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 static void __xsegbd_get(struct xsegbd_device *xsegbd_dev)
  56 {
  57         atomic_inc(&xsegbd_dev->usercount);
  58 }
  59
  60 static void __xsegbd_put(struct xsegbd_device *xsegbd_dev)
  61 {
  62         atomic_dec(&xsegbd_dev->usercount);
  63         wake_up(&xsegbd_dev->wq);
  64 }
  65
  66 static 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         /* wait for all pending operations on device to end */
 301         wait_event(xsegbd_dev->wq, atomic_read(&xsegbd_dev->usercount) <= 0);
 302         XSEGLOG("releasing id: %d", xsegbd_dev->id);
 303         if (xsegbd_dev->blk_queue)
 304                 blk_cleanup_queue(xsegbd_dev->blk_queue);
 305
 306
 307 //      if (xseg_free_requests(xsegbd_dev->xseg,
 308 //                      xsegbd_dev->src_portno, xsegbd_dev->nr_requests) < 0)
 309 //              XSEGLOG("Error trying to free requests!\n");
 310
 311
 312         //FIXME xseg_leave to free_up resources ?
 313         unregister_blkdev(xsegbd_dev->major, XSEGBD_NAME);
 314
 315         if (xsegbd_dev->blk_req_pending)
 316                 kfree(xsegbd_dev->blk_req_pending);
 317         xq_free(&xsegbd_dev->blk_queue_pending);
 318
 319         kfree(xsegbd_dev);
 320
 321         module_put(THIS_MODULE);
 322 }
 323
 324 /* ******************* */
 325 /* ** Critical Path ** */
 326 /* ******************* */
 327
 328 static void blk_to_xseg(struct xseg *xseg, struct xseg_request *xreq,
 329                         struct request *blkreq)
 330 {
 331         struct bio_vec *bvec;
 332         struct req_iterator iter;
 333         uint64_t off = 0;
 334         char *data = xseg_get_data(xseg, xreq);
 335         rq_for_each_segment(bvec, blkreq, iter) {
 336                 char *bdata = kmap_atomic(bvec->bv_page) + bvec->bv_offset;
 337                 memcpy(data + off, bdata, bvec->bv_len);
 338                 off += bvec->bv_len;
 339                 kunmap_atomic(bdata);
 340         }
 341 }
 342
 343 static void xseg_to_blk(struct xseg *xseg, struct xseg_request *xreq,
 344                         struct request *blkreq)
 345 {
 346         struct bio_vec *bvec;
 347         struct req_iterator iter;
 348         uint64_t off = 0;
 349         char *data = xseg_get_data(xseg, xreq);
 350         rq_for_each_segment(bvec, blkreq, iter) {
 351                 char *bdata = kmap_atomic(bvec->bv_page) + bvec->bv_offset;
 352                 memcpy(bdata, data + off, bvec->bv_len);
 353                 off += bvec->bv_len;
 354                 kunmap_atomic(bdata);
 355         }
 356 }
 357
 358 static void xseg_request_fn(struct request_queue *rq)
 359 {
 360         struct xseg_request *xreq;
 361         struct xsegbd_device *xsegbd_dev = rq->queuedata;
 362         struct request *blkreq;
 363         struct xsegbd_pending *pending;
 364         xqindex blkreq_idx;
 365         char *target;
 366         uint64_t datalen;
 367         xport p;
 368         int r;
 369         unsigned long flags;
 370
 371         __xsegbd_get(xsegbd_dev);
 372
 373         spin_unlock_irq(&xsegbd_dev->rqlock);
 374         for (;;) {
 375                 if (current_thread_info()->preempt_count || irqs_disabled()){
 376                         XSEGLOG("Current thread preempt_count: %d, irqs_disabled(): %lu ",
 377                                         current_thread_info()->preempt_count, irqs_disabled());
 378                 }
 379                 //XSEGLOG("Priority: %d", current_thread_info()->task->prio);
 380                 //XSEGLOG("Static priority: %d", current_thread_info()->task->static_prio);
 381                 //XSEGLOG("Normal priority: %d", current_thread_info()->task->normal_prio);
 382                 //XSEGLOG("Rt_priority: %u", current_thread_info()->task->rt_priority);
 383                 blkreq_idx = Noneidx;
 384                 xreq = xseg_get_request(xsegbd_dev->xseg, xsegbd_dev->src_portno,
 385                                 xsegbd_dev->dst_portno, X_ALLOC);
 386                 if (!xreq)
 387                         break;
 388
 389                 blkreq_idx = xq_pop_head(&xsegbd_dev->blk_queue_pending,
 390                                                 xsegbd_dev->src_portno);
 391                 if (blkreq_idx == Noneidx)
 392                         break;
 393
 394                 if (blkreq_idx >= xsegbd_dev->nr_requests) {
 395                         XSEGLOG("blkreq_idx >= xsegbd_dev->nr_requests");
 396                         BUG_ON(1);
 397                         break;
 398                 }
 399
 400
 401                 spin_lock_irqsave(&xsegbd_dev->rqlock, flags);
 402                 blkreq = blk_fetch_request(rq);
 403                 if (!blkreq){
 404                         spin_unlock_irqrestore(&xsegbd_dev->rqlock, flags);
 405                         break;
 406                 }
 407
 408                 if (blkreq->cmd_type != REQ_TYPE_FS) {
 409                         //we lose xreq here
 410                         XSEGLOG("non-fs cmd_type: %u. *shrug*", blkreq->cmd_type);
 411                         __blk_end_request_all(blkreq, 0);
 412                         spin_unlock_irqrestore(&xsegbd_dev->rqlock, flags);
 413                         continue;
 414                 }
 415                 spin_unlock_irqrestore(&xsegbd_dev->rqlock, flags);
 416                 if (current_thread_info()->preempt_count || irqs_disabled()){
 417                         XSEGLOG("Current thread preempt_count: %d, irqs_disabled(): %lu ",
 418                                         current_thread_info()->preempt_count, irqs_disabled());
 419                 }
 420
 421                 datalen = blk_rq_bytes(blkreq);
 422                 r = xseg_prep_request(xsegbd_dev->xseg, xreq,
 423                                         xsegbd_dev->targetlen, datalen);
 424                 if (r < 0) {
 425                         XSEGLOG("couldn't prep request");
 426                         blk_end_request_err(blkreq, r);
 427                         BUG_ON(1);
 428                         break;
 429                 }
 430                 r = -ENOMEM;
 431                 if (xreq->bufferlen - xsegbd_dev->targetlen < datalen){
 432                         XSEGLOG("malformed req buffers");
 433                         blk_end_request_err(blkreq, r);
 434                         BUG_ON(1);
 435                         break;
 436                 }
 437
 438                 target = xseg_get_target(xsegbd_dev->xseg, xreq);
 439                 strncpy(target, xsegbd_dev->target, xsegbd_dev->targetlen);
 440
 441                 pending = &xsegbd_dev->blk_req_pending[blkreq_idx];
 442                 pending->dev = xsegbd_dev;
 443                 pending->request = blkreq;
 444                 pending->comp = NULL;
 445
 446                 xreq->size = datalen;
 447                 xreq->offset = blk_rq_pos(blkreq) << 9;
 448                 xreq->priv = (uint64_t) blkreq_idx;
 449
 450                 /*
 451                 if (xreq->offset >= (sector_size << 9))
 452                         XSEGLOG("sector offset: %lu > %lu, flush:%u, fua:%u",
 453                                  blk_rq_pos(blkreq), sector_size,
 454                                  blkreq->cmd_flags & REQ_FLUSH,
 455                                  blkreq->cmd_flags & REQ_FUA);
 456                 */
 457
 458                 if (blkreq->cmd_flags & REQ_FLUSH)
 459                         xreq->flags |= XF_FLUSH;
 460
 461                 if (blkreq->cmd_flags & REQ_FUA)
 462                         xreq->flags |= XF_FUA;
 463
 464                 if (rq_data_dir(blkreq)) {
 465                         /* unlock for data transfers? */
 466                         blk_to_xseg(xsegbd_dev->xseg, xreq, blkreq);
 467                         xreq->op = X_WRITE;
 468                 } else {
 469                         xreq->op = X_READ;
 470                 }
 471
 472
 473                 r = -EIO;
 474                 p = xseg_submit(xsegbd_dev->xseg, xreq,
 475                                         xsegbd_dev->src_portno, X_ALLOC);
 476                 if (p == NoPort) {
 477                         XSEGLOG("coundn't submit req");
 478                         BUG_ON(1);
 479                         blk_end_request_err(blkreq, r);
 480                         break;
 481                 }
 482                 WARN_ON(xseg_signal(xsegbd_dev->xsegbd->xseg, p) < 0);
 483                 /* xsegbd_get here. will be put on receive */
 484                 __xsegbd_get(xsegbd_dev);
 485         }
 486         if (xreq)
 487                 BUG_ON(xseg_put_request(xsegbd_dev->xsegbd->xseg, xreq,
 488                                         xsegbd_dev->src_portno) == -1);
 489         if (blkreq_idx != Noneidx)
 490                 BUG_ON(xq_append_head(&xsegbd_dev->blk_queue_pending,
 491                                 blkreq_idx, xsegbd_dev->src_portno) == Noneidx);
 492         spin_lock_irq(&xsegbd_dev->rqlock);
 493         __xsegbd_put(xsegbd_dev);
 494 }
 495
 496 int update_dev_sectors_from_request(    struct xsegbd_device *xsegbd_dev,
 497                                         struct xseg_request *xreq       )
 498 {
 499         void *data;
 500         if (!xreq) {
 501                 XSEGLOG("Invalid xreq");
 502                 return -EIO;
 503         }
 504
 505         if (xreq->state & XS_FAILED)
 506                 return -ENOENT;
 507
 508         if (!(xreq->state & XS_SERVED))
 509                 return -EIO;
 510
 511         data = xseg_get_data(xsegbd_dev->xseg, xreq);
 512         if (!data) {
 513                 XSEGLOG("Invalid req data");
 514                 return -EIO;
 515         }
 516         if (!xsegbd_dev) {
 517                 XSEGLOG("Invalid xsegbd_dev");
 518                 return -ENOENT;
 519         }
 520         xsegbd_dev->sectors = *((uint64_t *) data) / 512ULL;
 521         return 0;
 522 }
 523
 524 static int xsegbd_get_size(struct xsegbd_device *xsegbd_dev)
 525 {
 526         struct xseg_request *xreq;
 527         char *target;
 528         uint64_t datalen;
 529         xqindex blkreq_idx;
 530         struct xsegbd_pending *pending;
 531         struct completion comp;
 532         xport p;
 533         void *data;
 534         int ret = -EBUSY, r;
 535
 536         __xsegbd_get(xsegbd_dev);
 537
 538         xreq = xseg_get_request(xsegbd_dev->xseg, xsegbd_dev->src_portno,
 539                         xsegbd_dev->dst_portno, X_ALLOC);
 540         if (!xreq)
 541                 goto out;
 542
 543         BUG_ON(xseg_prep_request(xsegbd_dev->xseg, xreq, xsegbd_dev->targetlen,
 544                                 sizeof(struct xseg_reply_info)));
 545
 546         init_completion(&comp);
 547         blkreq_idx = xq_pop_head(&xsegbd_dev->blk_queue_pending, 1);
 548         if (blkreq_idx == Noneidx)
 549                 goto out_put;
 550
 551         pending = &xsegbd_dev->blk_req_pending[blkreq_idx];
 552         pending->dev = xsegbd_dev;
 553         pending->request = NULL;
 554         pending->comp = &comp;
 555
 556
 557         xreq->priv = (uint64_t) blkreq_idx;
 558
 559         target = xseg_get_target(xsegbd_dev->xseg, xreq);
 560         strncpy(target, xsegbd_dev->target, xsegbd_dev->targetlen);
 561         xreq->size = xreq->datalen;
 562         xreq->offset = 0;
 563         xreq->op = X_INFO;
 564
 565         xseg_prepare_wait(xsegbd_dev->xseg, xsegbd_dev->src_portno);
 566         p = xseg_submit(xsegbd_dev->xseg, xreq,
 567                                 xsegbd_dev->src_portno, X_ALLOC);
 568         if ( p == NoPort) {
 569                 XSEGLOG("couldn't submit request");
 570                 BUG_ON(1);
 571                 goto out_queue;
 572         }
 573         WARN_ON(xseg_signal(xsegbd_dev->xseg, p) < 0);
 574         XSEGLOG("Before wait for completion, comp %lx [%llu]", (unsigned long) pending->comp, (unsigned long long) blkreq_idx);
 575         wait_for_completion_interruptible(&comp);
 576         XSEGLOG("Woken up after wait_for_completion_interruptible(), comp: %lx [%llu]", (unsigned long) pending->comp, (unsigned long long) blkreq_idx);
 577         ret = update_dev_sectors_from_request(xsegbd_dev, xreq);
 578         //XSEGLOG("get_size: sectors = %ld\n", (long)xsegbd_dev->sectors);
 579 out_put:
 580         BUG_ON(xseg_put_request(xsegbd_dev->xseg, xreq, xsegbd_dev->src_portno) == -1);
 581 out:
 582         __xsegbd_put(xsegbd_dev);
 583         return ret;
 584
 585 out_queue:
 586         pending->dev = NULL;
 587         pending->comp = NULL;
 588         xq_append_head(&xsegbd_dev->blk_queue_pending, blkreq_idx, 1);
 589
 590         goto out;
 591 }
 592
 593 static int xsegbd_mapclose(struct xsegbd_device *xsegbd_dev)
 594 {
 595         struct xseg_request *xreq;
 596         char *target;
 597         uint64_t datalen;
 598         xqindex blkreq_idx;
 599         struct xsegbd_pending *pending;
 600         struct completion comp;
 601         xport p;
 602         void *data;
 603         int ret = -EBUSY, r;
 604
 605         __xsegbd_get(xsegbd_dev);
 606         xreq = xseg_get_request(xsegbd_dev->xseg, xsegbd_dev->src_portno,
 607                         xsegbd_dev->dst_portno, X_ALLOC);
 608         if (!xreq)
 609                 goto out;
 610
 611         BUG_ON(xseg_prep_request(xsegbd_dev->xseg, xreq, xsegbd_dev->targetlen, 0));
 612
 613         init_completion(&comp);
 614         blkreq_idx = xq_pop_head(&xsegbd_dev->blk_queue_pending, 1);
 615         if (blkreq_idx == Noneidx)
 616                 goto out_put;
 617
 618         pending = &xsegbd_dev->blk_req_pending[blkreq_idx];
 619         pending->dev = xsegbd_dev;
 620         pending->request = NULL;
 621         pending->comp = &comp;
 622
 623
 624         xreq->priv = (uint64_t) blkreq_idx;
 625
 626         target = xseg_get_target(xsegbd_dev->xseg, xreq);
 627         strncpy(target, xsegbd_dev->target, xsegbd_dev->targetlen);
 628         xreq->size = xreq->datalen;
 629         xreq->offset = 0;
 630         xreq->op = X_CLOSE;
 631
 632         xseg_prepare_wait(xsegbd_dev->xseg, xsegbd_dev->src_portno);
 633         p = xseg_submit(xsegbd_dev->xseg, xreq,
 634                                 xsegbd_dev->src_portno, X_ALLOC);
 635         if ( p == NoPort) {
 636                 XSEGLOG("couldn't submit request");
 637                 BUG_ON(1);
 638                 goto out_queue;
 639         }
 640         WARN_ON(xseg_signal(xsegbd_dev->xseg, p) < 0);
 641         wait_for_completion_interruptible(&comp);
 642         ret = 0;
 643         if (xreq->state & XS_FAILED)
 644                 XSEGLOG("Couldn't close disk on mapper");
 645 out_put:
 646         BUG_ON(xseg_put_request(xsegbd_dev->xseg, xreq, xsegbd_dev->src_portno) == -1);
 647 out:
 648         __xsegbd_put(xsegbd_dev);
 649         return ret;
 650
 651 out_queue:
 652         pending->dev = NULL;
 653         pending->comp = NULL;
 654         xq_append_head(&xsegbd_dev->blk_queue_pending, blkreq_idx, 1);
 655
 656         goto out;
 657 }
 658
 659 static void xseg_callback(xport portno)
 660 {
 661         struct xsegbd_device *xsegbd_dev;
 662         struct xseg_request *xreq;
 663         struct request *blkreq;
 664         struct xsegbd_pending *pending;
 665         unsigned long flags;
 666         xqindex blkreq_idx, ridx;
 667         int err;
 668         void *data;
 669
 670         xsegbd_dev  = __xsegbd_get_dev(portno);
 671         if (!xsegbd_dev) {
 672                 XSEGLOG("portno: %u has no xsegbd device assigned", portno);
 673                 WARN_ON(1);
 674                 return;
 675         }
 676
 677         for (;;) {
 678                 xseg_prepare_wait(xsegbd_dev->xseg, xsegbd_dev->src_portno);
 679                 xreq = xseg_receive(xsegbd_dev->xseg, portno, 0);
 680                 if (!xreq)
 681                         break;
 682
 683                 xseg_cancel_wait(xsegbd_dev->xseg, xsegbd_dev->src_portno);
 684
 685                 blkreq_idx = (xqindex) xreq->priv;
 686                 if (blkreq_idx >= xsegbd_dev->nr_requests) {
 687                         WARN_ON(1);
 688                         //FIXME maybe put request?
 689                         continue;
 690                 }
 691
 692                 pending = &xsegbd_dev->blk_req_pending[blkreq_idx];
 693                 if (pending->comp) {
 694                         /* someone is blocking on this request
 695                            and will handle it when we wake them up. */
 696                         complete(pending->comp);
 697                         /* the request is blocker's responsibility so
 698                            we will not put_request(); */
 699
 700                         continue;
 701                 }
 702
 703                 /* this is now treated as a block I/O request to end */
 704                 blkreq = pending->request;
 705                 pending->request = NULL;
 706                 if (xsegbd_dev != pending->dev) {
 707                         //FIXME maybe put request?
 708                         XSEGLOG("xsegbd_dev != pending->dev");
 709                         BUG_ON(1);
 710                         continue;
 711                 }
 712                 pending->dev = NULL;
 713                 if (!blkreq){
 714                         //FIXME maybe put request?
 715                         XSEGLOG("blkreq does not exist");
 716                         BUG_ON(1);
 717                         continue;
 718                 }
 719
 720                 err = -EIO;
 721                 if (!(xreq->state & XS_SERVED))
 722                         goto blk_end;
 723
 724                 if (xreq->serviced != blk_rq_bytes(blkreq))
 725                         goto blk_end;
 726
 727                 err = 0;
 728                 if (!rq_data_dir(blkreq)){
 729                         xseg_to_blk(xsegbd_dev->xseg, xreq, blkreq);
 730                 }
 731 blk_end:
 732                 blk_end_request_all(blkreq, err);
 733
 734                 ridx = xq_append_head(&xsegbd_dev->blk_queue_pending,
 735                                         blkreq_idx, xsegbd_dev->src_portno);
 736                 if (ridx == Noneidx) {
 737                         XSEGLOG("couldnt append blkreq_idx");
 738                         WARN_ON(1);
 739                 }
 740
 741                 if (xseg_put_request(xsegbd_dev->xseg, xreq,
 742                                                 xsegbd_dev->src_portno) < 0){
 743                         XSEGLOG("couldn't put req");
 744                         BUG_ON(1);
 745                 }
 746                 __xsegbd_put(xsegbd_dev);
 747         }
 748         if (xsegbd_dev) {
 749                 spin_lock_irqsave(&xsegbd_dev->rqlock, flags);
 750                 xseg_request_fn(xsegbd_dev->blk_queue);
 751                 spin_unlock_irqrestore(&xsegbd_dev->rqlock, flags);
 752                 __xsegbd_put(xsegbd_dev);
 753         }
 754 }
 755
 756
 757 /* sysfs interface */
 758
 759 static struct bus_type xsegbd_bus_type = {
 760         .name   = "xsegbd",
 761 };
 762
 763 static ssize_t xsegbd_size_show(struct device *dev,
 764                                         struct device_attribute *attr, char *buf)
 765 {
 766         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 767
 768         return sprintf(buf, "%llu\n", (unsigned long long) xsegbd_dev->sectors * 512ULL);
 769 }
 770
 771 static ssize_t xsegbd_major_show(struct device *dev,
 772                                         struct device_attribute *attr, char *buf)
 773 {
 774         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 775
 776         return sprintf(buf, "%d\n", xsegbd_dev->major);
 777 }
 778
 779 static ssize_t xsegbd_srcport_show(struct device *dev,
 780                                         struct device_attribute *attr, char *buf)
 781 {
 782         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 783
 784         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->src_portno);
 785 }
 786
 787 static ssize_t xsegbd_dstport_show(struct device *dev,
 788                                         struct device_attribute *attr, char *buf)
 789 {
 790         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 791
 792         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->dst_portno);
 793 }
 794
 795 static ssize_t xsegbd_id_show(struct device *dev,
 796                                         struct device_attribute *attr, char *buf)
 797 {
 798         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 799
 800         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->id);
 801 }
 802
 803 static ssize_t xsegbd_reqs_show(struct device *dev,
 804                                         struct device_attribute *attr, char *buf)
 805 {
 806         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 807
 808         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->nr_requests);
 809 }
 810
 811 static ssize_t xsegbd_target_show(struct device *dev,
 812                                         struct device_attribute *attr, char *buf)
 813 {
 814         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 815
 816         return sprintf(buf, "%s\n", xsegbd_dev->target);
 817 }
 818
 819 static ssize_t xsegbd_image_refresh(struct device *dev,
 820                                         struct device_attribute *attr,
 821                                         const char *buf,
 822                                         size_t size)
 823 {
 824         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 825         int rc, ret = size;
 826
 827         mutex_lock_nested(&xsegbd_mutex, SINGLE_DEPTH_NESTING);
 828
 829         rc = xsegbd_get_size(xsegbd_dev);
 830         if (rc < 0) {
 831                 ret = rc;
 832                 goto out;
 833         }
 834
 835         set_capacity(xsegbd_dev->gd, xsegbd_dev->sectors);
 836
 837 out:
 838         mutex_unlock(&xsegbd_mutex);
 839         return ret;
 840 }
 841
 842 static ssize_t xsegbd_cleanup(struct device *dev,
 843                                         struct device_attribute *attr,
 844                                         const char *buf,
 845                                         size_t size)
 846 {
 847         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 848         int ret = size, i;
 849         struct request *blkreq = NULL;
 850         struct xsegbd_pending *pending = NULL;
 851         struct completion *comp = NULL;
 852
 853         mutex_lock_nested(&xsegbd_mutex, SINGLE_DEPTH_NESTING);
 854         xlock_acquire(&xsegbd_dev->blk_queue_pending.lock,
 855                                 xsegbd_dev->src_portno);
 856         for (i = 0; i < xsegbd_dev->nr_requests; i++) {
 857                 if (!__xq_check(&xsegbd_dev->blk_queue_pending, i)) {
 858                         pending = &xsegbd_dev->blk_req_pending[i];
 859                         blkreq = pending->request;
 860                         pending->request = NULL;
 861                         comp = pending->comp;
 862                         pending->comp = NULL;
 863                         if (blkreq){
 864                                 XSEGLOG("Cleaning up blkreq %lx [%d]", (unsigned long) blkreq, i);
 865                                 blk_end_request_all(blkreq, -EIO);
 866                         }
 867                         if (comp){
 868                                 XSEGLOG("Cleaning up comp %lx [%d]", (unsigned long) comp, i);
 869                                 complete(comp);
 870                         }
 871                         __xq_append_tail(&xsegbd_dev->blk_queue_pending, i);
 872                 }
 873         }
 874         xlock_release(&xsegbd_dev->blk_queue_pending.lock);
 875
 876         mutex_unlock(&xsegbd_mutex);
 877         return ret;
 878 }
 879
 880 static DEVICE_ATTR(size, S_IRUGO, xsegbd_size_show, NULL);
 881 static DEVICE_ATTR(major, S_IRUGO, xsegbd_major_show, NULL);
 882 static DEVICE_ATTR(srcport, S_IRUGO, xsegbd_srcport_show, NULL);
 883 static DEVICE_ATTR(dstport, S_IRUGO, xsegbd_dstport_show, NULL);
 884 static DEVICE_ATTR(id , S_IRUGO, xsegbd_id_show, NULL);
 885 static DEVICE_ATTR(reqs , S_IRUGO, xsegbd_reqs_show, NULL);
 886 static DEVICE_ATTR(target, S_IRUGO, xsegbd_target_show, NULL);
 887 static DEVICE_ATTR(refresh , S_IWUSR, NULL, xsegbd_image_refresh);
 888 static DEVICE_ATTR(cleanup , S_IWUSR, NULL, xsegbd_cleanup);
 889
 890 static struct attribute *xsegbd_attrs[] = {
 891         &dev_attr_size.attr,
 892         &dev_attr_major.attr,
 893         &dev_attr_srcport.attr,
 894         &dev_attr_dstport.attr,
 895         &dev_attr_id.attr,
 896         &dev_attr_reqs.attr,
 897         &dev_attr_target.attr,
 898         &dev_attr_refresh.attr,
 899         &dev_attr_cleanup.attr,
 900         NULL
 901 };
 902
 903 static struct attribute_group xsegbd_attr_group = {
 904         .attrs = xsegbd_attrs,
 905 };
 906
 907 static const struct attribute_group *xsegbd_attr_groups[] = {
 908         &xsegbd_attr_group,
 909         NULL
 910 };
 911
 912 static void xsegbd_sysfs_dev_release(struct device *dev)
 913 {
 914 }
 915
 916 static struct device_type xsegbd_device_type = {
 917         .name           = "xsegbd",
 918         .groups         = xsegbd_attr_groups,
 919         .release        = xsegbd_sysfs_dev_release,
 920 };
 921
 922 static void xsegbd_root_dev_release(struct device *dev)
 923 {
 924 }
 925
 926 static struct device xsegbd_root_dev = {
 927         .init_name      = "xsegbd",
 928         .release        = xsegbd_root_dev_release,
 929 };
 930
 931 static int xsegbd_bus_add_dev(struct xsegbd_device *xsegbd_dev)
 932 {
 933         int ret = -ENOMEM;
 934         struct device *dev;
 935
 936         mutex_lock_nested(&xsegbd_mutex, SINGLE_DEPTH_NESTING);
 937         dev = &xsegbd_dev->dev;
 938
 939         dev->bus = &xsegbd_bus_type;
 940         dev->type = &xsegbd_device_type;
 941         dev->parent = &xsegbd_root_dev;
 942         dev->release = xsegbd_dev_release;
 943         dev_set_name(dev, "%d", xsegbd_dev->id);
 944
 945         ret = device_register(dev);
 946
 947         mutex_unlock(&xsegbd_mutex);
 948         return ret;
 949 }
 950
 951 static void xsegbd_bus_del_dev(struct xsegbd_device *xsegbd_dev)
 952 {
 953         device_unregister(&xsegbd_dev->dev);
 954 }
 955
 956 static ssize_t xsegbd_add(struct bus_type *bus, const char *buf, size_t count)
 957 {
 958         struct xsegbd_device *xsegbd_dev;
 959         struct xseg_port *port;
 960         ssize_t ret = -ENOMEM;
 961
 962         if (!try_module_get(THIS_MODULE))
 963                 return -ENODEV;
 964
 965         xsegbd_dev = kzalloc(sizeof(*xsegbd_dev), GFP_KERNEL);
 966         if (!xsegbd_dev)
 967                 goto out;
 968
 969         spin_lock_init(&xsegbd_dev->rqlock);
 970         INIT_LIST_HEAD(&xsegbd_dev->node);
 971         init_waitqueue_head(&xsegbd_dev->wq);
 972         atomic_set(&xsegbd_dev->usercount, 0);
 973
 974         /* parse cmd */
 975         if (sscanf(buf, "%" __stringify(XSEGBD_TARGET_NAMELEN) "s "
 976                         "%d:%d:%d", xsegbd_dev->target, &xsegbd_dev->src_portno,
 977                         &xsegbd_dev->dst_portno, &xsegbd_dev->nr_requests) < 3) {
 978                 ret = -EINVAL;
 979                 goto out_dev;
 980         }
 981         xsegbd_dev->targetlen = strlen(xsegbd_dev->target);
 982
 983         spin_lock(&xsegbd_devices_lock);
 984         if (xsegbd_devices[xsegbd_dev->src_portno] != NULL) {
 985                 ret = -EINVAL;
 986                 goto out_unlock;
 987         }
 988         xsegbd_devices[xsegbd_dev->src_portno] = xsegbd_dev;
 989         xsegbd_dev->id = xsegbd_dev->src_portno;
 990         spin_unlock(&xsegbd_devices_lock);
 991
 992         XSEGLOG("registering block device major %d", major);
 993         ret = register_blkdev(major, XSEGBD_NAME);
 994         if (ret < 0) {
 995                 XSEGLOG("cannot register block device!");
 996                 ret = -EBUSY;
 997                 goto out_delentry;
 998         }
 999         xsegbd_dev->major = ret;
1000         XSEGLOG("registered block device major %d", xsegbd_dev->major);
1001
1002         ret = xsegbd_bus_add_dev(xsegbd_dev);
1003         if (ret)
1004                 goto out_blkdev;
1005
1006         if (!xq_alloc_seq(&xsegbd_dev->blk_queue_pending,
1007                                 xsegbd_dev->nr_requests,
1008                                 xsegbd_dev->nr_requests))
1009                 goto out_bus;
1010
1011         xsegbd_dev->blk_req_pending = kzalloc(
1012                         xsegbd_dev->nr_requests *sizeof(struct xsegbd_pending),
1013                                    GFP_KERNEL);
1014         if (!xsegbd_dev->blk_req_pending)
1015                 goto out_freeq;
1016
1017
1018         XSEGLOG("joining segment");
1019         //FIXME use xsebd module config for now
1020         xsegbd_dev->xseg = xseg_join(   xsegbd.config.type,
1021                                         xsegbd.config.name,
1022                                         "segdev",
1023                                         xseg_callback           );
1024         if (!xsegbd_dev->xseg)
1025                 goto out_freepending;
1026
1027
1028         XSEGLOG("%s binding to source port %u (destination %u)", xsegbd_dev->target,
1029                         xsegbd_dev->src_portno, xsegbd_dev->dst_portno);
1030         port = xseg_bind_port(xsegbd_dev->xseg, xsegbd_dev->src_portno, NULL);
1031         if (!port) {
1032                 XSEGLOG("cannot bind to port");
1033                 ret = -EFAULT;
1034
1035                 goto out_xseg;
1036         }
1037
1038         if (xsegbd_dev->src_portno != xseg_portno(xsegbd_dev->xseg, port)) {
1039                 XSEGLOG("portno != xsegbd_dev->src_portno");
1040                 BUG_ON(1);
1041                 ret = -EFAULT;
1042                 goto out_xseg;
1043         }
1044
1045
1046         /* make sure we don't get any requests until we're ready to handle them */
1047         xseg_cancel_wait(xsegbd_dev->xseg, xseg_portno(xsegbd_dev->xseg, port));
1048
1049         ret = xsegbd_dev_init(xsegbd_dev);
1050         if (ret)
1051                 goto out_xseg;
1052
1053         xseg_prepare_wait(xsegbd_dev->xseg, xseg_portno(xsegbd_dev->xseg, port));
1054         return count;
1055
1056 out_xseg:
1057         xseg_leave(xsegbd_dev->xseg);
1058
1059 out_freepending:
1060         kfree(xsegbd_dev->blk_req_pending);
1061
1062 out_freeq:
1063         xq_free(&xsegbd_dev->blk_queue_pending);
1064
1065 out_bus:
1066         xsegbd_bus_del_dev(xsegbd_dev);
1067         return ret;
1068
1069 out_blkdev:
1070         unregister_blkdev(xsegbd_dev->major, XSEGBD_NAME);
1071
1072 out_delentry:
1073         spin_lock(&xsegbd_devices_lock);
1074         xsegbd_devices[xsegbd_dev->src_portno] = NULL;
1075
1076 out_unlock:
1077         spin_unlock(&xsegbd_devices_lock);
1078
1079 out_dev:
1080         kfree(xsegbd_dev);
1081
1082 out:
1083         return ret;
1084 }
1085
1086 static ssize_t xsegbd_remove(struct bus_type *bus, const char *buf, size_t count)
1087 {
1088         struct xsegbd_device *xsegbd_dev = NULL;
1089         int id, ret;
1090         unsigned long ul_id;
1091
1092         ret = strict_strtoul(buf, 10, &ul_id);
1093         if (ret)
1094                 return ret;
1095
1096         id = (int) ul_id;
1097         if (id != ul_id)
1098                 return -EINVAL;
1099
1100         mutex_lock_nested(&xsegbd_mutex, SINGLE_DEPTH_NESTING);
1101
1102         ret = count;
1103         xsegbd_dev = __xsegbd_get_dev(id);
1104         if (!xsegbd_dev) {
1105                 ret = -ENOENT;
1106                 goto out_unlock;
1107         }
1108         __xsegbd_put(xsegbd_dev);
1109         xsegbd_bus_del_dev(xsegbd_dev);
1110
1111 out_unlock:
1112         mutex_unlock(&xsegbd_mutex);
1113         return ret;
1114 }
1115
1116 static struct bus_attribute xsegbd_bus_attrs[] = {
1117         __ATTR(add, S_IWUSR, NULL, xsegbd_add),
1118         __ATTR(remove, S_IWUSR, NULL, xsegbd_remove),
1119         __ATTR_NULL
1120 };
1121
1122 static int xsegbd_sysfs_init(void)
1123 {
1124         int ret;
1125
1126         ret = device_register(&xsegbd_root_dev);
1127         if (ret < 0)
1128                 return ret;
1129
1130         xsegbd_bus_type.bus_attrs = xsegbd_bus_attrs;
1131         ret = bus_register(&xsegbd_bus_type);
1132         if (ret < 0)
1133                 device_unregister(&xsegbd_root_dev);
1134
1135         return ret;
1136 }
1137
1138 static void xsegbd_sysfs_cleanup(void)
1139 {
1140         bus_unregister(&xsegbd_bus_type);
1141         device_unregister(&xsegbd_root_dev);
1142 }
1143
1144 /* *************************** */
1145 /* ** Module Initialization ** */
1146 /* *************************** */
1147
1148 static int __init xsegbd_init(void)
1149 {
1150         int ret = -ENOMEM;
1151         xsegbd_devices = kzalloc(max_dev * sizeof(struct xsegbd_devices *), GFP_KERNEL);
1152         if (!xsegbd_devices)
1153                 goto out;
1154
1155         spin_lock_init(&xsegbd_devices_lock);
1156
1157         ret = -ENOSYS;
1158         ret = xsegbd_xseg_init();
1159         if (ret)
1160                 goto out_free;
1161
1162         ret = xsegbd_sysfs_init();
1163         if (ret)
1164                 goto out_xseg;
1165
1166         XSEGLOG("initialization complete");
1167
1168 out:
1169         return ret;
1170
1171 out_xseg:
1172         xsegbd_xseg_quit();
1173
1174 out_free:
1175         kfree(xsegbd_devices);
1176
1177         goto out;
1178 }
1179
1180 static void __exit xsegbd_exit(void)
1181 {
1182         xsegbd_sysfs_cleanup();
1183         xsegbd_xseg_quit();
1184 }
1185
1186 module_init(xsegbd_init);
1187 module_exit(xsegbd_exit);
1188