Evgeniy Polyakov | ce0d9d7 | 2009-01-14 02:05:27 +0300 | [diff] [blame] | 1 | /* |
| 2 | * 2007+ Copyright (c) Evgeniy Polyakov <zbr@ioremap.net> |
| 3 | * All rights reserved. |
| 4 | * |
| 5 | * This program is free software; you can redistribute it and/or modify |
| 6 | * it under the terms of the GNU General Public License as published by |
| 7 | * the Free Software Foundation; either version 2 of the License, or |
| 8 | * (at your option) any later version. |
| 9 | * |
| 10 | * This program is distributed in the hope that it will be useful, |
| 11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 13 | * GNU General Public License for more details. |
| 14 | */ |
| 15 | |
| 16 | #include <linux/module.h> |
| 17 | #include <linux/kernel.h> |
| 18 | #include <linux/blkdev.h> |
| 19 | #include <linux/bio.h> |
| 20 | #include <linux/buffer_head.h> |
| 21 | #include <linux/connector.h> |
| 22 | #include <linux/dst.h> |
| 23 | #include <linux/device.h> |
| 24 | #include <linux/jhash.h> |
| 25 | #include <linux/idr.h> |
| 26 | #include <linux/init.h> |
| 27 | #include <linux/namei.h> |
| 28 | #include <linux/slab.h> |
| 29 | #include <linux/socket.h> |
| 30 | |
| 31 | #include <linux/in.h> |
| 32 | #include <linux/in6.h> |
| 33 | |
| 34 | #include <net/sock.h> |
| 35 | |
| 36 | static int dst_major; |
| 37 | |
| 38 | static DEFINE_MUTEX(dst_hash_lock); |
| 39 | static struct list_head *dst_hashtable; |
| 40 | static unsigned int dst_hashtable_size = 128; |
| 41 | module_param(dst_hashtable_size, uint, 0644); |
| 42 | |
| 43 | static char dst_name[] = "Dementianting goldfish"; |
| 44 | |
| 45 | static DEFINE_IDR(dst_index_idr); |
| 46 | static struct cb_id cn_dst_id = { CN_DST_IDX, CN_DST_VAL }; |
| 47 | |
| 48 | /* |
| 49 | * DST sysfs tree for device called 'storage': |
| 50 | * |
| 51 | * /sys/bus/dst/devices/storage/ |
| 52 | * /sys/bus/dst/devices/storage/type : 192.168.4.80:1025 |
| 53 | * /sys/bus/dst/devices/storage/size : 800 |
| 54 | * /sys/bus/dst/devices/storage/name : storage |
| 55 | */ |
| 56 | |
| 57 | static int dst_dev_match(struct device *dev, struct device_driver *drv) |
| 58 | { |
| 59 | return 1; |
| 60 | } |
| 61 | |
| 62 | static struct bus_type dst_dev_bus_type = { |
| 63 | .name = "dst", |
| 64 | .match = &dst_dev_match, |
| 65 | }; |
| 66 | |
| 67 | static void dst_node_release(struct device *dev) |
| 68 | { |
| 69 | struct dst_info *info = container_of(dev, struct dst_info, device); |
| 70 | |
| 71 | kfree(info); |
| 72 | } |
| 73 | |
| 74 | static struct device dst_node_dev = { |
| 75 | .bus = &dst_dev_bus_type, |
| 76 | .release = &dst_node_release |
| 77 | }; |
| 78 | |
| 79 | /* |
| 80 | * Setting size of the node after it was changed. |
| 81 | */ |
| 82 | static void dst_node_set_size(struct dst_node *n) |
| 83 | { |
| 84 | struct block_device *bdev; |
| 85 | |
| 86 | set_capacity(n->disk, n->size >> 9); |
| 87 | |
| 88 | bdev = bdget_disk(n->disk, 0); |
| 89 | if (bdev) { |
| 90 | mutex_lock(&bdev->bd_inode->i_mutex); |
| 91 | i_size_write(bdev->bd_inode, n->size); |
| 92 | mutex_unlock(&bdev->bd_inode->i_mutex); |
| 93 | bdput(bdev); |
| 94 | } |
| 95 | } |
| 96 | |
| 97 | /* |
| 98 | * Distributed storage request processing function. |
| 99 | */ |
| 100 | static int dst_request(struct request_queue *q, struct bio *bio) |
| 101 | { |
| 102 | struct dst_node *n = q->queuedata; |
Evgeniy Polyakov | 30c7c1c | 2009-01-19 20:20:37 +0300 | [diff] [blame] | 103 | int err = -EIO; |
| 104 | |
| 105 | if (bio_empty_barrier(bio) && !q->prepare_discard_fn) { |
| 106 | /* |
| 107 | * This is a dirty^Wnice hack, but if we complete this |
| 108 | * operation with -EOPNOTSUPP like intended, XFS |
| 109 | * will stuck and freeze the machine. This may be |
| 110 | * not particulary XFS problem though, but it is the |
| 111 | * only FS which sends empty barrier at umount time |
| 112 | * I worked with. |
| 113 | * |
| 114 | * Empty barriers are not allowed anyway, see 51fd77bd9f512 |
Jens Axboe | 1f98a13 | 2009-09-11 14:32:04 +0200 | [diff] [blame] | 115 | * for example, although later it was changed to |
| 116 | * bio_rw_flagged(bio, BIO_RW_DISCARD) only, which does not |
| 117 | * work in this case. |
Evgeniy Polyakov | 30c7c1c | 2009-01-19 20:20:37 +0300 | [diff] [blame] | 118 | */ |
| 119 | //err = -EOPNOTSUPP; |
| 120 | err = 0; |
| 121 | goto end_io; |
| 122 | } |
Evgeniy Polyakov | ce0d9d7 | 2009-01-14 02:05:27 +0300 | [diff] [blame] | 123 | |
| 124 | bio_get(bio); |
| 125 | |
| 126 | return dst_process_bio(n, bio); |
Evgeniy Polyakov | 30c7c1c | 2009-01-19 20:20:37 +0300 | [diff] [blame] | 127 | |
| 128 | end_io: |
| 129 | bio_endio(bio, err); |
| 130 | return err; |
Evgeniy Polyakov | ce0d9d7 | 2009-01-14 02:05:27 +0300 | [diff] [blame] | 131 | } |
| 132 | |
| 133 | /* |
| 134 | * Open/close callbacks for appropriate block device. |
| 135 | */ |
| 136 | static int dst_bdev_open(struct block_device *bdev, fmode_t mode) |
| 137 | { |
| 138 | struct dst_node *n = bdev->bd_disk->private_data; |
| 139 | |
| 140 | dst_node_get(n); |
| 141 | return 0; |
| 142 | } |
| 143 | |
| 144 | static int dst_bdev_release(struct gendisk *disk, fmode_t mode) |
| 145 | { |
| 146 | struct dst_node *n = disk->private_data; |
| 147 | |
| 148 | dst_node_put(n); |
| 149 | return 0; |
| 150 | } |
| 151 | |
| 152 | static struct block_device_operations dst_blk_ops = { |
| 153 | .open = dst_bdev_open, |
| 154 | .release = dst_bdev_release, |
| 155 | .owner = THIS_MODULE, |
| 156 | }; |
| 157 | |
| 158 | /* |
| 159 | * Block layer binding - disk is created when array is fully configured |
| 160 | * by userspace request. |
| 161 | */ |
| 162 | static int dst_node_create_disk(struct dst_node *n) |
| 163 | { |
| 164 | int err = -ENOMEM; |
| 165 | u32 index = 0; |
| 166 | |
| 167 | n->queue = blk_init_queue(NULL, NULL); |
| 168 | if (!n->queue) |
| 169 | goto err_out_exit; |
| 170 | |
| 171 | n->queue->queuedata = n; |
| 172 | blk_queue_make_request(n->queue, dst_request); |
| 173 | blk_queue_max_phys_segments(n->queue, n->max_pages); |
| 174 | blk_queue_max_hw_segments(n->queue, n->max_pages); |
| 175 | |
| 176 | err = -ENOMEM; |
| 177 | n->disk = alloc_disk(1); |
| 178 | if (!n->disk) |
| 179 | goto err_out_free_queue; |
| 180 | |
| 181 | if (!(n->state->permissions & DST_PERM_WRITE)) { |
| 182 | printk(KERN_INFO "DST node %s attached read-only.\n", n->name); |
| 183 | set_disk_ro(n->disk, 1); |
| 184 | } |
| 185 | |
| 186 | if (!idr_pre_get(&dst_index_idr, GFP_KERNEL)) |
| 187 | goto err_out_put; |
| 188 | |
| 189 | mutex_lock(&dst_hash_lock); |
| 190 | err = idr_get_new(&dst_index_idr, NULL, &index); |
| 191 | mutex_unlock(&dst_hash_lock); |
| 192 | if (err) |
| 193 | goto err_out_put; |
| 194 | |
| 195 | n->disk->major = dst_major; |
| 196 | n->disk->first_minor = index; |
| 197 | n->disk->fops = &dst_blk_ops; |
| 198 | n->disk->queue = n->queue; |
| 199 | n->disk->private_data = n; |
| 200 | snprintf(n->disk->disk_name, sizeof(n->disk->disk_name), "dst-%s", n->name); |
| 201 | |
| 202 | return 0; |
| 203 | |
| 204 | err_out_put: |
| 205 | put_disk(n->disk); |
| 206 | err_out_free_queue: |
| 207 | blk_cleanup_queue(n->queue); |
| 208 | err_out_exit: |
| 209 | return err; |
| 210 | } |
| 211 | |
| 212 | /* |
| 213 | * Sysfs machinery: show device's size. |
| 214 | */ |
| 215 | static ssize_t dst_show_size(struct device *dev, |
| 216 | struct device_attribute *attr, char *buf) |
| 217 | { |
| 218 | struct dst_info *info = container_of(dev, struct dst_info, device); |
| 219 | |
| 220 | return sprintf(buf, "%llu\n", info->size); |
| 221 | } |
| 222 | |
| 223 | /* |
| 224 | * Show local exported device. |
| 225 | */ |
| 226 | static ssize_t dst_show_local(struct device *dev, |
| 227 | struct device_attribute *attr, char *buf) |
| 228 | { |
| 229 | struct dst_info *info = container_of(dev, struct dst_info, device); |
| 230 | |
| 231 | return sprintf(buf, "%s\n", info->local); |
| 232 | } |
| 233 | |
| 234 | /* |
| 235 | * Shows type of the remote node - device major/minor number |
| 236 | * for local nodes and address (af_inet ipv4/ipv6 only) for remote nodes. |
| 237 | */ |
| 238 | static ssize_t dst_show_type(struct device *dev, |
| 239 | struct device_attribute *attr, char *buf) |
| 240 | { |
| 241 | struct dst_info *info = container_of(dev, struct dst_info, device); |
| 242 | int family = info->net.addr.sa_family; |
| 243 | |
| 244 | if (family == AF_INET) { |
| 245 | struct sockaddr_in *sin = (struct sockaddr_in *)&info->net.addr; |
| 246 | return sprintf(buf, "%u.%u.%u.%u:%d\n", |
| 247 | NIPQUAD(sin->sin_addr.s_addr), ntohs(sin->sin_port)); |
| 248 | } else if (family == AF_INET6) { |
| 249 | struct sockaddr_in6 *sin = (struct sockaddr_in6 *)&info->net.addr; |
| 250 | return sprintf(buf, |
| 251 | "%pi6:%d\n", |
| 252 | &sin->sin6_addr, ntohs(sin->sin6_port)); |
| 253 | } else { |
| 254 | int i, sz = PAGE_SIZE - 2; /* 0 symbol and '\n' below */ |
| 255 | int size, addrlen = info->net.addr.sa_data_len; |
| 256 | unsigned char *a = (unsigned char *)&info->net.addr.sa_data; |
| 257 | char *buf_orig = buf; |
| 258 | |
| 259 | size = snprintf(buf, sz, "family: %d, addrlen: %u, addr: ", |
| 260 | family, addrlen); |
| 261 | sz -= size; |
| 262 | buf += size; |
| 263 | |
| 264 | for (i=0; i<addrlen; ++i) { |
| 265 | if (sz < 3) |
| 266 | break; |
| 267 | |
| 268 | size = snprintf(buf, sz, "%02x ", a[i]); |
| 269 | sz -= size; |
| 270 | buf += size; |
| 271 | } |
| 272 | buf += sprintf(buf, "\n"); |
| 273 | |
| 274 | return buf - buf_orig; |
| 275 | } |
| 276 | return 0; |
| 277 | } |
| 278 | |
| 279 | static struct device_attribute dst_node_attrs[] = { |
| 280 | __ATTR(size, 0444, dst_show_size, NULL), |
| 281 | __ATTR(type, 0444, dst_show_type, NULL), |
| 282 | __ATTR(local, 0444, dst_show_local, NULL), |
| 283 | }; |
| 284 | |
| 285 | static int dst_create_node_attributes(struct dst_node *n) |
| 286 | { |
| 287 | int err, i; |
| 288 | |
| 289 | for (i=0; i<ARRAY_SIZE(dst_node_attrs); ++i) { |
| 290 | err = device_create_file(&n->info->device, |
| 291 | &dst_node_attrs[i]); |
| 292 | if (err) |
| 293 | goto err_out_remove_all; |
| 294 | } |
| 295 | return 0; |
| 296 | |
| 297 | err_out_remove_all: |
| 298 | while (--i >= 0) |
| 299 | device_remove_file(&n->info->device, |
| 300 | &dst_node_attrs[i]); |
| 301 | |
| 302 | return err; |
| 303 | } |
| 304 | |
| 305 | static void dst_remove_node_attributes(struct dst_node *n) |
| 306 | { |
| 307 | int i; |
| 308 | |
| 309 | for (i=0; i<ARRAY_SIZE(dst_node_attrs); ++i) |
| 310 | device_remove_file(&n->info->device, |
| 311 | &dst_node_attrs[i]); |
| 312 | } |
| 313 | |
| 314 | /* |
| 315 | * Sysfs cleanup and initialization. |
| 316 | * Shows number of useful parameters. |
| 317 | */ |
| 318 | static void dst_node_sysfs_exit(struct dst_node *n) |
| 319 | { |
| 320 | if (n->info) { |
| 321 | dst_remove_node_attributes(n); |
| 322 | device_unregister(&n->info->device); |
| 323 | n->info = NULL; |
| 324 | } |
| 325 | } |
| 326 | |
| 327 | static int dst_node_sysfs_init(struct dst_node *n) |
| 328 | { |
| 329 | int err; |
| 330 | |
| 331 | n->info = kzalloc(sizeof(struct dst_info), GFP_KERNEL); |
| 332 | if (!n->info) |
| 333 | return -ENOMEM; |
| 334 | |
| 335 | memcpy(&n->info->device, &dst_node_dev, sizeof(struct device)); |
| 336 | n->info->size = n->size; |
| 337 | |
Greg Kroah-Hartman | fdc4f2e | 2009-01-25 10:05:08 -0800 | [diff] [blame] | 338 | dev_set_name(&n->info->device, "dst-%s", n->name); |
Evgeniy Polyakov | ce0d9d7 | 2009-01-14 02:05:27 +0300 | [diff] [blame] | 339 | err = device_register(&n->info->device); |
| 340 | if (err) { |
| 341 | dprintk(KERN_ERR "Failed to register node '%s', err: %d.\n", |
| 342 | n->name, err); |
| 343 | goto err_out_exit; |
| 344 | } |
| 345 | |
| 346 | dst_create_node_attributes(n); |
| 347 | |
| 348 | return 0; |
| 349 | |
| 350 | err_out_exit: |
| 351 | kfree(n->info); |
| 352 | n->info = NULL; |
| 353 | return err; |
| 354 | } |
| 355 | |
| 356 | /* |
| 357 | * DST node hash tables machinery. |
| 358 | */ |
| 359 | static inline unsigned int dst_hash(char *str, unsigned int size) |
| 360 | { |
| 361 | return (jhash(str, size, 0) % dst_hashtable_size); |
| 362 | } |
| 363 | |
| 364 | static void dst_node_remove(struct dst_node *n) |
| 365 | { |
| 366 | mutex_lock(&dst_hash_lock); |
| 367 | list_del_init(&n->node_entry); |
| 368 | mutex_unlock(&dst_hash_lock); |
| 369 | } |
| 370 | |
| 371 | static void dst_node_add(struct dst_node *n) |
| 372 | { |
| 373 | unsigned hash = dst_hash(n->name, sizeof(n->name)); |
| 374 | |
| 375 | mutex_lock(&dst_hash_lock); |
| 376 | list_add_tail(&n->node_entry, &dst_hashtable[hash]); |
| 377 | mutex_unlock(&dst_hash_lock); |
| 378 | } |
| 379 | |
| 380 | /* |
| 381 | * Cleaning node when it is about to be freed. |
| 382 | * There are still users of the socket though, |
| 383 | * so connection cleanup should be protected. |
| 384 | */ |
| 385 | static void dst_node_cleanup(struct dst_node *n) |
| 386 | { |
| 387 | struct dst_state *st = n->state; |
| 388 | |
| 389 | if (!st) |
| 390 | return; |
| 391 | |
| 392 | if (n->queue) { |
| 393 | blk_cleanup_queue(n->queue); |
| 394 | |
| 395 | mutex_lock(&dst_hash_lock); |
| 396 | idr_remove(&dst_index_idr, n->disk->first_minor); |
| 397 | mutex_unlock(&dst_hash_lock); |
| 398 | |
| 399 | put_disk(n->disk); |
| 400 | } |
| 401 | |
| 402 | if (n->bdev) { |
| 403 | sync_blockdev(n->bdev); |
| 404 | blkdev_put(n->bdev, FMODE_READ|FMODE_WRITE); |
| 405 | } |
| 406 | |
| 407 | dst_state_lock(st); |
| 408 | st->need_exit = 1; |
| 409 | dst_state_exit_connected(st); |
| 410 | dst_state_unlock(st); |
| 411 | |
| 412 | wake_up(&st->thread_wait); |
| 413 | |
| 414 | dst_state_put(st); |
| 415 | n->state = NULL; |
| 416 | } |
| 417 | |
| 418 | /* |
| 419 | * Free security attributes attached to given node. |
| 420 | */ |
| 421 | static void dst_security_exit(struct dst_node *n) |
| 422 | { |
| 423 | struct dst_secure *s, *tmp; |
| 424 | |
| 425 | list_for_each_entry_safe(s, tmp, &n->security_list, sec_entry) { |
| 426 | list_del(&s->sec_entry); |
| 427 | kfree(s); |
| 428 | } |
| 429 | } |
| 430 | |
| 431 | /* |
| 432 | * Free node when there are no more users. |
| 433 | * Actually node has to be freed on behalf od userspace process, |
| 434 | * since there are number of threads, which are embedded in the |
| 435 | * node, so they can not exit and free node from there, that is |
| 436 | * why there is a wakeup if reference counter is not equal to zero. |
| 437 | */ |
| 438 | void dst_node_put(struct dst_node *n) |
| 439 | { |
| 440 | if (unlikely(!n)) |
| 441 | return; |
| 442 | |
| 443 | dprintk("%s: n: %p, refcnt: %d.\n", |
| 444 | __func__, n, atomic_read(&n->refcnt)); |
| 445 | |
| 446 | if (atomic_dec_and_test(&n->refcnt)) { |
| 447 | dst_node_remove(n); |
| 448 | n->trans_scan_timeout = 0; |
| 449 | dst_node_cleanup(n); |
| 450 | thread_pool_destroy(n->pool); |
| 451 | dst_node_sysfs_exit(n); |
| 452 | dst_node_crypto_exit(n); |
| 453 | dst_security_exit(n); |
| 454 | dst_node_trans_exit(n); |
| 455 | |
| 456 | kfree(n); |
| 457 | |
| 458 | dprintk("%s: freed n: %p.\n", __func__, n); |
| 459 | } else { |
| 460 | wake_up(&n->wait); |
| 461 | } |
| 462 | } |
| 463 | |
| 464 | /* |
| 465 | * This function finds devices major/minor numbers for given pathname. |
| 466 | */ |
| 467 | static int dst_lookup_device(const char *path, dev_t *dev) |
| 468 | { |
| 469 | int err; |
| 470 | struct nameidata nd; |
| 471 | struct inode *inode; |
| 472 | |
| 473 | err = path_lookup(path, LOOKUP_FOLLOW, &nd); |
| 474 | if (err) |
| 475 | return err; |
| 476 | |
| 477 | inode = nd.path.dentry->d_inode; |
| 478 | if (!inode) { |
| 479 | err = -ENOENT; |
| 480 | goto out; |
| 481 | } |
| 482 | |
| 483 | if (!S_ISBLK(inode->i_mode)) { |
| 484 | err = -ENOTBLK; |
| 485 | goto out; |
| 486 | } |
| 487 | |
| 488 | *dev = inode->i_rdev; |
| 489 | |
| 490 | out: |
| 491 | path_put(&nd.path); |
| 492 | return err; |
| 493 | } |
| 494 | |
| 495 | /* |
| 496 | * Setting up export device: lookup by the name, get its size |
| 497 | * and setup listening socket, which will accept clients, which |
| 498 | * will submit IO for given storage. |
| 499 | */ |
| 500 | static int dst_setup_export(struct dst_node *n, struct dst_ctl *ctl, |
| 501 | struct dst_export_ctl *le) |
| 502 | { |
| 503 | int err; |
| 504 | dev_t dev = 0; /* gcc likes to scream here */ |
| 505 | |
| 506 | snprintf(n->info->local, sizeof(n->info->local), "%s", le->device); |
| 507 | |
| 508 | err = dst_lookup_device(le->device, &dev); |
| 509 | if (err) |
| 510 | return err; |
| 511 | |
| 512 | n->bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE); |
| 513 | if (!n->bdev) |
| 514 | return -ENODEV; |
| 515 | |
| 516 | if (n->size != 0) |
| 517 | n->size = min_t(loff_t, n->bdev->bd_inode->i_size, n->size); |
| 518 | else |
| 519 | n->size = n->bdev->bd_inode->i_size; |
| 520 | |
| 521 | n->info->size = n->size; |
| 522 | err = dst_node_init_listened(n, le); |
| 523 | if (err) |
| 524 | goto err_out_cleanup; |
| 525 | |
| 526 | return 0; |
| 527 | |
| 528 | err_out_cleanup: |
| 529 | blkdev_put(n->bdev, FMODE_READ|FMODE_WRITE); |
| 530 | n->bdev = NULL; |
| 531 | |
| 532 | return err; |
| 533 | } |
| 534 | |
| 535 | /* Empty thread pool callbacks for the network processing threads. */ |
| 536 | static inline void *dst_thread_network_init(void *data) |
| 537 | { |
| 538 | dprintk("%s: data: %p.\n", __func__, data); |
| 539 | return data; |
| 540 | } |
| 541 | |
| 542 | static inline void dst_thread_network_cleanup(void *data) |
| 543 | { |
| 544 | dprintk("%s: data: %p.\n", __func__, data); |
| 545 | } |
| 546 | |
| 547 | /* |
| 548 | * Allocate DST node and initialize some of its parameters. |
| 549 | */ |
| 550 | static struct dst_node *dst_alloc_node(struct dst_ctl *ctl, |
| 551 | int (*start)(struct dst_node *), |
| 552 | int num) |
| 553 | { |
| 554 | struct dst_node *n; |
| 555 | int err; |
| 556 | |
| 557 | n = kzalloc(sizeof(struct dst_node), GFP_KERNEL); |
| 558 | if (!n) |
| 559 | return NULL; |
| 560 | |
| 561 | INIT_LIST_HEAD(&n->node_entry); |
| 562 | |
| 563 | INIT_LIST_HEAD(&n->security_list); |
| 564 | mutex_init(&n->security_lock); |
| 565 | |
| 566 | init_waitqueue_head(&n->wait); |
| 567 | |
| 568 | n->trans_scan_timeout = msecs_to_jiffies(ctl->trans_scan_timeout); |
| 569 | if (!n->trans_scan_timeout) |
| 570 | n->trans_scan_timeout = HZ; |
| 571 | |
| 572 | n->trans_max_retries = ctl->trans_max_retries; |
| 573 | if (!n->trans_max_retries) |
| 574 | n->trans_max_retries = 10; |
| 575 | |
| 576 | /* |
| 577 | * Pretty much arbitrary default numbers. |
| 578 | * 32 matches maximum number of pages in bio originated from ext3 (31). |
| 579 | */ |
| 580 | n->max_pages = ctl->max_pages; |
| 581 | if (!n->max_pages) |
| 582 | n->max_pages = 32; |
| 583 | |
| 584 | if (n->max_pages > 1024) |
| 585 | n->max_pages = 1024; |
| 586 | |
| 587 | n->start = start; |
| 588 | n->size = ctl->size; |
| 589 | |
| 590 | atomic_set(&n->refcnt, 1); |
| 591 | atomic_long_set(&n->gen, 0); |
| 592 | snprintf(n->name, sizeof(n->name), "%s", ctl->name); |
| 593 | |
| 594 | err = dst_node_sysfs_init(n); |
| 595 | if (err) |
| 596 | goto err_out_free; |
| 597 | |
| 598 | n->pool = thread_pool_create(num, n->name, dst_thread_network_init, |
| 599 | dst_thread_network_cleanup, n); |
| 600 | if (IS_ERR(n->pool)) { |
| 601 | err = PTR_ERR(n->pool); |
| 602 | goto err_out_sysfs_exit; |
| 603 | } |
| 604 | |
| 605 | dprintk("%s: n: %p, name: %s.\n", __func__, n, n->name); |
| 606 | |
| 607 | return n; |
| 608 | |
| 609 | err_out_sysfs_exit: |
| 610 | dst_node_sysfs_exit(n); |
| 611 | err_out_free: |
| 612 | kfree(n); |
| 613 | return NULL; |
| 614 | } |
| 615 | |
| 616 | /* |
| 617 | * Starting a node, connected to the remote server: |
| 618 | * register block device and initialize transaction mechanism. |
| 619 | * In revers order though. |
| 620 | * |
| 621 | * It will autonegotiate some parameters with the remote node |
| 622 | * and update local if needed. |
| 623 | * |
| 624 | * Transaction initialization should be the last thing before |
| 625 | * starting the node, since transaction should include not only |
| 626 | * block IO, but also crypto related data (if any), which are |
| 627 | * initialized separately. |
| 628 | */ |
| 629 | static int dst_start_remote(struct dst_node *n) |
| 630 | { |
| 631 | int err; |
| 632 | |
| 633 | err = dst_node_trans_init(n, sizeof(struct dst_trans)); |
| 634 | if (err) |
| 635 | return err; |
| 636 | |
| 637 | err = dst_node_create_disk(n); |
| 638 | if (err) |
| 639 | return err; |
| 640 | |
| 641 | dst_node_set_size(n); |
| 642 | add_disk(n->disk); |
| 643 | |
| 644 | dprintk("DST: started remote node '%s', minor: %d.\n", n->name, n->disk->first_minor); |
| 645 | |
| 646 | return 0; |
| 647 | } |
| 648 | |
| 649 | /* |
| 650 | * Adding remote node and initialize connection. |
| 651 | */ |
| 652 | static int dst_add_remote(struct dst_node *n, struct dst_ctl *ctl, |
| 653 | void *data, unsigned int size) |
| 654 | { |
| 655 | int err; |
| 656 | struct dst_network_ctl *rctl = data; |
| 657 | |
| 658 | if (n) |
| 659 | return -EEXIST; |
| 660 | |
| 661 | if (size != sizeof(struct dst_network_ctl)) |
| 662 | return -EINVAL; |
| 663 | |
| 664 | n = dst_alloc_node(ctl, dst_start_remote, 1); |
| 665 | if (!n) |
| 666 | return -ENOMEM; |
| 667 | |
| 668 | memcpy(&n->info->net, rctl, sizeof(struct dst_network_ctl)); |
| 669 | err = dst_node_init_connected(n, rctl); |
| 670 | if (err) |
| 671 | goto err_out_free; |
| 672 | |
| 673 | dst_node_add(n); |
| 674 | |
| 675 | return 0; |
| 676 | |
| 677 | err_out_free: |
| 678 | dst_node_put(n); |
| 679 | return err; |
| 680 | } |
| 681 | |
| 682 | /* |
| 683 | * Adding export node: initializing block device and listening socket. |
| 684 | */ |
| 685 | static int dst_add_export(struct dst_node *n, struct dst_ctl *ctl, |
| 686 | void *data, unsigned int size) |
| 687 | { |
| 688 | int err; |
| 689 | struct dst_export_ctl *le = data; |
| 690 | |
| 691 | if (n) |
| 692 | return -EEXIST; |
| 693 | |
| 694 | if (size != sizeof(struct dst_export_ctl)) |
| 695 | return -EINVAL; |
| 696 | |
| 697 | n = dst_alloc_node(ctl, dst_start_export, 2); |
| 698 | if (!n) |
| 699 | return -EINVAL; |
| 700 | |
| 701 | err = dst_setup_export(n, ctl, le); |
| 702 | if (err) |
| 703 | goto err_out_free; |
| 704 | |
| 705 | dst_node_add(n); |
| 706 | |
| 707 | return 0; |
| 708 | |
| 709 | err_out_free: |
| 710 | dst_node_put(n); |
| 711 | return err; |
| 712 | } |
| 713 | |
| 714 | static int dst_node_remove_unload(struct dst_node *n) |
| 715 | { |
| 716 | printk(KERN_INFO "STOPPED name: '%s', size: %llu.\n", |
| 717 | n->name, n->size); |
| 718 | |
| 719 | if (n->disk) |
| 720 | del_gendisk(n->disk); |
| 721 | |
| 722 | dst_node_remove(n); |
| 723 | dst_node_sysfs_exit(n); |
| 724 | |
| 725 | /* |
| 726 | * This is not a hack. Really. |
| 727 | * Node's reference counter allows to implement fine grained |
| 728 | * node freeing, but since all transactions (which hold node's |
| 729 | * reference counter) are processed in the dedicated thread, |
| 730 | * it is possible that reference will hit zero in that thread, |
| 731 | * so we will not be able to exit thread and cleanup the node. |
| 732 | * |
| 733 | * So, we remove disk, so no new activity is possible, and |
| 734 | * wait until all pending transaction are completed (either |
| 735 | * in receiving thread or by timeout in workqueue), in this |
| 736 | * case reference counter will be less or equal to 2 (once set in |
| 737 | * dst_alloc_node() and then in connector message parser; |
| 738 | * or when we force module unloading, and connector message |
| 739 | * parser does not hold a reference, in this case reference |
| 740 | * counter will be equal to 1), |
| 741 | * and subsequent dst_node_put() calls will free the node. |
| 742 | */ |
| 743 | dprintk("%s: going to sleep with %d refcnt.\n", __func__, atomic_read(&n->refcnt)); |
| 744 | wait_event(n->wait, atomic_read(&n->refcnt) <= 2); |
| 745 | |
| 746 | dst_node_put(n); |
| 747 | return 0; |
| 748 | } |
| 749 | |
| 750 | /* |
| 751 | * Remove node from the hash table. |
| 752 | */ |
| 753 | static int dst_del_node(struct dst_node *n, struct dst_ctl *ctl, |
| 754 | void *data, unsigned int size) |
| 755 | { |
| 756 | if (!n) |
| 757 | return -ENODEV; |
| 758 | |
| 759 | return dst_node_remove_unload(n); |
| 760 | } |
| 761 | |
| 762 | /* |
| 763 | * Initialize crypto processing for given node. |
| 764 | */ |
| 765 | static int dst_crypto_init(struct dst_node *n, struct dst_ctl *ctl, |
| 766 | void *data, unsigned int size) |
| 767 | { |
| 768 | struct dst_crypto_ctl *crypto = data; |
| 769 | |
| 770 | if (!n) |
| 771 | return -ENODEV; |
| 772 | |
| 773 | if (size != sizeof(struct dst_crypto_ctl) + crypto->hash_keysize + |
| 774 | crypto->cipher_keysize) |
| 775 | return -EINVAL; |
| 776 | |
| 777 | if (n->trans_cache) |
| 778 | return -EEXIST; |
| 779 | |
| 780 | return dst_node_crypto_init(n, crypto); |
| 781 | } |
| 782 | |
| 783 | /* |
| 784 | * Security attributes for given node. |
| 785 | */ |
| 786 | static int dst_security_init(struct dst_node *n, struct dst_ctl *ctl, |
| 787 | void *data, unsigned int size) |
| 788 | { |
| 789 | struct dst_secure *s; |
| 790 | |
| 791 | if (!n) |
| 792 | return -ENODEV; |
| 793 | |
| 794 | if (size != sizeof(struct dst_secure_user)) |
| 795 | return -EINVAL; |
| 796 | |
| 797 | s = kmalloc(sizeof(struct dst_secure), GFP_KERNEL); |
| 798 | if (!s) |
| 799 | return -ENOMEM; |
| 800 | |
| 801 | memcpy(&s->sec, data, size); |
| 802 | |
| 803 | mutex_lock(&n->security_lock); |
| 804 | list_add_tail(&s->sec_entry, &n->security_list); |
| 805 | mutex_unlock(&n->security_lock); |
| 806 | |
| 807 | return 0; |
| 808 | } |
| 809 | |
| 810 | /* |
| 811 | * Kill'em all! |
| 812 | */ |
| 813 | static int dst_start_node(struct dst_node *n, struct dst_ctl *ctl, |
| 814 | void *data, unsigned int size) |
| 815 | { |
| 816 | int err; |
| 817 | |
| 818 | if (!n) |
| 819 | return -ENODEV; |
| 820 | |
| 821 | if (n->trans_cache) |
| 822 | return 0; |
| 823 | |
| 824 | err = n->start(n); |
| 825 | if (err) |
| 826 | return err; |
| 827 | |
| 828 | printk(KERN_INFO "STARTED name: '%s', size: %llu.\n", n->name, n->size); |
| 829 | return 0; |
| 830 | } |
| 831 | |
| 832 | typedef int (*dst_command_func)(struct dst_node *n, struct dst_ctl *ctl, |
| 833 | void *data, unsigned int size); |
| 834 | |
| 835 | /* |
| 836 | * List of userspace commands. |
| 837 | */ |
| 838 | static dst_command_func dst_commands[] = { |
| 839 | [DST_ADD_REMOTE] = &dst_add_remote, |
| 840 | [DST_ADD_EXPORT] = &dst_add_export, |
| 841 | [DST_DEL_NODE] = &dst_del_node, |
| 842 | [DST_CRYPTO] = &dst_crypto_init, |
| 843 | [DST_SECURITY] = &dst_security_init, |
| 844 | [DST_START] = &dst_start_node, |
| 845 | }; |
| 846 | |
| 847 | /* |
| 848 | * Configuration parser. |
| 849 | */ |
Mike Frysinger | 0741241 | 2009-07-17 10:13:21 -0700 | [diff] [blame] | 850 | static void cn_dst_callback(struct cn_msg *msg) |
Evgeniy Polyakov | ce0d9d7 | 2009-01-14 02:05:27 +0300 | [diff] [blame] | 851 | { |
| 852 | struct dst_ctl *ctl; |
Evgeniy Polyakov | ce0d9d7 | 2009-01-14 02:05:27 +0300 | [diff] [blame] | 853 | int err; |
| 854 | struct dst_ctl_ack ack; |
| 855 | struct dst_node *n = NULL, *tmp; |
| 856 | unsigned int hash; |
| 857 | |
| 858 | if (msg->len < sizeof(struct dst_ctl)) { |
| 859 | err = -EBADMSG; |
| 860 | goto out; |
| 861 | } |
| 862 | |
| 863 | ctl = (struct dst_ctl *)msg->data; |
| 864 | |
| 865 | if (ctl->cmd >= DST_CMD_MAX) { |
| 866 | err = -EINVAL; |
| 867 | goto out; |
| 868 | } |
| 869 | hash = dst_hash(ctl->name, sizeof(ctl->name)); |
| 870 | |
| 871 | mutex_lock(&dst_hash_lock); |
| 872 | list_for_each_entry(tmp, &dst_hashtable[hash], node_entry) { |
| 873 | if (!memcmp(tmp->name, ctl->name, sizeof(tmp->name))) { |
| 874 | n = tmp; |
| 875 | dst_node_get(n); |
| 876 | break; |
| 877 | } |
| 878 | } |
| 879 | mutex_unlock(&dst_hash_lock); |
| 880 | |
| 881 | err = dst_commands[ctl->cmd](n, ctl, msg->data + sizeof(struct dst_ctl), |
| 882 | msg->len - sizeof(struct dst_ctl)); |
| 883 | |
| 884 | dst_node_put(n); |
| 885 | out: |
| 886 | memcpy(&ack.msg, msg, sizeof(struct cn_msg)); |
| 887 | |
| 888 | ack.msg.ack = msg->ack + 1; |
| 889 | ack.msg.len = sizeof(struct dst_ctl_ack) - sizeof(struct cn_msg); |
| 890 | |
| 891 | ack.error = err; |
| 892 | |
| 893 | cn_netlink_send(&ack.msg, 0, GFP_KERNEL); |
| 894 | } |
| 895 | |
| 896 | /* |
| 897 | * Global initialization: sysfs, hash table, block device registration, |
| 898 | * connector and various caches. |
| 899 | */ |
| 900 | static int __init dst_sysfs_init(void) |
| 901 | { |
| 902 | return bus_register(&dst_dev_bus_type); |
| 903 | } |
| 904 | |
| 905 | static void dst_sysfs_exit(void) |
| 906 | { |
| 907 | bus_unregister(&dst_dev_bus_type); |
| 908 | } |
| 909 | |
| 910 | static int __init dst_hashtable_init(void) |
| 911 | { |
| 912 | unsigned int i; |
| 913 | |
| 914 | dst_hashtable = kcalloc(dst_hashtable_size, sizeof(struct list_head), |
| 915 | GFP_KERNEL); |
| 916 | if (!dst_hashtable) |
| 917 | return -ENOMEM; |
| 918 | |
| 919 | for (i=0; i<dst_hashtable_size; ++i) |
| 920 | INIT_LIST_HEAD(&dst_hashtable[i]); |
| 921 | |
| 922 | return 0; |
| 923 | } |
| 924 | |
| 925 | static void dst_hashtable_exit(void) |
| 926 | { |
| 927 | unsigned int i; |
| 928 | struct dst_node *n, *tmp; |
| 929 | |
| 930 | for (i=0; i<dst_hashtable_size; ++i) { |
| 931 | list_for_each_entry_safe(n, tmp, &dst_hashtable[i], node_entry) { |
| 932 | dst_node_remove_unload(n); |
| 933 | } |
| 934 | } |
| 935 | |
| 936 | kfree(dst_hashtable); |
| 937 | } |
| 938 | |
| 939 | static int __init dst_sys_init(void) |
| 940 | { |
| 941 | int err = -ENOMEM; |
| 942 | |
| 943 | err = dst_hashtable_init(); |
| 944 | if (err) |
| 945 | goto err_out_exit; |
| 946 | |
| 947 | err = dst_export_init(); |
| 948 | if (err) |
| 949 | goto err_out_hashtable_exit; |
| 950 | |
| 951 | err = register_blkdev(dst_major, DST_NAME); |
| 952 | if (err < 0) |
| 953 | goto err_out_export_exit; |
| 954 | if (err) |
| 955 | dst_major = err; |
| 956 | |
| 957 | err = dst_sysfs_init(); |
| 958 | if (err) |
| 959 | goto err_out_unregister; |
| 960 | |
| 961 | err = cn_add_callback(&cn_dst_id, "DST", cn_dst_callback); |
| 962 | if (err) |
| 963 | goto err_out_sysfs_exit; |
| 964 | |
| 965 | printk(KERN_INFO "Distributed storage, '%s' release.\n", dst_name); |
| 966 | |
| 967 | return 0; |
| 968 | |
| 969 | err_out_sysfs_exit: |
| 970 | dst_sysfs_exit(); |
| 971 | err_out_unregister: |
| 972 | unregister_blkdev(dst_major, DST_NAME); |
| 973 | err_out_export_exit: |
| 974 | dst_export_exit(); |
| 975 | err_out_hashtable_exit: |
| 976 | dst_hashtable_exit(); |
| 977 | err_out_exit: |
| 978 | return err; |
| 979 | } |
| 980 | |
| 981 | static void __exit dst_sys_exit(void) |
| 982 | { |
| 983 | cn_del_callback(&cn_dst_id); |
| 984 | unregister_blkdev(dst_major, DST_NAME); |
| 985 | dst_hashtable_exit(); |
| 986 | dst_sysfs_exit(); |
| 987 | dst_export_exit(); |
| 988 | } |
| 989 | |
| 990 | module_init(dst_sys_init); |
| 991 | module_exit(dst_sys_exit); |
| 992 | |
| 993 | MODULE_DESCRIPTION("Distributed storage"); |
| 994 | MODULE_AUTHOR("Evgeniy Polyakov <zbr@ioremap.net>"); |
| 995 | MODULE_LICENSE("GPL"); |