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 #define DEBUG_SUBSYSTEM S_LNET
  43 #define LUSTRE_TRACEFILE_PRIVATE
  44 #include "tracefile.h"
  45
  46 #include "../../include/linux/libcfs/libcfs.h"
  47
  48 /* XXX move things up to the top, comment */
  49 union cfs_trace_data_union (*cfs_trace_data[TCD_MAX_TYPES])[NR_CPUS] __cacheline_aligned;
  50
  51 char cfs_tracefile[TRACEFILE_NAME_SIZE];
  52 long long cfs_tracefile_size = CFS_TRACEFILE_SIZE;
  53 static struct tracefiled_ctl trace_tctl;
  54 static DEFINE_MUTEX(cfs_trace_thread_mutex);
  55 static int thread_running;
  56
  57 static atomic_t cfs_tage_allocated = ATOMIC_INIT(0);
  58
  59 static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
  60                                          struct cfs_trace_cpu_data *tcd);
  61
  62 static inline struct cfs_trace_page *
  63 cfs_tage_from_list(struct list_head *list)
  64 {
  65         return list_entry(list, struct cfs_trace_page, linkage);
  66 }
  67
  68 static struct cfs_trace_page *cfs_tage_alloc(gfp_t gfp)
  69 {
  70         struct page         *page;
  71         struct cfs_trace_page *tage;
  72
  73         /* My caller is trying to free memory */
  74         if (!in_interrupt() && memory_pressure_get())
  75                 return NULL;
  76
  77         /*
  78          * Don't spam console with allocation failures: they will be reported
  79          * by upper layer anyway.
  80          */
  81         gfp |= __GFP_NOWARN;
  82         page = alloc_page(gfp);
  83         if (page == NULL)
  84                 return NULL;
  85
  86         tage = kmalloc(sizeof(*tage), gfp);
  87         if (tage == NULL) {
  88                 __free_page(page);
  89                 return NULL;
  90         }
  91
  92         tage->page = page;
  93         atomic_inc(&cfs_tage_allocated);
  94         return tage;
  95 }
  96
  97 static void cfs_tage_free(struct cfs_trace_page *tage)
  98 {
  99         __LASSERT(tage != NULL);
 100         __LASSERT(tage->page != NULL);
 101
 102         __free_page(tage->page);
 103         kfree(tage);
 104         atomic_dec(&cfs_tage_allocated);
 105 }
 106
 107 static void cfs_tage_to_tail(struct cfs_trace_page *tage,
 108                              struct list_head *queue)
 109 {
 110         __LASSERT(tage != NULL);
 111         __LASSERT(queue != NULL);
 112
 113         list_move_tail(&tage->linkage, queue);
 114 }
 115
 116 int cfs_trace_refill_stock(struct cfs_trace_cpu_data *tcd, gfp_t gfp,
 117                            struct list_head *stock)
 118 {
 119         int i;
 120
 121         /*
 122          * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
 123          * from here: this will lead to infinite recursion.
 124          */
 125
 126         for (i = 0; i + tcd->tcd_cur_stock_pages < TCD_STOCK_PAGES ; ++i) {
 127                 struct cfs_trace_page *tage;
 128
 129                 tage = cfs_tage_alloc(gfp);
 130                 if (tage == NULL)
 131                         break;
 132                 list_add_tail(&tage->linkage, stock);
 133         }
 134         return i;
 135 }
 136
 137 /* return a page that has 'len' bytes left at the end */
 138 static struct cfs_trace_page *
 139 cfs_trace_get_tage_try(struct cfs_trace_cpu_data *tcd, unsigned long len)
 140 {
 141         struct cfs_trace_page *tage;
 142
 143         if (tcd->tcd_cur_pages > 0) {
 144                 __LASSERT(!list_empty(&tcd->tcd_pages));
 145                 tage = cfs_tage_from_list(tcd->tcd_pages.prev);
 146                 if (tage->used + len <= PAGE_CACHE_SIZE)
 147                         return tage;
 148         }
 149
 150         if (tcd->tcd_cur_pages < tcd->tcd_max_pages) {
 151                 if (tcd->tcd_cur_stock_pages > 0) {
 152                         tage = cfs_tage_from_list(tcd->tcd_stock_pages.prev);
 153                         --tcd->tcd_cur_stock_pages;
 154                         list_del_init(&tage->linkage);
 155                 } else {
 156                         tage = cfs_tage_alloc(GFP_ATOMIC);
 157                         if (unlikely(tage == NULL)) {
 158                                 if ((!memory_pressure_get() ||
 159                                      in_interrupt()) && printk_ratelimit())
 160                                         printk(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 = smp_processor_id();
 169                 tage->type = tcd->tcd_type;
 170                 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                         wake_up(&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(KERN_WARNING "debug daemon buffer overflowed; discarding 10%% of pages (%d of %ld)\n",
 199                        pgcount + 1, tcd->tcd_cur_pages);
 200
 201         INIT_LIST_HEAD(&pc.pc_pages);
 202
 203         list_for_each_entry_safe(tage, tmp, &tcd->tcd_pages, linkage) {
 204                 if (pgcount-- == 0)
 205                         break;
 206
 207                 list_move_tail(&tage->linkage, &pc.pc_pages);
 208                 tcd->tcd_cur_pages--;
 209         }
 210         put_pages_on_tcd_daemon_list(&pc, tcd);
 211 }
 212
 213 /* return a page that has 'len' bytes left at the end */
 214 static struct cfs_trace_page *cfs_trace_get_tage(struct cfs_trace_cpu_data *tcd,
 215                                                  unsigned long len)
 216 {
 217         struct cfs_trace_page *tage;
 218
 219         /*
 220          * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
 221          * from here: this will lead to infinite recursion.
 222          */
 223
 224         if (len > PAGE_CACHE_SIZE) {
 225                 pr_err("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_msg(struct libcfs_debug_msg_data *msgdata,
 243                      const char *format, ...)
 244 {
 245         va_list args;
 246         int     rc;
 247
 248         va_start(args, format);
 249         rc = libcfs_debug_vmsg2(msgdata, format, args, NULL);
 250         va_end(args);
 251
 252         return rc;
 253 }
 254 EXPORT_SYMBOL(libcfs_debug_msg);
 255
 256 int libcfs_debug_vmsg2(struct libcfs_debug_msg_data *msgdata,
 257                        const char *format1, va_list args,
 258                        const char *format2, ...)
 259 {
 260         struct cfs_trace_cpu_data *tcd = NULL;
 261         struct ptldebug_header     header = {0};
 262         struct cfs_trace_page     *tage;
 263         /* string_buf is used only if tcd != NULL, and is always set then */
 264         char                  *string_buf = NULL;
 265         char                  *debug_buf;
 266         int                     known_size;
 267         int                     needed = 85; /* average message length */
 268         int                     max_nob;
 269         va_list             ap;
 270         int                     depth;
 271         int                     i;
 272         int                     remain;
 273         int                     mask = msgdata->msg_mask;
 274         const char              *file = kbasename(msgdata->msg_file);
 275         struct cfs_debug_limit_state   *cdls = msgdata->msg_cdls;
 276
 277         tcd = cfs_trace_get_tcd();
 278
 279         /* cfs_trace_get_tcd() grabs a lock, which disables preemption and
 280          * pins us to a particular CPU.  This avoids an smp_processor_id()
 281          * warning on Linux when debugging is enabled. */
 282         cfs_set_ptldebug_header(&header, msgdata, CDEBUG_STACK());
 283
 284         if (tcd == NULL)                /* arch may not log in IRQ context */
 285                 goto console;
 286
 287         if (tcd->tcd_cur_pages == 0)
 288                 header.ph_flags |= PH_FLAG_FIRST_RECORD;
 289
 290         if (tcd->tcd_shutting_down) {
 291                 cfs_trace_put_tcd(tcd);
 292                 tcd = NULL;
 293                 goto console;
 294         }
 295
 296         depth = __current_nesting_level();
 297         known_size = strlen(file) + 1 + depth;
 298         if (msgdata->msg_fn)
 299                 known_size += strlen(msgdata->msg_fn) + 1;
 300
 301         if (libcfs_debug_binary)
 302                 known_size += sizeof(header);
 303
 304         /*/
 305          * '2' used because vsnprintf return real size required for output
 306          * _without_ terminating NULL.
 307          * if needed is to small for this format.
 308          */
 309         for (i = 0; i < 2; i++) {
 310                 tage = cfs_trace_get_tage(tcd, needed + known_size + 1);
 311                 if (tage == NULL) {
 312                         if (needed + known_size > PAGE_CACHE_SIZE)
 313                                 mask |= D_ERROR;
 314
 315                         cfs_trace_put_tcd(tcd);
 316                         tcd = NULL;
 317                         goto console;
 318                 }
 319
 320                 string_buf = (char *)page_address(tage->page) +
 321                                         tage->used + known_size;
 322
 323                 max_nob = PAGE_CACHE_SIZE - tage->used - known_size;
 324                 if (max_nob <= 0) {
 325                         printk(KERN_EMERG "negative max_nob: %d\n",
 326                                max_nob);
 327                         mask |= D_ERROR;
 328                         cfs_trace_put_tcd(tcd);
 329                         tcd = NULL;
 330                         goto console;
 331                 }
 332
 333                 needed = 0;
 334                 if (format1) {
 335                         va_copy(ap, args);
 336                         needed = vsnprintf(string_buf, max_nob, format1, ap);
 337                         va_end(ap);
 338                 }
 339
 340                 if (format2) {
 341                         remain = max_nob - needed;
 342                         if (remain < 0)
 343                                 remain = 0;
 344
 345                         va_start(ap, format2);
 346                         needed += vsnprintf(string_buf + needed, remain,
 347                                             format2, ap);
 348                         va_end(ap);
 349                 }
 350
 351                 if (needed < max_nob) /* well. printing ok.. */
 352                         break;
 353         }
 354
 355         if (*(string_buf+needed-1) != '\n')
 356                 printk(KERN_INFO "format at %s:%d:%s doesn't end in newline\n",
 357                        file, msgdata->msg_line, msgdata->msg_fn);
 358
 359         header.ph_len = known_size + needed;
 360         debug_buf = (char *)page_address(tage->page) + tage->used;
 361
 362         if (libcfs_debug_binary) {
 363                 memcpy(debug_buf, &header, sizeof(header));
 364                 tage->used += sizeof(header);
 365                 debug_buf += sizeof(header);
 366         }
 367
 368         /* indent message according to the nesting level */
 369         while (depth-- > 0) {
 370                 *(debug_buf++) = '.';
 371                 ++tage->used;
 372         }
 373
 374         strcpy(debug_buf, file);
 375         tage->used += strlen(file) + 1;
 376         debug_buf += strlen(file) + 1;
 377
 378         if (msgdata->msg_fn) {
 379                 strcpy(debug_buf, msgdata->msg_fn);
 380                 tage->used += strlen(msgdata->msg_fn) + 1;
 381                 debug_buf += strlen(msgdata->msg_fn) + 1;
 382         }
 383
 384         __LASSERT(debug_buf == string_buf);
 385
 386         tage->used += needed;
 387         __LASSERT (tage->used <= PAGE_CACHE_SIZE);
 388
 389 console:
 390         if ((mask & libcfs_printk) == 0) {
 391                 /* no console output requested */
 392                 if (tcd != NULL)
 393                         cfs_trace_put_tcd(tcd);
 394                 return 1;
 395         }
 396
 397         if (cdls != NULL) {
 398                 if (libcfs_console_ratelimit &&
 399                     cdls->cdls_next != 0 &&     /* not first time ever */
 400                     !cfs_time_after(cfs_time_current(), cdls->cdls_next)) {
 401                         /* skipping a console message */
 402                         cdls->cdls_count++;
 403                         if (tcd != NULL)
 404                                 cfs_trace_put_tcd(tcd);
 405                         return 1;
 406                 }
 407
 408                 if (cfs_time_after(cfs_time_current(), cdls->cdls_next +
 409                                                        libcfs_console_max_delay
 410                                                        + cfs_time_seconds(10))) {
 411                         /* last timeout was a long time ago */
 412                         cdls->cdls_delay /= libcfs_console_backoff * 4;
 413                 } else {
 414                         cdls->cdls_delay *= libcfs_console_backoff;
 415                 }
 416
 417                 if (cdls->cdls_delay < libcfs_console_min_delay)
 418                         cdls->cdls_delay = libcfs_console_min_delay;
 419                 else if (cdls->cdls_delay > libcfs_console_max_delay)
 420                         cdls->cdls_delay = libcfs_console_max_delay;
 421
 422                 /* ensure cdls_next is never zero after it's been seen */
 423                 cdls->cdls_next = (cfs_time_current() + cdls->cdls_delay) | 1;
 424         }
 425
 426         if (tcd != NULL) {
 427                 cfs_print_to_console(&header, mask, string_buf, needed, file,
 428                                      msgdata->msg_fn);
 429                 cfs_trace_put_tcd(tcd);
 430         } else {
 431                 string_buf = cfs_trace_get_console_buffer();
 432
 433                 needed = 0;
 434                 if (format1 != NULL) {
 435                         va_copy(ap, args);
 436                         needed = vsnprintf(string_buf,
 437                                            CFS_TRACE_CONSOLE_BUFFER_SIZE,
 438                                            format1, ap);
 439                         va_end(ap);
 440                 }
 441                 if (format2 != NULL) {
 442                         remain = CFS_TRACE_CONSOLE_BUFFER_SIZE - needed;
 443                         if (remain > 0) {
 444                                 va_start(ap, format2);
 445                                 needed += vsnprintf(string_buf+needed, remain,
 446                                                     format2, ap);
 447                                 va_end(ap);
 448                         }
 449                 }
 450                 cfs_print_to_console(&header, mask,
 451                                      string_buf, needed, file, msgdata->msg_fn);
 452
 453                 put_cpu();
 454         }
 455
 456         if (cdls != NULL && cdls->cdls_count != 0) {
 457                 string_buf = cfs_trace_get_console_buffer();
 458
 459                 needed = snprintf(string_buf, CFS_TRACE_CONSOLE_BUFFER_SIZE,
 460                                   "Skipped %d previous similar message%s\n",
 461                                   cdls->cdls_count,
 462                                   (cdls->cdls_count > 1) ? "s" : "");
 463
 464                 cfs_print_to_console(&header, mask,
 465                                      string_buf, needed, file, msgdata->msg_fn);
 466
 467                 put_cpu();
 468                 cdls->cdls_count = 0;
 469         }
 470
 471         return 0;
 472 }
 473 EXPORT_SYMBOL(libcfs_debug_vmsg2);
 474
 475 void
 476 cfs_trace_assertion_failed(const char *str,
 477                            struct libcfs_debug_msg_data *msgdata)
 478 {
 479         struct ptldebug_header hdr;
 480
 481         libcfs_panic_in_progress = 1;
 482         libcfs_catastrophe = 1;
 483         mb();
 484
 485         cfs_set_ptldebug_header(&hdr, msgdata, CDEBUG_STACK());
 486
 487         cfs_print_to_console(&hdr, D_EMERG, str, strlen(str),
 488                              msgdata->msg_file, msgdata->msg_fn);
 489
 490         panic("Lustre debug assertion failure\n");
 491
 492         /* not reached */
 493 }
 494
 495 static void
 496 panic_collect_pages(struct page_collection *pc)
 497 {
 498         /* Do the collect_pages job on a single CPU: assumes that all other
 499          * CPUs have been stopped during a panic.  If this isn't true for some
 500          * arch, this will have to be implemented separately in each arch.  */
 501         int                     i;
 502         int                     j;
 503         struct cfs_trace_cpu_data *tcd;
 504
 505         INIT_LIST_HEAD(&pc->pc_pages);
 506
 507         cfs_tcd_for_each(tcd, i, j) {
 508                 list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
 509                 tcd->tcd_cur_pages = 0;
 510
 511                 if (pc->pc_want_daemon_pages) {
 512                         list_splice_init(&tcd->tcd_daemon_pages,
 513                                              &pc->pc_pages);
 514                         tcd->tcd_cur_daemon_pages = 0;
 515                 }
 516         }
 517 }
 518
 519 static void collect_pages_on_all_cpus(struct page_collection *pc)
 520 {
 521         struct cfs_trace_cpu_data *tcd;
 522         int i, cpu;
 523
 524         for_each_possible_cpu(cpu) {
 525                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
 526                         list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
 527                         tcd->tcd_cur_pages = 0;
 528                         if (pc->pc_want_daemon_pages) {
 529                                 list_splice_init(&tcd->tcd_daemon_pages,
 530                                                      &pc->pc_pages);
 531                                 tcd->tcd_cur_daemon_pages = 0;
 532                         }
 533                 }
 534         }
 535 }
 536
 537 static void collect_pages(struct page_collection *pc)
 538 {
 539         INIT_LIST_HEAD(&pc->pc_pages);
 540
 541         if (libcfs_panic_in_progress)
 542                 panic_collect_pages(pc);
 543         else
 544                 collect_pages_on_all_cpus(pc);
 545 }
 546
 547 static void put_pages_back_on_all_cpus(struct page_collection *pc)
 548 {
 549         struct cfs_trace_cpu_data *tcd;
 550         struct list_head *cur_head;
 551         struct cfs_trace_page *tage;
 552         struct cfs_trace_page *tmp;
 553         int i, cpu;
 554
 555         for_each_possible_cpu(cpu) {
 556                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
 557                         cur_head = tcd->tcd_pages.next;
 558
 559                         list_for_each_entry_safe(tage, tmp, &pc->pc_pages,
 560                                                  linkage) {
 561
 562                                 __LASSERT_TAGE_INVARIANT(tage);
 563
 564                                 if (tage->cpu != cpu || tage->type != i)
 565                                         continue;
 566
 567                                 cfs_tage_to_tail(tage, cur_head);
 568                                 tcd->tcd_cur_pages++;
 569                         }
 570                 }
 571         }
 572 }
 573
 574 static void put_pages_back(struct page_collection *pc)
 575 {
 576         if (!libcfs_panic_in_progress)
 577                 put_pages_back_on_all_cpus(pc);
 578 }
 579
 580 /* Add pages to a per-cpu debug daemon ringbuffer.  This buffer makes sure that
 581  * we have a good amount of data at all times for dumping during an LBUG, even
 582  * if we have been steadily writing (and otherwise discarding) pages via the
 583  * debug daemon. */
 584 static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
 585                                          struct cfs_trace_cpu_data *tcd)
 586 {
 587         struct cfs_trace_page *tage;
 588         struct cfs_trace_page *tmp;
 589
 590         list_for_each_entry_safe(tage, tmp, &pc->pc_pages, linkage) {
 591
 592                 __LASSERT_TAGE_INVARIANT(tage);
 593
 594                 if (tage->cpu != tcd->tcd_cpu || tage->type != tcd->tcd_type)
 595                         continue;
 596
 597                 cfs_tage_to_tail(tage, &tcd->tcd_daemon_pages);
 598                 tcd->tcd_cur_daemon_pages++;
 599
 600                 if (tcd->tcd_cur_daemon_pages > tcd->tcd_max_pages) {
 601                         struct cfs_trace_page *victim;
 602
 603                         __LASSERT(!list_empty(&tcd->tcd_daemon_pages));
 604                         victim = cfs_tage_from_list(tcd->tcd_daemon_pages.next);
 605
 606                         __LASSERT_TAGE_INVARIANT(victim);
 607
 608                         list_del(&victim->linkage);
 609                         cfs_tage_free(victim);
 610                         tcd->tcd_cur_daemon_pages--;
 611                 }
 612         }
 613 }
 614
 615 static void put_pages_on_daemon_list(struct page_collection *pc)
 616 {
 617         struct cfs_trace_cpu_data *tcd;
 618         int i, cpu;
 619
 620         for_each_possible_cpu(cpu) {
 621                 cfs_tcd_for_each_type_lock(tcd, i, cpu)
 622                         put_pages_on_tcd_daemon_list(pc, tcd);
 623         }
 624 }
 625
 626 void cfs_trace_debug_print(void)
 627 {
 628         struct page_collection pc;
 629         struct cfs_trace_page *tage;
 630         struct cfs_trace_page *tmp;
 631
 632         pc.pc_want_daemon_pages = 1;
 633         collect_pages(&pc);
 634         list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
 635                 char *p, *file, *fn;
 636                 struct page *page;
 637
 638                 __LASSERT_TAGE_INVARIANT(tage);
 639
 640                 page = tage->page;
 641                 p = page_address(page);
 642                 while (p < ((char *)page_address(page) + tage->used)) {
 643                         struct ptldebug_header *hdr;
 644                         int len;
 645
 646                         hdr = (void *)p;
 647                         p += sizeof(*hdr);
 648                         file = p;
 649                         p += strlen(file) + 1;
 650                         fn = p;
 651                         p += strlen(fn) + 1;
 652                         len = hdr->ph_len - (int)(p - (char *)hdr);
 653
 654                         cfs_print_to_console(hdr, D_EMERG, p, len, file, fn);
 655
 656                         p += len;
 657                 }
 658
 659                 list_del(&tage->linkage);
 660                 cfs_tage_free(tage);
 661         }
 662 }
 663
 664 int cfs_tracefile_dump_all_pages(char *filename)
 665 {
 666         struct page_collection  pc;
 667         struct file             *filp;
 668         struct cfs_trace_page   *tage;
 669         struct cfs_trace_page   *tmp;
 670         char                    *buf;
 671         int rc;
 672
 673         DECL_MMSPACE;
 674
 675         cfs_tracefile_write_lock();
 676
 677         filp = filp_open(filename, O_CREAT|O_EXCL|O_WRONLY|O_LARGEFILE, 0600);
 678         if (IS_ERR(filp)) {
 679                 rc = PTR_ERR(filp);
 680                 filp = NULL;
 681                 pr_err("LustreError: can't open %s for dump: rc %d\n",
 682                         filename, rc);
 683                 goto out;
 684         }
 685
 686         pc.pc_want_daemon_pages = 1;
 687         collect_pages(&pc);
 688         if (list_empty(&pc.pc_pages)) {
 689                 rc = 0;
 690                 goto close;
 691         }
 692
 693         /* ok, for now, just write the pages.  in the future we'll be building
 694          * iobufs with the pages and calling generic_direct_IO */
 695         MMSPACE_OPEN;
 696         list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
 697
 698                 __LASSERT_TAGE_INVARIANT(tage);
 699
 700                 buf = kmap(tage->page);
 701                 rc = vfs_write(filp, (__force const char __user *)buf,
 702                                tage->used, &filp->f_pos);
 703                 kunmap(tage->page);
 704
 705                 if (rc != (int)tage->used) {
 706                         printk(KERN_WARNING "wanted to write %u but wrote %d\n",
 707                                tage->used, rc);
 708                         put_pages_back(&pc);
 709                         __LASSERT(list_empty(&pc.pc_pages));
 710                         break;
 711                 }
 712                 list_del(&tage->linkage);
 713                 cfs_tage_free(tage);
 714         }
 715         MMSPACE_CLOSE;
 716         rc = vfs_fsync(filp, 1);
 717         if (rc)
 718                 pr_err("sync returns %d\n", rc);
 719 close:
 720         filp_close(filp, NULL);
 721 out:
 722         cfs_tracefile_write_unlock();
 723         return rc;
 724 }
 725
 726 void cfs_trace_flush_pages(void)
 727 {
 728         struct page_collection pc;
 729         struct cfs_trace_page *tage;
 730         struct cfs_trace_page *tmp;
 731
 732         pc.pc_want_daemon_pages = 1;
 733         collect_pages(&pc);
 734         list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
 735
 736                 __LASSERT_TAGE_INVARIANT(tage);
 737
 738                 list_del(&tage->linkage);
 739                 cfs_tage_free(tage);
 740         }
 741 }
 742
 743 int cfs_trace_copyin_string(char *knl_buffer, int knl_buffer_nob,
 744                             const char __user *usr_buffer, int usr_buffer_nob)
 745 {
 746         int    nob;
 747
 748         if (usr_buffer_nob > knl_buffer_nob)
 749                 return -EOVERFLOW;
 750
 751         if (copy_from_user((void *)knl_buffer,
 752                            usr_buffer, usr_buffer_nob))
 753                 return -EFAULT;
 754
 755         nob = strnlen(knl_buffer, usr_buffer_nob);
 756         while (nob-- >= 0)                    /* strip trailing whitespace */
 757                 if (!isspace(knl_buffer[nob]))
 758                         break;
 759
 760         if (nob < 0)                        /* empty string */
 761                 return -EINVAL;
 762
 763         if (nob == knl_buffer_nob)            /* no space to terminate */
 764                 return -EOVERFLOW;
 765
 766         knl_buffer[nob + 1] = 0;                /* terminate */
 767         return 0;
 768 }
 769 EXPORT_SYMBOL(cfs_trace_copyin_string);
 770
 771 int cfs_trace_copyout_string(char __user *usr_buffer, int usr_buffer_nob,
 772                              const char *knl_buffer, char *append)
 773 {
 774         /* NB if 'append' != NULL, it's a single character to append to the
 775          * copied out string - usually "\n", for /proc entries and "" (i.e. a
 776          * terminating zero byte) for sysctl entries */
 777         int   nob = strlen(knl_buffer);
 778
 779         if (nob > usr_buffer_nob)
 780                 nob = usr_buffer_nob;
 781
 782         if (copy_to_user(usr_buffer, knl_buffer, nob))
 783                 return -EFAULT;
 784
 785         if (append != NULL && nob < usr_buffer_nob) {
 786                 if (copy_to_user(usr_buffer + nob, append, 1))
 787                         return -EFAULT;
 788
 789                 nob++;
 790         }
 791
 792         return nob;
 793 }
 794 EXPORT_SYMBOL(cfs_trace_copyout_string);
 795
 796 int cfs_trace_allocate_string_buffer(char **str, int nob)
 797 {
 798         if (nob > 2 * PAGE_CACHE_SIZE)      /* string must be "sensible" */
 799                 return -EINVAL;
 800
 801         *str = kmalloc(nob, GFP_KERNEL | __GFP_ZERO);
 802         if (*str == NULL)
 803                 return -ENOMEM;
 804
 805         return 0;
 806 }
 807
 808 int cfs_trace_dump_debug_buffer_usrstr(void __user *usr_str, int usr_str_nob)
 809 {
 810         char     *str;
 811         int        rc;
 812
 813         rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
 814         if (rc != 0)
 815                 return rc;
 816
 817         rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
 818                                      usr_str, usr_str_nob);
 819         if (rc != 0)
 820                 goto out;
 821
 822         if (str[0] != '/') {
 823                 rc = -EINVAL;
 824                 goto out;
 825         }
 826         rc = cfs_tracefile_dump_all_pages(str);
 827 out:
 828         kfree(str);
 829         return rc;
 830 }
 831
 832 int cfs_trace_daemon_command(char *str)
 833 {
 834         int       rc = 0;
 835
 836         cfs_tracefile_write_lock();
 837
 838         if (strcmp(str, "stop") == 0) {
 839                 cfs_tracefile_write_unlock();
 840                 cfs_trace_stop_thread();
 841                 cfs_tracefile_write_lock();
 842                 memset(cfs_tracefile, 0, sizeof(cfs_tracefile));
 843
 844         } else if (strncmp(str, "size=", 5) == 0) {
 845                 cfs_tracefile_size = simple_strtoul(str + 5, NULL, 0);
 846                 if (cfs_tracefile_size < 10 || cfs_tracefile_size > 20480)
 847                         cfs_tracefile_size = CFS_TRACEFILE_SIZE;
 848                 else
 849                         cfs_tracefile_size <<= 20;
 850
 851         } else if (strlen(str) >= sizeof(cfs_tracefile)) {
 852                 rc = -ENAMETOOLONG;
 853         } else if (str[0] != '/') {
 854                 rc = -EINVAL;
 855         } else {
 856                 strcpy(cfs_tracefile, str);
 857
 858                 printk(KERN_INFO
 859                        "Lustre: debug daemon will attempt to start writing to %s (%lukB max)\n",
 860                        cfs_tracefile,
 861                        (long)(cfs_tracefile_size >> 10));
 862
 863                 cfs_trace_start_thread();
 864         }
 865
 866         cfs_tracefile_write_unlock();
 867         return rc;
 868 }
 869
 870 int cfs_trace_daemon_command_usrstr(void __user *usr_str, int usr_str_nob)
 871 {
 872         char *str;
 873         int   rc;
 874
 875         rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
 876         if (rc != 0)
 877                 return rc;
 878
 879         rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
 880                                  usr_str, usr_str_nob);
 881         if (rc == 0)
 882                 rc = cfs_trace_daemon_command(str);
 883
 884         kfree(str);
 885         return rc;
 886 }
 887
 888 int cfs_trace_set_debug_mb(int mb)
 889 {
 890         int i;
 891         int j;
 892         int pages;
 893         int limit = cfs_trace_max_debug_mb();
 894         struct cfs_trace_cpu_data *tcd;
 895
 896         if (mb < num_possible_cpus()) {
 897                 printk(KERN_WARNING
 898                        "Lustre: %d MB is too small for debug buffer size, setting it to %d MB.\n",
 899                        mb, num_possible_cpus());
 900                 mb = num_possible_cpus();
 901         }
 902
 903         if (mb > limit) {
 904                 printk(KERN_WARNING
 905                        "Lustre: %d MB is too large for debug buffer size, setting it to %d MB.\n",
 906                        mb, limit);
 907                 mb = limit;
 908         }
 909
 910         mb /= num_possible_cpus();
 911         pages = mb << (20 - PAGE_CACHE_SHIFT);
 912
 913         cfs_tracefile_write_lock();
 914
 915         cfs_tcd_for_each(tcd, i, j)
 916                 tcd->tcd_max_pages = (pages * tcd->tcd_pages_factor) / 100;
 917
 918         cfs_tracefile_write_unlock();
 919
 920         return 0;
 921 }
 922
 923 int cfs_trace_get_debug_mb(void)
 924 {
 925         int i;
 926         int j;
 927         struct cfs_trace_cpu_data *tcd;
 928         int total_pages = 0;
 929
 930         cfs_tracefile_read_lock();
 931
 932         cfs_tcd_for_each(tcd, i, j)
 933                 total_pages += tcd->tcd_max_pages;
 934
 935         cfs_tracefile_read_unlock();
 936
 937         return (total_pages >> (20 - PAGE_CACHE_SHIFT)) + 1;
 938 }
 939
 940 static int tracefiled(void *arg)
 941 {
 942         struct page_collection pc;
 943         struct tracefiled_ctl *tctl = arg;
 944         struct cfs_trace_page *tage;
 945         struct cfs_trace_page *tmp;
 946         struct file *filp;
 947         char *buf;
 948         int last_loop = 0;
 949         int rc;
 950
 951         DECL_MMSPACE;
 952
 953         /* we're started late enough that we pick up init's fs context */
 954         /* this is so broken in uml?  what on earth is going on? */
 955
 956         complete(&tctl->tctl_start);
 957
 958         while (1) {
 959                 wait_queue_t __wait;
 960
 961                 pc.pc_want_daemon_pages = 0;
 962                 collect_pages(&pc);
 963                 if (list_empty(&pc.pc_pages))
 964                         goto end_loop;
 965
 966                 filp = NULL;
 967                 cfs_tracefile_read_lock();
 968                 if (cfs_tracefile[0] != 0) {
 969                         filp = filp_open(cfs_tracefile,
 970                                          O_CREAT | O_RDWR | O_LARGEFILE,
 971                                          0600);
 972                         if (IS_ERR(filp)) {
 973                                 rc = PTR_ERR(filp);
 974                                 filp = NULL;
 975                                 printk(KERN_WARNING "couldn't open %s: %d\n",
 976                                        cfs_tracefile, rc);
 977                         }
 978                 }
 979                 cfs_tracefile_read_unlock();
 980                 if (filp == NULL) {
 981                         put_pages_on_daemon_list(&pc);
 982                         __LASSERT(list_empty(&pc.pc_pages));
 983                         goto end_loop;
 984                 }
 985
 986                 MMSPACE_OPEN;
 987
 988                 list_for_each_entry_safe(tage, tmp, &pc.pc_pages,
 989                                                    linkage) {
 990                         static loff_t f_pos;
 991
 992                         __LASSERT_TAGE_INVARIANT(tage);
 993
 994                         if (f_pos >= (off_t)cfs_tracefile_size)
 995                                 f_pos = 0;
 996                         else if (f_pos > i_size_read(file_inode(filp)))
 997                                 f_pos = i_size_read(file_inode(filp));
 998
 999                         buf = kmap(tage->page);
1000                         rc = vfs_write(filp, (__force const char __user *)buf,
1001                                        tage->used, &f_pos);
1002                         kunmap(tage->page);
1003
1004                         if (rc != (int)tage->used) {
1005                                 printk(KERN_WARNING "wanted to write %u but wrote %d\n",
1006                                        tage->used, rc);
1007                                 put_pages_back(&pc);
1008                                 __LASSERT(list_empty(&pc.pc_pages));
1009                                 break;
1010                         }
1011                 }
1012                 MMSPACE_CLOSE;
1013
1014                 filp_close(filp, NULL);
1015                 put_pages_on_daemon_list(&pc);
1016                 if (!list_empty(&pc.pc_pages)) {
1017                         int i;
1018
1019                         printk(KERN_ALERT "Lustre: trace pages aren't empty\n");
1020                         pr_err("total cpus(%d): ",
1021                                 num_possible_cpus());
1022                         for (i = 0; i < num_possible_cpus(); i++)
1023                                 if (cpu_online(i))
1024                                         pr_cont("%d(on) ", i);
1025                                 else
1026                                         pr_cont("%d(off) ", i);
1027                         pr_cont("\n");
1028
1029                         i = 0;
1030                         list_for_each_entry_safe(tage, tmp, &pc.pc_pages,
1031                                                      linkage)
1032                                 pr_err("page %d belongs to cpu %d\n",
1033                                         ++i, tage->cpu);
1034                         pr_err("There are %d pages unwritten\n", i);
1035                 }
1036                 __LASSERT(list_empty(&pc.pc_pages));
1037 end_loop:
1038                 if (atomic_read(&tctl->tctl_shutdown)) {
1039                         if (last_loop == 0) {
1040                                 last_loop = 1;
1041                                 continue;
1042                         } else {
1043                                 break;
1044                         }
1045                 }
1046                 init_waitqueue_entry(&__wait, current);
1047                 add_wait_queue(&tctl->tctl_waitq, &__wait);
1048                 set_current_state(TASK_INTERRUPTIBLE);
1049                 schedule_timeout(cfs_time_seconds(1));
1050                 remove_wait_queue(&tctl->tctl_waitq, &__wait);
1051         }
1052         complete(&tctl->tctl_stop);
1053         return 0;
1054 }
1055
1056 int cfs_trace_start_thread(void)
1057 {
1058         struct tracefiled_ctl *tctl = &trace_tctl;
1059         struct task_struct *task;
1060         int rc = 0;
1061
1062         mutex_lock(&cfs_trace_thread_mutex);
1063         if (thread_running)
1064                 goto out;
1065
1066         init_completion(&tctl->tctl_start);
1067         init_completion(&tctl->tctl_stop);
1068         init_waitqueue_head(&tctl->tctl_waitq);
1069         atomic_set(&tctl->tctl_shutdown, 0);
1070
1071         task = kthread_run(tracefiled, tctl, "ktracefiled");
1072         if (IS_ERR(task)) {
1073                 rc = PTR_ERR(task);
1074                 goto out;
1075         }
1076
1077         wait_for_completion(&tctl->tctl_start);
1078         thread_running = 1;
1079 out:
1080         mutex_unlock(&cfs_trace_thread_mutex);
1081         return rc;
1082 }
1083
1084 void cfs_trace_stop_thread(void)
1085 {
1086         struct tracefiled_ctl *tctl = &trace_tctl;
1087
1088         mutex_lock(&cfs_trace_thread_mutex);
1089         if (thread_running) {
1090                 printk(KERN_INFO
1091                        "Lustre: shutting down debug daemon thread...\n");
1092                 atomic_set(&tctl->tctl_shutdown, 1);
1093                 wait_for_completion(&tctl->tctl_stop);
1094                 thread_running = 0;
1095         }
1096         mutex_unlock(&cfs_trace_thread_mutex);
1097 }
1098
1099 int cfs_tracefile_init(int max_pages)
1100 {
1101         struct cfs_trace_cpu_data *tcd;
1102         int                 i;
1103         int                 j;
1104         int                 rc;
1105         int                 factor;
1106
1107         rc = cfs_tracefile_init_arch();
1108         if (rc != 0)
1109                 return rc;
1110
1111         cfs_tcd_for_each(tcd, i, j) {
1112                 /* tcd_pages_factor is initialized int tracefile_init_arch. */
1113                 factor = tcd->tcd_pages_factor;
1114                 INIT_LIST_HEAD(&tcd->tcd_pages);
1115                 INIT_LIST_HEAD(&tcd->tcd_stock_pages);
1116                 INIT_LIST_HEAD(&tcd->tcd_daemon_pages);
1117                 tcd->tcd_cur_pages = 0;
1118                 tcd->tcd_cur_stock_pages = 0;
1119                 tcd->tcd_cur_daemon_pages = 0;
1120                 tcd->tcd_max_pages = (max_pages * factor) / 100;
1121                 LASSERT(tcd->tcd_max_pages > 0);
1122                 tcd->tcd_shutting_down = 0;
1123         }
1124
1125         return 0;
1126 }
1127
1128 static void trace_cleanup_on_all_cpus(void)
1129 {
1130         struct cfs_trace_cpu_data *tcd;
1131         struct cfs_trace_page *tage;
1132         struct cfs_trace_page *tmp;
1133         int i, cpu;
1134
1135         for_each_possible_cpu(cpu) {
1136                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
1137                         tcd->tcd_shutting_down = 1;
1138
1139                         list_for_each_entry_safe(tage, tmp, &tcd->tcd_pages,
1140                                                            linkage) {
1141                                 __LASSERT_TAGE_INVARIANT(tage);
1142
1143                                 list_del(&tage->linkage);
1144                                 cfs_tage_free(tage);
1145                         }
1146
1147                         tcd->tcd_cur_pages = 0;
1148                 }
1149         }
1150 }
1151
1152 static void cfs_trace_cleanup(void)
1153 {
1154         struct page_collection pc;
1155
1156         INIT_LIST_HEAD(&pc.pc_pages);
1157
1158         trace_cleanup_on_all_cpus();
1159
1160         cfs_tracefile_fini_arch();
1161 }
1162
1163 void cfs_tracefile_exit(void)
1164 {
1165         cfs_trace_stop_thread();
1166         cfs_trace_cleanup();
1167 }