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 struct xsegbd xsegbd;
  49 static struct xsegbd_device **xsegbd_devices; /* indexed by portno */
  50 static DEFINE_MUTEX(xsegbd_mutex);
  51 static DEFINE_SPINLOCK(xsegbd_devices_lock);
  52
  53
  54
  55 static struct xsegbd_device *__xsegbd_get_dev(unsigned long id)
  56 {
  57         struct xsegbd_device *xsegbd_dev = NULL;
  58
  59         spin_lock(&xsegbd_devices_lock);
  60         xsegbd_dev = xsegbd_devices[id];
  61         spin_unlock(&xsegbd_devices_lock);
  62
  63         return xsegbd_dev;
  64 }
  65
  66 /* ************************* */
  67 /* ***** sysfs helpers ***** */
  68 /* ************************* */
  69
  70 static struct xsegbd_device *dev_to_xsegbd(struct device *dev)
  71 {
  72         return container_of(dev, struct xsegbd_device, dev);
  73 }
  74
  75 static struct device *xsegbd_get_dev(struct xsegbd_device *xsegbd_dev)
  76 {
  77         /* FIXME */
  78         return get_device(&xsegbd_dev->dev);
  79 }
  80
  81 static void xsegbd_put_dev(struct xsegbd_device *xsegbd_dev)
  82 {
  83         put_device(&xsegbd_dev->dev);
  84 }
  85
  86 /* ************************* */
  87 /* ** XSEG Initialization ** */
  88 /* ************************* */
  89
  90 static void xseg_callback(uint32_t portno);
  91
  92 int xsegbd_xseg_init(void)
  93 {
  94         int r;
  95
  96         if (!xsegbd.name[0])
  97                 strncpy(xsegbd.name, name, XSEGBD_SEGMENT_NAMELEN);
  98
  99         r = xseg_initialize();
 100         if (r) {
 101                 XSEGLOG("cannot initialize 'segdev' peer");
 102                 goto err;
 103         }
 104
 105         r = xseg_parse_spec(spec, &xsegbd.config);
 106         if (r)
 107                 goto err;
 108
 109         if (strncmp(xsegbd.config.type, "segdev", 16))
 110                 XSEGLOG("WARNING: unexpected segment type '%s' vs 'segdev'",
 111                          xsegbd.config.type);
 112
 113         /* leave it here for now */
 114         XSEGLOG("joining segment");
 115         xsegbd.xseg = xseg_join(        xsegbd.config.type,
 116                                         xsegbd.config.name,
 117                                         "segdev",
 118                                         xseg_callback           );
 119         if (!xsegbd.xseg) {
 120                 XSEGLOG("cannot find segment");
 121                 r = -ENODEV;
 122                 goto err;
 123         }
 124
 125         return 0;
 126 err:
 127         return r;
 128
 129 }
 130
 131 int xsegbd_xseg_quit(void)
 132 {
 133         struct segdev *segdev;
 134
 135         /* make sure to unmap the segment first */
 136         segdev = segdev_get(0);
 137         clear_bit(SEGDEV_RESERVED, &segdev->flags);
 138         xsegbd.xseg->priv->segment_type.ops.unmap(xsegbd.xseg, xsegbd.xseg->segment_size);
 139         segdev_put(segdev);
 140
 141         return 0;
 142 }
 143
 144
 145 /* ***************************** */
 146 /* ** Block Device Operations ** */
 147 /* ***************************** */
 148
 149 static int xsegbd_open(struct block_device *bdev, fmode_t mode)
 150 {
 151         struct gendisk *disk = bdev->bd_disk;
 152         struct xsegbd_device *xsegbd_dev = disk->private_data;
 153
 154         xsegbd_get_dev(xsegbd_dev);
 155
 156         return 0;
 157 }
 158
 159 static int xsegbd_release(struct gendisk *gd, fmode_t mode)
 160 {
 161         struct xsegbd_device *xsegbd_dev = gd->private_data;
 162
 163         xsegbd_put_dev(xsegbd_dev);
 164
 165         return 0;
 166 }
 167
 168 static int xsegbd_ioctl(struct block_device *bdev, fmode_t mode,
 169                         unsigned int cmd, unsigned long arg)
 170 {
 171         return -ENOTTY;
 172 }
 173
 174 static const struct block_device_operations xsegbd_ops = {
 175         .owner          = THIS_MODULE,
 176         .open           = xsegbd_open,
 177         .release        = xsegbd_release,
 178         .ioctl          = xsegbd_ioctl
 179 };
 180
 181
 182 /* *************************** */
 183 /* ** Device Initialization ** */
 184 /* *************************** */
 185
 186 static void xseg_request_fn(struct request_queue *rq);
 187 static int xsegbd_get_size(struct xsegbd_device *xsegbd_dev);
 188
 189 static int xsegbd_dev_init(struct xsegbd_device *xsegbd_dev)
 190 {
 191         int ret = -ENOMEM;
 192         struct gendisk *disk;
 193         unsigned int max_request_size_bytes;
 194
 195         spin_lock_init(&xsegbd_dev->rqlock);
 196
 197         xsegbd_dev->xsegbd = &xsegbd;
 198
 199         xsegbd_dev->blk_queue = blk_alloc_queue(GFP_KERNEL);
 200         if (!xsegbd_dev->blk_queue)
 201                 goto out;
 202
 203         if (!blk_init_allocated_queue(xsegbd_dev->blk_queue,
 204                         xseg_request_fn, &xsegbd_dev->rqlock))
 205                 goto outqueue;
 206
 207         xsegbd_dev->blk_queue->queuedata = xsegbd_dev;
 208
 209         blk_queue_flush(xsegbd_dev->blk_queue, REQ_FLUSH | REQ_FUA);
 210         blk_queue_logical_block_size(xsegbd_dev->blk_queue, 512);
 211         blk_queue_physical_block_size(xsegbd_dev->blk_queue, blksize);
 212         blk_queue_bounce_limit(xsegbd_dev->blk_queue, BLK_BOUNCE_ANY);
 213
 214         //blk_queue_max_segments(dev->blk_queue, 512);
 215
 216         max_request_size_bytes = XSEGBD_MAX_REQUEST_SIZE;
 217         blk_queue_max_hw_sectors(xsegbd_dev->blk_queue, max_request_size_bytes >> 9);
 218         blk_queue_max_segment_size(xsegbd_dev->blk_queue, max_request_size_bytes);
 219         blk_queue_io_min(xsegbd_dev->blk_queue, max_request_size_bytes);
 220         blk_queue_io_opt(xsegbd_dev->blk_queue, max_request_size_bytes);
 221
 222         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, xsegbd_dev->blk_queue);
 223
 224         /* vkoukis says we don't need partitions */
 225         xsegbd_dev->gd = disk = alloc_disk(1);
 226         if (!disk)
 227                 goto outqueue;
 228
 229         disk->major = xsegbd_dev->major;
 230         disk->first_minor = 0; // id * XSEGBD_MINORS;
 231         disk->fops = &xsegbd_ops;
 232         disk->queue = xsegbd_dev->blk_queue;
 233         disk->private_data = xsegbd_dev;
 234         disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
 235         snprintf(disk->disk_name, 32, "xsegbd%u", xsegbd_dev->id);
 236
 237         ret = 0;
 238
 239         /* allow a non-zero sector_size parameter to override the disk size */
 240         if (sector_size)
 241                 xsegbd_dev->sectors = sector_size;
 242         else {
 243                 ret = xsegbd_get_size(xsegbd_dev);
 244                 if (ret)
 245                         goto outdisk;
 246         }
 247
 248         set_capacity(disk, xsegbd_dev->sectors);
 249         XSEGLOG("xsegbd active...");
 250         add_disk(disk); /* immediately activates the device */
 251
 252         return 0;
 253
 254
 255 outdisk:
 256         put_disk(xsegbd_dev->gd);
 257 outqueue:
 258         blk_cleanup_queue(xsegbd_dev->blk_queue);
 259 out:
 260         xsegbd_dev->gd = NULL;
 261         return ret;
 262 }
 263
 264 static void xsegbd_dev_release(struct device *dev)
 265 {
 266         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 267
 268         xseg_cancel_wait(xsegbd_dev->xseg, xsegbd_dev->src_portno);
 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 //      if (xseg_free_requests(xsegbd_dev->xseg,
 280 //                      xsegbd_dev->src_portno, xsegbd_dev->nr_requests) < 0)
 281 //              XSEGLOG("Error trying to free requests!\n");
 282
 283
 284         unregister_blkdev(xsegbd_dev->major, XSEGBD_NAME);
 285
 286         spin_lock(&xsegbd_devices_lock);
 287         BUG_ON(xsegbd_devices[xsegbd_dev->src_portno] != xsegbd_dev);
 288         xsegbd_devices[xsegbd_dev->src_portno] = NULL;
 289         spin_unlock(&xsegbd_devices_lock);
 290
 291         if (xsegbd_dev->blk_req_pending)
 292                 kfree(xsegbd_dev->blk_req_pending);
 293         xq_free(&xsegbd_dev->blk_queue_pending);
 294
 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_get_data(xseg, xreq);
 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_get_data(xseg, xreq);
 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 xsegbd_pending *pending;
 340         xqindex blkreq_idx;
 341         char *target;
 342         uint64_t datalen;
 343         xport p;
 344         int r;
 345
 346         for (;;) {
 347                 blkreq_idx = Noneidx;
 348                 xreq = xseg_get_request(xsegbd_dev->xseg, xsegbd_dev->src_portno,
 349                                 xsegbd_dev->dst_portno, X_ALLOC);
 350                 if (!xreq)
 351                         break;
 352
 353                 blkreq_idx = xq_pop_head(&xsegbd_dev->blk_queue_pending,
 354                                                 xsegbd_dev->src_portno);
 355                 if (blkreq_idx == Noneidx)
 356                         break;
 357
 358                 if (blkreq_idx >= xsegbd_dev->nr_requests) {
 359                         XSEGLOG("blkreq_idx >= xsegbd_dev->nr_requests");
 360                         BUG_ON(1);
 361                         break;
 362                 }
 363
 364                 blkreq = blk_fetch_request(rq);
 365                 if (!blkreq)
 366                         break;
 367
 368                 if (blkreq->cmd_type != REQ_TYPE_FS) {
 369                         //we lose xreq here
 370                         XSEGLOG("non-fs cmd_type: %u. *shrug*", blkreq->cmd_type);
 371                         __blk_end_request_all(blkreq, 0);
 372                         continue;
 373                 }
 374
 375                 datalen = blk_rq_bytes(blkreq);
 376                 r = xseg_prep_request(xsegbd_dev->xseg, xreq,
 377                                         xsegbd_dev->targetlen, datalen);
 378                 if (r < 0) {
 379                         XSEGLOG("couldn't prep request");
 380                         __blk_end_request_err(blkreq, r);
 381                         BUG_ON(1);
 382                         break;
 383                 }
 384                 r = -ENOMEM;
 385                 if (xreq->bufferlen - xsegbd_dev->targetlen < datalen){
 386                         XSEGLOG("malformed req buffers");
 387                         __blk_end_request_err(blkreq, r);
 388                         BUG_ON(1);
 389                         break;
 390                 }
 391
 392                 target = xseg_get_target(xsegbd_dev->xseg, xreq);
 393                 strncpy(target, xsegbd_dev->target, xsegbd_dev->targetlen);
 394
 395                 pending = &xsegbd_dev->blk_req_pending[blkreq_idx];
 396                 pending->dev = xsegbd_dev;
 397                 pending->request = blkreq;
 398                 pending->comp = NULL;
 399
 400                 xreq->size = datalen;
 401                 xreq->offset = blk_rq_pos(blkreq) << 9;
 402                 xreq->priv = (uint64_t) blkreq_idx;
 403
 404                 /*
 405                 if (xreq->offset >= (sector_size << 9))
 406                         XSEGLOG("sector offset: %lu > %lu, flush:%u, fua:%u",
 407                                  blk_rq_pos(blkreq), sector_size,
 408                                  blkreq->cmd_flags & REQ_FLUSH,
 409                                  blkreq->cmd_flags & REQ_FUA);
 410                 */
 411
 412                 if (blkreq->cmd_flags & REQ_FLUSH)
 413                         xreq->flags |= XF_FLUSH;
 414
 415                 if (blkreq->cmd_flags & REQ_FUA)
 416                         xreq->flags |= XF_FUA;
 417
 418                 if (rq_data_dir(blkreq)) {
 419                         /* unlock for data transfers? */
 420                         blk_to_xseg(xsegbd_dev->xseg, xreq, blkreq);
 421                         xreq->op = X_WRITE;
 422                 } else {
 423                         xreq->op = X_READ;
 424                 }
 425
 426
 427                 r = -EIO;
 428                 p = xseg_submit(xsegbd_dev->xseg, xreq,
 429                                         xsegbd_dev->src_portno, X_ALLOC);
 430                 if (p == NoPort) {
 431                         XSEGLOG("coundn't submit req");
 432                         BUG_ON(1);
 433                         __blk_end_request_err(blkreq, r);
 434                         break;
 435                 }
 436                 WARN_ON(xseg_signal(xsegbd_dev->xsegbd->xseg, p) < 0);
 437         }
 438         if (xreq)
 439                 BUG_ON(xseg_put_request(xsegbd_dev->xsegbd->xseg, xreq,
 440                                         xsegbd_dev->src_portno) == -1);
 441         if (blkreq_idx != Noneidx)
 442                 BUG_ON(xq_append_head(&xsegbd_dev->blk_queue_pending,
 443                                 blkreq_idx, xsegbd_dev->src_portno) == Noneidx);
 444 }
 445
 446 int update_dev_sectors_from_request(    struct xsegbd_device *xsegbd_dev,
 447                                         struct xseg_request *xreq       )
 448 {
 449         void *data;
 450         if (!xreq) {
 451                 XSEGLOG("Invalid xreq");
 452                 return -EIO;
 453         }
 454
 455         if (xreq->state & XS_FAILED)
 456                 return -ENOENT;
 457
 458         if (!(xreq->state & XS_SERVED))
 459                 return -EIO;
 460
 461         data = xseg_get_data(xsegbd_dev->xseg, xreq);
 462         if (!data) {
 463                 XSEGLOG("Invalid req data");
 464                 return -EIO;
 465         }
 466         xsegbd_dev->sectors = *((uint64_t *) data) / 512ULL;
 467         return 0;
 468 }
 469
 470 static int xsegbd_get_size(struct xsegbd_device *xsegbd_dev)
 471 {
 472         struct xseg_request *xreq;
 473         char *target;
 474         uint64_t datalen;
 475         xqindex blkreq_idx;
 476         struct xsegbd_pending *pending;
 477         struct completion comp;
 478         xport p;
 479         void *data;
 480         int ret = -EBUSY, r;
 481         xreq = xseg_get_request(xsegbd_dev->xseg, xsegbd_dev->src_portno,
 482                         xsegbd_dev->dst_portno, X_ALLOC);
 483         if (!xreq)
 484                 goto out;
 485
 486         datalen = sizeof(uint64_t);
 487         BUG_ON(xseg_prep_request(xsegbd_dev->xseg, xreq, xsegbd_dev->targetlen, datalen));
 488         BUG_ON(xreq->bufferlen - xsegbd_dev->targetlen < datalen);
 489
 490         init_completion(&comp);
 491         blkreq_idx = xq_pop_head(&xsegbd_dev->blk_queue_pending, 1);
 492         if (blkreq_idx == Noneidx)
 493                 goto out;
 494
 495         pending = &xsegbd_dev->blk_req_pending[blkreq_idx];
 496         pending->dev = xsegbd_dev;
 497         pending->request = NULL;
 498         pending->comp = &comp;
 499
 500
 501         xreq->priv = (uint64_t) blkreq_idx;
 502
 503         target = xseg_get_target(xsegbd_dev->xseg, xreq);
 504         strncpy(target, xsegbd_dev->target, xsegbd_dev->targetlen);
 505         xreq->size = datalen;
 506         xreq->offset = 0;
 507         xreq->op = X_INFO;
 508
 509         xseg_prepare_wait(xsegbd_dev->xseg, xsegbd_dev->src_portno);
 510         p = xseg_submit(xsegbd_dev->xseg, xreq,
 511                                 xsegbd_dev->src_portno, X_ALLOC);
 512         if ( p == NoPort) {
 513                 XSEGLOG("couldn't submit request");
 514                 BUG_ON(1);
 515                 goto out_queue;
 516         }
 517         WARN_ON(xseg_signal(xsegbd_dev->xseg, p) < 0);
 518         XSEGLOG("Before wait for completion, xreq %lx", (unsigned long) xreq);
 519         wait_for_completion_interruptible(&comp);
 520         XSEGLOG("Woken up after wait_for_completion_interruptible(), xreq: %lx", (unsigned long) xreq);
 521         ret = update_dev_sectors_from_request(xsegbd_dev, xreq);
 522         //XSEGLOG("get_size: sectors = %ld\n", (long)xsegbd_dev->sectors);
 523 out:
 524         BUG_ON(xseg_put_request(xsegbd_dev->xseg, xreq, xsegbd_dev->src_portno) == -1);
 525         return ret;
 526
 527 out_queue:
 528         xq_append_head(&xsegbd_dev->blk_queue_pending, blkreq_idx, 1);
 529
 530         goto out;
 531 }
 532
 533 static void xseg_callback(xport portno)
 534 {
 535         struct xsegbd_device *xsegbd_dev;
 536         struct xseg_request *xreq;
 537         struct request *blkreq;
 538         struct xsegbd_pending *pending;
 539         unsigned long flags;
 540         xqindex blkreq_idx, ridx;
 541         int err;
 542         void *data;
 543
 544         xsegbd_dev  = __xsegbd_get_dev(portno);
 545         if (!xsegbd_dev) {
 546                 XSEGLOG("portno: %u has no xsegbd device assigned", portno);
 547                 WARN_ON(1);
 548                 return;
 549         }
 550
 551         for (;;) {
 552                 xseg_prepare_wait(xsegbd_dev->xseg, xsegbd_dev->src_portno);
 553                 xreq = xseg_receive(xsegbd_dev->xseg, portno);
 554                 if (!xreq)
 555                         break;
 556
 557                 xseg_cancel_wait(xsegbd_dev->xseg, xsegbd_dev->src_portno);
 558
 559                 blkreq_idx = (xqindex) xreq->priv;
 560                 if (blkreq_idx >= xsegbd_dev->nr_requests) {
 561                         WARN_ON(1);
 562                         //FIXME maybe put request?
 563                         continue;
 564                 }
 565
 566                 pending = &xsegbd_dev->blk_req_pending[blkreq_idx];
 567                 if (pending->comp) {
 568                         /* someone is blocking on this request
 569                            and will handle it when we wake them up. */
 570                         complete(pending->comp);
 571                         /* the request is blocker's responsibility so
 572                            we will not put_request(); */
 573                         continue;
 574                 }
 575
 576                 /* this is now treated as a block I/O request to end */
 577                 blkreq = pending->request;
 578                 pending->request = NULL;
 579                 if (xsegbd_dev != pending->dev) {
 580                         //FIXME maybe put request?
 581                         XSEGLOG("xsegbd_dev != pending->dev");
 582                         BUG_ON(1);
 583                         continue;
 584                 }
 585                 pending->dev = NULL;
 586                 if (!blkreq){
 587                         //FIXME maybe put request?
 588                         XSEGLOG("blkreq does not exist");
 589                         BUG_ON(1);
 590                         continue;
 591                 }
 592
 593                 err = -EIO;
 594                 if (!(xreq->state & XS_SERVED))
 595                         goto blk_end;
 596
 597                 if (xreq->serviced != blk_rq_bytes(blkreq))
 598                         goto blk_end;
 599
 600                 err = 0;
 601                 /* unlock for data transfer? */
 602                 if (!rq_data_dir(blkreq)){
 603                         xseg_to_blk(xsegbd_dev->xseg, xreq, blkreq);
 604                 }
 605 blk_end:
 606                 blk_end_request_all(blkreq, err);
 607
 608                 ridx = xq_append_head(&xsegbd_dev->blk_queue_pending,
 609                                         blkreq_idx, xsegbd_dev->src_portno);
 610                 if (ridx == Noneidx) {
 611                         XSEGLOG("couldnt append blkreq_idx");
 612                         WARN_ON(1);
 613                 }
 614
 615                 if (xseg_put_request(xsegbd_dev->xseg, xreq,
 616                                                 xsegbd_dev->src_portno) < 0){
 617                         XSEGLOG("couldn't put req");
 618                         BUG_ON(1);
 619                 }
 620         }
 621
 622         if (xsegbd_dev) {
 623                 spin_lock_irqsave(&xsegbd_dev->rqlock, flags);
 624                 xseg_request_fn(xsegbd_dev->blk_queue);
 625                 spin_unlock_irqrestore(&xsegbd_dev->rqlock, flags);
 626         }
 627 }
 628
 629
 630 /* sysfs interface */
 631
 632 static struct bus_type xsegbd_bus_type = {
 633         .name   = "xsegbd",
 634 };
 635
 636 static ssize_t xsegbd_size_show(struct device *dev,
 637                                         struct device_attribute *attr, char *buf)
 638 {
 639         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 640
 641         return sprintf(buf, "%llu\n", (unsigned long long) xsegbd_dev->sectors * 512ULL);
 642 }
 643
 644 static ssize_t xsegbd_major_show(struct device *dev,
 645                                         struct device_attribute *attr, char *buf)
 646 {
 647         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 648
 649         return sprintf(buf, "%d\n", xsegbd_dev->major);
 650 }
 651
 652 static ssize_t xsegbd_srcport_show(struct device *dev,
 653                                         struct device_attribute *attr, char *buf)
 654 {
 655         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 656
 657         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->src_portno);
 658 }
 659
 660 static ssize_t xsegbd_dstport_show(struct device *dev,
 661                                         struct device_attribute *attr, char *buf)
 662 {
 663         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 664
 665         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->dst_portno);
 666 }
 667
 668 static ssize_t xsegbd_id_show(struct device *dev,
 669                                         struct device_attribute *attr, char *buf)
 670 {
 671         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 672
 673         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->id);
 674 }
 675
 676 static ssize_t xsegbd_reqs_show(struct device *dev,
 677                                         struct device_attribute *attr, char *buf)
 678 {
 679         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 680
 681         return sprintf(buf, "%u\n", (unsigned) xsegbd_dev->nr_requests);
 682 }
 683
 684 static ssize_t xsegbd_target_show(struct device *dev,
 685                                         struct device_attribute *attr, char *buf)
 686 {
 687         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 688
 689         return sprintf(buf, "%s\n", xsegbd_dev->target);
 690 }
 691
 692 static ssize_t xsegbd_image_refresh(struct device *dev,
 693                                         struct device_attribute *attr,
 694                                         const char *buf,
 695                                         size_t size)
 696 {
 697         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 698         int rc, ret = size;
 699
 700         mutex_lock_nested(&xsegbd_mutex, SINGLE_DEPTH_NESTING);
 701
 702         rc = xsegbd_get_size(xsegbd_dev);
 703         if (rc < 0) {
 704                 ret = rc;
 705                 goto out;
 706         }
 707
 708         set_capacity(xsegbd_dev->gd, xsegbd_dev->sectors);
 709
 710 out:
 711         mutex_unlock(&xsegbd_mutex);
 712         return ret;
 713 }
 714
 715 static ssize_t xsegbd_cleanup(struct device *dev,
 716                                         struct device_attribute *attr,
 717                                         const char *buf,
 718                                         size_t size)
 719 {
 720         struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
 721         int ret = size, i;
 722         struct request *blkreq = NULL;
 723         struct xsegbd_pending *pending = NULL;
 724         struct completion *comp = NULL;
 725
 726         mutex_lock_nested(&xsegbd_mutex, SINGLE_DEPTH_NESTING);
 727         for (i = 0; i < xsegbd_dev->nr_requests; i++) {
 728                 xlock_acquire(&xsegbd_dev->blk_queue_pending.lock,
 729                                 xsegbd_dev->src_portno);
 730                 if (!__xq_check(&xsegbd_dev->blk_queue_pending, i)) {
 731                         pending = &xsegbd_dev->blk_req_pending[i];
 732                         blkreq = pending->request;
 733                         pending->request = NULL;
 734                         comp = pending->comp;
 735                         pending->comp = NULL;
 736                         if (blkreq)
 737                                 blk_end_request_all(blkreq, -EIO);
 738                         if (comp)
 739                                 complete(comp);
 740                 }
 741                 xlock_release(&xsegbd_dev->blk_queue_pending.lock);
 742         }
 743
 744         mutex_unlock(&xsegbd_mutex);
 745         return ret;
 746 }
 747
 748 static DEVICE_ATTR(size, S_IRUGO, xsegbd_size_show, NULL);
 749 static DEVICE_ATTR(major, S_IRUGO, xsegbd_major_show, NULL);
 750 static DEVICE_ATTR(srcport, S_IRUGO, xsegbd_srcport_show, NULL);
 751 static DEVICE_ATTR(dstport, S_IRUGO, xsegbd_dstport_show, NULL);
 752 static DEVICE_ATTR(id , S_IRUGO, xsegbd_id_show, NULL);
 753 static DEVICE_ATTR(reqs , S_IRUGO, xsegbd_reqs_show, NULL);
 754 static DEVICE_ATTR(target, S_IRUGO, xsegbd_target_show, NULL);
 755 static DEVICE_ATTR(refresh , S_IWUSR, NULL, xsegbd_image_refresh);
 756 static DEVICE_ATTR(cleanup , S_IWUSR, NULL, xsegbd_cleanup);
 757
 758 static struct attribute *xsegbd_attrs[] = {
 759         &dev_attr_size.attr,
 760         &dev_attr_major.attr,
 761         &dev_attr_srcport.attr,
 762         &dev_attr_dstport.attr,
 763         &dev_attr_id.attr,
 764         &dev_attr_reqs.attr,
 765         &dev_attr_target.attr,
 766         &dev_attr_refresh.attr,
 767         &dev_attr_cleanup.attr,
 768         NULL
 769 };
 770
 771 static struct attribute_group xsegbd_attr_group = {
 772         .attrs = xsegbd_attrs,
 773 };
 774
 775 static const struct attribute_group *xsegbd_attr_groups[] = {
 776         &xsegbd_attr_group,
 777         NULL
 778 };
 779
 780 static void xsegbd_sysfs_dev_release(struct device *dev)
 781 {
 782 }
 783
 784 static struct device_type xsegbd_device_type = {
 785         .name           = "xsegbd",
 786         .groups         = xsegbd_attr_groups,
 787         .release        = xsegbd_sysfs_dev_release,
 788 };
 789
 790 static void xsegbd_root_dev_release(struct device *dev)
 791 {
 792 }
 793
 794 static struct device xsegbd_root_dev = {
 795         .init_name      = "xsegbd",
 796         .release        = xsegbd_root_dev_release,
 797 };
 798
 799 static int xsegbd_bus_add_dev(struct xsegbd_device *xsegbd_dev)
 800 {
 801         int ret = -ENOMEM;
 802         struct device *dev;
 803
 804         mutex_lock_nested(&xsegbd_mutex, SINGLE_DEPTH_NESTING);
 805         dev = &xsegbd_dev->dev;
 806
 807         dev->bus = &xsegbd_bus_type;
 808         dev->type = &xsegbd_device_type;
 809         dev->parent = &xsegbd_root_dev;
 810         dev->release = xsegbd_dev_release;
 811         dev_set_name(dev, "%d", xsegbd_dev->id);
 812
 813         ret = device_register(dev);
 814
 815         mutex_unlock(&xsegbd_mutex);
 816         return ret;
 817 }
 818
 819 static void xsegbd_bus_del_dev(struct xsegbd_device *xsegbd_dev)
 820 {
 821         device_unregister(&xsegbd_dev->dev);
 822 }
 823
 824 static ssize_t xsegbd_add(struct bus_type *bus, const char *buf, size_t count)
 825 {
 826         struct xsegbd_device *xsegbd_dev;
 827         struct xseg_port *port;
 828         ssize_t ret = -ENOMEM;
 829
 830         if (!try_module_get(THIS_MODULE))
 831                 return -ENODEV;
 832
 833         xsegbd_dev = kzalloc(sizeof(*xsegbd_dev), GFP_KERNEL);
 834         if (!xsegbd_dev)
 835                 goto out;
 836
 837         spin_lock_init(&xsegbd_dev->rqlock);
 838         INIT_LIST_HEAD(&xsegbd_dev->node);
 839
 840         /* parse cmd */
 841         if (sscanf(buf, "%" __stringify(XSEGBD_TARGET_NAMELEN) "s "
 842                         "%d:%d:%d", xsegbd_dev->target, &xsegbd_dev->src_portno,
 843                         &xsegbd_dev->dst_portno, &xsegbd_dev->nr_requests) < 3) {
 844                 ret = -EINVAL;
 845                 goto out_dev;
 846         }
 847         xsegbd_dev->targetlen = strlen(xsegbd_dev->target);
 848
 849         spin_lock(&xsegbd_devices_lock);
 850         if (xsegbd_devices[xsegbd_dev->src_portno] != NULL) {
 851                 ret = -EINVAL;
 852                 goto out_unlock;
 853         }
 854         xsegbd_devices[xsegbd_dev->src_portno] = xsegbd_dev;
 855         xsegbd_dev->id = xsegbd_dev->src_portno;
 856         spin_unlock(&xsegbd_devices_lock);
 857
 858         XSEGLOG("registering block device major %d", major);
 859         ret = register_blkdev(major, XSEGBD_NAME);
 860         if (ret < 0) {
 861                 XSEGLOG("cannot register block device!");
 862                 ret = -EBUSY;
 863                 goto out_delentry;
 864         }
 865         xsegbd_dev->major = ret;
 866         XSEGLOG("registered block device major %d", xsegbd_dev->major);
 867
 868         ret = xsegbd_bus_add_dev(xsegbd_dev);
 869         if (ret)
 870                 goto out_blkdev;
 871
 872         if (!xq_alloc_seq(&xsegbd_dev->blk_queue_pending,
 873                                 xsegbd_dev->nr_requests,
 874                                 xsegbd_dev->nr_requests))
 875                 goto out_bus;
 876
 877         xsegbd_dev->blk_req_pending = kzalloc(
 878                         xsegbd_dev->nr_requests *sizeof(struct xsegbd_pending),
 879                                    GFP_KERNEL);
 880         if (!xsegbd_dev->blk_req_pending)
 881                 goto out_freeq;
 882
 883
 884         XSEGLOG("joining segment");
 885         //FIXME use xsebd module config for now
 886         xsegbd_dev->xseg = xseg_join(   xsegbd.config.type,
 887                                         xsegbd.config.name,
 888                                         "segdev",
 889                                         xseg_callback           );
 890         if (!xsegbd_dev->xseg)
 891                 goto out_freepending;
 892
 893
 894         XSEGLOG("binding to source port %u (destination %u)",
 895                         xsegbd_dev->src_portno, xsegbd_dev->dst_portno);
 896         port = xseg_bind_port(xsegbd_dev->xseg, xsegbd_dev->src_portno);
 897         if (!port) {
 898                 XSEGLOG("cannot bind to port");
 899                 ret = -EFAULT;
 900
 901                 goto out_xseg;
 902         }
 903
 904         if (xsegbd_dev->src_portno != xseg_portno(xsegbd_dev->xseg, port)) {
 905                 XSEGLOG("portno != xsegbd_dev->src_portno");
 906                 BUG_ON(1);
 907                 ret = -EFAULT;
 908                 goto out_xseg;
 909         }
 910
 911         /* make sure we don't get any requests until we're ready to handle them */
 912         xseg_cancel_wait(xsegbd_dev->xseg, xseg_portno(xsegbd_dev->xseg, port));
 913
 914         ret = xsegbd_dev_init(xsegbd_dev);
 915         if (ret)
 916                 goto out_xseg;
 917
 918         xseg_prepare_wait(xsegbd_dev->xseg, xseg_portno(xsegbd_dev->xseg, port));
 919         return count;
 920
 921 out_xseg:
 922         xseg_leave(xsegbd_dev->xseg);
 923
 924 out_freepending:
 925         kfree(xsegbd_dev->blk_req_pending);
 926
 927 out_freeq:
 928         xq_free(&xsegbd_dev->blk_queue_pending);
 929
 930 out_bus:
 931         xsegbd_bus_del_dev(xsegbd_dev);
 932         return ret;
 933
 934 out_blkdev:
 935         unregister_blkdev(xsegbd_dev->major, XSEGBD_NAME);
 936
 937 out_delentry:
 938         spin_lock(&xsegbd_devices_lock);
 939         xsegbd_devices[xsegbd_dev->src_portno] = NULL;
 940
 941 out_unlock:
 942         spin_unlock(&xsegbd_devices_lock);
 943
 944 out_dev:
 945         kfree(xsegbd_dev);
 946
 947 out:
 948         return ret;
 949 }
 950
 951 static ssize_t xsegbd_remove(struct bus_type *bus, const char *buf, size_t count)
 952 {
 953         struct xsegbd_device *xsegbd_dev = NULL;
 954         int id, ret;
 955         unsigned long ul_id;
 956
 957         ret = strict_strtoul(buf, 10, &ul_id);
 958         if (ret)
 959                 return ret;
 960
 961         id = (int) ul_id;
 962         if (id != ul_id)
 963                 return -EINVAL;
 964
 965         mutex_lock_nested(&xsegbd_mutex, SINGLE_DEPTH_NESTING);
 966
 967         ret = count;
 968         xsegbd_dev = __xsegbd_get_dev(id);
 969         if (!xsegbd_dev) {
 970                 ret = -ENOENT;
 971                 goto out_unlock;
 972         }
 973         xsegbd_bus_del_dev(xsegbd_dev);
 974
 975 out_unlock:
 976         mutex_unlock(&xsegbd_mutex);
 977         return ret;
 978 }
 979
 980 static struct bus_attribute xsegbd_bus_attrs[] = {
 981         __ATTR(add, S_IWUSR, NULL, xsegbd_add),
 982         __ATTR(remove, S_IWUSR, NULL, xsegbd_remove),
 983         __ATTR_NULL
 984 };
 985
 986 static int xsegbd_sysfs_init(void)
 987 {
 988         int ret;
 989
 990         ret = device_register(&xsegbd_root_dev);
 991         if (ret < 0)
 992                 return ret;
 993
 994         xsegbd_bus_type.bus_attrs = xsegbd_bus_attrs;
 995         ret = bus_register(&xsegbd_bus_type);
 996         if (ret < 0)
 997                 device_unregister(&xsegbd_root_dev);
 998
 999         return ret;
1000 }
1001
1002 static void xsegbd_sysfs_cleanup(void)
1003 {
1004         bus_unregister(&xsegbd_bus_type);
1005         device_unregister(&xsegbd_root_dev);
1006 }
1007
1008 /* *************************** */
1009 /* ** Module Initialization ** */
1010 /* *************************** */
1011
1012 static int __init xsegbd_init(void)
1013 {
1014         int ret = -ENOMEM;
1015         xsegbd_devices = kzalloc(max_dev * sizeof(struct xsegbd_devices *), GFP_KERNEL);
1016         if (!xsegbd_devices)
1017                 goto out;
1018
1019         spin_lock_init(&xsegbd_devices_lock);
1020
1021         ret = -ENOSYS;
1022         ret = xsegbd_xseg_init();
1023         if (ret)
1024                 goto out_free;
1025
1026         ret = xsegbd_sysfs_init();
1027         if (ret)
1028                 goto out_xseg;
1029
1030         XSEGLOG("initialization complete");
1031
1032 out:
1033         return ret;
1034
1035 out_xseg:
1036         xsegbd_xseg_quit();
1037
1038 out_free:
1039         kfree(xsegbd_devices);
1040
1041         goto out;
1042 }
1043
1044 static void __exit xsegbd_exit(void)
1045 {
1046         xsegbd_sysfs_cleanup();
1047         xsegbd_xseg_quit();
1048 }
1049
1050 module_init(xsegbd_init);
1051 module_exit(xsegbd_exit);
1052