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)
  84                 return NULL;
  85
  86         tage = kmalloc(sizeof(*tage), gfp);
  87         if (!tage) {
  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         __free_page(tage->page);
 100         kfree(tage);
 101         atomic_dec(&cfs_tage_allocated);
 102 }
 103
 104 static void cfs_tage_to_tail(struct cfs_trace_page *tage,
 105                              struct list_head *queue)
 106 {
 107         list_move_tail(&tage->linkage, queue);
 108 }
 109
 110 int cfs_trace_refill_stock(struct cfs_trace_cpu_data *tcd, gfp_t gfp,
 111                            struct list_head *stock)
 112 {
 113         int i;
 114
 115         /*
 116          * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
 117          * from here: this will lead to infinite recursion.
 118          */
 119
 120         for (i = 0; i + tcd->tcd_cur_stock_pages < TCD_STOCK_PAGES ; ++i) {
 121                 struct cfs_trace_page *tage;
 122
 123                 tage = cfs_tage_alloc(gfp);
 124                 if (!tage)
 125                         break;
 126                 list_add_tail(&tage->linkage, stock);
 127         }
 128         return i;
 129 }
 130
 131 /* return a page that has 'len' bytes left at the end */
 132 static struct cfs_trace_page *
 133 cfs_trace_get_tage_try(struct cfs_trace_cpu_data *tcd, unsigned long len)
 134 {
 135         struct cfs_trace_page *tage;
 136
 137         if (tcd->tcd_cur_pages > 0) {
 138                 __LASSERT(!list_empty(&tcd->tcd_pages));
 139                 tage = cfs_tage_from_list(tcd->tcd_pages.prev);
 140                 if (tage->used + len <= PAGE_CACHE_SIZE)
 141                         return tage;
 142         }
 143
 144         if (tcd->tcd_cur_pages < tcd->tcd_max_pages) {
 145                 if (tcd->tcd_cur_stock_pages > 0) {
 146                         tage = cfs_tage_from_list(tcd->tcd_stock_pages.prev);
 147                         --tcd->tcd_cur_stock_pages;
 148                         list_del_init(&tage->linkage);
 149                 } else {
 150                         tage = cfs_tage_alloc(GFP_ATOMIC);
 151                         if (unlikely(!tage)) {
 152                                 if ((!memory_pressure_get() ||
 153                                      in_interrupt()) && printk_ratelimit())
 154                                         printk(KERN_WARNING
 155                                                "cannot allocate a tage (%ld)\n",
 156                                                tcd->tcd_cur_pages);
 157                                 return NULL;
 158                         }
 159                 }
 160
 161                 tage->used = 0;
 162                 tage->cpu = smp_processor_id();
 163                 tage->type = tcd->tcd_type;
 164                 list_add_tail(&tage->linkage, &tcd->tcd_pages);
 165                 tcd->tcd_cur_pages++;
 166
 167                 if (tcd->tcd_cur_pages > 8 && thread_running) {
 168                         struct tracefiled_ctl *tctl = &trace_tctl;
 169                         /*
 170                          * wake up tracefiled to process some pages.
 171                          */
 172                         wake_up(&tctl->tctl_waitq);
 173                 }
 174                 return tage;
 175         }
 176         return NULL;
 177 }
 178
 179 static void cfs_tcd_shrink(struct cfs_trace_cpu_data *tcd)
 180 {
 181         int pgcount = tcd->tcd_cur_pages / 10;
 182         struct page_collection pc;
 183         struct cfs_trace_page *tage;
 184         struct cfs_trace_page *tmp;
 185
 186         /*
 187          * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
 188          * from here: this will lead to infinite recursion.
 189          */
 190
 191         if (printk_ratelimit())
 192                 printk(KERN_WARNING "debug daemon buffer overflowed; discarding 10%% of pages (%d of %ld)\n",
 193                        pgcount + 1, tcd->tcd_cur_pages);
 194
 195         INIT_LIST_HEAD(&pc.pc_pages);
 196
 197         list_for_each_entry_safe(tage, tmp, &tcd->tcd_pages, linkage) {
 198                 if (pgcount-- == 0)
 199                         break;
 200
 201                 list_move_tail(&tage->linkage, &pc.pc_pages);
 202                 tcd->tcd_cur_pages--;
 203         }
 204         put_pages_on_tcd_daemon_list(&pc, tcd);
 205 }
 206
 207 /* return a page that has 'len' bytes left at the end */
 208 static struct cfs_trace_page *cfs_trace_get_tage(struct cfs_trace_cpu_data *tcd,
 209                                                  unsigned long len)
 210 {
 211         struct cfs_trace_page *tage;
 212
 213         /*
 214          * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
 215          * from here: this will lead to infinite recursion.
 216          */
 217
 218         if (len > PAGE_CACHE_SIZE) {
 219                 pr_err("cowardly refusing to write %lu bytes in a page\n", len);
 220                 return NULL;
 221         }
 222
 223         tage = cfs_trace_get_tage_try(tcd, len);
 224         if (tage)
 225                 return tage;
 226         if (thread_running)
 227                 cfs_tcd_shrink(tcd);
 228         if (tcd->tcd_cur_pages > 0) {
 229                 tage = cfs_tage_from_list(tcd->tcd_pages.next);
 230                 tage->used = 0;
 231                 cfs_tage_to_tail(tage, &tcd->tcd_pages);
 232         }
 233         return tage;
 234 }
 235
 236 int libcfs_debug_msg(struct libcfs_debug_msg_data *msgdata,
 237                      const char *format, ...)
 238 {
 239         va_list args;
 240         int     rc;
 241
 242         va_start(args, format);
 243         rc = libcfs_debug_vmsg2(msgdata, format, args, NULL);
 244         va_end(args);
 245
 246         return rc;
 247 }
 248 EXPORT_SYMBOL(libcfs_debug_msg);
 249
 250 int libcfs_debug_vmsg2(struct libcfs_debug_msg_data *msgdata,
 251                        const char *format1, va_list args,
 252                        const char *format2, ...)
 253 {
 254         struct cfs_trace_cpu_data *tcd = NULL;
 255         struct ptldebug_header     header = {0};
 256         struct cfs_trace_page     *tage;
 257         /* string_buf is used only if tcd != NULL, and is always set then */
 258         char                  *string_buf = NULL;
 259         char                  *debug_buf;
 260         int                     known_size;
 261         int                     needed = 85; /* average message length */
 262         int                     max_nob;
 263         va_list             ap;
 264         int                     depth;
 265         int                     i;
 266         int                     remain;
 267         int                     mask = msgdata->msg_mask;
 268         const char              *file = kbasename(msgdata->msg_file);
 269         struct cfs_debug_limit_state   *cdls = msgdata->msg_cdls;
 270
 271         tcd = cfs_trace_get_tcd();
 272
 273         /* cfs_trace_get_tcd() grabs a lock, which disables preemption and
 274          * pins us to a particular CPU.  This avoids an smp_processor_id()
 275          * warning on Linux when debugging is enabled. */
 276         cfs_set_ptldebug_header(&header, msgdata, CDEBUG_STACK());
 277
 278         if (!tcd)               /* arch may not log in IRQ context */
 279                 goto console;
 280
 281         if (tcd->tcd_cur_pages == 0)
 282                 header.ph_flags |= PH_FLAG_FIRST_RECORD;
 283
 284         if (tcd->tcd_shutting_down) {
 285                 cfs_trace_put_tcd(tcd);
 286                 tcd = NULL;
 287                 goto console;
 288         }
 289
 290         depth = __current_nesting_level();
 291         known_size = strlen(file) + 1 + depth;
 292         if (msgdata->msg_fn)
 293                 known_size += strlen(msgdata->msg_fn) + 1;
 294
 295         if (libcfs_debug_binary)
 296                 known_size += sizeof(header);
 297
 298         /*/
 299          * '2' used because vsnprintf return real size required for output
 300          * _without_ terminating NULL.
 301          * if needed is to small for this format.
 302          */
 303         for (i = 0; i < 2; i++) {
 304                 tage = cfs_trace_get_tage(tcd, needed + known_size + 1);
 305                 if (!tage) {
 306                         if (needed + known_size > PAGE_CACHE_SIZE)
 307                                 mask |= D_ERROR;
 308
 309                         cfs_trace_put_tcd(tcd);
 310                         tcd = NULL;
 311                         goto console;
 312                 }
 313
 314                 string_buf = (char *)page_address(tage->page) +
 315                                         tage->used + known_size;
 316
 317                 max_nob = PAGE_CACHE_SIZE - tage->used - known_size;
 318                 if (max_nob <= 0) {
 319                         printk(KERN_EMERG "negative max_nob: %d\n",
 320                                max_nob);
 321                         mask |= D_ERROR;
 322                         cfs_trace_put_tcd(tcd);
 323                         tcd = NULL;
 324                         goto console;
 325                 }
 326
 327                 needed = 0;
 328                 if (format1) {
 329                         va_copy(ap, args);
 330                         needed = vsnprintf(string_buf, max_nob, format1, ap);
 331                         va_end(ap);
 332                 }
 333
 334                 if (format2) {
 335                         remain = max_nob - needed;
 336                         if (remain < 0)
 337                                 remain = 0;
 338
 339                         va_start(ap, format2);
 340                         needed += vsnprintf(string_buf + needed, remain,
 341                                             format2, ap);
 342                         va_end(ap);
 343                 }
 344
 345                 if (needed < max_nob) /* well. printing ok.. */
 346                         break;
 347         }
 348
 349         if (*(string_buf + needed - 1) != '\n')
 350                 printk(KERN_INFO "format at %s:%d:%s doesn't end in newline\n",
 351                        file, msgdata->msg_line, msgdata->msg_fn);
 352
 353         header.ph_len = known_size + needed;
 354         debug_buf = (char *)page_address(tage->page) + tage->used;
 355
 356         if (libcfs_debug_binary) {
 357                 memcpy(debug_buf, &header, sizeof(header));
 358                 tage->used += sizeof(header);
 359                 debug_buf += sizeof(header);
 360         }
 361
 362         /* indent message according to the nesting level */
 363         while (depth-- > 0) {
 364                 *(debug_buf++) = '.';
 365                 ++tage->used;
 366         }
 367
 368         strcpy(debug_buf, file);
 369         tage->used += strlen(file) + 1;
 370         debug_buf += strlen(file) + 1;
 371
 372         if (msgdata->msg_fn) {
 373                 strcpy(debug_buf, msgdata->msg_fn);
 374                 tage->used += strlen(msgdata->msg_fn) + 1;
 375                 debug_buf += strlen(msgdata->msg_fn) + 1;
 376         }
 377
 378         __LASSERT(debug_buf == string_buf);
 379
 380         tage->used += needed;
 381         __LASSERT(tage->used <= PAGE_CACHE_SIZE);
 382
 383 console:
 384         if ((mask & libcfs_printk) == 0) {
 385                 /* no console output requested */
 386                 if (tcd)
 387                         cfs_trace_put_tcd(tcd);
 388                 return 1;
 389         }
 390
 391         if (cdls) {
 392                 if (libcfs_console_ratelimit &&
 393                     cdls->cdls_next != 0 &&     /* not first time ever */
 394                     !cfs_time_after(cfs_time_current(), cdls->cdls_next)) {
 395                         /* skipping a console message */
 396                         cdls->cdls_count++;
 397                         if (tcd)
 398                                 cfs_trace_put_tcd(tcd);
 399                         return 1;
 400                 }
 401
 402                 if (cfs_time_after(cfs_time_current(), cdls->cdls_next +
 403                                                        libcfs_console_max_delay
 404                                                        + cfs_time_seconds(10))) {
 405                         /* last timeout was a long time ago */
 406                         cdls->cdls_delay /= libcfs_console_backoff * 4;
 407                 } else {
 408                         cdls->cdls_delay *= libcfs_console_backoff;
 409                 }
 410
 411                 if (cdls->cdls_delay < libcfs_console_min_delay)
 412                         cdls->cdls_delay = libcfs_console_min_delay;
 413                 else if (cdls->cdls_delay > libcfs_console_max_delay)
 414                         cdls->cdls_delay = libcfs_console_max_delay;
 415
 416                 /* ensure cdls_next is never zero after it's been seen */
 417                 cdls->cdls_next = (cfs_time_current() + cdls->cdls_delay) | 1;
 418         }
 419
 420         if (tcd) {
 421                 cfs_print_to_console(&header, mask, string_buf, needed, file,
 422                                      msgdata->msg_fn);
 423                 cfs_trace_put_tcd(tcd);
 424         } else {
 425                 string_buf = cfs_trace_get_console_buffer();
 426
 427                 needed = 0;
 428                 if (format1) {
 429                         va_copy(ap, args);
 430                         needed = vsnprintf(string_buf,
 431                                            CFS_TRACE_CONSOLE_BUFFER_SIZE,
 432                                            format1, ap);
 433                         va_end(ap);
 434                 }
 435                 if (format2) {
 436                         remain = CFS_TRACE_CONSOLE_BUFFER_SIZE - needed;
 437                         if (remain > 0) {
 438                                 va_start(ap, format2);
 439                                 needed += vsnprintf(string_buf + needed, remain,
 440                                                     format2, ap);
 441                                 va_end(ap);
 442                         }
 443                 }
 444                 cfs_print_to_console(&header, mask,
 445                                      string_buf, needed, file, msgdata->msg_fn);
 446
 447                 put_cpu();
 448         }
 449
 450         if (cdls && cdls->cdls_count != 0) {
 451                 string_buf = cfs_trace_get_console_buffer();
 452
 453                 needed = snprintf(string_buf, CFS_TRACE_CONSOLE_BUFFER_SIZE,
 454                                   "Skipped %d previous similar message%s\n",
 455                                   cdls->cdls_count,
 456                                   (cdls->cdls_count > 1) ? "s" : "");
 457
 458                 cfs_print_to_console(&header, mask,
 459                                      string_buf, needed, file, msgdata->msg_fn);
 460
 461                 put_cpu();
 462                 cdls->cdls_count = 0;
 463         }
 464
 465         return 0;
 466 }
 467 EXPORT_SYMBOL(libcfs_debug_vmsg2);
 468
 469 void
 470 cfs_trace_assertion_failed(const char *str,
 471                            struct libcfs_debug_msg_data *msgdata)
 472 {
 473         struct ptldebug_header hdr;
 474
 475         libcfs_panic_in_progress = 1;
 476         libcfs_catastrophe = 1;
 477         mb();
 478
 479         cfs_set_ptldebug_header(&hdr, msgdata, CDEBUG_STACK());
 480
 481         cfs_print_to_console(&hdr, D_EMERG, str, strlen(str),
 482                              msgdata->msg_file, msgdata->msg_fn);
 483
 484         panic("Lustre debug assertion failure\n");
 485
 486         /* not reached */
 487 }
 488
 489 static void
 490 panic_collect_pages(struct page_collection *pc)
 491 {
 492         /* Do the collect_pages job on a single CPU: assumes that all other
 493          * CPUs have been stopped during a panic.  If this isn't true for some
 494          * arch, this will have to be implemented separately in each arch.  */
 495         int                     i;
 496         int                     j;
 497         struct cfs_trace_cpu_data *tcd;
 498
 499         INIT_LIST_HEAD(&pc->pc_pages);
 500
 501         cfs_tcd_for_each(tcd, i, j) {
 502                 list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
 503                 tcd->tcd_cur_pages = 0;
 504
 505                 if (pc->pc_want_daemon_pages) {
 506                         list_splice_init(&tcd->tcd_daemon_pages, &pc->pc_pages);
 507                         tcd->tcd_cur_daemon_pages = 0;
 508                 }
 509         }
 510 }
 511
 512 static void collect_pages_on_all_cpus(struct page_collection *pc)
 513 {
 514         struct cfs_trace_cpu_data *tcd;
 515         int i, cpu;
 516
 517         for_each_possible_cpu(cpu) {
 518                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
 519                         list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
 520                         tcd->tcd_cur_pages = 0;
 521                         if (pc->pc_want_daemon_pages) {
 522                                 list_splice_init(&tcd->tcd_daemon_pages,
 523                                                  &pc->pc_pages);
 524                                 tcd->tcd_cur_daemon_pages = 0;
 525                         }
 526                 }
 527         }
 528 }
 529
 530 static void collect_pages(struct page_collection *pc)
 531 {
 532         INIT_LIST_HEAD(&pc->pc_pages);
 533
 534         if (libcfs_panic_in_progress)
 535                 panic_collect_pages(pc);
 536         else
 537                 collect_pages_on_all_cpus(pc);
 538 }
 539
 540 static void put_pages_back_on_all_cpus(struct page_collection *pc)
 541 {
 542         struct cfs_trace_cpu_data *tcd;
 543         struct list_head *cur_head;
 544         struct cfs_trace_page *tage;
 545         struct cfs_trace_page *tmp;
 546         int i, cpu;
 547
 548         for_each_possible_cpu(cpu) {
 549                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
 550                         cur_head = tcd->tcd_pages.next;
 551
 552                         list_for_each_entry_safe(tage, tmp, &pc->pc_pages,
 553                                                  linkage) {
 554
 555                                 __LASSERT_TAGE_INVARIANT(tage);
 556
 557                                 if (tage->cpu != cpu || tage->type != i)
 558                                         continue;
 559
 560                                 cfs_tage_to_tail(tage, cur_head);
 561                                 tcd->tcd_cur_pages++;
 562                         }
 563                 }
 564         }
 565 }
 566
 567 static void put_pages_back(struct page_collection *pc)
 568 {
 569         if (!libcfs_panic_in_progress)
 570                 put_pages_back_on_all_cpus(pc);
 571 }
 572
 573 /* Add pages to a per-cpu debug daemon ringbuffer.  This buffer makes sure that
 574  * we have a good amount of data at all times for dumping during an LBUG, even
 575  * if we have been steadily writing (and otherwise discarding) pages via the
 576  * debug daemon. */
 577 static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
 578                                          struct cfs_trace_cpu_data *tcd)
 579 {
 580         struct cfs_trace_page *tage;
 581         struct cfs_trace_page *tmp;
 582
 583         list_for_each_entry_safe(tage, tmp, &pc->pc_pages, linkage) {
 584
 585                 __LASSERT_TAGE_INVARIANT(tage);
 586
 587                 if (tage->cpu != tcd->tcd_cpu || tage->type != tcd->tcd_type)
 588                         continue;
 589
 590                 cfs_tage_to_tail(tage, &tcd->tcd_daemon_pages);
 591                 tcd->tcd_cur_daemon_pages++;
 592
 593                 if (tcd->tcd_cur_daemon_pages > tcd->tcd_max_pages) {
 594                         struct cfs_trace_page *victim;
 595
 596                         __LASSERT(!list_empty(&tcd->tcd_daemon_pages));
 597                         victim = cfs_tage_from_list(tcd->tcd_daemon_pages.next);
 598
 599                         __LASSERT_TAGE_INVARIANT(victim);
 600
 601                         list_del(&victim->linkage);
 602                         cfs_tage_free(victim);
 603                         tcd->tcd_cur_daemon_pages--;
 604                 }
 605         }
 606 }
 607
 608 static void put_pages_on_daemon_list(struct page_collection *pc)
 609 {
 610         struct cfs_trace_cpu_data *tcd;
 611         int i, cpu;
 612
 613         for_each_possible_cpu(cpu) {
 614                 cfs_tcd_for_each_type_lock(tcd, i, cpu)
 615                         put_pages_on_tcd_daemon_list(pc, tcd);
 616         }
 617 }
 618
 619 void cfs_trace_debug_print(void)
 620 {
 621         struct page_collection pc;
 622         struct cfs_trace_page *tage;
 623         struct cfs_trace_page *tmp;
 624
 625         pc.pc_want_daemon_pages = 1;
 626         collect_pages(&pc);
 627         list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
 628                 char *p, *file, *fn;
 629                 struct page *page;
 630
 631                 __LASSERT_TAGE_INVARIANT(tage);
 632
 633                 page = tage->page;
 634                 p = page_address(page);
 635                 while (p < ((char *)page_address(page) + tage->used)) {
 636                         struct ptldebug_header *hdr;
 637                         int len;
 638
 639                         hdr = (void *)p;
 640                         p += sizeof(*hdr);
 641                         file = p;
 642                         p += strlen(file) + 1;
 643                         fn = p;
 644                         p += strlen(fn) + 1;
 645                         len = hdr->ph_len - (int)(p - (char *)hdr);
 646
 647                         cfs_print_to_console(hdr, D_EMERG, p, len, file, fn);
 648
 649                         p += len;
 650                 }
 651
 652                 list_del(&tage->linkage);
 653                 cfs_tage_free(tage);
 654         }
 655 }
 656
 657 int cfs_tracefile_dump_all_pages(char *filename)
 658 {
 659         struct page_collection  pc;
 660         struct file             *filp;
 661         struct cfs_trace_page   *tage;
 662         struct cfs_trace_page   *tmp;
 663         char                    *buf;
 664         int rc;
 665
 666         DECL_MMSPACE;
 667
 668         cfs_tracefile_write_lock();
 669
 670         filp = filp_open(filename, O_CREAT | O_EXCL | O_WRONLY | O_LARGEFILE,
 671                          0600);
 672         if (IS_ERR(filp)) {
 673                 rc = PTR_ERR(filp);
 674                 filp = NULL;
 675                 pr_err("LustreError: can't open %s for dump: rc %d\n",
 676                        filename, rc);
 677                 goto out;
 678         }
 679
 680         pc.pc_want_daemon_pages = 1;
 681         collect_pages(&pc);
 682         if (list_empty(&pc.pc_pages)) {
 683                 rc = 0;
 684                 goto close;
 685         }
 686
 687         /* ok, for now, just write the pages.  in the future we'll be building
 688          * iobufs with the pages and calling generic_direct_IO */
 689         MMSPACE_OPEN;
 690         list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
 691
 692                 __LASSERT_TAGE_INVARIANT(tage);
 693
 694                 buf = kmap(tage->page);
 695                 rc = vfs_write(filp, (__force const char __user *)buf,
 696                                tage->used, &filp->f_pos);
 697                 kunmap(tage->page);
 698
 699                 if (rc != (int)tage->used) {
 700                         printk(KERN_WARNING "wanted to write %u but wrote %d\n",
 701                                tage->used, rc);
 702                         put_pages_back(&pc);
 703                         __LASSERT(list_empty(&pc.pc_pages));
 704                         break;
 705                 }
 706                 list_del(&tage->linkage);
 707                 cfs_tage_free(tage);
 708         }
 709         MMSPACE_CLOSE;
 710         rc = vfs_fsync(filp, 1);
 711         if (rc)
 712                 pr_err("sync returns %d\n", rc);
 713 close:
 714         filp_close(filp, NULL);
 715 out:
 716         cfs_tracefile_write_unlock();
 717         return rc;
 718 }
 719
 720 void cfs_trace_flush_pages(void)
 721 {
 722         struct page_collection pc;
 723         struct cfs_trace_page *tage;
 724         struct cfs_trace_page *tmp;
 725
 726         pc.pc_want_daemon_pages = 1;
 727         collect_pages(&pc);
 728         list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
 729
 730                 __LASSERT_TAGE_INVARIANT(tage);
 731
 732                 list_del(&tage->linkage);
 733                 cfs_tage_free(tage);
 734         }
 735 }
 736
 737 int cfs_trace_copyin_string(char *knl_buffer, int knl_buffer_nob,
 738                             const char __user *usr_buffer, int usr_buffer_nob)
 739 {
 740         int    nob;
 741
 742         if (usr_buffer_nob > knl_buffer_nob)
 743                 return -EOVERFLOW;
 744
 745         if (copy_from_user((void *)knl_buffer,
 746                            usr_buffer, usr_buffer_nob))
 747                 return -EFAULT;
 748
 749         nob = strnlen(knl_buffer, usr_buffer_nob);
 750         while (nob-- >= 0)                    /* strip trailing whitespace */
 751                 if (!isspace(knl_buffer[nob]))
 752                         break;
 753
 754         if (nob < 0)                        /* empty string */
 755                 return -EINVAL;
 756
 757         if (nob == knl_buffer_nob)            /* no space to terminate */
 758                 return -EOVERFLOW;
 759
 760         knl_buffer[nob + 1] = 0;                /* terminate */
 761         return 0;
 762 }
 763 EXPORT_SYMBOL(cfs_trace_copyin_string);
 764
 765 int cfs_trace_copyout_string(char __user *usr_buffer, int usr_buffer_nob,
 766                              const char *knl_buffer, char *append)
 767 {
 768         /*
 769          * NB if 'append' != NULL, it's a single character to append to the
 770          * copied out string - usually "\n" or "" (i.e. a terminating zero byte)
 771          */
 772         int   nob = strlen(knl_buffer);
 773
 774         if (nob > usr_buffer_nob)
 775                 nob = usr_buffer_nob;
 776
 777         if (copy_to_user(usr_buffer, knl_buffer, nob))
 778                 return -EFAULT;
 779
 780         if (append && nob < usr_buffer_nob) {
 781                 if (copy_to_user(usr_buffer + nob, append, 1))
 782                         return -EFAULT;
 783
 784                 nob++;
 785         }
 786
 787         return nob;
 788 }
 789 EXPORT_SYMBOL(cfs_trace_copyout_string);
 790
 791 int cfs_trace_allocate_string_buffer(char **str, int nob)
 792 {
 793         if (nob > 2 * PAGE_CACHE_SIZE)      /* string must be "sensible" */
 794                 return -EINVAL;
 795
 796         *str = kmalloc(nob, GFP_KERNEL | __GFP_ZERO);
 797         if (!*str)
 798                 return -ENOMEM;
 799
 800         return 0;
 801 }
 802
 803 int cfs_trace_dump_debug_buffer_usrstr(void __user *usr_str, int usr_str_nob)
 804 {
 805         char     *str;
 806         int        rc;
 807
 808         rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
 809         if (rc != 0)
 810                 return rc;
 811
 812         rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
 813                                      usr_str, usr_str_nob);
 814         if (rc != 0)
 815                 goto out;
 816
 817         if (str[0] != '/') {
 818                 rc = -EINVAL;
 819                 goto out;
 820         }
 821         rc = cfs_tracefile_dump_all_pages(str);
 822 out:
 823         kfree(str);
 824         return rc;
 825 }
 826
 827 int cfs_trace_daemon_command(char *str)
 828 {
 829         int       rc = 0;
 830
 831         cfs_tracefile_write_lock();
 832
 833         if (strcmp(str, "stop") == 0) {
 834                 cfs_tracefile_write_unlock();
 835                 cfs_trace_stop_thread();
 836                 cfs_tracefile_write_lock();
 837                 memset(cfs_tracefile, 0, sizeof(cfs_tracefile));
 838
 839         } else if (strncmp(str, "size=", 5) == 0) {
 840                 cfs_tracefile_size = simple_strtoul(str + 5, NULL, 0);
 841                 if (cfs_tracefile_size < 10 || cfs_tracefile_size > 20480)
 842                         cfs_tracefile_size = CFS_TRACEFILE_SIZE;
 843                 else
 844                         cfs_tracefile_size <<= 20;
 845
 846         } else if (strlen(str) >= sizeof(cfs_tracefile)) {
 847                 rc = -ENAMETOOLONG;
 848         } else if (str[0] != '/') {
 849                 rc = -EINVAL;
 850         } else {
 851                 strcpy(cfs_tracefile, str);
 852
 853                 printk(KERN_INFO
 854                        "Lustre: debug daemon will attempt to start writing to %s (%lukB max)\n",
 855                        cfs_tracefile,
 856                        (long)(cfs_tracefile_size >> 10));
 857
 858                 cfs_trace_start_thread();
 859         }
 860
 861         cfs_tracefile_write_unlock();
 862         return rc;
 863 }
 864
 865 int cfs_trace_daemon_command_usrstr(void __user *usr_str, int usr_str_nob)
 866 {
 867         char *str;
 868         int   rc;
 869
 870         rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
 871         if (rc != 0)
 872                 return rc;
 873
 874         rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
 875                                      usr_str, usr_str_nob);
 876         if (rc == 0)
 877                 rc = cfs_trace_daemon_command(str);
 878
 879         kfree(str);
 880         return rc;
 881 }
 882
 883 int cfs_trace_set_debug_mb(int mb)
 884 {
 885         int i;
 886         int j;
 887         int pages;
 888         int limit = cfs_trace_max_debug_mb();
 889         struct cfs_trace_cpu_data *tcd;
 890
 891         if (mb < num_possible_cpus()) {
 892                 printk(KERN_WARNING
 893                        "Lustre: %d MB is too small for debug buffer size, setting it to %d MB.\n",
 894                        mb, num_possible_cpus());
 895                 mb = num_possible_cpus();
 896         }
 897
 898         if (mb > limit) {
 899                 printk(KERN_WARNING
 900                        "Lustre: %d MB is too large for debug buffer size, setting it to %d MB.\n",
 901                        mb, limit);
 902                 mb = limit;
 903         }
 904
 905         mb /= num_possible_cpus();
 906         pages = mb << (20 - PAGE_CACHE_SHIFT);
 907
 908         cfs_tracefile_write_lock();
 909
 910         cfs_tcd_for_each(tcd, i, j)
 911                 tcd->tcd_max_pages = (pages * tcd->tcd_pages_factor) / 100;
 912
 913         cfs_tracefile_write_unlock();
 914
 915         return 0;
 916 }
 917
 918 int cfs_trace_get_debug_mb(void)
 919 {
 920         int i;
 921         int j;
 922         struct cfs_trace_cpu_data *tcd;
 923         int total_pages = 0;
 924
 925         cfs_tracefile_read_lock();
 926
 927         cfs_tcd_for_each(tcd, i, j)
 928                 total_pages += tcd->tcd_max_pages;
 929
 930         cfs_tracefile_read_unlock();
 931
 932         return (total_pages >> (20 - PAGE_CACHE_SHIFT)) + 1;
 933 }
 934
 935 static int tracefiled(void *arg)
 936 {
 937         struct page_collection pc;
 938         struct tracefiled_ctl *tctl = arg;
 939         struct cfs_trace_page *tage;
 940         struct cfs_trace_page *tmp;
 941         struct file *filp;
 942         char *buf;
 943         int last_loop = 0;
 944         int rc;
 945
 946         DECL_MMSPACE;
 947
 948         /* we're started late enough that we pick up init's fs context */
 949         /* this is so broken in uml?  what on earth is going on? */
 950
 951         complete(&tctl->tctl_start);
 952
 953         while (1) {
 954                 wait_queue_t __wait;
 955
 956                 pc.pc_want_daemon_pages = 0;
 957                 collect_pages(&pc);
 958                 if (list_empty(&pc.pc_pages))
 959                         goto end_loop;
 960
 961                 filp = NULL;
 962                 cfs_tracefile_read_lock();
 963                 if (cfs_tracefile[0] != 0) {
 964                         filp = filp_open(cfs_tracefile,
 965                                          O_CREAT | O_RDWR | O_LARGEFILE,
 966                                          0600);
 967                         if (IS_ERR(filp)) {
 968                                 rc = PTR_ERR(filp);
 969                                 filp = NULL;
 970                                 printk(KERN_WARNING "couldn't open %s: %d\n",
 971                                        cfs_tracefile, rc);
 972                         }
 973                 }
 974                 cfs_tracefile_read_unlock();
 975                 if (!filp) {
 976                         put_pages_on_daemon_list(&pc);
 977                         __LASSERT(list_empty(&pc.pc_pages));
 978                         goto end_loop;
 979                 }
 980
 981                 MMSPACE_OPEN;
 982
 983                 list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
 984                         static loff_t f_pos;
 985
 986                         __LASSERT_TAGE_INVARIANT(tage);
 987
 988                         if (f_pos >= (off_t)cfs_tracefile_size)
 989                                 f_pos = 0;
 990                         else if (f_pos > i_size_read(file_inode(filp)))
 991                                 f_pos = i_size_read(file_inode(filp));
 992
 993                         buf = kmap(tage->page);
 994                         rc = vfs_write(filp, (__force const char __user *)buf,
 995                                        tage->used, &f_pos);
 996                         kunmap(tage->page);
 997
 998                         if (rc != (int)tage->used) {
 999                                 printk(KERN_WARNING "wanted to write %u but wrote %d\n",
1000                                        tage->used, rc);
1001                                 put_pages_back(&pc);
1002                                 __LASSERT(list_empty(&pc.pc_pages));
1003                                 break;
1004                         }
1005                 }
1006                 MMSPACE_CLOSE;
1007
1008                 filp_close(filp, NULL);
1009                 put_pages_on_daemon_list(&pc);
1010                 if (!list_empty(&pc.pc_pages)) {
1011                         int i;
1012
1013                         printk(KERN_ALERT "Lustre: trace pages aren't empty\n");
1014                         pr_err("total cpus(%d): ", num_possible_cpus());
1015                         for (i = 0; i < num_possible_cpus(); i++)
1016                                 if (cpu_online(i))
1017                                         pr_cont("%d(on) ", i);
1018                                 else
1019                                         pr_cont("%d(off) ", i);
1020                         pr_cont("\n");
1021
1022                         i = 0;
1023                         list_for_each_entry_safe(tage, tmp, &pc.pc_pages,
1024                                                  linkage)
1025                                 pr_err("page %d belongs to cpu %d\n",
1026                                        ++i, tage->cpu);
1027                         pr_err("There are %d pages unwritten\n", i);
1028                 }
1029                 __LASSERT(list_empty(&pc.pc_pages));
1030 end_loop:
1031                 if (atomic_read(&tctl->tctl_shutdown)) {
1032                         if (last_loop == 0) {
1033                                 last_loop = 1;
1034                                 continue;
1035                         } else {
1036                                 break;
1037                         }
1038                 }
1039                 init_waitqueue_entry(&__wait, current);
1040                 add_wait_queue(&tctl->tctl_waitq, &__wait);
1041                 set_current_state(TASK_INTERRUPTIBLE);
1042                 schedule_timeout(cfs_time_seconds(1));
1043                 remove_wait_queue(&tctl->tctl_waitq, &__wait);
1044         }
1045         complete(&tctl->tctl_stop);
1046         return 0;
1047 }
1048
1049 int cfs_trace_start_thread(void)
1050 {
1051         struct tracefiled_ctl *tctl = &trace_tctl;
1052         struct task_struct *task;
1053         int rc = 0;
1054
1055         mutex_lock(&cfs_trace_thread_mutex);
1056         if (thread_running)
1057                 goto out;
1058
1059         init_completion(&tctl->tctl_start);
1060         init_completion(&tctl->tctl_stop);
1061         init_waitqueue_head(&tctl->tctl_waitq);
1062         atomic_set(&tctl->tctl_shutdown, 0);
1063
1064         task = kthread_run(tracefiled, tctl, "ktracefiled");
1065         if (IS_ERR(task)) {
1066                 rc = PTR_ERR(task);
1067                 goto out;
1068         }
1069
1070         wait_for_completion(&tctl->tctl_start);
1071         thread_running = 1;
1072 out:
1073         mutex_unlock(&cfs_trace_thread_mutex);
1074         return rc;
1075 }
1076
1077 void cfs_trace_stop_thread(void)
1078 {
1079         struct tracefiled_ctl *tctl = &trace_tctl;
1080
1081         mutex_lock(&cfs_trace_thread_mutex);
1082         if (thread_running) {
1083                 printk(KERN_INFO
1084                        "Lustre: shutting down debug daemon thread...\n");
1085                 atomic_set(&tctl->tctl_shutdown, 1);
1086                 wait_for_completion(&tctl->tctl_stop);
1087                 thread_running = 0;
1088         }
1089         mutex_unlock(&cfs_trace_thread_mutex);
1090 }
1091
1092 int cfs_tracefile_init(int max_pages)
1093 {
1094         struct cfs_trace_cpu_data *tcd;
1095         int                 i;
1096         int                 j;
1097         int                 rc;
1098         int                 factor;
1099
1100         rc = cfs_tracefile_init_arch();
1101         if (rc != 0)
1102                 return rc;
1103
1104         cfs_tcd_for_each(tcd, i, j) {
1105                 /* tcd_pages_factor is initialized int tracefile_init_arch. */
1106                 factor = tcd->tcd_pages_factor;
1107                 INIT_LIST_HEAD(&tcd->tcd_pages);
1108                 INIT_LIST_HEAD(&tcd->tcd_stock_pages);
1109                 INIT_LIST_HEAD(&tcd->tcd_daemon_pages);
1110                 tcd->tcd_cur_pages = 0;
1111                 tcd->tcd_cur_stock_pages = 0;
1112                 tcd->tcd_cur_daemon_pages = 0;
1113                 tcd->tcd_max_pages = (max_pages * factor) / 100;
1114                 LASSERT(tcd->tcd_max_pages > 0);
1115                 tcd->tcd_shutting_down = 0;
1116         }
1117
1118         return 0;
1119 }
1120
1121 static void trace_cleanup_on_all_cpus(void)
1122 {
1123         struct cfs_trace_cpu_data *tcd;
1124         struct cfs_trace_page *tage;
1125         struct cfs_trace_page *tmp;
1126         int i, cpu;
1127
1128         for_each_possible_cpu(cpu) {
1129                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
1130                         tcd->tcd_shutting_down = 1;
1131
1132                         list_for_each_entry_safe(tage, tmp, &tcd->tcd_pages,
1133                                                  linkage) {
1134                                 __LASSERT_TAGE_INVARIANT(tage);
1135
1136                                 list_del(&tage->linkage);
1137                                 cfs_tage_free(tage);
1138                         }
1139
1140                         tcd->tcd_cur_pages = 0;
1141                 }
1142         }
1143 }
1144
1145 static void cfs_trace_cleanup(void)
1146 {
1147         struct page_collection pc;
1148
1149         INIT_LIST_HEAD(&pc.pc_pages);
1150
1151         trace_cleanup_on_all_cpus();
1152
1153         cfs_tracefile_fini_arch();
1154 }
1155
1156 void cfs_tracefile_exit(void)
1157 {
1158         cfs_trace_stop_thread();
1159         cfs_trace_cleanup();
1160 }