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