kernel/drivers/staging/lustre/lustre/libcfs/tracefile.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * libcfs/libcfs/tracefile.c
  37  *
  38  * Author: Zach Brown <zab@clusterfs.com>
  39  * Author: Phil Schwan <phil@clusterfs.com>
  40  */
  41
  42 #define DEBUG_SUBSYSTEM S_LNET
  43 #define LUSTRE_TRACEFILE_PRIVATE
  44 #include "tracefile.h"
  45
  46 #include "../../include/linux/libcfs/libcfs.h"
  47
  48 /* XXX move things up to the top, comment */
  49 union cfs_trace_data_union (*cfs_trace_data[TCD_MAX_TYPES])[NR_CPUS] __cacheline_aligned;
  50
  51 char cfs_tracefile[TRACEFILE_NAME_SIZE];
  52 long long cfs_tracefile_size = CFS_TRACEFILE_SIZE;
  53 static struct tracefiled_ctl trace_tctl;
  54 static DEFINE_MUTEX(cfs_trace_thread_mutex);
  55 static int thread_running;
  56
  57 static atomic_t cfs_tage_allocated = ATOMIC_INIT(0);
  58
  59 static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
  60                                          struct cfs_trace_cpu_data *tcd);
  61
  62 static inline struct cfs_trace_page *
  63 cfs_tage_from_list(struct list_head *list)
  64 {
  65         return list_entry(list, struct cfs_trace_page, linkage);
  66 }
  67
  68 static struct cfs_trace_page *cfs_tage_alloc(gfp_t gfp)
  69 {
  70         struct page         *page;
  71         struct cfs_trace_page *tage;
  72
  73         /* My caller is trying to free memory */
  74         if (!in_interrupt() && memory_pressure_get())
  75                 return NULL;
  76
  77         /*
  78          * Don't spam console with allocation failures: they will be reported
  79          * by upper layer anyway.
  80          */
  81         gfp |= __GFP_NOWARN;
  82         page = alloc_page(gfp);
  83         if (page == NULL)
  84                 return NULL;
  85
  86         tage = kmalloc(sizeof(*tage), gfp);
  87         if (tage == NULL) {
  88                 __free_page(page);
  89                 return NULL;
  90         }
  91
  92         tage->page = page;
  93         atomic_inc(&cfs_tage_allocated);
  94         return tage;
  95 }
  96
  97 static void cfs_tage_free(struct cfs_trace_page *tage)
  98 {
  99         __LASSERT(tage != NULL);
 100         __LASSERT(tage->page != NULL);
 101
 102         __free_page(tage->page);
 103         kfree(tage);
 104         atomic_dec(&cfs_tage_allocated);
 105 }
 106
 107 static void cfs_tage_to_tail(struct cfs_trace_page *tage,
 108                              struct list_head *queue)
 109 {
 110         __LASSERT(tage != NULL);
 111         __LASSERT(queue != NULL);
 112
 113         list_move_tail(&tage->linkage, queue);
 114 }
 115
 116 int cfs_trace_refill_stock(struct cfs_trace_cpu_data *tcd, gfp_t gfp,
 117                            struct list_head *stock)
 118 {
 119         int i;
 120
 121         /*
 122          * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
 123          * from here: this will lead to infinite recursion.
 124          */
 125
 126         for (i = 0; i + tcd->tcd_cur_stock_pages < TCD_STOCK_PAGES ; ++i) {
 127                 struct cfs_trace_page *tage;
 128
 129                 tage = cfs_tage_alloc(gfp);
 130                 if (tage == NULL)
 131                         break;
 132                 list_add_tail(&tage->linkage, stock);
 133         }
 134         return i;
 135 }
 136
 137 /* return a page that has 'len' bytes left at the end */
 138 static struct cfs_trace_page *
 139 cfs_trace_get_tage_try(struct cfs_trace_cpu_data *tcd, unsigned long len)
 140 {
 141         struct cfs_trace_page *tage;
 142
 143         if (tcd->tcd_cur_pages > 0) {
 144                 __LASSERT(!list_empty(&tcd->tcd_pages));
 145                 tage = cfs_tage_from_list(tcd->tcd_pages.prev);
 146                 if (tage->used + len <= PAGE_CACHE_SIZE)
 147                         return tage;
 148         }
 149
 150         if (tcd->tcd_cur_pages < tcd->tcd_max_pages) {
 151                 if (tcd->tcd_cur_stock_pages > 0) {
 152                         tage = cfs_tage_from_list(tcd->tcd_stock_pages.prev);
 153                         --tcd->tcd_cur_stock_pages;
 154                         list_del_init(&tage->linkage);
 155                 } else {
 156                         tage = cfs_tage_alloc(GFP_ATOMIC);
 157                         if (unlikely(tage == NULL)) {
 158                                 if ((!memory_pressure_get() ||
 159                                      in_interrupt()) && printk_ratelimit())
 160                                         printk(KERN_WARNING
 161                                                "cannot allocate a tage (%ld)\n",
 162                                                tcd->tcd_cur_pages);
 163                                 return NULL;
 164                         }
 165                 }
 166
 167                 tage->used = 0;
 168                 tage->cpu = smp_processor_id();
 169                 tage->type = tcd->tcd_type;
 170                 list_add_tail(&tage->linkage, &tcd->tcd_pages);
 171                 tcd->tcd_cur_pages++;
 172
 173                 if (tcd->tcd_cur_pages > 8 && thread_running) {
 174                         struct tracefiled_ctl *tctl = &trace_tctl;
 175                         /*
 176                          * wake up tracefiled to process some pages.
 177                          */
 178                         wake_up(&tctl->tctl_waitq);
 179                 }
 180                 return tage;
 181         }
 182         return NULL;
 183 }
 184
 185 static void cfs_tcd_shrink(struct cfs_trace_cpu_data *tcd)
 186 {
 187         int pgcount = tcd->tcd_cur_pages / 10;
 188         struct page_collection pc;
 189         struct cfs_trace_page *tage;
 190         struct cfs_trace_page *tmp;
 191
 192         /*
 193          * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
 194          * from here: this will lead to infinite recursion.
 195          */
 196
 197         if (printk_ratelimit())
 198                 printk(KERN_WARNING "debug daemon buffer overflowed; discarding 10%% of pages (%d of %ld)\n",
 199                        pgcount + 1, tcd->tcd_cur_pages);
 200
 201         INIT_LIST_HEAD(&pc.pc_pages);
 202         spin_lock_init(&pc.pc_lock);
 203
 204         list_for_each_entry_safe(tage, tmp, &tcd->tcd_pages, linkage) {
 205                 if (pgcount-- == 0)
 206                         break;
 207
 208                 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 > PAGE_CACHE_SIZE) {
 226                 pr_err("cowardly refusing to write %lu bytes in a page\n", len);
 227                 return NULL;
 228         }
 229
 230         tage = cfs_trace_get_tage_try(tcd, len);
 231         if (tage != NULL)
 232                 return tage;
 233         if (thread_running)
 234                 cfs_tcd_shrink(tcd);
 235         if (tcd->tcd_cur_pages > 0) {
 236                 tage = cfs_tage_from_list(tcd->tcd_pages.next);
 237                 tage->used = 0;
 238                 cfs_tage_to_tail(tage, &tcd->tcd_pages);
 239         }
 240         return tage;
 241 }
 242
 243 int libcfs_debug_msg(struct libcfs_debug_msg_data *msgdata,
 244                      const char *format, ...)
 245 {
 246         va_list args;
 247         int     rc;
 248
 249         va_start(args, format);
 250         rc = libcfs_debug_vmsg2(msgdata, format, args, NULL);
 251         va_end(args);
 252
 253         return rc;
 254 }
 255 EXPORT_SYMBOL(libcfs_debug_msg);
 256
 257 int libcfs_debug_vmsg2(struct libcfs_debug_msg_data *msgdata,
 258                        const char *format1, va_list args,
 259                        const char *format2, ...)
 260 {
 261         struct cfs_trace_cpu_data *tcd = NULL;
 262         struct ptldebug_header     header = {0};
 263         struct cfs_trace_page     *tage;
 264         /* string_buf is used only if tcd != NULL, and is always set then */
 265         char                  *string_buf = NULL;
 266         char                  *debug_buf;
 267         int                     known_size;
 268         int                     needed = 85; /* average message length */
 269         int                     max_nob;
 270         va_list             ap;
 271         int                     depth;
 272         int                     i;
 273         int                     remain;
 274         int                     mask = msgdata->msg_mask;
 275         const char              *file = kbasename(msgdata->msg_file);
 276         struct cfs_debug_limit_state   *cdls = msgdata->msg_cdls;
 277
 278         tcd = cfs_trace_get_tcd();
 279
 280         /* cfs_trace_get_tcd() grabs a lock, which disables preemption and
 281          * pins us to a particular CPU.  This avoids an smp_processor_id()
 282          * warning on Linux when debugging is enabled. */
 283         cfs_set_ptldebug_header(&header, msgdata, CDEBUG_STACK());
 284
 285         if (tcd == NULL)                /* arch may not log in IRQ context */
 286                 goto console;
 287
 288         if (tcd->tcd_cur_pages == 0)
 289                 header.ph_flags |= PH_FLAG_FIRST_RECORD;
 290
 291         if (tcd->tcd_shutting_down) {
 292                 cfs_trace_put_tcd(tcd);
 293                 tcd = NULL;
 294                 goto console;
 295         }
 296
 297         depth = __current_nesting_level();
 298         known_size = strlen(file) + 1 + depth;
 299         if (msgdata->msg_fn)
 300                 known_size += strlen(msgdata->msg_fn) + 1;
 301
 302         if (libcfs_debug_binary)
 303                 known_size += sizeof(header);
 304
 305         /*/
 306          * '2' used because vsnprintf return real size required for output
 307          * _without_ terminating NULL.
 308          * if needed is to small for this format.
 309          */
 310         for (i = 0; i < 2; i++) {
 311                 tage = cfs_trace_get_tage(tcd, needed + known_size + 1);
 312                 if (tage == NULL) {
 313                         if (needed + known_size > PAGE_CACHE_SIZE)
 314                                 mask |= D_ERROR;
 315
 316                         cfs_trace_put_tcd(tcd);
 317                         tcd = NULL;
 318                         goto console;
 319                 }
 320
 321                 string_buf = (char *)page_address(tage->page) +
 322                                         tage->used + known_size;
 323
 324                 max_nob = PAGE_CACHE_SIZE - tage->used - known_size;
 325                 if (max_nob <= 0) {
 326                         printk(KERN_EMERG "negative max_nob: %d\n",
 327                                max_nob);
 328                         mask |= D_ERROR;
 329                         cfs_trace_put_tcd(tcd);
 330                         tcd = NULL;
 331                         goto console;
 332                 }
 333
 334                 needed = 0;
 335                 if (format1) {
 336                         va_copy(ap, args);
 337                         needed = vsnprintf(string_buf, max_nob, format1, ap);
 338                         va_end(ap);
 339                 }
 340
 341                 if (format2) {
 342                         remain = max_nob - needed;
 343                         if (remain < 0)
 344                                 remain = 0;
 345
 346                         va_start(ap, format2);
 347                         needed += vsnprintf(string_buf + needed, remain,
 348                                             format2, ap);
 349                         va_end(ap);
 350                 }
 351
 352                 if (needed < max_nob) /* well. printing ok.. */
 353                         break;
 354         }
 355
 356         if (*(string_buf+needed-1) != '\n')
 357                 printk(KERN_INFO "format at %s:%d:%s doesn't end in newline\n",
 358                        file, msgdata->msg_line, msgdata->msg_fn);
 359
 360         header.ph_len = known_size + needed;
 361         debug_buf = (char *)page_address(tage->page) + tage->used;
 362
 363         if (libcfs_debug_binary) {
 364                 memcpy(debug_buf, &header, sizeof(header));
 365                 tage->used += sizeof(header);
 366                 debug_buf += sizeof(header);
 367         }
 368
 369         /* indent message according to the nesting level */
 370         while (depth-- > 0) {
 371                 *(debug_buf++) = '.';
 372                 ++tage->used;
 373         }
 374
 375         strcpy(debug_buf, file);
 376         tage->used += strlen(file) + 1;
 377         debug_buf += strlen(file) + 1;
 378
 379         if (msgdata->msg_fn) {
 380                 strcpy(debug_buf, msgdata->msg_fn);
 381                 tage->used += strlen(msgdata->msg_fn) + 1;
 382                 debug_buf += strlen(msgdata->msg_fn) + 1;
 383         }
 384
 385         __LASSERT(debug_buf == string_buf);
 386
 387         tage->used += needed;
 388         __LASSERT (tage->used <= PAGE_CACHE_SIZE);
 389
 390 console:
 391         if ((mask & libcfs_printk) == 0) {
 392                 /* no console output requested */
 393                 if (tcd != NULL)
 394                         cfs_trace_put_tcd(tcd);
 395                 return 1;
 396         }
 397
 398         if (cdls != NULL) {
 399                 if (libcfs_console_ratelimit &&
 400                     cdls->cdls_next != 0 &&     /* not first time ever */
 401                     !cfs_time_after(cfs_time_current(), cdls->cdls_next)) {
 402                         /* skipping a console message */
 403                         cdls->cdls_count++;
 404                         if (tcd != NULL)
 405                                 cfs_trace_put_tcd(tcd);
 406                         return 1;
 407                 }
 408
 409                 if (cfs_time_after(cfs_time_current(), cdls->cdls_next +
 410                                                        libcfs_console_max_delay
 411                                                        + cfs_time_seconds(10))) {
 412                         /* last timeout was a long time ago */
 413                         cdls->cdls_delay /= libcfs_console_backoff * 4;
 414                 } else {
 415                         cdls->cdls_delay *= libcfs_console_backoff;
 416                 }
 417
 418                 if (cdls->cdls_delay < libcfs_console_min_delay)
 419                         cdls->cdls_delay = libcfs_console_min_delay;
 420                 else if (cdls->cdls_delay > libcfs_console_max_delay)
 421                         cdls->cdls_delay = libcfs_console_max_delay;
 422
 423                 /* ensure cdls_next is never zero after it's been seen */
 424                 cdls->cdls_next = (cfs_time_current() + cdls->cdls_delay) | 1;
 425         }
 426
 427         if (tcd != NULL) {
 428                 cfs_print_to_console(&header, mask, string_buf, needed, file,
 429                                      msgdata->msg_fn);
 430                 cfs_trace_put_tcd(tcd);
 431         } else {
 432                 string_buf = cfs_trace_get_console_buffer();
 433
 434                 needed = 0;
 435                 if (format1 != NULL) {
 436                         va_copy(ap, args);
 437                         needed = vsnprintf(string_buf,
 438                                            CFS_TRACE_CONSOLE_BUFFER_SIZE,
 439                                            format1, ap);
 440                         va_end(ap);
 441                 }
 442                 if (format2 != NULL) {
 443                         remain = CFS_TRACE_CONSOLE_BUFFER_SIZE - needed;
 444                         if (remain > 0) {
 445                                 va_start(ap, format2);
 446                                 needed += vsnprintf(string_buf+needed, remain,
 447                                                     format2, ap);
 448                                 va_end(ap);
 449                         }
 450                 }
 451                 cfs_print_to_console(&header, mask,
 452                                      string_buf, needed, file, msgdata->msg_fn);
 453
 454                 cfs_trace_put_console_buffer(string_buf);
 455         }
 456
 457         if (cdls != NULL && cdls->cdls_count != 0) {
 458                 string_buf = cfs_trace_get_console_buffer();
 459
 460                 needed = snprintf(string_buf, CFS_TRACE_CONSOLE_BUFFER_SIZE,
 461                                   "Skipped %d previous similar message%s\n",
 462                                   cdls->cdls_count,
 463                                   (cdls->cdls_count > 1) ? "s" : "");
 464
 465                 cfs_print_to_console(&header, mask,
 466                                      string_buf, needed, file, msgdata->msg_fn);
 467
 468                 cfs_trace_put_console_buffer(string_buf);
 469                 cdls->cdls_count = 0;
 470         }
 471
 472         return 0;
 473 }
 474 EXPORT_SYMBOL(libcfs_debug_vmsg2);
 475
 476 void
 477 cfs_trace_assertion_failed(const char *str,
 478                            struct libcfs_debug_msg_data *msgdata)
 479 {
 480         struct ptldebug_header hdr;
 481
 482         libcfs_panic_in_progress = 1;
 483         libcfs_catastrophe = 1;
 484         mb();
 485
 486         cfs_set_ptldebug_header(&hdr, msgdata, CDEBUG_STACK());
 487
 488         cfs_print_to_console(&hdr, D_EMERG, str, strlen(str),
 489                              msgdata->msg_file, msgdata->msg_fn);
 490
 491         panic("Lustre debug assertion failure\n");
 492
 493         /* not reached */
 494 }
 495
 496 static void
 497 panic_collect_pages(struct page_collection *pc)
 498 {
 499         /* Do the collect_pages job on a single CPU: assumes that all other
 500          * CPUs have been stopped during a panic.  If this isn't true for some
 501          * arch, this will have to be implemented separately in each arch.  */
 502         int                     i;
 503         int                     j;
 504         struct cfs_trace_cpu_data *tcd;
 505
 506         INIT_LIST_HEAD(&pc->pc_pages);
 507
 508         cfs_tcd_for_each(tcd, i, j) {
 509                 list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
 510                 tcd->tcd_cur_pages = 0;
 511
 512                 if (pc->pc_want_daemon_pages) {
 513                         list_splice_init(&tcd->tcd_daemon_pages,
 514                                              &pc->pc_pages);
 515                         tcd->tcd_cur_daemon_pages = 0;
 516                 }
 517         }
 518 }
 519
 520 static void collect_pages_on_all_cpus(struct page_collection *pc)
 521 {
 522         struct cfs_trace_cpu_data *tcd;
 523         int i, cpu;
 524
 525         spin_lock(&pc->pc_lock);
 526         for_each_possible_cpu(cpu) {
 527                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
 528                         list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
 529                         tcd->tcd_cur_pages = 0;
 530                         if (pc->pc_want_daemon_pages) {
 531                                 list_splice_init(&tcd->tcd_daemon_pages,
 532                                                      &pc->pc_pages);
 533                                 tcd->tcd_cur_daemon_pages = 0;
 534                         }
 535                 }
 536         }
 537         spin_unlock(&pc->pc_lock);
 538 }
 539
 540 static void collect_pages(struct page_collection *pc)
 541 {
 542         INIT_LIST_HEAD(&pc->pc_pages);
 543
 544         if (libcfs_panic_in_progress)
 545                 panic_collect_pages(pc);
 546         else
 547                 collect_pages_on_all_cpus(pc);
 548 }
 549
 550 static void put_pages_back_on_all_cpus(struct page_collection *pc)
 551 {
 552         struct cfs_trace_cpu_data *tcd;
 553         struct list_head *cur_head;
 554         struct cfs_trace_page *tage;
 555         struct cfs_trace_page *tmp;
 556         int i, cpu;
 557
 558         spin_lock(&pc->pc_lock);
 559         for_each_possible_cpu(cpu) {
 560                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
 561                         cur_head = tcd->tcd_pages.next;
 562
 563                         list_for_each_entry_safe(tage, tmp, &pc->pc_pages,
 564                                                  linkage) {
 565
 566                                 __LASSERT_TAGE_INVARIANT(tage);
 567
 568                                 if (tage->cpu != cpu || tage->type != i)
 569                                         continue;
 570
 571                                 cfs_tage_to_tail(tage, cur_head);
 572                                 tcd->tcd_cur_pages++;
 573                         }
 574                 }
 575         }
 576         spin_unlock(&pc->pc_lock);
 577 }
 578
 579 static void put_pages_back(struct page_collection *pc)
 580 {
 581         if (!libcfs_panic_in_progress)
 582                 put_pages_back_on_all_cpus(pc);
 583 }
 584
 585 /* Add pages to a per-cpu debug daemon ringbuffer.  This buffer makes sure that
 586  * we have a good amount of data at all times for dumping during an LBUG, even
 587  * if we have been steadily writing (and otherwise discarding) pages via the
 588  * debug daemon. */
 589 static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
 590                                          struct cfs_trace_cpu_data *tcd)
 591 {
 592         struct cfs_trace_page *tage;
 593         struct cfs_trace_page *tmp;
 594
 595         spin_lock(&pc->pc_lock);
 596         list_for_each_entry_safe(tage, tmp, &pc->pc_pages, linkage) {
 597
 598                 __LASSERT_TAGE_INVARIANT(tage);
 599
 600                 if (tage->cpu != tcd->tcd_cpu || tage->type != tcd->tcd_type)
 601                         continue;
 602
 603                 cfs_tage_to_tail(tage, &tcd->tcd_daemon_pages);
 604                 tcd->tcd_cur_daemon_pages++;
 605
 606                 if (tcd->tcd_cur_daemon_pages > tcd->tcd_max_pages) {
 607                         struct cfs_trace_page *victim;
 608
 609                         __LASSERT(!list_empty(&tcd->tcd_daemon_pages));
 610                         victim = cfs_tage_from_list(tcd->tcd_daemon_pages.next);
 611
 612                         __LASSERT_TAGE_INVARIANT(victim);
 613
 614                         list_del(&victim->linkage);
 615                         cfs_tage_free(victim);
 616                         tcd->tcd_cur_daemon_pages--;
 617                 }
 618         }
 619         spin_unlock(&pc->pc_lock);
 620 }
 621
 622 static void put_pages_on_daemon_list(struct page_collection *pc)
 623 {
 624         struct cfs_trace_cpu_data *tcd;
 625         int i, cpu;
 626
 627         for_each_possible_cpu(cpu) {
 628                 cfs_tcd_for_each_type_lock(tcd, i, cpu)
 629                         put_pages_on_tcd_daemon_list(pc, tcd);
 630         }
 631 }
 632
 633 void cfs_trace_debug_print(void)
 634 {
 635         struct page_collection pc;
 636         struct cfs_trace_page *tage;
 637         struct cfs_trace_page *tmp;
 638
 639         spin_lock_init(&pc.pc_lock);
 640
 641         pc.pc_want_daemon_pages = 1;
 642         collect_pages(&pc);
 643         list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
 644                 char *p, *file, *fn;
 645                 struct page *page;
 646
 647                 __LASSERT_TAGE_INVARIANT(tage);
 648
 649                 page = tage->page;
 650                 p = page_address(page);
 651                 while (p < ((char *)page_address(page) + tage->used)) {
 652                         struct ptldebug_header *hdr;
 653                         int len;
 654
 655                         hdr = (void *)p;
 656                         p += sizeof(*hdr);
 657                         file = p;
 658                         p += strlen(file) + 1;
 659                         fn = p;
 660                         p += strlen(fn) + 1;
 661                         len = hdr->ph_len - (int)(p - (char *)hdr);
 662
 663                         cfs_print_to_console(hdr, D_EMERG, p, len, file, fn);
 664
 665                         p += len;
 666                 }
 667
 668                 list_del(&tage->linkage);
 669                 cfs_tage_free(tage);
 670         }
 671 }
 672
 673 int cfs_tracefile_dump_all_pages(char *filename)
 674 {
 675         struct page_collection  pc;
 676         struct file             *filp;
 677         struct cfs_trace_page   *tage;
 678         struct cfs_trace_page   *tmp;
 679         char                    *buf;
 680         int rc;
 681
 682         DECL_MMSPACE;
 683
 684         cfs_tracefile_write_lock();
 685
 686         filp = filp_open(filename, O_CREAT|O_EXCL|O_WRONLY|O_LARGEFILE, 0600);
 687         if (IS_ERR(filp)) {
 688                 rc = PTR_ERR(filp);
 689                 filp = NULL;
 690                 pr_err("LustreError: can't open %s for dump: rc %d\n",
 691                         filename, rc);
 692                 goto out;
 693         }
 694
 695         spin_lock_init(&pc.pc_lock);
 696         pc.pc_want_daemon_pages = 1;
 697         collect_pages(&pc);
 698         if (list_empty(&pc.pc_pages)) {
 699                 rc = 0;
 700                 goto close;
 701         }
 702
 703         /* ok, for now, just write the pages.  in the future we'll be building
 704          * iobufs with the pages and calling generic_direct_IO */
 705         MMSPACE_OPEN;
 706         list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
 707
 708                 __LASSERT_TAGE_INVARIANT(tage);
 709
 710                 buf = kmap(tage->page);
 711                 rc = vfs_write(filp, (__force const char __user *)buf,
 712                                tage->used, &filp->f_pos);
 713                 kunmap(tage->page);
 714
 715                 if (rc != (int)tage->used) {
 716                         printk(KERN_WARNING "wanted to write %u but wrote %d\n",
 717                                tage->used, rc);
 718                         put_pages_back(&pc);
 719                         __LASSERT(list_empty(&pc.pc_pages));
 720                         break;
 721                 }
 722                 list_del(&tage->linkage);
 723                 cfs_tage_free(tage);
 724         }
 725         MMSPACE_CLOSE;
 726         rc = vfs_fsync(filp, 1);
 727         if (rc)
 728                 pr_err("sync returns %d\n", rc);
 729 close:
 730         filp_close(filp, NULL);
 731 out:
 732         cfs_tracefile_write_unlock();
 733         return rc;
 734 }
 735
 736 void cfs_trace_flush_pages(void)
 737 {
 738         struct page_collection pc;
 739         struct cfs_trace_page *tage;
 740         struct cfs_trace_page *tmp;
 741
 742         spin_lock_init(&pc.pc_lock);
 743
 744         pc.pc_want_daemon_pages = 1;
 745         collect_pages(&pc);
 746         list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
 747
 748                 __LASSERT_TAGE_INVARIANT(tage);
 749
 750                 list_del(&tage->linkage);
 751                 cfs_tage_free(tage);
 752         }
 753 }
 754
 755 int cfs_trace_copyin_string(char *knl_buffer, int knl_buffer_nob,
 756                             const char __user *usr_buffer, int usr_buffer_nob)
 757 {
 758         int    nob;
 759
 760         if (usr_buffer_nob > knl_buffer_nob)
 761                 return -EOVERFLOW;
 762
 763         if (copy_from_user((void *)knl_buffer,
 764                            usr_buffer, usr_buffer_nob))
 765                 return -EFAULT;
 766
 767         nob = strnlen(knl_buffer, usr_buffer_nob);
 768         while (nob-- >= 0)                    /* strip trailing whitespace */
 769                 if (!isspace(knl_buffer[nob]))
 770                         break;
 771
 772         if (nob < 0)                        /* empty string */
 773                 return -EINVAL;
 774
 775         if (nob == knl_buffer_nob)            /* no space to terminate */
 776                 return -EOVERFLOW;
 777
 778         knl_buffer[nob + 1] = 0;                /* terminate */
 779         return 0;
 780 }
 781 EXPORT_SYMBOL(cfs_trace_copyin_string);
 782
 783 int cfs_trace_copyout_string(char __user *usr_buffer, int usr_buffer_nob,
 784                              const char *knl_buffer, char *append)
 785 {
 786         /* NB if 'append' != NULL, it's a single character to append to the
 787          * copied out string - usually "\n", for /proc entries and "" (i.e. a
 788          * terminating zero byte) for sysctl entries */
 789         int   nob = strlen(knl_buffer);
 790
 791         if (nob > usr_buffer_nob)
 792                 nob = usr_buffer_nob;
 793
 794         if (copy_to_user(usr_buffer, knl_buffer, nob))
 795                 return -EFAULT;
 796
 797         if (append != NULL && nob < usr_buffer_nob) {
 798                 if (copy_to_user(usr_buffer + nob, append, 1))
 799                         return -EFAULT;
 800
 801                 nob++;
 802         }
 803
 804         return nob;
 805 }
 806 EXPORT_SYMBOL(cfs_trace_copyout_string);
 807
 808 int cfs_trace_allocate_string_buffer(char **str, int nob)
 809 {
 810         if (nob > 2 * PAGE_CACHE_SIZE)      /* string must be "sensible" */
 811                 return -EINVAL;
 812
 813         *str = kmalloc(nob, GFP_KERNEL | __GFP_ZERO);
 814         if (*str == NULL)
 815                 return -ENOMEM;
 816
 817         return 0;
 818 }
 819
 820 void cfs_trace_free_string_buffer(char *str, int nob)
 821 {
 822         kfree(str);
 823 }
 824
 825 int cfs_trace_dump_debug_buffer_usrstr(void __user *usr_str, int usr_str_nob)
 826 {
 827         char     *str;
 828         int        rc;
 829
 830         rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
 831         if (rc != 0)
 832                 return rc;
 833
 834         rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
 835                                      usr_str, usr_str_nob);
 836         if (rc != 0)
 837                 goto out;
 838
 839         if (str[0] != '/') {
 840                 rc = -EINVAL;
 841                 goto out;
 842         }
 843         rc = cfs_tracefile_dump_all_pages(str);
 844 out:
 845         cfs_trace_free_string_buffer(str, usr_str_nob + 1);
 846         return rc;
 847 }
 848
 849 int cfs_trace_daemon_command(char *str)
 850 {
 851         int       rc = 0;
 852
 853         cfs_tracefile_write_lock();
 854
 855         if (strcmp(str, "stop") == 0) {
 856                 cfs_tracefile_write_unlock();
 857                 cfs_trace_stop_thread();
 858                 cfs_tracefile_write_lock();
 859                 memset(cfs_tracefile, 0, sizeof(cfs_tracefile));
 860
 861         } else if (strncmp(str, "size=", 5) == 0) {
 862                 cfs_tracefile_size = simple_strtoul(str + 5, NULL, 0);
 863                 if (cfs_tracefile_size < 10 || cfs_tracefile_size > 20480)
 864                         cfs_tracefile_size = CFS_TRACEFILE_SIZE;
 865                 else
 866                         cfs_tracefile_size <<= 20;
 867
 868         } else if (strlen(str) >= sizeof(cfs_tracefile)) {
 869                 rc = -ENAMETOOLONG;
 870         } else if (str[0] != '/') {
 871                 rc = -EINVAL;
 872         } else {
 873                 strcpy(cfs_tracefile, str);
 874
 875                 printk(KERN_INFO
 876                        "Lustre: debug daemon will attempt to start writing to %s (%lukB max)\n",
 877                        cfs_tracefile,
 878                        (long)(cfs_tracefile_size >> 10));
 879
 880                 cfs_trace_start_thread();
 881         }
 882
 883         cfs_tracefile_write_unlock();
 884         return rc;
 885 }
 886
 887 int cfs_trace_daemon_command_usrstr(void __user *usr_str, int usr_str_nob)
 888 {
 889         char *str;
 890         int   rc;
 891
 892         rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
 893         if (rc != 0)
 894                 return rc;
 895
 896         rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
 897                                  usr_str, usr_str_nob);
 898         if (rc == 0)
 899                 rc = cfs_trace_daemon_command(str);
 900
 901         cfs_trace_free_string_buffer(str, usr_str_nob + 1);
 902         return rc;
 903 }
 904
 905 int cfs_trace_set_debug_mb(int mb)
 906 {
 907         int i;
 908         int j;
 909         int pages;
 910         int limit = cfs_trace_max_debug_mb();
 911         struct cfs_trace_cpu_data *tcd;
 912
 913         if (mb < num_possible_cpus()) {
 914                 printk(KERN_WARNING
 915                        "Lustre: %d MB is too small for debug buffer size, setting it to %d MB.\n",
 916                        mb, num_possible_cpus());
 917                 mb = num_possible_cpus();
 918         }
 919
 920         if (mb > limit) {
 921                 printk(KERN_WARNING
 922                        "Lustre: %d MB is too large for debug buffer size, setting it to %d MB.\n",
 923                        mb, limit);
 924                 mb = limit;
 925         }
 926
 927         mb /= num_possible_cpus();
 928         pages = mb << (20 - PAGE_CACHE_SHIFT);
 929
 930         cfs_tracefile_write_lock();
 931
 932         cfs_tcd_for_each(tcd, i, j)
 933                 tcd->tcd_max_pages = (pages * tcd->tcd_pages_factor) / 100;
 934
 935         cfs_tracefile_write_unlock();
 936
 937         return 0;
 938 }
 939
 940 int cfs_trace_get_debug_mb(void)
 941 {
 942         int i;
 943         int j;
 944         struct cfs_trace_cpu_data *tcd;
 945         int total_pages = 0;
 946
 947         cfs_tracefile_read_lock();
 948
 949         cfs_tcd_for_each(tcd, i, j)
 950                 total_pages += tcd->tcd_max_pages;
 951
 952         cfs_tracefile_read_unlock();
 953
 954         return (total_pages >> (20 - PAGE_CACHE_SHIFT)) + 1;
 955 }
 956
 957 static int tracefiled(void *arg)
 958 {
 959         struct page_collection pc;
 960         struct tracefiled_ctl *tctl = arg;
 961         struct cfs_trace_page *tage;
 962         struct cfs_trace_page *tmp;
 963         struct file *filp;
 964         char *buf;
 965         int last_loop = 0;
 966         int rc;
 967
 968         DECL_MMSPACE;
 969
 970         /* we're started late enough that we pick up init's fs context */
 971         /* this is so broken in uml?  what on earth is going on? */
 972
 973         spin_lock_init(&pc.pc_lock);
 974         complete(&tctl->tctl_start);
 975
 976         while (1) {
 977                 wait_queue_t __wait;
 978
 979                 pc.pc_want_daemon_pages = 0;
 980                 collect_pages(&pc);
 981                 if (list_empty(&pc.pc_pages))
 982                         goto end_loop;
 983
 984                 filp = NULL;
 985                 cfs_tracefile_read_lock();
 986                 if (cfs_tracefile[0] != 0) {
 987                         filp = filp_open(cfs_tracefile,
 988                                          O_CREAT | O_RDWR | O_LARGEFILE,
 989                                          0600);
 990                         if (IS_ERR(filp)) {
 991                                 rc = PTR_ERR(filp);
 992                                 filp = NULL;
 993                                 printk(KERN_WARNING "couldn't open %s: %d\n",
 994                                        cfs_tracefile, rc);
 995                         }
 996                 }
 997                 cfs_tracefile_read_unlock();
 998                 if (filp == NULL) {
 999                         put_pages_on_daemon_list(&pc);
1000                         __LASSERT(list_empty(&pc.pc_pages));
1001                         goto end_loop;
1002                 }
1003
1004                 MMSPACE_OPEN;
1005
1006                 list_for_each_entry_safe(tage, tmp, &pc.pc_pages,
1007                                                    linkage) {
1008                         static loff_t f_pos;
1009
1010                         __LASSERT_TAGE_INVARIANT(tage);
1011
1012                         if (f_pos >= (off_t)cfs_tracefile_size)
1013                                 f_pos = 0;
1014                         else if (f_pos > i_size_read(file_inode(filp)))
1015                                 f_pos = i_size_read(file_inode(filp));
1016
1017                         buf = kmap(tage->page);
1018                         rc = vfs_write(filp, (__force const char __user *)buf,
1019                                        tage->used, &f_pos);
1020                         kunmap(tage->page);
1021
1022                         if (rc != (int)tage->used) {
1023                                 printk(KERN_WARNING "wanted to write %u but wrote %d\n",
1024                                        tage->used, rc);
1025                                 put_pages_back(&pc);
1026                                 __LASSERT(list_empty(&pc.pc_pages));
1027                                 break;
1028                         }
1029                 }
1030                 MMSPACE_CLOSE;
1031
1032                 filp_close(filp, NULL);
1033                 put_pages_on_daemon_list(&pc);
1034                 if (!list_empty(&pc.pc_pages)) {
1035                         int i;
1036
1037                         printk(KERN_ALERT "Lustre: trace pages aren't empty\n");
1038                         pr_err("total cpus(%d): ",
1039                                 num_possible_cpus());
1040                         for (i = 0; i < num_possible_cpus(); i++)
1041                                 if (cpu_online(i))
1042                                         pr_cont("%d(on) ", i);
1043                                 else
1044                                         pr_cont("%d(off) ", i);
1045                         pr_cont("\n");
1046
1047                         i = 0;
1048                         list_for_each_entry_safe(tage, tmp, &pc.pc_pages,
1049                                                      linkage)
1050                                 pr_err("page %d belongs to cpu %d\n",
1051                                         ++i, tage->cpu);
1052                         pr_err("There are %d pages unwritten\n", i);
1053                 }
1054                 __LASSERT(list_empty(&pc.pc_pages));
1055 end_loop:
1056                 if (atomic_read(&tctl->tctl_shutdown)) {
1057                         if (last_loop == 0) {
1058                                 last_loop = 1;
1059                                 continue;
1060                         } else {
1061                                 break;
1062                         }
1063                 }
1064                 init_waitqueue_entry(&__wait, current);
1065                 add_wait_queue(&tctl->tctl_waitq, &__wait);
1066                 set_current_state(TASK_INTERRUPTIBLE);
1067                 schedule_timeout(cfs_time_seconds(1));
1068                 remove_wait_queue(&tctl->tctl_waitq, &__wait);
1069         }
1070         complete(&tctl->tctl_stop);
1071         return 0;
1072 }
1073
1074 int cfs_trace_start_thread(void)
1075 {
1076         struct tracefiled_ctl *tctl = &trace_tctl;
1077         int rc = 0;
1078
1079         mutex_lock(&cfs_trace_thread_mutex);
1080         if (thread_running)
1081                 goto out;
1082
1083         init_completion(&tctl->tctl_start);
1084         init_completion(&tctl->tctl_stop);
1085         init_waitqueue_head(&tctl->tctl_waitq);
1086         atomic_set(&tctl->tctl_shutdown, 0);
1087
1088         if (IS_ERR(kthread_run(tracefiled, tctl, "ktracefiled"))) {
1089                 rc = -ECHILD;
1090                 goto out;
1091         }
1092
1093         wait_for_completion(&tctl->tctl_start);
1094         thread_running = 1;
1095 out:
1096         mutex_unlock(&cfs_trace_thread_mutex);
1097         return rc;
1098 }
1099
1100 void cfs_trace_stop_thread(void)
1101 {
1102         struct tracefiled_ctl *tctl = &trace_tctl;
1103
1104         mutex_lock(&cfs_trace_thread_mutex);
1105         if (thread_running) {
1106                 printk(KERN_INFO
1107                        "Lustre: shutting down debug daemon thread...\n");
1108                 atomic_set(&tctl->tctl_shutdown, 1);
1109                 wait_for_completion(&tctl->tctl_stop);
1110                 thread_running = 0;
1111         }
1112         mutex_unlock(&cfs_trace_thread_mutex);
1113 }
1114
1115 int cfs_tracefile_init(int max_pages)
1116 {
1117         struct cfs_trace_cpu_data *tcd;
1118         int                 i;
1119         int                 j;
1120         int                 rc;
1121         int                 factor;
1122
1123         rc = cfs_tracefile_init_arch();
1124         if (rc != 0)
1125                 return rc;
1126
1127         cfs_tcd_for_each(tcd, i, j) {
1128                 /* tcd_pages_factor is initialized int tracefile_init_arch. */
1129                 factor = tcd->tcd_pages_factor;
1130                 INIT_LIST_HEAD(&tcd->tcd_pages);
1131                 INIT_LIST_HEAD(&tcd->tcd_stock_pages);
1132                 INIT_LIST_HEAD(&tcd->tcd_daemon_pages);
1133                 tcd->tcd_cur_pages = 0;
1134                 tcd->tcd_cur_stock_pages = 0;
1135                 tcd->tcd_cur_daemon_pages = 0;
1136                 tcd->tcd_max_pages = (max_pages * factor) / 100;
1137                 LASSERT(tcd->tcd_max_pages > 0);
1138                 tcd->tcd_shutting_down = 0;
1139         }
1140
1141         return 0;
1142 }
1143
1144 static void trace_cleanup_on_all_cpus(void)
1145 {
1146         struct cfs_trace_cpu_data *tcd;
1147         struct cfs_trace_page *tage;
1148         struct cfs_trace_page *tmp;
1149         int i, cpu;
1150
1151         for_each_possible_cpu(cpu) {
1152                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
1153                         tcd->tcd_shutting_down = 1;
1154
1155                         list_for_each_entry_safe(tage, tmp, &tcd->tcd_pages,
1156                                                            linkage) {
1157                                 __LASSERT_TAGE_INVARIANT(tage);
1158
1159                                 list_del(&tage->linkage);
1160                                 cfs_tage_free(tage);
1161                         }
1162
1163                         tcd->tcd_cur_pages = 0;
1164                 }
1165         }
1166 }
1167
1168 static void cfs_trace_cleanup(void)
1169 {
1170         struct page_collection pc;
1171
1172         INIT_LIST_HEAD(&pc.pc_pages);
1173         spin_lock_init(&pc.pc_lock);
1174
1175         trace_cleanup_on_all_cpus();
1176
1177         cfs_tracefile_fini_arch();
1178 }
1179
1180 void cfs_tracefile_exit(void)
1181 {
1182         cfs_trace_stop_thread();
1183         cfs_trace_cleanup();
1184 }