libcfs/libcfs/tracefile.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  */
  30 /*
  31  * This file is part of Lustre, http://www.lustre.org/
  32  * Lustre is a trademark of Sun Microsystems, Inc.
  33  *
  34  * libcfs/libcfs/tracefile.c
  35  *
  36  * Author: Zach Brown <zab@clusterfs.com>
  37  * Author: Phil Schwan <phil@clusterfs.com>
  38  */
  39
  40
  41 #define DEBUG_SUBSYSTEM S_LNET
  42 #define LUSTRE_TRACEFILE_PRIVATE
  43 #include "tracefile.h"
  44
  45 #include <libcfs/libcfs.h>
  46
  47 /* XXX move things up to the top, comment */
  48 union cfs_trace_data_union (*cfs_trace_data[TCD_MAX_TYPES])[CFS_NR_CPUS] __cacheline_aligned;
  49
  50 char cfs_tracefile[TRACEFILE_NAME_SIZE];
  51 long long cfs_tracefile_size = CFS_TRACEFILE_SIZE;
  52 static struct tracefiled_ctl trace_tctl;
  53 struct mutex cfs_trace_thread_mutex;
  54 static int thread_running = 0;
  55
  56 cfs_atomic_t cfs_tage_allocated = CFS_ATOMIC_INIT(0);
  57
  58 static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
  59                                          struct cfs_trace_cpu_data *tcd);
  60
  61 static inline struct cfs_trace_page *
  62 cfs_tage_from_list(cfs_list_t *list)
  63 {
  64         return cfs_list_entry(list, struct cfs_trace_page, linkage);
  65 }
  66
  67 static struct cfs_trace_page *cfs_tage_alloc(int gfp)
  68 {
  69         cfs_page_t            *page;
  70         struct cfs_trace_page *tage;
  71
  72         /* My caller is trying to free memory */
  73         if (!cfs_in_interrupt() && cfs_memory_pressure_get())
  74                 return NULL;
  75
  76         /*
  77          * Don't spam console with allocation failures: they will be reported
  78          * by upper layer anyway.
  79          */
  80         gfp |= CFS_ALLOC_NOWARN;
  81         page = cfs_alloc_page(gfp);
  82         if (page == NULL)
  83                 return NULL;
  84
  85         tage = cfs_alloc(sizeof(*tage), gfp);
  86         if (tage == NULL) {
  87                 cfs_free_page(page);
  88                 return NULL;
  89         }
  90
  91         tage->page = page;
  92         cfs_atomic_inc(&cfs_tage_allocated);
  93         return tage;
  94 }
  95
  96 static void cfs_tage_free(struct cfs_trace_page *tage)
  97 {
  98         __LASSERT(tage != NULL);
  99         __LASSERT(tage->page != NULL);
 100
 101         cfs_free_page(tage->page);
 102         cfs_free(tage);
 103         cfs_atomic_dec(&cfs_tage_allocated);
 104 }
 105
 106 static void cfs_tage_to_tail(struct cfs_trace_page *tage,
 107                              cfs_list_t *queue)
 108 {
 109         __LASSERT(tage != NULL);
 110         __LASSERT(queue != NULL);
 111
 112         cfs_list_move_tail(&tage->linkage, queue);
 113 }
 114
 115 int cfs_trace_refill_stock(struct cfs_trace_cpu_data *tcd, int gfp,
 116                            cfs_list_t *stock)
 117 {
 118         int i;
 119
 120         /*
 121          * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
 122          * from here: this will lead to infinite recursion.
 123          */
 124
 125         for (i = 0; i + tcd->tcd_cur_stock_pages < TCD_STOCK_PAGES ; ++ i) {
 126                 struct cfs_trace_page *tage;
 127
 128                 tage = cfs_tage_alloc(gfp);
 129                 if (tage == NULL)
 130                         break;
 131                 cfs_list_add_tail(&tage->linkage, stock);
 132         }
 133         return i;
 134 }
 135
 136 /* return a page that has 'len' bytes left at the end */
 137 static struct cfs_trace_page *
 138 cfs_trace_get_tage_try(struct cfs_trace_cpu_data *tcd, unsigned long len)
 139 {
 140         struct cfs_trace_page *tage;
 141
 142         if (tcd->tcd_cur_pages > 0) {
 143                 __LASSERT(!cfs_list_empty(&tcd->tcd_pages));
 144                 tage = cfs_tage_from_list(tcd->tcd_pages.prev);
 145                 if (tage->used + len <= CFS_PAGE_SIZE)
 146                         return tage;
 147         }
 148
 149         if (tcd->tcd_cur_pages < tcd->tcd_max_pages) {
 150                 if (tcd->tcd_cur_stock_pages > 0) {
 151                         tage = cfs_tage_from_list(tcd->tcd_stock_pages.prev);
 152                         -- tcd->tcd_cur_stock_pages;
 153                         cfs_list_del_init(&tage->linkage);
 154                 } else {
 155                         tage = cfs_tage_alloc(CFS_ALLOC_ATOMIC);
 156                         if (tage == NULL) {
 157                                 if (printk_ratelimit())
 158                                         printk(CFS_KERN_WARNING
 159                                                "cannot allocate a tage (%ld)\n",
 160                                        tcd->tcd_cur_pages);
 161                                 return NULL;
 162                         }
 163                 }
 164
 165                 tage->used = 0;
 166                 tage->cpu = cfs_smp_processor_id();
 167                 tage->type = tcd->tcd_type;
 168                 cfs_list_add_tail(&tage->linkage, &tcd->tcd_pages);
 169                 tcd->tcd_cur_pages++;
 170
 171                 if (tcd->tcd_cur_pages > 8 && thread_running) {
 172                         struct tracefiled_ctl *tctl = &trace_tctl;
 173                         /*
 174                          * wake up tracefiled to process some pages.
 175                          */
 176                         cfs_waitq_signal(&tctl->tctl_waitq);
 177                 }
 178                 return tage;
 179         }
 180         return NULL;
 181 }
 182
 183 static void cfs_tcd_shrink(struct cfs_trace_cpu_data *tcd)
 184 {
 185         int pgcount = tcd->tcd_cur_pages / 10;
 186         struct page_collection pc;
 187         struct cfs_trace_page *tage;
 188         struct cfs_trace_page *tmp;
 189
 190         /*
 191          * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
 192          * from here: this will lead to infinite recursion.
 193          */
 194
 195         if (printk_ratelimit())
 196                 printk(CFS_KERN_WARNING "debug daemon buffer overflowed; "
 197                        "discarding 10%% of pages (%d of %ld)\n",
 198                        pgcount + 1, tcd->tcd_cur_pages);
 199
 200         CFS_INIT_LIST_HEAD(&pc.pc_pages);
 201         spin_lock_init(&pc.pc_lock);
 202
 203         cfs_list_for_each_entry_safe_typed(tage, tmp, &tcd->tcd_pages,
 204                                            struct cfs_trace_page, linkage) {
 205                 if (pgcount-- == 0)
 206                         break;
 207
 208                 cfs_list_move_tail(&tage->linkage, &pc.pc_pages);
 209                 tcd->tcd_cur_pages--;
 210         }
 211         put_pages_on_tcd_daemon_list(&pc, tcd);
 212 }
 213
 214 /* return a page that has 'len' bytes left at the end */
 215 static struct cfs_trace_page *cfs_trace_get_tage(struct cfs_trace_cpu_data *tcd,
 216                                                  unsigned long len)
 217 {
 218         struct cfs_trace_page *tage;
 219
 220         /*
 221          * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
 222          * from here: this will lead to infinite recursion.
 223          */
 224
 225         if (len > CFS_PAGE_SIZE) {
 226                 printk(CFS_KERN_ERR
 227                        "cowardly refusing to write %lu bytes in a page\n", len);
 228                 return NULL;
 229         }
 230
 231         tage = cfs_trace_get_tage_try(tcd, len);
 232         if (tage != NULL)
 233                 return tage;
 234         if (thread_running)
 235                 cfs_tcd_shrink(tcd);
 236         if (tcd->tcd_cur_pages > 0) {
 237                 tage = cfs_tage_from_list(tcd->tcd_pages.next);
 238                 tage->used = 0;
 239                 cfs_tage_to_tail(tage, &tcd->tcd_pages);
 240         }
 241         return tage;
 242 }
 243
 244 int libcfs_debug_msg(struct libcfs_debug_msg_data *msgdata,
 245                      const char *format, ...)
 246 {
 247         va_list args;
 248         int     rc;
 249
 250         va_start(args, format);
 251         rc = libcfs_debug_vmsg2(msgdata, format, args, NULL);
 252         va_end(args);
 253
 254         return rc;
 255 }
 256 EXPORT_SYMBOL(libcfs_debug_msg);
 257
 258 int libcfs_debug_vmsg2(struct libcfs_debug_msg_data *msgdata,
 259                        const char *format1, va_list args,
 260                        const char *format2, ...)
 261 {
 262         struct cfs_trace_cpu_data *tcd = NULL;
 263         struct ptldebug_header     header = {0};
 264         struct cfs_trace_page     *tage;
 265         /* string_buf is used only if tcd != NULL, and is always set then */
 266         char                      *string_buf = NULL;
 267         char                      *debug_buf;
 268         int                        known_size;
 269         int                        needed = 85; /* average message length */
 270         int                        max_nob;
 271         va_list                    ap;
 272         int                        depth;
 273         int                        i;
 274         int                        remain;
 275         int                        mask = msgdata->msg_mask;
 276         char                      *file = (char *)msgdata->msg_file;
 277         cfs_debug_limit_state_t   *cdls = msgdata->msg_cdls;
 278
 279         if (strchr(file, '/'))
 280                 file = strrchr(file, '/') + 1;
 281
 282         tcd = cfs_trace_get_tcd();
 283
 284         /* cfs_trace_get_tcd() grabs a lock, which disables preemption and
 285          * pins us to a particular CPU.  This avoids an smp_processor_id()
 286          * warning on Linux when debugging is enabled. */
 287         cfs_set_ptldebug_header(&header, msgdata, CDEBUG_STACK());
 288
 289         if (tcd == NULL)                /* arch may not log in IRQ context */
 290                 goto console;
 291
 292         if (tcd->tcd_cur_pages == 0)
 293                 header.ph_flags |= PH_FLAG_FIRST_RECORD;
 294
 295         if (tcd->tcd_shutting_down) {
 296                 cfs_trace_put_tcd(tcd);
 297                 tcd = NULL;
 298                 goto console;
 299         }
 300
 301         depth = __current_nesting_level();
 302         known_size = strlen(file) + 1 + depth;
 303         if (msgdata->msg_fn)
 304                 known_size += strlen(msgdata->msg_fn) + 1;
 305
 306         if (libcfs_debug_binary)
 307                 known_size += sizeof(header);
 308
 309         /*/
 310          * '2' used because vsnprintf return real size required for output
 311          * _without_ terminating NULL.
 312          * if needed is to small for this format.
 313          */
 314         for (i = 0; i < 2; i++) {
 315                 tage = cfs_trace_get_tage(tcd, needed + known_size + 1);
 316                 if (tage == NULL) {
 317                         if (needed + known_size > CFS_PAGE_SIZE)
 318                                 mask |= D_ERROR;
 319
 320                         cfs_trace_put_tcd(tcd);
 321                         tcd = NULL;
 322                         goto console;
 323                 }
 324
 325                 string_buf = (char *)cfs_page_address(tage->page) +
 326                                         tage->used + known_size;
 327
 328                 max_nob = CFS_PAGE_SIZE - tage->used - known_size;
 329                 if (max_nob <= 0) {
 330                         printk(CFS_KERN_EMERG "negative max_nob: %d\n",
 331                                max_nob);
 332                         mask |= D_ERROR;
 333                         cfs_trace_put_tcd(tcd);
 334                         tcd = NULL;
 335                         goto console;
 336                 }
 337
 338                 needed = 0;
 339                 if (format1) {
 340                         va_copy(ap, args);
 341                         needed = vsnprintf(string_buf, max_nob, format1, ap);
 342                         va_end(ap);
 343                 }
 344
 345                 if (format2) {
 346                         remain = max_nob - needed;
 347                         if (remain < 0)
 348                                 remain = 0;
 349
 350                         va_start(ap, format2);
 351                         needed += vsnprintf(string_buf + needed, remain,
 352                                             format2, ap);
 353                         va_end(ap);
 354                 }
 355
 356                 if (needed < max_nob) /* well. printing ok.. */
 357                         break;
 358         }
 359
 360         if (*(string_buf+needed-1) != '\n')
 361                 printk(CFS_KERN_INFO "format at %s:%d:%s doesn't end in "
 362                        "newline\n", file, msgdata->msg_line, msgdata->msg_fn);
 363
 364         header.ph_len = known_size + needed;
 365         debug_buf = (char *)cfs_page_address(tage->page) + tage->used;
 366
 367         if (libcfs_debug_binary) {
 368                 memcpy(debug_buf, &header, sizeof(header));
 369                 tage->used += sizeof(header);
 370                 debug_buf += sizeof(header);
 371         }
 372
 373         /* indent message according to the nesting level */
 374         while (depth-- > 0) {
 375                 *(debug_buf++) = '.';
 376                 ++ tage->used;
 377         }
 378
 379         strcpy(debug_buf, file);
 380         tage->used += strlen(file) + 1;
 381         debug_buf += strlen(file) + 1;
 382
 383         if (msgdata->msg_fn) {
 384                 strcpy(debug_buf, msgdata->msg_fn);
 385                 tage->used += strlen(msgdata->msg_fn) + 1;
 386                 debug_buf += strlen(msgdata->msg_fn) + 1;
 387         }
 388
 389         __LASSERT(debug_buf == string_buf);
 390
 391         tage->used += needed;
 392         __LASSERT (tage->used <= CFS_PAGE_SIZE);
 393
 394 console:
 395         if ((mask & libcfs_printk) == 0) {
 396                 /* no console output requested */
 397                 if (tcd != NULL)
 398                         cfs_trace_put_tcd(tcd);
 399                 return 1;
 400         }
 401
 402         if (cdls != NULL) {
 403                 if (libcfs_console_ratelimit &&
 404                     cdls->cdls_next != 0 &&     /* not first time ever */
 405                     !cfs_time_after(cfs_time_current(), cdls->cdls_next)) {
 406                         /* skipping a console message */
 407                         cdls->cdls_count++;
 408                         if (tcd != NULL)
 409                                 cfs_trace_put_tcd(tcd);
 410                         return 1;
 411                 }
 412
 413                 if (cfs_time_after(cfs_time_current(), cdls->cdls_next +
 414                                                        libcfs_console_max_delay
 415                                                        + cfs_time_seconds(10))) {
 416                         /* last timeout was a long time ago */
 417                         cdls->cdls_delay /= libcfs_console_backoff * 4;
 418                 } else {
 419                         cdls->cdls_delay *= libcfs_console_backoff;
 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
 427                 /* ensure cdls_next is never zero after it's been seen */
 428                 cdls->cdls_next = (cfs_time_current() + cdls->cdls_delay) | 1;
 429         }
 430
 431         if (tcd != NULL) {
 432                 cfs_print_to_console(&header, mask, string_buf, needed, file,
 433                                      msgdata->msg_fn);
 434                 cfs_trace_put_tcd(tcd);
 435         } else {
 436                 string_buf = cfs_trace_get_console_buffer();
 437
 438                 needed = 0;
 439                 if (format1 != NULL) {
 440                         va_copy(ap, args);
 441                         needed = vsnprintf(string_buf,
 442                                            CFS_TRACE_CONSOLE_BUFFER_SIZE,
 443                                            format1, ap);
 444                         va_end(ap);
 445                 }
 446                 if (format2 != NULL) {
 447                         remain = CFS_TRACE_CONSOLE_BUFFER_SIZE - needed;
 448                         if (remain > 0) {
 449                                 va_start(ap, format2);
 450                                 needed += vsnprintf(string_buf+needed, remain,
 451                                                     format2, ap);
 452                                 va_end(ap);
 453                         }
 454                 }
 455                 cfs_print_to_console(&header, mask,
 456                                      string_buf, needed, file, msgdata->msg_fn);
 457
 458                 cfs_trace_put_console_buffer(string_buf);
 459         }
 460
 461         if (cdls != NULL && cdls->cdls_count != 0) {
 462                 string_buf = cfs_trace_get_console_buffer();
 463
 464                 needed = snprintf(string_buf, CFS_TRACE_CONSOLE_BUFFER_SIZE,
 465                                   "Skipped %d previous similar message%s\n",
 466                                   cdls->cdls_count,
 467                                   (cdls->cdls_count > 1) ? "s" : "");
 468
 469                 cfs_print_to_console(&header, mask,
 470                                      string_buf, needed, file, msgdata->msg_fn);
 471
 472                 cfs_trace_put_console_buffer(string_buf);
 473                 cdls->cdls_count = 0;
 474         }
 475
 476         return 0;
 477 }
 478 EXPORT_SYMBOL(libcfs_debug_vmsg2);
 479
 480 void
 481 cfs_trace_assertion_failed(const char *str,
 482                            struct libcfs_debug_msg_data *msgdata)
 483 {
 484         struct ptldebug_header hdr;
 485
 486         libcfs_panic_in_progress = 1;
 487         libcfs_catastrophe = 1;
 488         cfs_mb();
 489
 490         cfs_set_ptldebug_header(&hdr, msgdata, CDEBUG_STACK());
 491
 492         cfs_print_to_console(&hdr, D_EMERG, str, strlen(str),
 493                              msgdata->msg_file, msgdata->msg_fn);
 494
 495         LIBCFS_PANIC("Lustre debug assertion failure\n");
 496
 497         /* not reached */
 498 }
 499
 500 static void
 501 panic_collect_pages(struct page_collection *pc)
 502 {
 503         /* Do the collect_pages job on a single CPU: assumes that all other
 504          * CPUs have been stopped during a panic.  If this isn't true for some
 505          * arch, this will have to be implemented separately in each arch.  */
 506         int                        i;
 507         int                        j;
 508         struct cfs_trace_cpu_data *tcd;
 509
 510         CFS_INIT_LIST_HEAD(&pc->pc_pages);
 511
 512         cfs_tcd_for_each(tcd, i, j) {
 513                 cfs_list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
 514                 tcd->tcd_cur_pages = 0;
 515
 516                 if (pc->pc_want_daemon_pages) {
 517                         cfs_list_splice_init(&tcd->tcd_daemon_pages,
 518                                              &pc->pc_pages);
 519                         tcd->tcd_cur_daemon_pages = 0;
 520                 }
 521         }
 522 }
 523
 524 static void collect_pages_on_all_cpus(struct page_collection *pc)
 525 {
 526         struct cfs_trace_cpu_data *tcd;
 527         int i, cpu;
 528
 529         spin_lock(&pc->pc_lock);
 530         cfs_for_each_possible_cpu(cpu) {
 531                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
 532                         cfs_list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
 533                         tcd->tcd_cur_pages = 0;
 534                         if (pc->pc_want_daemon_pages) {
 535                                 cfs_list_splice_init(&tcd->tcd_daemon_pages,
 536                                                      &pc->pc_pages);
 537                                 tcd->tcd_cur_daemon_pages = 0;
 538                         }
 539                 }
 540         }
 541         spin_unlock(&pc->pc_lock);
 542 }
 543
 544 static void collect_pages(struct page_collection *pc)
 545 {
 546         CFS_INIT_LIST_HEAD(&pc->pc_pages);
 547
 548         if (libcfs_panic_in_progress)
 549                 panic_collect_pages(pc);
 550         else
 551                 collect_pages_on_all_cpus(pc);
 552 }
 553
 554 static void put_pages_back_on_all_cpus(struct page_collection *pc)
 555 {
 556         struct cfs_trace_cpu_data *tcd;
 557         cfs_list_t *cur_head;
 558         struct cfs_trace_page *tage;
 559         struct cfs_trace_page *tmp;
 560         int i, cpu;
 561
 562         spin_lock(&pc->pc_lock);
 563         cfs_for_each_possible_cpu(cpu) {
 564                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
 565                         cur_head = tcd->tcd_pages.next;
 566
 567                         cfs_list_for_each_entry_safe_typed(tage, tmp,
 568                                                            &pc->pc_pages,
 569                                                            struct cfs_trace_page,
 570                                                            linkage) {
 571
 572                                 __LASSERT_TAGE_INVARIANT(tage);
 573
 574                                 if (tage->cpu != cpu || tage->type != i)
 575                                         continue;
 576
 577                                 cfs_tage_to_tail(tage, cur_head);
 578                                 tcd->tcd_cur_pages++;
 579                         }
 580                 }
 581         }
 582         spin_unlock(&pc->pc_lock);
 583 }
 584
 585 static void put_pages_back(struct page_collection *pc)
 586 {
 587         if (!libcfs_panic_in_progress)
 588                 put_pages_back_on_all_cpus(pc);
 589 }
 590
 591 /* Add pages to a per-cpu debug daemon ringbuffer.  This buffer makes sure that
 592  * we have a good amount of data at all times for dumping during an LBUG, even
 593  * if we have been steadily writing (and otherwise discarding) pages via the
 594  * debug daemon. */
 595 static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
 596                                          struct cfs_trace_cpu_data *tcd)
 597 {
 598         struct cfs_trace_page *tage;
 599         struct cfs_trace_page *tmp;
 600
 601         spin_lock(&pc->pc_lock);
 602         cfs_list_for_each_entry_safe_typed(tage, tmp, &pc->pc_pages,
 603                                            struct cfs_trace_page, linkage) {
 604
 605                 __LASSERT_TAGE_INVARIANT(tage);
 606
 607                 if (tage->cpu != tcd->tcd_cpu || tage->type != tcd->tcd_type)
 608                         continue;
 609
 610                 cfs_tage_to_tail(tage, &tcd->tcd_daemon_pages);
 611                 tcd->tcd_cur_daemon_pages++;
 612
 613                 if (tcd->tcd_cur_daemon_pages > tcd->tcd_max_pages) {
 614                         struct cfs_trace_page *victim;
 615
 616                         __LASSERT(!cfs_list_empty(&tcd->tcd_daemon_pages));
 617                         victim = cfs_tage_from_list(tcd->tcd_daemon_pages.next);
 618
 619                         __LASSERT_TAGE_INVARIANT(victim);
 620
 621                         cfs_list_del(&victim->linkage);
 622                         cfs_tage_free(victim);
 623                         tcd->tcd_cur_daemon_pages--;
 624                 }
 625         }
 626         spin_unlock(&pc->pc_lock);
 627 }
 628
 629 static void put_pages_on_daemon_list(struct page_collection *pc)
 630 {
 631         struct cfs_trace_cpu_data *tcd;
 632         int i, cpu;
 633
 634         cfs_for_each_possible_cpu(cpu) {
 635                 cfs_tcd_for_each_type_lock(tcd, i, cpu)
 636                         put_pages_on_tcd_daemon_list(pc, tcd);
 637         }
 638 }
 639
 640 void cfs_trace_debug_print(void)
 641 {
 642         struct page_collection pc;
 643         struct cfs_trace_page *tage;
 644         struct cfs_trace_page *tmp;
 645
 646         spin_lock_init(&pc.pc_lock);
 647
 648         pc.pc_want_daemon_pages = 1;
 649         collect_pages(&pc);
 650         cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
 651                                            struct cfs_trace_page, linkage) {
 652                 char *p, *file, *fn;
 653                 cfs_page_t *page;
 654
 655                 __LASSERT_TAGE_INVARIANT(tage);
 656
 657                 page = tage->page;
 658                 p = cfs_page_address(page);
 659                 while (p < ((char *)cfs_page_address(page) + tage->used)) {
 660                         struct ptldebug_header *hdr;
 661                         int len;
 662                         hdr = (void *)p;
 663                         p += sizeof(*hdr);
 664                         file = p;
 665                         p += strlen(file) + 1;
 666                         fn = p;
 667                         p += strlen(fn) + 1;
 668                         len = hdr->ph_len - (int)(p - (char *)hdr);
 669
 670                         cfs_print_to_console(hdr, D_EMERG, p, len, file, fn);
 671
 672                         p += len;
 673                 }
 674
 675                 cfs_list_del(&tage->linkage);
 676                 cfs_tage_free(tage);
 677         }
 678 }
 679
 680 int cfs_tracefile_dump_all_pages(char *filename)
 681 {
 682         struct page_collection pc;
 683         cfs_file_t *filp;
 684         struct cfs_trace_page *tage;
 685         struct cfs_trace_page *tmp;
 686         int rc;
 687
 688         CFS_DECL_MMSPACE;
 689
 690         cfs_tracefile_write_lock();
 691
 692         filp = cfs_filp_open(filename,
 693                              O_CREAT|O_EXCL|O_WRONLY|O_LARGEFILE, 0600, &rc);
 694         if (!filp) {
 695                 if (rc != -EEXIST)
 696                         printk(CFS_KERN_ERR
 697                                "LustreError: can't open %s for dump: rc %d\n",
 698                                filename, rc);
 699                 goto out;
 700         }
 701
 702         spin_lock_init(&pc.pc_lock);
 703         pc.pc_want_daemon_pages = 1;
 704         collect_pages(&pc);
 705         if (cfs_list_empty(&pc.pc_pages)) {
 706                 rc = 0;
 707                 goto close;
 708         }
 709
 710         /* ok, for now, just write the pages.  in the future we'll be building
 711          * iobufs with the pages and calling generic_direct_IO */
 712         CFS_MMSPACE_OPEN;
 713         cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
 714                                            struct cfs_trace_page, linkage) {
 715
 716                 __LASSERT_TAGE_INVARIANT(tage);
 717
 718                 rc = cfs_filp_write(filp, cfs_page_address(tage->page),
 719                                     tage->used, cfs_filp_poff(filp));
 720                 if (rc != (int)tage->used) {
 721                         printk(CFS_KERN_WARNING "wanted to write %u but wrote "
 722                                "%d\n", tage->used, rc);
 723                         put_pages_back(&pc);
 724                         __LASSERT(cfs_list_empty(&pc.pc_pages));
 725                         break;
 726                 }
 727                 cfs_list_del(&tage->linkage);
 728                 cfs_tage_free(tage);
 729         }
 730         CFS_MMSPACE_CLOSE;
 731         rc = cfs_filp_fsync(filp);
 732         if (rc)
 733                 printk(CFS_KERN_ERR "sync returns %d\n", rc);
 734  close:
 735         cfs_filp_close(filp);
 736  out:
 737         cfs_tracefile_write_unlock();
 738         return rc;
 739 }
 740
 741 void cfs_trace_flush_pages(void)
 742 {
 743         struct page_collection pc;
 744         struct cfs_trace_page *tage;
 745         struct cfs_trace_page *tmp;
 746
 747         spin_lock_init(&pc.pc_lock);
 748
 749         pc.pc_want_daemon_pages = 1;
 750         collect_pages(&pc);
 751         cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
 752                                            struct cfs_trace_page, linkage) {
 753
 754                 __LASSERT_TAGE_INVARIANT(tage);
 755
 756                 cfs_list_del(&tage->linkage);
 757                 cfs_tage_free(tage);
 758         }
 759 }
 760
 761 int cfs_trace_copyin_string(char *knl_buffer, int knl_buffer_nob,
 762                             const char *usr_buffer, int usr_buffer_nob)
 763 {
 764         int    nob;
 765
 766         if (usr_buffer_nob > knl_buffer_nob)
 767                 return -EOVERFLOW;
 768
 769         if (cfs_copy_from_user((void *)knl_buffer,
 770                            (void *)usr_buffer, usr_buffer_nob))
 771                 return -EFAULT;
 772
 773         nob = strnlen(knl_buffer, usr_buffer_nob);
 774         while (nob-- >= 0)                      /* strip trailing whitespace */
 775                 if (!isspace(knl_buffer[nob]))
 776                         break;
 777
 778         if (nob < 0)                            /* empty string */
 779                 return -EINVAL;
 780
 781         if (nob == knl_buffer_nob)              /* no space to terminate */
 782                 return -EOVERFLOW;
 783
 784         knl_buffer[nob + 1] = 0;                /* terminate */
 785         return 0;
 786 }
 787 EXPORT_SYMBOL(cfs_trace_copyin_string);
 788
 789 int cfs_trace_copyout_string(char *usr_buffer, int usr_buffer_nob,
 790                              const char *knl_buffer, char *append)
 791 {
 792         /* NB if 'append' != NULL, it's a single character to append to the
 793          * copied out string - usually "\n", for /proc entries and "" (i.e. a
 794          * terminating zero byte) for sysctl entries */
 795         int   nob = strlen(knl_buffer);
 796
 797         if (nob > usr_buffer_nob)
 798                 nob = usr_buffer_nob;
 799
 800         if (cfs_copy_to_user(usr_buffer, knl_buffer, nob))
 801                 return -EFAULT;
 802
 803         if (append != NULL && nob < usr_buffer_nob) {
 804                 if (cfs_copy_to_user(usr_buffer + nob, append, 1))
 805                         return -EFAULT;
 806
 807                 nob++;
 808         }
 809
 810         return nob;
 811 }
 812 EXPORT_SYMBOL(cfs_trace_copyout_string);
 813
 814 int cfs_trace_allocate_string_buffer(char **str, int nob)
 815 {
 816         if (nob > 2 * CFS_PAGE_SIZE)            /* string must be "sensible" */
 817                 return -EINVAL;
 818
 819         *str = cfs_alloc(nob, CFS_ALLOC_STD | CFS_ALLOC_ZERO);
 820         if (*str == NULL)
 821                 return -ENOMEM;
 822
 823         return 0;
 824 }
 825
 826 void cfs_trace_free_string_buffer(char *str, int nob)
 827 {
 828         cfs_free(str);
 829 }
 830
 831 int cfs_trace_dump_debug_buffer_usrstr(void *usr_str, int usr_str_nob)
 832 {
 833         char         *str;
 834         int           rc;
 835
 836         rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
 837         if (rc != 0)
 838                 return rc;
 839
 840         rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
 841                                      usr_str, usr_str_nob);
 842         if (rc != 0)
 843                 goto out;
 844
 845 #if !defined(__WINNT__)
 846         if (str[0] != '/') {
 847                 rc = -EINVAL;
 848                 goto out;
 849         }
 850 #endif
 851         rc = cfs_tracefile_dump_all_pages(str);
 852 out:
 853         cfs_trace_free_string_buffer(str, usr_str_nob + 1);
 854         return rc;
 855 }
 856
 857 int cfs_trace_daemon_command(char *str)
 858 {
 859         int       rc = 0;
 860
 861         cfs_tracefile_write_lock();
 862
 863         if (strcmp(str, "stop") == 0) {
 864                 cfs_tracefile_write_unlock();
 865                 cfs_trace_stop_thread();
 866                 cfs_tracefile_write_lock();
 867                 memset(cfs_tracefile, 0, sizeof(cfs_tracefile));
 868
 869         } else if (strncmp(str, "size=", 5) == 0) {
 870                 cfs_tracefile_size = simple_strtoul(str + 5, NULL, 0);
 871                 if (cfs_tracefile_size < 10 || cfs_tracefile_size > 20480)
 872                         cfs_tracefile_size = CFS_TRACEFILE_SIZE;
 873                 else
 874                         cfs_tracefile_size <<= 20;
 875
 876         } else if (strlen(str) >= sizeof(cfs_tracefile)) {
 877                 rc = -ENAMETOOLONG;
 878 #ifndef __WINNT__
 879         } else if (str[0] != '/') {
 880                 rc = -EINVAL;
 881 #endif
 882         } else {
 883                 strcpy(cfs_tracefile, str);
 884
 885                 printk(CFS_KERN_INFO
 886                        "Lustre: debug daemon will attempt to start writing "
 887                        "to %s (%lukB max)\n", cfs_tracefile,
 888                        (long)(cfs_tracefile_size >> 10));
 889
 890                 cfs_trace_start_thread();
 891         }
 892
 893         cfs_tracefile_write_unlock();
 894         return rc;
 895 }
 896
 897 int cfs_trace_daemon_command_usrstr(void *usr_str, int usr_str_nob)
 898 {
 899         char *str;
 900         int   rc;
 901
 902         rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
 903         if (rc != 0)
 904                 return rc;
 905
 906         rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
 907                                  usr_str, usr_str_nob);
 908         if (rc == 0)
 909                 rc = cfs_trace_daemon_command(str);
 910
 911         cfs_trace_free_string_buffer(str, usr_str_nob + 1);
 912         return rc;
 913 }
 914
 915 int cfs_trace_set_debug_mb(int mb)
 916 {
 917         int i;
 918         int j;
 919         int pages;
 920         int limit = cfs_trace_max_debug_mb();
 921         struct cfs_trace_cpu_data *tcd;
 922
 923         if (mb < cfs_num_possible_cpus()) {
 924                 printk(CFS_KERN_WARNING
 925                        "Lustre: %d MB is too small for debug buffer size, "
 926                        "setting it to %d MB.\n", mb, cfs_num_possible_cpus());
 927                 mb = cfs_num_possible_cpus();
 928         }
 929
 930         if (mb > limit) {
 931                 printk(CFS_KERN_WARNING
 932                        "Lustre: %d MB is too large for debug buffer size, "
 933                        "setting it to %d MB.\n", mb, limit);
 934                 mb = limit;
 935         }
 936
 937         mb /= cfs_num_possible_cpus();
 938         pages = mb << (20 - CFS_PAGE_SHIFT);
 939
 940         cfs_tracefile_write_lock();
 941
 942         cfs_tcd_for_each(tcd, i, j)
 943                 tcd->tcd_max_pages = (pages * tcd->tcd_pages_factor) / 100;
 944
 945         cfs_tracefile_write_unlock();
 946
 947         return 0;
 948 }
 949
 950 int cfs_trace_set_debug_mb_usrstr(void *usr_str, int usr_str_nob)
 951 {
 952         char     str[32];
 953         int      rc;
 954
 955         rc = cfs_trace_copyin_string(str, sizeof(str), usr_str, usr_str_nob);
 956         if (rc < 0)
 957                 return rc;
 958
 959         return cfs_trace_set_debug_mb(simple_strtoul(str, NULL, 0));
 960 }
 961
 962 int cfs_trace_get_debug_mb(void)
 963 {
 964         int i;
 965         int j;
 966         struct cfs_trace_cpu_data *tcd;
 967         int total_pages = 0;
 968
 969         cfs_tracefile_read_lock();
 970
 971         cfs_tcd_for_each(tcd, i, j)
 972                 total_pages += tcd->tcd_max_pages;
 973
 974         cfs_tracefile_read_unlock();
 975
 976         return (total_pages >> (20 - CFS_PAGE_SHIFT)) + 1;
 977 }
 978
 979 static int tracefiled(void *arg)
 980 {
 981         struct page_collection pc;
 982         struct tracefiled_ctl *tctl = arg;
 983         struct cfs_trace_page *tage;
 984         struct cfs_trace_page *tmp;
 985         cfs_file_t *filp;
 986         int last_loop = 0;
 987         int rc;
 988
 989         CFS_DECL_MMSPACE;
 990
 991         /* we're started late enough that we pick up init's fs context */
 992         /* this is so broken in uml?  what on earth is going on? */
 993         cfs_daemonize("ktracefiled");
 994
 995         spin_lock_init(&pc.pc_lock);
 996         complete(&tctl->tctl_start);
 997
 998         while (1) {
 999                 cfs_waitlink_t __wait;
1000
1001                 pc.pc_want_daemon_pages = 0;
1002                 collect_pages(&pc);
1003                 if (cfs_list_empty(&pc.pc_pages))
1004                         goto end_loop;
1005
1006                 filp = NULL;
1007                 cfs_tracefile_read_lock();
1008                 if (cfs_tracefile[0] != 0) {
1009                         filp = cfs_filp_open(cfs_tracefile,
1010                                              O_CREAT | O_RDWR | O_LARGEFILE,
1011                                              0600, &rc);
1012                         if (!(filp))
1013                                 printk(CFS_KERN_WARNING "couldn't open %s: "
1014                                        "%d\n", cfs_tracefile, rc);
1015                 }
1016                 cfs_tracefile_read_unlock();
1017                 if (filp == NULL) {
1018                         put_pages_on_daemon_list(&pc);
1019                         __LASSERT(cfs_list_empty(&pc.pc_pages));
1020                         goto end_loop;
1021                 }
1022
1023                 CFS_MMSPACE_OPEN;
1024
1025                 cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
1026                                                    struct cfs_trace_page,
1027                                                    linkage) {
1028                         static loff_t f_pos;
1029
1030                         __LASSERT_TAGE_INVARIANT(tage);
1031
1032                         if (f_pos >= (off_t)cfs_tracefile_size)
1033                                 f_pos = 0;
1034                         else if (f_pos > (off_t)cfs_filp_size(filp))
1035                                 f_pos = cfs_filp_size(filp);
1036
1037                         rc = cfs_filp_write(filp, cfs_page_address(tage->page),
1038                                             tage->used, &f_pos);
1039                         if (rc != (int)tage->used) {
1040                                 printk(CFS_KERN_WARNING "wanted to write %u "
1041                                        "but wrote %d\n", tage->used, rc);
1042                                 put_pages_back(&pc);
1043                                 __LASSERT(cfs_list_empty(&pc.pc_pages));
1044                         }
1045                 }
1046                 CFS_MMSPACE_CLOSE;
1047
1048                 cfs_filp_close(filp);
1049                 put_pages_on_daemon_list(&pc);
1050                 if (!cfs_list_empty(&pc.pc_pages)) {
1051                         int i;
1052
1053                         printk(CFS_KERN_ALERT "Lustre: trace pages aren't "
1054                                " empty\n");
1055                         printk(CFS_KERN_ERR "total cpus(%d): ",
1056                                cfs_num_possible_cpus());
1057                         for (i = 0; i < cfs_num_possible_cpus(); i++)
1058                                 if (cpu_online(i))
1059                                         printk(CFS_KERN_ERR "%d(on) ", i);
1060                                 else
1061                                         printk(CFS_KERN_ERR "%d(off) ", i);
1062                         printk(CFS_KERN_ERR "\n");
1063
1064                         i = 0;
1065                         cfs_list_for_each_entry_safe(tage, tmp, &pc.pc_pages,
1066                                                      linkage)
1067                                 printk(CFS_KERN_ERR "page %d belongs to cpu "
1068                                        "%d\n", ++i, tage->cpu);
1069                         printk(CFS_KERN_ERR "There are %d pages unwritten\n",
1070                                i);
1071                 }
1072                 __LASSERT(cfs_list_empty(&pc.pc_pages));
1073 end_loop:
1074                 if (cfs_atomic_read(&tctl->tctl_shutdown)) {
1075                         if (last_loop == 0) {
1076                                 last_loop = 1;
1077                                 continue;
1078                         } else {
1079                                 break;
1080                         }
1081                 }
1082                 cfs_waitlink_init(&__wait);
1083                 cfs_waitq_add(&tctl->tctl_waitq, &__wait);
1084                 cfs_set_current_state(CFS_TASK_INTERRUPTIBLE);
1085                 cfs_waitq_timedwait(&__wait, CFS_TASK_INTERRUPTIBLE,
1086                                     cfs_time_seconds(1));
1087                 cfs_waitq_del(&tctl->tctl_waitq, &__wait);
1088         }
1089         complete(&tctl->tctl_stop);
1090         return 0;
1091 }
1092
1093 int cfs_trace_start_thread(void)
1094 {
1095         struct tracefiled_ctl *tctl = &trace_tctl;
1096         int rc = 0;
1097
1098         mutex_lock(&cfs_trace_thread_mutex);
1099         if (thread_running)
1100                 goto out;
1101
1102         init_completion(&tctl->tctl_start);
1103         init_completion(&tctl->tctl_stop);
1104         cfs_waitq_init(&tctl->tctl_waitq);
1105         cfs_atomic_set(&tctl->tctl_shutdown, 0);
1106
1107         if (cfs_create_thread(tracefiled, tctl, 0) < 0) {
1108                 rc = -ECHILD;
1109                 goto out;
1110         }
1111
1112         wait_for_completion(&tctl->tctl_start);
1113         thread_running = 1;
1114 out:
1115         mutex_unlock(&cfs_trace_thread_mutex);
1116         return rc;
1117 }
1118
1119 void cfs_trace_stop_thread(void)
1120 {
1121         struct tracefiled_ctl *tctl = &trace_tctl;
1122
1123         mutex_lock(&cfs_trace_thread_mutex);
1124         if (thread_running) {
1125                 printk(CFS_KERN_INFO
1126                        "Lustre: shutting down debug daemon thread...\n");
1127                 cfs_atomic_set(&tctl->tctl_shutdown, 1);
1128                 wait_for_completion(&tctl->tctl_stop);
1129                 thread_running = 0;
1130         }
1131         mutex_unlock(&cfs_trace_thread_mutex);
1132 }
1133
1134 int cfs_tracefile_init(int max_pages)
1135 {
1136         struct cfs_trace_cpu_data *tcd;
1137         int                    i;
1138         int                    j;
1139         int                    rc;
1140         int                    factor;
1141
1142         rc = cfs_tracefile_init_arch();
1143         if (rc != 0)
1144                 return rc;
1145
1146         cfs_tcd_for_each(tcd, i, j) {
1147                 /* tcd_pages_factor is initialized int tracefile_init_arch. */
1148                 factor = tcd->tcd_pages_factor;
1149                 CFS_INIT_LIST_HEAD(&tcd->tcd_pages);
1150                 CFS_INIT_LIST_HEAD(&tcd->tcd_stock_pages);
1151                 CFS_INIT_LIST_HEAD(&tcd->tcd_daemon_pages);
1152                 tcd->tcd_cur_pages = 0;
1153                 tcd->tcd_cur_stock_pages = 0;
1154                 tcd->tcd_cur_daemon_pages = 0;
1155                 tcd->tcd_max_pages = (max_pages * factor) / 100;
1156                 LASSERT(tcd->tcd_max_pages > 0);
1157                 tcd->tcd_shutting_down = 0;
1158         }
1159
1160         return 0;
1161 }
1162
1163 static void trace_cleanup_on_all_cpus(void)
1164 {
1165         struct cfs_trace_cpu_data *tcd;
1166         struct cfs_trace_page *tage;
1167         struct cfs_trace_page *tmp;
1168         int i, cpu;
1169
1170         cfs_for_each_possible_cpu(cpu) {
1171                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
1172                         tcd->tcd_shutting_down = 1;
1173
1174                         cfs_list_for_each_entry_safe_typed(tage, tmp,
1175                                                            &tcd->tcd_pages,
1176                                                            struct cfs_trace_page,
1177                                                            linkage) {
1178                                 __LASSERT_TAGE_INVARIANT(tage);
1179
1180                                 cfs_list_del(&tage->linkage);
1181                                 cfs_tage_free(tage);
1182                         }
1183
1184                         tcd->tcd_cur_pages = 0;
1185                 }
1186         }
1187 }
1188
1189 static void cfs_trace_cleanup(void)
1190 {
1191         struct page_collection pc;
1192
1193         CFS_INIT_LIST_HEAD(&pc.pc_pages);
1194         spin_lock_init(&pc.pc_lock);
1195
1196         trace_cleanup_on_all_cpus();
1197
1198         cfs_tracefile_fini_arch();
1199 }
1200
1201 void cfs_tracefile_exit(void)
1202 {
1203         cfs_trace_stop_thread();
1204         cfs_trace_cleanup();
1205 }