libcfs/libcfs/tracefile.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * libcfs/libcfs/tracefile.c
  37  *
  38  * Author: Zach Brown <zab@clusterfs.com>
  39  * Author: Phil Schwan <phil@clusterfs.com>
  40  */
  41
  42
  43 #define DEBUG_SUBSYSTEM S_LNET
  44 #define LUSTRE_TRACEFILE_PRIVATE
  45 #include "tracefile.h"
  46
  47 #include <libcfs/libcfs.h>
  48
  49 /* XXX move things up to the top, comment */
  50 union cfs_trace_data_union (*cfs_trace_data[TCD_MAX_TYPES])[CFS_NR_CPUS] __cacheline_aligned;
  51
  52 char cfs_tracefile[TRACEFILE_NAME_SIZE];
  53 long long cfs_tracefile_size = CFS_TRACEFILE_SIZE;
  54 static struct tracefiled_ctl trace_tctl;
  55 cfs_mutex_t 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         cfs_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 libcfs_assertion_failed(const char *expr, struct libcfs_debug_msg_data *msgdata)
 484 {
 485         libcfs_debug_msg(msgdata, "ASSERTION(%s) failed\n", expr);
 486         /* cfs_enter_debugger(); */
 487         lbug_with_loc(msgdata);
 488 }
 489 EXPORT_SYMBOL(libcfs_assertion_failed);
 490
 491 void
 492 cfs_trace_assertion_failed(const char *str,
 493                            struct libcfs_debug_msg_data *msgdata)
 494 {
 495         struct ptldebug_header hdr;
 496
 497         libcfs_panic_in_progress = 1;
 498         libcfs_catastrophe = 1;
 499         cfs_mb();
 500
 501         cfs_set_ptldebug_header(&hdr, msgdata, CDEBUG_STACK());
 502
 503         cfs_print_to_console(&hdr, D_EMERG, str, strlen(str),
 504                              msgdata->msg_file, msgdata->msg_fn);
 505
 506         LIBCFS_PANIC("Lustre debug assertion failure\n");
 507
 508         /* not reached */
 509 }
 510
 511 static void
 512 panic_collect_pages(struct page_collection *pc)
 513 {
 514         /* Do the collect_pages job on a single CPU: assumes that all other
 515          * CPUs have been stopped during a panic.  If this isn't true for some
 516          * arch, this will have to be implemented separately in each arch.  */
 517         int                        i;
 518         int                        j;
 519         struct cfs_trace_cpu_data *tcd;
 520
 521         CFS_INIT_LIST_HEAD(&pc->pc_pages);
 522
 523         cfs_tcd_for_each(tcd, i, j) {
 524                 cfs_list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
 525                 tcd->tcd_cur_pages = 0;
 526
 527                 if (pc->pc_want_daemon_pages) {
 528                         cfs_list_splice_init(&tcd->tcd_daemon_pages,
 529                                              &pc->pc_pages);
 530                         tcd->tcd_cur_daemon_pages = 0;
 531                 }
 532         }
 533 }
 534
 535 static void collect_pages_on_all_cpus(struct page_collection *pc)
 536 {
 537         struct cfs_trace_cpu_data *tcd;
 538         int i, cpu;
 539
 540         cfs_spin_lock(&pc->pc_lock);
 541         cfs_for_each_possible_cpu(cpu) {
 542                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
 543                         cfs_list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
 544                         tcd->tcd_cur_pages = 0;
 545                         if (pc->pc_want_daemon_pages) {
 546                                 cfs_list_splice_init(&tcd->tcd_daemon_pages,
 547                                                      &pc->pc_pages);
 548                                 tcd->tcd_cur_daemon_pages = 0;
 549                         }
 550                 }
 551         }
 552         cfs_spin_unlock(&pc->pc_lock);
 553 }
 554
 555 static void collect_pages(struct page_collection *pc)
 556 {
 557         CFS_INIT_LIST_HEAD(&pc->pc_pages);
 558
 559         if (libcfs_panic_in_progress)
 560                 panic_collect_pages(pc);
 561         else
 562                 collect_pages_on_all_cpus(pc);
 563 }
 564
 565 static void put_pages_back_on_all_cpus(struct page_collection *pc)
 566 {
 567         struct cfs_trace_cpu_data *tcd;
 568         cfs_list_t *cur_head;
 569         struct cfs_trace_page *tage;
 570         struct cfs_trace_page *tmp;
 571         int i, cpu;
 572
 573         cfs_spin_lock(&pc->pc_lock);
 574         cfs_for_each_possible_cpu(cpu) {
 575                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
 576                         cur_head = tcd->tcd_pages.next;
 577
 578                         cfs_list_for_each_entry_safe_typed(tage, tmp,
 579                                                            &pc->pc_pages,
 580                                                            struct cfs_trace_page,
 581                                                            linkage) {
 582
 583                                 __LASSERT_TAGE_INVARIANT(tage);
 584
 585                                 if (tage->cpu != cpu || tage->type != i)
 586                                         continue;
 587
 588                                 cfs_tage_to_tail(tage, cur_head);
 589                                 tcd->tcd_cur_pages++;
 590                         }
 591                 }
 592         }
 593         cfs_spin_unlock(&pc->pc_lock);
 594 }
 595
 596 static void put_pages_back(struct page_collection *pc)
 597 {
 598         if (!libcfs_panic_in_progress)
 599                 put_pages_back_on_all_cpus(pc);
 600 }
 601
 602 /* Add pages to a per-cpu debug daemon ringbuffer.  This buffer makes sure that
 603  * we have a good amount of data at all times for dumping during an LBUG, even
 604  * if we have been steadily writing (and otherwise discarding) pages via the
 605  * debug daemon. */
 606 static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
 607                                          struct cfs_trace_cpu_data *tcd)
 608 {
 609         struct cfs_trace_page *tage;
 610         struct cfs_trace_page *tmp;
 611
 612         cfs_spin_lock(&pc->pc_lock);
 613         cfs_list_for_each_entry_safe_typed(tage, tmp, &pc->pc_pages,
 614                                            struct cfs_trace_page, linkage) {
 615
 616                 __LASSERT_TAGE_INVARIANT(tage);
 617
 618                 if (tage->cpu != tcd->tcd_cpu || tage->type != tcd->tcd_type)
 619                         continue;
 620
 621                 cfs_tage_to_tail(tage, &tcd->tcd_daemon_pages);
 622                 tcd->tcd_cur_daemon_pages++;
 623
 624                 if (tcd->tcd_cur_daemon_pages > tcd->tcd_max_pages) {
 625                         struct cfs_trace_page *victim;
 626
 627                         __LASSERT(!cfs_list_empty(&tcd->tcd_daemon_pages));
 628                         victim = cfs_tage_from_list(tcd->tcd_daemon_pages.next);
 629
 630                         __LASSERT_TAGE_INVARIANT(victim);
 631
 632                         cfs_list_del(&victim->linkage);
 633                         cfs_tage_free(victim);
 634                         tcd->tcd_cur_daemon_pages--;
 635                 }
 636         }
 637         cfs_spin_unlock(&pc->pc_lock);
 638 }
 639
 640 static void put_pages_on_daemon_list(struct page_collection *pc)
 641 {
 642         struct cfs_trace_cpu_data *tcd;
 643         int i, cpu;
 644
 645         cfs_for_each_possible_cpu(cpu) {
 646                 cfs_tcd_for_each_type_lock(tcd, i, cpu)
 647                         put_pages_on_tcd_daemon_list(pc, tcd);
 648         }
 649 }
 650
 651 void cfs_trace_debug_print(void)
 652 {
 653         struct page_collection pc;
 654         struct cfs_trace_page *tage;
 655         struct cfs_trace_page *tmp;
 656
 657         cfs_spin_lock_init(&pc.pc_lock);
 658
 659         pc.pc_want_daemon_pages = 1;
 660         collect_pages(&pc);
 661         cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
 662                                            struct cfs_trace_page, linkage) {
 663                 char *p, *file, *fn;
 664                 cfs_page_t *page;
 665
 666                 __LASSERT_TAGE_INVARIANT(tage);
 667
 668                 page = tage->page;
 669                 p = cfs_page_address(page);
 670                 while (p < ((char *)cfs_page_address(page) + tage->used)) {
 671                         struct ptldebug_header *hdr;
 672                         int len;
 673                         hdr = (void *)p;
 674                         p += sizeof(*hdr);
 675                         file = p;
 676                         p += strlen(file) + 1;
 677                         fn = p;
 678                         p += strlen(fn) + 1;
 679                         len = hdr->ph_len - (int)(p - (char *)hdr);
 680
 681                         cfs_print_to_console(hdr, D_EMERG, p, len, file, fn);
 682
 683                         p += len;
 684                 }
 685
 686                 cfs_list_del(&tage->linkage);
 687                 cfs_tage_free(tage);
 688         }
 689 }
 690
 691 int cfs_tracefile_dump_all_pages(char *filename)
 692 {
 693         struct page_collection pc;
 694         cfs_file_t *filp;
 695         struct cfs_trace_page *tage;
 696         struct cfs_trace_page *tmp;
 697         int rc;
 698
 699         CFS_DECL_MMSPACE;
 700
 701         cfs_tracefile_write_lock();
 702
 703         filp = cfs_filp_open(filename,
 704                              O_CREAT|O_EXCL|O_WRONLY|O_LARGEFILE, 0600, &rc);
 705         if (!filp) {
 706                 if (rc != -EEXIST)
 707                         printk(CFS_KERN_ERR
 708                                "LustreError: can't open %s for dump: rc %d\n",
 709                                filename, rc);
 710                 goto out;
 711         }
 712
 713         cfs_spin_lock_init(&pc.pc_lock);
 714         pc.pc_want_daemon_pages = 1;
 715         collect_pages(&pc);
 716         if (cfs_list_empty(&pc.pc_pages)) {
 717                 rc = 0;
 718                 goto close;
 719         }
 720
 721         /* ok, for now, just write the pages.  in the future we'll be building
 722          * iobufs with the pages and calling generic_direct_IO */
 723         CFS_MMSPACE_OPEN;
 724         cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
 725                                            struct cfs_trace_page, linkage) {
 726
 727                 __LASSERT_TAGE_INVARIANT(tage);
 728
 729                 rc = cfs_filp_write(filp, cfs_page_address(tage->page),
 730                                     tage->used, cfs_filp_poff(filp));
 731                 if (rc != (int)tage->used) {
 732                         printk(CFS_KERN_WARNING "wanted to write %u but wrote "
 733                                "%d\n", tage->used, rc);
 734                         put_pages_back(&pc);
 735                         __LASSERT(cfs_list_empty(&pc.pc_pages));
 736                         break;
 737                 }
 738                 cfs_list_del(&tage->linkage);
 739                 cfs_tage_free(tage);
 740         }
 741         CFS_MMSPACE_CLOSE;
 742         rc = cfs_filp_fsync(filp);
 743         if (rc)
 744                 printk(CFS_KERN_ERR "sync returns %d\n", rc);
 745  close:
 746         cfs_filp_close(filp);
 747  out:
 748         cfs_tracefile_write_unlock();
 749         return rc;
 750 }
 751
 752 void cfs_trace_flush_pages(void)
 753 {
 754         struct page_collection pc;
 755         struct cfs_trace_page *tage;
 756         struct cfs_trace_page *tmp;
 757
 758         cfs_spin_lock_init(&pc.pc_lock);
 759
 760         pc.pc_want_daemon_pages = 1;
 761         collect_pages(&pc);
 762         cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
 763                                            struct cfs_trace_page, linkage) {
 764
 765                 __LASSERT_TAGE_INVARIANT(tage);
 766
 767                 cfs_list_del(&tage->linkage);
 768                 cfs_tage_free(tage);
 769         }
 770 }
 771
 772 int cfs_trace_copyin_string(char *knl_buffer, int knl_buffer_nob,
 773                             const char *usr_buffer, int usr_buffer_nob)
 774 {
 775         int    nob;
 776
 777         if (usr_buffer_nob > knl_buffer_nob)
 778                 return -EOVERFLOW;
 779
 780         if (cfs_copy_from_user((void *)knl_buffer,
 781                            (void *)usr_buffer, usr_buffer_nob))
 782                 return -EFAULT;
 783
 784         nob = strnlen(knl_buffer, usr_buffer_nob);
 785         while (nob-- >= 0)                      /* strip trailing whitespace */
 786                 if (!isspace(knl_buffer[nob]))
 787                         break;
 788
 789         if (nob < 0)                            /* empty string */
 790                 return -EINVAL;
 791
 792         if (nob == knl_buffer_nob)              /* no space to terminate */
 793                 return -EOVERFLOW;
 794
 795         knl_buffer[nob + 1] = 0;                /* terminate */
 796         return 0;
 797 }
 798
 799 int cfs_trace_copyout_string(char *usr_buffer, int usr_buffer_nob,
 800                              const char *knl_buffer, char *append)
 801 {
 802         /* NB if 'append' != NULL, it's a single character to append to the
 803          * copied out string - usually "\n", for /proc entries and "" (i.e. a
 804          * terminating zero byte) for sysctl entries */
 805         int   nob = strlen(knl_buffer);
 806
 807         if (nob > usr_buffer_nob)
 808                 nob = usr_buffer_nob;
 809
 810         if (cfs_copy_to_user(usr_buffer, knl_buffer, nob))
 811                 return -EFAULT;
 812
 813         if (append != NULL && nob < usr_buffer_nob) {
 814                 if (cfs_copy_to_user(usr_buffer + nob, append, 1))
 815                         return -EFAULT;
 816
 817                 nob++;
 818         }
 819
 820         return nob;
 821 }
 822 EXPORT_SYMBOL(cfs_trace_copyout_string);
 823
 824 int cfs_trace_allocate_string_buffer(char **str, int nob)
 825 {
 826         if (nob > 2 * CFS_PAGE_SIZE)            /* string must be "sensible" */
 827                 return -EINVAL;
 828
 829         *str = cfs_alloc(nob, CFS_ALLOC_STD | CFS_ALLOC_ZERO);
 830         if (*str == NULL)
 831                 return -ENOMEM;
 832
 833         return 0;
 834 }
 835
 836 void cfs_trace_free_string_buffer(char *str, int nob)
 837 {
 838         cfs_free(str);
 839 }
 840
 841 int cfs_trace_dump_debug_buffer_usrstr(void *usr_str, int usr_str_nob)
 842 {
 843         char         *str;
 844         int           rc;
 845
 846         rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
 847         if (rc != 0)
 848                 return rc;
 849
 850         rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
 851                                      usr_str, usr_str_nob);
 852         if (rc != 0)
 853                 goto out;
 854
 855 #if !defined(__WINNT__)
 856         if (str[0] != '/') {
 857                 rc = -EINVAL;
 858                 goto out;
 859         }
 860 #endif
 861         rc = cfs_tracefile_dump_all_pages(str);
 862 out:
 863         cfs_trace_free_string_buffer(str, usr_str_nob + 1);
 864         return rc;
 865 }
 866
 867 int cfs_trace_daemon_command(char *str)
 868 {
 869         int       rc = 0;
 870
 871         cfs_tracefile_write_lock();
 872
 873         if (strcmp(str, "stop") == 0) {
 874                 cfs_tracefile_write_unlock();
 875                 cfs_trace_stop_thread();
 876                 cfs_tracefile_write_lock();
 877                 memset(cfs_tracefile, 0, sizeof(cfs_tracefile));
 878
 879         } else if (strncmp(str, "size=", 5) == 0) {
 880                 cfs_tracefile_size = simple_strtoul(str + 5, NULL, 0);
 881                 if (cfs_tracefile_size < 10 || cfs_tracefile_size > 20480)
 882                         cfs_tracefile_size = CFS_TRACEFILE_SIZE;
 883                 else
 884                         cfs_tracefile_size <<= 20;
 885
 886         } else if (strlen(str) >= sizeof(cfs_tracefile)) {
 887                 rc = -ENAMETOOLONG;
 888 #ifndef __WINNT__
 889         } else if (str[0] != '/') {
 890                 rc = -EINVAL;
 891 #endif
 892         } else {
 893                 strcpy(cfs_tracefile, str);
 894
 895                 printk(CFS_KERN_INFO
 896                        "Lustre: debug daemon will attempt to start writing "
 897                        "to %s (%lukB max)\n", cfs_tracefile,
 898                        (long)(cfs_tracefile_size >> 10));
 899
 900                 cfs_trace_start_thread();
 901         }
 902
 903         cfs_tracefile_write_unlock();
 904         return rc;
 905 }
 906
 907 int cfs_trace_daemon_command_usrstr(void *usr_str, int usr_str_nob)
 908 {
 909         char *str;
 910         int   rc;
 911
 912         rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
 913         if (rc != 0)
 914                 return rc;
 915
 916         rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
 917                                  usr_str, usr_str_nob);
 918         if (rc == 0)
 919                 rc = cfs_trace_daemon_command(str);
 920
 921         cfs_trace_free_string_buffer(str, usr_str_nob + 1);
 922         return rc;
 923 }
 924
 925 int cfs_trace_set_debug_mb(int mb)
 926 {
 927         int i;
 928         int j;
 929         int pages;
 930         int limit = cfs_trace_max_debug_mb();
 931         struct cfs_trace_cpu_data *tcd;
 932
 933         if (mb < cfs_num_possible_cpus()) {
 934                 printk(CFS_KERN_WARNING
 935                        "Lustre: %d MB is too small for debug buffer size, "
 936                        "setting it to %d MB.\n", mb, cfs_num_possible_cpus());
 937                 mb = cfs_num_possible_cpus();
 938         }
 939
 940         if (mb > limit) {
 941                 printk(CFS_KERN_WARNING
 942                        "Lustre: %d MB is too large for debug buffer size, "
 943                        "setting it to %d MB.\n", mb, limit);
 944                 mb = limit;
 945         }
 946
 947         mb /= cfs_num_possible_cpus();
 948         pages = mb << (20 - CFS_PAGE_SHIFT);
 949
 950         cfs_tracefile_write_lock();
 951
 952         cfs_tcd_for_each(tcd, i, j)
 953                 tcd->tcd_max_pages = (pages * tcd->tcd_pages_factor) / 100;
 954
 955         cfs_tracefile_write_unlock();
 956
 957         return 0;
 958 }
 959
 960 int cfs_trace_set_debug_mb_usrstr(void *usr_str, int usr_str_nob)
 961 {
 962         char     str[32];
 963         int      rc;
 964
 965         rc = cfs_trace_copyin_string(str, sizeof(str), usr_str, usr_str_nob);
 966         if (rc < 0)
 967                 return rc;
 968
 969         return cfs_trace_set_debug_mb(simple_strtoul(str, NULL, 0));
 970 }
 971
 972 int cfs_trace_get_debug_mb(void)
 973 {
 974         int i;
 975         int j;
 976         struct cfs_trace_cpu_data *tcd;
 977         int total_pages = 0;
 978
 979         cfs_tracefile_read_lock();
 980
 981         cfs_tcd_for_each(tcd, i, j)
 982                 total_pages += tcd->tcd_max_pages;
 983
 984         cfs_tracefile_read_unlock();
 985
 986         return (total_pages >> (20 - CFS_PAGE_SHIFT)) + 1;
 987 }
 988
 989 static int tracefiled(void *arg)
 990 {
 991         struct page_collection pc;
 992         struct tracefiled_ctl *tctl = arg;
 993         struct cfs_trace_page *tage;
 994         struct cfs_trace_page *tmp;
 995         cfs_file_t *filp;
 996         int last_loop = 0;
 997         int rc;
 998
 999         CFS_DECL_MMSPACE;
1000
1001         /* we're started late enough that we pick up init's fs context */
1002         /* this is so broken in uml?  what on earth is going on? */
1003         cfs_daemonize("ktracefiled");
1004
1005         cfs_spin_lock_init(&pc.pc_lock);
1006         cfs_complete(&tctl->tctl_start);
1007
1008         while (1) {
1009                 cfs_waitlink_t __wait;
1010
1011                 pc.pc_want_daemon_pages = 0;
1012                 collect_pages(&pc);
1013                 if (cfs_list_empty(&pc.pc_pages))
1014                         goto end_loop;
1015
1016                 filp = NULL;
1017                 cfs_tracefile_read_lock();
1018                 if (cfs_tracefile[0] != 0) {
1019                         filp = cfs_filp_open(cfs_tracefile,
1020                                              O_CREAT | O_RDWR | O_LARGEFILE,
1021                                              0600, &rc);
1022                         if (!(filp))
1023                                 printk(CFS_KERN_WARNING "couldn't open %s: "
1024                                        "%d\n", cfs_tracefile, rc);
1025                 }
1026                 cfs_tracefile_read_unlock();
1027                 if (filp == NULL) {
1028                         put_pages_on_daemon_list(&pc);
1029                         __LASSERT(cfs_list_empty(&pc.pc_pages));
1030                         goto end_loop;
1031                 }
1032
1033                 CFS_MMSPACE_OPEN;
1034
1035                 cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
1036                                                    struct cfs_trace_page,
1037                                                    linkage) {
1038                         static loff_t f_pos;
1039
1040                         __LASSERT_TAGE_INVARIANT(tage);
1041
1042                         if (f_pos >= (off_t)cfs_tracefile_size)
1043                                 f_pos = 0;
1044                         else if (f_pos > (off_t)cfs_filp_size(filp))
1045                                 f_pos = cfs_filp_size(filp);
1046
1047                         rc = cfs_filp_write(filp, cfs_page_address(tage->page),
1048                                             tage->used, &f_pos);
1049                         if (rc != (int)tage->used) {
1050                                 printk(CFS_KERN_WARNING "wanted to write %u "
1051                                        "but wrote %d\n", tage->used, rc);
1052                                 put_pages_back(&pc);
1053                                 __LASSERT(cfs_list_empty(&pc.pc_pages));
1054                         }
1055                 }
1056                 CFS_MMSPACE_CLOSE;
1057
1058                 cfs_filp_close(filp);
1059                 put_pages_on_daemon_list(&pc);
1060                 if (!cfs_list_empty(&pc.pc_pages)) {
1061                         int i;
1062
1063                         printk(CFS_KERN_ALERT "Lustre: trace pages aren't "
1064                                " empty\n");
1065                         printk(CFS_KERN_ERR "total cpus(%d): ",
1066                                cfs_num_possible_cpus());
1067                         for (i = 0; i < cfs_num_possible_cpus(); i++)
1068                                 if (cpu_online(i))
1069                                         printk(CFS_KERN_ERR "%d(on) ", i);
1070                                 else
1071                                         printk(CFS_KERN_ERR "%d(off) ", i);
1072                         printk(CFS_KERN_ERR "\n");
1073
1074                         i = 0;
1075                         cfs_list_for_each_entry_safe(tage, tmp, &pc.pc_pages,
1076                                                      linkage)
1077                                 printk(CFS_KERN_ERR "page %d belongs to cpu "
1078                                        "%d\n", ++i, tage->cpu);
1079                         printk(CFS_KERN_ERR "There are %d pages unwritten\n",
1080                                i);
1081                 }
1082                 __LASSERT(cfs_list_empty(&pc.pc_pages));
1083 end_loop:
1084                 if (cfs_atomic_read(&tctl->tctl_shutdown)) {
1085                         if (last_loop == 0) {
1086                                 last_loop = 1;
1087                                 continue;
1088                         } else {
1089                                 break;
1090                         }
1091                 }
1092                 cfs_waitlink_init(&__wait);
1093                 cfs_waitq_add(&tctl->tctl_waitq, &__wait);
1094                 cfs_set_current_state(CFS_TASK_INTERRUPTIBLE);
1095                 cfs_waitq_timedwait(&__wait, CFS_TASK_INTERRUPTIBLE,
1096                                     cfs_time_seconds(1));
1097                 cfs_waitq_del(&tctl->tctl_waitq, &__wait);
1098         }
1099         cfs_complete(&tctl->tctl_stop);
1100         return 0;
1101 }
1102
1103 int cfs_trace_start_thread(void)
1104 {
1105         struct tracefiled_ctl *tctl = &trace_tctl;
1106         int rc = 0;
1107
1108         cfs_mutex_lock(&cfs_trace_thread_mutex);
1109         if (thread_running)
1110                 goto out;
1111
1112         cfs_init_completion(&tctl->tctl_start);
1113         cfs_init_completion(&tctl->tctl_stop);
1114         cfs_waitq_init(&tctl->tctl_waitq);
1115         cfs_atomic_set(&tctl->tctl_shutdown, 0);
1116
1117         if (cfs_create_thread(tracefiled, tctl, 0) < 0) {
1118                 rc = -ECHILD;
1119                 goto out;
1120         }
1121
1122         cfs_wait_for_completion(&tctl->tctl_start);
1123         thread_running = 1;
1124 out:
1125         cfs_mutex_unlock(&cfs_trace_thread_mutex);
1126         return rc;
1127 }
1128
1129 void cfs_trace_stop_thread(void)
1130 {
1131         struct tracefiled_ctl *tctl = &trace_tctl;
1132
1133         cfs_mutex_lock(&cfs_trace_thread_mutex);
1134         if (thread_running) {
1135                 printk(CFS_KERN_INFO
1136                        "Lustre: shutting down debug daemon thread...\n");
1137                 cfs_atomic_set(&tctl->tctl_shutdown, 1);
1138                 cfs_wait_for_completion(&tctl->tctl_stop);
1139                 thread_running = 0;
1140         }
1141         cfs_mutex_unlock(&cfs_trace_thread_mutex);
1142 }
1143
1144 int cfs_tracefile_init(int max_pages)
1145 {
1146         struct cfs_trace_cpu_data *tcd;
1147         int                    i;
1148         int                    j;
1149         int                    rc;
1150         int                    factor;
1151
1152         rc = cfs_tracefile_init_arch();
1153         if (rc != 0)
1154                 return rc;
1155
1156         cfs_tcd_for_each(tcd, i, j) {
1157                 /* tcd_pages_factor is initialized int tracefile_init_arch. */
1158                 factor = tcd->tcd_pages_factor;
1159                 CFS_INIT_LIST_HEAD(&tcd->tcd_pages);
1160                 CFS_INIT_LIST_HEAD(&tcd->tcd_stock_pages);
1161                 CFS_INIT_LIST_HEAD(&tcd->tcd_daemon_pages);
1162                 tcd->tcd_cur_pages = 0;
1163                 tcd->tcd_cur_stock_pages = 0;
1164                 tcd->tcd_cur_daemon_pages = 0;
1165                 tcd->tcd_max_pages = (max_pages * factor) / 100;
1166                 LASSERT(tcd->tcd_max_pages > 0);
1167                 tcd->tcd_shutting_down = 0;
1168         }
1169
1170         return 0;
1171 }
1172
1173 static void trace_cleanup_on_all_cpus(void)
1174 {
1175         struct cfs_trace_cpu_data *tcd;
1176         struct cfs_trace_page *tage;
1177         struct cfs_trace_page *tmp;
1178         int i, cpu;
1179
1180         cfs_for_each_possible_cpu(cpu) {
1181                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
1182                         tcd->tcd_shutting_down = 1;
1183
1184                         cfs_list_for_each_entry_safe_typed(tage, tmp,
1185                                                            &tcd->tcd_pages,
1186                                                            struct cfs_trace_page,
1187                                                            linkage) {
1188                                 __LASSERT_TAGE_INVARIANT(tage);
1189
1190                                 cfs_list_del(&tage->linkage);
1191                                 cfs_tage_free(tage);
1192                         }
1193
1194                         tcd->tcd_cur_pages = 0;
1195                 }
1196         }
1197 }
1198
1199 static void cfs_trace_cleanup(void)
1200 {
1201         struct page_collection pc;
1202
1203         CFS_INIT_LIST_HEAD(&pc.pc_pages);
1204         cfs_spin_lock_init(&pc.pc_lock);
1205
1206         trace_cleanup_on_all_cpus();
1207
1208         cfs_tracefile_fini_arch();
1209 }
1210
1211 void cfs_tracefile_exit(void)
1212 {
1213         cfs_trace_stop_thread();
1214         cfs_trace_cleanup();
1215 }