1 /* 2 * Copyright (c) 2004 Topspin Communications. All rights reserved. 3 * Copyright (c) 2005 Voltaire, Inc. All rights reserved. 4 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved. 5 * Copyright (c) 2008 Cisco. All rights reserved. 6 * 7 * This software is available to you under a choice of one of two 8 * licenses. You may choose to be licensed under the terms of the GNU 9 * General Public License (GPL) Version 2, available from the file 10 * COPYING in the main directory of this source tree, or the 11 * OpenIB.org BSD license below: 12 * 13 * Redistribution and use in source and binary forms, with or 14 * without modification, are permitted provided that the following 15 * conditions are met: 16 * 17 * - Redistributions of source code must retain the above 18 * copyright notice, this list of conditions and the following 19 * disclaimer. 20 * 21 * - Redistributions in binary form must reproduce the above 22 * copyright notice, this list of conditions and the following 23 * disclaimer in the documentation and/or other materials 24 * provided with the distribution. 25 * 26 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 27 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 28 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 29 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 30 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 31 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 32 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 33 * SOFTWARE. 34 */ 35 36 #define pr_fmt(fmt) "user_mad: " fmt 37 38 #include <linux/module.h> 39 #include <linux/init.h> 40 #include <linux/device.h> 41 #include <linux/err.h> 42 #include <linux/fs.h> 43 #include <linux/cdev.h> 44 #include <linux/dma-mapping.h> 45 #include <linux/poll.h> 46 #include <linux/mutex.h> 47 #include <linux/kref.h> 48 #include <linux/compat.h> 49 #include <linux/sched.h> 50 #include <linux/semaphore.h> 51 #include <linux/slab.h> 52 53 #include <asm/uaccess.h> 54 55 #include <rdma/ib_mad.h> 56 #include <rdma/ib_user_mad.h> 57 58 MODULE_AUTHOR("Roland Dreier"); 59 MODULE_DESCRIPTION("InfiniBand userspace MAD packet access"); 60 MODULE_LICENSE("Dual BSD/GPL"); 61 62 enum { 63 IB_UMAD_MAX_PORTS = 64, 64 IB_UMAD_MAX_AGENTS = 32, 65 66 IB_UMAD_MAJOR = 231, 67 IB_UMAD_MINOR_BASE = 0 68 }; 69 70 /* 71 * Our lifetime rules for these structs are the following: 72 * device special file is opened, we take a reference on the 73 * ib_umad_port's struct ib_umad_device. We drop these 74 * references in the corresponding close(). 75 * 76 * In addition to references coming from open character devices, there 77 * is one more reference to each ib_umad_device representing the 78 * module's reference taken when allocating the ib_umad_device in 79 * ib_umad_add_one(). 80 * 81 * When destroying an ib_umad_device, we drop the module's reference. 82 */ 83 84 struct ib_umad_port { 85 struct cdev cdev; 86 struct device *dev; 87 88 struct cdev sm_cdev; 89 struct device *sm_dev; 90 struct semaphore sm_sem; 91 92 struct mutex file_mutex; 93 struct list_head file_list; 94 95 struct ib_device *ib_dev; 96 struct ib_umad_device *umad_dev; 97 int dev_num; 98 u8 port_num; 99 }; 100 101 struct ib_umad_device { 102 struct kobject kobj; 103 struct ib_umad_port port[0]; 104 }; 105 106 struct ib_umad_file { 107 struct mutex mutex; 108 struct ib_umad_port *port; 109 struct list_head recv_list; 110 struct list_head send_list; 111 struct list_head port_list; 112 spinlock_t send_lock; 113 wait_queue_head_t recv_wait; 114 struct ib_mad_agent *agent[IB_UMAD_MAX_AGENTS]; 115 int agents_dead; 116 u8 use_pkey_index; 117 u8 already_used; 118 }; 119 120 struct ib_umad_packet { 121 struct ib_mad_send_buf *msg; 122 struct ib_mad_recv_wc *recv_wc; 123 struct list_head list; 124 int length; 125 struct ib_user_mad mad; 126 }; 127 128 static struct class *umad_class; 129 130 static const dev_t base_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE); 131 132 static DEFINE_SPINLOCK(port_lock); 133 static DECLARE_BITMAP(dev_map, IB_UMAD_MAX_PORTS); 134 135 static void ib_umad_add_one(struct ib_device *device); 136 static void ib_umad_remove_one(struct ib_device *device, void *client_data); 137 138 static void ib_umad_release_dev(struct kobject *kobj) 139 { 140 struct ib_umad_device *dev = 141 container_of(kobj, struct ib_umad_device, kobj); 142 143 kfree(dev); 144 } 145 146 static struct kobj_type ib_umad_dev_ktype = { 147 .release = ib_umad_release_dev, 148 }; 149 150 static int hdr_size(struct ib_umad_file *file) 151 { 152 return file->use_pkey_index ? sizeof (struct ib_user_mad_hdr) : 153 sizeof (struct ib_user_mad_hdr_old); 154 } 155 156 /* caller must hold file->mutex */ 157 static struct ib_mad_agent *__get_agent(struct ib_umad_file *file, int id) 158 { 159 return file->agents_dead ? NULL : file->agent[id]; 160 } 161 162 static int queue_packet(struct ib_umad_file *file, 163 struct ib_mad_agent *agent, 164 struct ib_umad_packet *packet) 165 { 166 int ret = 1; 167 168 mutex_lock(&file->mutex); 169 170 for (packet->mad.hdr.id = 0; 171 packet->mad.hdr.id < IB_UMAD_MAX_AGENTS; 172 packet->mad.hdr.id++) 173 if (agent == __get_agent(file, packet->mad.hdr.id)) { 174 list_add_tail(&packet->list, &file->recv_list); 175 wake_up_interruptible(&file->recv_wait); 176 ret = 0; 177 break; 178 } 179 180 mutex_unlock(&file->mutex); 181 182 return ret; 183 } 184 185 static void dequeue_send(struct ib_umad_file *file, 186 struct ib_umad_packet *packet) 187 { 188 spin_lock_irq(&file->send_lock); 189 list_del(&packet->list); 190 spin_unlock_irq(&file->send_lock); 191 } 192 193 static void send_handler(struct ib_mad_agent *agent, 194 struct ib_mad_send_wc *send_wc) 195 { 196 struct ib_umad_file *file = agent->context; 197 struct ib_umad_packet *packet = send_wc->send_buf->context[0]; 198 199 dequeue_send(file, packet); 200 ib_destroy_ah(packet->msg->ah); 201 ib_free_send_mad(packet->msg); 202 203 if (send_wc->status == IB_WC_RESP_TIMEOUT_ERR) { 204 packet->length = IB_MGMT_MAD_HDR; 205 packet->mad.hdr.status = ETIMEDOUT; 206 if (!queue_packet(file, agent, packet)) 207 return; 208 } 209 kfree(packet); 210 } 211 212 static void recv_handler(struct ib_mad_agent *agent, 213 struct ib_mad_recv_wc *mad_recv_wc) 214 { 215 struct ib_umad_file *file = agent->context; 216 struct ib_umad_packet *packet; 217 218 if (mad_recv_wc->wc->status != IB_WC_SUCCESS) 219 goto err1; 220 221 packet = kzalloc(sizeof *packet, GFP_KERNEL); 222 if (!packet) 223 goto err1; 224 225 packet->length = mad_recv_wc->mad_len; 226 packet->recv_wc = mad_recv_wc; 227 228 packet->mad.hdr.status = 0; 229 packet->mad.hdr.length = hdr_size(file) + mad_recv_wc->mad_len; 230 packet->mad.hdr.qpn = cpu_to_be32(mad_recv_wc->wc->src_qp); 231 packet->mad.hdr.lid = cpu_to_be16(mad_recv_wc->wc->slid); 232 packet->mad.hdr.sl = mad_recv_wc->wc->sl; 233 packet->mad.hdr.path_bits = mad_recv_wc->wc->dlid_path_bits; 234 packet->mad.hdr.pkey_index = mad_recv_wc->wc->pkey_index; 235 packet->mad.hdr.grh_present = !!(mad_recv_wc->wc->wc_flags & IB_WC_GRH); 236 if (packet->mad.hdr.grh_present) { 237 struct ib_ah_attr ah_attr; 238 239 ib_init_ah_from_wc(agent->device, agent->port_num, 240 mad_recv_wc->wc, mad_recv_wc->recv_buf.grh, 241 &ah_attr); 242 243 packet->mad.hdr.gid_index = ah_attr.grh.sgid_index; 244 packet->mad.hdr.hop_limit = ah_attr.grh.hop_limit; 245 packet->mad.hdr.traffic_class = ah_attr.grh.traffic_class; 246 memcpy(packet->mad.hdr.gid, &ah_attr.grh.dgid, 16); 247 packet->mad.hdr.flow_label = cpu_to_be32(ah_attr.grh.flow_label); 248 } 249 250 if (queue_packet(file, agent, packet)) 251 goto err2; 252 return; 253 254 err2: 255 kfree(packet); 256 err1: 257 ib_free_recv_mad(mad_recv_wc); 258 } 259 260 static ssize_t copy_recv_mad(struct ib_umad_file *file, char __user *buf, 261 struct ib_umad_packet *packet, size_t count) 262 { 263 struct ib_mad_recv_buf *recv_buf; 264 int left, seg_payload, offset, max_seg_payload; 265 size_t seg_size; 266 267 recv_buf = &packet->recv_wc->recv_buf; 268 seg_size = packet->recv_wc->mad_seg_size; 269 270 /* We need enough room to copy the first (or only) MAD segment. */ 271 if ((packet->length <= seg_size && 272 count < hdr_size(file) + packet->length) || 273 (packet->length > seg_size && 274 count < hdr_size(file) + seg_size)) 275 return -EINVAL; 276 277 if (copy_to_user(buf, &packet->mad, hdr_size(file))) 278 return -EFAULT; 279 280 buf += hdr_size(file); 281 seg_payload = min_t(int, packet->length, seg_size); 282 if (copy_to_user(buf, recv_buf->mad, seg_payload)) 283 return -EFAULT; 284 285 if (seg_payload < packet->length) { 286 /* 287 * Multipacket RMPP MAD message. Copy remainder of message. 288 * Note that last segment may have a shorter payload. 289 */ 290 if (count < hdr_size(file) + packet->length) { 291 /* 292 * The buffer is too small, return the first RMPP segment, 293 * which includes the RMPP message length. 294 */ 295 return -ENOSPC; 296 } 297 offset = ib_get_mad_data_offset(recv_buf->mad->mad_hdr.mgmt_class); 298 max_seg_payload = seg_size - offset; 299 300 for (left = packet->length - seg_payload, buf += seg_payload; 301 left; left -= seg_payload, buf += seg_payload) { 302 recv_buf = container_of(recv_buf->list.next, 303 struct ib_mad_recv_buf, list); 304 seg_payload = min(left, max_seg_payload); 305 if (copy_to_user(buf, ((void *) recv_buf->mad) + offset, 306 seg_payload)) 307 return -EFAULT; 308 } 309 } 310 return hdr_size(file) + packet->length; 311 } 312 313 static ssize_t copy_send_mad(struct ib_umad_file *file, char __user *buf, 314 struct ib_umad_packet *packet, size_t count) 315 { 316 ssize_t size = hdr_size(file) + packet->length; 317 318 if (count < size) 319 return -EINVAL; 320 321 if (copy_to_user(buf, &packet->mad, hdr_size(file))) 322 return -EFAULT; 323 324 buf += hdr_size(file); 325 326 if (copy_to_user(buf, packet->mad.data, packet->length)) 327 return -EFAULT; 328 329 return size; 330 } 331 332 static ssize_t ib_umad_read(struct file *filp, char __user *buf, 333 size_t count, loff_t *pos) 334 { 335 struct ib_umad_file *file = filp->private_data; 336 struct ib_umad_packet *packet; 337 ssize_t ret; 338 339 if (count < hdr_size(file)) 340 return -EINVAL; 341 342 mutex_lock(&file->mutex); 343 344 while (list_empty(&file->recv_list)) { 345 mutex_unlock(&file->mutex); 346 347 if (filp->f_flags & O_NONBLOCK) 348 return -EAGAIN; 349 350 if (wait_event_interruptible(file->recv_wait, 351 !list_empty(&file->recv_list))) 352 return -ERESTARTSYS; 353 354 mutex_lock(&file->mutex); 355 } 356 357 packet = list_entry(file->recv_list.next, struct ib_umad_packet, list); 358 list_del(&packet->list); 359 360 mutex_unlock(&file->mutex); 361 362 if (packet->recv_wc) 363 ret = copy_recv_mad(file, buf, packet, count); 364 else 365 ret = copy_send_mad(file, buf, packet, count); 366 367 if (ret < 0) { 368 /* Requeue packet */ 369 mutex_lock(&file->mutex); 370 list_add(&packet->list, &file->recv_list); 371 mutex_unlock(&file->mutex); 372 } else { 373 if (packet->recv_wc) 374 ib_free_recv_mad(packet->recv_wc); 375 kfree(packet); 376 } 377 return ret; 378 } 379 380 static int copy_rmpp_mad(struct ib_mad_send_buf *msg, const char __user *buf) 381 { 382 int left, seg; 383 384 /* Copy class specific header */ 385 if ((msg->hdr_len > IB_MGMT_RMPP_HDR) && 386 copy_from_user(msg->mad + IB_MGMT_RMPP_HDR, buf + IB_MGMT_RMPP_HDR, 387 msg->hdr_len - IB_MGMT_RMPP_HDR)) 388 return -EFAULT; 389 390 /* All headers are in place. Copy data segments. */ 391 for (seg = 1, left = msg->data_len, buf += msg->hdr_len; left > 0; 392 seg++, left -= msg->seg_size, buf += msg->seg_size) { 393 if (copy_from_user(ib_get_rmpp_segment(msg, seg), buf, 394 min(left, msg->seg_size))) 395 return -EFAULT; 396 } 397 return 0; 398 } 399 400 static int same_destination(struct ib_user_mad_hdr *hdr1, 401 struct ib_user_mad_hdr *hdr2) 402 { 403 if (!hdr1->grh_present && !hdr2->grh_present) 404 return (hdr1->lid == hdr2->lid); 405 406 if (hdr1->grh_present && hdr2->grh_present) 407 return !memcmp(hdr1->gid, hdr2->gid, 16); 408 409 return 0; 410 } 411 412 static int is_duplicate(struct ib_umad_file *file, 413 struct ib_umad_packet *packet) 414 { 415 struct ib_umad_packet *sent_packet; 416 struct ib_mad_hdr *sent_hdr, *hdr; 417 418 hdr = (struct ib_mad_hdr *) packet->mad.data; 419 list_for_each_entry(sent_packet, &file->send_list, list) { 420 sent_hdr = (struct ib_mad_hdr *) sent_packet->mad.data; 421 422 if ((hdr->tid != sent_hdr->tid) || 423 (hdr->mgmt_class != sent_hdr->mgmt_class)) 424 continue; 425 426 /* 427 * No need to be overly clever here. If two new operations have 428 * the same TID, reject the second as a duplicate. This is more 429 * restrictive than required by the spec. 430 */ 431 if (!ib_response_mad(hdr)) { 432 if (!ib_response_mad(sent_hdr)) 433 return 1; 434 continue; 435 } else if (!ib_response_mad(sent_hdr)) 436 continue; 437 438 if (same_destination(&packet->mad.hdr, &sent_packet->mad.hdr)) 439 return 1; 440 } 441 442 return 0; 443 } 444 445 static ssize_t ib_umad_write(struct file *filp, const char __user *buf, 446 size_t count, loff_t *pos) 447 { 448 struct ib_umad_file *file = filp->private_data; 449 struct ib_umad_packet *packet; 450 struct ib_mad_agent *agent; 451 struct ib_ah_attr ah_attr; 452 struct ib_ah *ah; 453 struct ib_rmpp_mad *rmpp_mad; 454 __be64 *tid; 455 int ret, data_len, hdr_len, copy_offset, rmpp_active; 456 u8 base_version; 457 458 if (count < hdr_size(file) + IB_MGMT_RMPP_HDR) 459 return -EINVAL; 460 461 packet = kzalloc(sizeof *packet + IB_MGMT_RMPP_HDR, GFP_KERNEL); 462 if (!packet) 463 return -ENOMEM; 464 465 if (copy_from_user(&packet->mad, buf, hdr_size(file))) { 466 ret = -EFAULT; 467 goto err; 468 } 469 470 if (packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) { 471 ret = -EINVAL; 472 goto err; 473 } 474 475 buf += hdr_size(file); 476 477 if (copy_from_user(packet->mad.data, buf, IB_MGMT_RMPP_HDR)) { 478 ret = -EFAULT; 479 goto err; 480 } 481 482 mutex_lock(&file->mutex); 483 484 agent = __get_agent(file, packet->mad.hdr.id); 485 if (!agent) { 486 ret = -EINVAL; 487 goto err_up; 488 } 489 490 memset(&ah_attr, 0, sizeof ah_attr); 491 ah_attr.dlid = be16_to_cpu(packet->mad.hdr.lid); 492 ah_attr.sl = packet->mad.hdr.sl; 493 ah_attr.src_path_bits = packet->mad.hdr.path_bits; 494 ah_attr.port_num = file->port->port_num; 495 if (packet->mad.hdr.grh_present) { 496 ah_attr.ah_flags = IB_AH_GRH; 497 memcpy(ah_attr.grh.dgid.raw, packet->mad.hdr.gid, 16); 498 ah_attr.grh.sgid_index = packet->mad.hdr.gid_index; 499 ah_attr.grh.flow_label = be32_to_cpu(packet->mad.hdr.flow_label); 500 ah_attr.grh.hop_limit = packet->mad.hdr.hop_limit; 501 ah_attr.grh.traffic_class = packet->mad.hdr.traffic_class; 502 } 503 504 ah = ib_create_ah(agent->qp->pd, &ah_attr); 505 if (IS_ERR(ah)) { 506 ret = PTR_ERR(ah); 507 goto err_up; 508 } 509 510 rmpp_mad = (struct ib_rmpp_mad *) packet->mad.data; 511 hdr_len = ib_get_mad_data_offset(rmpp_mad->mad_hdr.mgmt_class); 512 513 if (ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class) 514 && ib_mad_kernel_rmpp_agent(agent)) { 515 copy_offset = IB_MGMT_RMPP_HDR; 516 rmpp_active = ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) & 517 IB_MGMT_RMPP_FLAG_ACTIVE; 518 } else { 519 copy_offset = IB_MGMT_MAD_HDR; 520 rmpp_active = 0; 521 } 522 523 base_version = ((struct ib_mad_hdr *)&packet->mad.data)->base_version; 524 data_len = count - hdr_size(file) - hdr_len; 525 packet->msg = ib_create_send_mad(agent, 526 be32_to_cpu(packet->mad.hdr.qpn), 527 packet->mad.hdr.pkey_index, rmpp_active, 528 hdr_len, data_len, GFP_KERNEL, 529 base_version); 530 if (IS_ERR(packet->msg)) { 531 ret = PTR_ERR(packet->msg); 532 goto err_ah; 533 } 534 535 packet->msg->ah = ah; 536 packet->msg->timeout_ms = packet->mad.hdr.timeout_ms; 537 packet->msg->retries = packet->mad.hdr.retries; 538 packet->msg->context[0] = packet; 539 540 /* Copy MAD header. Any RMPP header is already in place. */ 541 memcpy(packet->msg->mad, packet->mad.data, IB_MGMT_MAD_HDR); 542 543 if (!rmpp_active) { 544 if (copy_from_user(packet->msg->mad + copy_offset, 545 buf + copy_offset, 546 hdr_len + data_len - copy_offset)) { 547 ret = -EFAULT; 548 goto err_msg; 549 } 550 } else { 551 ret = copy_rmpp_mad(packet->msg, buf); 552 if (ret) 553 goto err_msg; 554 } 555 556 /* 557 * Set the high-order part of the transaction ID to make MADs from 558 * different agents unique, and allow routing responses back to the 559 * original requestor. 560 */ 561 if (!ib_response_mad(packet->msg->mad)) { 562 tid = &((struct ib_mad_hdr *) packet->msg->mad)->tid; 563 *tid = cpu_to_be64(((u64) agent->hi_tid) << 32 | 564 (be64_to_cpup(tid) & 0xffffffff)); 565 rmpp_mad->mad_hdr.tid = *tid; 566 } 567 568 if (!ib_mad_kernel_rmpp_agent(agent) 569 && ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class) 570 && (ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) & IB_MGMT_RMPP_FLAG_ACTIVE)) { 571 spin_lock_irq(&file->send_lock); 572 list_add_tail(&packet->list, &file->send_list); 573 spin_unlock_irq(&file->send_lock); 574 } else { 575 spin_lock_irq(&file->send_lock); 576 ret = is_duplicate(file, packet); 577 if (!ret) 578 list_add_tail(&packet->list, &file->send_list); 579 spin_unlock_irq(&file->send_lock); 580 if (ret) { 581 ret = -EINVAL; 582 goto err_msg; 583 } 584 } 585 586 ret = ib_post_send_mad(packet->msg, NULL); 587 if (ret) 588 goto err_send; 589 590 mutex_unlock(&file->mutex); 591 return count; 592 593 err_send: 594 dequeue_send(file, packet); 595 err_msg: 596 ib_free_send_mad(packet->msg); 597 err_ah: 598 ib_destroy_ah(ah); 599 err_up: 600 mutex_unlock(&file->mutex); 601 err: 602 kfree(packet); 603 return ret; 604 } 605 606 static unsigned int ib_umad_poll(struct file *filp, struct poll_table_struct *wait) 607 { 608 struct ib_umad_file *file = filp->private_data; 609 610 /* we will always be able to post a MAD send */ 611 unsigned int mask = POLLOUT | POLLWRNORM; 612 613 poll_wait(filp, &file->recv_wait, wait); 614 615 if (!list_empty(&file->recv_list)) 616 mask |= POLLIN | POLLRDNORM; 617 618 return mask; 619 } 620 621 static int ib_umad_reg_agent(struct ib_umad_file *file, void __user *arg, 622 int compat_method_mask) 623 { 624 struct ib_user_mad_reg_req ureq; 625 struct ib_mad_reg_req req; 626 struct ib_mad_agent *agent = NULL; 627 int agent_id; 628 int ret; 629 630 mutex_lock(&file->port->file_mutex); 631 mutex_lock(&file->mutex); 632 633 if (!file->port->ib_dev) { 634 dev_notice(file->port->dev, 635 "ib_umad_reg_agent: invalid device\n"); 636 ret = -EPIPE; 637 goto out; 638 } 639 640 if (copy_from_user(&ureq, arg, sizeof ureq)) { 641 ret = -EFAULT; 642 goto out; 643 } 644 645 if (ureq.qpn != 0 && ureq.qpn != 1) { 646 dev_notice(file->port->dev, 647 "ib_umad_reg_agent: invalid QPN %d specified\n", 648 ureq.qpn); 649 ret = -EINVAL; 650 goto out; 651 } 652 653 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id) 654 if (!__get_agent(file, agent_id)) 655 goto found; 656 657 dev_notice(file->port->dev, 658 "ib_umad_reg_agent: Max Agents (%u) reached\n", 659 IB_UMAD_MAX_AGENTS); 660 ret = -ENOMEM; 661 goto out; 662 663 found: 664 if (ureq.mgmt_class) { 665 memset(&req, 0, sizeof(req)); 666 req.mgmt_class = ureq.mgmt_class; 667 req.mgmt_class_version = ureq.mgmt_class_version; 668 memcpy(req.oui, ureq.oui, sizeof req.oui); 669 670 if (compat_method_mask) { 671 u32 *umm = (u32 *) ureq.method_mask; 672 int i; 673 674 for (i = 0; i < BITS_TO_LONGS(IB_MGMT_MAX_METHODS); ++i) 675 req.method_mask[i] = 676 umm[i * 2] | ((u64) umm[i * 2 + 1] << 32); 677 } else 678 memcpy(req.method_mask, ureq.method_mask, 679 sizeof req.method_mask); 680 } 681 682 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num, 683 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI, 684 ureq.mgmt_class ? &req : NULL, 685 ureq.rmpp_version, 686 send_handler, recv_handler, file, 0); 687 if (IS_ERR(agent)) { 688 ret = PTR_ERR(agent); 689 agent = NULL; 690 goto out; 691 } 692 693 if (put_user(agent_id, 694 (u32 __user *) (arg + offsetof(struct ib_user_mad_reg_req, id)))) { 695 ret = -EFAULT; 696 goto out; 697 } 698 699 if (!file->already_used) { 700 file->already_used = 1; 701 if (!file->use_pkey_index) { 702 dev_warn(file->port->dev, 703 "process %s did not enable P_Key index support.\n", 704 current->comm); 705 dev_warn(file->port->dev, 706 " Documentation/infiniband/user_mad.txt has info on the new ABI.\n"); 707 } 708 } 709 710 file->agent[agent_id] = agent; 711 ret = 0; 712 713 out: 714 mutex_unlock(&file->mutex); 715 716 if (ret && agent) 717 ib_unregister_mad_agent(agent); 718 719 mutex_unlock(&file->port->file_mutex); 720 721 return ret; 722 } 723 724 static int ib_umad_reg_agent2(struct ib_umad_file *file, void __user *arg) 725 { 726 struct ib_user_mad_reg_req2 ureq; 727 struct ib_mad_reg_req req; 728 struct ib_mad_agent *agent = NULL; 729 int agent_id; 730 int ret; 731 732 mutex_lock(&file->port->file_mutex); 733 mutex_lock(&file->mutex); 734 735 if (!file->port->ib_dev) { 736 dev_notice(file->port->dev, 737 "ib_umad_reg_agent2: invalid device\n"); 738 ret = -EPIPE; 739 goto out; 740 } 741 742 if (copy_from_user(&ureq, arg, sizeof(ureq))) { 743 ret = -EFAULT; 744 goto out; 745 } 746 747 if (ureq.qpn != 0 && ureq.qpn != 1) { 748 dev_notice(file->port->dev, 749 "ib_umad_reg_agent2: invalid QPN %d specified\n", 750 ureq.qpn); 751 ret = -EINVAL; 752 goto out; 753 } 754 755 if (ureq.flags & ~IB_USER_MAD_REG_FLAGS_CAP) { 756 dev_notice(file->port->dev, 757 "ib_umad_reg_agent2 failed: invalid registration flags specified 0x%x; supported 0x%x\n", 758 ureq.flags, IB_USER_MAD_REG_FLAGS_CAP); 759 ret = -EINVAL; 760 761 if (put_user((u32)IB_USER_MAD_REG_FLAGS_CAP, 762 (u32 __user *) (arg + offsetof(struct 763 ib_user_mad_reg_req2, flags)))) 764 ret = -EFAULT; 765 766 goto out; 767 } 768 769 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id) 770 if (!__get_agent(file, agent_id)) 771 goto found; 772 773 dev_notice(file->port->dev, 774 "ib_umad_reg_agent2: Max Agents (%u) reached\n", 775 IB_UMAD_MAX_AGENTS); 776 ret = -ENOMEM; 777 goto out; 778 779 found: 780 if (ureq.mgmt_class) { 781 memset(&req, 0, sizeof(req)); 782 req.mgmt_class = ureq.mgmt_class; 783 req.mgmt_class_version = ureq.mgmt_class_version; 784 if (ureq.oui & 0xff000000) { 785 dev_notice(file->port->dev, 786 "ib_umad_reg_agent2 failed: oui invalid 0x%08x\n", 787 ureq.oui); 788 ret = -EINVAL; 789 goto out; 790 } 791 req.oui[2] = ureq.oui & 0x0000ff; 792 req.oui[1] = (ureq.oui & 0x00ff00) >> 8; 793 req.oui[0] = (ureq.oui & 0xff0000) >> 16; 794 memcpy(req.method_mask, ureq.method_mask, 795 sizeof(req.method_mask)); 796 } 797 798 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num, 799 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI, 800 ureq.mgmt_class ? &req : NULL, 801 ureq.rmpp_version, 802 send_handler, recv_handler, file, 803 ureq.flags); 804 if (IS_ERR(agent)) { 805 ret = PTR_ERR(agent); 806 agent = NULL; 807 goto out; 808 } 809 810 if (put_user(agent_id, 811 (u32 __user *)(arg + 812 offsetof(struct ib_user_mad_reg_req2, id)))) { 813 ret = -EFAULT; 814 goto out; 815 } 816 817 if (!file->already_used) { 818 file->already_used = 1; 819 file->use_pkey_index = 1; 820 } 821 822 file->agent[agent_id] = agent; 823 ret = 0; 824 825 out: 826 mutex_unlock(&file->mutex); 827 828 if (ret && agent) 829 ib_unregister_mad_agent(agent); 830 831 mutex_unlock(&file->port->file_mutex); 832 833 return ret; 834 } 835 836 837 static int ib_umad_unreg_agent(struct ib_umad_file *file, u32 __user *arg) 838 { 839 struct ib_mad_agent *agent = NULL; 840 u32 id; 841 int ret = 0; 842 843 if (get_user(id, arg)) 844 return -EFAULT; 845 846 mutex_lock(&file->port->file_mutex); 847 mutex_lock(&file->mutex); 848 849 if (id >= IB_UMAD_MAX_AGENTS || !__get_agent(file, id)) { 850 ret = -EINVAL; 851 goto out; 852 } 853 854 agent = file->agent[id]; 855 file->agent[id] = NULL; 856 857 out: 858 mutex_unlock(&file->mutex); 859 860 if (agent) 861 ib_unregister_mad_agent(agent); 862 863 mutex_unlock(&file->port->file_mutex); 864 865 return ret; 866 } 867 868 static long ib_umad_enable_pkey(struct ib_umad_file *file) 869 { 870 int ret = 0; 871 872 mutex_lock(&file->mutex); 873 if (file->already_used) 874 ret = -EINVAL; 875 else 876 file->use_pkey_index = 1; 877 mutex_unlock(&file->mutex); 878 879 return ret; 880 } 881 882 static long ib_umad_ioctl(struct file *filp, unsigned int cmd, 883 unsigned long arg) 884 { 885 switch (cmd) { 886 case IB_USER_MAD_REGISTER_AGENT: 887 return ib_umad_reg_agent(filp->private_data, (void __user *) arg, 0); 888 case IB_USER_MAD_UNREGISTER_AGENT: 889 return ib_umad_unreg_agent(filp->private_data, (__u32 __user *) arg); 890 case IB_USER_MAD_ENABLE_PKEY: 891 return ib_umad_enable_pkey(filp->private_data); 892 case IB_USER_MAD_REGISTER_AGENT2: 893 return ib_umad_reg_agent2(filp->private_data, (void __user *) arg); 894 default: 895 return -ENOIOCTLCMD; 896 } 897 } 898 899 #ifdef CONFIG_COMPAT 900 static long ib_umad_compat_ioctl(struct file *filp, unsigned int cmd, 901 unsigned long arg) 902 { 903 switch (cmd) { 904 case IB_USER_MAD_REGISTER_AGENT: 905 return ib_umad_reg_agent(filp->private_data, compat_ptr(arg), 1); 906 case IB_USER_MAD_UNREGISTER_AGENT: 907 return ib_umad_unreg_agent(filp->private_data, compat_ptr(arg)); 908 case IB_USER_MAD_ENABLE_PKEY: 909 return ib_umad_enable_pkey(filp->private_data); 910 case IB_USER_MAD_REGISTER_AGENT2: 911 return ib_umad_reg_agent2(filp->private_data, compat_ptr(arg)); 912 default: 913 return -ENOIOCTLCMD; 914 } 915 } 916 #endif 917 918 /* 919 * ib_umad_open() does not need the BKL: 920 * 921 * - the ib_umad_port structures are properly reference counted, and 922 * everything else is purely local to the file being created, so 923 * races against other open calls are not a problem; 924 * - the ioctl method does not affect any global state outside of the 925 * file structure being operated on; 926 */ 927 static int ib_umad_open(struct inode *inode, struct file *filp) 928 { 929 struct ib_umad_port *port; 930 struct ib_umad_file *file; 931 int ret = -ENXIO; 932 933 port = container_of(inode->i_cdev, struct ib_umad_port, cdev); 934 935 mutex_lock(&port->file_mutex); 936 937 if (!port->ib_dev) 938 goto out; 939 940 ret = -ENOMEM; 941 file = kzalloc(sizeof *file, GFP_KERNEL); 942 if (!file) 943 goto out; 944 945 mutex_init(&file->mutex); 946 spin_lock_init(&file->send_lock); 947 INIT_LIST_HEAD(&file->recv_list); 948 INIT_LIST_HEAD(&file->send_list); 949 init_waitqueue_head(&file->recv_wait); 950 951 file->port = port; 952 filp->private_data = file; 953 954 list_add_tail(&file->port_list, &port->file_list); 955 956 ret = nonseekable_open(inode, filp); 957 if (ret) { 958 list_del(&file->port_list); 959 kfree(file); 960 goto out; 961 } 962 963 kobject_get(&port->umad_dev->kobj); 964 965 out: 966 mutex_unlock(&port->file_mutex); 967 return ret; 968 } 969 970 static int ib_umad_close(struct inode *inode, struct file *filp) 971 { 972 struct ib_umad_file *file = filp->private_data; 973 struct ib_umad_device *dev = file->port->umad_dev; 974 struct ib_umad_packet *packet, *tmp; 975 int already_dead; 976 int i; 977 978 mutex_lock(&file->port->file_mutex); 979 mutex_lock(&file->mutex); 980 981 already_dead = file->agents_dead; 982 file->agents_dead = 1; 983 984 list_for_each_entry_safe(packet, tmp, &file->recv_list, list) { 985 if (packet->recv_wc) 986 ib_free_recv_mad(packet->recv_wc); 987 kfree(packet); 988 } 989 990 list_del(&file->port_list); 991 992 mutex_unlock(&file->mutex); 993 994 if (!already_dead) 995 for (i = 0; i < IB_UMAD_MAX_AGENTS; ++i) 996 if (file->agent[i]) 997 ib_unregister_mad_agent(file->agent[i]); 998 999 mutex_unlock(&file->port->file_mutex); 1000 1001 kfree(file); 1002 kobject_put(&dev->kobj); 1003 1004 return 0; 1005 } 1006 1007 static const struct file_operations umad_fops = { 1008 .owner = THIS_MODULE, 1009 .read = ib_umad_read, 1010 .write = ib_umad_write, 1011 .poll = ib_umad_poll, 1012 .unlocked_ioctl = ib_umad_ioctl, 1013 #ifdef CONFIG_COMPAT 1014 .compat_ioctl = ib_umad_compat_ioctl, 1015 #endif 1016 .open = ib_umad_open, 1017 .release = ib_umad_close, 1018 .llseek = no_llseek, 1019 }; 1020 1021 static int ib_umad_sm_open(struct inode *inode, struct file *filp) 1022 { 1023 struct ib_umad_port *port; 1024 struct ib_port_modify props = { 1025 .set_port_cap_mask = IB_PORT_SM 1026 }; 1027 int ret; 1028 1029 port = container_of(inode->i_cdev, struct ib_umad_port, sm_cdev); 1030 1031 if (filp->f_flags & O_NONBLOCK) { 1032 if (down_trylock(&port->sm_sem)) { 1033 ret = -EAGAIN; 1034 goto fail; 1035 } 1036 } else { 1037 if (down_interruptible(&port->sm_sem)) { 1038 ret = -ERESTARTSYS; 1039 goto fail; 1040 } 1041 } 1042 1043 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props); 1044 if (ret) 1045 goto err_up_sem; 1046 1047 filp->private_data = port; 1048 1049 ret = nonseekable_open(inode, filp); 1050 if (ret) 1051 goto err_clr_sm_cap; 1052 1053 kobject_get(&port->umad_dev->kobj); 1054 1055 return 0; 1056 1057 err_clr_sm_cap: 1058 swap(props.set_port_cap_mask, props.clr_port_cap_mask); 1059 ib_modify_port(port->ib_dev, port->port_num, 0, &props); 1060 1061 err_up_sem: 1062 up(&port->sm_sem); 1063 1064 fail: 1065 return ret; 1066 } 1067 1068 static int ib_umad_sm_close(struct inode *inode, struct file *filp) 1069 { 1070 struct ib_umad_port *port = filp->private_data; 1071 struct ib_port_modify props = { 1072 .clr_port_cap_mask = IB_PORT_SM 1073 }; 1074 int ret = 0; 1075 1076 mutex_lock(&port->file_mutex); 1077 if (port->ib_dev) 1078 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props); 1079 mutex_unlock(&port->file_mutex); 1080 1081 up(&port->sm_sem); 1082 1083 kobject_put(&port->umad_dev->kobj); 1084 1085 return ret; 1086 } 1087 1088 static const struct file_operations umad_sm_fops = { 1089 .owner = THIS_MODULE, 1090 .open = ib_umad_sm_open, 1091 .release = ib_umad_sm_close, 1092 .llseek = no_llseek, 1093 }; 1094 1095 static struct ib_client umad_client = { 1096 .name = "umad", 1097 .add = ib_umad_add_one, 1098 .remove = ib_umad_remove_one 1099 }; 1100 1101 static ssize_t show_ibdev(struct device *dev, struct device_attribute *attr, 1102 char *buf) 1103 { 1104 struct ib_umad_port *port = dev_get_drvdata(dev); 1105 1106 if (!port) 1107 return -ENODEV; 1108 1109 return sprintf(buf, "%s\n", port->ib_dev->name); 1110 } 1111 static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL); 1112 1113 static ssize_t show_port(struct device *dev, struct device_attribute *attr, 1114 char *buf) 1115 { 1116 struct ib_umad_port *port = dev_get_drvdata(dev); 1117 1118 if (!port) 1119 return -ENODEV; 1120 1121 return sprintf(buf, "%d\n", port->port_num); 1122 } 1123 static DEVICE_ATTR(port, S_IRUGO, show_port, NULL); 1124 1125 static CLASS_ATTR_STRING(abi_version, S_IRUGO, 1126 __stringify(IB_USER_MAD_ABI_VERSION)); 1127 1128 static dev_t overflow_maj; 1129 static DECLARE_BITMAP(overflow_map, IB_UMAD_MAX_PORTS); 1130 static int find_overflow_devnum(struct ib_device *device) 1131 { 1132 int ret; 1133 1134 if (!overflow_maj) { 1135 ret = alloc_chrdev_region(&overflow_maj, 0, IB_UMAD_MAX_PORTS * 2, 1136 "infiniband_mad"); 1137 if (ret) { 1138 dev_err(&device->dev, 1139 "couldn't register dynamic device number\n"); 1140 return ret; 1141 } 1142 } 1143 1144 ret = find_first_zero_bit(overflow_map, IB_UMAD_MAX_PORTS); 1145 if (ret >= IB_UMAD_MAX_PORTS) 1146 return -1; 1147 1148 return ret; 1149 } 1150 1151 static int ib_umad_init_port(struct ib_device *device, int port_num, 1152 struct ib_umad_device *umad_dev, 1153 struct ib_umad_port *port) 1154 { 1155 int devnum; 1156 dev_t base; 1157 1158 spin_lock(&port_lock); 1159 devnum = find_first_zero_bit(dev_map, IB_UMAD_MAX_PORTS); 1160 if (devnum >= IB_UMAD_MAX_PORTS) { 1161 spin_unlock(&port_lock); 1162 devnum = find_overflow_devnum(device); 1163 if (devnum < 0) 1164 return -1; 1165 1166 spin_lock(&port_lock); 1167 port->dev_num = devnum + IB_UMAD_MAX_PORTS; 1168 base = devnum + overflow_maj; 1169 set_bit(devnum, overflow_map); 1170 } else { 1171 port->dev_num = devnum; 1172 base = devnum + base_dev; 1173 set_bit(devnum, dev_map); 1174 } 1175 spin_unlock(&port_lock); 1176 1177 port->ib_dev = device; 1178 port->port_num = port_num; 1179 sema_init(&port->sm_sem, 1); 1180 mutex_init(&port->file_mutex); 1181 INIT_LIST_HEAD(&port->file_list); 1182 1183 cdev_init(&port->cdev, &umad_fops); 1184 port->cdev.owner = THIS_MODULE; 1185 port->cdev.kobj.parent = &umad_dev->kobj; 1186 kobject_set_name(&port->cdev.kobj, "umad%d", port->dev_num); 1187 if (cdev_add(&port->cdev, base, 1)) 1188 goto err_cdev; 1189 1190 port->dev = device_create(umad_class, device->dma_device, 1191 port->cdev.dev, port, 1192 "umad%d", port->dev_num); 1193 if (IS_ERR(port->dev)) 1194 goto err_cdev; 1195 1196 if (device_create_file(port->dev, &dev_attr_ibdev)) 1197 goto err_dev; 1198 if (device_create_file(port->dev, &dev_attr_port)) 1199 goto err_dev; 1200 1201 base += IB_UMAD_MAX_PORTS; 1202 cdev_init(&port->sm_cdev, &umad_sm_fops); 1203 port->sm_cdev.owner = THIS_MODULE; 1204 port->sm_cdev.kobj.parent = &umad_dev->kobj; 1205 kobject_set_name(&port->sm_cdev.kobj, "issm%d", port->dev_num); 1206 if (cdev_add(&port->sm_cdev, base, 1)) 1207 goto err_sm_cdev; 1208 1209 port->sm_dev = device_create(umad_class, device->dma_device, 1210 port->sm_cdev.dev, port, 1211 "issm%d", port->dev_num); 1212 if (IS_ERR(port->sm_dev)) 1213 goto err_sm_cdev; 1214 1215 if (device_create_file(port->sm_dev, &dev_attr_ibdev)) 1216 goto err_sm_dev; 1217 if (device_create_file(port->sm_dev, &dev_attr_port)) 1218 goto err_sm_dev; 1219 1220 return 0; 1221 1222 err_sm_dev: 1223 device_destroy(umad_class, port->sm_cdev.dev); 1224 1225 err_sm_cdev: 1226 cdev_del(&port->sm_cdev); 1227 1228 err_dev: 1229 device_destroy(umad_class, port->cdev.dev); 1230 1231 err_cdev: 1232 cdev_del(&port->cdev); 1233 if (port->dev_num < IB_UMAD_MAX_PORTS) 1234 clear_bit(devnum, dev_map); 1235 else 1236 clear_bit(devnum, overflow_map); 1237 1238 return -1; 1239 } 1240 1241 static void ib_umad_kill_port(struct ib_umad_port *port) 1242 { 1243 struct ib_umad_file *file; 1244 int id; 1245 1246 dev_set_drvdata(port->dev, NULL); 1247 dev_set_drvdata(port->sm_dev, NULL); 1248 1249 device_destroy(umad_class, port->cdev.dev); 1250 device_destroy(umad_class, port->sm_cdev.dev); 1251 1252 cdev_del(&port->cdev); 1253 cdev_del(&port->sm_cdev); 1254 1255 mutex_lock(&port->file_mutex); 1256 1257 port->ib_dev = NULL; 1258 1259 list_for_each_entry(file, &port->file_list, port_list) { 1260 mutex_lock(&file->mutex); 1261 file->agents_dead = 1; 1262 mutex_unlock(&file->mutex); 1263 1264 for (id = 0; id < IB_UMAD_MAX_AGENTS; ++id) 1265 if (file->agent[id]) 1266 ib_unregister_mad_agent(file->agent[id]); 1267 } 1268 1269 mutex_unlock(&port->file_mutex); 1270 1271 if (port->dev_num < IB_UMAD_MAX_PORTS) 1272 clear_bit(port->dev_num, dev_map); 1273 else 1274 clear_bit(port->dev_num - IB_UMAD_MAX_PORTS, overflow_map); 1275 } 1276 1277 static void ib_umad_add_one(struct ib_device *device) 1278 { 1279 struct ib_umad_device *umad_dev; 1280 int s, e, i; 1281 int count = 0; 1282 1283 s = rdma_start_port(device); 1284 e = rdma_end_port(device); 1285 1286 umad_dev = kzalloc(sizeof *umad_dev + 1287 (e - s + 1) * sizeof (struct ib_umad_port), 1288 GFP_KERNEL); 1289 if (!umad_dev) 1290 return; 1291 1292 kobject_init(&umad_dev->kobj, &ib_umad_dev_ktype); 1293 1294 for (i = s; i <= e; ++i) { 1295 if (!rdma_cap_ib_mad(device, i)) 1296 continue; 1297 1298 umad_dev->port[i - s].umad_dev = umad_dev; 1299 1300 if (ib_umad_init_port(device, i, umad_dev, 1301 &umad_dev->port[i - s])) 1302 goto err; 1303 1304 count++; 1305 } 1306 1307 if (!count) 1308 goto free; 1309 1310 ib_set_client_data(device, &umad_client, umad_dev); 1311 1312 return; 1313 1314 err: 1315 while (--i >= s) { 1316 if (!rdma_cap_ib_mad(device, i)) 1317 continue; 1318 1319 ib_umad_kill_port(&umad_dev->port[i - s]); 1320 } 1321 free: 1322 kobject_put(&umad_dev->kobj); 1323 } 1324 1325 static void ib_umad_remove_one(struct ib_device *device, void *client_data) 1326 { 1327 struct ib_umad_device *umad_dev = client_data; 1328 int i; 1329 1330 if (!umad_dev) 1331 return; 1332 1333 for (i = 0; i <= rdma_end_port(device) - rdma_start_port(device); ++i) { 1334 if (rdma_cap_ib_mad(device, i + rdma_start_port(device))) 1335 ib_umad_kill_port(&umad_dev->port[i]); 1336 } 1337 1338 kobject_put(&umad_dev->kobj); 1339 } 1340 1341 static char *umad_devnode(struct device *dev, umode_t *mode) 1342 { 1343 return kasprintf(GFP_KERNEL, "infiniband/%s", dev_name(dev)); 1344 } 1345 1346 static int __init ib_umad_init(void) 1347 { 1348 int ret; 1349 1350 ret = register_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2, 1351 "infiniband_mad"); 1352 if (ret) { 1353 pr_err("couldn't register device number\n"); 1354 goto out; 1355 } 1356 1357 umad_class = class_create(THIS_MODULE, "infiniband_mad"); 1358 if (IS_ERR(umad_class)) { 1359 ret = PTR_ERR(umad_class); 1360 pr_err("couldn't create class infiniband_mad\n"); 1361 goto out_chrdev; 1362 } 1363 1364 umad_class->devnode = umad_devnode; 1365 1366 ret = class_create_file(umad_class, &class_attr_abi_version.attr); 1367 if (ret) { 1368 pr_err("couldn't create abi_version attribute\n"); 1369 goto out_class; 1370 } 1371 1372 ret = ib_register_client(&umad_client); 1373 if (ret) { 1374 pr_err("couldn't register ib_umad client\n"); 1375 goto out_class; 1376 } 1377 1378 return 0; 1379 1380 out_class: 1381 class_destroy(umad_class); 1382 1383 out_chrdev: 1384 unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2); 1385 1386 out: 1387 return ret; 1388 } 1389 1390 static void __exit ib_umad_cleanup(void) 1391 { 1392 ib_unregister_client(&umad_client); 1393 class_destroy(umad_class); 1394 unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2); 1395 if (overflow_maj) 1396 unregister_chrdev_region(overflow_maj, IB_UMAD_MAX_PORTS * 2); 1397 } 1398 1399 module_init(ib_umad_init); 1400 module_exit(ib_umad_cleanup); 1401