code.grnet.gr Git - archipelago/blob - xseg/peers/kernel/xsegbd.c

   1 /* xsegbd.c
   2  *
   3  */
   4
   5 #include <linux/module.h>
   6 #include <linux/moduleparam.h>
   7 #include <linux/init.h>
   8 #include <linux/sched.h>
   9 #include <linux/kernel.h>
  10 #include <linux/slab.h>
  11 #include <linux/fs.h>
  12 #include <linux/errno.h>
  13 #include <linux/timer.h>
  14 #include <linux/types.h>
  15 #include <linux/vmalloc.h>
  16 #include <linux/genhd.h>
  17 #include <linux/blkdev.h>
  18 #include <linux/bio.h>
  19 #include <linux/device.h>
  20 #include <linux/completion.h>
  21
  22 #include <sys/kernel/segdev.h>
  23 #include "xsegbd.h"
  24
  25 #define XSEGBD_MINORS 1
  26 /* define max request size to be used in xsegbd */
  27 //FIXME should we make this 4MB instead of 256KB ?
  28 #define XSEGBD_MAX_REQUEST_SIZE 262144U
  29
  30 MODULE_DESCRIPTION("xsegbd");
  31 MODULE_AUTHOR("XSEG");
  32 MODULE_LICENSE("GPL");
  33
  34 static long sector_size = 0;
  35 static long blksize = 512;
  36 static int major = 0;
  37 static int max_dev = 1024;
  38 static char name[XSEGBD_SEGMENT_NAMELEN] = "xsegbd";
  39 static char spec[256] = "segdev:xsegbd:4:1024:12";
  40
  41 module_param(sector_size, long, 0644);
  42 module_param(blksize, long, 0644);
  43 module_param(max_dev, int, 0644);
  44 module_param(major, int, 0644);
  45 module_param_string(name, name, sizeof(name), 0644);
  46 module_param_string(spec, spec, sizeof(spec), 0644);
  47
  48 //static spinlock_t __lock;
  49 static struct xsegbd xsegbd;
  50 static struct xsegbd_device **xsegbd_devices; /* indexed by portno */
  51 static DEFINE_MUTEX(xsegbd_mutex);
  52 static DEFINE_SPINLOCK(xsegbd_devices_lock);
  53
  54
  55
  56 static struct xsegbd_device *__xsegbd_get_dev(unsigned long id)
  57 {
  58         struct xsegbd_device *xsegbd_dev = NULL;
  59
  60         spin_lock(&xsegbd_devices_lock);
  61         xsegbd_dev = xsegbd_devices[id];
  62         spin_unlock(&xsegbd_devices_lock);
  63
  64         return xsegbd_dev;
  65 }
  66
  67 /* ************************* */
  68 /* ***** sysfs helpers ***** */
  69 /* ************************* */
  70
  71 static struct xsegbd_device *dev_to_xsegbd(struct device *dev)
  72 {
  73         return container_of(dev, struct xsegbd_device, dev);
  74 }
  75
  76 static struct device *xsegbd_get_dev(struct xsegbd_device *xsegbd_dev)
  77 {
  78         /* FIXME */
  79         return get_device(&xsegbd_dev->dev);
  80 }
  81
  82 static void xsegbd_put_dev(struct xsegbd_device *xsegbd_dev)
  83 {
  84         put_device(&xsegbd_dev->dev);
  85 }
  86
  87 /* ************************* */
  88 /* ** XSEG Initialization ** */
  89 /* ************************* */
  90
  91 static void xseg_callback(struct xseg *xseg, uint32_t portno);
  92
  93 int xsegbd_xseg_init(void)
  94 {
  95         int r;
  96
  97         if (!xsegbd.name[0])
  98                 strncpy(xsegbd.name, name, XSEGBD_SEGMENT_NAMELEN);
  99
 100         r = xseg_initialize();
 101         if (r) {
 102                 XSEGLOG("cannot initialize 'segdev' peer");
 103                 goto err;
 104         }
 105
 106         r = xseg_parse_spec(spec, &xsegbd.config);
 107         if (r)
 108                 goto err;
 109
 110         if (strncmp(xsegbd.config.type, "segdev", 16))
 111                 XSEGLOG("WARNING: unexpected segment type '%s' vs 'segdev'",
 112                          xsegbd.config.type);
 113
 114         /* leave it here for now */
 115         XSEGLOG("joining segment");
 116         xsegbd.xseg = xseg_join(        xsegbd.config.type,
 117                                         xsegbd.config.name,
 118                                         "segdev",
 119                                         xseg_callback           );
 120         if (!xsegbd.xseg) {
 121                 XSEGLOG("cannot find segment");
 122                 r = -ENODEV;
 123                 goto err;
 124         }
 125
 126         return 0;
 127 err:
 128         return r;
 129
 130 }
 131
 132 int xsegbd_xseg_quit(void)
 133 {
 134         struct segdev *segdev;
 135
 136         /* make sure to unmap the segment first */
 137         segdev = segdev_get(0);
 138         clear_bit(SEGDEV_RESERVED, &segdev->flags);
 139         xsegbd.xseg->priv->segment_type.ops.unmap(xsegbd.xseg, xsegbd.xseg->segment_size);
 140         segdev_put(segdev);
 141
 142         return 0;
 143 }
 144
 145
 146 /* ***************************** */
 147 /* ** Block Device Operations ** */
 148 /* ***************************** */
 149
 150 static int xsegbd_open(struct block_device *bdev, fmode_t mode)
 151 {
 152         struct gendisk *disk = bdev->bd_disk;
 153         struct xsegbd_device *xsegbd_dev = disk->private_data;
 154
 155         xsegbd_get_dev(xsegbd_dev);
 156
 157         return 0;
 158 }
 159
 160 static int xsegbd_release(struct gendisk *gd, fmode_t mode)
 161 {
 162         struct xsegbd_device *xsegbd_dev = gd->private_data;
 163
 164         xsegbd_put_dev(xsegbd_dev);
 165
 166         return 0;
 167 }
 168
 169 static int xsegbd_ioctl(struct block_device *bdev, fmode_t mode,
 170                         unsigned int cmd, unsigned long arg)
 171 {
 172         return -ENOTTY;
 173 }
 174
 175 static const struct block_device_operations xsegbd_ops = {
 176         .owner          = THIS_MODULE,
 177         .open           = xsegbd_open,
 178         .release        = xsegbd_release,
 179         .ioctl          = xsegbd_ioctl
 180 };
 181
 182
 183 /* *************************** */
 184 /* ** Device Initialization ** */
 185 /* *************************** */
 186
 187 static void xseg_request_fn(struct request_queue *rq);
 188 static int xsegbd_get_size(struct xsegbd_device *xsegbd_dev);
 189
 190 static int xsegbd_dev_init(struct xsegbd_device *xsegbd_dev)
 191 {
 192         int ret = -ENOMEM;
 193         struct gendisk *disk;
 194         unsigned int max_request_size_bytes;
 195
 196         spin_lock_init(&xsegbd_dev->rqlock);
 197
 198         xsegbd_dev->xsegbd = &xsegbd;
 199
 200         xsegbd_dev->blk_queue = blk_alloc_queue(GFP_KERNEL);
 201         if (!xsegbd_dev->blk_queue)
 202                 goto out;
 203
 204         blk_init_allocated_queue(xsegbd_dev->blk_queue, xseg_request_fn, &xsegbd_dev->rqlock);
 205         xsegbd_dev->blk_queue->queuedata = xsegbd_dev;
 206
 207         blk_queue_flush(xsegbd_dev->blk_queue, REQ_FLUSH | REQ_FUA);
 208         blk_queue_logical_block_size(xsegbd_dev->blk_queue, 512);
 209         blk_queue_physical_block_size(xsegbd_dev->blk_queue, blksize);
 210         blk_queue_bounce_limit(xsegbd_dev->blk_queue, BLK_BOUNCE_ANY);
 211
 212         //blk_queue_max_segments(dev->blk_queue, 512);
 213
 214         max_request_size_bytes = XSEGBD_MAX_REQUEST_SIZE;
 215         blk_queue_max_hw_sectors(xsegbd_dev->blk_queue, max_request_size_bytes >> 9);
 216         blk_queue_max_segment_size(xsegbd_dev->blk_queue, max_request_size_bytes);
 217         blk_queue_io_min(xsegbd_dev->blk_queue, max_request_size_bytes);
 218         blk_queue_io_opt(xsegbd_dev->blk_queue, max_request_size_bytes);
 219
 220         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, xsegbd_dev->blk_queue);
 221
 222         /* vkoukis says we don't need partitions */
 223         xsegbd_dev->gd = disk = alloc_disk(1);
 224         if (!disk)
 225                 /* FIXME: We call xsegbd_dev_release if something goes wrong, to cleanup
 226                  * disks/queues/etc.
 227                  * Would it be better to do the cleanup here, and conditionally cleanup
 228                  * in dev_release?
 229                  */
 230                 goto out;
 231
 232         disk->major = xsegbd_dev->major;
 233         disk->first_minor = 0; // id * XSEGBD_MINORS;
 234         disk->fops = &xsegbd_ops;
 235         disk->queue = xsegbd_dev->blk_queue;
 236         disk->private_data = xsegbd_dev;
 237         disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
 238         snprintf(disk->disk_name, 32, "xsegbd%u", xsegbd_dev->id);
 239
 240         ret = 0;
 241
 242         /* allow a non-zero sector_size parameter to override the disk size */
 243         if (sector_size)
 244                 xsegbd_dev->sectors = sector_size;
 245         else {
 246                 ret = xsegbd_get_size(xsegbd_dev);
 247                 if (ret)
 248                         goto out;
 249         }
 250
 251         set_capacity(disk, xsegbd_dev->sectors);
 252         XSEGLOG("xsegbd active...");
 253         add_disk(disk); /* immediately activates the device */
 254
 255         return 0;
 256
 257 out:
 258         return ret;
 259 }
 260
 261 static void xsegbd_dev_release(struct device *dev)
 262 {
 263         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 264
 265         /* cleanup gendisk and blk_queue the right way */
 266         if (xsegbd_dev->gd) {
 267                 if (xsegbd_dev->gd->flags & GENHD_FL_UP)
 268                         del_gendisk(xsegbd_dev->gd);
 269
 270                 blk_cleanup_queue(xsegbd_dev->blk_queue);
 271                 put_disk(xsegbd_dev->gd);
 272         }
 273
 274         /* xsegbd actually does not need to use waiting.
 275          * maybe we can use xseg_cancel_wait for clarity
 276          * with the xseg_segdev kernel driver to convert
 277          * this to a noop
 278          */
 279         xseg_cancel_wait(xsegbd_dev->xseg, xsegbd_dev->src_portno);
 280
 281         if (xseg_free_requests(xsegbd_dev->xseg,
 282                         xsegbd_dev->src_portno, xsegbd_dev->nr_requests) < 0)
 283                 XSEGLOG("Error trying to free requests!\n");
 284
 285
 286         unregister_blkdev(xsegbd_dev->major, XSEGBD_NAME);
 287
 288         spin_lock(&xsegbd_devices_lock);
 289         BUG_ON(xsegbd_devices[xsegbd_dev->src_portno] != xsegbd_dev);
 290         xsegbd_devices[xsegbd_dev->src_portno] = NULL;
 291         spin_unlock(&xsegbd_devices_lock);
 292
 293         if (xsegbd_dev->blk_req_pending)
 294                 kfree(xsegbd_dev->blk_req_pending);
 295         xq_free(&xsegbd_dev->blk_queue_pending);
 296
 297         kfree(xsegbd_dev);
 298
 299         module_put(THIS_MODULE);
 300 }
 301
 302 /* ******************* */
 303 /* ** Critical Path ** */
 304 /* ******************* */
 305
 306 static void blk_to_xseg(struct xseg *xseg, struct xseg_request *xreq,
 307                         struct request *blkreq)
 308 {
 309         struct bio_vec *bvec;
 310         struct req_iterator iter;
 311         uint64_t off = 0;
 312         char *data = xseg_get_data(xseg, xreq);
 313         rq_for_each_segment(bvec, blkreq, iter) {
 314                 char *bdata = kmap_atomic(bvec->bv_page) + bvec->bv_offset;
 315                 memcpy(data + off, bdata, bvec->bv_len);
 316                 off += bvec->bv_len;
 317                 kunmap_atomic(bdata);
 318         }
 319 }
 320
 321 static void xseg_to_blk(struct xseg *xseg, struct xseg_request *xreq,
 322                         struct request *blkreq)
 323 {
 324         struct bio_vec *bvec;
 325         struct req_iterator iter;
 326         uint64_t off = 0;
 327         char *data = xseg_get_data(xseg, xreq);
 328         rq_for_each_segment(bvec, blkreq, iter) {
 329                 char *bdata = kmap_atomic(bvec->bv_page) + bvec->bv_offset;
 330                 memcpy(bdata, data + off, bvec->bv_len);
 331                 off += bvec->bv_len;
 332                 kunmap_atomic(bdata);
 333         }
 334 }
 335
 336 static void xseg_request_fn(struct request_queue *rq)
 337 {
 338         struct xseg_request *xreq;
 339         struct xsegbd_device *xsegbd_dev = rq->queuedata;
 340         struct request *blkreq;
 341         struct xsegbd_pending *pending;
 342         xqindex blkreq_idx;
 343         char *target;
 344         uint64_t datalen;
 345         xport p;
 346         int r;
 347
 348         for (;;) {
 349                 blkreq_idx = Noneidx;
 350                 xreq = xseg_get_request(xsegbd_dev->xseg, xsegbd_dev->src_portno,
 351                                 xsegbd_dev->dst_portno, X_ALLOC);
 352                 if (!xreq)
 353                         break;
 354
 355                 blkreq_idx = xq_pop_head(&xsegbd_dev->blk_queue_pending,
 356                                                 xsegbd_dev->src_portno);
 357                 if (blkreq_idx == Noneidx)
 358                         break;
 359
 360                 blkreq = blk_fetch_request(rq);
 361                 if (!blkreq)
 362                         break;
 363
 364                 if (blkreq->cmd_type != REQ_TYPE_FS) {
 365                         XSEGLOG("non-fs cmd_type: %u. *shrug*", blkreq->cmd_type);
 366                         __blk_end_request_all(blkreq, 0);
 367                 }
 368
 369                 datalen = blk_rq_bytes(blkreq);
 370                 BUG_ON(xseg_prep_request(xsegbd_dev->xseg, xreq,
 371                                         xsegbd_dev->targetlen, datalen));
 372                 BUG_ON(xreq->bufferlen - xsegbd_dev->targetlen < datalen);
 373
 374                 target = xseg_get_target(xsegbd_dev->xseg, xreq);
 375                 strncpy(target, xsegbd_dev->target, xsegbd_dev->targetlen);
 376                 BUG_ON(blkreq_idx == Noneidx);
 377                 pending = &xsegbd_dev->blk_req_pending[blkreq_idx];
 378                 pending->dev = xsegbd_dev;
 379                 pending->request = blkreq;
 380                 pending->comp = NULL;
 381                 xreq->size = datalen;
 382                 xreq->offset = blk_rq_pos(blkreq) << 9;
 383                 /*
 384                 if (xreq->offset >= (sector_size << 9))
 385                         XSEGLOG("sector offset: %lu > %lu, flush:%u, fua:%u",
 386                                  blk_rq_pos(blkreq), sector_size,
 387                                  blkreq->cmd_flags & REQ_FLUSH,
 388                                  blkreq->cmd_flags & REQ_FUA);
 389                 */
 390
 391                 if (blkreq->cmd_flags & REQ_FLUSH)
 392                         xreq->flags |= XF_FLUSH;
 393
 394                 if (blkreq->cmd_flags & REQ_FUA)
 395                         xreq->flags |= XF_FUA;
 396
 397                 XSEGLOG("xreq: %lx size: %llu offset: %llu, blkreq_idx: %llu",
 398                                 xreq, xreq->size, xreq->offset, blkreq_idx);
 399
 400                 if (rq_data_dir(blkreq)) {
 401                         /* unlock for data transfers? */
 402                         blk_to_xseg(xsegbd_dev->xseg, xreq, blkreq);
 403                         XSEGLOG("xreq: %lx size: %llu offset: %llu, blkreq_idx: %llu completed blk_to_xseg",
 404                                 xreq, xreq->size, xreq->offset, blkreq_idx);
 405                         xreq->op = X_WRITE;
 406                 } else {
 407                         xreq->op = X_READ;
 408                 }
 409
 410                 //maybe put this in loop start, and on break,
 411                 //just do xseg_get_req_data
 412                 spin_lock(&xsegbd_dev->reqdatalock);
 413                 r = xseg_set_req_data(xsegbd_dev->xseg, xreq, (void *) blkreq_idx);
 414                 spin_unlock(&xsegbd_dev->reqdatalock);
 415                 BUG_ON(r < 0);
 416                 XSEGLOG("xreq: %lx size: %llu offset: %llu, blkreq_idx: %llu set req data",
 417                                 xreq, xreq->size, xreq->offset, blkreq_idx);
 418
 419                 BUG_ON((p = xseg_submit(xsegbd_dev->xseg, xreq,
 420                                         xsegbd_dev->src_portno, X_ALLOC)) == NoPort);
 421                 XSEGLOG("xreq: %lx size: %llu offset: %llu, blkreq_idx: %llu submitted",
 422                                 xreq, xreq->size, xreq->offset, blkreq_idx);
 423                 WARN_ON(xseg_signal(xsegbd_dev->xsegbd->xseg, p) < 0);
 424         }
 425         if (xreq)
 426                 BUG_ON(xseg_put_request(xsegbd_dev->xsegbd->xseg, xreq,
 427                                         xsegbd_dev->src_portno) == -1);
 428         if (blkreq_idx != Noneidx)
 429                 BUG_ON(xq_append_head(&xsegbd_dev->blk_queue_pending,
 430                                         blkreq_idx, xsegbd_dev->src_portno) == Noneidx);
 431 }
 432
 433 int update_dev_sectors_from_request(    struct xsegbd_device *xsegbd_dev,
 434                                         struct xseg_request *xreq       )
 435 {
 436         void *data;
 437
 438         if (xreq->state & XS_FAILED)
 439                 return -ENOENT;
 440
 441         if (!(xreq->state & XS_SERVED))
 442                 return -EIO;
 443
 444         data = xseg_get_data(xsegbd_dev->xseg, xreq);
 445         xsegbd_dev->sectors = *((uint64_t *) data) / 512ULL;
 446         return 0;
 447 }
 448
 449 static int xsegbd_get_size(struct xsegbd_device *xsegbd_dev)
 450 {
 451         struct xseg_request *xreq;
 452         char *target;
 453         uint64_t datalen;
 454         xqindex blkreq_idx;
 455         struct xsegbd_pending *pending;
 456         struct completion comp;
 457         xport p;
 458         void *data;
 459         int ret = -EBUSY, r;
 460         xreq = xseg_get_request(xsegbd_dev->xseg, xsegbd_dev->src_portno,
 461                         xsegbd_dev->dst_portno, X_ALLOC);
 462         if (!xreq)
 463                 goto out;
 464
 465         datalen = sizeof(uint64_t);
 466         BUG_ON(xseg_prep_request(xsegbd_dev->xseg, xreq, xsegbd_dev->targetlen, datalen));
 467         BUG_ON(xreq->bufferlen - xsegbd_dev->targetlen < datalen);
 468
 469         init_completion(&comp);
 470         blkreq_idx = xq_pop_head(&xsegbd_dev->blk_queue_pending, 1);
 471         if (blkreq_idx == Noneidx)
 472                 goto out;
 473
 474         pending = &xsegbd_dev->blk_req_pending[blkreq_idx];
 475         pending->dev = xsegbd_dev;
 476         pending->request = NULL;
 477         pending->comp = &comp;
 478
 479
 480         spin_lock(&xsegbd_dev->reqdatalock);
 481         r = xseg_set_req_data(xsegbd_dev->xseg, xreq, (void *) blkreq_idx);
 482         spin_unlock(&xsegbd_dev->reqdatalock);
 483         if (r < 0)
 484                 goto out_queue;
 485         XSEGLOG("for req: %lx, set data %llu (lx: %lx)", xreq, blkreq_idx, (void *) blkreq_idx);
 486
 487         target = xseg_get_target(xsegbd_dev->xseg, xreq);
 488         strncpy(target, xsegbd_dev->target, xsegbd_dev->targetlen);
 489         xreq->size = datalen;
 490         xreq->offset = 0;
 491
 492         xreq->op = X_INFO;
 493
 494         /* waiting is not needed.
 495          * but it should be better to use xseg_prepare_wait
 496          * and the xseg_segdev kernel driver, would be a no op
 497          */
 498
 499         xseg_prepare_wait(xsegbd_dev->xseg, xsegbd_dev->src_portno);
 500         p = xseg_submit(xsegbd_dev->xseg, xreq,
 501                                 xsegbd_dev->src_portno, X_ALLOC);
 502         BUG_ON(p == NoPort);
 503         if ( p == NoPort) {
 504                 goto out_data;
 505         }
 506         WARN_ON(xseg_signal(xsegbd_dev->xseg, p) < 0);
 507
 508         wait_for_completion_interruptible(&comp);
 509         XSEGLOG("Woken up after wait_for_completion_interruptible()\n");
 510         ret = update_dev_sectors_from_request(xsegbd_dev, xreq);
 511         XSEGLOG("get_size: sectors = %ld\n", (long)xsegbd_dev->sectors);
 512 out:
 513         BUG_ON(xseg_put_request(xsegbd_dev->xseg, xreq, xsegbd_dev->src_portno) < 0);
 514         return ret;
 515
 516 out_data:
 517         spin_lock(&xsegbd_dev->reqdatalock);
 518         r = xseg_get_req_data(xsegbd_dev->xseg, xreq, &data);
 519         spin_unlock(&xsegbd_dev->reqdatalock);
 520 out_queue:
 521         xq_append_head(&xsegbd_dev->blk_queue_pending, blkreq_idx, 1);
 522
 523         goto out;
 524 }
 525
 526 static void xseg_callback(struct xseg *xseg, xport portno)
 527 {
 528         struct xsegbd_device *xsegbd_dev;
 529         struct xseg_request *xreq;
 530         struct request *blkreq;
 531         struct xsegbd_pending *pending;
 532         unsigned long flags;
 533         xqindex blkreq_idx;
 534         int err;
 535         void *data;
 536
 537         xsegbd_dev  = __xsegbd_get_dev(portno);
 538         if (!xsegbd_dev) {
 539                 WARN_ON(3);
 540                 return;
 541         }
 542
 543         for (;;) {
 544                 xreq = xseg_receive(xsegbd_dev->xseg, portno);
 545                 if (!xreq)
 546                         break;
 547
 548                 spin_lock(&xsegbd_dev->reqdatalock);
 549                 err = xseg_get_req_data(xsegbd_dev->xseg, xreq, &data);
 550                 spin_unlock(&xsegbd_dev->reqdatalock);
 551                 XSEGLOG("for req: %lx, got data %llu (lx %lx)", xreq, (xqindex) data, data);
 552                 if (err < 0) {
 553                         WARN_ON(2);
 554                         //maybe put request?
 555                         continue;
 556                 }
 557
 558                 blkreq_idx = (xqindex) data;
 559                 if (blkreq_idx >= xsegbd_dev->nr_requests) {
 560                         WARN_ON(1);
 561                         //maybe put request?
 562                         continue;
 563                 }
 564
 565                 pending = &xsegbd_dev->blk_req_pending[blkreq_idx];
 566                 if (pending->comp) {
 567                         /* someone is blocking on this request
 568                            and will handle it when we wake them up. */
 569                         complete(pending->comp);
 570                         /* the request is blocker's responsibility so
 571                            we will not put_request(); */
 572                         continue;
 573                 }
 574
 575                 /* this is now treated as a block I/O request to end */
 576                 blkreq = pending->request;
 577                 pending->request = NULL;
 578                 //xsegbd_dev = pending->dev;
 579                 BUG_ON(xsegbd_dev != pending->dev);
 580                 pending->dev = NULL;
 581                 WARN_ON(!blkreq);
 582
 583                 if (!(xreq->state & XS_SERVED))
 584                         goto blk_end;
 585
 586                 if (xreq->serviced != blk_rq_bytes(blkreq))
 587                         goto blk_end;
 588
 589                 /* unlock for data transfer? */
 590                 if (!rq_data_dir(blkreq)){
 591                         xseg_to_blk(xsegbd_dev->xseg, xreq, blkreq);
 592                         XSEGLOG("for req: %lx, completed xseg_to_blk", xreq);
 593                 }
 594
 595                 err = 0;
 596 blk_end:
 597                 blk_end_request_all(blkreq, err);
 598                 XSEGLOG("for req: %lx, completed", xreq);
 599                 xq_append_head(&xsegbd_dev->blk_queue_pending, blkreq_idx, 1);
 600                 BUG_ON(xseg_put_request(xsegbd_dev->xseg, xreq, xsegbd_dev->src_portno) < 0);
 601         }
 602
 603         if (xsegbd_dev) {
 604                 spin_lock_irqsave(&xsegbd_dev->rqlock, flags);
 605                 xseg_request_fn(xsegbd_dev->blk_queue);
 606                 spin_unlock_irqrestore(&xsegbd_dev->rqlock, flags);
 607         }
 608 }
 609
 610
 611 /* sysfs interface */
 612
 613 static struct bus_type xsegbd_bus_type = {
 614         .name   = "xsegbd",
 615 };
 616
 617 static ssize_t xsegbd_size_show(struct device *dev,
 618                                         struct device_attribute *attr, char *buf)
 619 {
 620         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 621
 622         return sprintf(buf, "%llu\n", (unsigned long long) xsegbd_dev->sectors * 512ULL);
 623 }
 624
 625 static ssize_t xsegbd_major_show(struct device *dev,
 626                                         struct device_attribute *attr, char *buf)
 627 {
 628         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 629
 630         return sprintf(buf, "%d\n", xsegbd_dev->major);
 631 }
 632
 633 static ssize_t xsegbd_srcport_show(struct device *dev,
 634                                         struct device_attribute *attr, char *buf)
 635 {
 636         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 637
 638         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->src_portno);
 639 }
 640
 641 static ssize_t xsegbd_dstport_show(struct device *dev,
 642                                         struct device_attribute *attr, char *buf)
 643 {
 644         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 645
 646         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->dst_portno);
 647 }
 648
 649 static ssize_t xsegbd_id_show(struct device *dev,
 650                                         struct device_attribute *attr, char *buf)
 651 {
 652         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 653
 654         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->id);
 655 }
 656
 657 static ssize_t xsegbd_reqs_show(struct device *dev,
 658                                         struct device_attribute *attr, char *buf)
 659 {
 660         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 661
 662         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->nr_requests);
 663 }
 664
 665 static ssize_t xsegbd_target_show(struct device *dev,
 666                                         struct device_attribute *attr, char *buf)
 667 {
 668         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 669
 670         return sprintf(buf, "%s\n", xsegbd_dev->target);
 671 }
 672
 673 static ssize_t xsegbd_image_refresh(struct device *dev,
 674                                         struct device_attribute *attr,
 675                                         const char *buf,
 676                                         size_t size)
 677 {
 678         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 679         int rc, ret = size;
 680
 681         mutex_lock_nested(&xsegbd_mutex, SINGLE_DEPTH_NESTING);
 682
 683         rc = xsegbd_get_size(xsegbd_dev);
 684         if (rc < 0) {
 685                 ret = rc;
 686                 goto out;
 687         }
 688
 689         set_capacity(xsegbd_dev->gd, xsegbd_dev->sectors);
 690
 691 out:
 692         mutex_unlock(&xsegbd_mutex);
 693         return ret;
 694 }
 695
 696 static DEVICE_ATTR(size, S_IRUGO, xsegbd_size_show, NULL);
 697 static DEVICE_ATTR(major, S_IRUGO, xsegbd_major_show, NULL);
 698 static DEVICE_ATTR(srcport, S_IRUGO, xsegbd_srcport_show, NULL);
 699 static DEVICE_ATTR(dstport, S_IRUGO, xsegbd_dstport_show, NULL);
 700 static DEVICE_ATTR(id , S_IRUGO, xsegbd_id_show, NULL);
 701 static DEVICE_ATTR(reqs , S_IRUGO, xsegbd_reqs_show, NULL);
 702 static DEVICE_ATTR(target, S_IRUGO, xsegbd_target_show, NULL);
 703 static DEVICE_ATTR(refresh , S_IWUSR, NULL, xsegbd_image_refresh);
 704
 705 static struct attribute *xsegbd_attrs[] = {
 706         &dev_attr_size.attr,
 707         &dev_attr_major.attr,
 708         &dev_attr_srcport.attr,
 709         &dev_attr_dstport.attr,
 710         &dev_attr_id.attr,
 711         &dev_attr_reqs.attr,
 712         &dev_attr_target.attr,
 713         &dev_attr_refresh.attr,
 714         NULL
 715 };
 716
 717 static struct attribute_group xsegbd_attr_group = {
 718         .attrs = xsegbd_attrs,
 719 };
 720
 721 static const struct attribute_group *xsegbd_attr_groups[] = {
 722         &xsegbd_attr_group,
 723         NULL
 724 };
 725
 726 static void xsegbd_sysfs_dev_release(struct device *dev)
 727 {
 728 }
 729
 730 static struct device_type xsegbd_device_type = {
 731         .name           = "xsegbd",
 732         .groups         = xsegbd_attr_groups,
 733         .release        = xsegbd_sysfs_dev_release,
 734 };
 735
 736 static void xsegbd_root_dev_release(struct device *dev)
 737 {
 738 }
 739
 740 static struct device xsegbd_root_dev = {
 741         .init_name      = "xsegbd",
 742         .release        = xsegbd_root_dev_release,
 743 };
 744
 745 static int xsegbd_bus_add_dev(struct xsegbd_device *xsegbd_dev)
 746 {
 747         int ret = -ENOMEM;
 748         struct device *dev;
 749
 750         mutex_lock_nested(&xsegbd_mutex, SINGLE_DEPTH_NESTING);
 751         dev = &xsegbd_dev->dev;
 752
 753         dev->bus = &xsegbd_bus_type;
 754         dev->type = &xsegbd_device_type;
 755         dev->parent = &xsegbd_root_dev;
 756         dev->release = xsegbd_dev_release;
 757         dev_set_name(dev, "%d", xsegbd_dev->id);
 758
 759         ret = device_register(dev);
 760
 761         mutex_unlock(&xsegbd_mutex);
 762         return ret;
 763 }
 764
 765 static void xsegbd_bus_del_dev(struct xsegbd_device *xsegbd_dev)
 766 {
 767         device_unregister(&xsegbd_dev->dev);
 768 }
 769
 770 static ssize_t xsegbd_add(struct bus_type *bus, const char *buf, size_t count)
 771 {
 772         struct xsegbd_device *xsegbd_dev;
 773         struct xseg_port *port;
 774         ssize_t ret = -ENOMEM;
 775
 776         if (!try_module_get(THIS_MODULE))
 777                 return -ENODEV;
 778
 779         xsegbd_dev = kzalloc(sizeof(*xsegbd_dev), GFP_KERNEL);
 780         if (!xsegbd_dev)
 781                 goto out;
 782
 783         spin_lock_init(&xsegbd_dev->rqlock);
 784         spin_lock_init(&xsegbd_dev->reqdatalock);
 785         INIT_LIST_HEAD(&xsegbd_dev->node);
 786
 787         /* parse cmd */
 788         if (sscanf(buf, "%" __stringify(XSEGBD_TARGET_NAMELEN) "s "
 789                         "%d:%d:%d", xsegbd_dev->target, &xsegbd_dev->src_portno,
 790                         &xsegbd_dev->dst_portno, &xsegbd_dev->nr_requests) < 3) {
 791                 ret = -EINVAL;
 792                 goto out_dev;
 793         }
 794         xsegbd_dev->targetlen = strlen(xsegbd_dev->target);
 795
 796         spin_lock(&xsegbd_devices_lock);
 797         if (xsegbd_devices[xsegbd_dev->src_portno] != NULL) {
 798                 ret = -EINVAL;
 799                 goto out_unlock;
 800         }
 801         xsegbd_devices[xsegbd_dev->src_portno] = xsegbd_dev;
 802         xsegbd_dev->id = xsegbd_dev->src_portno;
 803         spin_unlock(&xsegbd_devices_lock);
 804
 805         XSEGLOG("registering block device major %d", major);
 806         ret = register_blkdev(major, XSEGBD_NAME);
 807         if (ret < 0) {
 808                 XSEGLOG("cannot register block device!");
 809                 ret = -EBUSY;
 810                 goto out_delentry;
 811         }
 812         xsegbd_dev->major = ret;
 813         XSEGLOG("registered block device major %d", xsegbd_dev->major);
 814
 815         ret = xsegbd_bus_add_dev(xsegbd_dev);
 816         if (ret)
 817                 goto out_blkdev;
 818
 819         if (!xq_alloc_seq(&xsegbd_dev->blk_queue_pending,
 820                                 xsegbd_dev->nr_requests,
 821                                 xsegbd_dev->nr_requests))
 822                 goto out_bus;
 823
 824         xsegbd_dev->blk_req_pending = kzalloc(
 825                         xsegbd_dev->nr_requests *sizeof(struct xsegbd_pending),
 826                                    GFP_KERNEL);
 827         if (!xsegbd_dev->blk_req_pending)
 828                 goto out_freeq;
 829
 830
 831         XSEGLOG("joining segment");
 832         //FIXME use xsebd module config for now
 833         xsegbd_dev->xseg = xseg_join(   xsegbd.config.type,
 834                                         xsegbd.config.name,
 835                                         "segdev",
 836                                         xseg_callback           );
 837         if (!xsegbd_dev->xseg)
 838                 goto out_freepending;
 839
 840
 841         XSEGLOG("binding to source port %u (destination %u)",
 842                         xsegbd_dev->src_portno, xsegbd_dev->dst_portno);
 843         port = xseg_bind_port(xsegbd_dev->xseg, xsegbd_dev->src_portno);
 844         if (!port) {
 845                 XSEGLOG("cannot bind to port");
 846                 ret = -EFAULT;
 847
 848                 goto out_xseg;
 849         }
 850         //FIXME rollback here
 851         BUG_ON(xsegbd_dev->src_portno != xseg_portno(xsegbd_dev->xseg, port));
 852
 853         /* make sure we don't get any requests until we're ready to handle them */
 854         xseg_cancel_wait(xsegbd_dev->xseg, xseg_portno(xsegbd_dev->xseg, port));
 855
 856         ret = xsegbd_dev_init(xsegbd_dev);
 857         if (ret)
 858                 goto out_xseg;
 859
 860         return count;
 861
 862 out_xseg:
 863         xseg_leave(xsegbd_dev->xseg);
 864
 865 out_freepending:
 866         kfree(xsegbd_dev->blk_req_pending);
 867
 868 out_freeq:
 869         xq_free(&xsegbd_dev->blk_queue_pending);
 870
 871 out_bus:
 872         xsegbd_bus_del_dev(xsegbd_dev);
 873
 874         return ret;
 875
 876 out_blkdev:
 877         unregister_blkdev(xsegbd_dev->major, XSEGBD_NAME);
 878
 879 out_delentry:
 880         spin_lock(&xsegbd_devices_lock);
 881         xsegbd_devices[xsegbd_dev->src_portno] = NULL;
 882
 883 out_unlock:
 884         spin_unlock(&xsegbd_devices_lock);
 885
 886 out_dev:
 887         kfree(xsegbd_dev);
 888
 889 out:
 890         return ret;
 891 }
 892
 893 static ssize_t xsegbd_remove(struct bus_type *bus, const char *buf, size_t count)
 894 {
 895         struct xsegbd_device *xsegbd_dev = NULL;
 896         int id, ret;
 897         unsigned long ul_id;
 898
 899         ret = strict_strtoul(buf, 10, &ul_id);
 900         if (ret)
 901                 return ret;
 902
 903         id = (int) ul_id;
 904         if (id != ul_id)
 905                 return -EINVAL;
 906
 907         mutex_lock_nested(&xsegbd_mutex, SINGLE_DEPTH_NESTING);
 908
 909         ret = count;
 910         xsegbd_dev = __xsegbd_get_dev(id);
 911         if (!xsegbd_dev) {
 912                 ret = -ENOENT;
 913                 goto out_unlock;
 914         }
 915         xsegbd_bus_del_dev(xsegbd_dev);
 916
 917 out_unlock:
 918         mutex_unlock(&xsegbd_mutex);
 919         return ret;
 920 }
 921
 922 static struct bus_attribute xsegbd_bus_attrs[] = {
 923         __ATTR(add, S_IWUSR, NULL, xsegbd_add),
 924         __ATTR(remove, S_IWUSR, NULL, xsegbd_remove),
 925         __ATTR_NULL
 926 };
 927
 928 static int xsegbd_sysfs_init(void)
 929 {
 930         int ret;
 931
 932         ret = device_register(&xsegbd_root_dev);
 933         if (ret < 0)
 934                 return ret;
 935
 936         xsegbd_bus_type.bus_attrs = xsegbd_bus_attrs;
 937         ret = bus_register(&xsegbd_bus_type);
 938         if (ret < 0)
 939                 device_unregister(&xsegbd_root_dev);
 940
 941         return ret;
 942 }
 943
 944 static void xsegbd_sysfs_cleanup(void)
 945 {
 946         bus_unregister(&xsegbd_bus_type);
 947         device_unregister(&xsegbd_root_dev);
 948 }
 949
 950 /* *************************** */
 951 /* ** Module Initialization ** */
 952 /* *************************** */
 953
 954 static int __init xsegbd_init(void)
 955 {
 956         int ret = -ENOMEM;
 957         xsegbd_devices = kzalloc(max_dev * sizeof(struct xsegbd_devices *), GFP_KERNEL);
 958         if (!xsegbd_devices)
 959                 goto out;
 960
 961         spin_lock_init(&xsegbd_devices_lock);
 962
 963         ret = -ENOSYS;
 964         ret = xsegbd_xseg_init();
 965         if (ret)
 966                 goto out_free;
 967
 968         ret = xsegbd_sysfs_init();
 969         if (ret)
 970                 goto out_xseg;
 971
 972         XSEGLOG("initialization complete");
 973
 974 out:
 975         return ret;
 976
 977 out_xseg:
 978         xsegbd_xseg_quit();
 979
 980 out_free:
 981         kfree(xsegbd_devices);
 982
 983         goto out;
 984 }
 985
 986 static void __exit xsegbd_exit(void)
 987 {
 988         xsegbd_sysfs_cleanup();
 989         xsegbd_xseg_quit();
 990 }
 991
 992 module_init(xsegbd_init);
 993 module_exit(xsegbd_exit);
 994