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
  27 MODULE_DESCRIPTION("xsegbd");
  28 MODULE_AUTHOR("XSEG");
  29 MODULE_LICENSE("GPL");
  30
  31 static long sector_size = 0;
  32 static long blksize = 512;
  33 static int major = 0;
  34 static int max_nr_pending = 1024;
  35 static char name[XSEGBD_SEGMENT_NAMELEN] = "xsegbd";
  36 static char spec[256] = "segdev:xsegbd:4:512:64:1024:12";
  37
  38 module_param(sector_size, long, 0644);
  39 module_param(blksize, long, 0644);
  40 module_param(max_nr_pending, int, 0644);
  41 module_param(major, int, 0644);
  42 module_param_string(name, name, sizeof(name), 0644);
  43 module_param_string(spec, spec, sizeof(spec), 0644);
  44
  45 struct pending {
  46         struct request *request;
  47         struct completion *comp;
  48         struct xsegbd_device *dev;
  49 };
  50
  51 static struct xq blk_queue_pending;
  52 static struct pending *blk_req_pending;
  53 static unsigned int nr_pending;
  54 static spinlock_t __lock;
  55 static struct xsegbd xsegbd;
  56 static DEFINE_MUTEX(xsegbd_mutex);
  57 static LIST_HEAD(xsegbd_dev_list);
  58 static DEFINE_SPINLOCK(xsegbd_dev_list_lock);
  59
  60 /* ************************* */
  61 /* ***** sysfs helpers ***** */
  62 /* ************************* */
  63
  64 static struct xsegbd_device *dev_to_xsegbd(struct device *dev)
  65 {
  66         return container_of(dev, struct xsegbd_device, dev);
  67 }
  68
  69 static struct device *xsegbd_get_dev(struct xsegbd_device *xsegbd_dev)
  70 {
  71         /* FIXME */
  72         return get_device(&xsegbd_dev->dev);
  73 }
  74
  75 static void xsegbd_put_dev(struct xsegbd_device *xsegbd_dev)
  76 {
  77         put_device(&xsegbd_dev->dev);
  78 }
  79
  80 /* ************************* */
  81 /* ** XSEG Initialization ** */
  82 /* ************************* */
  83
  84 static void xseg_callback(struct xseg *xseg, uint32_t portno);
  85
  86 int xsegbd_xseg_init(void)
  87 {
  88         int r;
  89
  90         if (!xsegbd.name[0])
  91                 strncpy(xsegbd.name, name, XSEGBD_SEGMENT_NAMELEN);
  92
  93         r = xseg_initialize();
  94         if (r) {
  95                 XSEGLOG("cannot initialize 'segdev' peer");
  96                 goto err;
  97         }
  98
  99         r = xseg_parse_spec(spec, &xsegbd.config);
 100         if (r)
 101                 goto err;
 102
 103         if (strncmp(xsegbd.config.type, "segdev", 16))
 104                 XSEGLOG("WARNING: unexpected segment type '%s' vs 'segdev'",
 105                          xsegbd.config.type);
 106
 107         XSEGLOG("joining segment");
 108         xsegbd.xseg = xseg_join(        xsegbd.config.type,
 109                                         xsegbd.config.name,
 110                                         "segdev",
 111                                         xseg_callback           );
 112         if (!xsegbd.xseg) {
 113                 XSEGLOG("cannot find segment");
 114                 r = -ENODEV;
 115                 goto err;
 116         }
 117
 118         return 0;
 119 err:
 120         return r;
 121
 122 }
 123
 124 int xsegbd_xseg_quit(void)
 125 {
 126         struct segdev *segdev;
 127
 128         /* make sure to unmap the segment first */
 129         segdev = segdev_get(0);
 130         clear_bit(SEGDEV_RESERVED, &segdev->flags);
 131         xsegbd.xseg->priv->segment_type.ops.unmap(xsegbd.xseg, xsegbd.xseg->segment_size);
 132         segdev_put(segdev);
 133
 134         return 0;
 135 }
 136
 137
 138 /* ***************************** */
 139 /* ** Block Device Operations ** */
 140 /* ***************************** */
 141
 142 static int xsegbd_open(struct block_device *bdev, fmode_t mode)
 143 {
 144         struct gendisk *disk = bdev->bd_disk;
 145         struct xsegbd_device *xsegbd_dev = disk->private_data;
 146
 147         xsegbd_get_dev(xsegbd_dev);
 148
 149         return 0;
 150 }
 151
 152 static int xsegbd_release(struct gendisk *gd, fmode_t mode)
 153 {
 154         struct xsegbd_device *xsegbd_dev = gd->private_data;
 155
 156         xsegbd_put_dev(xsegbd_dev);
 157
 158         return 0;
 159 }
 160
 161 static int xsegbd_ioctl(struct block_device *bdev, fmode_t mode,
 162                         unsigned int cmd, unsigned long arg)
 163 {
 164         return -ENOTTY;
 165 }
 166
 167 static const struct block_device_operations xsegbd_ops = {
 168         .owner          = THIS_MODULE,
 169         .open           = xsegbd_open,
 170         .release        = xsegbd_release,
 171         .ioctl          = xsegbd_ioctl
 172 };
 173
 174
 175 /* *************************** */
 176 /* ** Device Initialization ** */
 177 /* *************************** */
 178
 179 static void xseg_request_fn(struct request_queue *rq);
 180 static int xsegbd_get_size(struct xsegbd_device *xsegbd_dev);
 181
 182 static int xsegbd_dev_init(struct xsegbd_device *xsegbd_dev)
 183 {
 184         int ret = -ENOMEM;
 185         struct gendisk *disk;
 186         unsigned int max_request_size_bytes;
 187
 188         spin_lock_init(&xsegbd_dev->lock);
 189
 190         xsegbd_dev->xsegbd = &xsegbd;
 191
 192         xsegbd_dev->blk_queue = blk_alloc_queue(GFP_KERNEL);
 193         if (!xsegbd_dev->blk_queue)
 194                 goto out;
 195
 196         blk_init_allocated_queue(xsegbd_dev->blk_queue, xseg_request_fn, &xsegbd_dev->lock);
 197         xsegbd_dev->blk_queue->queuedata = xsegbd_dev;
 198
 199         blk_queue_flush(xsegbd_dev->blk_queue, REQ_FLUSH | REQ_FUA);
 200         blk_queue_logical_block_size(xsegbd_dev->blk_queue, 512);
 201         blk_queue_physical_block_size(xsegbd_dev->blk_queue, blksize);
 202         blk_queue_bounce_limit(xsegbd_dev->blk_queue, BLK_BOUNCE_ANY);
 203
 204         //blk_queue_max_segments(dev->blk_queue, 512);
 205         /* calculate maximum block request size
 206          * request size in pages * page_size
 207          * leave one page in buffer for name
 208          */
 209         max_request_size_bytes =
 210                  (unsigned int) (xsegbd.config.request_size - 1) *
 211                                 ( 1 << xsegbd.config.page_shift) ;
 212         blk_queue_max_hw_sectors(xsegbd_dev->blk_queue, max_request_size_bytes >> 9);
 213         blk_queue_max_segment_size(xsegbd_dev->blk_queue, max_request_size_bytes);
 214         blk_queue_io_min(xsegbd_dev->blk_queue, max_request_size_bytes);
 215         blk_queue_io_opt(xsegbd_dev->blk_queue, max_request_size_bytes);
 216
 217         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, xsegbd_dev->blk_queue);
 218
 219         /* vkoukis says we don't need partitions */
 220         xsegbd_dev->gd = disk = alloc_disk(1);
 221         if (!disk)
 222                 goto out_disk;
 223
 224         disk->major = xsegbd_dev->major;
 225         disk->first_minor = 0; // id * XSEGBD_MINORS;
 226         disk->fops = &xsegbd_ops;
 227         disk->queue = xsegbd_dev->blk_queue;
 228         disk->private_data = xsegbd_dev;
 229         disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
 230         snprintf(disk->disk_name, 32, "xsegbd%u", xsegbd_dev->id);
 231
 232         ret = 0;
 233         spin_lock_irq(&__lock);
 234         if (nr_pending + xsegbd_dev->nr_requests > max_nr_pending)
 235                 ret = -ENOBUFS;
 236         else
 237                 nr_pending += xsegbd_dev->nr_requests;
 238         spin_unlock_irq(&__lock);
 239
 240         if (ret)
 241                 goto out_disk;
 242
 243         /* allow a non-zero sector_size parameter to override the disk size */
 244         if (sector_size)
 245                 xsegbd_dev->sectors = sector_size;
 246         else {
 247                 ret = xsegbd_get_size(xsegbd_dev);
 248                 if (ret)
 249                         goto out_disk;
 250         }
 251
 252
 253         set_capacity(disk, xsegbd_dev->sectors);
 254         XSEGLOG("xsegbd active...");
 255         add_disk(disk); /* immediately activates the device */
 256
 257         return 0;
 258
 259 out_disk:
 260         put_disk(disk);
 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         struct xseg_port *port;
 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         /* reset the port's waitcue (aka cancel_wait) */
 280         if ((port = &xsegbd.xseg->ports[xsegbd_dev->src_portno]) != NULL) {
 281                 port->waitcue = (long) NULL;
 282
 283                 if (xseg_free_requests(xsegbd.xseg, xsegbd_dev->src_portno, xsegbd_dev->nr_requests) != 0)
 284                         XSEGLOG("Error trying to free requests!\n");
 285         }
 286
 287         WARN_ON(nr_pending < xsegbd_dev->nr_requests);
 288         spin_lock_irq(&__lock);
 289         nr_pending -= xsegbd_dev->nr_requests;
 290         spin_unlock_irq(&__lock);
 291
 292         unregister_blkdev(xsegbd_dev->major, XSEGBD_NAME);
 293
 294         spin_lock(&xsegbd_dev_list_lock);
 295         list_del_init(&xsegbd_dev->node);
 296         spin_unlock(&xsegbd_dev_list_lock);
 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_TAKE_PTR(xreq->data, xseg->segment);
 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_TAKE_PTR(xreq->data, xseg->segment);
 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 pending *pending;
 342         xqindex blkreq_idx;
 343         char *target;
 344         uint64_t datalen;
 345
 346         for (;;) {
 347                 xreq = xseg_get_request(xsegbd.xseg, xsegbd_dev->src_portno);
 348                 if (!xreq)
 349                         break;
 350
 351                 blkreq = blk_fetch_request(rq);
 352                 if (!blkreq)
 353                         break;
 354
 355                 if (blkreq->cmd_type != REQ_TYPE_FS) {
 356                         XSEGLOG("non-fs cmd_type: %u. *shrug*", blkreq->cmd_type);
 357                         __blk_end_request_all(blkreq, 0);
 358                 }
 359
 360
 361                 datalen = blk_rq_bytes(blkreq);
 362                 BUG_ON(xreq->bufferlen - xsegbd_dev->targetlen < datalen);
 363                 BUG_ON(xseg_prep_request(xreq, xsegbd_dev->targetlen, datalen));
 364
 365                 target = XSEG_TAKE_PTR(xreq->target, xsegbd.xseg->segment);
 366                 strncpy(target, xsegbd_dev->target, xsegbd_dev->targetlen);
 367                 blkreq_idx = xq_pop_head(&blk_queue_pending);
 368                 BUG_ON(blkreq_idx == None);
 369                 pending = &blk_req_pending[blkreq_idx];
 370                 pending->dev = xsegbd_dev;
 371                 pending->request = blkreq;
 372                 pending->comp = NULL;
 373                 xreq->priv = (uint64_t)blkreq_idx;
 374                 xreq->size = datalen;
 375                 xreq->offset = blk_rq_pos(blkreq) << 9;
 376                 /*
 377                 if (xreq->offset >= (sector_size << 9))
 378                         XSEGLOG("sector offset: %lu > %lu, flush:%u, fua:%u",
 379                                  blk_rq_pos(blkreq), sector_size,
 380                                  blkreq->cmd_flags & REQ_FLUSH,
 381                                  blkreq->cmd_flags & REQ_FUA);
 382                 */
 383
 384                 if (blkreq->cmd_flags & REQ_FLUSH)
 385                         xreq->flags |= XF_FLUSH;
 386
 387                 if (blkreq->cmd_flags & REQ_FUA)
 388                         xreq->flags |= XF_FUA;
 389
 390                 if (rq_data_dir(blkreq)) {
 391                         /* unlock for data transfers? */
 392                         blk_to_xseg(xsegbd.xseg, xreq, blkreq);
 393                         xreq->op = X_WRITE;
 394                 } else {
 395                         xreq->op = X_READ;
 396                 }
 397
 398                 BUG_ON(xseg_submit(xsegbd.xseg, xsegbd_dev->dst_portno, xreq) == NoSerial);
 399         }
 400
 401         /* TODO:
 402          * This is going to happen at least once.
 403          * Add a WARN_ON when debugging find out why it happens more than once.
 404          */
 405         xseg_signal(xsegbd_dev->xsegbd->xseg, xsegbd_dev->dst_portno);
 406         if (xreq)
 407                 xseg_put_request(xsegbd_dev->xsegbd->xseg, xsegbd_dev->src_portno, xreq);
 408 }
 409
 410 int update_dev_sectors_from_request(    struct xsegbd_device *xsegbd_dev,
 411                                         struct xseg_request *xreq       )
 412 {
 413         void *data;
 414
 415         if (xreq->state & XS_FAILED)
 416                 return -ENOENT;
 417
 418         if (!(xreq->state & XS_SERVED))
 419                 return -EIO;
 420
 421         data = XSEG_TAKE_PTR(xreq->data, xsegbd.xseg->segment);
 422         xsegbd_dev->sectors = *((uint64_t *) data) / 512ULL;
 423         return 0;
 424 }
 425
 426 static int xsegbd_get_size(struct xsegbd_device *xsegbd_dev)
 427 {
 428         struct xseg_request *xreq;
 429         struct xseg_port *port;
 430         char *target;
 431         uint64_t datalen;
 432         xqindex blkreq_idx;
 433         struct pending *pending;
 434         struct completion comp;
 435         int ret = -EBUSY;
 436
 437         xreq = xseg_get_request(xsegbd.xseg, xsegbd_dev->src_portno);
 438         if (!xreq)
 439                 goto out;
 440
 441         datalen = sizeof(uint64_t);
 442         BUG_ON(xreq->bufferlen - xsegbd_dev->targetlen < datalen);
 443         BUG_ON(xseg_prep_request(xreq, xsegbd_dev->targetlen, datalen));
 444
 445         init_completion(&comp);
 446         blkreq_idx = xq_pop_head(&blk_queue_pending);
 447         BUG_ON(blkreq_idx == None);
 448         pending = &blk_req_pending[blkreq_idx];
 449         pending->dev = xsegbd_dev;
 450         pending->request = NULL;
 451         pending->comp = &comp;
 452         xreq->priv = (uint64_t)blkreq_idx;
 453
 454         target = XSEG_TAKE_PTR(xreq->target, xsegbd.xseg->segment);
 455         strncpy(target, xsegbd_dev->target, xsegbd_dev->targetlen);
 456         xreq->size = datalen;
 457         xreq->offset = 0;
 458
 459         xreq->op = X_INFO;
 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         xseg_signal(xsegbd.xseg, xsegbd_dev->dst_portno);
 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         xseg_put_request(xsegbd.xseg, xsegbd_dev->src_portno, xreq);
 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);
 537                 xseg_put_request(xseg, xreq->portno, xreq);
 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 = kstrtoul(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