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_nr_pending = 1024;
  38 static char name[XSEGBD_SEGMENT_NAMELEN] = "xsegbd";
  39 static char spec[256] = "segdev:xsegbd:4:256:12";
  40
  41 module_param(sector_size, long, 0644);
  42 module_param(blksize, long, 0644);
  43 module_param(max_nr_pending, 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 struct pending {
  49         struct request *request;
  50         struct completion *comp;
  51         struct xsegbd_device *dev;
  52 };
  53
  54 static struct xq blk_queue_pending;
  55 static struct pending *blk_req_pending;
  56 static unsigned int nr_pending;
  57 static spinlock_t __lock;
  58 static struct xsegbd xsegbd;
  59 static DEFINE_MUTEX(xsegbd_mutex);
  60 static LIST_HEAD(xsegbd_dev_list);
  61 static DEFINE_SPINLOCK(xsegbd_dev_list_lock);
  62
  63 /* ************************* */
  64 /* ***** sysfs helpers ***** */
  65 /* ************************* */
  66
  67 static struct xsegbd_device *dev_to_xsegbd(struct device *dev)
  68 {
  69         return container_of(dev, struct xsegbd_device, dev);
  70 }
  71
  72 static struct device *xsegbd_get_dev(struct xsegbd_device *xsegbd_dev)
  73 {
  74         /* FIXME */
  75         return get_device(&xsegbd_dev->dev);
  76 }
  77
  78 static void xsegbd_put_dev(struct xsegbd_device *xsegbd_dev)
  79 {
  80         put_device(&xsegbd_dev->dev);
  81 }
  82
  83 /* ************************* */
  84 /* ** XSEG Initialization ** */
  85 /* ************************* */
  86
  87 static void xseg_callback(struct xseg *xseg, uint32_t portno);
  88
  89 int xsegbd_xseg_init(void)
  90 {
  91         int r;
  92
  93         if (!xsegbd.name[0])
  94                 strncpy(xsegbd.name, name, XSEGBD_SEGMENT_NAMELEN);
  95
  96         r = xseg_initialize();
  97         if (r) {
  98                 XSEGLOG("cannot initialize 'segdev' peer");
  99                 goto err;
 100         }
 101
 102         r = xseg_parse_spec(spec, &xsegbd.config);
 103         if (r)
 104                 goto err;
 105
 106         if (strncmp(xsegbd.config.type, "segdev", 16))
 107                 XSEGLOG("WARNING: unexpected segment type '%s' vs 'segdev'",
 108                          xsegbd.config.type);
 109
 110         XSEGLOG("joining segment");
 111         xsegbd.xseg = xseg_join(        xsegbd.config.type,
 112                                         xsegbd.config.name,
 113                                         "segdev",
 114                                         xseg_callback           );
 115         if (!xsegbd.xseg) {
 116                 XSEGLOG("cannot find segment");
 117                 r = -ENODEV;
 118                 goto err;
 119         }
 120
 121         return 0;
 122 err:
 123         return r;
 124
 125 }
 126
 127 int xsegbd_xseg_quit(void)
 128 {
 129         struct segdev *segdev;
 130
 131         /* make sure to unmap the segment first */
 132         segdev = segdev_get(0);
 133         clear_bit(SEGDEV_RESERVED, &segdev->flags);
 134         xsegbd.xseg->priv->segment_type.ops.unmap(xsegbd.xseg, xsegbd.xseg->segment_size);
 135         segdev_put(segdev);
 136
 137         return 0;
 138 }
 139
 140
 141 /* ***************************** */
 142 /* ** Block Device Operations ** */
 143 /* ***************************** */
 144
 145 static int xsegbd_open(struct block_device *bdev, fmode_t mode)
 146 {
 147         struct gendisk *disk = bdev->bd_disk;
 148         struct xsegbd_device *xsegbd_dev = disk->private_data;
 149
 150         xsegbd_get_dev(xsegbd_dev);
 151
 152         return 0;
 153 }
 154
 155 static int xsegbd_release(struct gendisk *gd, fmode_t mode)
 156 {
 157         struct xsegbd_device *xsegbd_dev = gd->private_data;
 158
 159         xsegbd_put_dev(xsegbd_dev);
 160
 161         return 0;
 162 }
 163
 164 static int xsegbd_ioctl(struct block_device *bdev, fmode_t mode,
 165                         unsigned int cmd, unsigned long arg)
 166 {
 167         return -ENOTTY;
 168 }
 169
 170 static const struct block_device_operations xsegbd_ops = {
 171         .owner          = THIS_MODULE,
 172         .open           = xsegbd_open,
 173         .release        = xsegbd_release,
 174         .ioctl          = xsegbd_ioctl
 175 };
 176
 177
 178 /* *************************** */
 179 /* ** Device Initialization ** */
 180 /* *************************** */
 181
 182 static void xseg_request_fn(struct request_queue *rq);
 183 static int xsegbd_get_size(struct xsegbd_device *xsegbd_dev);
 184
 185 static int xsegbd_dev_init(struct xsegbd_device *xsegbd_dev)
 186 {
 187         int ret = -ENOMEM;
 188         struct gendisk *disk;
 189         unsigned int max_request_size_bytes;
 190
 191         spin_lock_init(&xsegbd_dev->lock);
 192
 193         xsegbd_dev->xsegbd = &xsegbd;
 194
 195         xsegbd_dev->blk_queue = blk_alloc_queue(GFP_KERNEL);
 196         if (!xsegbd_dev->blk_queue)
 197                 goto out;
 198
 199         blk_init_allocated_queue(xsegbd_dev->blk_queue, xseg_request_fn, &xsegbd_dev->lock);
 200         xsegbd_dev->blk_queue->queuedata = xsegbd_dev;
 201
 202         blk_queue_flush(xsegbd_dev->blk_queue, REQ_FLUSH | REQ_FUA);
 203         blk_queue_logical_block_size(xsegbd_dev->blk_queue, 512);
 204         blk_queue_physical_block_size(xsegbd_dev->blk_queue, blksize);
 205         blk_queue_bounce_limit(xsegbd_dev->blk_queue, BLK_BOUNCE_ANY);
 206
 207         //blk_queue_max_segments(dev->blk_queue, 512);
 208
 209         max_request_size_bytes = XSEGBD_MAX_REQUEST_SIZE;
 210         blk_queue_max_hw_sectors(xsegbd_dev->blk_queue, max_request_size_bytes >> 9);
 211         blk_queue_max_segment_size(xsegbd_dev->blk_queue, max_request_size_bytes);
 212         blk_queue_io_min(xsegbd_dev->blk_queue, max_request_size_bytes);
 213         blk_queue_io_opt(xsegbd_dev->blk_queue, max_request_size_bytes);
 214
 215         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, xsegbd_dev->blk_queue);
 216
 217         /* vkoukis says we don't need partitions */
 218         xsegbd_dev->gd = disk = alloc_disk(1);
 219         if (!disk)
 220                 /* FIXME: We call xsegbd_dev_release if something goes wrong, to cleanup
 221                  * disks/queues/etc.
 222                  * Would it be better to do the cleanup here, and conditionally cleanup
 223                  * in dev_release?
 224                  */
 225                 goto out;
 226
 227         disk->major = xsegbd_dev->major;
 228         disk->first_minor = 0; // id * XSEGBD_MINORS;
 229         disk->fops = &xsegbd_ops;
 230         disk->queue = xsegbd_dev->blk_queue;
 231         disk->private_data = xsegbd_dev;
 232         disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
 233         snprintf(disk->disk_name, 32, "xsegbd%u", xsegbd_dev->id);
 234
 235         ret = 0;
 236         spin_lock_irq(&__lock);
 237         if (nr_pending + xsegbd_dev->nr_requests > max_nr_pending)
 238                 ret = -ENOBUFS;
 239         else
 240                 nr_pending += xsegbd_dev->nr_requests;
 241         spin_unlock_irq(&__lock);
 242
 243         if (ret)
 244                 goto out;
 245
 246         /* allow a non-zero sector_size parameter to override the disk size */
 247         if (sector_size)
 248                 xsegbd_dev->sectors = sector_size;
 249         else {
 250                 ret = xsegbd_get_size(xsegbd_dev);
 251                 if (ret)
 252                         goto out;
 253         }
 254
 255         set_capacity(disk, xsegbd_dev->sectors);
 256         XSEGLOG("xsegbd active...");
 257         add_disk(disk); /* immediately activates the device */
 258
 259         return 0;
 260
 261 out:
 262         return ret;
 263 }
 264
 265 static void xsegbd_dev_release(struct device *dev)
 266 {
 267         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 268
 269         /* cleanup gendisk and blk_queue the right way */
 270         if (xsegbd_dev->gd) {
 271                 if (xsegbd_dev->gd->flags & GENHD_FL_UP)
 272                         del_gendisk(xsegbd_dev->gd);
 273
 274                 blk_cleanup_queue(xsegbd_dev->blk_queue);
 275                 put_disk(xsegbd_dev->gd);
 276         }
 277
 278         /* xsegbd actually does not need use waiting.
 279          * maybe we use xseg_cancel_wait for clarity
 280          * with xseg_segdev kernel driver convert this
 281          * to a noop
 282          */
 283 //      xseg_cancel_wait(xseg, xsegbd_dev->src_portno);
 284
 285         if (xseg_free_requests(xsegbd.xseg, xsegbd_dev->src_portno, xsegbd_dev->nr_requests) != 0)
 286                 XSEGLOG("Error trying to free requests!\n");
 287
 288         WARN_ON(nr_pending < xsegbd_dev->nr_requests);
 289         spin_lock_irq(&__lock);
 290         nr_pending -= xsegbd_dev->nr_requests;
 291         spin_unlock_irq(&__lock);
 292
 293         unregister_blkdev(xsegbd_dev->major, XSEGBD_NAME);
 294
 295         spin_lock(&xsegbd_dev_list_lock);
 296         list_del_init(&xsegbd_dev->node);
 297         spin_unlock(&xsegbd_dev_list_lock);
 298         kfree(xsegbd_dev);
 299
 300         module_put(THIS_MODULE);
 301 }
 302
 303 /* ******************* */
 304 /* ** Critical Path ** */
 305 /* ******************* */
 306
 307 static void blk_to_xseg(struct xseg *xseg, struct xseg_request *xreq,
 308                         struct request *blkreq)
 309 {
 310         struct bio_vec *bvec;
 311         struct req_iterator iter;
 312         uint64_t off = 0;
 313         char *data = XSEG_TAKE_PTR(xreq->data, xseg->segment);
 314         rq_for_each_segment(bvec, blkreq, iter) {
 315                 char *bdata = kmap_atomic(bvec->bv_page) + bvec->bv_offset;
 316                 memcpy(data + off, bdata, bvec->bv_len);
 317                 off += bvec->bv_len;
 318                 kunmap_atomic(bdata);
 319         }
 320 }
 321
 322 static void xseg_to_blk(struct xseg *xseg, struct xseg_request *xreq,
 323                         struct request *blkreq)
 324 {
 325         struct bio_vec *bvec;
 326         struct req_iterator iter;
 327         uint64_t off = 0;
 328         char *data = XSEG_TAKE_PTR(xreq->data, xseg->segment);
 329         rq_for_each_segment(bvec, blkreq, iter) {
 330                 char *bdata = kmap_atomic(bvec->bv_page) + bvec->bv_offset;
 331                 memcpy(bdata, data + off, bvec->bv_len);
 332                 off += bvec->bv_len;
 333                 kunmap_atomic(bdata);
 334         }
 335 }
 336
 337 static void xseg_request_fn(struct request_queue *rq)
 338 {
 339         struct xseg_request *xreq;
 340         struct xsegbd_device *xsegbd_dev = rq->queuedata;
 341         struct request *blkreq;
 342         struct pending *pending;
 343         xqindex blkreq_idx;
 344         char *target;
 345         uint64_t datalen;
 346
 347         for (;;) {
 348                 xreq = xseg_get_request(xsegbd.xseg, xsegbd_dev->src_portno);
 349                 if (!xreq)
 350                         break;
 351
 352                 blkreq = blk_fetch_request(rq);
 353                 if (!blkreq)
 354                         break;
 355
 356                 if (blkreq->cmd_type != REQ_TYPE_FS) {
 357                         XSEGLOG("non-fs cmd_type: %u. *shrug*", blkreq->cmd_type);
 358                         __blk_end_request_all(blkreq, 0);
 359                 }
 360
 361
 362                 datalen = blk_rq_bytes(blkreq);
 363                 BUG_ON(xseg_prep_request(xsegbd.xseg, xreq, xsegbd_dev->targetlen, datalen));
 364                 BUG_ON(xreq->bufferlen - xsegbd_dev->targetlen < datalen);
 365
 366                 target = XSEG_TAKE_PTR(xreq->target, xsegbd.xseg->segment);
 367                 strncpy(target, xsegbd_dev->target, xsegbd_dev->targetlen);
 368                 blkreq_idx = xq_pop_head(&blk_queue_pending, 1);
 369                 BUG_ON(blkreq_idx == Noneidx);
 370                 pending = &blk_req_pending[blkreq_idx];
 371                 pending->dev = xsegbd_dev;
 372                 pending->request = blkreq;
 373                 pending->comp = NULL;
 374                 xreq->priv = (uint64_t)blkreq_idx;
 375                 xreq->size = datalen;
 376                 xreq->offset = blk_rq_pos(blkreq) << 9;
 377                 /*
 378                 if (xreq->offset >= (sector_size << 9))
 379                         XSEGLOG("sector offset: %lu > %lu, flush:%u, fua:%u",
 380                                  blk_rq_pos(blkreq), sector_size,
 381                                  blkreq->cmd_flags & REQ_FLUSH,
 382                                  blkreq->cmd_flags & REQ_FUA);
 383                 */
 384
 385                 if (blkreq->cmd_flags & REQ_FLUSH)
 386                         xreq->flags |= XF_FLUSH;
 387
 388                 if (blkreq->cmd_flags & REQ_FUA)
 389                         xreq->flags |= XF_FUA;
 390
 391                 if (rq_data_dir(blkreq)) {
 392                         /* unlock for data transfers? */
 393                         blk_to_xseg(xsegbd.xseg, xreq, blkreq);
 394                         xreq->op = X_WRITE;
 395                 } else {
 396                         xreq->op = X_READ;
 397                 }
 398
 399                 BUG_ON(xseg_submit(xsegbd.xseg, xsegbd_dev->dst_portno, xreq) == NoSerial);
 400         }
 401
 402         WARN_ON(xseg_signal(xsegbd_dev->xsegbd->xseg, xsegbd_dev->dst_portno) < 0);
 403         if (xreq)
 404                 BUG_ON(xseg_put_request(xsegbd_dev->xsegbd->xseg, xsegbd_dev->src_portno, xreq) == NoSerial);
 405 }
 406
 407 int update_dev_sectors_from_request(    struct xsegbd_device *xsegbd_dev,
 408                                         struct xseg_request *xreq       )
 409 {
 410         void *data;
 411
 412         if (xreq->state & XS_FAILED)
 413                 return -ENOENT;
 414
 415         if (!(xreq->state & XS_SERVED))
 416                 return -EIO;
 417
 418         data = XSEG_TAKE_PTR(xreq->data, xsegbd.xseg->segment);
 419         xsegbd_dev->sectors = *((uint64_t *) data) / 512ULL;
 420         return 0;
 421 }
 422
 423 static int xsegbd_get_size(struct xsegbd_device *xsegbd_dev)
 424 {
 425         struct xseg_request *xreq;
 426         char *target;
 427         uint64_t datalen;
 428         xqindex blkreq_idx;
 429         struct pending *pending;
 430         struct completion comp;
 431         int ret = -EBUSY;
 432
 433         xreq = xseg_get_request(xsegbd.xseg, xsegbd_dev->src_portno);
 434         if (!xreq)
 435                 goto out;
 436
 437         datalen = sizeof(uint64_t);
 438         BUG_ON(xseg_prep_request(xsegbd.xseg, xreq, xsegbd_dev->targetlen, datalen));
 439         BUG_ON(xreq->bufferlen - xsegbd_dev->targetlen < datalen);
 440
 441         init_completion(&comp);
 442         blkreq_idx = xq_pop_head(&blk_queue_pending, 1);
 443         BUG_ON(blkreq_idx == Noneidx);
 444         pending = &blk_req_pending[blkreq_idx];
 445         pending->dev = xsegbd_dev;
 446         pending->request = NULL;
 447         pending->comp = &comp;
 448         xreq->priv = (uint64_t)blkreq_idx;
 449
 450         target = XSEG_TAKE_PTR(xreq->target, xsegbd.xseg->segment);
 451         strncpy(target, xsegbd_dev->target, xsegbd_dev->targetlen);
 452         xreq->size = datalen;
 453         xreq->offset = 0;
 454
 455         xreq->op = X_INFO;
 456
 457         /* waiting is not needed.
 458          * but it should be better to use xseg_prepare_wait
 459          * and the xseg_segdev kernel driver, would be a no op
 460          */
 461 //      port = &xsegbd.xseg->ports[xsegbd_dev->src_portno];
 462 //      port->waitcue = (uint64_t)(long)xsegbd_dev;
 463
 464         BUG_ON(xseg_submit(xsegbd.xseg, xsegbd_dev->dst_portno, xreq) == NoSerial);
 465         WARN_ON(xseg_signal(xsegbd.xseg, xsegbd_dev->dst_portno) < 0);
 466
 467         wait_for_completion_interruptible(&comp);
 468         XSEGLOG("Woken up after wait_for_completion_interruptible()\n");
 469         ret = update_dev_sectors_from_request(xsegbd_dev, xreq);
 470         XSEGLOG("get_size: sectors = %ld\n", (long)xsegbd_dev->sectors);
 471 out:
 472         BUG_ON(xseg_put_request(xsegbd.xseg, xsegbd_dev->src_portno, xreq) == NoSerial);
 473         return ret;
 474 }
 475
 476 static void xseg_callback(struct xseg *xseg, uint32_t portno)
 477 {
 478         struct xsegbd_device *xsegbd_dev = NULL, *old_dev = NULL;
 479         struct xseg_request *xreq;
 480         struct request *blkreq;
 481         struct pending *pending;
 482         unsigned long flags;
 483         uint32_t blkreq_idx;
 484         int err;
 485
 486         for (;;) {
 487                 xreq = xseg_receive(xseg, portno);
 488                 if (!xreq)
 489                         break;
 490
 491                 /* we rely upon our peers to not have touched ->priv */
 492                 blkreq_idx = (uint64_t)xreq->priv;
 493                 if (blkreq_idx >= max_nr_pending) {
 494                         WARN_ON(1);
 495                         continue;
 496                 }
 497
 498                 pending = &blk_req_pending[blkreq_idx];
 499                 if (pending->comp) {
 500                         /* someone is blocking on this request
 501                            and will handle it when we wake them up. */
 502                         complete(pending->comp);
 503                         /* the request is blocker's responsibility so
 504                            we will not put_request(); */
 505                         continue;
 506                 }
 507
 508                 /* this is now treated as a block I/O request to end */
 509                 blkreq = pending->request;
 510                 pending->request = NULL;
 511                 xsegbd_dev = pending->dev;
 512                 pending->dev = NULL;
 513                 WARN_ON(!blkreq);
 514
 515                 if ((xsegbd_dev != old_dev) && old_dev) {
 516                         spin_lock_irqsave(&old_dev->lock, flags);
 517                         xseg_request_fn(old_dev->blk_queue);
 518                         spin_unlock_irqrestore(&old_dev->lock, flags);
 519                 }
 520
 521                 old_dev = xsegbd_dev;
 522
 523                 if (!(xreq->state & XS_SERVED))
 524                         goto blk_end;
 525
 526                 if (xreq->serviced != blk_rq_bytes(blkreq))
 527                         goto blk_end;
 528
 529                 /* unlock for data transfer? */
 530                 if (!rq_data_dir(blkreq))
 531                         xseg_to_blk(xseg, xreq, blkreq);
 532
 533                 err = 0;
 534 blk_end:
 535                 blk_end_request_all(blkreq, err);
 536                 xq_append_head(&blk_queue_pending, blkreq_idx, 1);
 537                 BUG_ON(xseg_put_request(xseg, xreq->portno, xreq) == NoSerial);
 538         }
 539
 540         if (xsegbd_dev) {
 541                 spin_lock_irqsave(&xsegbd_dev->lock, flags);
 542                 xseg_request_fn(xsegbd_dev->blk_queue);
 543                 spin_unlock_irqrestore(&xsegbd_dev->lock, flags);
 544         }
 545 }
 546
 547
 548 /* sysfs interface */
 549
 550 static struct bus_type xsegbd_bus_type = {
 551         .name   = "xsegbd",
 552 };
 553
 554 static ssize_t xsegbd_size_show(struct device *dev,
 555                                         struct device_attribute *attr, char *buf)
 556 {
 557         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 558
 559         return sprintf(buf, "%llu\n", (unsigned long long) xsegbd_dev->sectors * 512ULL);
 560 }
 561
 562 static ssize_t xsegbd_major_show(struct device *dev,
 563                                         struct device_attribute *attr, char *buf)
 564 {
 565         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 566
 567         return sprintf(buf, "%d\n", xsegbd_dev->major);
 568 }
 569
 570 static ssize_t xsegbd_srcport_show(struct device *dev,
 571                                         struct device_attribute *attr, char *buf)
 572 {
 573         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 574
 575         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->src_portno);
 576 }
 577
 578 static ssize_t xsegbd_dstport_show(struct device *dev,
 579                                         struct device_attribute *attr, char *buf)
 580 {
 581         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 582
 583         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->dst_portno);
 584 }
 585
 586 static ssize_t xsegbd_id_show(struct device *dev,
 587                                         struct device_attribute *attr, char *buf)
 588 {
 589         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 590
 591         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->id);
 592 }
 593
 594 static ssize_t xsegbd_reqs_show(struct device *dev,
 595                                         struct device_attribute *attr, char *buf)
 596 {
 597         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 598
 599         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->nr_requests);
 600 }
 601
 602 static ssize_t xsegbd_target_show(struct device *dev,
 603                                         struct device_attribute *attr, char *buf)
 604 {
 605         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 606
 607         return sprintf(buf, "%s\n", xsegbd_dev->target);
 608 }
 609
 610 static ssize_t xsegbd_image_refresh(struct device *dev,
 611                                         struct device_attribute *attr,
 612                                         const char *buf,
 613                                         size_t size)
 614 {
 615         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 616         int rc, ret = size;
 617
 618         mutex_lock_nested(&xsegbd_mutex, SINGLE_DEPTH_NESTING);
 619
 620         rc = xsegbd_get_size(xsegbd_dev);
 621         if (rc < 0) {
 622                 ret = rc;
 623                 goto out;
 624         }
 625
 626         set_capacity(xsegbd_dev->gd, xsegbd_dev->sectors);
 627
 628 out:
 629         mutex_unlock(&xsegbd_mutex);
 630         return ret;
 631 }
 632
 633 static DEVICE_ATTR(size, S_IRUGO, xsegbd_size_show, NULL);
 634 static DEVICE_ATTR(major, S_IRUGO, xsegbd_major_show, NULL);
 635 static DEVICE_ATTR(srcport, S_IRUGO, xsegbd_srcport_show, NULL);
 636 static DEVICE_ATTR(dstport, S_IRUGO, xsegbd_dstport_show, NULL);
 637 static DEVICE_ATTR(id , S_IRUGO, xsegbd_id_show, NULL);
 638 static DEVICE_ATTR(reqs , S_IRUGO, xsegbd_reqs_show, NULL);
 639 static DEVICE_ATTR(target, S_IRUGO, xsegbd_target_show, NULL);
 640 static DEVICE_ATTR(refresh , S_IWUSR, NULL, xsegbd_image_refresh);
 641
 642 static struct attribute *xsegbd_attrs[] = {
 643         &dev_attr_size.attr,
 644         &dev_attr_major.attr,
 645         &dev_attr_srcport.attr,
 646         &dev_attr_dstport.attr,
 647         &dev_attr_id.attr,
 648         &dev_attr_reqs.attr,
 649         &dev_attr_target.attr,
 650         &dev_attr_refresh.attr,
 651         NULL
 652 };
 653
 654 static struct attribute_group xsegbd_attr_group = {
 655         .attrs = xsegbd_attrs,
 656 };
 657
 658 static const struct attribute_group *xsegbd_attr_groups[] = {
 659         &xsegbd_attr_group,
 660         NULL
 661 };
 662
 663 static void xsegbd_sysfs_dev_release(struct device *dev)
 664 {
 665 }
 666
 667 static struct device_type xsegbd_device_type = {
 668         .name           = "xsegbd",
 669         .groups         = xsegbd_attr_groups,
 670         .release        = xsegbd_sysfs_dev_release,
 671 };
 672
 673 static void xsegbd_root_dev_release(struct device *dev)
 674 {
 675 }
 676
 677 static struct device xsegbd_root_dev = {
 678         .init_name      = "xsegbd",
 679         .release        = xsegbd_root_dev_release,
 680 };
 681
 682 static int xsegbd_bus_add_dev(struct xsegbd_device *xsegbd_dev)
 683 {
 684         int ret = -ENOMEM;
 685         struct device *dev;
 686
 687         mutex_lock_nested(&xsegbd_mutex, SINGLE_DEPTH_NESTING);
 688         dev = &xsegbd_dev->dev;
 689
 690         dev->bus = &xsegbd_bus_type;
 691         dev->type = &xsegbd_device_type;
 692         dev->parent = &xsegbd_root_dev;
 693         dev->release = xsegbd_dev_release;
 694         dev_set_name(dev, "%d", xsegbd_dev->id);
 695
 696         ret = device_register(dev);
 697
 698         mutex_unlock(&xsegbd_mutex);
 699         return ret;
 700 }
 701
 702 static void xsegbd_bus_del_dev(struct xsegbd_device *xsegbd_dev)
 703 {
 704         device_unregister(&xsegbd_dev->dev);
 705 }
 706
 707 static ssize_t xsegbd_add(struct bus_type *bus, const char *buf, size_t count)
 708 {
 709         struct xsegbd_device *xsegbd_dev;
 710         struct xseg_port *xport;
 711         ssize_t ret = -ENOMEM;
 712         int new_id = 0;
 713         struct list_head *tmp;
 714
 715         if (!try_module_get(THIS_MODULE))
 716                 return -ENODEV;
 717
 718         xsegbd_dev = kzalloc(sizeof(*xsegbd_dev), GFP_KERNEL);
 719         if (!xsegbd_dev)
 720                 goto out;
 721
 722         spin_lock_init(&xsegbd_dev->lock);
 723         INIT_LIST_HEAD(&xsegbd_dev->node);
 724
 725         /* parse cmd */
 726         if (sscanf(buf, "%" __stringify(XSEGBD_TARGET_NAMELEN) "s "
 727                         "%d:%d:%d", xsegbd_dev->target, &xsegbd_dev->src_portno,
 728                         &xsegbd_dev->dst_portno, &xsegbd_dev->nr_requests) < 3) {
 729                 ret = -EINVAL;
 730                 goto out_dev;
 731         }
 732         xsegbd_dev->targetlen = strlen(xsegbd_dev->target);
 733
 734         spin_lock(&xsegbd_dev_list_lock);
 735
 736         list_for_each(tmp, &xsegbd_dev_list) {
 737                 struct xsegbd_device *entry;
 738
 739                 entry = list_entry(tmp, struct xsegbd_device, node);
 740
 741                 if (entry->src_portno == xsegbd_dev->src_portno) {
 742                         ret = -EINVAL;
 743                         goto out_unlock;
 744                 }
 745
 746                 if (entry->id >= new_id)
 747                         new_id = entry->id + 1;
 748         }
 749
 750         xsegbd_dev->id = new_id;
 751
 752         list_add_tail(&xsegbd_dev->node, &xsegbd_dev_list);
 753
 754         spin_unlock(&xsegbd_dev_list_lock);
 755
 756         XSEGLOG("registering block device major %d", major);
 757         ret = register_blkdev(major, XSEGBD_NAME);
 758         if (ret < 0) {
 759                 XSEGLOG("cannot register block device!");
 760                 ret = -EBUSY;
 761                 goto out_delentry;
 762         }
 763         xsegbd_dev->major = ret;
 764         XSEGLOG("registered block device major %d", xsegbd_dev->major);
 765
 766         ret = xsegbd_bus_add_dev(xsegbd_dev);
 767         if (ret)
 768                 goto out_blkdev;
 769
 770         XSEGLOG("binding to source port %u (destination %u)",
 771                         xsegbd_dev->src_portno, xsegbd_dev->dst_portno);
 772         xport = xseg_bind_port(xsegbd.xseg, xsegbd_dev->src_portno);
 773         if (!xport) {
 774                 XSEGLOG("cannot bind to port");
 775                 ret = -EFAULT;
 776
 777                 goto out_bus;
 778         }
 779         /* make sure we don't get any requests until we're ready to handle them */
 780         xport->waitcue = (long) NULL;
 781
 782         XSEGLOG("allocating %u requests", xsegbd_dev->nr_requests);
 783         if (xseg_alloc_requests(xsegbd.xseg, xsegbd_dev->src_portno, xsegbd_dev->nr_requests)) {
 784                 XSEGLOG("cannot allocate requests");
 785                 ret = -EFAULT;
 786
 787                 goto out_bus;
 788         }
 789
 790         ret = xsegbd_dev_init(xsegbd_dev);
 791         if (ret)
 792                 goto out_bus;
 793
 794         return count;
 795
 796 out_bus:
 797         xsegbd_bus_del_dev(xsegbd_dev);
 798
 799         return ret;
 800
 801 out_blkdev:
 802         unregister_blkdev(xsegbd_dev->major, XSEGBD_NAME);
 803
 804 out_delentry:
 805         spin_lock(&xsegbd_dev_list_lock);
 806         list_del_init(&xsegbd_dev->node);
 807
 808 out_unlock:
 809         spin_unlock(&xsegbd_dev_list_lock);
 810
 811 out_dev:
 812         kfree(xsegbd_dev);
 813
 814 out:
 815         return ret;
 816 }
 817
 818 static struct xsegbd_device *__xsegbd_get_dev(unsigned long id)
 819 {
 820         struct list_head *tmp;
 821         struct xsegbd_device *xsegbd_dev;
 822
 823
 824         spin_lock(&xsegbd_dev_list_lock);
 825         list_for_each(tmp, &xsegbd_dev_list) {
 826                 xsegbd_dev = list_entry(tmp, struct xsegbd_device, node);
 827                 if (xsegbd_dev->id == id) {
 828                         spin_unlock(&xsegbd_dev_list_lock);
 829                         return xsegbd_dev;
 830                 }
 831         }
 832         spin_unlock(&xsegbd_dev_list_lock);
 833         return NULL;
 834 }
 835
 836 static ssize_t xsegbd_remove(struct bus_type *bus, const char *buf, size_t count)
 837 {
 838         struct xsegbd_device *xsegbd_dev = NULL;
 839         int id, ret;
 840         unsigned long ul_id;
 841
 842         ret = strict_strtoul(buf, 10, &ul_id);
 843         if (ret)
 844                 return ret;
 845
 846         id = (int) ul_id;
 847         if (id != ul_id)
 848                 return -EINVAL;
 849
 850         mutex_lock_nested(&xsegbd_mutex, SINGLE_DEPTH_NESTING);
 851
 852         ret = count;
 853         xsegbd_dev = __xsegbd_get_dev(id);
 854         if (!xsegbd_dev) {
 855                 ret = -ENOENT;
 856                 goto out_unlock;
 857         }
 858
 859         xsegbd_bus_del_dev(xsegbd_dev);
 860
 861 out_unlock:
 862         mutex_unlock(&xsegbd_mutex);
 863         return ret;
 864 }
 865
 866 static struct bus_attribute xsegbd_bus_attrs[] = {
 867         __ATTR(add, S_IWUSR, NULL, xsegbd_add),
 868         __ATTR(remove, S_IWUSR, NULL, xsegbd_remove),
 869         __ATTR_NULL
 870 };
 871
 872 static int xsegbd_sysfs_init(void)
 873 {
 874         int ret;
 875
 876         ret = device_register(&xsegbd_root_dev);
 877         if (ret < 0)
 878                 return ret;
 879
 880         xsegbd_bus_type.bus_attrs = xsegbd_bus_attrs;
 881         ret = bus_register(&xsegbd_bus_type);
 882         if (ret < 0)
 883                 device_unregister(&xsegbd_root_dev);
 884
 885         return ret;
 886 }
 887
 888 static void xsegbd_sysfs_cleanup(void)
 889 {
 890         bus_unregister(&xsegbd_bus_type);
 891         device_unregister(&xsegbd_root_dev);
 892 }
 893
 894 /* *************************** */
 895 /* ** Module Initialization ** */
 896 /* *************************** */
 897
 898 static int __init xsegbd_init(void)
 899 {
 900         int ret = -ENOMEM;
 901
 902         if (!xq_alloc_seq(&blk_queue_pending, max_nr_pending, max_nr_pending))
 903                 goto out;
 904
 905         blk_req_pending = kzalloc(sizeof(struct pending) * max_nr_pending, GFP_KERNEL);
 906         if (!blk_req_pending)
 907                 goto out_queue;
 908
 909         ret = -ENOSYS;
 910         ret = xsegbd_xseg_init();
 911         if (ret)
 912                 goto out_pending;
 913
 914         ret = xsegbd_sysfs_init();
 915         if (ret)
 916                 goto out_xseg;
 917
 918         XSEGLOG("initialization complete");
 919
 920 out:
 921         return ret;
 922
 923 out_xseg:
 924         xsegbd_xseg_quit();
 925 out_pending:
 926         kfree(blk_req_pending);
 927 out_queue:
 928         xq_free(&blk_queue_pending);
 929         goto out;
 930 }
 931
 932 static void __exit xsegbd_exit(void)
 933 {
 934         xsegbd_sysfs_cleanup();
 935         xsegbd_xseg_quit();
 936 }
 937
 938 module_init(xsegbd_init);
 939 module_exit(xsegbd_exit);
 940