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