drivers/staging/lustre/lustre/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 "../../include/linux/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])[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 atomic_t cfs_tage_allocated = 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(struct list_head *list)
  65 {
  66         return list_entry(list, struct cfs_trace_page, linkage);
  67 }
  68
  69 static struct cfs_trace_page *cfs_tage_alloc(gfp_t gfp)
  70 {
  71         struct page         *page;
  72         struct cfs_trace_page *tage;
  73
  74         /* My caller is trying to free memory */
  75         if (!in_interrupt() && 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 |= __GFP_NOWARN;
  83         page = alloc_page(gfp);
  84         if (page == NULL)
  85                 return NULL;
  86
  87         tage = kmalloc(sizeof(*tage), gfp);
  88         if (tage == NULL) {
  89                 __free_page(page);
  90                 return NULL;
  91         }
  92
  93         tage->page = page;
  94         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         __free_page(tage->page);
 104         kfree(tage);
 105         atomic_dec(&cfs_tage_allocated);
 106 }
 107
 108 static void cfs_tage_to_tail(struct cfs_trace_page *tage,
 109                              struct list_head *queue)
 110 {
 111         __LASSERT(tage != NULL);
 112         __LASSERT(queue != NULL);
 113
 114         list_move_tail(&tage->linkage, queue);
 115 }
 116
 117 int cfs_trace_refill_stock(struct cfs_trace_cpu_data *tcd, gfp_t gfp,
 118                            struct list_head *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                 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(!list_empty(&tcd->tcd_pages));
 146                 tage = cfs_tage_from_list(tcd->tcd_pages.prev);
 147                 if (tage->used + len <= PAGE_CACHE_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                         list_del_init(&tage->linkage);
 156                 } else {
 157                         tage = cfs_tage_alloc(GFP_ATOMIC);
 158                         if (unlikely(tage == NULL)) {
 159                                 if ((!memory_pressure_get() ||
 160                                      in_interrupt()) && printk_ratelimit())
 161                                         printk(KERN_WARNING
 162                                                "cannot allocate a tage (%ld)\n",
 163                                                tcd->tcd_cur_pages);
 164                                 return NULL;
 165                         }
 166                 }
 167
 168                 tage->used = 0;
 169                 tage->cpu = smp_processor_id();
 170                 tage->type = tcd->tcd_type;
 171                 list_add_tail(&tage->linkage, &tcd->tcd_pages);
 172                 tcd->tcd_cur_pages++;
 173
 174                 if (tcd->tcd_cur_pages > 8 && thread_running) {
 175                         struct tracefiled_ctl *tctl = &trace_tctl;
 176                         /*
 177                          * wake up tracefiled to process some pages.
 178                          */
 179                         wake_up(&tctl->tctl_waitq);
 180                 }
 181                 return tage;
 182         }
 183         return NULL;
 184 }
 185
 186 static void cfs_tcd_shrink(struct cfs_trace_cpu_data *tcd)
 187 {
 188         int pgcount = tcd->tcd_cur_pages / 10;
 189         struct page_collection pc;
 190         struct cfs_trace_page *tage;
 191         struct cfs_trace_page *tmp;
 192
 193         /*
 194          * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
 195          * from here: this will lead to infinite recursion.
 196          */
 197
 198         if (printk_ratelimit())
 199                 printk(KERN_WARNING "debug daemon buffer overflowed; "
 200                        "discarding 10%% of pages (%d of %ld)\n",
 201                        pgcount + 1, tcd->tcd_cur_pages);
 202
 203         INIT_LIST_HEAD(&pc.pc_pages);
 204         spin_lock_init(&pc.pc_lock);
 205
 206         list_for_each_entry_safe(tage, tmp, &tcd->tcd_pages, linkage) {
 207                 if (pgcount-- == 0)
 208                         break;
 209
 210                 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 > PAGE_CACHE_SIZE) {
 228                 printk(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         const char              *file = kbasename(msgdata->msg_file);
 279         struct cfs_debug_limit_state   *cdls = msgdata->msg_cdls;
 280
 281         tcd = cfs_trace_get_tcd();
 282
 283         /* cfs_trace_get_tcd() grabs a lock, which disables preemption and
 284          * pins us to a particular CPU.  This avoids an smp_processor_id()
 285          * warning on Linux when debugging is enabled. */
 286         cfs_set_ptldebug_header(&header, msgdata, CDEBUG_STACK());
 287
 288         if (tcd == NULL)                /* arch may not log in IRQ context */
 289                 goto console;
 290
 291         if (tcd->tcd_cur_pages == 0)
 292                 header.ph_flags |= PH_FLAG_FIRST_RECORD;
 293
 294         if (tcd->tcd_shutting_down) {
 295                 cfs_trace_put_tcd(tcd);
 296                 tcd = NULL;
 297                 goto console;
 298         }
 299
 300         depth = __current_nesting_level();
 301         known_size = strlen(file) + 1 + depth;
 302         if (msgdata->msg_fn)
 303                 known_size += strlen(msgdata->msg_fn) + 1;
 304
 305         if (libcfs_debug_binary)
 306                 known_size += sizeof(header);
 307
 308         /*/
 309          * '2' used because vsnprintf return real size required for output
 310          * _without_ terminating NULL.
 311          * if needed is to small for this format.
 312          */
 313         for (i = 0; i < 2; i++) {
 314                 tage = cfs_trace_get_tage(tcd, needed + known_size + 1);
 315                 if (tage == NULL) {
 316                         if (needed + known_size > PAGE_CACHE_SIZE)
 317                                 mask |= D_ERROR;
 318
 319                         cfs_trace_put_tcd(tcd);
 320                         tcd = NULL;
 321                         goto console;
 322                 }
 323
 324                 string_buf = (char *)page_address(tage->page) +
 325                                         tage->used + known_size;
 326
 327                 max_nob = PAGE_CACHE_SIZE - tage->used - known_size;
 328                 if (max_nob <= 0) {
 329                         printk(KERN_EMERG "negative max_nob: %d\n",
 330                                max_nob);
 331                         mask |= D_ERROR;
 332                         cfs_trace_put_tcd(tcd);
 333                         tcd = NULL;
 334                         goto console;
 335                 }
 336
 337                 needed = 0;
 338                 if (format1) {
 339                         va_copy(ap, args);
 340                         needed = vsnprintf(string_buf, max_nob, format1, ap);
 341                         va_end(ap);
 342                 }
 343
 344                 if (format2) {
 345                         remain = max_nob - needed;
 346                         if (remain < 0)
 347                                 remain = 0;
 348
 349                         va_start(ap, format2);
 350                         needed += vsnprintf(string_buf + needed, remain,
 351                                             format2, ap);
 352                         va_end(ap);
 353                 }
 354
 355                 if (needed < max_nob) /* well. printing ok.. */
 356                         break;
 357         }
 358
 359         if (*(string_buf+needed-1) != '\n')
 360                 printk(KERN_INFO "format at %s:%d:%s doesn't end in "
 361                        "newline\n", file, msgdata->msg_line, msgdata->msg_fn);
 362
 363         header.ph_len = known_size + needed;
 364         debug_buf = (char *)page_address(tage->page) + tage->used;
 365
 366         if (libcfs_debug_binary) {
 367                 memcpy(debug_buf, &header, sizeof(header));
 368                 tage->used += sizeof(header);
 369                 debug_buf += sizeof(header);
 370         }
 371
 372         /* indent message according to the nesting level */
 373         while (depth-- > 0) {
 374                 *(debug_buf++) = '.';
 375                 ++ tage->used;
 376         }
 377
 378         strcpy(debug_buf, file);
 379         tage->used += strlen(file) + 1;
 380         debug_buf += strlen(file) + 1;
 381
 382         if (msgdata->msg_fn) {
 383                 strcpy(debug_buf, msgdata->msg_fn);
 384                 tage->used += strlen(msgdata->msg_fn) + 1;
 385                 debug_buf += strlen(msgdata->msg_fn) + 1;
 386         }
 387
 388         __LASSERT(debug_buf == string_buf);
 389
 390         tage->used += needed;
 391         __LASSERT (tage->used <= PAGE_CACHE_SIZE);
 392
 393 console:
 394         if ((mask & libcfs_printk) == 0) {
 395                 /* no console output requested */
 396                 if (tcd != NULL)
 397                         cfs_trace_put_tcd(tcd);
 398                 return 1;
 399         }
 400
 401         if (cdls != NULL) {
 402                 if (libcfs_console_ratelimit &&
 403                     cdls->cdls_next != 0 &&     /* not first time ever */
 404                     !cfs_time_after(cfs_time_current(), cdls->cdls_next)) {
 405                         /* skipping a console message */
 406                         cdls->cdls_count++;
 407                         if (tcd != NULL)
 408                                 cfs_trace_put_tcd(tcd);
 409                         return 1;
 410                 }
 411
 412                 if (cfs_time_after(cfs_time_current(), cdls->cdls_next +
 413                                                        libcfs_console_max_delay
 414                                                        + cfs_time_seconds(10))) {
 415                         /* last timeout was a long time ago */
 416                         cdls->cdls_delay /= libcfs_console_backoff * 4;
 417                 } else {
 418                         cdls->cdls_delay *= libcfs_console_backoff;
 419                 }
 420
 421                 if (cdls->cdls_delay < libcfs_console_min_delay)
 422                         cdls->cdls_delay = libcfs_console_min_delay;
 423                 else if (cdls->cdls_delay > libcfs_console_max_delay)
 424                         cdls->cdls_delay = libcfs_console_max_delay;
 425
 426                 /* ensure cdls_next is never zero after it's been seen */
 427                 cdls->cdls_next = (cfs_time_current() + cdls->cdls_delay) | 1;
 428         }
 429
 430         if (tcd != NULL) {
 431                 cfs_print_to_console(&header, mask, string_buf, needed, file,
 432                                      msgdata->msg_fn);
 433                 cfs_trace_put_tcd(tcd);
 434         } else {
 435                 string_buf = cfs_trace_get_console_buffer();
 436
 437                 needed = 0;
 438                 if (format1 != NULL) {
 439                         va_copy(ap, args);
 440                         needed = vsnprintf(string_buf,
 441                                            CFS_TRACE_CONSOLE_BUFFER_SIZE,
 442                                            format1, ap);
 443                         va_end(ap);
 444                 }
 445                 if (format2 != NULL) {
 446                         remain = CFS_TRACE_CONSOLE_BUFFER_SIZE - needed;
 447                         if (remain > 0) {
 448                                 va_start(ap, format2);
 449                                 needed += vsnprintf(string_buf+needed, remain,
 450                                                     format2, ap);
 451                                 va_end(ap);
 452                         }
 453                 }
 454                 cfs_print_to_console(&header, mask,
 455                                      string_buf, needed, file, msgdata->msg_fn);
 456
 457                 cfs_trace_put_console_buffer(string_buf);
 458         }
 459
 460         if (cdls != NULL && cdls->cdls_count != 0) {
 461                 string_buf = cfs_trace_get_console_buffer();
 462
 463                 needed = snprintf(string_buf, CFS_TRACE_CONSOLE_BUFFER_SIZE,
 464                                   "Skipped %d previous similar message%s\n",
 465                                   cdls->cdls_count,
 466                                   (cdls->cdls_count > 1) ? "s" : "");
 467
 468                 cfs_print_to_console(&header, mask,
 469                                      string_buf, needed, file, msgdata->msg_fn);
 470
 471                 cfs_trace_put_console_buffer(string_buf);
 472                 cdls->cdls_count = 0;
 473         }
 474
 475         return 0;
 476 }
 477 EXPORT_SYMBOL(libcfs_debug_vmsg2);
 478
 479 void
 480 cfs_trace_assertion_failed(const char *str,
 481                            struct libcfs_debug_msg_data *msgdata)
 482 {
 483         struct ptldebug_header hdr;
 484
 485         libcfs_panic_in_progress = 1;
 486         libcfs_catastrophe = 1;
 487         mb();
 488
 489         cfs_set_ptldebug_header(&hdr, msgdata, CDEBUG_STACK());
 490
 491         cfs_print_to_console(&hdr, D_EMERG, str, strlen(str),
 492                              msgdata->msg_file, msgdata->msg_fn);
 493
 494         panic("Lustre debug assertion failure\n");
 495
 496         /* not reached */
 497 }
 498
 499 static void
 500 panic_collect_pages(struct page_collection *pc)
 501 {
 502         /* Do the collect_pages job on a single CPU: assumes that all other
 503          * CPUs have been stopped during a panic.  If this isn't true for some
 504          * arch, this will have to be implemented separately in each arch.  */
 505         int                     i;
 506         int                     j;
 507         struct cfs_trace_cpu_data *tcd;
 508
 509         INIT_LIST_HEAD(&pc->pc_pages);
 510
 511         cfs_tcd_for_each(tcd, i, j) {
 512                 list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
 513                 tcd->tcd_cur_pages = 0;
 514
 515                 if (pc->pc_want_daemon_pages) {
 516                         list_splice_init(&tcd->tcd_daemon_pages,
 517                                              &pc->pc_pages);
 518                         tcd->tcd_cur_daemon_pages = 0;
 519                 }
 520         }
 521 }
 522
 523 static void collect_pages_on_all_cpus(struct page_collection *pc)
 524 {
 525         struct cfs_trace_cpu_data *tcd;
 526         int i, cpu;
 527
 528         spin_lock(&pc->pc_lock);
 529         for_each_possible_cpu(cpu) {
 530                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
 531                         list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
 532                         tcd->tcd_cur_pages = 0;
 533                         if (pc->pc_want_daemon_pages) {
 534                                 list_splice_init(&tcd->tcd_daemon_pages,
 535                                                      &pc->pc_pages);
 536                                 tcd->tcd_cur_daemon_pages = 0;
 537                         }
 538                 }
 539         }
 540         spin_unlock(&pc->pc_lock);
 541 }
 542
 543 static void collect_pages(struct page_collection *pc)
 544 {
 545         INIT_LIST_HEAD(&pc->pc_pages);
 546
 547         if (libcfs_panic_in_progress)
 548                 panic_collect_pages(pc);
 549         else
 550                 collect_pages_on_all_cpus(pc);
 551 }
 552
 553 static void put_pages_back_on_all_cpus(struct page_collection *pc)
 554 {
 555         struct cfs_trace_cpu_data *tcd;
 556         struct list_head *cur_head;
 557         struct cfs_trace_page *tage;
 558         struct cfs_trace_page *tmp;
 559         int i, cpu;
 560
 561         spin_lock(&pc->pc_lock);
 562         for_each_possible_cpu(cpu) {
 563                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
 564                         cur_head = tcd->tcd_pages.next;
 565
 566                         list_for_each_entry_safe(tage, tmp, &pc->pc_pages,
 567                                                  linkage) {
 568
 569                                 __LASSERT_TAGE_INVARIANT(tage);
 570
 571                                 if (tage->cpu != cpu || tage->type != i)
 572                                         continue;
 573
 574                                 cfs_tage_to_tail(tage, cur_head);
 575                                 tcd->tcd_cur_pages++;
 576                         }
 577                 }
 578         }
 579         spin_unlock(&pc->pc_lock);
 580 }
 581
 582 static void put_pages_back(struct page_collection *pc)
 583 {
 584         if (!libcfs_panic_in_progress)
 585                 put_pages_back_on_all_cpus(pc);
 586 }
 587
 588 /* Add pages to a per-cpu debug daemon ringbuffer.  This buffer makes sure that
 589  * we have a good amount of data at all times for dumping during an LBUG, even
 590  * if we have been steadily writing (and otherwise discarding) pages via the
 591  * debug daemon. */
 592 static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
 593                                          struct cfs_trace_cpu_data *tcd)
 594 {
 595         struct cfs_trace_page *tage;
 596         struct cfs_trace_page *tmp;
 597
 598         spin_lock(&pc->pc_lock);
 599         list_for_each_entry_safe(tage, tmp, &pc->pc_pages, linkage) {
 600
 601                 __LASSERT_TAGE_INVARIANT(tage);
 602
 603                 if (tage->cpu != tcd->tcd_cpu || tage->type != tcd->tcd_type)
 604                         continue;
 605
 606                 cfs_tage_to_tail(tage, &tcd->tcd_daemon_pages);
 607                 tcd->tcd_cur_daemon_pages++;
 608
 609                 if (tcd->tcd_cur_daemon_pages > tcd->tcd_max_pages) {
 610                         struct cfs_trace_page *victim;
 611
 612                         __LASSERT(!list_empty(&tcd->tcd_daemon_pages));
 613                         victim = cfs_tage_from_list(tcd->tcd_daemon_pages.next);
 614
 615                         __LASSERT_TAGE_INVARIANT(victim);
 616
 617                         list_del(&victim->linkage);
 618                         cfs_tage_free(victim);
 619                         tcd->tcd_cur_daemon_pages--;
 620                 }
 621         }
 622         spin_unlock(&pc->pc_lock);
 623 }
 624
 625 static void put_pages_on_daemon_list(struct page_collection *pc)
 626 {
 627         struct cfs_trace_cpu_data *tcd;
 628         int i, cpu;
 629
 630         for_each_possible_cpu(cpu) {
 631                 cfs_tcd_for_each_type_lock(tcd, i, cpu)
 632                         put_pages_on_tcd_daemon_list(pc, tcd);
 633         }
 634 }
 635
 636 void cfs_trace_debug_print(void)
 637 {
 638         struct page_collection pc;
 639         struct cfs_trace_page *tage;
 640         struct cfs_trace_page *tmp;
 641
 642         spin_lock_init(&pc.pc_lock);
 643
 644         pc.pc_want_daemon_pages = 1;
 645         collect_pages(&pc);
 646         list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
 647                 char *p, *file, *fn;
 648                 struct page *page;
 649
 650                 __LASSERT_TAGE_INVARIANT(tage);
 651
 652                 page = tage->page;
 653                 p = page_address(page);
 654                 while (p < ((char *)page_address(page) + tage->used)) {
 655                         struct ptldebug_header *hdr;
 656                         int len;
 657                         hdr = (void *)p;
 658                         p += sizeof(*hdr);
 659                         file = p;
 660                         p += strlen(file) + 1;
 661                         fn = p;
 662                         p += strlen(fn) + 1;
 663                         len = hdr->ph_len - (int)(p - (char *)hdr);
 664
 665                         cfs_print_to_console(hdr, D_EMERG, p, len, file, fn);
 666
 667                         p += len;
 668                 }
 669
 670                 list_del(&tage->linkage);
 671                 cfs_tage_free(tage);
 672         }
 673 }
 674
 675 int cfs_tracefile_dump_all_pages(char *filename)
 676 {
 677         struct page_collection  pc;
 678         struct file             *filp;
 679         struct cfs_trace_page   *tage;
 680         struct cfs_trace_page   *tmp;
 681         char                    *buf;
 682         int rc;
 683
 684         DECL_MMSPACE;
 685
 686         cfs_tracefile_write_lock();
 687
 688         filp = filp_open(filename, O_CREAT|O_EXCL|O_WRONLY|O_LARGEFILE, 0600);
 689         if (IS_ERR(filp)) {
 690                 rc = PTR_ERR(filp);
 691                 filp = NULL;
 692                 printk(KERN_ERR "LustreError: can't open %s for dump: rc %d\n",
 693                       filename, rc);
 694                 goto out;
 695         }
 696
 697         spin_lock_init(&pc.pc_lock);
 698         pc.pc_want_daemon_pages = 1;
 699         collect_pages(&pc);
 700         if (list_empty(&pc.pc_pages)) {
 701                 rc = 0;
 702                 goto close;
 703         }
 704
 705         /* ok, for now, just write the pages.  in the future we'll be building
 706          * iobufs with the pages and calling generic_direct_IO */
 707         MMSPACE_OPEN;
 708         list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
 709
 710                 __LASSERT_TAGE_INVARIANT(tage);
 711
 712                 buf = kmap(tage->page);
 713                 rc = vfs_write(filp, (__force const char __user *)buf,
 714                                tage->used, &filp->f_pos);
 715                 kunmap(tage->page);
 716
 717                 if (rc != (int)tage->used) {
 718                         printk(KERN_WARNING "wanted to write %u but wrote "
 719                                "%d\n", tage->used, rc);
 720                         put_pages_back(&pc);
 721                         __LASSERT(list_empty(&pc.pc_pages));
 722                         break;
 723                 }
 724                 list_del(&tage->linkage);
 725                 cfs_tage_free(tage);
 726         }
 727         MMSPACE_CLOSE;
 728         rc = vfs_fsync(filp, 1);
 729         if (rc)
 730                 printk(KERN_ERR "sync returns %d\n", rc);
 731 close:
 732         filp_close(filp, NULL);
 733 out:
 734         cfs_tracefile_write_unlock();
 735         return rc;
 736 }
 737
 738 void cfs_trace_flush_pages(void)
 739 {
 740         struct page_collection pc;
 741         struct cfs_trace_page *tage;
 742         struct cfs_trace_page *tmp;
 743
 744         spin_lock_init(&pc.pc_lock);
 745
 746         pc.pc_want_daemon_pages = 1;
 747         collect_pages(&pc);
 748         list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
 749
 750                 __LASSERT_TAGE_INVARIANT(tage);
 751
 752                 list_del(&tage->linkage);
 753                 cfs_tage_free(tage);
 754         }
 755 }
 756
 757 int cfs_trace_copyin_string(char *knl_buffer, int knl_buffer_nob,
 758                             const char __user *usr_buffer, int usr_buffer_nob)
 759 {
 760         int    nob;
 761
 762         if (usr_buffer_nob > knl_buffer_nob)
 763                 return -EOVERFLOW;
 764
 765         if (copy_from_user((void *)knl_buffer,
 766                            usr_buffer, usr_buffer_nob))
 767                 return -EFAULT;
 768
 769         nob = strnlen(knl_buffer, usr_buffer_nob);
 770         while (nob-- >= 0)                    /* strip trailing whitespace */
 771                 if (!isspace(knl_buffer[nob]))
 772                         break;
 773
 774         if (nob < 0)                        /* empty string */
 775                 return -EINVAL;
 776
 777         if (nob == knl_buffer_nob)            /* no space to terminate */
 778                 return -EOVERFLOW;
 779
 780         knl_buffer[nob + 1] = 0;                /* terminate */
 781         return 0;
 782 }
 783 EXPORT_SYMBOL(cfs_trace_copyin_string);
 784
 785 int cfs_trace_copyout_string(char __user *usr_buffer, int usr_buffer_nob,
 786                              const char *knl_buffer, char *append)
 787 {
 788         /* NB if 'append' != NULL, it's a single character to append to the
 789          * copied out string - usually "\n", for /proc entries and "" (i.e. a
 790          * terminating zero byte) for sysctl entries */
 791         int   nob = strlen(knl_buffer);
 792
 793         if (nob > usr_buffer_nob)
 794                 nob = usr_buffer_nob;
 795
 796         if (copy_to_user(usr_buffer, knl_buffer, nob))
 797                 return -EFAULT;
 798
 799         if (append != NULL && nob < usr_buffer_nob) {
 800                 if (copy_to_user(usr_buffer + nob, append, 1))
 801                         return -EFAULT;
 802
 803                 nob++;
 804         }
 805
 806         return nob;
 807 }
 808 EXPORT_SYMBOL(cfs_trace_copyout_string);
 809
 810 int cfs_trace_allocate_string_buffer(char **str, int nob)
 811 {
 812         if (nob > 2 * PAGE_CACHE_SIZE)      /* string must be "sensible" */
 813                 return -EINVAL;
 814
 815         *str = kmalloc(nob, GFP_IOFS | __GFP_ZERO);
 816         if (*str == NULL)
 817                 return -ENOMEM;
 818
 819         return 0;
 820 }
 821
 822 void cfs_trace_free_string_buffer(char *str, int nob)
 823 {
 824         kfree(str);
 825 }
 826
 827 int cfs_trace_dump_debug_buffer_usrstr(void __user *usr_str, int usr_str_nob)
 828 {
 829         char     *str;
 830         int        rc;
 831
 832         rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
 833         if (rc != 0)
 834                 return rc;
 835
 836         rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
 837                                      usr_str, usr_str_nob);
 838         if (rc != 0)
 839                 goto out;
 840
 841         if (str[0] != '/') {
 842                 rc = -EINVAL;
 843                 goto out;
 844         }
 845         rc = cfs_tracefile_dump_all_pages(str);
 846 out:
 847         cfs_trace_free_string_buffer(str, usr_str_nob + 1);
 848         return rc;
 849 }
 850
 851 int cfs_trace_daemon_command(char *str)
 852 {
 853         int       rc = 0;
 854
 855         cfs_tracefile_write_lock();
 856
 857         if (strcmp(str, "stop") == 0) {
 858                 cfs_tracefile_write_unlock();
 859                 cfs_trace_stop_thread();
 860                 cfs_tracefile_write_lock();
 861                 memset(cfs_tracefile, 0, sizeof(cfs_tracefile));
 862
 863         } else if (strncmp(str, "size=", 5) == 0) {
 864                 cfs_tracefile_size = simple_strtoul(str + 5, NULL, 0);
 865                 if (cfs_tracefile_size < 10 || cfs_tracefile_size > 20480)
 866                         cfs_tracefile_size = CFS_TRACEFILE_SIZE;
 867                 else
 868                         cfs_tracefile_size <<= 20;
 869
 870         } else if (strlen(str) >= sizeof(cfs_tracefile)) {
 871                 rc = -ENAMETOOLONG;
 872         } else if (str[0] != '/') {
 873                 rc = -EINVAL;
 874         } else {
 875                 strcpy(cfs_tracefile, str);
 876
 877                 printk(KERN_INFO
 878                        "Lustre: debug daemon will attempt to start writing "
 879                        "to %s (%lukB max)\n", cfs_tracefile,
 880                        (long)(cfs_tracefile_size >> 10));
 881
 882                 cfs_trace_start_thread();
 883         }
 884
 885         cfs_tracefile_write_unlock();
 886         return rc;
 887 }
 888
 889 int cfs_trace_daemon_command_usrstr(void __user *usr_str, int usr_str_nob)
 890 {
 891         char *str;
 892         int   rc;
 893
 894         rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
 895         if (rc != 0)
 896                 return rc;
 897
 898         rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
 899                                  usr_str, usr_str_nob);
 900         if (rc == 0)
 901                 rc = cfs_trace_daemon_command(str);
 902
 903         cfs_trace_free_string_buffer(str, usr_str_nob + 1);
 904         return rc;
 905 }
 906
 907 int cfs_trace_set_debug_mb(int mb)
 908 {
 909         int i;
 910         int j;
 911         int pages;
 912         int limit = cfs_trace_max_debug_mb();
 913         struct cfs_trace_cpu_data *tcd;
 914
 915         if (mb < num_possible_cpus()) {
 916                 printk(KERN_WARNING
 917                        "Lustre: %d MB is too small for debug buffer size, "
 918                        "setting it to %d MB.\n", mb, num_possible_cpus());
 919                 mb = num_possible_cpus();
 920         }
 921
 922         if (mb > limit) {
 923                 printk(KERN_WARNING
 924                        "Lustre: %d MB is too large for debug buffer size, "
 925                        "setting it to %d MB.\n", mb, limit);
 926                 mb = limit;
 927         }
 928
 929         mb /= num_possible_cpus();
 930         pages = mb << (20 - PAGE_CACHE_SHIFT);
 931
 932         cfs_tracefile_write_lock();
 933
 934         cfs_tcd_for_each(tcd, i, j)
 935                 tcd->tcd_max_pages = (pages * tcd->tcd_pages_factor) / 100;
 936
 937         cfs_tracefile_write_unlock();
 938
 939         return 0;
 940 }
 941
 942 int cfs_trace_set_debug_mb_usrstr(void __user *usr_str, int usr_str_nob)
 943 {
 944         char     str[32];
 945         int      rc;
 946
 947         rc = cfs_trace_copyin_string(str, sizeof(str), usr_str, usr_str_nob);
 948         if (rc < 0)
 949                 return rc;
 950
 951         return cfs_trace_set_debug_mb(simple_strtoul(str, NULL, 0));
 952 }
 953
 954 int cfs_trace_get_debug_mb(void)
 955 {
 956         int i;
 957         int j;
 958         struct cfs_trace_cpu_data *tcd;
 959         int total_pages = 0;
 960
 961         cfs_tracefile_read_lock();
 962
 963         cfs_tcd_for_each(tcd, i, j)
 964                 total_pages += tcd->tcd_max_pages;
 965
 966         cfs_tracefile_read_unlock();
 967
 968         return (total_pages >> (20 - PAGE_CACHE_SHIFT)) + 1;
 969 }
 970
 971 static int tracefiled(void *arg)
 972 {
 973         struct page_collection pc;
 974         struct tracefiled_ctl *tctl = arg;
 975         struct cfs_trace_page *tage;
 976         struct cfs_trace_page *tmp;
 977         struct file *filp;
 978         char *buf;
 979         int last_loop = 0;
 980         int rc;
 981
 982         DECL_MMSPACE;
 983
 984         /* we're started late enough that we pick up init's fs context */
 985         /* this is so broken in uml?  what on earth is going on? */
 986
 987         spin_lock_init(&pc.pc_lock);
 988         complete(&tctl->tctl_start);
 989
 990         while (1) {
 991                 wait_queue_t __wait;
 992
 993                 pc.pc_want_daemon_pages = 0;
 994                 collect_pages(&pc);
 995                 if (list_empty(&pc.pc_pages))
 996                         goto end_loop;
 997
 998                 filp = NULL;
 999                 cfs_tracefile_read_lock();
1000                 if (cfs_tracefile[0] != 0) {
1001                         filp = filp_open(cfs_tracefile,
1002                                          O_CREAT | O_RDWR | O_LARGEFILE,
1003                                          0600);
1004                         if (IS_ERR(filp)) {
1005                                 rc = PTR_ERR(filp);
1006                                 filp = NULL;
1007                                 printk(KERN_WARNING "couldn't open %s: "
1008                                        "%d\n", cfs_tracefile, rc);
1009                         }
1010                 }
1011                 cfs_tracefile_read_unlock();
1012                 if (filp == NULL) {
1013                         put_pages_on_daemon_list(&pc);
1014                         __LASSERT(list_empty(&pc.pc_pages));
1015                         goto end_loop;
1016                 }
1017
1018                 MMSPACE_OPEN;
1019
1020                 list_for_each_entry_safe(tage, tmp, &pc.pc_pages,
1021                                                    linkage) {
1022                         static loff_t f_pos;
1023
1024                         __LASSERT_TAGE_INVARIANT(tage);
1025
1026                         if (f_pos >= (off_t)cfs_tracefile_size)
1027                                 f_pos = 0;
1028                         else if (f_pos > i_size_read(filp->f_dentry->d_inode))
1029                                 f_pos = i_size_read(filp->f_dentry->d_inode);
1030
1031                         buf = kmap(tage->page);
1032                         rc = vfs_write(filp, (__force const char __user *)buf,
1033                                        tage->used, &f_pos);
1034                         kunmap(tage->page);
1035
1036                         if (rc != (int)tage->used) {
1037                                 printk(KERN_WARNING "wanted to write %u "
1038                                        "but wrote %d\n", tage->used, rc);
1039                                 put_pages_back(&pc);
1040                                 __LASSERT(list_empty(&pc.pc_pages));
1041                         }
1042                 }
1043                 MMSPACE_CLOSE;
1044
1045                 filp_close(filp, NULL);
1046                 put_pages_on_daemon_list(&pc);
1047                 if (!list_empty(&pc.pc_pages)) {
1048                         int i;
1049
1050                         printk(KERN_ALERT "Lustre: trace pages aren't "
1051                                " empty\n");
1052                         printk(KERN_ERR "total cpus(%d): ",
1053                                num_possible_cpus());
1054                         for (i = 0; i < num_possible_cpus(); i++)
1055                                 if (cpu_online(i))
1056                                         printk(KERN_ERR "%d(on) ", i);
1057                                 else
1058                                         printk(KERN_ERR "%d(off) ", i);
1059                         printk(KERN_ERR "\n");
1060
1061                         i = 0;
1062                         list_for_each_entry_safe(tage, tmp, &pc.pc_pages,
1063                                                      linkage)
1064                                 printk(KERN_ERR "page %d belongs to cpu "
1065                                        "%d\n", ++i, tage->cpu);
1066                         printk(KERN_ERR "There are %d pages unwritten\n",
1067                                i);
1068                 }
1069                 __LASSERT(list_empty(&pc.pc_pages));
1070 end_loop:
1071                 if (atomic_read(&tctl->tctl_shutdown)) {
1072                         if (last_loop == 0) {
1073                                 last_loop = 1;
1074                                 continue;
1075                         } else {
1076                                 break;
1077                         }
1078                 }
1079                 init_waitqueue_entry(&__wait, current);
1080                 add_wait_queue(&tctl->tctl_waitq, &__wait);
1081                 set_current_state(TASK_INTERRUPTIBLE);
1082                 schedule_timeout(cfs_time_seconds(1));
1083                 remove_wait_queue(&tctl->tctl_waitq, &__wait);
1084         }
1085         complete(&tctl->tctl_stop);
1086         return 0;
1087 }
1088
1089 int cfs_trace_start_thread(void)
1090 {
1091         struct tracefiled_ctl *tctl = &trace_tctl;
1092         int rc = 0;
1093
1094         mutex_lock(&cfs_trace_thread_mutex);
1095         if (thread_running)
1096                 goto out;
1097
1098         init_completion(&tctl->tctl_start);
1099         init_completion(&tctl->tctl_stop);
1100         init_waitqueue_head(&tctl->tctl_waitq);
1101         atomic_set(&tctl->tctl_shutdown, 0);
1102
1103         if (IS_ERR(kthread_run(tracefiled, tctl, "ktracefiled"))) {
1104                 rc = -ECHILD;
1105                 goto out;
1106         }
1107
1108         wait_for_completion(&tctl->tctl_start);
1109         thread_running = 1;
1110 out:
1111         mutex_unlock(&cfs_trace_thread_mutex);
1112         return rc;
1113 }
1114
1115 void cfs_trace_stop_thread(void)
1116 {
1117         struct tracefiled_ctl *tctl = &trace_tctl;
1118
1119         mutex_lock(&cfs_trace_thread_mutex);
1120         if (thread_running) {
1121                 printk(KERN_INFO
1122                        "Lustre: shutting down debug daemon thread...\n");
1123                 atomic_set(&tctl->tctl_shutdown, 1);
1124                 wait_for_completion(&tctl->tctl_stop);
1125                 thread_running = 0;
1126         }
1127         mutex_unlock(&cfs_trace_thread_mutex);
1128 }
1129
1130 int cfs_tracefile_init(int max_pages)
1131 {
1132         struct cfs_trace_cpu_data *tcd;
1133         int                 i;
1134         int                 j;
1135         int                 rc;
1136         int                 factor;
1137
1138         rc = cfs_tracefile_init_arch();
1139         if (rc != 0)
1140                 return rc;
1141
1142         cfs_tcd_for_each(tcd, i, j) {
1143                 /* tcd_pages_factor is initialized int tracefile_init_arch. */
1144                 factor = tcd->tcd_pages_factor;
1145                 INIT_LIST_HEAD(&tcd->tcd_pages);
1146                 INIT_LIST_HEAD(&tcd->tcd_stock_pages);
1147                 INIT_LIST_HEAD(&tcd->tcd_daemon_pages);
1148                 tcd->tcd_cur_pages = 0;
1149                 tcd->tcd_cur_stock_pages = 0;
1150                 tcd->tcd_cur_daemon_pages = 0;
1151                 tcd->tcd_max_pages = (max_pages * factor) / 100;
1152                 LASSERT(tcd->tcd_max_pages > 0);
1153                 tcd->tcd_shutting_down = 0;
1154         }
1155
1156         return 0;
1157 }
1158
1159 static void trace_cleanup_on_all_cpus(void)
1160 {
1161         struct cfs_trace_cpu_data *tcd;
1162         struct cfs_trace_page *tage;
1163         struct cfs_trace_page *tmp;
1164         int i, cpu;
1165
1166         for_each_possible_cpu(cpu) {
1167                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
1168                         tcd->tcd_shutting_down = 1;
1169
1170                         list_for_each_entry_safe(tage, tmp, &tcd->tcd_pages,
1171                                                            linkage) {
1172                                 __LASSERT_TAGE_INVARIANT(tage);
1173
1174                                 list_del(&tage->linkage);
1175                                 cfs_tage_free(tage);
1176                         }
1177
1178                         tcd->tcd_cur_pages = 0;
1179                 }
1180         }
1181 }
1182
1183 static void cfs_trace_cleanup(void)
1184 {
1185         struct page_collection pc;
1186
1187         INIT_LIST_HEAD(&pc.pc_pages);
1188         spin_lock_init(&pc.pc_lock);
1189
1190         trace_cleanup_on_all_cpus();
1191
1192         cfs_tracefile_fini_arch();
1193 }
1194
1195 void cfs_tracefile_exit(void)
1196 {
1197         cfs_trace_stop_thread();
1198         cfs_trace_cleanup();
1199 }