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                         XSEGLOG("non-fs cmd_type: %u. *shrug*", blkreq->cmd_type);
 366                         __blk_end_request_all(blkreq, 0);
 367                 }
 368
 369
 370                 datalen = blk_rq_bytes(blkreq);
 371                 BUG_ON(xseg_prep_request(xsegbd_dev->xseg, xreq,
 372                                         xsegbd_dev->targetlen, datalen));
 373                 BUG_ON(xreq->bufferlen - xsegbd_dev->targetlen < datalen);
 374
 375                 target = xseg_get_target(xsegbd_dev->xseg, xreq);
 376                 strncpy(target, xsegbd_dev->target, xsegbd_dev->targetlen);
 377                 BUG_ON(blkreq_idx == Noneidx);
 378                 pending = &xsegbd_dev->blk_req_pending[blkreq_idx];
 379                 pending->dev = xsegbd_dev;
 380                 pending->request = blkreq;
 381                 pending->comp = NULL;
 382                 xreq->priv = (uint64_t)blkreq_idx;
 383                 xreq->size = datalen;
 384                 xreq->offset = blk_rq_pos(blkreq) << 9;
 385                 /*
 386                 if (xreq->offset >= (sector_size << 9))
 387                         XSEGLOG("sector offset: %lu > %lu, flush:%u, fua:%u",
 388                                  blk_rq_pos(blkreq), sector_size,
 389                                  blkreq->cmd_flags & REQ_FLUSH,
 390                                  blkreq->cmd_flags & REQ_FUA);
 391                 */
 392
 393                 if (blkreq->cmd_flags & REQ_FLUSH)
 394                         xreq->flags |= XF_FLUSH;
 395
 396                 if (blkreq->cmd_flags & REQ_FUA)
 397                         xreq->flags |= XF_FUA;
 398
 399                 if (rq_data_dir(blkreq)) {
 400                         /* unlock for data transfers? */
 401                         blk_to_xseg(xsegbd_dev->xseg, xreq, blkreq);
 402                         xreq->op = X_WRITE;
 403                 } else {
 404                         xreq->op = X_READ;
 405                 }
 406
 407                 //maybe put this in loop start, and on break,
 408                 //just do xseg_get_req_data
 409                 spin_lock(&xsegbd_dev->reqdatalock);
 410                 r = xseg_set_req_data(xsegbd_dev->xseg, xreq, (void *) blkreq_idx);
 411                 spin_unlock(&xsegbd_dev->reqdatalock);
 412                 BUG_ON(r < 0);
 413
 414                 BUG_ON((p = xseg_submit(xsegbd_dev->xseg, xreq,
 415                                         xsegbd_dev->src_portno, X_ALLOC)) == NoPort);
 416                 WARN_ON(xseg_signal(xsegbd_dev->xsegbd->xseg, p) < 0);
 417         }
 418         if (xreq)
 419                 BUG_ON(xseg_put_request(xsegbd_dev->xsegbd->xseg, xreq,
 420                                         xsegbd_dev->src_portno) == -1);
 421         if (blkreq_idx != Noneidx)
 422                 BUG_ON(xq_append_head(&xsegbd_dev->blk_queue_pending,
 423                                         blkreq_idx, xsegbd_dev->src_portno) == Noneidx);
 424 }
 425
 426 int update_dev_sectors_from_request(    struct xsegbd_device *xsegbd_dev,
 427                                         struct xseg_request *xreq       )
 428 {
 429         void *data;
 430
 431         if (xreq->state & XS_FAILED)
 432                 return -ENOENT;
 433
 434         if (!(xreq->state & XS_SERVED))
 435                 return -EIO;
 436
 437         data = xseg_get_data(xsegbd_dev->xseg, xreq);
 438         xsegbd_dev->sectors = *((uint64_t *) data) / 512ULL;
 439         return 0;
 440 }
 441
 442 static int xsegbd_get_size(struct xsegbd_device *xsegbd_dev)
 443 {
 444         struct xseg_request *xreq;
 445         char *target;
 446         uint64_t datalen;
 447         xqindex blkreq_idx;
 448         struct xsegbd_pending *pending;
 449         struct completion comp;
 450         xport p;
 451         void *data;
 452         int ret = -EBUSY, r;
 453         xreq = xseg_get_request(xsegbd_dev->xseg, xsegbd_dev->src_portno,
 454                         xsegbd_dev->dst_portno, X_ALLOC);
 455         if (!xreq)
 456                 goto out;
 457
 458         datalen = sizeof(uint64_t);
 459         BUG_ON(xseg_prep_request(xsegbd_dev->xseg, xreq, xsegbd_dev->targetlen, datalen));
 460         BUG_ON(xreq->bufferlen - xsegbd_dev->targetlen < datalen);
 461
 462         init_completion(&comp);
 463         blkreq_idx = xq_pop_head(&xsegbd_dev->blk_queue_pending, 1);
 464         if (blkreq_idx == Noneidx)
 465                 goto out;
 466
 467         pending = &xsegbd_dev->blk_req_pending[blkreq_idx];
 468         pending->dev = xsegbd_dev;
 469         pending->request = NULL;
 470         pending->comp = &comp;
 471
 472
 473         spin_lock(&xsegbd_dev->reqdatalock);
 474         r = xseg_set_req_data(xsegbd_dev->xseg, xreq, (void *) blkreq_idx);
 475         spin_unlock(&xsegbd_dev->reqdatalock);
 476         if (r < 0)
 477                 goto out_queue;
 478         XSEGLOG("for req: %lx, set data %llu (lx: %lx)", xreq, blkreq_idx, (void *) blkreq_idx);
 479
 480         target = xseg_get_target(xsegbd_dev->xseg, xreq);
 481         strncpy(target, xsegbd_dev->target, xsegbd_dev->targetlen);
 482         xreq->size = datalen;
 483         xreq->offset = 0;
 484
 485         xreq->op = X_INFO;
 486
 487         /* waiting is not needed.
 488          * but it should be better to use xseg_prepare_wait
 489          * and the xseg_segdev kernel driver, would be a no op
 490          */
 491
 492         xseg_prepare_wait(xsegbd_dev->xseg, xsegbd_dev->src_portno);
 493         p = xseg_submit(xsegbd_dev->xseg, xreq,
 494                                 xsegbd_dev->src_portno, X_ALLOC);
 495         BUG_ON(p == NoPort);
 496         if ( p == NoPort) {
 497                 goto out_data;
 498         }
 499         WARN_ON(xseg_signal(xsegbd_dev->xseg, p) < 0);
 500
 501         wait_for_completion_interruptible(&comp);
 502         XSEGLOG("Woken up after wait_for_completion_interruptible()\n");
 503         ret = update_dev_sectors_from_request(xsegbd_dev, xreq);
 504         XSEGLOG("get_size: sectors = %ld\n", (long)xsegbd_dev->sectors);
 505 out:
 506         BUG_ON(xseg_put_request(xsegbd_dev->xseg, xreq, xsegbd_dev->src_portno) < 0);
 507         return ret;
 508
 509 out_data:
 510         spin_lock(&xsegbd_dev->reqdatalock);
 511         r = xseg_get_req_data(xsegbd_dev->xseg, xreq, &data);
 512         spin_unlock(&xsegbd_dev->reqdatalock);
 513 out_queue:
 514         xq_append_head(&xsegbd_dev->blk_queue_pending, blkreq_idx, 1);
 515
 516         goto out;
 517 }
 518
 519 static void xseg_callback(struct xseg *xseg, xport portno)
 520 {
 521         struct xsegbd_device *xsegbd_dev;
 522         struct xseg_request *xreq;
 523         struct request *blkreq;
 524         struct xsegbd_pending *pending;
 525         unsigned long flags;
 526         xqindex blkreq_idx;
 527         int err;
 528         void *data;
 529
 530         xsegbd_dev  = __xsegbd_get_dev(portno);
 531         if (!xsegbd_dev) {
 532                 WARN_ON(3);
 533                 return;
 534         }
 535
 536         for (;;) {
 537                 xreq = xseg_receive(xsegbd_dev->xseg, portno);
 538                 if (!xreq)
 539                         break;
 540
 541                 spin_lock(&xsegbd_dev->reqdatalock);
 542                 err = xseg_get_req_data(xsegbd_dev->xseg, xreq, &data);
 543                 spin_unlock(&xsegbd_dev->reqdatalock);
 544                 XSEGLOG("for req: %lx, got data %llu (lx %lx)", xreq, (xqindex) data, data);
 545                 if (err < 0) {
 546                         WARN_ON(2);
 547                         //maybe put request?
 548                         continue;
 549                 }
 550
 551                 blkreq_idx = (xqindex) data;
 552                 if (blkreq_idx >= xsegbd_dev->nr_requests) {
 553                         WARN_ON(1);
 554                         //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                 //xsegbd_dev = pending->dev;
 572                 BUG_ON(xsegbd_dev != pending->dev);
 573                 pending->dev = NULL;
 574                 WARN_ON(!blkreq);
 575
 576                 if (!(xreq->state & XS_SERVED))
 577                         goto blk_end;
 578
 579                 if (xreq->serviced != blk_rq_bytes(blkreq))
 580                         goto blk_end;
 581
 582                 /* unlock for data transfer? */
 583                 if (!rq_data_dir(blkreq))
 584                         xseg_to_blk(xsegbd_dev->xseg, xreq, blkreq);
 585
 586                 err = 0;
 587 blk_end:
 588                 blk_end_request_all(blkreq, err);
 589                 xq_append_head(&xsegbd_dev->blk_queue_pending, blkreq_idx, 1);
 590                 BUG_ON(xseg_put_request(xsegbd_dev->xseg, xreq, xsegbd_dev->src_portno) < 0);
 591         }
 592
 593         if (xsegbd_dev) {
 594                 spin_lock_irqsave(&xsegbd_dev->rqlock, flags);
 595                 xseg_request_fn(xsegbd_dev->blk_queue);
 596                 spin_unlock_irqrestore(&xsegbd_dev->rqlock, flags);
 597         }
 598 }
 599
 600
 601 /* sysfs interface */
 602
 603 static struct bus_type xsegbd_bus_type = {
 604         .name   = "xsegbd",
 605 };
 606
 607 static ssize_t xsegbd_size_show(struct device *dev,
 608                                         struct device_attribute *attr, char *buf)
 609 {
 610         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 611
 612         return sprintf(buf, "%llu\n", (unsigned long long) xsegbd_dev->sectors * 512ULL);
 613 }
 614
 615 static ssize_t xsegbd_major_show(struct device *dev,
 616                                         struct device_attribute *attr, char *buf)
 617 {
 618         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 619
 620         return sprintf(buf, "%d\n", xsegbd_dev->major);
 621 }
 622
 623 static ssize_t xsegbd_srcport_show(struct device *dev,
 624                                         struct device_attribute *attr, char *buf)
 625 {
 626         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 627
 628         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->src_portno);
 629 }
 630
 631 static ssize_t xsegbd_dstport_show(struct device *dev,
 632                                         struct device_attribute *attr, char *buf)
 633 {
 634         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 635
 636         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->dst_portno);
 637 }
 638
 639 static ssize_t xsegbd_id_show(struct device *dev,
 640                                         struct device_attribute *attr, char *buf)
 641 {
 642         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 643
 644         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->id);
 645 }
 646
 647 static ssize_t xsegbd_reqs_show(struct device *dev,
 648                                         struct device_attribute *attr, char *buf)
 649 {
 650         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 651
 652         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->nr_requests);
 653 }
 654
 655 static ssize_t xsegbd_target_show(struct device *dev,
 656                                         struct device_attribute *attr, char *buf)
 657 {
 658         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 659
 660         return sprintf(buf, "%s\n", xsegbd_dev->target);
 661 }
 662
 663 static ssize_t xsegbd_image_refresh(struct device *dev,
 664                                         struct device_attribute *attr,
 665                                         const char *buf,
 666                                         size_t size)
 667 {
 668         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 669         int rc, ret = size;
 670
 671         mutex_lock_nested(&xsegbd_mutex, SINGLE_DEPTH_NESTING);
 672
 673         rc = xsegbd_get_size(xsegbd_dev);
 674         if (rc < 0) {
 675                 ret = rc;
 676                 goto out;
 677         }
 678
 679         set_capacity(xsegbd_dev->gd, xsegbd_dev->sectors);
 680
 681 out:
 682         mutex_unlock(&xsegbd_mutex);
 683         return ret;
 684 }
 685
 686 static DEVICE_ATTR(size, S_IRUGO, xsegbd_size_show, NULL);
 687 static DEVICE_ATTR(major, S_IRUGO, xsegbd_major_show, NULL);
 688 static DEVICE_ATTR(srcport, S_IRUGO, xsegbd_srcport_show, NULL);
 689 static DEVICE_ATTR(dstport, S_IRUGO, xsegbd_dstport_show, NULL);
 690 static DEVICE_ATTR(id , S_IRUGO, xsegbd_id_show, NULL);
 691 static DEVICE_ATTR(reqs , S_IRUGO, xsegbd_reqs_show, NULL);
 692 static DEVICE_ATTR(target, S_IRUGO, xsegbd_target_show, NULL);
 693 static DEVICE_ATTR(refresh , S_IWUSR, NULL, xsegbd_image_refresh);
 694
 695 static struct attribute *xsegbd_attrs[] = {
 696         &dev_attr_size.attr,
 697         &dev_attr_major.attr,
 698         &dev_attr_srcport.attr,
 699         &dev_attr_dstport.attr,
 700         &dev_attr_id.attr,
 701         &dev_attr_reqs.attr,
 702         &dev_attr_target.attr,
 703         &dev_attr_refresh.attr,
 704         NULL
 705 };
 706
 707 static struct attribute_group xsegbd_attr_group = {
 708         .attrs = xsegbd_attrs,
 709 };
 710
 711 static const struct attribute_group *xsegbd_attr_groups[] = {
 712         &xsegbd_attr_group,
 713         NULL
 714 };
 715
 716 static void xsegbd_sysfs_dev_release(struct device *dev)
 717 {
 718 }
 719
 720 static struct device_type xsegbd_device_type = {
 721         .name           = "xsegbd",
 722         .groups         = xsegbd_attr_groups,
 723         .release        = xsegbd_sysfs_dev_release,
 724 };
 725
 726 static void xsegbd_root_dev_release(struct device *dev)
 727 {
 728 }
 729
 730 static struct device xsegbd_root_dev = {
 731         .init_name      = "xsegbd",
 732         .release        = xsegbd_root_dev_release,
 733 };
 734
 735 static int xsegbd_bus_add_dev(struct xsegbd_device *xsegbd_dev)
 736 {
 737         int ret = -ENOMEM;
 738         struct device *dev;
 739
 740         mutex_lock_nested(&xsegbd_mutex, SINGLE_DEPTH_NESTING);
 741         dev = &xsegbd_dev->dev;
 742
 743         dev->bus = &xsegbd_bus_type;
 744         dev->type = &xsegbd_device_type;
 745         dev->parent = &xsegbd_root_dev;
 746         dev->release = xsegbd_dev_release;
 747         dev_set_name(dev, "%d", xsegbd_dev->id);
 748
 749         ret = device_register(dev);
 750
 751         mutex_unlock(&xsegbd_mutex);
 752         return ret;
 753 }
 754
 755 static void xsegbd_bus_del_dev(struct xsegbd_device *xsegbd_dev)
 756 {
 757         device_unregister(&xsegbd_dev->dev);
 758 }
 759
 760 static ssize_t xsegbd_add(struct bus_type *bus, const char *buf, size_t count)
 761 {
 762         struct xsegbd_device *xsegbd_dev;
 763         struct xseg_port *port;
 764         ssize_t ret = -ENOMEM;
 765
 766         if (!try_module_get(THIS_MODULE))
 767                 return -ENODEV;
 768
 769         xsegbd_dev = kzalloc(sizeof(*xsegbd_dev), GFP_KERNEL);
 770         if (!xsegbd_dev)
 771                 goto out;
 772
 773         spin_lock_init(&xsegbd_dev->rqlock);
 774         spin_lock_init(&xsegbd_dev->reqdatalock);
 775         INIT_LIST_HEAD(&xsegbd_dev->node);
 776
 777         /* parse cmd */
 778         if (sscanf(buf, "%" __stringify(XSEGBD_TARGET_NAMELEN) "s "
 779                         "%d:%d:%d", xsegbd_dev->target, &xsegbd_dev->src_portno,
 780                         &xsegbd_dev->dst_portno, &xsegbd_dev->nr_requests) < 3) {
 781                 ret = -EINVAL;
 782                 goto out_dev;
 783         }
 784         xsegbd_dev->targetlen = strlen(xsegbd_dev->target);
 785
 786         spin_lock(&xsegbd_devices_lock);
 787         if (xsegbd_devices[xsegbd_dev->src_portno] != NULL) {
 788                 ret = -EINVAL;
 789                 goto out_unlock;
 790         }
 791         xsegbd_devices[xsegbd_dev->src_portno] = xsegbd_dev;
 792         xsegbd_dev->id = xsegbd_dev->src_portno;
 793         spin_unlock(&xsegbd_devices_lock);
 794
 795         XSEGLOG("registering block device major %d", major);
 796         ret = register_blkdev(major, XSEGBD_NAME);
 797         if (ret < 0) {
 798                 XSEGLOG("cannot register block device!");
 799                 ret = -EBUSY;
 800                 goto out_delentry;
 801         }
 802         xsegbd_dev->major = ret;
 803         XSEGLOG("registered block device major %d", xsegbd_dev->major);
 804
 805         ret = xsegbd_bus_add_dev(xsegbd_dev);
 806         if (ret)
 807                 goto out_blkdev;
 808
 809         if (!xq_alloc_seq(&xsegbd_dev->blk_queue_pending,
 810                                 xsegbd_dev->nr_requests,
 811                                 xsegbd_dev->nr_requests))
 812                 goto out_bus;
 813
 814         xsegbd_dev->blk_req_pending = kzalloc(
 815                         xsegbd_dev->nr_requests *sizeof(struct xsegbd_pending),
 816                                    GFP_KERNEL);
 817         if (!xsegbd_dev->blk_req_pending)
 818                 goto out_freeq;
 819
 820
 821         XSEGLOG("joining segment");
 822         //FIXME use xsebd module config for now
 823         xsegbd_dev->xseg = xseg_join(   xsegbd.config.type,
 824                                         xsegbd.config.name,
 825                                         "segdev",
 826                                         xseg_callback           );
 827         if (!xsegbd_dev->xseg)
 828                 goto out_freepending;
 829
 830
 831         XSEGLOG("binding to source port %u (destination %u)",
 832                         xsegbd_dev->src_portno, xsegbd_dev->dst_portno);
 833         port = xseg_bind_port(xsegbd_dev->xseg, xsegbd_dev->src_portno);
 834         if (!port) {
 835                 XSEGLOG("cannot bind to port");
 836                 ret = -EFAULT;
 837
 838                 goto out_xseg;
 839         }
 840         //FIXME rollback here
 841         BUG_ON(xsegbd_dev->src_portno != xseg_portno(xsegbd_dev->xseg, port));
 842
 843         /* make sure we don't get any requests until we're ready to handle them */
 844         xseg_cancel_wait(xsegbd_dev->xseg, xseg_portno(xsegbd_dev->xseg, port));
 845
 846         ret = xsegbd_dev_init(xsegbd_dev);
 847         if (ret)
 848                 goto out_xseg;
 849
 850         return count;
 851
 852 out_xseg:
 853         xseg_leave(xsegbd_dev->xseg);
 854
 855 out_freepending:
 856         kfree(xsegbd_dev->blk_req_pending);
 857
 858 out_freeq:
 859         xq_free(&xsegbd_dev->blk_queue_pending);
 860
 861 out_bus:
 862         xsegbd_bus_del_dev(xsegbd_dev);
 863
 864         return ret;
 865
 866 out_blkdev:
 867         unregister_blkdev(xsegbd_dev->major, XSEGBD_NAME);
 868
 869 out_delentry:
 870         spin_lock(&xsegbd_devices_lock);
 871         xsegbd_devices[xsegbd_dev->src_portno] = NULL;
 872
 873 out_unlock:
 874         spin_unlock(&xsegbd_devices_lock);
 875
 876 out_dev:
 877         kfree(xsegbd_dev);
 878
 879 out:
 880         return ret;
 881 }
 882
 883 static ssize_t xsegbd_remove(struct bus_type *bus, const char *buf, size_t count)
 884 {
 885         struct xsegbd_device *xsegbd_dev = NULL;
 886         int id, ret;
 887         unsigned long ul_id;
 888
 889         ret = strict_strtoul(buf, 10, &ul_id);
 890         if (ret)
 891                 return ret;
 892
 893         id = (int) ul_id;
 894         if (id != ul_id)
 895                 return -EINVAL;
 896
 897         mutex_lock_nested(&xsegbd_mutex, SINGLE_DEPTH_NESTING);
 898
 899         ret = count;
 900         xsegbd_dev = __xsegbd_get_dev(id);
 901         if (!xsegbd_dev) {
 902                 ret = -ENOENT;
 903                 goto out_unlock;
 904         }
 905         xsegbd_bus_del_dev(xsegbd_dev);
 906
 907 out_unlock:
 908         mutex_unlock(&xsegbd_mutex);
 909         return ret;
 910 }
 911
 912 static struct bus_attribute xsegbd_bus_attrs[] = {
 913         __ATTR(add, S_IWUSR, NULL, xsegbd_add),
 914         __ATTR(remove, S_IWUSR, NULL, xsegbd_remove),
 915         __ATTR_NULL
 916 };
 917
 918 static int xsegbd_sysfs_init(void)
 919 {
 920         int ret;
 921
 922         ret = device_register(&xsegbd_root_dev);
 923         if (ret < 0)
 924                 return ret;
 925
 926         xsegbd_bus_type.bus_attrs = xsegbd_bus_attrs;
 927         ret = bus_register(&xsegbd_bus_type);
 928         if (ret < 0)
 929                 device_unregister(&xsegbd_root_dev);
 930
 931         return ret;
 932 }
 933
 934 static void xsegbd_sysfs_cleanup(void)
 935 {
 936         bus_unregister(&xsegbd_bus_type);
 937         device_unregister(&xsegbd_root_dev);
 938 }
 939
 940 /* *************************** */
 941 /* ** Module Initialization ** */
 942 /* *************************** */
 943
 944 static int __init xsegbd_init(void)
 945 {
 946         int ret = -ENOMEM;
 947         xsegbd_devices = kzalloc(max_dev * sizeof(struct xsegbd_devices *), GFP_KERNEL);
 948         if (!xsegbd_devices)
 949                 goto out;
 950
 951         spin_lock_init(&xsegbd_devices_lock);
 952
 953         ret = -ENOSYS;
 954         ret = xsegbd_xseg_init();
 955         if (ret)
 956                 goto out_free;
 957
 958         ret = xsegbd_sysfs_init();
 959         if (ret)
 960                 goto out_xseg;
 961
 962         XSEGLOG("initialization complete");
 963
 964 out:
 965         return ret;
 966
 967 out_xseg:
 968         xsegbd_xseg_quit();
 969
 970 out_free:
 971         kfree(xsegbd_devices);
 972
 973         goto out;
 974 }
 975
 976 static void __exit xsegbd_exit(void)
 977 {
 978         xsegbd_sysfs_cleanup();
 979         xsegbd_xseg_quit();
 980 }
 981
 982 module_init(xsegbd_init);
 983 module_exit(xsegbd_exit);
 984