code.grnet.gr Git - archipelago/blob - xseg/peers/kernel/xsegbd.c

   1 /* xsegbd.c
   2  *
   3  */
   4
   5 #include <linux/module.h>
   6 #include <linux/moduleparam.h>
   7 #include <linux/init.h>
   8 #include <linux/sched.h>
   9 #include <linux/kernel.h>
  10 #include <linux/slab.h>
  11 #include <linux/fs.h>
  12 #include <linux/errno.h>
  13 #include <linux/timer.h>
  14 #include <linux/types.h>
  15 #include <linux/vmalloc.h>
  16 #include <linux/genhd.h>
  17 #include <linux/blkdev.h>
  18 #include <linux/bio.h>
  19 #include <linux/device.h>
  20 #include <linux/completion.h>
  21 #include <linux/wait.h>
  22 #include <sys/kernel/segdev.h>
  23 #include "xsegbd.h"
  24 #include <xseg/protocol.h>
  25
  26 #define XSEGBD_MINORS 1
  27 /* define max request size to be used in xsegbd */
  28 //FIXME should we make this 4MB instead of 256KB ?
  29 #define XSEGBD_MAX_REQUEST_SIZE 262144U
  30
  31 MODULE_DESCRIPTION("xsegbd");
  32 MODULE_AUTHOR("XSEG");
  33 MODULE_LICENSE("GPL");
  34
  35 static long sector_size = 0;
  36 static long blksize = 512;
  37 static int major = 0;
  38 static int max_dev = 1024;
  39 static char name[XSEGBD_SEGMENT_NAMELEN] = "xsegbd";
  40 static char spec[256] = "segdev:xsegbd:4:1024:12";
  41
  42 module_param(sector_size, long, 0644);
  43 module_param(blksize, long, 0644);
  44 module_param(max_dev, int, 0644);
  45 module_param(major, int, 0644);
  46 module_param_string(name, name, sizeof(name), 0644);
  47 module_param_string(spec, spec, sizeof(spec), 0644);
  48
  49 static struct xsegbd xsegbd;
  50 static struct xsegbd_device **xsegbd_devices; /* indexed by portno */
  51 static DEFINE_MUTEX(xsegbd_mutex);
  52 static DEFINE_SPINLOCK(xsegbd_devices_lock);
  53
  54
  55 void __xsegbd_get(struct xsegbd_device *xsegbd_dev)
  56 {
  57         atomic_inc(&xsegbd_dev->usercount);
  58 }
  59
  60 void __xsegbd_put(struct xsegbd_device *xsegbd_dev)
  61 {
  62         if (atomic_dec_and_test(&xsegbd_dev->usercount))
  63                 wake_up(&xsegbd_dev->wq);
  64 }
  65
  66 struct xsegbd_device *__xsegbd_get_dev(unsigned long id)
  67 {
  68         struct xsegbd_device *xsegbd_dev = NULL;
  69
  70         spin_lock(&xsegbd_devices_lock);
  71         xsegbd_dev = xsegbd_devices[id];
  72         if (xsegbd_dev)
  73                 __xsegbd_get(xsegbd_dev);
  74         spin_unlock(&xsegbd_devices_lock);
  75
  76         return xsegbd_dev;
  77 }
  78
  79 /* ************************* */
  80 /* ***** sysfs helpers ***** */
  81 /* ************************* */
  82
  83 static struct xsegbd_device *dev_to_xsegbd(struct device *dev)
  84 {
  85         return container_of(dev, struct xsegbd_device, dev);
  86 }
  87
  88 static struct device *xsegbd_get_dev(struct xsegbd_device *xsegbd_dev)
  89 {
  90         /* FIXME */
  91         return get_device(&xsegbd_dev->dev);
  92 }
  93
  94 static void xsegbd_put_dev(struct xsegbd_device *xsegbd_dev)
  95 {
  96         put_device(&xsegbd_dev->dev);
  97 }
  98
  99 /* ************************* */
 100 /* ** XSEG Initialization ** */
 101 /* ************************* */
 102
 103 static void xseg_callback(uint32_t portno);
 104
 105 int xsegbd_xseg_init(void)
 106 {
 107         int r;
 108
 109         if (!xsegbd.name[0])
 110                 strncpy(xsegbd.name, name, XSEGBD_SEGMENT_NAMELEN);
 111
 112         r = xseg_initialize();
 113         if (r) {
 114                 XSEGLOG("cannot initialize 'segdev' peer");
 115                 goto err;
 116         }
 117
 118         r = xseg_parse_spec(spec, &xsegbd.config);
 119         if (r)
 120                 goto err;
 121
 122         if (strncmp(xsegbd.config.type, "segdev", 16))
 123                 XSEGLOG("WARNING: unexpected segment type '%s' vs 'segdev'",
 124                          xsegbd.config.type);
 125
 126         /* leave it here for now */
 127         XSEGLOG("joining segment");
 128         xsegbd.xseg = xseg_join(        xsegbd.config.type,
 129                                         xsegbd.config.name,
 130                                         "segdev",
 131                                         xseg_callback           );
 132         if (!xsegbd.xseg) {
 133                 XSEGLOG("cannot find segment");
 134                 r = -ENODEV;
 135                 goto err;
 136         }
 137
 138         return 0;
 139 err:
 140         return r;
 141
 142 }
 143
 144 int xsegbd_xseg_quit(void)
 145 {
 146         struct segdev *segdev;
 147
 148         /* make sure to unmap the segment first */
 149         segdev = segdev_get(0);
 150         clear_bit(SEGDEV_RESERVED, &segdev->flags);
 151         xsegbd.xseg->priv->segment_type.ops.unmap(xsegbd.xseg, xsegbd.xseg->segment_size);
 152         segdev_put(segdev);
 153
 154         return 0;
 155 }
 156
 157
 158 /* ***************************** */
 159 /* ** Block Device Operations ** */
 160 /* ***************************** */
 161
 162 static int xsegbd_open(struct block_device *bdev, fmode_t mode)
 163 {
 164         struct gendisk *disk = bdev->bd_disk;
 165         struct xsegbd_device *xsegbd_dev = disk->private_data;
 166
 167         xsegbd_get_dev(xsegbd_dev);
 168
 169         return 0;
 170 }
 171
 172 static int xsegbd_release(struct gendisk *gd, fmode_t mode)
 173 {
 174         struct xsegbd_device *xsegbd_dev = gd->private_data;
 175
 176         xsegbd_put_dev(xsegbd_dev);
 177
 178         return 0;
 179 }
 180
 181 static int xsegbd_ioctl(struct block_device *bdev, fmode_t mode,
 182                         unsigned int cmd, unsigned long arg)
 183 {
 184         return -ENOTTY;
 185 }
 186
 187 static const struct block_device_operations xsegbd_ops = {
 188         .owner          = THIS_MODULE,
 189         .open           = xsegbd_open,
 190         .release        = xsegbd_release,
 191         .ioctl          = xsegbd_ioctl
 192 };
 193
 194
 195 /* *************************** */
 196 /* ** Device Initialization ** */
 197 /* *************************** */
 198
 199 static void xseg_request_fn(struct request_queue *rq);
 200 static int xsegbd_get_size(struct xsegbd_device *xsegbd_dev);
 201 static int xsegbd_mapclose(struct xsegbd_device *xsegbd_dev);
 202
 203 static int xsegbd_dev_init(struct xsegbd_device *xsegbd_dev)
 204 {
 205         int ret = -ENOMEM;
 206         struct gendisk *disk;
 207         unsigned int max_request_size_bytes;
 208
 209         spin_lock_init(&xsegbd_dev->rqlock);
 210
 211         xsegbd_dev->xsegbd = &xsegbd;
 212
 213         xsegbd_dev->blk_queue = blk_alloc_queue(GFP_KERNEL);
 214         if (!xsegbd_dev->blk_queue)
 215                 goto out;
 216
 217         if (!blk_init_allocated_queue(xsegbd_dev->blk_queue,
 218                         xseg_request_fn, &xsegbd_dev->rqlock))
 219                 goto outqueue;
 220
 221         xsegbd_dev->blk_queue->queuedata = xsegbd_dev;
 222
 223         blk_queue_flush(xsegbd_dev->blk_queue, REQ_FLUSH | REQ_FUA);
 224         blk_queue_logical_block_size(xsegbd_dev->blk_queue, 512);
 225         blk_queue_physical_block_size(xsegbd_dev->blk_queue, blksize);
 226         blk_queue_bounce_limit(xsegbd_dev->blk_queue, BLK_BOUNCE_ANY);
 227
 228         //blk_queue_max_segments(dev->blk_queue, 512);
 229
 230         max_request_size_bytes = XSEGBD_MAX_REQUEST_SIZE;
 231         blk_queue_max_hw_sectors(xsegbd_dev->blk_queue, max_request_size_bytes >> 9);
 232         blk_queue_max_segment_size(xsegbd_dev->blk_queue, max_request_size_bytes);
 233         blk_queue_io_min(xsegbd_dev->blk_queue, max_request_size_bytes);
 234         blk_queue_io_opt(xsegbd_dev->blk_queue, max_request_size_bytes);
 235
 236         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, xsegbd_dev->blk_queue);
 237
 238         /* vkoukis says we don't need partitions */
 239         xsegbd_dev->gd = disk = alloc_disk(1);
 240         if (!disk)
 241                 goto outqueue;
 242
 243         disk->major = xsegbd_dev->major;
 244         disk->first_minor = 0; // id * XSEGBD_MINORS;
 245         disk->fops = &xsegbd_ops;
 246         disk->queue = xsegbd_dev->blk_queue;
 247         disk->private_data = xsegbd_dev;
 248         disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
 249         snprintf(disk->disk_name, 32, "xsegbd%u", xsegbd_dev->id);
 250
 251         ret = 0;
 252
 253         /* allow a non-zero sector_size parameter to override the disk size */
 254         if (sector_size)
 255                 xsegbd_dev->sectors = sector_size;
 256         else {
 257                 ret = xsegbd_get_size(xsegbd_dev);
 258                 if (ret)
 259                         goto outdisk;
 260         }
 261
 262         set_capacity(disk, xsegbd_dev->sectors);
 263         XSEGLOG("xsegbd active...");
 264         add_disk(disk); /* immediately activates the device */
 265
 266         return 0;
 267
 268
 269 outdisk:
 270         put_disk(xsegbd_dev->gd);
 271 outqueue:
 272         blk_cleanup_queue(xsegbd_dev->blk_queue);
 273 out:
 274         xsegbd_dev->blk_queue = NULL;
 275         xsegbd_dev->gd = NULL;
 276         return ret;
 277 }
 278
 279 static void xsegbd_dev_release(struct device *dev)
 280 {
 281         int ret;
 282         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 283
 284
 285         /* cleanup gendisk and blk_queue the right way */
 286         if (xsegbd_dev->gd) {
 287                 if (xsegbd_dev->gd->flags & GENHD_FL_UP)
 288                         del_gendisk(xsegbd_dev->gd);
 289
 290                 put_disk(xsegbd_dev->gd);
 291                 xsegbd_mapclose(xsegbd_dev);
 292         }
 293
 294         spin_lock(&xsegbd_devices_lock);
 295         BUG_ON(xsegbd_devices[xsegbd_dev->src_portno] != xsegbd_dev);
 296         xsegbd_devices[xsegbd_dev->src_portno] = NULL;
 297         spin_unlock(&xsegbd_devices_lock);
 298
 299 //      xseg_cancel_wait(xsegbd_dev->xseg, xsegbd_dev->src_portno);
 300         xseg_quit_local_signal(xsegbd_dev->xseg, xsegbd_dev->src_portno);
 301         /* wait for all pending operations on device to end */
 302         wait_event(xsegbd_dev->wq, atomic_read(&xsegbd_dev->usercount) <= 0);
 303         XSEGLOG("releasing id: %d", xsegbd_dev->id);
 304         if (xsegbd_dev->blk_queue)
 305                 blk_cleanup_queue(xsegbd_dev->blk_queue);
 306
 307
 308 //      if (xseg_free_requests(xsegbd_dev->xseg,
 309 //                      xsegbd_dev->src_portno, xsegbd_dev->nr_requests) < 0)
 310 //              XSEGLOG("Error trying to free requests!\n");
 311
 312
 313         //FIXME xseg_leave to free_up resources ?
 314         unregister_blkdev(xsegbd_dev->major, XSEGBD_NAME);
 315
 316         if (xsegbd_dev->blk_req_pending)
 317                 kfree(xsegbd_dev->blk_req_pending);
 318         xq_free(&xsegbd_dev->blk_queue_pending);
 319
 320         kfree(xsegbd_dev);
 321
 322         module_put(THIS_MODULE);
 323 }
 324
 325 /* ******************* */
 326 /* ** Critical Path ** */
 327 /* ******************* */
 328
 329 static void blk_to_xseg(struct xseg *xseg, struct xseg_request *xreq,
 330                         struct request *blkreq)
 331 {
 332         struct bio_vec *bvec;
 333         struct req_iterator iter;
 334         uint64_t off = 0;
 335         char *data = xseg_get_data(xseg, xreq);
 336         rq_for_each_segment(bvec, blkreq, iter) {
 337                 char *bdata = kmap_atomic(bvec->bv_page) + bvec->bv_offset;
 338                 memcpy(data + off, bdata, bvec->bv_len);
 339                 off += bvec->bv_len;
 340                 kunmap_atomic(bdata);
 341         }
 342 }
 343
 344 static void xseg_to_blk(struct xseg *xseg, struct xseg_request *xreq,
 345                         struct request *blkreq)
 346 {
 347         struct bio_vec *bvec;
 348         struct req_iterator iter;
 349         uint64_t off = 0;
 350         char *data = xseg_get_data(xseg, xreq);
 351         rq_for_each_segment(bvec, blkreq, iter) {
 352                 char *bdata = kmap_atomic(bvec->bv_page) + bvec->bv_offset;
 353                 memcpy(bdata, data + off, bvec->bv_len);
 354                 off += bvec->bv_len;
 355                 kunmap_atomic(bdata);
 356         }
 357 }
 358
 359 static void xseg_request_fn(struct request_queue *rq)
 360 {
 361         struct xseg_request *xreq;
 362         struct xsegbd_device *xsegbd_dev = rq->queuedata;
 363         struct request *blkreq;
 364         struct xsegbd_pending *pending;
 365         xqindex blkreq_idx;
 366         char *target;
 367         uint64_t datalen;
 368         xport p;
 369         int r;
 370         unsigned long flags;
 371
 372         __xsegbd_get(xsegbd_dev);
 373
 374         spin_unlock_irq(&xsegbd_dev->rqlock);
 375         for (;;) {
 376                 if (current_thread_info()->preempt_count || irqs_disabled()){
 377                         XSEGLOG("Current thread preempt_count: %d, irqs_disabled(): %lu ",
 378                                         current_thread_info()->preempt_count, irqs_disabled());
 379                 }
 380                 //XSEGLOG("Priority: %d", current_thread_info()->task->prio);
 381                 //XSEGLOG("Static priority: %d", current_thread_info()->task->static_prio);
 382                 //XSEGLOG("Normal priority: %d", current_thread_info()->task->normal_prio);
 383                 //XSEGLOG("Rt_priority: %u", current_thread_info()->task->rt_priority);
 384                 blkreq_idx = Noneidx;
 385                 xreq = xseg_get_request(xsegbd_dev->xseg, xsegbd_dev->src_portno,
 386                                 xsegbd_dev->dst_portno, X_ALLOC);
 387                 if (!xreq)
 388                         break;
 389
 390                 blkreq_idx = xq_pop_head(&xsegbd_dev->blk_queue_pending,
 391                                                 xsegbd_dev->src_portno);
 392                 if (blkreq_idx == Noneidx)
 393                         break;
 394
 395                 if (blkreq_idx >= xsegbd_dev->nr_requests) {
 396                         XSEGLOG("blkreq_idx >= xsegbd_dev->nr_requests");
 397                         BUG_ON(1);
 398                         break;
 399                 }
 400
 401
 402                 spin_lock_irqsave(&xsegbd_dev->rqlock, flags);
 403                 blkreq = blk_fetch_request(rq);
 404                 if (!blkreq){
 405                         spin_unlock_irqrestore(&xsegbd_dev->rqlock, flags);
 406                         break;
 407                 }
 408
 409                 if (blkreq->cmd_type != REQ_TYPE_FS) {
 410                         //we lose xreq here
 411                         XSEGLOG("non-fs cmd_type: %u. *shrug*", blkreq->cmd_type);
 412                         __blk_end_request_all(blkreq, 0);
 413                         spin_unlock_irqrestore(&xsegbd_dev->rqlock, flags);
 414                         continue;
 415                 }
 416                 spin_unlock_irqrestore(&xsegbd_dev->rqlock, flags);
 417                 if (current_thread_info()->preempt_count || irqs_disabled()){
 418                         XSEGLOG("Current thread preempt_count: %d, irqs_disabled(): %lu ",
 419                                         current_thread_info()->preempt_count, irqs_disabled());
 420                 }
 421
 422                 datalen = blk_rq_bytes(blkreq);
 423                 r = xseg_prep_request(xsegbd_dev->xseg, xreq,
 424                                         xsegbd_dev->targetlen, datalen);
 425                 if (r < 0) {
 426                         XSEGLOG("couldn't prep request");
 427                         blk_end_request_err(blkreq, r);
 428                         BUG_ON(1);
 429                         break;
 430                 }
 431                 r = -ENOMEM;
 432                 if (xreq->bufferlen - xsegbd_dev->targetlen < datalen){
 433                         XSEGLOG("malformed req buffers");
 434                         blk_end_request_err(blkreq, r);
 435                         BUG_ON(1);
 436                         break;
 437                 }
 438
 439                 target = xseg_get_target(xsegbd_dev->xseg, xreq);
 440                 strncpy(target, xsegbd_dev->target, xsegbd_dev->targetlen);
 441
 442                 pending = &xsegbd_dev->blk_req_pending[blkreq_idx];
 443                 pending->dev = xsegbd_dev;
 444                 pending->request = blkreq;
 445                 pending->comp = NULL;
 446
 447                 xreq->size = datalen;
 448                 xreq->offset = blk_rq_pos(blkreq) << 9;
 449                 xreq->priv = (uint64_t) blkreq_idx;
 450
 451                 /*
 452                 if (xreq->offset >= (sector_size << 9))
 453                         XSEGLOG("sector offset: %lu > %lu, flush:%u, fua:%u",
 454                                  blk_rq_pos(blkreq), sector_size,
 455                                  blkreq->cmd_flags & REQ_FLUSH,
 456                                  blkreq->cmd_flags & REQ_FUA);
 457                 */
 458
 459                 if (blkreq->cmd_flags & REQ_FLUSH)
 460                         xreq->flags |= XF_FLUSH;
 461
 462                 if (blkreq->cmd_flags & REQ_FUA)
 463                         xreq->flags |= XF_FUA;
 464
 465                 if (rq_data_dir(blkreq)) {
 466                         /* unlock for data transfers? */
 467                         blk_to_xseg(xsegbd_dev->xseg, xreq, blkreq);
 468                         xreq->op = X_WRITE;
 469                 } else {
 470                         xreq->op = X_READ;
 471                 }
 472
 473
 474                 r = -EIO;
 475                 /* xsegbd_get here. will be put on receive */
 476                 __xsegbd_get(xsegbd_dev);
 477                 p = xseg_submit(xsegbd_dev->xseg, xreq,
 478                                         xsegbd_dev->src_portno, X_ALLOC);
 479                 if (p == NoPort) {
 480                         XSEGLOG("coundn't submit req");
 481                         WARN_ON(1);
 482                         blk_end_request_err(blkreq, r);
 483                         __xsegbd_put(xsegbd_dev);
 484                         break;
 485                 }
 486                 WARN_ON(xseg_signal(xsegbd_dev->xsegbd->xseg, p) < 0);
 487         }
 488         if (xreq)
 489                 BUG_ON(xseg_put_request(xsegbd_dev->xsegbd->xseg, xreq,
 490                                         xsegbd_dev->src_portno) == -1);
 491         if (blkreq_idx != Noneidx)
 492                 BUG_ON(xq_append_head(&xsegbd_dev->blk_queue_pending,
 493                                 blkreq_idx, xsegbd_dev->src_portno) == Noneidx);
 494         spin_lock_irq(&xsegbd_dev->rqlock);
 495         __xsegbd_put(xsegbd_dev);
 496 }
 497
 498 int update_dev_sectors_from_request(    struct xsegbd_device *xsegbd_dev,
 499                                         struct xseg_request *xreq       )
 500 {
 501         void *data;
 502         if (!xreq) {
 503                 XSEGLOG("Invalid xreq");
 504                 return -EIO;
 505         }
 506
 507         if (xreq->state & XS_FAILED)
 508                 return -ENOENT;
 509
 510         if (!(xreq->state & XS_SERVED))
 511                 return -EIO;
 512
 513         data = xseg_get_data(xsegbd_dev->xseg, xreq);
 514         if (!data) {
 515                 XSEGLOG("Invalid req data");
 516                 return -EIO;
 517         }
 518         if (!xsegbd_dev) {
 519                 XSEGLOG("Invalid xsegbd_dev");
 520                 return -ENOENT;
 521         }
 522         xsegbd_dev->sectors = *((uint64_t *) data) / 512ULL;
 523         return 0;
 524 }
 525
 526 static int xsegbd_get_size(struct xsegbd_device *xsegbd_dev)
 527 {
 528         struct xseg_request *xreq;
 529         char *target;
 530         uint64_t datalen;
 531         xqindex blkreq_idx;
 532         struct xsegbd_pending *pending;
 533         struct completion comp;
 534         xport p;
 535         void *data;
 536         int ret = -EBUSY, r;
 537
 538         __xsegbd_get(xsegbd_dev);
 539
 540         xreq = xseg_get_request(xsegbd_dev->xseg, xsegbd_dev->src_portno,
 541                         xsegbd_dev->dst_portno, X_ALLOC);
 542         if (!xreq)
 543                 goto out;
 544
 545         BUG_ON(xseg_prep_request(xsegbd_dev->xseg, xreq, xsegbd_dev->targetlen,
 546                                 sizeof(struct xseg_reply_info)));
 547
 548         init_completion(&comp);
 549         blkreq_idx = xq_pop_head(&xsegbd_dev->blk_queue_pending, 1);
 550         if (blkreq_idx == Noneidx)
 551                 goto out_put;
 552
 553         pending = &xsegbd_dev->blk_req_pending[blkreq_idx];
 554         pending->dev = xsegbd_dev;
 555         pending->request = NULL;
 556         pending->comp = &comp;
 557
 558
 559         xreq->priv = (uint64_t) blkreq_idx;
 560
 561         target = xseg_get_target(xsegbd_dev->xseg, xreq);
 562         strncpy(target, xsegbd_dev->target, xsegbd_dev->targetlen);
 563         xreq->size = xreq->datalen;
 564         xreq->offset = 0;
 565         xreq->op = X_INFO;
 566
 567         xseg_prepare_wait(xsegbd_dev->xseg, xsegbd_dev->src_portno);
 568         p = xseg_submit(xsegbd_dev->xseg, xreq,
 569                                 xsegbd_dev->src_portno, X_ALLOC);
 570         if ( p == NoPort) {
 571                 XSEGLOG("couldn't submit request");
 572                 BUG_ON(1);
 573                 goto out_queue;
 574         }
 575         WARN_ON(xseg_signal(xsegbd_dev->xseg, p) < 0);
 576         XSEGLOG("Before wait for completion, comp %lx [%llu]", (unsigned long) pending->comp, (unsigned long long) blkreq_idx);
 577         wait_for_completion_interruptible(&comp);
 578         XSEGLOG("Woken up after wait_for_completion_interruptible(), comp: %lx [%llu]", (unsigned long) pending->comp, (unsigned long long) blkreq_idx);
 579         ret = update_dev_sectors_from_request(xsegbd_dev, xreq);
 580         //XSEGLOG("get_size: sectors = %ld\n", (long)xsegbd_dev->sectors);
 581 out_put:
 582         BUG_ON(xseg_put_request(xsegbd_dev->xseg, xreq, xsegbd_dev->src_portno) == -1);
 583 out:
 584         __xsegbd_put(xsegbd_dev);
 585         return ret;
 586
 587 out_queue:
 588         pending->dev = NULL;
 589         pending->comp = NULL;
 590         xq_append_head(&xsegbd_dev->blk_queue_pending, blkreq_idx, 1);
 591
 592         goto out;
 593 }
 594
 595 static int xsegbd_mapclose(struct xsegbd_device *xsegbd_dev)
 596 {
 597         struct xseg_request *xreq;
 598         char *target;
 599         uint64_t datalen;
 600         xqindex blkreq_idx;
 601         struct xsegbd_pending *pending;
 602         struct completion comp;
 603         xport p;
 604         void *data;
 605         int ret = -EBUSY, r;
 606
 607         __xsegbd_get(xsegbd_dev);
 608         xreq = xseg_get_request(xsegbd_dev->xseg, xsegbd_dev->src_portno,
 609                         xsegbd_dev->dst_portno, X_ALLOC);
 610         if (!xreq)
 611                 goto out;
 612
 613         BUG_ON(xseg_prep_request(xsegbd_dev->xseg, xreq, xsegbd_dev->targetlen, 0));
 614
 615         init_completion(&comp);
 616         blkreq_idx = xq_pop_head(&xsegbd_dev->blk_queue_pending, 1);
 617         if (blkreq_idx == Noneidx)
 618                 goto out_put;
 619
 620         pending = &xsegbd_dev->blk_req_pending[blkreq_idx];
 621         pending->dev = xsegbd_dev;
 622         pending->request = NULL;
 623         pending->comp = &comp;
 624
 625
 626         xreq->priv = (uint64_t) blkreq_idx;
 627
 628         target = xseg_get_target(xsegbd_dev->xseg, xreq);
 629         strncpy(target, xsegbd_dev->target, xsegbd_dev->targetlen);
 630         xreq->size = xreq->datalen;
 631         xreq->offset = 0;
 632         xreq->op = X_CLOSE;
 633
 634         xseg_prepare_wait(xsegbd_dev->xseg, xsegbd_dev->src_portno);
 635         p = xseg_submit(xsegbd_dev->xseg, xreq,
 636                                 xsegbd_dev->src_portno, X_ALLOC);
 637         if ( p == NoPort) {
 638                 XSEGLOG("couldn't submit request");
 639                 BUG_ON(1);
 640                 goto out_queue;
 641         }
 642         WARN_ON(xseg_signal(xsegbd_dev->xseg, p) < 0);
 643         wait_for_completion_interruptible(&comp);
 644         ret = 0;
 645         if (xreq->state & XS_FAILED)
 646                 XSEGLOG("Couldn't close disk on mapper");
 647 out_put:
 648         BUG_ON(xseg_put_request(xsegbd_dev->xseg, xreq, xsegbd_dev->src_portno) == -1);
 649 out:
 650         __xsegbd_put(xsegbd_dev);
 651         return ret;
 652
 653 out_queue:
 654         pending->dev = NULL;
 655         pending->comp = NULL;
 656         xq_append_head(&xsegbd_dev->blk_queue_pending, blkreq_idx, 1);
 657
 658         goto out;
 659 }
 660
 661 static void xseg_callback(xport portno)
 662 {
 663         struct xsegbd_device *xsegbd_dev;
 664         struct xseg_request *xreq;
 665         struct request *blkreq;
 666         struct xsegbd_pending *pending;
 667         unsigned long flags;
 668         xqindex blkreq_idx, ridx;
 669         int err;
 670         void *data;
 671
 672         xsegbd_dev  = __xsegbd_get_dev(portno);
 673         if (!xsegbd_dev) {
 674                 XSEGLOG("portno: %u has no xsegbd device assigned", portno);
 675                 WARN_ON(1);
 676                 return;
 677         }
 678
 679         for (;;) {
 680                 xseg_prepare_wait(xsegbd_dev->xseg, xsegbd_dev->src_portno);
 681                 xreq = xseg_receive(xsegbd_dev->xseg, portno, 0);
 682                 if (!xreq)
 683                         break;
 684
 685 //              xseg_cancel_wait(xsegbd_dev->xseg, xsegbd_dev->src_portno);
 686
 687                 blkreq_idx = (xqindex) xreq->priv;
 688                 if (blkreq_idx >= xsegbd_dev->nr_requests) {
 689                         WARN_ON(1);
 690                         //FIXME maybe put request?
 691                         continue;
 692                 }
 693
 694                 pending = &xsegbd_dev->blk_req_pending[blkreq_idx];
 695                 if (pending->comp) {
 696                         /* someone is blocking on this request
 697                            and will handle it when we wake them up. */
 698                         complete(pending->comp);
 699                         /* the request is blocker's responsibility so
 700                            we will not put_request(); */
 701
 702                         continue;
 703                 }
 704
 705                 /* this is now treated as a block I/O request to end */
 706                 blkreq = pending->request;
 707                 pending->request = NULL;
 708                 if (xsegbd_dev != pending->dev) {
 709                         //FIXME maybe put request?
 710                         XSEGLOG("xsegbd_dev != pending->dev");
 711                         BUG_ON(1);
 712                         continue;
 713                 }
 714                 pending->dev = NULL;
 715                 if (!blkreq){
 716                         //FIXME maybe put request?
 717                         XSEGLOG("blkreq does not exist");
 718                         BUG_ON(1);
 719                         continue;
 720                 }
 721
 722                 err = -EIO;
 723                 if (!(xreq->state & XS_SERVED))
 724                         goto blk_end;
 725
 726                 if (xreq->serviced != blk_rq_bytes(blkreq))
 727                         goto blk_end;
 728
 729                 err = 0;
 730                 if (!rq_data_dir(blkreq)){
 731                         xseg_to_blk(xsegbd_dev->xseg, xreq, blkreq);
 732                 }
 733 blk_end:
 734                 blk_end_request_all(blkreq, err);
 735
 736                 ridx = xq_append_head(&xsegbd_dev->blk_queue_pending,
 737                                         blkreq_idx, xsegbd_dev->src_portno);
 738                 if (ridx == Noneidx) {
 739                         XSEGLOG("couldnt append blkreq_idx");
 740                         WARN_ON(1);
 741                 }
 742
 743                 if (xseg_put_request(xsegbd_dev->xseg, xreq,
 744                                                 xsegbd_dev->src_portno) < 0){
 745                         XSEGLOG("couldn't put req");
 746                         BUG_ON(1);
 747                 }
 748                 __xsegbd_put(xsegbd_dev);
 749         }
 750         if (xsegbd_dev) {
 751                 spin_lock_irqsave(&xsegbd_dev->rqlock, flags);
 752                 xseg_request_fn(xsegbd_dev->blk_queue);
 753                 spin_unlock_irqrestore(&xsegbd_dev->rqlock, flags);
 754                 __xsegbd_put(xsegbd_dev);
 755         }
 756 }
 757
 758
 759 /* sysfs interface */
 760
 761 static struct bus_type xsegbd_bus_type = {
 762         .name   = "xsegbd",
 763 };
 764
 765 static ssize_t xsegbd_size_show(struct device *dev,
 766                                         struct device_attribute *attr, char *buf)
 767 {
 768         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 769
 770         return sprintf(buf, "%llu\n", (unsigned long long) xsegbd_dev->sectors * 512ULL);
 771 }
 772
 773 static ssize_t xsegbd_major_show(struct device *dev,
 774                                         struct device_attribute *attr, char *buf)
 775 {
 776         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 777
 778         return sprintf(buf, "%d\n", xsegbd_dev->major);
 779 }
 780
 781 static ssize_t xsegbd_srcport_show(struct device *dev,
 782                                         struct device_attribute *attr, char *buf)
 783 {
 784         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 785
 786         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->src_portno);
 787 }
 788
 789 static ssize_t xsegbd_dstport_show(struct device *dev,
 790                                         struct device_attribute *attr, char *buf)
 791 {
 792         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 793
 794         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->dst_portno);
 795 }
 796
 797 static ssize_t xsegbd_id_show(struct device *dev,
 798                                         struct device_attribute *attr, char *buf)
 799 {
 800         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 801
 802         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->id);
 803 }
 804
 805 static ssize_t xsegbd_reqs_show(struct device *dev,
 806                                         struct device_attribute *attr, char *buf)
 807 {
 808         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 809
 810         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->nr_requests);
 811 }
 812
 813 static ssize_t xsegbd_target_show(struct device *dev,
 814                                         struct device_attribute *attr, char *buf)
 815 {
 816         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 817
 818         return sprintf(buf, "%s\n", xsegbd_dev->target);
 819 }
 820
 821 static ssize_t xsegbd_image_refresh(struct device *dev,
 822                                         struct device_attribute *attr,
 823                                         const char *buf,
 824                                         size_t size)
 825 {
 826         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 827         int rc, ret = size;
 828
 829         mutex_lock_nested(&xsegbd_mutex, SINGLE_DEPTH_NESTING);
 830
 831         rc = xsegbd_get_size(xsegbd_dev);
 832         if (rc < 0) {
 833                 ret = rc;
 834                 goto out;
 835         }
 836
 837         set_capacity(xsegbd_dev->gd, xsegbd_dev->sectors);
 838
 839 out:
 840         mutex_unlock(&xsegbd_mutex);
 841         return ret;
 842 }
 843
 844 //FIXME
 845 static ssize_t xsegbd_cleanup(struct device *dev,
 846                                         struct device_attribute *attr,
 847                                         const char *buf,
 848                                         size_t size)
 849 {
 850         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 851         int ret = size, i;
 852         struct request *blkreq = NULL;
 853         struct xsegbd_pending *pending = NULL;
 854         struct completion *comp = NULL;
 855
 856         mutex_lock_nested(&xsegbd_mutex, SINGLE_DEPTH_NESTING);
 857         xlock_acquire(&xsegbd_dev->blk_queue_pending.lock,
 858                                 xsegbd_dev->src_portno);
 859         for (i = 0; i < xsegbd_dev->nr_requests; i++) {
 860                 if (!__xq_check(&xsegbd_dev->blk_queue_pending, i)) {
 861                         pending = &xsegbd_dev->blk_req_pending[i];
 862                         blkreq = pending->request;
 863                         pending->request = NULL;
 864                         comp = pending->comp;
 865                         pending->comp = NULL;
 866                         if (blkreq){
 867                                 XSEGLOG("Cleaning up blkreq %lx [%d]", (unsigned long) blkreq, i);
 868                                 blk_end_request_all(blkreq, -EIO);
 869                         }
 870                         if (comp){
 871                                 XSEGLOG("Cleaning up comp %lx [%d]", (unsigned long) comp, i);
 872                                 complete(comp);
 873                         }
 874                         __xq_append_tail(&xsegbd_dev->blk_queue_pending, i);
 875                 }
 876         }
 877         xlock_release(&xsegbd_dev->blk_queue_pending.lock);
 878
 879         mutex_unlock(&xsegbd_mutex);
 880         return ret;
 881 }
 882
 883 static DEVICE_ATTR(size, S_IRUGO, xsegbd_size_show, NULL);
 884 static DEVICE_ATTR(major, S_IRUGO, xsegbd_major_show, NULL);
 885 static DEVICE_ATTR(srcport, S_IRUGO, xsegbd_srcport_show, NULL);
 886 static DEVICE_ATTR(dstport, S_IRUGO, xsegbd_dstport_show, NULL);
 887 static DEVICE_ATTR(id , S_IRUGO, xsegbd_id_show, NULL);
 888 static DEVICE_ATTR(reqs , S_IRUGO, xsegbd_reqs_show, NULL);
 889 static DEVICE_ATTR(target, S_IRUGO, xsegbd_target_show, NULL);
 890 static DEVICE_ATTR(refresh , S_IWUSR, NULL, xsegbd_image_refresh);
 891 static DEVICE_ATTR(cleanup , S_IWUSR, NULL, xsegbd_cleanup);
 892
 893 static struct attribute *xsegbd_attrs[] = {
 894         &dev_attr_size.attr,
 895         &dev_attr_major.attr,
 896         &dev_attr_srcport.attr,
 897         &dev_attr_dstport.attr,
 898         &dev_attr_id.attr,
 899         &dev_attr_reqs.attr,
 900         &dev_attr_target.attr,
 901         &dev_attr_refresh.attr,
 902         &dev_attr_cleanup.attr,
 903         NULL
 904 };
 905
 906 static struct attribute_group xsegbd_attr_group = {
 907         .attrs = xsegbd_attrs,
 908 };
 909
 910 static const struct attribute_group *xsegbd_attr_groups[] = {
 911         &xsegbd_attr_group,
 912         NULL
 913 };
 914
 915 static void xsegbd_sysfs_dev_release(struct device *dev)
 916 {
 917 }
 918
 919 static struct device_type xsegbd_device_type = {
 920         .name           = "xsegbd",
 921         .groups         = xsegbd_attr_groups,
 922         .release        = xsegbd_sysfs_dev_release,
 923 };
 924
 925 static void xsegbd_root_dev_release(struct device *dev)
 926 {
 927 }
 928
 929 static struct device xsegbd_root_dev = {
 930         .init_name      = "xsegbd",
 931         .release        = xsegbd_root_dev_release,
 932 };
 933
 934 static int xsegbd_bus_add_dev(struct xsegbd_device *xsegbd_dev)
 935 {
 936         int ret = -ENOMEM;
 937         struct device *dev;
 938
 939         mutex_lock_nested(&xsegbd_mutex, SINGLE_DEPTH_NESTING);
 940         dev = &xsegbd_dev->dev;
 941
 942         dev->bus = &xsegbd_bus_type;
 943         dev->type = &xsegbd_device_type;
 944         dev->parent = &xsegbd_root_dev;
 945         dev->release = xsegbd_dev_release;
 946         dev_set_name(dev, "%d", xsegbd_dev->id);
 947
 948         ret = device_register(dev);
 949
 950         mutex_unlock(&xsegbd_mutex);
 951         return ret;
 952 }
 953
 954 static void xsegbd_bus_del_dev(struct xsegbd_device *xsegbd_dev)
 955 {
 956         device_unregister(&xsegbd_dev->dev);
 957 }
 958
 959 static ssize_t xsegbd_add(struct bus_type *bus, const char *buf, size_t count)
 960 {
 961         struct xsegbd_device *xsegbd_dev;
 962         struct xseg_port *port;
 963         ssize_t ret = -ENOMEM;
 964
 965         if (!try_module_get(THIS_MODULE))
 966                 return -ENODEV;
 967
 968         xsegbd_dev = kzalloc(sizeof(*xsegbd_dev), GFP_KERNEL);
 969         if (!xsegbd_dev)
 970                 goto out;
 971
 972         spin_lock_init(&xsegbd_dev->rqlock);
 973         INIT_LIST_HEAD(&xsegbd_dev->node);
 974         init_waitqueue_head(&xsegbd_dev->wq);
 975         atomic_set(&xsegbd_dev->usercount, 0);
 976
 977         /* parse cmd */
 978         if (sscanf(buf, "%" __stringify(XSEGBD_TARGET_NAMELEN) "s "
 979                         "%d:%d:%d", xsegbd_dev->target, &xsegbd_dev->src_portno,
 980                         &xsegbd_dev->dst_portno, &xsegbd_dev->nr_requests) < 3) {
 981                 ret = -EINVAL;
 982                 goto out_dev;
 983         }
 984         xsegbd_dev->targetlen = strlen(xsegbd_dev->target);
 985
 986         spin_lock(&xsegbd_devices_lock);
 987         if (xsegbd_devices[xsegbd_dev->src_portno] != NULL) {
 988                 ret = -EINVAL;
 989                 goto out_unlock;
 990         }
 991         xsegbd_devices[xsegbd_dev->src_portno] = xsegbd_dev;
 992         xsegbd_dev->id = xsegbd_dev->src_portno;
 993         spin_unlock(&xsegbd_devices_lock);
 994
 995         XSEGLOG("registering block device major %d", major);
 996         ret = register_blkdev(major, XSEGBD_NAME);
 997         if (ret < 0) {
 998                 XSEGLOG("cannot register block device!");
 999                 ret = -EBUSY;
1000                 goto out_delentry;
1001         }
1002         xsegbd_dev->major = ret;
1003         XSEGLOG("registered block device major %d", xsegbd_dev->major);
1004
1005         ret = xsegbd_bus_add_dev(xsegbd_dev);
1006         if (ret)
1007                 goto out_blkdev;
1008
1009         if (!xq_alloc_seq(&xsegbd_dev->blk_queue_pending,
1010                                 xsegbd_dev->nr_requests,
1011                                 xsegbd_dev->nr_requests))
1012                 goto out_bus;
1013
1014         xsegbd_dev->blk_req_pending = kzalloc(
1015                         xsegbd_dev->nr_requests *sizeof(struct xsegbd_pending),
1016                                    GFP_KERNEL);
1017         if (!xsegbd_dev->blk_req_pending)
1018                 goto out_freeq;
1019
1020
1021         XSEGLOG("joining segment");
1022         //FIXME use xsebd module config for now
1023         xsegbd_dev->xseg = xseg_join(   xsegbd.config.type,
1024                                         xsegbd.config.name,
1025                                         "segdev",
1026                                         xseg_callback           );
1027         if (!xsegbd_dev->xseg)
1028                 goto out_freepending;
1029         __sync_synchronize();
1030
1031         XSEGLOG("%s binding to source port %u (destination %u)", xsegbd_dev->target,
1032                         xsegbd_dev->src_portno, xsegbd_dev->dst_portno);
1033         port = xseg_bind_port(xsegbd_dev->xseg, xsegbd_dev->src_portno, NULL);
1034         if (!port) {
1035                 XSEGLOG("cannot bind to port");
1036                 ret = -EFAULT;
1037
1038                 goto out_xseg;
1039         }
1040
1041         if (xsegbd_dev->src_portno != xseg_portno(xsegbd_dev->xseg, port)) {
1042                 XSEGLOG("portno != xsegbd_dev->src_portno");
1043                 BUG_ON(1);
1044                 ret = -EFAULT;
1045                 goto out_xseg;
1046         }
1047         xseg_init_local_signal(xsegbd_dev->xseg, xsegbd_dev->src_portno);
1048
1049
1050         /* make sure we don't get any requests until we're ready to handle them */
1051         xseg_cancel_wait(xsegbd_dev->xseg, xseg_portno(xsegbd_dev->xseg, port));
1052
1053         ret = xsegbd_dev_init(xsegbd_dev);
1054         if (ret)
1055                 goto out_signal;
1056
1057         xseg_prepare_wait(xsegbd_dev->xseg, xseg_portno(xsegbd_dev->xseg, port));
1058         return count;
1059
1060 out_signal:
1061         xseg_quit_local_signal(xsegbd_dev->xseg, xsegbd_dev->src_portno);
1062 out_xseg:
1063         xseg_leave(xsegbd_dev->xseg);
1064
1065 out_freepending:
1066         kfree(xsegbd_dev->blk_req_pending);
1067
1068 out_freeq:
1069         xq_free(&xsegbd_dev->blk_queue_pending);
1070
1071 out_bus:
1072         xsegbd_bus_del_dev(xsegbd_dev);
1073         return ret;
1074
1075 out_blkdev:
1076         unregister_blkdev(xsegbd_dev->major, XSEGBD_NAME);
1077
1078 out_delentry:
1079         spin_lock(&xsegbd_devices_lock);
1080         xsegbd_devices[xsegbd_dev->src_portno] = NULL;
1081
1082 out_unlock:
1083         spin_unlock(&xsegbd_devices_lock);
1084
1085 out_dev:
1086         kfree(xsegbd_dev);
1087
1088 out:
1089         return ret;
1090 }
1091
1092 static ssize_t xsegbd_remove(struct bus_type *bus, const char *buf, size_t count)
1093 {
1094         struct xsegbd_device *xsegbd_dev = NULL;
1095         int id, ret;
1096         unsigned long ul_id;
1097
1098         ret = strict_strtoul(buf, 10, &ul_id);
1099         if (ret)
1100                 return ret;
1101
1102         id = (int) ul_id;
1103         if (id != ul_id)
1104                 return -EINVAL;
1105
1106         mutex_lock_nested(&xsegbd_mutex, SINGLE_DEPTH_NESTING);
1107
1108         ret = count;
1109         xsegbd_dev = __xsegbd_get_dev(id);
1110         if (!xsegbd_dev) {
1111                 ret = -ENOENT;
1112                 goto out_unlock;
1113         }
1114         __xsegbd_put(xsegbd_dev);
1115         xsegbd_bus_del_dev(xsegbd_dev);
1116
1117 out_unlock:
1118         mutex_unlock(&xsegbd_mutex);
1119         return ret;
1120 }
1121
1122 static struct bus_attribute xsegbd_bus_attrs[] = {
1123         __ATTR(add, S_IWUSR, NULL, xsegbd_add),
1124         __ATTR(remove, S_IWUSR, NULL, xsegbd_remove),
1125         __ATTR_NULL
1126 };
1127
1128 static int xsegbd_sysfs_init(void)
1129 {
1130         int ret;
1131
1132         ret = device_register(&xsegbd_root_dev);
1133         if (ret < 0)
1134                 return ret;
1135
1136         xsegbd_bus_type.bus_attrs = xsegbd_bus_attrs;
1137         ret = bus_register(&xsegbd_bus_type);
1138         if (ret < 0)
1139                 device_unregister(&xsegbd_root_dev);
1140
1141         return ret;
1142 }
1143
1144 static void xsegbd_sysfs_cleanup(void)
1145 {
1146         bus_unregister(&xsegbd_bus_type);
1147         device_unregister(&xsegbd_root_dev);
1148 }
1149
1150 /* *************************** */
1151 /* ** Module Initialization ** */
1152 /* *************************** */
1153
1154 static int __init xsegbd_init(void)
1155 {
1156         int ret = -ENOMEM;
1157         xsegbd_devices = kzalloc(max_dev * sizeof(struct xsegbd_devices *), GFP_KERNEL);
1158         if (!xsegbd_devices)
1159                 goto out;
1160
1161         spin_lock_init(&xsegbd_devices_lock);
1162
1163         ret = -ENOSYS;
1164         ret = xsegbd_xseg_init();
1165         if (ret)
1166                 goto out_free;
1167
1168         ret = xsegbd_sysfs_init();
1169         if (ret)
1170                 goto out_xseg;
1171
1172         XSEGLOG("initialization complete");
1173
1174 out:
1175         return ret;
1176
1177 out_xseg:
1178         xsegbd_xseg_quit();
1179
1180 out_free:
1181         kfree(xsegbd_devices);
1182
1183         goto out;
1184 }
1185
1186 static void __exit xsegbd_exit(void)
1187 {
1188         xsegbd_sysfs_cleanup();
1189         xsegbd_xseg_quit();
1190 }
1191
1192 module_init(xsegbd_init);
1193 module_exit(xsegbd_exit);
1194