ib_user_mad.c revision 331769
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/device.h> 40#include <linux/err.h> 41#include <linux/fs.h> 42#include <linux/cdev.h> 43#include <linux/dma-mapping.h> 44#include <linux/poll.h> 45#include <linux/mutex.h> 46#include <linux/kref.h> 47#include <linux/compat.h> 48#include <linux/sched.h> 49#include <linux/semaphore.h> 50#include <linux/slab.h> 51 52#include <asm/uaccess.h> 53 54#include <rdma/ib_mad.h> 55#include <rdma/ib_user_mad.h> 56 57MODULE_AUTHOR("Roland Dreier"); 58MODULE_DESCRIPTION("InfiniBand userspace MAD packet access"); 59MODULE_LICENSE("Dual BSD/GPL"); 60 61enum { 62 IB_UMAD_MAX_PORTS = 64, 63 IB_UMAD_MAX_AGENTS = 32, 64 65 IB_UMAD_MAJOR = 231, 66 IB_UMAD_MINOR_BASE = 0 67}; 68 69/* 70 * Our lifetime rules for these structs are the following: 71 * device special file is opened, we take a reference on the 72 * ib_umad_port's struct ib_umad_device. We drop these 73 * references in the corresponding close(). 74 * 75 * In addition to references coming from open character devices, there 76 * is one more reference to each ib_umad_device representing the 77 * module's reference taken when allocating the ib_umad_device in 78 * ib_umad_add_one(). 79 * 80 * When destroying an ib_umad_device, we drop the module's reference. 81 */ 82 83struct ib_umad_port { 84 struct cdev cdev; 85 struct device *dev; 86 87 struct cdev sm_cdev; 88 struct device *sm_dev; 89 struct semaphore sm_sem; 90 91 struct mutex file_mutex; 92 struct list_head file_list; 93 94 struct ib_device *ib_dev; 95 struct ib_umad_device *umad_dev; 96 int dev_num; 97 u8 port_num; 98}; 99 100struct ib_umad_device { 101 struct kobject kobj; 102 struct ib_umad_port port[0]; 103}; 104 105struct ib_umad_file { 106 struct mutex mutex; 107 struct ib_umad_port *port; 108 struct list_head recv_list; 109 struct list_head send_list; 110 struct list_head port_list; 111 spinlock_t send_lock; 112 wait_queue_head_t recv_wait; 113 struct ib_mad_agent *agent[IB_UMAD_MAX_AGENTS]; 114 int agents_dead; 115 u8 use_pkey_index; 116 u8 already_used; 117}; 118 119struct ib_umad_packet { 120 struct ib_mad_send_buf *msg; 121 struct ib_mad_recv_wc *recv_wc; 122 struct list_head list; 123 int length; 124 struct ib_user_mad mad; 125}; 126 127static struct class *umad_class; 128 129static const dev_t base_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE); 130 131static DEFINE_SPINLOCK(port_lock); 132static DECLARE_BITMAP(dev_map, IB_UMAD_MAX_PORTS); 133 134static void ib_umad_add_one(struct ib_device *device); 135static void ib_umad_remove_one(struct ib_device *device, void *client_data); 136 137static void ib_umad_release_dev(struct kobject *kobj) 138{ 139 struct ib_umad_device *dev = 140 container_of(kobj, struct ib_umad_device, kobj); 141 142 kfree(dev); 143} 144 145static struct kobj_type ib_umad_dev_ktype = { 146 .release = ib_umad_release_dev, 147}; 148 149static int hdr_size(struct ib_umad_file *file) 150{ 151 return file->use_pkey_index ? sizeof (struct ib_user_mad_hdr) : 152 sizeof (struct ib_user_mad_hdr_old); 153} 154 155/* caller must hold file->mutex */ 156static struct ib_mad_agent *__get_agent(struct ib_umad_file *file, int id) 157{ 158 return file->agents_dead ? NULL : file->agent[id]; 159} 160 161static int queue_packet(struct ib_umad_file *file, 162 struct ib_mad_agent *agent, 163 struct ib_umad_packet *packet) 164{ 165 int ret = 1; 166 167 mutex_lock(&file->mutex); 168 169 for (packet->mad.hdr.id = 0; 170 packet->mad.hdr.id < IB_UMAD_MAX_AGENTS; 171 packet->mad.hdr.id++) 172 if (agent == __get_agent(file, packet->mad.hdr.id)) { 173 list_add_tail(&packet->list, &file->recv_list); 174 wake_up_interruptible(&file->recv_wait); 175 ret = 0; 176 break; 177 } 178 179 mutex_unlock(&file->mutex); 180 181 return ret; 182} 183 184static void dequeue_send(struct ib_umad_file *file, 185 struct ib_umad_packet *packet) 186{ 187 spin_lock_irq(&file->send_lock); 188 list_del(&packet->list); 189 spin_unlock_irq(&file->send_lock); 190} 191 192static void send_handler(struct ib_mad_agent *agent, 193 struct ib_mad_send_wc *send_wc) 194{ 195 struct ib_umad_file *file = agent->context; 196 struct ib_umad_packet *packet = send_wc->send_buf->context[0]; 197 198 dequeue_send(file, packet); 199 ib_destroy_ah(packet->msg->ah); 200 ib_free_send_mad(packet->msg); 201 202 if (send_wc->status == IB_WC_RESP_TIMEOUT_ERR) { 203 packet->length = IB_MGMT_MAD_HDR; 204 packet->mad.hdr.status = ETIMEDOUT; 205 if (!queue_packet(file, agent, packet)) 206 return; 207 } 208 kfree(packet); 209} 210 211static void recv_handler(struct ib_mad_agent *agent, 212 struct ib_mad_send_buf *send_buf, 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 254err2: 255 kfree(packet); 256err1: 257 ib_free_recv_mad(mad_recv_wc); 258} 259 260static 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, (char *)recv_buf->mad + offset, 306 seg_payload)) 307 return -EFAULT; 308 } 309 } 310 return hdr_size(file) + packet->length; 311} 312 313static 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 332static 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 380static 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((char *)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 400static 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 412static 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 445static 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((char *)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 593err_send: 594 dequeue_send(file, packet); 595err_msg: 596 ib_free_send_mad(packet->msg); 597err_ah: 598 ib_destroy_ah(ah); 599err_up: 600 mutex_unlock(&file->mutex); 601err: 602 kfree(packet); 603 return ret; 604} 605 606static 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 621static 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 663found: 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 *) ((char *)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 713out: 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 724static 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 const u32 flags = IB_USER_MAD_REG_FLAGS_CAP; 757 dev_notice(file->port->dev, 758 "ib_umad_reg_agent2 failed: invalid registration flags specified 0x%x; supported 0x%x\n", 759 ureq.flags, IB_USER_MAD_REG_FLAGS_CAP); 760 ret = -EINVAL; 761 762 if (put_user(flags, 763 (u32 __user *) ((char *)arg + offsetof(struct 764 ib_user_mad_reg_req2, flags)))) 765 ret = -EFAULT; 766 767 goto out; 768 } 769 770 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id) 771 if (!__get_agent(file, agent_id)) 772 goto found; 773 774 dev_notice(file->port->dev, 775 "ib_umad_reg_agent2: Max Agents (%u) reached\n", 776 IB_UMAD_MAX_AGENTS); 777 ret = -ENOMEM; 778 goto out; 779 780found: 781 if (ureq.mgmt_class) { 782 memset(&req, 0, sizeof(req)); 783 req.mgmt_class = ureq.mgmt_class; 784 req.mgmt_class_version = ureq.mgmt_class_version; 785 if (ureq.oui & 0xff000000) { 786 dev_notice(file->port->dev, 787 "ib_umad_reg_agent2 failed: oui invalid 0x%08x\n", 788 ureq.oui); 789 ret = -EINVAL; 790 goto out; 791 } 792 req.oui[2] = ureq.oui & 0x0000ff; 793 req.oui[1] = (ureq.oui & 0x00ff00) >> 8; 794 req.oui[0] = (ureq.oui & 0xff0000) >> 16; 795 memcpy(req.method_mask, ureq.method_mask, 796 sizeof(req.method_mask)); 797 } 798 799 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num, 800 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI, 801 ureq.mgmt_class ? &req : NULL, 802 ureq.rmpp_version, 803 send_handler, recv_handler, file, 804 ureq.flags); 805 if (IS_ERR(agent)) { 806 ret = PTR_ERR(agent); 807 agent = NULL; 808 goto out; 809 } 810 811 if (put_user(agent_id, 812 (u32 __user *)((char *)arg + 813 offsetof(struct ib_user_mad_reg_req2, id)))) { 814 ret = -EFAULT; 815 goto out; 816 } 817 818 if (!file->already_used) { 819 file->already_used = 1; 820 file->use_pkey_index = 1; 821 } 822 823 file->agent[agent_id] = agent; 824 ret = 0; 825 826out: 827 mutex_unlock(&file->mutex); 828 829 if (ret && agent) 830 ib_unregister_mad_agent(agent); 831 832 mutex_unlock(&file->port->file_mutex); 833 834 return ret; 835} 836 837 838static int ib_umad_unreg_agent(struct ib_umad_file *file, u32 __user *arg) 839{ 840 struct ib_mad_agent *agent = NULL; 841 u32 id; 842 int ret = 0; 843 844 if (get_user(id, arg)) 845 return -EFAULT; 846 847 mutex_lock(&file->port->file_mutex); 848 mutex_lock(&file->mutex); 849 850 if (id >= IB_UMAD_MAX_AGENTS || !__get_agent(file, id)) { 851 ret = -EINVAL; 852 goto out; 853 } 854 855 agent = file->agent[id]; 856 file->agent[id] = NULL; 857 858out: 859 mutex_unlock(&file->mutex); 860 861 if (agent) 862 ib_unregister_mad_agent(agent); 863 864 mutex_unlock(&file->port->file_mutex); 865 866 return ret; 867} 868 869static long ib_umad_enable_pkey(struct ib_umad_file *file) 870{ 871 int ret = 0; 872 873 mutex_lock(&file->mutex); 874 if (file->already_used) 875 ret = -EINVAL; 876 else 877 file->use_pkey_index = 1; 878 mutex_unlock(&file->mutex); 879 880 return ret; 881} 882 883static long ib_umad_ioctl(struct file *filp, unsigned int cmd, 884 unsigned long arg) 885{ 886 switch (cmd) { 887 case IB_USER_MAD_REGISTER_AGENT: 888 return ib_umad_reg_agent(filp->private_data, (void __user *) arg, 0); 889 case IB_USER_MAD_UNREGISTER_AGENT: 890 return ib_umad_unreg_agent(filp->private_data, (__u32 __user *) arg); 891 case IB_USER_MAD_ENABLE_PKEY: 892 return ib_umad_enable_pkey(filp->private_data); 893 case IB_USER_MAD_REGISTER_AGENT2: 894 return ib_umad_reg_agent2(filp->private_data, (void __user *) arg); 895 default: 896 return -ENOIOCTLCMD; 897 } 898} 899 900#ifdef CONFIG_COMPAT 901static long ib_umad_compat_ioctl(struct file *filp, unsigned int cmd, 902 unsigned long arg) 903{ 904 switch (cmd) { 905 case IB_USER_MAD_REGISTER_AGENT: 906 return ib_umad_reg_agent(filp->private_data, compat_ptr(arg), 1); 907 case IB_USER_MAD_UNREGISTER_AGENT: 908 return ib_umad_unreg_agent(filp->private_data, compat_ptr(arg)); 909 case IB_USER_MAD_ENABLE_PKEY: 910 return ib_umad_enable_pkey(filp->private_data); 911 case IB_USER_MAD_REGISTER_AGENT2: 912 return ib_umad_reg_agent2(filp->private_data, compat_ptr(arg)); 913 default: 914 return -ENOIOCTLCMD; 915 } 916} 917#endif 918 919/* 920 * ib_umad_open() does not need the BKL: 921 * 922 * - the ib_umad_port structures are properly reference counted, and 923 * everything else is purely local to the file being created, so 924 * races against other open calls are not a problem; 925 * - the ioctl method does not affect any global state outside of the 926 * file structure being operated on; 927 */ 928static int ib_umad_open(struct inode *inode, struct file *filp) 929{ 930 struct ib_umad_port *port; 931 struct ib_umad_file *file; 932 int ret = -ENXIO; 933 934 port = container_of(inode->i_cdev->si_drv1, struct ib_umad_port, cdev); 935 936 mutex_lock(&port->file_mutex); 937 938 if (!port->ib_dev) 939 goto out; 940 941 ret = -ENOMEM; 942 file = kzalloc(sizeof *file, GFP_KERNEL); 943 if (!file) 944 goto out; 945 946 mutex_init(&file->mutex); 947 spin_lock_init(&file->send_lock); 948 INIT_LIST_HEAD(&file->recv_list); 949 INIT_LIST_HEAD(&file->send_list); 950 init_waitqueue_head(&file->recv_wait); 951 952 file->port = port; 953 filp->private_data = file; 954 955 list_add_tail(&file->port_list, &port->file_list); 956 957 ret = nonseekable_open(inode, filp); 958 if (ret) { 959 list_del(&file->port_list); 960 kfree(file); 961 goto out; 962 } 963 964 kobject_get(&port->umad_dev->kobj); 965 966out: 967 mutex_unlock(&port->file_mutex); 968 return ret; 969} 970 971static int ib_umad_close(struct inode *inode, struct file *filp) 972{ 973 struct ib_umad_file *file = filp->private_data; 974 struct ib_umad_device *dev = file->port->umad_dev; 975 struct ib_umad_packet *packet, *tmp; 976 int already_dead; 977 int i; 978 979 mutex_lock(&file->port->file_mutex); 980 mutex_lock(&file->mutex); 981 982 already_dead = file->agents_dead; 983 file->agents_dead = 1; 984 985 list_for_each_entry_safe(packet, tmp, &file->recv_list, list) { 986 if (packet->recv_wc) 987 ib_free_recv_mad(packet->recv_wc); 988 kfree(packet); 989 } 990 991 list_del(&file->port_list); 992 993 mutex_unlock(&file->mutex); 994 995 if (!already_dead) 996 for (i = 0; i < IB_UMAD_MAX_AGENTS; ++i) 997 if (file->agent[i]) 998 ib_unregister_mad_agent(file->agent[i]); 999 1000 mutex_unlock(&file->port->file_mutex); 1001 1002 kfree(file); 1003 kobject_put(&dev->kobj); 1004 1005 return 0; 1006} 1007 1008static const struct file_operations umad_fops = { 1009 .owner = THIS_MODULE, 1010 .read = ib_umad_read, 1011 .write = ib_umad_write, 1012 .poll = ib_umad_poll, 1013 .unlocked_ioctl = ib_umad_ioctl, 1014#ifdef CONFIG_COMPAT 1015 .compat_ioctl = ib_umad_compat_ioctl, 1016#endif 1017 .open = ib_umad_open, 1018 .release = ib_umad_close, 1019 .llseek = no_llseek, 1020}; 1021 1022static int ib_umad_sm_open(struct inode *inode, struct file *filp) 1023{ 1024 struct ib_umad_port *port; 1025 struct ib_port_modify props = { 1026 .set_port_cap_mask = IB_PORT_SM 1027 }; 1028 int ret; 1029 1030 port = container_of(inode->i_cdev->si_drv1, struct ib_umad_port, sm_cdev); 1031 1032 if (filp->f_flags & O_NONBLOCK) { 1033 if (down_trylock(&port->sm_sem)) { 1034 ret = -EAGAIN; 1035 goto fail; 1036 } 1037 } else { 1038 if (down_interruptible(&port->sm_sem)) { 1039 ret = -ERESTARTSYS; 1040 goto fail; 1041 } 1042 } 1043 1044 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props); 1045 if (ret) 1046 goto err_up_sem; 1047 1048 filp->private_data = port; 1049 1050 ret = nonseekable_open(inode, filp); 1051 if (ret) 1052 goto err_clr_sm_cap; 1053 1054 kobject_get(&port->umad_dev->kobj); 1055 1056 return 0; 1057 1058err_clr_sm_cap: 1059 swap(props.set_port_cap_mask, props.clr_port_cap_mask); 1060 ib_modify_port(port->ib_dev, port->port_num, 0, &props); 1061 1062err_up_sem: 1063 up(&port->sm_sem); 1064 1065fail: 1066 return ret; 1067} 1068 1069static int ib_umad_sm_close(struct inode *inode, struct file *filp) 1070{ 1071 struct ib_umad_port *port = filp->private_data; 1072 struct ib_port_modify props = { 1073 .clr_port_cap_mask = IB_PORT_SM 1074 }; 1075 int ret = 0; 1076 1077 mutex_lock(&port->file_mutex); 1078 if (port->ib_dev) 1079 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props); 1080 mutex_unlock(&port->file_mutex); 1081 1082 up(&port->sm_sem); 1083 1084 kobject_put(&port->umad_dev->kobj); 1085 1086 return ret; 1087} 1088 1089static const struct file_operations umad_sm_fops = { 1090 .owner = THIS_MODULE, 1091 .open = ib_umad_sm_open, 1092 .release = ib_umad_sm_close, 1093 .llseek = no_llseek, 1094}; 1095 1096static struct ib_client umad_client = { 1097 .name = "umad", 1098 .add = ib_umad_add_one, 1099 .remove = ib_umad_remove_one 1100}; 1101 1102static ssize_t show_ibdev(struct device *dev, struct device_attribute *attr, 1103 char *buf) 1104{ 1105 struct ib_umad_port *port = dev_get_drvdata(dev); 1106 1107 if (!port) 1108 return -ENODEV; 1109 1110 return sprintf(buf, "%s\n", port->ib_dev->name); 1111} 1112static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL); 1113 1114static ssize_t show_port(struct device *dev, struct device_attribute *attr, 1115 char *buf) 1116{ 1117 struct ib_umad_port *port = dev_get_drvdata(dev); 1118 1119 if (!port) 1120 return -ENODEV; 1121 1122 return sprintf(buf, "%d\n", port->port_num); 1123} 1124static DEVICE_ATTR(port, S_IRUGO, show_port, NULL); 1125 1126static CLASS_ATTR_STRING(abi_version, S_IRUGO, 1127 __stringify(IB_USER_MAD_ABI_VERSION)); 1128 1129static dev_t overflow_maj; 1130static DECLARE_BITMAP(overflow_map, IB_UMAD_MAX_PORTS); 1131static int find_overflow_devnum(struct ib_device *device) 1132{ 1133 int ret; 1134 1135 if (!overflow_maj) { 1136 ret = alloc_chrdev_region(&overflow_maj, 0, IB_UMAD_MAX_PORTS * 2, 1137 "infiniband_mad"); 1138 if (ret) { 1139 dev_err(&device->dev, 1140 "couldn't register dynamic device number\n"); 1141 return ret; 1142 } 1143 } 1144 1145 ret = find_first_zero_bit(overflow_map, IB_UMAD_MAX_PORTS); 1146 if (ret >= IB_UMAD_MAX_PORTS) 1147 return -1; 1148 1149 return ret; 1150} 1151 1152static int ib_umad_init_port(struct ib_device *device, int port_num, 1153 struct ib_umad_device *umad_dev, 1154 struct ib_umad_port *port) 1155{ 1156 int devnum; 1157 dev_t base; 1158 1159 spin_lock(&port_lock); 1160 devnum = find_first_zero_bit(dev_map, IB_UMAD_MAX_PORTS); 1161 if (devnum >= IB_UMAD_MAX_PORTS) { 1162 spin_unlock(&port_lock); 1163 devnum = find_overflow_devnum(device); 1164 if (devnum < 0) 1165 return -1; 1166 1167 spin_lock(&port_lock); 1168 port->dev_num = devnum + IB_UMAD_MAX_PORTS; 1169 base = devnum + overflow_maj; 1170 set_bit(devnum, overflow_map); 1171 } else { 1172 port->dev_num = devnum; 1173 base = devnum + base_dev; 1174 set_bit(devnum, dev_map); 1175 } 1176 spin_unlock(&port_lock); 1177 1178 port->ib_dev = device; 1179 port->port_num = port_num; 1180 sema_init(&port->sm_sem, 1); 1181 mutex_init(&port->file_mutex); 1182 INIT_LIST_HEAD(&port->file_list); 1183 1184 cdev_init(&port->cdev, &umad_fops); 1185 port->cdev.owner = THIS_MODULE; 1186 port->cdev.kobj.parent = &umad_dev->kobj; 1187 kobject_set_name(&port->cdev.kobj, "umad%d", port->dev_num); 1188 if (cdev_add(&port->cdev, base, 1)) 1189 goto err_cdev; 1190 1191 port->dev = device_create(umad_class, device->dma_device, 1192 port->cdev.dev, port, 1193 "umad%d", port->dev_num); 1194 if (IS_ERR(port->dev)) 1195 goto err_cdev; 1196 1197 if (device_create_file(port->dev, &dev_attr_ibdev)) 1198 goto err_dev; 1199 if (device_create_file(port->dev, &dev_attr_port)) 1200 goto err_dev; 1201 1202 base += IB_UMAD_MAX_PORTS; 1203 cdev_init(&port->sm_cdev, &umad_sm_fops); 1204 port->sm_cdev.owner = THIS_MODULE; 1205 port->sm_cdev.kobj.parent = &umad_dev->kobj; 1206 kobject_set_name(&port->sm_cdev.kobj, "issm%d", port->dev_num); 1207 if (cdev_add(&port->sm_cdev, base, 1)) 1208 goto err_sm_cdev; 1209 1210 port->sm_dev = device_create(umad_class, device->dma_device, 1211 port->sm_cdev.dev, port, 1212 "issm%d", port->dev_num); 1213 if (IS_ERR(port->sm_dev)) 1214 goto err_sm_cdev; 1215 1216 if (device_create_file(port->sm_dev, &dev_attr_ibdev)) 1217 goto err_sm_dev; 1218 if (device_create_file(port->sm_dev, &dev_attr_port)) 1219 goto err_sm_dev; 1220 1221 return 0; 1222 1223err_sm_dev: 1224 device_destroy(umad_class, port->sm_cdev.dev); 1225 1226err_sm_cdev: 1227 cdev_del(&port->sm_cdev); 1228 1229err_dev: 1230 device_destroy(umad_class, port->cdev.dev); 1231 1232err_cdev: 1233 cdev_del(&port->cdev); 1234 if (port->dev_num < IB_UMAD_MAX_PORTS) 1235 clear_bit(devnum, dev_map); 1236 else 1237 clear_bit(devnum, overflow_map); 1238 1239 return -1; 1240} 1241 1242static void ib_umad_kill_port(struct ib_umad_port *port) 1243{ 1244 struct ib_umad_file *file; 1245 int id; 1246 1247 dev_set_drvdata(port->dev, NULL); 1248 dev_set_drvdata(port->sm_dev, NULL); 1249 1250 device_destroy(umad_class, port->cdev.dev); 1251 device_destroy(umad_class, port->sm_cdev.dev); 1252 1253 cdev_del(&port->cdev); 1254 cdev_del(&port->sm_cdev); 1255 1256 mutex_lock(&port->file_mutex); 1257 1258 port->ib_dev = NULL; 1259 1260 list_for_each_entry(file, &port->file_list, port_list) { 1261 mutex_lock(&file->mutex); 1262 file->agents_dead = 1; 1263 mutex_unlock(&file->mutex); 1264 1265 for (id = 0; id < IB_UMAD_MAX_AGENTS; ++id) 1266 if (file->agent[id]) 1267 ib_unregister_mad_agent(file->agent[id]); 1268 } 1269 1270 mutex_unlock(&port->file_mutex); 1271 1272 if (port->dev_num < IB_UMAD_MAX_PORTS) 1273 clear_bit(port->dev_num, dev_map); 1274 else 1275 clear_bit(port->dev_num - IB_UMAD_MAX_PORTS, overflow_map); 1276} 1277 1278static void ib_umad_add_one(struct ib_device *device) 1279{ 1280 struct ib_umad_device *umad_dev; 1281 int s, e, i; 1282 int count = 0; 1283 1284 s = rdma_start_port(device); 1285 e = rdma_end_port(device); 1286 1287 umad_dev = kzalloc(sizeof *umad_dev + 1288 (e - s + 1) * sizeof (struct ib_umad_port), 1289 GFP_KERNEL); 1290 if (!umad_dev) 1291 return; 1292 1293 kobject_init(&umad_dev->kobj, &ib_umad_dev_ktype); 1294 1295 for (i = s; i <= e; ++i) { 1296 if (!rdma_cap_ib_mad(device, i)) 1297 continue; 1298 1299 umad_dev->port[i - s].umad_dev = umad_dev; 1300 1301 if (ib_umad_init_port(device, i, umad_dev, 1302 &umad_dev->port[i - s])) 1303 goto err; 1304 1305 count++; 1306 } 1307 1308 if (!count) 1309 goto free; 1310 1311 ib_set_client_data(device, &umad_client, umad_dev); 1312 1313 return; 1314 1315err: 1316 while (--i >= s) { 1317 if (!rdma_cap_ib_mad(device, i)) 1318 continue; 1319 1320 ib_umad_kill_port(&umad_dev->port[i - s]); 1321 } 1322free: 1323 kobject_put(&umad_dev->kobj); 1324} 1325 1326static void ib_umad_remove_one(struct ib_device *device, void *client_data) 1327{ 1328 struct ib_umad_device *umad_dev = client_data; 1329 int i; 1330 1331 if (!umad_dev) 1332 return; 1333 1334 for (i = 0; i <= rdma_end_port(device) - rdma_start_port(device); ++i) { 1335 if (rdma_cap_ib_mad(device, i + rdma_start_port(device))) 1336 ib_umad_kill_port(&umad_dev->port[i]); 1337 } 1338 1339 kobject_put(&umad_dev->kobj); 1340} 1341 1342static char *umad_devnode(struct device *dev, umode_t *mode) 1343{ 1344 return kasprintf(GFP_KERNEL, "infiniband/%s", dev_name(dev)); 1345} 1346 1347static int __init ib_umad_init(void) 1348{ 1349 int ret; 1350 1351 ret = register_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2, 1352 "infiniband_mad"); 1353 if (ret) { 1354 pr_err("couldn't register device number\n"); 1355 goto out; 1356 } 1357 1358 umad_class = class_create(THIS_MODULE, "infiniband_mad"); 1359 if (IS_ERR(umad_class)) { 1360 ret = PTR_ERR(umad_class); 1361 pr_err("couldn't create class infiniband_mad\n"); 1362 goto out_chrdev; 1363 } 1364 1365 umad_class->devnode = umad_devnode; 1366 1367 ret = class_create_file(umad_class, &class_attr_abi_version.attr); 1368 if (ret) { 1369 pr_err("couldn't create abi_version attribute\n"); 1370 goto out_class; 1371 } 1372 1373 ret = ib_register_client(&umad_client); 1374 if (ret) { 1375 pr_err("couldn't register ib_umad client\n"); 1376 goto out_class; 1377 } 1378 1379 return 0; 1380 1381out_class: 1382 class_destroy(umad_class); 1383 1384out_chrdev: 1385 unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2); 1386 1387out: 1388 return ret; 1389} 1390 1391static void __exit ib_umad_cleanup(void) 1392{ 1393 ib_unregister_client(&umad_client); 1394 class_destroy(umad_class); 1395 unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2); 1396 if (overflow_maj) 1397 unregister_chrdev_region(overflow_maj, IB_UMAD_MAX_PORTS * 2); 1398} 1399 1400module_init_order(ib_umad_init, SI_ORDER_THIRD); 1401module_exit(ib_umad_cleanup); 1402