block/blktrace.c

   1 /*
   2  * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License version 2 as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11  * GNU General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  16  *
  17  */
  18 #include <linux/kernel.h>
  19 #include <linux/blkdev.h>
  20 #include <linux/blktrace_api.h>
  21 #include <linux/percpu.h>
  22 #include <linux/init.h>
  23 #include <linux/mutex.h>
  24 #include <linux/debugfs.h>
  25 #include <linux/time.h>
  26 #include <asm/uaccess.h>
  27
  28 static unsigned int blktrace_seq __read_mostly = 1;
  29
  30 /*
  31  * Send out a notify message.
  32  */
  33 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  34                        const void *data, size_t len)
  35 {
  36         struct blk_io_trace *t;
  37
  38         t = relay_reserve(bt->rchan, sizeof(*t) + len);
  39         if (t) {
  40                 const int cpu = smp_processor_id();
  41
  42                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
  43                 t->time = ktime_to_ns(ktime_get());
  44                 t->device = bt->dev;
  45                 t->action = action;
  46                 t->pid = pid;
  47                 t->cpu = cpu;
  48                 t->pdu_len = len;
  49                 memcpy((void *) t + sizeof(*t), data, len);
  50         }
  51 }
  52
  53 /*
  54  * Send out a notify for this process, if we haven't done so since a trace
  55  * started
  56  */
  57 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
  58 {
  59         tsk->btrace_seq = blktrace_seq;
  60         trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
  61 }
  62
  63 static void trace_note_time(struct blk_trace *bt)
  64 {
  65         struct timespec now;
  66         unsigned long flags;
  67         u32 words[2];
  68
  69         getnstimeofday(&now);
  70         words[0] = now.tv_sec;
  71         words[1] = now.tv_nsec;
  72
  73         local_irq_save(flags);
  74         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
  75         local_irq_restore(flags);
  76 }
  77
  78 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
  79 {
  80         int n;
  81         va_list args;
  82         unsigned long flags;
  83         char *buf;
  84
  85         local_irq_save(flags);
  86         buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
  87         va_start(args, fmt);
  88         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
  89         va_end(args);
  90
  91         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
  92         local_irq_restore(flags);
  93 }
  94 EXPORT_SYMBOL_GPL(__trace_note_message);
  95
  96 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
  97                          pid_t pid)
  98 {
  99         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 100                 return 1;
 101         if (sector < bt->start_lba || sector > bt->end_lba)
 102                 return 1;
 103         if (bt->pid && pid != bt->pid)
 104                 return 1;
 105
 106         return 0;
 107 }
 108
 109 /*
 110  * Data direction bit lookup
 111  */
 112 static u32 ddir_act[2] __read_mostly = { BLK_TC_ACT(BLK_TC_READ), BLK_TC_ACT(BLK_TC_WRITE) };
 113
 114 /*
 115  * Bio action bits of interest
 116  */
 117 static u32 bio_act[9] __read_mostly = { 0, BLK_TC_ACT(BLK_TC_BARRIER), BLK_TC_ACT(BLK_TC_SYNC), 0, BLK_TC_ACT(BLK_TC_AHEAD), 0, 0, 0, BLK_TC_ACT(BLK_TC_META) };
 118
 119 /*
 120  * More could be added as needed, taking care to increment the decrementer
 121  * to get correct indexing
 122  */
 123 #define trace_barrier_bit(rw)   \
 124         (((rw) & (1 << BIO_RW_BARRIER)) >> (BIO_RW_BARRIER - 0))
 125 #define trace_sync_bit(rw)      \
 126         (((rw) & (1 << BIO_RW_SYNC)) >> (BIO_RW_SYNC - 1))
 127 #define trace_ahead_bit(rw)     \
 128         (((rw) & (1 << BIO_RW_AHEAD)) << (2 - BIO_RW_AHEAD))
 129 #define trace_meta_bit(rw)      \
 130         (((rw) & (1 << BIO_RW_META)) >> (BIO_RW_META - 3))
 131
 132 /*
 133  * The worker for the various blk_add_trace*() types. Fills out a
 134  * blk_io_trace structure and places it in a per-cpu subbuffer.
 135  */
 136 void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 137                      int rw, u32 what, int error, int pdu_len, void *pdu_data)
 138 {
 139         struct task_struct *tsk = current;
 140         struct blk_io_trace *t;
 141         unsigned long flags;
 142         unsigned long *sequence;
 143         pid_t pid;
 144         int cpu;
 145
 146         if (unlikely(bt->trace_state != Blktrace_running))
 147                 return;
 148
 149         what |= ddir_act[rw & WRITE];
 150         what |= bio_act[trace_barrier_bit(rw)];
 151         what |= bio_act[trace_sync_bit(rw)];
 152         what |= bio_act[trace_ahead_bit(rw)];
 153         what |= bio_act[trace_meta_bit(rw)];
 154
 155         pid = tsk->pid;
 156         if (unlikely(act_log_check(bt, what, sector, pid)))
 157                 return;
 158
 159         /*
 160          * A word about the locking here - we disable interrupts to reserve
 161          * some space in the relay per-cpu buffer, to prevent an irq
 162          * from coming in and stepping on our toes.
 163          */
 164         local_irq_save(flags);
 165
 166         if (unlikely(tsk->btrace_seq != blktrace_seq))
 167                 trace_note_tsk(bt, tsk);
 168
 169         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
 170         if (t) {
 171                 cpu = smp_processor_id();
 172                 sequence = per_cpu_ptr(bt->sequence, cpu);
 173
 174                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 175                 t->sequence = ++(*sequence);
 176                 t->time = ktime_to_ns(ktime_get());
 177                 t->sector = sector;
 178                 t->bytes = bytes;
 179                 t->action = what;
 180                 t->pid = pid;
 181                 t->device = bt->dev;
 182                 t->cpu = cpu;
 183                 t->error = error;
 184                 t->pdu_len = pdu_len;
 185
 186                 if (pdu_len)
 187                         memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
 188         }
 189
 190         local_irq_restore(flags);
 191 }
 192
 193 EXPORT_SYMBOL_GPL(__blk_add_trace);
 194
 195 static struct dentry *blk_tree_root;
 196 static DEFINE_MUTEX(blk_tree_mutex);
 197 static unsigned int root_users;
 198
 199 static inline void blk_remove_root(void)
 200 {
 201         if (blk_tree_root) {
 202                 debugfs_remove(blk_tree_root);
 203                 blk_tree_root = NULL;
 204         }
 205 }
 206
 207 static void blk_remove_tree(struct dentry *dir)
 208 {
 209         mutex_lock(&blk_tree_mutex);
 210         debugfs_remove(dir);
 211         if (--root_users == 0)
 212                 blk_remove_root();
 213         mutex_unlock(&blk_tree_mutex);
 214 }
 215
 216 static struct dentry *blk_create_tree(const char *blk_name)
 217 {
 218         struct dentry *dir = NULL;
 219         int created = 0;
 220
 221         mutex_lock(&blk_tree_mutex);
 222
 223         if (!blk_tree_root) {
 224                 blk_tree_root = debugfs_create_dir("block", NULL);
 225                 if (!blk_tree_root)
 226                         goto err;
 227                 created = 1;
 228         }
 229
 230         dir = debugfs_create_dir(blk_name, blk_tree_root);
 231         if (dir)
 232                 root_users++;
 233         else {
 234                 /* Delete root only if we created it */
 235                 if (created)
 236                         blk_remove_root();
 237         }
 238
 239 err:
 240         mutex_unlock(&blk_tree_mutex);
 241         return dir;
 242 }
 243
 244 static void blk_trace_cleanup(struct blk_trace *bt)
 245 {
 246         relay_close(bt->rchan);
 247         debugfs_remove(bt->dropped_file);
 248         blk_remove_tree(bt->dir);
 249         free_percpu(bt->sequence);
 250         free_percpu(bt->msg_data);
 251         kfree(bt);
 252 }
 253
 254 int blk_trace_remove(struct request_queue *q)
 255 {
 256         struct blk_trace *bt;
 257
 258         bt = xchg(&q->blk_trace, NULL);
 259         if (!bt)
 260                 return -EINVAL;
 261
 262         if (bt->trace_state == Blktrace_setup ||
 263             bt->trace_state == Blktrace_stopped)
 264                 blk_trace_cleanup(bt);
 265
 266         return 0;
 267 }
 268 EXPORT_SYMBOL_GPL(blk_trace_remove);
 269
 270 static int blk_dropped_open(struct inode *inode, struct file *filp)
 271 {
 272         filp->private_data = inode->i_private;
 273
 274         return 0;
 275 }
 276
 277 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 278                                 size_t count, loff_t *ppos)
 279 {
 280         struct blk_trace *bt = filp->private_data;
 281         char buf[16];
 282
 283         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 284
 285         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 286 }
 287
 288 static const struct file_operations blk_dropped_fops = {
 289         .owner =        THIS_MODULE,
 290         .open =         blk_dropped_open,
 291         .read =         blk_dropped_read,
 292 };
 293
 294 /*
 295  * Keep track of how many times we encountered a full subbuffer, to aid
 296  * the user space app in telling how many lost events there were.
 297  */
 298 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 299                                      void *prev_subbuf, size_t prev_padding)
 300 {
 301         struct blk_trace *bt;
 302
 303         if (!relay_buf_full(buf))
 304                 return 1;
 305
 306         bt = buf->chan->private_data;
 307         atomic_inc(&bt->dropped);
 308         return 0;
 309 }
 310
 311 static int blk_remove_buf_file_callback(struct dentry *dentry)
 312 {
 313         debugfs_remove(dentry);
 314         return 0;
 315 }
 316
 317 static struct dentry *blk_create_buf_file_callback(const char *filename,
 318                                                    struct dentry *parent,
 319                                                    int mode,
 320                                                    struct rchan_buf *buf,
 321                                                    int *is_global)
 322 {
 323         return debugfs_create_file(filename, mode, parent, buf,
 324                                         &relay_file_operations);
 325 }
 326
 327 static struct rchan_callbacks blk_relay_callbacks = {
 328         .subbuf_start           = blk_subbuf_start_callback,
 329         .create_buf_file        = blk_create_buf_file_callback,
 330         .remove_buf_file        = blk_remove_buf_file_callback,
 331 };
 332
 333 /*
 334  * Setup everything required to start tracing
 335  */
 336 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 337                         struct blk_user_trace_setup *buts)
 338 {
 339         struct blk_trace *old_bt, *bt = NULL;
 340         struct dentry *dir = NULL;
 341         int ret, i;
 342
 343         if (!buts->buf_size || !buts->buf_nr)
 344                 return -EINVAL;
 345
 346         strcpy(buts->name, name);
 347
 348         /*
 349          * some device names have larger paths - convert the slashes
 350          * to underscores for this to work as expected
 351          */
 352         for (i = 0; i < strlen(buts->name); i++)
 353                 if (buts->name[i] == '/')
 354                         buts->name[i] = '_';
 355
 356         ret = -ENOMEM;
 357         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 358         if (!bt)
 359                 goto err;
 360
 361         bt->sequence = alloc_percpu(unsigned long);
 362         if (!bt->sequence)
 363                 goto err;
 364
 365         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG);
 366         if (!bt->msg_data)
 367                 goto err;
 368
 369         ret = -ENOENT;
 370         dir = blk_create_tree(buts->name);
 371         if (!dir)
 372                 goto err;
 373
 374         bt->dir = dir;
 375         bt->dev = dev;
 376         atomic_set(&bt->dropped, 0);
 377
 378         ret = -EIO;
 379         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt, &blk_dropped_fops);
 380         if (!bt->dropped_file)
 381                 goto err;
 382
 383         bt->rchan = relay_open("trace", dir, buts->buf_size,
 384                                 buts->buf_nr, &blk_relay_callbacks, bt);
 385         if (!bt->rchan)
 386                 goto err;
 387
 388         bt->act_mask = buts->act_mask;
 389         if (!bt->act_mask)
 390                 bt->act_mask = (u16) -1;
 391
 392         bt->start_lba = buts->start_lba;
 393         bt->end_lba = buts->end_lba;
 394         if (!bt->end_lba)
 395                 bt->end_lba = -1ULL;
 396
 397         bt->pid = buts->pid;
 398         bt->trace_state = Blktrace_setup;
 399
 400         ret = -EBUSY;
 401         old_bt = xchg(&q->blk_trace, bt);
 402         if (old_bt) {
 403                 (void) xchg(&q->blk_trace, old_bt);
 404                 goto err;
 405         }
 406
 407         return 0;
 408 err:
 409         if (dir)
 410                 blk_remove_tree(dir);
 411         if (bt) {
 412                 if (bt->dropped_file)
 413                         debugfs_remove(bt->dropped_file);
 414                 free_percpu(bt->sequence);
 415                 free_percpu(bt->msg_data);
 416                 if (bt->rchan)
 417                         relay_close(bt->rchan);
 418                 kfree(bt);
 419         }
 420         return ret;
 421 }
 422
 423 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 424                     char __user *arg)
 425 {
 426         struct blk_user_trace_setup buts;
 427         int ret;
 428
 429         ret = copy_from_user(&buts, arg, sizeof(buts));
 430         if (ret)
 431                 return -EFAULT;
 432
 433         ret = do_blk_trace_setup(q, name, dev, &buts);
 434         if (ret)
 435                 return ret;
 436
 437         if (copy_to_user(arg, &buts, sizeof(buts)))
 438                 return -EFAULT;
 439
 440         return 0;
 441 }
 442 EXPORT_SYMBOL_GPL(blk_trace_setup);
 443
 444 int blk_trace_startstop(struct request_queue *q, int start)
 445 {
 446         struct blk_trace *bt;
 447         int ret;
 448
 449         if ((bt = q->blk_trace) == NULL)
 450                 return -EINVAL;
 451
 452         /*
 453          * For starting a trace, we can transition from a setup or stopped
 454          * trace. For stopping a trace, the state must be running
 455          */
 456         ret = -EINVAL;
 457         if (start) {
 458                 if (bt->trace_state == Blktrace_setup ||
 459                     bt->trace_state == Blktrace_stopped) {
 460                         blktrace_seq++;
 461                         smp_mb();
 462                         bt->trace_state = Blktrace_running;
 463
 464                         trace_note_time(bt);
 465                         ret = 0;
 466                 }
 467         } else {
 468                 if (bt->trace_state == Blktrace_running) {
 469                         bt->trace_state = Blktrace_stopped;
 470                         relay_flush(bt->rchan);
 471                         ret = 0;
 472                 }
 473         }
 474
 475         return ret;
 476 }
 477 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 478
 479 /**
 480  * blk_trace_ioctl: - handle the ioctls associated with tracing
 481  * @bdev:       the block device
 482  * @cmd:        the ioctl cmd
 483  * @arg:        the argument data, if any
 484  *
 485  **/
 486 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 487 {
 488         struct request_queue *q;
 489         int ret, start = 0;
 490         char b[BDEVNAME_SIZE];
 491
 492         q = bdev_get_queue(bdev);
 493         if (!q)
 494                 return -ENXIO;
 495
 496         mutex_lock(&bdev->bd_mutex);
 497
 498         switch (cmd) {
 499         case BLKTRACESETUP:
 500                 bdevname(bdev, b);
 501                 ret = blk_trace_setup(q, b, bdev->bd_dev, arg);
 502                 break;
 503         case BLKTRACESTART:
 504                 start = 1;
 505         case BLKTRACESTOP:
 506                 ret = blk_trace_startstop(q, start);
 507                 break;
 508         case BLKTRACETEARDOWN:
 509                 ret = blk_trace_remove(q);
 510                 break;
 511         default:
 512                 ret = -ENOTTY;
 513                 break;
 514         }
 515
 516         mutex_unlock(&bdev->bd_mutex);
 517         return ret;
 518 }
 519
 520 /**
 521  * blk_trace_shutdown: - stop and cleanup trace structures
 522  * @q:    the request queue associated with the device
 523  *
 524  **/
 525 void blk_trace_shutdown(struct request_queue *q)
 526 {
 527         if (q->blk_trace) {
 528                 blk_trace_startstop(q, 0);
 529                 blk_trace_remove(q);
 530         }
 531 }