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         port = &xsegbd.xseg->ports[xsegbd_dev->src_portno];
 281         port->waitcue = (long) NULL;
 282
 283         xseg_free_requests(xsegbd.xseg, xsegbd_dev->src_portno, xsegbd_dev->nr_requests);
 284
 285         WARN_ON(nr_pending < xsegbd_dev->nr_requests);
 286         spin_lock_irq(&__lock);
 287         nr_pending -= xsegbd_dev->nr_requests;
 288         spin_unlock_irq(&__lock);
 289
 290         unregister_blkdev(xsegbd_dev->major, XSEGBD_NAME);
 291
 292         spin_lock(&xsegbd_dev_list_lock);
 293         list_del_init(&xsegbd_dev->node);
 294         spin_unlock(&xsegbd_dev_list_lock);
 295         kfree(xsegbd_dev);
 296
 297         module_put(THIS_MODULE);
 298 }
 299
 300 /* ******************* */
 301 /* ** Critical Path ** */
 302 /* ******************* */
 303
 304 static void blk_to_xseg(struct xseg *xseg, struct xseg_request *xreq,
 305                         struct request *blkreq)
 306 {
 307         struct bio_vec *bvec;
 308         struct req_iterator iter;
 309         uint64_t off = 0;
 310         char *data = XSEG_TAKE_PTR(xreq->data, xseg->segment);
 311         rq_for_each_segment(bvec, blkreq, iter) {
 312                 char *bdata = kmap_atomic(bvec->bv_page) + bvec->bv_offset;
 313                 memcpy(data + off, bdata, bvec->bv_len);
 314                 off += bvec->bv_len;
 315                 kunmap_atomic(bdata);
 316         }
 317 }
 318
 319 static void xseg_to_blk(struct xseg *xseg, struct xseg_request *xreq,
 320                         struct request *blkreq)
 321 {
 322         struct bio_vec *bvec;
 323         struct req_iterator iter;
 324         uint64_t off = 0;
 325         char *data = XSEG_TAKE_PTR(xreq->data, xseg->segment);
 326         rq_for_each_segment(bvec, blkreq, iter) {
 327                 char *bdata = kmap_atomic(bvec->bv_page) + bvec->bv_offset;
 328                 memcpy(bdata, data + off, bvec->bv_len);
 329                 off += bvec->bv_len;
 330                 kunmap_atomic(bdata);
 331         }
 332 }
 333
 334 static void xseg_request_fn(struct request_queue *rq)
 335 {
 336         struct xseg_request *xreq;
 337         struct xsegbd_device *xsegbd_dev = rq->queuedata;
 338         struct request *blkreq;
 339         struct pending *pending;
 340         xqindex blkreq_idx;
 341         char *name;
 342         uint64_t datasize;
 343
 344         for (;;) {
 345                 xreq = xseg_get_request(xsegbd.xseg, xsegbd_dev->src_portno);
 346                 if (!xreq)
 347                         break;
 348
 349                 blkreq = blk_fetch_request(rq);
 350                 if (!blkreq)
 351                         break;
 352
 353                 if (blkreq->cmd_type != REQ_TYPE_FS) {
 354                         XSEGLOG("non-fs cmd_type: %u. *shrug*", blkreq->cmd_type);
 355                         __blk_end_request_all(blkreq, 0);
 356                 }
 357
 358
 359                 datasize = blk_rq_bytes(blkreq);
 360                 BUG_ON(xreq->buffersize - xsegbd_dev->namesize < datasize);
 361                 BUG_ON(xseg_prep_request(xreq, xsegbd_dev->namesize, datasize));
 362
 363                 name = XSEG_TAKE_PTR(xreq->name, xsegbd.xseg->segment);
 364                 strncpy(name, xsegbd_dev->name, xsegbd_dev->namesize);
 365                 blkreq_idx = xq_pop_head(&blk_queue_pending);
 366                 BUG_ON(blkreq_idx == None);
 367                 pending = &blk_req_pending[blkreq_idx];
 368                 pending->dev = xsegbd_dev;
 369                 pending->request = blkreq;
 370                 pending->comp = NULL;
 371                 xreq->priv = (uint64_t)blkreq_idx;
 372                 xreq->size = datasize;
 373                 xreq->offset = blk_rq_pos(blkreq) << 9;
 374                 /*
 375                 if (xreq->offset >= (sector_size << 9))
 376                         XSEGLOG("sector offset: %lu > %lu, flush:%u, fua:%u",
 377                                  blk_rq_pos(blkreq), sector_size,
 378                                  blkreq->cmd_flags & REQ_FLUSH,
 379                                  blkreq->cmd_flags & REQ_FUA);
 380                 */
 381
 382                 if (blkreq->cmd_flags & REQ_FLUSH)
 383                         xreq->flags |= XF_FLUSH;
 384
 385                 if (blkreq->cmd_flags & REQ_FUA)
 386                         xreq->flags |= XF_FUA;
 387
 388                 if (rq_data_dir(blkreq)) {
 389                         /* unlock for data transfers? */
 390                         blk_to_xseg(xsegbd.xseg, xreq, blkreq);
 391                         xreq->op = X_WRITE;
 392                 } else {
 393                         xreq->op = X_READ;
 394                 }
 395
 396                 BUG_ON(xseg_submit(xsegbd.xseg, xsegbd_dev->dst_portno, xreq) == NoSerial);
 397         }
 398
 399         /* TODO:
 400          * This is going to happen at least once.
 401          * Add a WARN_ON when debugging find out why it happens more than once.
 402          */
 403         xseg_signal(xsegbd_dev->xsegbd->xseg, xsegbd_dev->dst_portno);
 404         if (xreq)
 405                 xseg_put_request(xsegbd_dev->xsegbd->xseg, xsegbd_dev->src_portno, xreq);
 406 }
 407
 408 int update_dev_sectors_from_request(    struct xsegbd_device *xsegbd_dev,
 409                                         struct xseg_request *xreq       )
 410 {
 411         void *data;
 412
 413         if (xreq->state & XS_FAILED)
 414                 return -ENOENT;
 415
 416         if (!(xreq->state & XS_SERVED))
 417                 return -EIO;
 418
 419         data = XSEG_TAKE_PTR(xreq->data, xsegbd.xseg->segment);
 420         xsegbd_dev->sectors = *((uint64_t *) data) / 512ULL;
 421         return 0;
 422 }
 423
 424 static int xsegbd_get_size(struct xsegbd_device *xsegbd_dev)
 425 {
 426         struct xseg_request *xreq;
 427         struct xseg_port *port;
 428         char *name;
 429         uint64_t datasize;
 430         xqindex blkreq_idx;
 431         struct pending *pending;
 432         struct completion comp;
 433         int ret = -EBUSY;
 434
 435         xreq = xseg_get_request(xsegbd.xseg, xsegbd_dev->src_portno);
 436         if (!xreq)
 437                 goto out;
 438
 439         datasize = sizeof(uint64_t);
 440         BUG_ON(xreq->buffersize - xsegbd_dev->namesize < datasize);
 441         BUG_ON(xseg_prep_request(xreq, xsegbd_dev->namesize, datasize));
 442
 443         init_completion(&comp);
 444         blkreq_idx = xq_pop_head(&blk_queue_pending);
 445         BUG_ON(blkreq_idx == None);
 446         pending = &blk_req_pending[blkreq_idx];
 447         pending->dev = xsegbd_dev;
 448         pending->request = NULL;
 449         pending->comp = &comp;
 450         xreq->priv = (uint64_t)blkreq_idx;
 451
 452         name = XSEG_TAKE_PTR(xreq->name, xsegbd.xseg->segment);
 453         strncpy(name, xsegbd_dev->name, xsegbd_dev->namesize);
 454         xreq->size = datasize;
 455         xreq->offset = 0;
 456
 457         xreq->op = X_INFO;
 458
 459         port = &xsegbd.xseg->ports[xsegbd_dev->src_portno];
 460         port->waitcue = (uint64_t)(long)xsegbd_dev;
 461
 462         BUG_ON(xseg_submit(xsegbd.xseg, xsegbd_dev->dst_portno, xreq) == NoSerial);
 463         xseg_signal(xsegbd.xseg, xsegbd_dev->dst_portno);
 464
 465         wait_for_completion_interruptible(&comp);
 466         XSEGLOG("Woken up after wait_for_completion_interruptible()\n");
 467         ret = update_dev_sectors_from_request(xsegbd_dev, xreq);
 468         XSEGLOG("get_size: sectors = %ld\n", (long)xsegbd_dev->sectors);
 469 out:
 470         xseg_put_request(xsegbd.xseg, xsegbd_dev->src_portno, xreq);
 471         return ret;
 472 }
 473
 474 static void xseg_callback(struct xseg *xseg, uint32_t portno)
 475 {
 476         struct xsegbd_device *xsegbd_dev = NULL, *old_dev = NULL;
 477         struct xseg_request *xreq;
 478         struct request *blkreq;
 479         struct pending *pending;
 480         unsigned long flags;
 481         uint32_t blkreq_idx;
 482         int err;
 483
 484         for (;;) {
 485                 xreq = xseg_receive(xseg, portno);
 486                 if (!xreq)
 487                         break;
 488
 489                 /* we rely upon our peers to not have touched ->priv */
 490                 blkreq_idx = (uint64_t)xreq->priv;
 491                 if (blkreq_idx >= max_nr_pending) {
 492                         WARN_ON(1);
 493                         continue;
 494                 }
 495
 496                 pending = &blk_req_pending[blkreq_idx];
 497                 if (pending->comp) {
 498                         /* someone is blocking on this request
 499                            and will handle it when we wake them up. */
 500                         complete(pending->comp);
 501                         /* the request is blocker's responsibility so
 502                            we will not put_request(); */
 503                         continue;
 504                 }
 505
 506                 /* this is now treated as a block I/O request to end */
 507                 blkreq = pending->request;
 508                 pending->request = NULL;
 509                 xsegbd_dev = pending->dev;
 510                 pending->dev = NULL;
 511                 WARN_ON(!blkreq);
 512
 513                 if ((xsegbd_dev != old_dev) && old_dev) {
 514                         spin_lock_irqsave(&old_dev->lock, flags);
 515                         xseg_request_fn(old_dev->blk_queue);
 516                         spin_unlock_irqrestore(&old_dev->lock, flags);
 517                 }
 518
 519                 old_dev = xsegbd_dev;
 520
 521                 if (!(xreq->state & XS_SERVED))
 522                         goto blk_end;
 523
 524                 if (xreq->serviced != blk_rq_bytes(blkreq))
 525                         goto blk_end;
 526
 527                 /* unlock for data transfer? */
 528                 if (!rq_data_dir(blkreq))
 529                         xseg_to_blk(xseg, xreq, blkreq);
 530
 531                 err = 0;
 532 blk_end:
 533                 blk_end_request_all(blkreq, err);
 534                 xq_append_head(&blk_queue_pending, blkreq_idx);
 535                 xseg_put_request(xseg, xreq->portno, xreq);
 536         }
 537
 538         if (xsegbd_dev) {
 539                 spin_lock_irqsave(&xsegbd_dev->lock, flags);
 540                 xseg_request_fn(xsegbd_dev->blk_queue);
 541                 spin_unlock_irqrestore(&xsegbd_dev->lock, flags);
 542         }
 543 }
 544
 545
 546 /* sysfs interface */
 547
 548 static struct bus_type xsegbd_bus_type = {
 549         .name   = "xsegbd",
 550 };
 551
 552 static ssize_t xsegbd_size_show(struct device *dev,
 553                                         struct device_attribute *attr, char *buf)
 554 {
 555         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 556
 557         return sprintf(buf, "%llu\n", (unsigned long long) xsegbd_dev->sectors * 512ULL);
 558 }
 559
 560 static ssize_t xsegbd_major_show(struct device *dev,
 561                                         struct device_attribute *attr, char *buf)
 562 {
 563         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 564
 565         return sprintf(buf, "%d\n", xsegbd_dev->major);
 566 }
 567
 568 static ssize_t xsegbd_srcport_show(struct device *dev,
 569                                         struct device_attribute *attr, char *buf)
 570 {
 571         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 572
 573         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->src_portno);
 574 }
 575
 576 static ssize_t xsegbd_dstport_show(struct device *dev,
 577                                         struct device_attribute *attr, char *buf)
 578 {
 579         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 580
 581         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->dst_portno);
 582 }
 583
 584 static ssize_t xsegbd_id_show(struct device *dev,
 585                                         struct device_attribute *attr, char *buf)
 586 {
 587         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 588
 589         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->id);
 590 }
 591
 592 static ssize_t xsegbd_reqs_show(struct device *dev,
 593                                         struct device_attribute *attr, char *buf)
 594 {
 595         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 596
 597         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->nr_requests);
 598 }
 599
 600 static ssize_t xsegbd_name_show(struct device *dev,
 601                                         struct device_attribute *attr, char *buf)
 602 {
 603         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 604
 605         return sprintf(buf, "%s\n", xsegbd_dev->name);
 606 }
 607
 608 static DEVICE_ATTR(size, S_IRUGO, xsegbd_size_show, NULL);
 609 static DEVICE_ATTR(major, S_IRUGO, xsegbd_major_show, NULL);
 610 static DEVICE_ATTR(srcport, S_IRUGO, xsegbd_srcport_show, NULL);
 611 static DEVICE_ATTR(dstport, S_IRUGO, xsegbd_dstport_show, NULL);
 612 static DEVICE_ATTR(id , S_IRUGO, xsegbd_id_show, NULL);
 613 static DEVICE_ATTR(reqs , S_IRUGO, xsegbd_reqs_show, NULL);
 614 static DEVICE_ATTR(name , S_IRUGO, xsegbd_name_show, NULL);
 615
 616 static struct attribute *xsegbd_attrs[] = {
 617         &dev_attr_size.attr,
 618         &dev_attr_major.attr,
 619         &dev_attr_srcport.attr,
 620         &dev_attr_dstport.attr,
 621         &dev_attr_id.attr,
 622         &dev_attr_reqs.attr,
 623         &dev_attr_name.attr,
 624         NULL
 625 };
 626
 627 static struct attribute_group xsegbd_attr_group = {
 628         .attrs = xsegbd_attrs,
 629 };
 630
 631 static const struct attribute_group *xsegbd_attr_groups[] = {
 632         &xsegbd_attr_group,
 633         NULL
 634 };
 635
 636 static void xsegbd_sysfs_dev_release(struct device *dev)
 637 {
 638 }
 639
 640 static struct device_type xsegbd_device_type = {
 641         .name           = "xsegbd",
 642         .groups         = xsegbd_attr_groups,
 643         .release        = xsegbd_sysfs_dev_release,
 644 };
 645
 646 static void xsegbd_root_dev_release(struct device *dev)
 647 {
 648 }
 649
 650 static struct device xsegbd_root_dev = {
 651         .init_name      = "xsegbd",
 652         .release        = xsegbd_root_dev_release,
 653 };
 654
 655 static int xsegbd_bus_add_dev(struct xsegbd_device *xsegbd_dev)
 656 {
 657         int ret = -ENOMEM;
 658         struct device *dev;
 659
 660         mutex_lock_nested(&xsegbd_mutex, SINGLE_DEPTH_NESTING);
 661         dev = &xsegbd_dev->dev;
 662
 663         dev->bus = &xsegbd_bus_type;
 664         dev->type = &xsegbd_device_type;
 665         dev->parent = &xsegbd_root_dev;
 666         dev->release = xsegbd_dev_release;
 667         dev_set_name(dev, "%d", xsegbd_dev->id);
 668
 669         ret = device_register(dev);
 670
 671         mutex_unlock(&xsegbd_mutex);
 672         return ret;
 673 }
 674
 675 static void xsegbd_bus_del_dev(struct xsegbd_device *xsegbd_dev)
 676 {
 677         device_unregister(&xsegbd_dev->dev);
 678 }
 679
 680 static ssize_t xsegbd_add(struct bus_type *bus, const char *buf, size_t count)
 681 {
 682         struct xsegbd_device *xsegbd_dev;
 683         struct xseg_port *xport;
 684         ssize_t ret = -ENOMEM;
 685         int new_id = 0;
 686         struct list_head *tmp;
 687
 688         if (!try_module_get(THIS_MODULE))
 689                 return -ENODEV;
 690
 691         xsegbd_dev = kzalloc(sizeof(*xsegbd_dev), GFP_KERNEL);
 692         if (!xsegbd_dev)
 693                 goto out;
 694
 695         spin_lock_init(&xsegbd_dev->lock);
 696         INIT_LIST_HEAD(&xsegbd_dev->node);
 697
 698         /* parse cmd */
 699         if (sscanf(buf, "%" __stringify(XSEGBD_TARGET_NAMELEN) "s "
 700                         "%d:%d:%d", xsegbd_dev->name, &xsegbd_dev->src_portno,
 701                         &xsegbd_dev->dst_portno, &xsegbd_dev->nr_requests) < 3) {
 702                 ret = -EINVAL;
 703                 goto out_dev;
 704         }
 705         xsegbd_dev->namesize = strlen(xsegbd_dev->name);
 706
 707         spin_lock(&xsegbd_dev_list_lock);
 708
 709         list_for_each(tmp, &xsegbd_dev_list) {
 710                 struct xsegbd_device *entry;
 711
 712                 entry = list_entry(tmp, struct xsegbd_device, node);
 713
 714                 if (entry->src_portno == xsegbd_dev->src_portno) {
 715                         ret = -EINVAL;
 716                         goto out_unlock;
 717                 }
 718
 719                 if (entry->id >= new_id)
 720                         new_id = entry->id + 1;
 721         }
 722
 723         xsegbd_dev->id = new_id;
 724
 725         list_add_tail(&xsegbd_dev->node, &xsegbd_dev_list);
 726
 727         spin_unlock(&xsegbd_dev_list_lock);
 728
 729         XSEGLOG("registering block device major %d", major);
 730         ret = register_blkdev(major, XSEGBD_NAME);
 731         if (ret < 0) {
 732                 XSEGLOG("cannot register block device!");
 733                 ret = -EBUSY;
 734                 goto out_delentry;
 735         }
 736         xsegbd_dev->major = ret;
 737         XSEGLOG("registered block device major %d", xsegbd_dev->major);
 738
 739         ret = xsegbd_bus_add_dev(xsegbd_dev);
 740         if (ret)
 741                 goto out_blkdev;
 742
 743         XSEGLOG("binding to source port %u (destination %u)",
 744                         xsegbd_dev->src_portno, xsegbd_dev->dst_portno);
 745         xport = xseg_bind_port(xsegbd.xseg, xsegbd_dev->src_portno);
 746         if (!xport) {
 747                 XSEGLOG("cannot bind to port");
 748                 ret = -EFAULT;
 749
 750                 goto out_bus;
 751         }
 752         /* make sure we don't get any requests until we're ready to handle them */
 753         xport->waitcue = (long) NULL;
 754
 755         XSEGLOG("allocating %u requests", xsegbd_dev->nr_requests);
 756         if (xseg_alloc_requests(xsegbd.xseg, xsegbd_dev->src_portno, xsegbd_dev->nr_requests)) {
 757                 XSEGLOG("cannot allocate requests");
 758                 ret = -EFAULT;
 759
 760                 goto out_bus;
 761         }
 762
 763         ret = xsegbd_dev_init(xsegbd_dev);
 764         if (ret)
 765                 goto out_bus;
 766
 767         return count;
 768
 769 out_bus:
 770         xsegbd_bus_del_dev(xsegbd_dev);
 771
 772         return ret;
 773
 774 out_blkdev:
 775         unregister_blkdev(xsegbd_dev->major, XSEGBD_NAME);
 776
 777 out_delentry:
 778         spin_lock(&xsegbd_dev_list_lock);
 779         list_del_init(&xsegbd_dev->node);
 780
 781 out_unlock:
 782         spin_unlock(&xsegbd_dev_list_lock);
 783
 784 out_dev:
 785         kfree(xsegbd_dev);
 786
 787 out:
 788         return ret;
 789 }
 790
 791 static struct xsegbd_device *__xsegbd_get_dev(unsigned long id)
 792 {
 793         struct list_head *tmp;
 794         struct xsegbd_device *xsegbd_dev;
 795
 796
 797         spin_lock(&xsegbd_dev_list_lock);
 798         list_for_each(tmp, &xsegbd_dev_list) {
 799                 xsegbd_dev = list_entry(tmp, struct xsegbd_device, node);
 800                 if (xsegbd_dev->id == id) {
 801                         spin_unlock(&xsegbd_dev_list_lock);
 802                         return xsegbd_dev;
 803                 }
 804         }
 805         spin_unlock(&xsegbd_dev_list_lock);
 806         return NULL;
 807 }
 808
 809 static ssize_t xsegbd_remove(struct bus_type *bus, const char *buf, size_t count)
 810 {
 811         struct xsegbd_device *xsegbd_dev = NULL;
 812         int id, ret;
 813         unsigned long ul_id;
 814
 815         ret = kstrtoul(buf, 10, &ul_id);
 816         if (ret)
 817                 return ret;
 818
 819         id = (int) ul_id;
 820         if (id != ul_id)
 821                 return -EINVAL;
 822
 823         mutex_lock_nested(&xsegbd_mutex, SINGLE_DEPTH_NESTING);
 824
 825         ret = count;
 826         xsegbd_dev = __xsegbd_get_dev(id);
 827         if (!xsegbd_dev) {
 828                 ret = -ENOENT;
 829                 goto out_unlock;
 830         }
 831
 832         xsegbd_bus_del_dev(xsegbd_dev);
 833
 834 out_unlock:
 835         mutex_unlock(&xsegbd_mutex);
 836         return ret;
 837 }
 838
 839 static struct bus_attribute xsegbd_bus_attrs[] = {
 840         __ATTR(add, S_IWUSR, NULL, xsegbd_add),
 841         __ATTR(remove, S_IWUSR, NULL, xsegbd_remove),
 842         __ATTR_NULL
 843 };
 844
 845 static int xsegbd_sysfs_init(void)
 846 {
 847         int ret;
 848
 849         ret = device_register(&xsegbd_root_dev);
 850         if (ret < 0)
 851                 return ret;
 852
 853         xsegbd_bus_type.bus_attrs = xsegbd_bus_attrs;
 854         ret = bus_register(&xsegbd_bus_type);
 855         if (ret < 0)
 856                 device_unregister(&xsegbd_root_dev);
 857
 858         return ret;
 859 }
 860
 861 static void xsegbd_sysfs_cleanup(void)
 862 {
 863         bus_unregister(&xsegbd_bus_type);
 864         device_unregister(&xsegbd_root_dev);
 865 }
 866
 867 /* *************************** */
 868 /* ** Module Initialization ** */
 869 /* *************************** */
 870
 871 static int __init xsegbd_init(void)
 872 {
 873         int ret = -ENOMEM;
 874
 875         if (!xq_alloc_seq(&blk_queue_pending, max_nr_pending, max_nr_pending))
 876                 goto out;
 877
 878         blk_req_pending = kzalloc(sizeof(struct pending) * max_nr_pending, GFP_KERNEL);
 879         if (!blk_req_pending)
 880                 goto out_queue;
 881
 882         ret = -ENOSYS;
 883         ret = xsegbd_xseg_init();
 884         if (ret)
 885                 goto out_pending;
 886
 887         ret = xsegbd_sysfs_init();
 888         if (ret)
 889                 goto out_xseg;
 890
 891         XSEGLOG("initialization complete");
 892
 893 out:
 894         return ret;
 895
 896 out_xseg:
 897         xsegbd_xseg_quit();
 898 out_pending:
 899         kfree(blk_req_pending);
 900 out_queue:
 901         xq_free(&blk_queue_pending);
 902         goto out;
 903 }
 904
 905 static void __exit xsegbd_exit(void)
 906 {
 907         xsegbd_sysfs_cleanup();
 908         xsegbd_xseg_quit();
 909 }
 910
 911 module_init(xsegbd_init);
 912 module_exit(xsegbd_exit);
 913