_SIZE)            /* string must be "sensible" */
 818                 return -EINVAL;
 819
 820         *str = cfs_alloc(nob, CFS_ALLOC_STD | CFS_ALLOC_ZERO);
 821         if (*str == NULL)
 822                 return -ENOMEM;
 823
 824         return 0;
 825 }
 826
 827 void cfs_trace_free_string_buffer(char *str, int nob)
 828 {
 829         cfs_free(str);
 830 }
 831
 832 int cfs_trace_dump_debug_buffer_usrstr(void *usr_str, int usr_str_nob)
 833 {
 834         char         *str;
 835         int           rc;
 836
 837         rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
 838         if (rc != 0)
 839                 return rc;
 840
 841         rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
 842                                      usr_str, usr_str_nob);
 843         if (rc != 0)
 844                 goto out;
 845
 846 #if !defined(__WINNT__)
 847         if (str[0] != '/') {
 848                 rc = -EINVAL;
 849                 goto out;
 850         }
 851 #endif
 852         rc = cfs_tracefile_dump_all_pages(str);
 853 out:
 854         cfs_trace_free_string_buffer(str, usr_str_nob + 1);
 855         return rc;
 856 }
 857
 858 int cfs_trace_daemon_command(char *str)
 859 {
 860         int       rc = 0;
 861
 862         cfs_tracefile_write_lock();
 863
 864         if (strcmp(str, "stop") == 0) {
 865                 cfs_tracefile_write_unlock();
 866                 cfs_trace_stop_thread();
 867                 cfs_tracefile_write_lock();
 868                 memset(cfs_tracefile, 0, sizeof(cfs_tracefile));
 869
 870         } else if (strncmp(str, "size=", 5) == 0) {
 871                 cfs_tracefile_size = simple_strtoul(str + 5, NULL, 0);
 872                 if (cfs_tracefile_size < 10 || cfs_tracefile_size > 20480)
 873                         cfs_tracefile_size = CFS_TRACEFILE_SIZE;
 874                 else
 875                         cfs_tracefile_size <<= 20;
 876
 877         } else if (strlen(str) >= sizeof(cfs_tracefile)) {
 878                 rc = -ENAMETOOLONG;
 879 #ifndef __WINNT__
 880         } else if (str[0] != '/') {
 881                 rc = -EINVAL;
 882 #endif
 883         } else {
 884                 strcpy(cfs_tracefile, str);
 885
 886                 printk(CFS_KERN_INFO
 887                        "Lustre: debug daemon will attempt to start writing "
 888                        "to %s (%lukB max)\n", cfs_tracefile,
 889                        (long)(cfs_tracefile_size >> 10));
 890
 891                 cfs_trace_start_thread();
 892         }
 893
 894         cfs_tracefile_write_unlock();
 895         return rc;
 896 }
 897
 898 int cfs_trace_daemon_command_usrstr(void *usr_str, int usr_str_nob)
 899 {
 900         char *str;
 901         int   rc;
 902
 903         rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
 904         if (rc != 0)
 905                 return rc;
 906
 907         rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
 908                                  usr_str, usr_str_nob);
 909         if (rc == 0)
 910                 rc = cfs_trace_daemon_command(str);
 911
 912         cfs_trace_free_string_buffer(str, usr_str_nob + 1);
 913         return rc;
 914 }
 915
 916 int cfs_trace_set_debug_mb(int mb)
 917 {
 918         int i;
 919         int j;
 920         int pages;
 921         int limit = cfs_trace_max_debug_mb();
 922         struct cfs_trace_cpu_data *tcd;
 923
 924         if (mb < cfs_num_possible_cpus()) {
 925                 printk(CFS_KERN_WARNING
 926                        "Lustre: %d MB is too small for debug buffer size, "
 927                        "setting it to %d MB.\n", mb, cfs_num_possible_cpus());
 928                 mb = cfs_num_possible_cpus();
 929         }
 930
 931         if (mb > limit) {
 932                 printk(CFS_KERN_WARNING
 933                        "Lustre: %d MB is too large for debug buffer size, "
 934                        "setting it to %d MB.\n", mb, limit);
 935                 mb = limit;
 936         }
 937
 938         mb /= cfs_num_possible_cpus();
 939         pages = mb << (20 - CFS_PAGE_SHIFT);
 940
 941         cfs_tracefile_write_lock();
 942
 943         cfs_tcd_for_each(tcd, i, j)
 944                 tcd->tcd_max_pages = (pages * tcd->tcd_pages_factor) / 100;
 945
 946         cfs_tracefile_write_unlock();
 947
 948         return 0;
 949 }
 950
 951 int cfs_trace_set_debug_mb_usrstr(void *usr_str, int usr_str_nob)
 952 {
 953         char     str[32];
 954         int      rc;
 955
 956         rc = cfs_trace_copyin_string(str, sizeof(str), usr_str, usr_str_nob);
 957         if (rc < 0)
 958                 return rc;
 959
 960         return cfs_trace_set_debug_mb(simple_strtoul(str, NULL, 0));
 961 }
 962
 963 int cfs_trace_get_debug_mb(void)
 964 {
 965         int i;
 966         int j;
 967         struct cfs_trace_cpu_data *tcd;
 968         int total_pages = 0;
 969
 970         cfs_tracefile_read_lock();
 971
 972         cfs_tcd_for_each(tcd, i, j)
 973                 total_pages += tcd->tcd_max_pages;
 974
 975         cfs_tracefile_read_unlock();
 976
 977         return (total_pages >> (20 - CFS_PAGE_SHIFT)) + 1;
 978 }
 979
 980 static int tracefiled(void *arg)
 981 {
 982         struct page_collection pc;
 983         struct tracefiled_ctl *tctl = arg;
 984         struct cfs_trace_page *tage;
 985         struct cfs_trace_page *tmp;
 986         struct file *filp;
 987         int last_loop = 0;
 988         int rc;
 989
 990         CFS_DECL_MMSPACE;
 991
 992         /* we're started late enough that we pick up init's fs context */
 993         /* this is so broken in uml?  what on earth is going on? */
 994         cfs_daemonize("ktracefiled");
 995
 996         spin_lock_init(&pc.pc_lock);
 997         complete(&tctl->tctl_start);
 998
 999         while (1) {
1000                 cfs_waitlink_t __wait;
1001
1002                 pc.pc_want_daemon_pages = 0;
1003                 collect_pages(&pc);
1004                 if (cfs_list_empty(&pc.pc_pages))
1005                         goto end_loop;
1006
1007                 filp = NULL;
1008                 cfs_tracefile_read_lock();
1009                 if (cfs_tracefile[0] != 0) {
1010                         filp = filp_open(cfs_tracefile,
1011                                          O_CREAT | O_RDWR | O_LARGEFILE,
1012                                          0600);
1013                         if (IS_ERR(filp)) {
1014                                 rc = PTR_ERR(filp);
1015                                 filp = NULL;
1016                                 printk(CFS_KERN_WARNING "couldn't open %s: "
1017                                        "%d\n", cfs_tracefile, rc);
1018                         }
1019                 }
1020                 cfs_tracefile_read_unlock();
1021                 if (filp == NULL) {
1022                         put_pages_on_daemon_list(&pc);
1023                         __LASSERT(cfs_list_empty(&pc.pc_pages));
1024                         goto end_loop;
1025                 }
1026
1027                 CFS_MMSPACE_OPEN;
1028
1029                 cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
1030                                                    struct cfs_trace_page,
1031                                                    linkage) {
1032                         static loff_t f_pos;
1033
1034                         __LASSERT_TAGE_INVARIANT(tage);
1035
1036                         if (f_pos >= (off_t)cfs_tracefile_size)
1037                                 f_pos = 0;
1038                         else if (f_pos > (off_t)filp_size(filp))
1039                                 f_pos = filp_size(filp);
1040
1041                         rc = filp_write(filp, cfs_page_address(tage->page),
1042                                         tage->used, &f_pos);
1043                         if (rc != (int)tage->used) {
1044                                 printk(CFS_KERN_WARNING "wanted to write %u "
1045                                        "but wrote %d\n", tage->used, rc);
1046                                 put_pages_back(&pc);
1047                                 __LASSERT(cfs_list_empty(&pc.pc_pages));
1048                         }
1049                 }
1050                 CFS_MMSPACE_CLOSE;
1051
1052                 filp_close(filp, NULL);
1053                 put_pages_on_daemon_list(&pc);
1054                 if (!cfs_list_empty(&pc.pc_pages)) {
1055                         int i;
1056
1057                         printk(CFS_KERN_ALERT "Lustre: trace pages aren't "
1058                                " empty\n");
1059                         printk(CFS_KERN_ERR "total cpus(%d): ",
1060                                cfs_num_possible_cpus());
1061                         for (i = 0; i < cfs_num_possible_cpus(); i++)
1062                                 if (cpu_online(i))
1063                                         printk(CFS_KERN_ERR "%d(on) ", i);
1064                                 else
1065                                         printk(CFS_KERN_ERR "%d(off) ", i);
1066                         printk(CFS_KERN_ERR "\n");
1067
1068                         i = 0;
1069                         cfs_list_for_each_entry_safe(tage, tmp, &pc.pc_pages,
1070                                                      linkage)
1071                                 printk(CFS_KERN_ERR "page %d belongs to cpu "
1072                                        "%d\n", ++i, tage->cpu);
1073                         printk(CFS_KERN_ERR "There are %d pages unwritten\n",
1074                                i);
1075                 }
1076                 __LASSERT(cfs_list_empty(&pc.pc_pages));
1077 end_loop:
1078                 if (cfs_atomic_read(&tctl->tctl_shutdown)) {
1079                         if (last_loop == 0) {
1080                                 last_loop = 1;
1081                                 continue;
1082                         } else {
1083                                 break;
1084                         }
1085                 }
1086                 cfs_waitlink_init(&__wait);
1087                 cfs_waitq_add(&tctl->tctl_waitq, &__wait);
1088                 cfs_set_current_state(CFS_TASK_INTERRUPTIBLE);
1089                 cfs_waitq_timedwait(&__wait, CFS_TASK_INTERRUPTIBLE,
1090                                     cfs_time_seconds(1));
1091                 cfs_waitq_del(&tctl->tctl_waitq, &__wait);
1092         }
1093         complete(&tctl->tctl_stop);
1094         return 0;
1095 }
1096
1097 int cfs_trace_start_thread(void)
1098 {
1099         struct tracefiled_ctl *tctl = &trace_tctl;
1100         int rc = 0;
1101
1102         mutex_lock(&cfs_trace_thread_mutex);
1103         if (thread_running)
1104                 goto out;
1105
1106         init_completion(&tctl->tctl_start);
1107         init_completion(&tctl->tctl_stop);
1108         cfs_waitq_init(&tctl->tctl_waitq);
1109         cfs_atomic_set(&tctl->tctl_shutdown, 0);
1110
1111         if (cfs_create_thread(tracefiled, tctl, 0) < 0) {
1112                 rc = -ECHILD;
1113                 goto out;
1114         }
1115
1116         wait_for_completion(&tctl->tctl_start);
1117         thread_running = 1;
1118 out:
1119         mutex_unlock(&cfs_trace_thread_mutex);
1120         return rc;
1121 }
1122
1123 void cfs_trace_stop_thread(void)
1124 {
1125         struct tracefiled_ctl *tctl = &trace_tctl;
1126
1127         mutex_lock(&cfs_trace_thread_mutex);
1128         if (thread_running) {
1129                 printk(CFS_KERN_INFO
1130                        "Lustre: shutting down debug daemon thread...\n");
1131                 cfs_atomic_set(&tctl->tctl_shutdown, 1);
1132                 wait_for_completion(&tctl->tctl_stop);
1133                 thread_running = 0;
1134         }
1135         mutex_unlock(&cfs_trace_thread_mutex);
1136 }
1137
1138 int cfs_tracefile_init(int max_pages)
1139 {
1140         struct cfs_trace_cpu_data *tcd;
1141         int                    i;
1142         int                    j;
1143         int                    rc;
1144         int                    factor;
1145
1146         rc = cfs_tracefile_init_arch();
1147         if (rc != 0)
1148                 return rc;
1149
1150         cfs_tcd_for_each(tcd, i, j) {
1151                 /* tcd_pages_factor is initialized int tracefile_init_arch. */
1152                 factor = tcd->tcd_pages_factor;
1153                 CFS_INIT_LIST_HEAD(&tcd->tcd_pages);
1154                 CFS_INIT_LIST_HEAD(&tcd->tcd_stock_pages);
1155                 CFS_INIT_LIST_HEAD(&tcd->tcd_daemon_pages);
1156                 tcd->tcd_cur_pages = 0;
1157                 tcd->tcd_cur_stock_pages = 0;
1158                 tcd->tcd_cur_daemon_pages = 0;
1159                 tcd->tcd_max_pages = (max_pages * factor) / 100;
1160                 LASSERT(tcd->tcd_max_pages > 0);
1161                 tcd->tcd_shutting_down = 0;
1162         }
1163
1164         return 0;
1165 }
1166
1167 static void trace_cleanup_on_all_cpus(void)
1168 {
1169         struct cfs_trace_cpu_data *tcd;
1170         struct cfs_trace_page *tage;
1171         struct cfs_trace_page *tmp;
1172         int i, cpu;
1173
1174         cfs_for_each_possible_cpu(cpu) {
1175                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
1176                         tcd->tcd_shutting_down = 1;
1177
1178                         cfs_list_for_each_entry_safe_typed(tage, tmp,
1179                                                            &tcd->tcd_pages,
1180                                                            struct cfs_trace_page,
1181                                                            linkage) {
1182                                 __LASSERT_TAGE_INVARIANT(tage);
1183
1184                                 cfs_list_del(&tage->linkage);
1185                                 cfs_tage_free(tage);
1186                         }
1187
1188                         tcd->tcd_cur_pages = 0;
1189                 }
1190         }
1191 }
1192
1193 static void cfs_trace_cleanup(void)
1194 {
1195         struct page_collection pc;
1196
1197         CFS_INIT_LIST_HEAD(&pc.pc_pages);
1198         spin_lock_init(&pc.pc_lock);
1199
1200         trace_cleanup_on_all_cpus();
1201
1202         cfs_tracefile_fini_arch();
1203 }
1204
1205 void cfs_tracefile_exit(void)
1206 {
1207         cfs_trace_stop_thread();
1208         cfs_trace_cleanup();
1209 }