2 * Copyright (c) 2005 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005, 2006, 2007 Cisco Systems. All rights reserved.
4 * Copyright (c) 2005 PathScale, Inc. All rights reserved.
5 * Copyright (c) 2006 Mellanox Technologies. All rights reserved.
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:
13 * Redistribution and use in source and binary forms, with or
14 * without modification, are permitted provided that the following
17 * - Redistributions of source code must retain the above
18 * copyright notice, this list of conditions and the following
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.
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
36 #include <linux/file.h>
38 #include <linux/slab.h>
39 #include <linux/sched.h>
41 #include <asm/uaccess.h>
44 #include "core_priv.h"
46 struct uverbs_lock_class {
47 struct lock_class_key key;
51 static struct uverbs_lock_class pd_lock_class = { .name = "PD-uobj" };
52 static struct uverbs_lock_class mr_lock_class = { .name = "MR-uobj" };
53 static struct uverbs_lock_class mw_lock_class = { .name = "MW-uobj" };
54 static struct uverbs_lock_class cq_lock_class = { .name = "CQ-uobj" };
55 static struct uverbs_lock_class qp_lock_class = { .name = "QP-uobj" };
56 static struct uverbs_lock_class ah_lock_class = { .name = "AH-uobj" };
57 static struct uverbs_lock_class srq_lock_class = { .name = "SRQ-uobj" };
58 static struct uverbs_lock_class xrcd_lock_class = { .name = "XRCD-uobj" };
59 static struct uverbs_lock_class rule_lock_class = { .name = "RULE-uobj" };
62 * The ib_uobject locking scheme is as follows:
64 * - ib_uverbs_idr_lock protects the uverbs idrs themselves, so it
65 * needs to be held during all idr operations. When an object is
66 * looked up, a reference must be taken on the object's kref before
69 * - Each object also has an rwsem. This rwsem must be held for
70 * reading while an operation that uses the object is performed.
71 * For example, while registering an MR, the associated PD's
72 * uobject.mutex must be held for reading. The rwsem must be held
73 * for writing while initializing or destroying an object.
75 * - In addition, each object has a "live" flag. If this flag is not
76 * set, then lookups of the object will fail even if it is found in
77 * the idr. This handles a reader that blocks and does not acquire
78 * the rwsem until after the object is destroyed. The destroy
79 * operation will set the live flag to 0 and then drop the rwsem;
80 * this will allow the reader to acquire the rwsem, see that the
81 * live flag is 0, and then drop the rwsem and its reference to
82 * object. The underlying storage will not be freed until the last
83 * reference to the object is dropped.
86 static void init_uobj(struct ib_uobject *uobj, u64 user_handle,
87 struct ib_ucontext *context, struct uverbs_lock_class *c)
89 uobj->user_handle = user_handle;
90 uobj->context = context;
91 kref_init(&uobj->ref);
92 init_rwsem(&uobj->mutex);
93 lockdep_set_class_and_name(&uobj->mutex, &c->key, c->name);
97 static void release_uobj(struct kref *kref)
99 kfree(container_of(kref, struct ib_uobject, ref));
102 static void put_uobj(struct ib_uobject *uobj)
104 kref_put(&uobj->ref, release_uobj);
107 static void put_uobj_read(struct ib_uobject *uobj)
109 up_read(&uobj->mutex);
113 static void put_uobj_write(struct ib_uobject *uobj)
115 up_write(&uobj->mutex);
119 static int idr_add_uobj(struct idr *idr, struct ib_uobject *uobj)
123 idr_preload(GFP_KERNEL);
124 spin_lock(&ib_uverbs_idr_lock);
126 ret = idr_alloc(idr, uobj, 0, 0, GFP_NOWAIT);
130 spin_unlock(&ib_uverbs_idr_lock);
133 return ret < 0 ? ret : 0;
136 void idr_remove_uobj(struct idr *idr, struct ib_uobject *uobj)
138 spin_lock(&ib_uverbs_idr_lock);
139 idr_remove(idr, uobj->id);
140 spin_unlock(&ib_uverbs_idr_lock);
143 static struct ib_uobject *__idr_get_uobj(struct idr *idr, int id,
144 struct ib_ucontext *context)
146 struct ib_uobject *uobj;
148 spin_lock(&ib_uverbs_idr_lock);
149 uobj = idr_find(idr, id);
151 if (uobj->context == context)
152 kref_get(&uobj->ref);
156 spin_unlock(&ib_uverbs_idr_lock);
161 static struct ib_uobject *idr_read_uobj(struct idr *idr, int id,
162 struct ib_ucontext *context, int nested)
164 struct ib_uobject *uobj;
166 uobj = __idr_get_uobj(idr, id, context);
171 down_read_nested(&uobj->mutex, SINGLE_DEPTH_NESTING);
173 down_read(&uobj->mutex);
182 static struct ib_uobject *idr_write_uobj(struct idr *idr, int id,
183 struct ib_ucontext *context)
185 struct ib_uobject *uobj;
187 uobj = __idr_get_uobj(idr, id, context);
191 down_write(&uobj->mutex);
193 put_uobj_write(uobj);
200 static void *idr_read_obj(struct idr *idr, int id, struct ib_ucontext *context,
203 struct ib_uobject *uobj;
205 uobj = idr_read_uobj(idr, id, context, nested);
206 return uobj ? uobj->object : NULL;
209 static struct ib_pd *idr_read_pd(int pd_handle, struct ib_ucontext *context)
211 return idr_read_obj(&ib_uverbs_pd_idr, pd_handle, context, 0);
214 static void put_pd_read(struct ib_pd *pd)
216 put_uobj_read(pd->uobject);
219 static struct ib_cq *idr_read_cq(int cq_handle, struct ib_ucontext *context, int nested)
221 return idr_read_obj(&ib_uverbs_cq_idr, cq_handle, context, nested);
224 static void put_cq_read(struct ib_cq *cq)
226 put_uobj_read(cq->uobject);
229 static struct ib_ah *idr_read_ah(int ah_handle, struct ib_ucontext *context)
231 return idr_read_obj(&ib_uverbs_ah_idr, ah_handle, context, 0);
234 static void put_ah_read(struct ib_ah *ah)
236 put_uobj_read(ah->uobject);
239 static struct ib_qp *idr_read_qp(int qp_handle, struct ib_ucontext *context)
241 return idr_read_obj(&ib_uverbs_qp_idr, qp_handle, context, 0);
244 static struct ib_qp *idr_write_qp(int qp_handle, struct ib_ucontext *context)
246 struct ib_uobject *uobj;
248 uobj = idr_write_uobj(&ib_uverbs_qp_idr, qp_handle, context);
249 return uobj ? uobj->object : NULL;
252 static void put_qp_read(struct ib_qp *qp)
254 put_uobj_read(qp->uobject);
257 static void put_qp_write(struct ib_qp *qp)
259 put_uobj_write(qp->uobject);
262 static struct ib_srq *idr_read_srq(int srq_handle, struct ib_ucontext *context)
264 return idr_read_obj(&ib_uverbs_srq_idr, srq_handle, context, 0);
267 static void put_srq_read(struct ib_srq *srq)
269 put_uobj_read(srq->uobject);
272 static struct ib_xrcd *idr_read_xrcd(int xrcd_handle, struct ib_ucontext *context,
273 struct ib_uobject **uobj)
275 *uobj = idr_read_uobj(&ib_uverbs_xrcd_idr, xrcd_handle, context, 0);
276 return *uobj ? (*uobj)->object : NULL;
279 static void put_xrcd_read(struct ib_uobject *uobj)
284 ssize_t ib_uverbs_get_context(struct ib_uverbs_file *file,
285 const char __user *buf,
286 int in_len, int out_len)
288 struct ib_uverbs_get_context cmd;
289 struct ib_uverbs_get_context_resp resp;
290 struct ib_udata udata;
291 struct ib_device *ibdev = file->device->ib_dev;
292 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
293 struct ib_device_attr dev_attr;
295 struct ib_ucontext *ucontext;
299 if (out_len < sizeof resp)
302 if (copy_from_user(&cmd, buf, sizeof cmd))
305 mutex_lock(&file->mutex);
307 if (file->ucontext) {
312 INIT_UDATA(&udata, buf + sizeof cmd,
313 (unsigned long) cmd.response + sizeof resp,
314 in_len - sizeof cmd, out_len - sizeof resp);
316 ucontext = ibdev->alloc_ucontext(ibdev, &udata);
317 if (IS_ERR(ucontext)) {
318 ret = PTR_ERR(ucontext);
322 ucontext->device = ibdev;
323 INIT_LIST_HEAD(&ucontext->pd_list);
324 INIT_LIST_HEAD(&ucontext->mr_list);
325 INIT_LIST_HEAD(&ucontext->mw_list);
326 INIT_LIST_HEAD(&ucontext->cq_list);
327 INIT_LIST_HEAD(&ucontext->qp_list);
328 INIT_LIST_HEAD(&ucontext->srq_list);
329 INIT_LIST_HEAD(&ucontext->ah_list);
330 INIT_LIST_HEAD(&ucontext->xrcd_list);
331 INIT_LIST_HEAD(&ucontext->rule_list);
333 ucontext->tgid = get_task_pid(current->group_leader, PIDTYPE_PID);
335 ucontext->closing = 0;
337 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
338 ucontext->umem_tree = RB_ROOT;
339 init_rwsem(&ucontext->umem_rwsem);
340 ucontext->odp_mrs_count = 0;
341 INIT_LIST_HEAD(&ucontext->no_private_counters);
343 ret = ib_query_device(ibdev, &dev_attr);
346 if (!(dev_attr.device_cap_flags & IB_DEVICE_ON_DEMAND_PAGING))
347 ucontext->invalidate_range = NULL;
351 resp.num_comp_vectors = file->device->num_comp_vectors;
353 ret = get_unused_fd_flags(O_CLOEXEC);
358 filp = ib_uverbs_alloc_event_file(file, 1);
364 if (copy_to_user((void __user *) (unsigned long) cmd.response,
365 &resp, sizeof resp)) {
370 file->async_file = filp->private_data;
372 INIT_IB_EVENT_HANDLER(&file->event_handler, file->device->ib_dev,
373 ib_uverbs_event_handler);
374 ret = ib_register_event_handler(&file->event_handler);
378 kref_get(&file->async_file->ref);
379 kref_get(&file->ref);
380 file->ucontext = ucontext;
382 fd_install(resp.async_fd, filp);
384 mutex_unlock(&file->mutex);
392 put_unused_fd(resp.async_fd);
395 put_pid(ucontext->tgid);
396 ibdev->dealloc_ucontext(ucontext);
399 mutex_unlock(&file->mutex);
403 ssize_t ib_uverbs_query_device(struct ib_uverbs_file *file,
404 const char __user *buf,
405 int in_len, int out_len)
407 struct ib_uverbs_query_device cmd;
408 struct ib_uverbs_query_device_resp resp;
409 struct ib_device_attr attr;
412 if (out_len < sizeof resp)
415 if (copy_from_user(&cmd, buf, sizeof cmd))
418 ret = ib_query_device(file->device->ib_dev, &attr);
422 memset(&resp, 0, sizeof resp);
424 resp.fw_ver = attr.fw_ver;
425 resp.node_guid = file->device->ib_dev->node_guid;
426 resp.sys_image_guid = attr.sys_image_guid;
427 resp.max_mr_size = attr.max_mr_size;
428 resp.page_size_cap = attr.page_size_cap;
429 resp.vendor_id = attr.vendor_id;
430 resp.vendor_part_id = attr.vendor_part_id;
431 resp.hw_ver = attr.hw_ver;
432 resp.max_qp = attr.max_qp;
433 resp.max_qp_wr = attr.max_qp_wr;
434 resp.device_cap_flags = attr.device_cap_flags;
435 resp.max_sge = attr.max_sge;
436 resp.max_sge_rd = attr.max_sge_rd;
437 resp.max_cq = attr.max_cq;
438 resp.max_cqe = attr.max_cqe;
439 resp.max_mr = attr.max_mr;
440 resp.max_pd = attr.max_pd;
441 resp.max_qp_rd_atom = attr.max_qp_rd_atom;
442 resp.max_ee_rd_atom = attr.max_ee_rd_atom;
443 resp.max_res_rd_atom = attr.max_res_rd_atom;
444 resp.max_qp_init_rd_atom = attr.max_qp_init_rd_atom;
445 resp.max_ee_init_rd_atom = attr.max_ee_init_rd_atom;
446 resp.atomic_cap = attr.atomic_cap;
447 resp.max_ee = attr.max_ee;
448 resp.max_rdd = attr.max_rdd;
449 resp.max_mw = attr.max_mw;
450 resp.max_raw_ipv6_qp = attr.max_raw_ipv6_qp;
451 resp.max_raw_ethy_qp = attr.max_raw_ethy_qp;
452 resp.max_mcast_grp = attr.max_mcast_grp;
453 resp.max_mcast_qp_attach = attr.max_mcast_qp_attach;
454 resp.max_total_mcast_qp_attach = attr.max_total_mcast_qp_attach;
455 resp.max_ah = attr.max_ah;
456 resp.max_fmr = attr.max_fmr;
457 resp.max_map_per_fmr = attr.max_map_per_fmr;
458 resp.max_srq = attr.max_srq;
459 resp.max_srq_wr = attr.max_srq_wr;
460 resp.max_srq_sge = attr.max_srq_sge;
461 resp.max_pkeys = attr.max_pkeys;
462 resp.local_ca_ack_delay = attr.local_ca_ack_delay;
463 resp.phys_port_cnt = file->device->ib_dev->phys_port_cnt;
465 if (copy_to_user((void __user *) (unsigned long) cmd.response,
472 ssize_t ib_uverbs_query_port(struct ib_uverbs_file *file,
473 const char __user *buf,
474 int in_len, int out_len)
476 struct ib_uverbs_query_port cmd;
477 struct ib_uverbs_query_port_resp resp;
478 struct ib_port_attr attr;
481 if (out_len < sizeof resp)
484 if (copy_from_user(&cmd, buf, sizeof cmd))
487 ret = ib_query_port(file->device->ib_dev, cmd.port_num, &attr);
491 memset(&resp, 0, sizeof resp);
493 resp.state = attr.state;
494 resp.max_mtu = attr.max_mtu;
495 resp.active_mtu = attr.active_mtu;
496 resp.gid_tbl_len = attr.gid_tbl_len;
497 resp.port_cap_flags = attr.port_cap_flags;
498 resp.max_msg_sz = attr.max_msg_sz;
499 resp.bad_pkey_cntr = attr.bad_pkey_cntr;
500 resp.qkey_viol_cntr = attr.qkey_viol_cntr;
501 resp.pkey_tbl_len = attr.pkey_tbl_len;
503 resp.sm_lid = attr.sm_lid;
505 resp.max_vl_num = attr.max_vl_num;
506 resp.sm_sl = attr.sm_sl;
507 resp.subnet_timeout = attr.subnet_timeout;
508 resp.init_type_reply = attr.init_type_reply;
509 resp.active_width = attr.active_width;
510 resp.active_speed = attr.active_speed;
511 resp.phys_state = attr.phys_state;
512 resp.link_layer = rdma_port_get_link_layer(file->device->ib_dev,
515 if (copy_to_user((void __user *) (unsigned long) cmd.response,
522 ssize_t ib_uverbs_alloc_pd(struct ib_uverbs_file *file,
523 const char __user *buf,
524 int in_len, int out_len)
526 struct ib_uverbs_alloc_pd cmd;
527 struct ib_uverbs_alloc_pd_resp resp;
528 struct ib_udata udata;
529 struct ib_uobject *uobj;
533 if (out_len < sizeof resp)
536 if (copy_from_user(&cmd, buf, sizeof cmd))
539 INIT_UDATA(&udata, buf + sizeof cmd,
540 (unsigned long) cmd.response + sizeof resp,
541 in_len - sizeof cmd, out_len - sizeof resp);
543 uobj = kmalloc(sizeof *uobj, GFP_KERNEL);
547 init_uobj(uobj, 0, file->ucontext, &pd_lock_class);
548 down_write(&uobj->mutex);
550 pd = file->device->ib_dev->alloc_pd(file->device->ib_dev,
551 file->ucontext, &udata);
557 pd->device = file->device->ib_dev;
559 atomic_set(&pd->usecnt, 0);
562 ret = idr_add_uobj(&ib_uverbs_pd_idr, uobj);
566 memset(&resp, 0, sizeof resp);
567 resp.pd_handle = uobj->id;
569 if (copy_to_user((void __user *) (unsigned long) cmd.response,
570 &resp, sizeof resp)) {
575 mutex_lock(&file->mutex);
576 list_add_tail(&uobj->list, &file->ucontext->pd_list);
577 mutex_unlock(&file->mutex);
581 up_write(&uobj->mutex);
586 idr_remove_uobj(&ib_uverbs_pd_idr, uobj);
592 put_uobj_write(uobj);
596 ssize_t ib_uverbs_dealloc_pd(struct ib_uverbs_file *file,
597 const char __user *buf,
598 int in_len, int out_len)
600 struct ib_uverbs_dealloc_pd cmd;
601 struct ib_uobject *uobj;
604 if (copy_from_user(&cmd, buf, sizeof cmd))
607 uobj = idr_write_uobj(&ib_uverbs_pd_idr, cmd.pd_handle, file->ucontext);
611 ret = ib_dealloc_pd(uobj->object);
615 put_uobj_write(uobj);
620 idr_remove_uobj(&ib_uverbs_pd_idr, uobj);
622 mutex_lock(&file->mutex);
623 list_del(&uobj->list);
624 mutex_unlock(&file->mutex);
631 struct xrcd_table_entry {
633 struct ib_xrcd *xrcd;
637 static int xrcd_table_insert(struct ib_uverbs_device *dev,
639 struct ib_xrcd *xrcd)
641 struct xrcd_table_entry *entry, *scan;
642 struct rb_node **p = &dev->xrcd_tree.rb_node;
643 struct rb_node *parent = NULL;
645 entry = kmalloc(sizeof *entry, GFP_KERNEL);
650 entry->inode = inode;
654 scan = rb_entry(parent, struct xrcd_table_entry, node);
656 if (inode < scan->inode) {
658 } else if (inode > scan->inode) {
666 rb_link_node(&entry->node, parent, p);
667 rb_insert_color(&entry->node, &dev->xrcd_tree);
672 static struct xrcd_table_entry *xrcd_table_search(struct ib_uverbs_device *dev,
675 struct xrcd_table_entry *entry;
676 struct rb_node *p = dev->xrcd_tree.rb_node;
679 entry = rb_entry(p, struct xrcd_table_entry, node);
681 if (inode < entry->inode)
683 else if (inode > entry->inode)
692 static struct ib_xrcd *find_xrcd(struct ib_uverbs_device *dev, struct inode *inode)
694 struct xrcd_table_entry *entry;
696 entry = xrcd_table_search(dev, inode);
703 static void xrcd_table_delete(struct ib_uverbs_device *dev,
706 struct xrcd_table_entry *entry;
708 entry = xrcd_table_search(dev, inode);
711 rb_erase(&entry->node, &dev->xrcd_tree);
716 ssize_t ib_uverbs_open_xrcd(struct ib_uverbs_file *file,
717 const char __user *buf, int in_len,
720 struct ib_uverbs_open_xrcd cmd;
721 struct ib_uverbs_open_xrcd_resp resp;
722 struct ib_udata udata;
723 struct ib_uxrcd_object *obj;
724 struct ib_xrcd *xrcd = NULL;
725 struct fd f = {NULL, 0};
726 struct inode *inode = NULL;
730 if (out_len < sizeof resp)
733 if (copy_from_user(&cmd, buf, sizeof cmd))
736 INIT_UDATA(&udata, buf + sizeof cmd,
737 (unsigned long) cmd.response + sizeof resp,
738 in_len - sizeof cmd, out_len - sizeof resp);
740 mutex_lock(&file->device->xrcd_tree_mutex);
743 /* search for file descriptor */
747 goto err_tree_mutex_unlock;
750 inode = file_inode(f.file);
751 xrcd = find_xrcd(file->device, inode);
752 if (!xrcd && !(cmd.oflags & O_CREAT)) {
753 /* no file descriptor. Need CREATE flag */
755 goto err_tree_mutex_unlock;
758 if (xrcd && cmd.oflags & O_EXCL) {
760 goto err_tree_mutex_unlock;
764 obj = kmalloc(sizeof *obj, GFP_KERNEL);
767 goto err_tree_mutex_unlock;
770 init_uobj(&obj->uobject, 0, file->ucontext, &xrcd_lock_class);
772 down_write(&obj->uobject.mutex);
775 xrcd = file->device->ib_dev->alloc_xrcd(file->device->ib_dev,
776 file->ucontext, &udata);
783 xrcd->device = file->device->ib_dev;
784 atomic_set(&xrcd->usecnt, 0);
785 mutex_init(&xrcd->tgt_qp_mutex);
786 INIT_LIST_HEAD(&xrcd->tgt_qp_list);
790 atomic_set(&obj->refcnt, 0);
791 obj->uobject.object = xrcd;
792 ret = idr_add_uobj(&ib_uverbs_xrcd_idr, &obj->uobject);
796 memset(&resp, 0, sizeof resp);
797 resp.xrcd_handle = obj->uobject.id;
801 /* create new inode/xrcd table entry */
802 ret = xrcd_table_insert(file->device, inode, xrcd);
804 goto err_insert_xrcd;
806 atomic_inc(&xrcd->usecnt);
809 if (copy_to_user((void __user *) (unsigned long) cmd.response,
810 &resp, sizeof resp)) {
818 mutex_lock(&file->mutex);
819 list_add_tail(&obj->uobject.list, &file->ucontext->xrcd_list);
820 mutex_unlock(&file->mutex);
822 obj->uobject.live = 1;
823 up_write(&obj->uobject.mutex);
825 mutex_unlock(&file->device->xrcd_tree_mutex);
831 xrcd_table_delete(file->device, inode);
832 atomic_dec(&xrcd->usecnt);
836 idr_remove_uobj(&ib_uverbs_xrcd_idr, &obj->uobject);
839 ib_dealloc_xrcd(xrcd);
842 put_uobj_write(&obj->uobject);
844 err_tree_mutex_unlock:
848 mutex_unlock(&file->device->xrcd_tree_mutex);
853 ssize_t ib_uverbs_close_xrcd(struct ib_uverbs_file *file,
854 const char __user *buf, int in_len,
857 struct ib_uverbs_close_xrcd cmd;
858 struct ib_uobject *uobj;
859 struct ib_xrcd *xrcd = NULL;
860 struct inode *inode = NULL;
861 struct ib_uxrcd_object *obj;
865 if (copy_from_user(&cmd, buf, sizeof cmd))
868 mutex_lock(&file->device->xrcd_tree_mutex);
869 uobj = idr_write_uobj(&ib_uverbs_xrcd_idr, cmd.xrcd_handle, file->ucontext);
877 obj = container_of(uobj, struct ib_uxrcd_object, uobject);
878 if (atomic_read(&obj->refcnt)) {
879 put_uobj_write(uobj);
884 if (!inode || atomic_dec_and_test(&xrcd->usecnt)) {
885 ret = ib_dealloc_xrcd(uobj->object);
892 atomic_inc(&xrcd->usecnt);
894 put_uobj_write(uobj);
900 xrcd_table_delete(file->device, inode);
902 idr_remove_uobj(&ib_uverbs_xrcd_idr, uobj);
903 mutex_lock(&file->mutex);
904 list_del(&uobj->list);
905 mutex_unlock(&file->mutex);
911 mutex_unlock(&file->device->xrcd_tree_mutex);
915 void ib_uverbs_dealloc_xrcd(struct ib_uverbs_device *dev,
916 struct ib_xrcd *xrcd)
921 if (inode && !atomic_dec_and_test(&xrcd->usecnt))
924 ib_dealloc_xrcd(xrcd);
927 xrcd_table_delete(dev, inode);
930 ssize_t ib_uverbs_reg_mr(struct ib_uverbs_file *file,
931 const char __user *buf, int in_len,
934 struct ib_uverbs_reg_mr cmd;
935 struct ib_uverbs_reg_mr_resp resp;
936 struct ib_udata udata;
937 struct ib_uobject *uobj;
942 if (out_len < sizeof resp)
945 if (copy_from_user(&cmd, buf, sizeof cmd))
948 INIT_UDATA(&udata, buf + sizeof cmd,
949 (unsigned long) cmd.response + sizeof resp,
950 in_len - sizeof cmd, out_len - sizeof resp);
952 if ((cmd.start & ~PAGE_MASK) != (cmd.hca_va & ~PAGE_MASK))
955 ret = ib_check_mr_access(cmd.access_flags);
959 uobj = kmalloc(sizeof *uobj, GFP_KERNEL);
963 init_uobj(uobj, 0, file->ucontext, &mr_lock_class);
964 down_write(&uobj->mutex);
966 pd = idr_read_pd(cmd.pd_handle, file->ucontext);
972 if (cmd.access_flags & IB_ACCESS_ON_DEMAND) {
973 struct ib_device_attr attr;
975 ret = ib_query_device(pd->device, &attr);
976 if (ret || !(attr.device_cap_flags &
977 IB_DEVICE_ON_DEMAND_PAGING)) {
978 pr_debug("ODP support not available\n");
984 mr = pd->device->reg_user_mr(pd, cmd.start, cmd.length, cmd.hca_va,
985 cmd.access_flags, &udata);
991 mr->device = pd->device;
994 atomic_inc(&pd->usecnt);
995 atomic_set(&mr->usecnt, 0);
998 ret = idr_add_uobj(&ib_uverbs_mr_idr, uobj);
1002 memset(&resp, 0, sizeof resp);
1003 resp.lkey = mr->lkey;
1004 resp.rkey = mr->rkey;
1005 resp.mr_handle = uobj->id;
1007 if (copy_to_user((void __user *) (unsigned long) cmd.response,
1008 &resp, sizeof resp)) {
1015 mutex_lock(&file->mutex);
1016 list_add_tail(&uobj->list, &file->ucontext->mr_list);
1017 mutex_unlock(&file->mutex);
1021 up_write(&uobj->mutex);
1026 idr_remove_uobj(&ib_uverbs_mr_idr, uobj);
1035 put_uobj_write(uobj);
1039 ssize_t ib_uverbs_rereg_mr(struct ib_uverbs_file *file,
1040 const char __user *buf, int in_len,
1043 struct ib_uverbs_rereg_mr cmd;
1044 struct ib_uverbs_rereg_mr_resp resp;
1045 struct ib_udata udata;
1046 struct ib_pd *pd = NULL;
1048 struct ib_pd *old_pd;
1050 struct ib_uobject *uobj;
1052 if (out_len < sizeof(resp))
1055 if (copy_from_user(&cmd, buf, sizeof(cmd)))
1058 INIT_UDATA(&udata, buf + sizeof(cmd),
1059 (unsigned long) cmd.response + sizeof(resp),
1060 in_len - sizeof(cmd), out_len - sizeof(resp));
1062 if (cmd.flags & ~IB_MR_REREG_SUPPORTED || !cmd.flags)
1065 if ((cmd.flags & IB_MR_REREG_TRANS) &&
1066 (!cmd.start || !cmd.hca_va || 0 >= cmd.length ||
1067 (cmd.start & ~PAGE_MASK) != (cmd.hca_va & ~PAGE_MASK)))
1070 uobj = idr_write_uobj(&ib_uverbs_mr_idr, cmd.mr_handle,
1078 if (cmd.flags & IB_MR_REREG_ACCESS) {
1079 ret = ib_check_mr_access(cmd.access_flags);
1084 if (cmd.flags & IB_MR_REREG_PD) {
1085 pd = idr_read_pd(cmd.pd_handle, file->ucontext);
1092 if (atomic_read(&mr->usecnt)) {
1098 ret = mr->device->rereg_user_mr(mr, cmd.flags, cmd.start,
1099 cmd.length, cmd.hca_va,
1100 cmd.access_flags, pd, &udata);
1102 if (cmd.flags & IB_MR_REREG_PD) {
1103 atomic_inc(&pd->usecnt);
1105 atomic_dec(&old_pd->usecnt);
1111 memset(&resp, 0, sizeof(resp));
1112 resp.lkey = mr->lkey;
1113 resp.rkey = mr->rkey;
1115 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1116 &resp, sizeof(resp)))
1122 if (cmd.flags & IB_MR_REREG_PD)
1127 put_uobj_write(mr->uobject);
1132 ssize_t ib_uverbs_dereg_mr(struct ib_uverbs_file *file,
1133 const char __user *buf, int in_len,
1136 struct ib_uverbs_dereg_mr cmd;
1138 struct ib_uobject *uobj;
1141 if (copy_from_user(&cmd, buf, sizeof cmd))
1144 uobj = idr_write_uobj(&ib_uverbs_mr_idr, cmd.mr_handle, file->ucontext);
1150 ret = ib_dereg_mr(mr);
1154 put_uobj_write(uobj);
1159 idr_remove_uobj(&ib_uverbs_mr_idr, uobj);
1161 mutex_lock(&file->mutex);
1162 list_del(&uobj->list);
1163 mutex_unlock(&file->mutex);
1170 ssize_t ib_uverbs_alloc_mw(struct ib_uverbs_file *file,
1171 const char __user *buf, int in_len,
1174 struct ib_uverbs_alloc_mw cmd;
1175 struct ib_uverbs_alloc_mw_resp resp;
1176 struct ib_uobject *uobj;
1181 if (out_len < sizeof(resp))
1184 if (copy_from_user(&cmd, buf, sizeof(cmd)))
1187 uobj = kmalloc(sizeof(*uobj), GFP_KERNEL);
1191 init_uobj(uobj, 0, file->ucontext, &mw_lock_class);
1192 down_write(&uobj->mutex);
1194 pd = idr_read_pd(cmd.pd_handle, file->ucontext);
1200 mw = pd->device->alloc_mw(pd, cmd.mw_type);
1206 mw->device = pd->device;
1209 atomic_inc(&pd->usecnt);
1212 ret = idr_add_uobj(&ib_uverbs_mw_idr, uobj);
1216 memset(&resp, 0, sizeof(resp));
1217 resp.rkey = mw->rkey;
1218 resp.mw_handle = uobj->id;
1220 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1221 &resp, sizeof(resp))) {
1228 mutex_lock(&file->mutex);
1229 list_add_tail(&uobj->list, &file->ucontext->mw_list);
1230 mutex_unlock(&file->mutex);
1234 up_write(&uobj->mutex);
1239 idr_remove_uobj(&ib_uverbs_mw_idr, uobj);
1248 put_uobj_write(uobj);
1252 ssize_t ib_uverbs_dealloc_mw(struct ib_uverbs_file *file,
1253 const char __user *buf, int in_len,
1256 struct ib_uverbs_dealloc_mw cmd;
1258 struct ib_uobject *uobj;
1261 if (copy_from_user(&cmd, buf, sizeof(cmd)))
1264 uobj = idr_write_uobj(&ib_uverbs_mw_idr, cmd.mw_handle, file->ucontext);
1270 ret = ib_dealloc_mw(mw);
1274 put_uobj_write(uobj);
1279 idr_remove_uobj(&ib_uverbs_mw_idr, uobj);
1281 mutex_lock(&file->mutex);
1282 list_del(&uobj->list);
1283 mutex_unlock(&file->mutex);
1290 ssize_t ib_uverbs_create_comp_channel(struct ib_uverbs_file *file,
1291 const char __user *buf, int in_len,
1294 struct ib_uverbs_create_comp_channel cmd;
1295 struct ib_uverbs_create_comp_channel_resp resp;
1299 if (out_len < sizeof resp)
1302 if (copy_from_user(&cmd, buf, sizeof cmd))
1305 ret = get_unused_fd_flags(O_CLOEXEC);
1310 filp = ib_uverbs_alloc_event_file(file, 0);
1312 put_unused_fd(resp.fd);
1313 return PTR_ERR(filp);
1316 if (copy_to_user((void __user *) (unsigned long) cmd.response,
1317 &resp, sizeof resp)) {
1318 put_unused_fd(resp.fd);
1323 fd_install(resp.fd, filp);
1327 ssize_t ib_uverbs_create_cq(struct ib_uverbs_file *file,
1328 const char __user *buf, int in_len,
1331 struct ib_uverbs_create_cq cmd;
1332 struct ib_uverbs_create_cq_resp resp;
1333 struct ib_udata udata;
1334 struct ib_ucq_object *obj;
1335 struct ib_uverbs_event_file *ev_file = NULL;
1339 if (out_len < sizeof resp)
1342 if (copy_from_user(&cmd, buf, sizeof cmd))
1345 INIT_UDATA(&udata, buf + sizeof cmd,
1346 (unsigned long) cmd.response + sizeof resp,
1347 in_len - sizeof cmd, out_len - sizeof resp);
1349 if (cmd.comp_vector >= file->device->num_comp_vectors)
1352 obj = kmalloc(sizeof *obj, GFP_KERNEL);
1356 init_uobj(&obj->uobject, cmd.user_handle, file->ucontext, &cq_lock_class);
1357 down_write(&obj->uobject.mutex);
1359 if (cmd.comp_channel >= 0) {
1360 ev_file = ib_uverbs_lookup_comp_file(cmd.comp_channel);
1367 obj->uverbs_file = file;
1368 obj->comp_events_reported = 0;
1369 obj->async_events_reported = 0;
1370 INIT_LIST_HEAD(&obj->comp_list);
1371 INIT_LIST_HEAD(&obj->async_list);
1373 cq = file->device->ib_dev->create_cq(file->device->ib_dev, cmd.cqe,
1375 file->ucontext, &udata);
1381 cq->device = file->device->ib_dev;
1382 cq->uobject = &obj->uobject;
1383 cq->comp_handler = ib_uverbs_comp_handler;
1384 cq->event_handler = ib_uverbs_cq_event_handler;
1385 cq->cq_context = ev_file;
1386 atomic_set(&cq->usecnt, 0);
1388 obj->uobject.object = cq;
1389 ret = idr_add_uobj(&ib_uverbs_cq_idr, &obj->uobject);
1393 memset(&resp, 0, sizeof resp);
1394 resp.cq_handle = obj->uobject.id;
1397 if (copy_to_user((void __user *) (unsigned long) cmd.response,
1398 &resp, sizeof resp)) {
1403 mutex_lock(&file->mutex);
1404 list_add_tail(&obj->uobject.list, &file->ucontext->cq_list);
1405 mutex_unlock(&file->mutex);
1407 obj->uobject.live = 1;
1409 up_write(&obj->uobject.mutex);
1414 idr_remove_uobj(&ib_uverbs_cq_idr, &obj->uobject);
1421 ib_uverbs_release_ucq(file, ev_file, obj);
1424 put_uobj_write(&obj->uobject);
1428 ssize_t ib_uverbs_resize_cq(struct ib_uverbs_file *file,
1429 const char __user *buf, int in_len,
1432 struct ib_uverbs_resize_cq cmd;
1433 struct ib_uverbs_resize_cq_resp resp;
1434 struct ib_udata udata;
1438 if (copy_from_user(&cmd, buf, sizeof cmd))
1441 INIT_UDATA(&udata, buf + sizeof cmd,
1442 (unsigned long) cmd.response + sizeof resp,
1443 in_len - sizeof cmd, out_len - sizeof resp);
1445 cq = idr_read_cq(cmd.cq_handle, file->ucontext, 0);
1449 ret = cq->device->resize_cq(cq, cmd.cqe, &udata);
1455 if (copy_to_user((void __user *) (unsigned long) cmd.response,
1456 &resp, sizeof resp.cqe))
1462 return ret ? ret : in_len;
1465 static int copy_wc_to_user(void __user *dest, struct ib_wc *wc)
1467 struct ib_uverbs_wc tmp;
1469 tmp.wr_id = wc->wr_id;
1470 tmp.status = wc->status;
1471 tmp.opcode = wc->opcode;
1472 tmp.vendor_err = wc->vendor_err;
1473 tmp.byte_len = wc->byte_len;
1474 tmp.ex.imm_data = (__u32 __force) wc->ex.imm_data;
1475 tmp.qp_num = wc->qp->qp_num;
1476 tmp.src_qp = wc->src_qp;
1477 tmp.wc_flags = wc->wc_flags;
1478 tmp.pkey_index = wc->pkey_index;
1479 tmp.slid = wc->slid;
1481 tmp.dlid_path_bits = wc->dlid_path_bits;
1482 tmp.port_num = wc->port_num;
1485 if (copy_to_user(dest, &tmp, sizeof tmp))
1491 ssize_t ib_uverbs_poll_cq(struct ib_uverbs_file *file,
1492 const char __user *buf, int in_len,
1495 struct ib_uverbs_poll_cq cmd;
1496 struct ib_uverbs_poll_cq_resp resp;
1497 u8 __user *header_ptr;
1498 u8 __user *data_ptr;
1503 if (copy_from_user(&cmd, buf, sizeof cmd))
1506 cq = idr_read_cq(cmd.cq_handle, file->ucontext, 0);
1510 /* we copy a struct ib_uverbs_poll_cq_resp to user space */
1511 header_ptr = (void __user *)(unsigned long) cmd.response;
1512 data_ptr = header_ptr + sizeof resp;
1514 memset(&resp, 0, sizeof resp);
1515 while (resp.count < cmd.ne) {
1516 ret = ib_poll_cq(cq, 1, &wc);
1522 ret = copy_wc_to_user(data_ptr, &wc);
1526 data_ptr += sizeof(struct ib_uverbs_wc);
1530 if (copy_to_user(header_ptr, &resp, sizeof resp)) {
1542 ssize_t ib_uverbs_req_notify_cq(struct ib_uverbs_file *file,
1543 const char __user *buf, int in_len,
1546 struct ib_uverbs_req_notify_cq cmd;
1549 if (copy_from_user(&cmd, buf, sizeof cmd))
1552 cq = idr_read_cq(cmd.cq_handle, file->ucontext, 0);
1556 ib_req_notify_cq(cq, cmd.solicited_only ?
1557 IB_CQ_SOLICITED : IB_CQ_NEXT_COMP);
1564 ssize_t ib_uverbs_destroy_cq(struct ib_uverbs_file *file,
1565 const char __user *buf, int in_len,
1568 struct ib_uverbs_destroy_cq cmd;
1569 struct ib_uverbs_destroy_cq_resp resp;
1570 struct ib_uobject *uobj;
1572 struct ib_ucq_object *obj;
1573 struct ib_uverbs_event_file *ev_file;
1576 if (copy_from_user(&cmd, buf, sizeof cmd))
1579 uobj = idr_write_uobj(&ib_uverbs_cq_idr, cmd.cq_handle, file->ucontext);
1583 ev_file = cq->cq_context;
1584 obj = container_of(cq->uobject, struct ib_ucq_object, uobject);
1586 ret = ib_destroy_cq(cq);
1590 put_uobj_write(uobj);
1595 idr_remove_uobj(&ib_uverbs_cq_idr, uobj);
1597 mutex_lock(&file->mutex);
1598 list_del(&uobj->list);
1599 mutex_unlock(&file->mutex);
1601 ib_uverbs_release_ucq(file, ev_file, obj);
1603 memset(&resp, 0, sizeof resp);
1604 resp.comp_events_reported = obj->comp_events_reported;
1605 resp.async_events_reported = obj->async_events_reported;
1609 if (copy_to_user((void __user *) (unsigned long) cmd.response,
1610 &resp, sizeof resp))
1616 ssize_t ib_uverbs_create_qp(struct ib_uverbs_file *file,
1617 const char __user *buf, int in_len,
1620 struct ib_uverbs_create_qp cmd;
1621 struct ib_uverbs_create_qp_resp resp;
1622 struct ib_udata udata;
1623 struct ib_uqp_object *obj;
1624 struct ib_device *device;
1625 struct ib_pd *pd = NULL;
1626 struct ib_xrcd *xrcd = NULL;
1627 struct ib_uobject *uninitialized_var(xrcd_uobj);
1628 struct ib_cq *scq = NULL, *rcq = NULL;
1629 struct ib_srq *srq = NULL;
1631 struct ib_qp_init_attr attr;
1634 if (out_len < sizeof resp)
1637 if (copy_from_user(&cmd, buf, sizeof cmd))
1640 if (cmd.qp_type == IB_QPT_RAW_PACKET && !capable(CAP_NET_RAW))
1643 INIT_UDATA(&udata, buf + sizeof cmd,
1644 (unsigned long) cmd.response + sizeof resp,
1645 in_len - sizeof cmd, out_len - sizeof resp);
1647 obj = kzalloc(sizeof *obj, GFP_KERNEL);
1651 init_uobj(&obj->uevent.uobject, cmd.user_handle, file->ucontext, &qp_lock_class);
1652 down_write(&obj->uevent.uobject.mutex);
1654 if (cmd.qp_type == IB_QPT_XRC_TGT) {
1655 xrcd = idr_read_xrcd(cmd.pd_handle, file->ucontext, &xrcd_uobj);
1660 device = xrcd->device;
1662 if (cmd.qp_type == IB_QPT_XRC_INI) {
1663 cmd.max_recv_wr = cmd.max_recv_sge = 0;
1666 srq = idr_read_srq(cmd.srq_handle, file->ucontext);
1667 if (!srq || srq->srq_type != IB_SRQT_BASIC) {
1673 if (cmd.recv_cq_handle != cmd.send_cq_handle) {
1674 rcq = idr_read_cq(cmd.recv_cq_handle, file->ucontext, 0);
1682 scq = idr_read_cq(cmd.send_cq_handle, file->ucontext, !!rcq);
1684 pd = idr_read_pd(cmd.pd_handle, file->ucontext);
1690 device = pd->device;
1693 attr.event_handler = ib_uverbs_qp_event_handler;
1694 attr.qp_context = file;
1699 attr.sq_sig_type = cmd.sq_sig_all ? IB_SIGNAL_ALL_WR : IB_SIGNAL_REQ_WR;
1700 attr.qp_type = cmd.qp_type;
1701 attr.create_flags = 0;
1703 attr.cap.max_send_wr = cmd.max_send_wr;
1704 attr.cap.max_recv_wr = cmd.max_recv_wr;
1705 attr.cap.max_send_sge = cmd.max_send_sge;
1706 attr.cap.max_recv_sge = cmd.max_recv_sge;
1707 attr.cap.max_inline_data = cmd.max_inline_data;
1709 obj->uevent.events_reported = 0;
1710 INIT_LIST_HEAD(&obj->uevent.event_list);
1711 INIT_LIST_HEAD(&obj->mcast_list);
1713 if (cmd.qp_type == IB_QPT_XRC_TGT)
1714 qp = ib_create_qp(pd, &attr);
1716 qp = device->create_qp(pd, &attr, &udata);
1723 if (cmd.qp_type != IB_QPT_XRC_TGT) {
1725 qp->device = device;
1727 qp->send_cq = attr.send_cq;
1728 qp->recv_cq = attr.recv_cq;
1730 qp->event_handler = attr.event_handler;
1731 qp->qp_context = attr.qp_context;
1732 qp->qp_type = attr.qp_type;
1733 atomic_set(&qp->usecnt, 0);
1734 atomic_inc(&pd->usecnt);
1735 atomic_inc(&attr.send_cq->usecnt);
1737 atomic_inc(&attr.recv_cq->usecnt);
1739 atomic_inc(&attr.srq->usecnt);
1741 qp->uobject = &obj->uevent.uobject;
1743 obj->uevent.uobject.object = qp;
1744 ret = idr_add_uobj(&ib_uverbs_qp_idr, &obj->uevent.uobject);
1748 memset(&resp, 0, sizeof resp);
1749 resp.qpn = qp->qp_num;
1750 resp.qp_handle = obj->uevent.uobject.id;
1751 resp.max_recv_sge = attr.cap.max_recv_sge;
1752 resp.max_send_sge = attr.cap.max_send_sge;
1753 resp.max_recv_wr = attr.cap.max_recv_wr;
1754 resp.max_send_wr = attr.cap.max_send_wr;
1755 resp.max_inline_data = attr.cap.max_inline_data;
1757 if (copy_to_user((void __user *) (unsigned long) cmd.response,
1758 &resp, sizeof resp)) {
1764 obj->uxrcd = container_of(xrcd_uobj, struct ib_uxrcd_object,
1766 atomic_inc(&obj->uxrcd->refcnt);
1767 put_xrcd_read(xrcd_uobj);
1774 if (rcq && rcq != scq)
1779 mutex_lock(&file->mutex);
1780 list_add_tail(&obj->uevent.uobject.list, &file->ucontext->qp_list);
1781 mutex_unlock(&file->mutex);
1783 obj->uevent.uobject.live = 1;
1785 up_write(&obj->uevent.uobject.mutex);
1790 idr_remove_uobj(&ib_uverbs_qp_idr, &obj->uevent.uobject);
1797 put_xrcd_read(xrcd_uobj);
1802 if (rcq && rcq != scq)
1807 put_uobj_write(&obj->uevent.uobject);
1811 ssize_t ib_uverbs_open_qp(struct ib_uverbs_file *file,
1812 const char __user *buf, int in_len, int out_len)
1814 struct ib_uverbs_open_qp cmd;
1815 struct ib_uverbs_create_qp_resp resp;
1816 struct ib_udata udata;
1817 struct ib_uqp_object *obj;
1818 struct ib_xrcd *xrcd;
1819 struct ib_uobject *uninitialized_var(xrcd_uobj);
1821 struct ib_qp_open_attr attr;
1824 if (out_len < sizeof resp)
1827 if (copy_from_user(&cmd, buf, sizeof cmd))
1830 INIT_UDATA(&udata, buf + sizeof cmd,
1831 (unsigned long) cmd.response + sizeof resp,
1832 in_len - sizeof cmd, out_len - sizeof resp);
1834 obj = kmalloc(sizeof *obj, GFP_KERNEL);
1838 init_uobj(&obj->uevent.uobject, cmd.user_handle, file->ucontext, &qp_lock_class);
1839 down_write(&obj->uevent.uobject.mutex);
1841 xrcd = idr_read_xrcd(cmd.pd_handle, file->ucontext, &xrcd_uobj);
1847 attr.event_handler = ib_uverbs_qp_event_handler;
1848 attr.qp_context = file;
1849 attr.qp_num = cmd.qpn;
1850 attr.qp_type = cmd.qp_type;
1852 obj->uevent.events_reported = 0;
1853 INIT_LIST_HEAD(&obj->uevent.event_list);
1854 INIT_LIST_HEAD(&obj->mcast_list);
1856 qp = ib_open_qp(xrcd, &attr);
1862 qp->uobject = &obj->uevent.uobject;
1864 obj->uevent.uobject.object = qp;
1865 ret = idr_add_uobj(&ib_uverbs_qp_idr, &obj->uevent.uobject);
1869 memset(&resp, 0, sizeof resp);
1870 resp.qpn = qp->qp_num;
1871 resp.qp_handle = obj->uevent.uobject.id;
1873 if (copy_to_user((void __user *) (unsigned long) cmd.response,
1874 &resp, sizeof resp)) {
1879 obj->uxrcd = container_of(xrcd_uobj, struct ib_uxrcd_object, uobject);
1880 atomic_inc(&obj->uxrcd->refcnt);
1881 put_xrcd_read(xrcd_uobj);
1883 mutex_lock(&file->mutex);
1884 list_add_tail(&obj->uevent.uobject.list, &file->ucontext->qp_list);
1885 mutex_unlock(&file->mutex);
1887 obj->uevent.uobject.live = 1;
1889 up_write(&obj->uevent.uobject.mutex);
1894 idr_remove_uobj(&ib_uverbs_qp_idr, &obj->uevent.uobject);
1900 put_xrcd_read(xrcd_uobj);
1901 put_uobj_write(&obj->uevent.uobject);
1905 ssize_t ib_uverbs_query_qp(struct ib_uverbs_file *file,
1906 const char __user *buf, int in_len,
1909 struct ib_uverbs_query_qp cmd;
1910 struct ib_uverbs_query_qp_resp resp;
1912 struct ib_qp_attr *attr;
1913 struct ib_qp_init_attr *init_attr;
1916 if (copy_from_user(&cmd, buf, sizeof cmd))
1919 attr = kmalloc(sizeof *attr, GFP_KERNEL);
1920 init_attr = kmalloc(sizeof *init_attr, GFP_KERNEL);
1921 if (!attr || !init_attr) {
1926 qp = idr_read_qp(cmd.qp_handle, file->ucontext);
1932 ret = ib_query_qp(qp, attr, cmd.attr_mask, init_attr);
1939 memset(&resp, 0, sizeof resp);
1941 resp.qp_state = attr->qp_state;
1942 resp.cur_qp_state = attr->cur_qp_state;
1943 resp.path_mtu = attr->path_mtu;
1944 resp.path_mig_state = attr->path_mig_state;
1945 resp.qkey = attr->qkey;
1946 resp.rq_psn = attr->rq_psn;
1947 resp.sq_psn = attr->sq_psn;
1948 resp.dest_qp_num = attr->dest_qp_num;
1949 resp.qp_access_flags = attr->qp_access_flags;
1950 resp.pkey_index = attr->pkey_index;
1951 resp.alt_pkey_index = attr->alt_pkey_index;
1952 resp.sq_draining = attr->sq_draining;
1953 resp.max_rd_atomic = attr->max_rd_atomic;
1954 resp.max_dest_rd_atomic = attr->max_dest_rd_atomic;
1955 resp.min_rnr_timer = attr->min_rnr_timer;
1956 resp.port_num = attr->port_num;
1957 resp.timeout = attr->timeout;
1958 resp.retry_cnt = attr->retry_cnt;
1959 resp.rnr_retry = attr->rnr_retry;
1960 resp.alt_port_num = attr->alt_port_num;
1961 resp.alt_timeout = attr->alt_timeout;
1963 memcpy(resp.dest.dgid, attr->ah_attr.grh.dgid.raw, 16);
1964 resp.dest.flow_label = attr->ah_attr.grh.flow_label;
1965 resp.dest.sgid_index = attr->ah_attr.grh.sgid_index;
1966 resp.dest.hop_limit = attr->ah_attr.grh.hop_limit;
1967 resp.dest.traffic_class = attr->ah_attr.grh.traffic_class;
1968 resp.dest.dlid = attr->ah_attr.dlid;
1969 resp.dest.sl = attr->ah_attr.sl;
1970 resp.dest.src_path_bits = attr->ah_attr.src_path_bits;
1971 resp.dest.static_rate = attr->ah_attr.static_rate;
1972 resp.dest.is_global = !!(attr->ah_attr.ah_flags & IB_AH_GRH);
1973 resp.dest.port_num = attr->ah_attr.port_num;
1975 memcpy(resp.alt_dest.dgid, attr->alt_ah_attr.grh.dgid.raw, 16);
1976 resp.alt_dest.flow_label = attr->alt_ah_attr.grh.flow_label;
1977 resp.alt_dest.sgid_index = attr->alt_ah_attr.grh.sgid_index;
1978 resp.alt_dest.hop_limit = attr->alt_ah_attr.grh.hop_limit;
1979 resp.alt_dest.traffic_class = attr->alt_ah_attr.grh.traffic_class;
1980 resp.alt_dest.dlid = attr->alt_ah_attr.dlid;
1981 resp.alt_dest.sl = attr->alt_ah_attr.sl;
1982 resp.alt_dest.src_path_bits = attr->alt_ah_attr.src_path_bits;
1983 resp.alt_dest.static_rate = attr->alt_ah_attr.static_rate;
1984 resp.alt_dest.is_global = !!(attr->alt_ah_attr.ah_flags & IB_AH_GRH);
1985 resp.alt_dest.port_num = attr->alt_ah_attr.port_num;
1987 resp.max_send_wr = init_attr->cap.max_send_wr;
1988 resp.max_recv_wr = init_attr->cap.max_recv_wr;
1989 resp.max_send_sge = init_attr->cap.max_send_sge;
1990 resp.max_recv_sge = init_attr->cap.max_recv_sge;
1991 resp.max_inline_data = init_attr->cap.max_inline_data;
1992 resp.sq_sig_all = init_attr->sq_sig_type == IB_SIGNAL_ALL_WR;
1994 if (copy_to_user((void __user *) (unsigned long) cmd.response,
1995 &resp, sizeof resp))
2002 return ret ? ret : in_len;
2005 /* Remove ignored fields set in the attribute mask */
2006 static int modify_qp_mask(enum ib_qp_type qp_type, int mask)
2009 case IB_QPT_XRC_INI:
2010 return mask & ~(IB_QP_MAX_DEST_RD_ATOMIC | IB_QP_MIN_RNR_TIMER);
2011 case IB_QPT_XRC_TGT:
2012 return mask & ~(IB_QP_MAX_QP_RD_ATOMIC | IB_QP_RETRY_CNT |
2019 ssize_t ib_uverbs_modify_qp(struct ib_uverbs_file *file,
2020 const char __user *buf, int in_len,
2023 struct ib_uverbs_modify_qp cmd;
2024 struct ib_udata udata;
2026 struct ib_qp_attr *attr;
2029 if (copy_from_user(&cmd, buf, sizeof cmd))
2032 INIT_UDATA(&udata, buf + sizeof cmd, NULL, in_len - sizeof cmd,
2035 attr = kmalloc(sizeof *attr, GFP_KERNEL);
2039 qp = idr_read_qp(cmd.qp_handle, file->ucontext);
2045 attr->qp_state = cmd.qp_state;
2046 attr->cur_qp_state = cmd.cur_qp_state;
2047 attr->path_mtu = cmd.path_mtu;
2048 attr->path_mig_state = cmd.path_mig_state;
2049 attr->qkey = cmd.qkey;
2050 attr->rq_psn = cmd.rq_psn;
2051 attr->sq_psn = cmd.sq_psn;
2052 attr->dest_qp_num = cmd.dest_qp_num;
2053 attr->qp_access_flags = cmd.qp_access_flags;
2054 attr->pkey_index = cmd.pkey_index;
2055 attr->alt_pkey_index = cmd.alt_pkey_index;
2056 attr->en_sqd_async_notify = cmd.en_sqd_async_notify;
2057 attr->max_rd_atomic = cmd.max_rd_atomic;
2058 attr->max_dest_rd_atomic = cmd.max_dest_rd_atomic;
2059 attr->min_rnr_timer = cmd.min_rnr_timer;
2060 attr->port_num = cmd.port_num;
2061 attr->timeout = cmd.timeout;
2062 attr->retry_cnt = cmd.retry_cnt;
2063 attr->rnr_retry = cmd.rnr_retry;
2064 attr->alt_port_num = cmd.alt_port_num;
2065 attr->alt_timeout = cmd.alt_timeout;
2067 memcpy(attr->ah_attr.grh.dgid.raw, cmd.dest.dgid, 16);
2068 attr->ah_attr.grh.flow_label = cmd.dest.flow_label;
2069 attr->ah_attr.grh.sgid_index = cmd.dest.sgid_index;
2070 attr->ah_attr.grh.hop_limit = cmd.dest.hop_limit;
2071 attr->ah_attr.grh.traffic_class = cmd.dest.traffic_class;
2072 attr->ah_attr.dlid = cmd.dest.dlid;
2073 attr->ah_attr.sl = cmd.dest.sl;
2074 attr->ah_attr.src_path_bits = cmd.dest.src_path_bits;
2075 attr->ah_attr.static_rate = cmd.dest.static_rate;
2076 attr->ah_attr.ah_flags = cmd.dest.is_global ? IB_AH_GRH : 0;
2077 attr->ah_attr.port_num = cmd.dest.port_num;
2079 memcpy(attr->alt_ah_attr.grh.dgid.raw, cmd.alt_dest.dgid, 16);
2080 attr->alt_ah_attr.grh.flow_label = cmd.alt_dest.flow_label;
2081 attr->alt_ah_attr.grh.sgid_index = cmd.alt_dest.sgid_index;
2082 attr->alt_ah_attr.grh.hop_limit = cmd.alt_dest.hop_limit;
2083 attr->alt_ah_attr.grh.traffic_class = cmd.alt_dest.traffic_class;
2084 attr->alt_ah_attr.dlid = cmd.alt_dest.dlid;
2085 attr->alt_ah_attr.sl = cmd.alt_dest.sl;
2086 attr->alt_ah_attr.src_path_bits = cmd.alt_dest.src_path_bits;
2087 attr->alt_ah_attr.static_rate = cmd.alt_dest.static_rate;
2088 attr->alt_ah_attr.ah_flags = cmd.alt_dest.is_global ? IB_AH_GRH : 0;
2089 attr->alt_ah_attr.port_num = cmd.alt_dest.port_num;
2091 if (qp->real_qp == qp) {
2092 ret = ib_resolve_eth_l2_attrs(qp, attr, &cmd.attr_mask);
2095 ret = qp->device->modify_qp(qp, attr,
2096 modify_qp_mask(qp->qp_type, cmd.attr_mask), &udata);
2098 ret = ib_modify_qp(qp, attr, modify_qp_mask(qp->qp_type, cmd.attr_mask));
2115 ssize_t ib_uverbs_destroy_qp(struct ib_uverbs_file *file,
2116 const char __user *buf, int in_len,
2119 struct ib_uverbs_destroy_qp cmd;
2120 struct ib_uverbs_destroy_qp_resp resp;
2121 struct ib_uobject *uobj;
2123 struct ib_uqp_object *obj;
2126 if (copy_from_user(&cmd, buf, sizeof cmd))
2129 memset(&resp, 0, sizeof resp);
2131 uobj = idr_write_uobj(&ib_uverbs_qp_idr, cmd.qp_handle, file->ucontext);
2135 obj = container_of(uobj, struct ib_uqp_object, uevent.uobject);
2137 if (!list_empty(&obj->mcast_list)) {
2138 put_uobj_write(uobj);
2142 ret = ib_destroy_qp(qp);
2146 put_uobj_write(uobj);
2152 atomic_dec(&obj->uxrcd->refcnt);
2154 idr_remove_uobj(&ib_uverbs_qp_idr, uobj);
2156 mutex_lock(&file->mutex);
2157 list_del(&uobj->list);
2158 mutex_unlock(&file->mutex);
2160 ib_uverbs_release_uevent(file, &obj->uevent);
2162 resp.events_reported = obj->uevent.events_reported;
2166 if (copy_to_user((void __user *) (unsigned long) cmd.response,
2167 &resp, sizeof resp))
2173 ssize_t ib_uverbs_post_send(struct ib_uverbs_file *file,
2174 const char __user *buf, int in_len,
2177 struct ib_uverbs_post_send cmd;
2178 struct ib_uverbs_post_send_resp resp;
2179 struct ib_uverbs_send_wr *user_wr;
2180 struct ib_send_wr *wr = NULL, *last, *next, *bad_wr;
2184 ssize_t ret = -EINVAL;
2186 if (copy_from_user(&cmd, buf, sizeof cmd))
2189 if (in_len < sizeof cmd + cmd.wqe_size * cmd.wr_count +
2190 cmd.sge_count * sizeof (struct ib_uverbs_sge))
2193 if (cmd.wqe_size < sizeof (struct ib_uverbs_send_wr))
2196 user_wr = kmalloc(cmd.wqe_size, GFP_KERNEL);
2200 qp = idr_read_qp(cmd.qp_handle, file->ucontext);
2204 is_ud = qp->qp_type == IB_QPT_UD;
2207 for (i = 0; i < cmd.wr_count; ++i) {
2208 if (copy_from_user(user_wr,
2209 buf + sizeof cmd + i * cmd.wqe_size,
2215 if (user_wr->num_sge + sg_ind > cmd.sge_count) {
2220 next = kmalloc(ALIGN(sizeof *next, sizeof (struct ib_sge)) +
2221 user_wr->num_sge * sizeof (struct ib_sge),
2235 next->wr_id = user_wr->wr_id;
2236 next->num_sge = user_wr->num_sge;
2237 next->opcode = user_wr->opcode;
2238 next->send_flags = user_wr->send_flags;
2241 next->wr.ud.ah = idr_read_ah(user_wr->wr.ud.ah,
2243 if (!next->wr.ud.ah) {
2247 next->wr.ud.remote_qpn = user_wr->wr.ud.remote_qpn;
2248 next->wr.ud.remote_qkey = user_wr->wr.ud.remote_qkey;
2249 if (next->opcode == IB_WR_SEND_WITH_IMM)
2251 (__be32 __force) user_wr->ex.imm_data;
2253 switch (next->opcode) {
2254 case IB_WR_RDMA_WRITE_WITH_IMM:
2256 (__be32 __force) user_wr->ex.imm_data;
2257 case IB_WR_RDMA_WRITE:
2258 case IB_WR_RDMA_READ:
2259 next->wr.rdma.remote_addr =
2260 user_wr->wr.rdma.remote_addr;
2261 next->wr.rdma.rkey =
2262 user_wr->wr.rdma.rkey;
2264 case IB_WR_SEND_WITH_IMM:
2266 (__be32 __force) user_wr->ex.imm_data;
2268 case IB_WR_SEND_WITH_INV:
2269 next->ex.invalidate_rkey =
2270 user_wr->ex.invalidate_rkey;
2272 case IB_WR_ATOMIC_CMP_AND_SWP:
2273 case IB_WR_ATOMIC_FETCH_AND_ADD:
2274 next->wr.atomic.remote_addr =
2275 user_wr->wr.atomic.remote_addr;
2276 next->wr.atomic.compare_add =
2277 user_wr->wr.atomic.compare_add;
2278 next->wr.atomic.swap = user_wr->wr.atomic.swap;
2279 next->wr.atomic.rkey = user_wr->wr.atomic.rkey;
2286 if (next->num_sge) {
2287 next->sg_list = (void *) next +
2288 ALIGN(sizeof *next, sizeof (struct ib_sge));
2289 if (copy_from_user(next->sg_list,
2291 cmd.wr_count * cmd.wqe_size +
2292 sg_ind * sizeof (struct ib_sge),
2293 next->num_sge * sizeof (struct ib_sge))) {
2297 sg_ind += next->num_sge;
2299 next->sg_list = NULL;
2303 ret = qp->device->post_send(qp->real_qp, wr, &bad_wr);
2305 for (next = wr; next; next = next->next) {
2311 if (copy_to_user((void __user *) (unsigned long) cmd.response,
2312 &resp, sizeof resp))
2319 if (is_ud && wr->wr.ud.ah)
2320 put_ah_read(wr->wr.ud.ah);
2329 return ret ? ret : in_len;
2332 static struct ib_recv_wr *ib_uverbs_unmarshall_recv(const char __user *buf,
2338 struct ib_uverbs_recv_wr *user_wr;
2339 struct ib_recv_wr *wr = NULL, *last, *next;
2344 if (in_len < wqe_size * wr_count +
2345 sge_count * sizeof (struct ib_uverbs_sge))
2346 return ERR_PTR(-EINVAL);
2348 if (wqe_size < sizeof (struct ib_uverbs_recv_wr))
2349 return ERR_PTR(-EINVAL);
2351 user_wr = kmalloc(wqe_size, GFP_KERNEL);
2353 return ERR_PTR(-ENOMEM);
2357 for (i = 0; i < wr_count; ++i) {
2358 if (copy_from_user(user_wr, buf + i * wqe_size,
2364 if (user_wr->num_sge + sg_ind > sge_count) {
2369 next = kmalloc(ALIGN(sizeof *next, sizeof (struct ib_sge)) +
2370 user_wr->num_sge * sizeof (struct ib_sge),
2384 next->wr_id = user_wr->wr_id;
2385 next->num_sge = user_wr->num_sge;
2387 if (next->num_sge) {
2388 next->sg_list = (void *) next +
2389 ALIGN(sizeof *next, sizeof (struct ib_sge));
2390 if (copy_from_user(next->sg_list,
2391 buf + wr_count * wqe_size +
2392 sg_ind * sizeof (struct ib_sge),
2393 next->num_sge * sizeof (struct ib_sge))) {
2397 sg_ind += next->num_sge;
2399 next->sg_list = NULL;
2414 return ERR_PTR(ret);
2417 ssize_t ib_uverbs_post_recv(struct ib_uverbs_file *file,
2418 const char __user *buf, int in_len,
2421 struct ib_uverbs_post_recv cmd;
2422 struct ib_uverbs_post_recv_resp resp;
2423 struct ib_recv_wr *wr, *next, *bad_wr;
2425 ssize_t ret = -EINVAL;
2427 if (copy_from_user(&cmd, buf, sizeof cmd))
2430 wr = ib_uverbs_unmarshall_recv(buf + sizeof cmd,
2431 in_len - sizeof cmd, cmd.wr_count,
2432 cmd.sge_count, cmd.wqe_size);
2436 qp = idr_read_qp(cmd.qp_handle, file->ucontext);
2441 ret = qp->device->post_recv(qp->real_qp, wr, &bad_wr);
2446 for (next = wr; next; next = next->next) {
2452 if (copy_to_user((void __user *) (unsigned long) cmd.response,
2453 &resp, sizeof resp))
2463 return ret ? ret : in_len;
2466 ssize_t ib_uverbs_post_srq_recv(struct ib_uverbs_file *file,
2467 const char __user *buf, int in_len,
2470 struct ib_uverbs_post_srq_recv cmd;
2471 struct ib_uverbs_post_srq_recv_resp resp;
2472 struct ib_recv_wr *wr, *next, *bad_wr;
2474 ssize_t ret = -EINVAL;
2476 if (copy_from_user(&cmd, buf, sizeof cmd))
2479 wr = ib_uverbs_unmarshall_recv(buf + sizeof cmd,
2480 in_len - sizeof cmd, cmd.wr_count,
2481 cmd.sge_count, cmd.wqe_size);
2485 srq = idr_read_srq(cmd.srq_handle, file->ucontext);
2490 ret = srq->device->post_srq_recv(srq, wr, &bad_wr);
2495 for (next = wr; next; next = next->next) {
2501 if (copy_to_user((void __user *) (unsigned long) cmd.response,
2502 &resp, sizeof resp))
2512 return ret ? ret : in_len;
2515 ssize_t ib_uverbs_create_ah(struct ib_uverbs_file *file,
2516 const char __user *buf, int in_len,
2519 struct ib_uverbs_create_ah cmd;
2520 struct ib_uverbs_create_ah_resp resp;
2521 struct ib_uobject *uobj;
2524 struct ib_ah_attr attr;
2527 if (out_len < sizeof resp)
2530 if (copy_from_user(&cmd, buf, sizeof cmd))
2533 uobj = kmalloc(sizeof *uobj, GFP_KERNEL);
2537 init_uobj(uobj, cmd.user_handle, file->ucontext, &ah_lock_class);
2538 down_write(&uobj->mutex);
2540 pd = idr_read_pd(cmd.pd_handle, file->ucontext);
2546 attr.dlid = cmd.attr.dlid;
2547 attr.sl = cmd.attr.sl;
2548 attr.src_path_bits = cmd.attr.src_path_bits;
2549 attr.static_rate = cmd.attr.static_rate;
2550 attr.ah_flags = cmd.attr.is_global ? IB_AH_GRH : 0;
2551 attr.port_num = cmd.attr.port_num;
2552 attr.grh.flow_label = cmd.attr.grh.flow_label;
2553 attr.grh.sgid_index = cmd.attr.grh.sgid_index;
2554 attr.grh.hop_limit = cmd.attr.grh.hop_limit;
2555 attr.grh.traffic_class = cmd.attr.grh.traffic_class;
2557 memset(&attr.dmac, 0, sizeof(attr.dmac));
2558 memcpy(attr.grh.dgid.raw, cmd.attr.grh.dgid, 16);
2560 ah = ib_create_ah(pd, &attr);
2569 ret = idr_add_uobj(&ib_uverbs_ah_idr, uobj);
2573 resp.ah_handle = uobj->id;
2575 if (copy_to_user((void __user *) (unsigned long) cmd.response,
2576 &resp, sizeof resp)) {
2583 mutex_lock(&file->mutex);
2584 list_add_tail(&uobj->list, &file->ucontext->ah_list);
2585 mutex_unlock(&file->mutex);
2589 up_write(&uobj->mutex);
2594 idr_remove_uobj(&ib_uverbs_ah_idr, uobj);
2603 put_uobj_write(uobj);
2607 ssize_t ib_uverbs_destroy_ah(struct ib_uverbs_file *file,
2608 const char __user *buf, int in_len, int out_len)
2610 struct ib_uverbs_destroy_ah cmd;
2612 struct ib_uobject *uobj;
2615 if (copy_from_user(&cmd, buf, sizeof cmd))
2618 uobj = idr_write_uobj(&ib_uverbs_ah_idr, cmd.ah_handle, file->ucontext);
2623 ret = ib_destroy_ah(ah);
2627 put_uobj_write(uobj);
2632 idr_remove_uobj(&ib_uverbs_ah_idr, uobj);
2634 mutex_lock(&file->mutex);
2635 list_del(&uobj->list);
2636 mutex_unlock(&file->mutex);
2643 ssize_t ib_uverbs_attach_mcast(struct ib_uverbs_file *file,
2644 const char __user *buf, int in_len,
2647 struct ib_uverbs_attach_mcast cmd;
2649 struct ib_uqp_object *obj;
2650 struct ib_uverbs_mcast_entry *mcast;
2653 if (copy_from_user(&cmd, buf, sizeof cmd))
2656 qp = idr_write_qp(cmd.qp_handle, file->ucontext);
2660 obj = container_of(qp->uobject, struct ib_uqp_object, uevent.uobject);
2662 list_for_each_entry(mcast, &obj->mcast_list, list)
2663 if (cmd.mlid == mcast->lid &&
2664 !memcmp(cmd.gid, mcast->gid.raw, sizeof mcast->gid.raw)) {
2669 mcast = kmalloc(sizeof *mcast, GFP_KERNEL);
2675 mcast->lid = cmd.mlid;
2676 memcpy(mcast->gid.raw, cmd.gid, sizeof mcast->gid.raw);
2678 ret = ib_attach_mcast(qp, &mcast->gid, cmd.mlid);
2680 list_add_tail(&mcast->list, &obj->mcast_list);
2687 return ret ? ret : in_len;
2690 ssize_t ib_uverbs_detach_mcast(struct ib_uverbs_file *file,
2691 const char __user *buf, int in_len,
2694 struct ib_uverbs_detach_mcast cmd;
2695 struct ib_uqp_object *obj;
2697 struct ib_uverbs_mcast_entry *mcast;
2700 if (copy_from_user(&cmd, buf, sizeof cmd))
2703 qp = idr_write_qp(cmd.qp_handle, file->ucontext);
2707 ret = ib_detach_mcast(qp, (union ib_gid *) cmd.gid, cmd.mlid);
2711 obj = container_of(qp->uobject, struct ib_uqp_object, uevent.uobject);
2713 list_for_each_entry(mcast, &obj->mcast_list, list)
2714 if (cmd.mlid == mcast->lid &&
2715 !memcmp(cmd.gid, mcast->gid.raw, sizeof mcast->gid.raw)) {
2716 list_del(&mcast->list);
2724 return ret ? ret : in_len;
2727 static int kern_spec_to_ib_spec(struct ib_uverbs_flow_spec *kern_spec,
2728 union ib_flow_spec *ib_spec)
2730 if (kern_spec->reserved)
2733 ib_spec->type = kern_spec->type;
2735 switch (ib_spec->type) {
2736 case IB_FLOW_SPEC_ETH:
2737 ib_spec->eth.size = sizeof(struct ib_flow_spec_eth);
2738 if (ib_spec->eth.size != kern_spec->eth.size)
2740 memcpy(&ib_spec->eth.val, &kern_spec->eth.val,
2741 sizeof(struct ib_flow_eth_filter));
2742 memcpy(&ib_spec->eth.mask, &kern_spec->eth.mask,
2743 sizeof(struct ib_flow_eth_filter));
2745 case IB_FLOW_SPEC_IPV4:
2746 ib_spec->ipv4.size = sizeof(struct ib_flow_spec_ipv4);
2747 if (ib_spec->ipv4.size != kern_spec->ipv4.size)
2749 memcpy(&ib_spec->ipv4.val, &kern_spec->ipv4.val,
2750 sizeof(struct ib_flow_ipv4_filter));
2751 memcpy(&ib_spec->ipv4.mask, &kern_spec->ipv4.mask,
2752 sizeof(struct ib_flow_ipv4_filter));
2754 case IB_FLOW_SPEC_TCP:
2755 case IB_FLOW_SPEC_UDP:
2756 ib_spec->tcp_udp.size = sizeof(struct ib_flow_spec_tcp_udp);
2757 if (ib_spec->tcp_udp.size != kern_spec->tcp_udp.size)
2759 memcpy(&ib_spec->tcp_udp.val, &kern_spec->tcp_udp.val,
2760 sizeof(struct ib_flow_tcp_udp_filter));
2761 memcpy(&ib_spec->tcp_udp.mask, &kern_spec->tcp_udp.mask,
2762 sizeof(struct ib_flow_tcp_udp_filter));
2770 int ib_uverbs_ex_create_flow(struct ib_uverbs_file *file,
2771 struct ib_udata *ucore,
2772 struct ib_udata *uhw)
2774 struct ib_uverbs_create_flow cmd;
2775 struct ib_uverbs_create_flow_resp resp;
2776 struct ib_uobject *uobj;
2777 struct ib_flow *flow_id;
2778 struct ib_uverbs_flow_attr *kern_flow_attr;
2779 struct ib_flow_attr *flow_attr;
2786 if (ucore->inlen < sizeof(cmd))
2789 if (ucore->outlen < sizeof(resp))
2792 err = ib_copy_from_udata(&cmd, ucore, sizeof(cmd));
2796 ucore->inbuf += sizeof(cmd);
2797 ucore->inlen -= sizeof(cmd);
2802 if ((cmd.flow_attr.type == IB_FLOW_ATTR_SNIFFER &&
2803 !capable(CAP_NET_ADMIN)) || !capable(CAP_NET_RAW))
2806 if (cmd.flow_attr.num_of_specs > IB_FLOW_SPEC_SUPPORT_LAYERS)
2809 if (cmd.flow_attr.size > ucore->inlen ||
2810 cmd.flow_attr.size >
2811 (cmd.flow_attr.num_of_specs * sizeof(struct ib_uverbs_flow_spec)))
2814 if (cmd.flow_attr.reserved[0] ||
2815 cmd.flow_attr.reserved[1])
2818 if (cmd.flow_attr.num_of_specs) {
2819 kern_flow_attr = kmalloc(sizeof(*kern_flow_attr) + cmd.flow_attr.size,
2821 if (!kern_flow_attr)
2824 memcpy(kern_flow_attr, &cmd.flow_attr, sizeof(*kern_flow_attr));
2825 err = ib_copy_from_udata(kern_flow_attr + 1, ucore,
2826 cmd.flow_attr.size);
2830 kern_flow_attr = &cmd.flow_attr;
2833 uobj = kmalloc(sizeof(*uobj), GFP_KERNEL);
2838 init_uobj(uobj, 0, file->ucontext, &rule_lock_class);
2839 down_write(&uobj->mutex);
2841 qp = idr_read_qp(cmd.qp_handle, file->ucontext);
2847 flow_attr = kmalloc(sizeof(*flow_attr) + cmd.flow_attr.size, GFP_KERNEL);
2853 flow_attr->type = kern_flow_attr->type;
2854 flow_attr->priority = kern_flow_attr->priority;
2855 flow_attr->num_of_specs = kern_flow_attr->num_of_specs;
2856 flow_attr->port = kern_flow_attr->port;
2857 flow_attr->flags = kern_flow_attr->flags;
2858 flow_attr->size = sizeof(*flow_attr);
2860 kern_spec = kern_flow_attr + 1;
2861 ib_spec = flow_attr + 1;
2862 for (i = 0; i < flow_attr->num_of_specs &&
2863 cmd.flow_attr.size > offsetof(struct ib_uverbs_flow_spec, reserved) &&
2864 cmd.flow_attr.size >=
2865 ((struct ib_uverbs_flow_spec *)kern_spec)->size; i++) {
2866 err = kern_spec_to_ib_spec(kern_spec, ib_spec);
2870 ((union ib_flow_spec *) ib_spec)->size;
2871 cmd.flow_attr.size -= ((struct ib_uverbs_flow_spec *)kern_spec)->size;
2872 kern_spec += ((struct ib_uverbs_flow_spec *) kern_spec)->size;
2873 ib_spec += ((union ib_flow_spec *) ib_spec)->size;
2875 if (cmd.flow_attr.size || (i != flow_attr->num_of_specs)) {
2876 pr_warn("create flow failed, flow %d: %d bytes left from uverb cmd\n",
2877 i, cmd.flow_attr.size);
2881 flow_id = ib_create_flow(qp, flow_attr, IB_FLOW_DOMAIN_USER);
2882 if (IS_ERR(flow_id)) {
2883 err = PTR_ERR(flow_id);
2887 flow_id->uobject = uobj;
2888 uobj->object = flow_id;
2890 err = idr_add_uobj(&ib_uverbs_rule_idr, uobj);
2894 memset(&resp, 0, sizeof(resp));
2895 resp.flow_handle = uobj->id;
2897 err = ib_copy_to_udata(ucore,
2898 &resp, sizeof(resp));
2903 mutex_lock(&file->mutex);
2904 list_add_tail(&uobj->list, &file->ucontext->rule_list);
2905 mutex_unlock(&file->mutex);
2909 up_write(&uobj->mutex);
2911 if (cmd.flow_attr.num_of_specs)
2912 kfree(kern_flow_attr);
2915 idr_remove_uobj(&ib_uverbs_rule_idr, uobj);
2917 ib_destroy_flow(flow_id);
2923 put_uobj_write(uobj);
2925 if (cmd.flow_attr.num_of_specs)
2926 kfree(kern_flow_attr);
2930 int ib_uverbs_ex_destroy_flow(struct ib_uverbs_file *file,
2931 struct ib_udata *ucore,
2932 struct ib_udata *uhw)
2934 struct ib_uverbs_destroy_flow cmd;
2935 struct ib_flow *flow_id;
2936 struct ib_uobject *uobj;
2939 if (ucore->inlen < sizeof(cmd))
2942 ret = ib_copy_from_udata(&cmd, ucore, sizeof(cmd));
2949 uobj = idr_write_uobj(&ib_uverbs_rule_idr, cmd.flow_handle,
2953 flow_id = uobj->object;
2955 ret = ib_destroy_flow(flow_id);
2959 put_uobj_write(uobj);
2961 idr_remove_uobj(&ib_uverbs_rule_idr, uobj);
2963 mutex_lock(&file->mutex);
2964 list_del(&uobj->list);
2965 mutex_unlock(&file->mutex);
2972 static int __uverbs_create_xsrq(struct ib_uverbs_file *file,
2973 struct ib_uverbs_create_xsrq *cmd,
2974 struct ib_udata *udata)
2976 struct ib_uverbs_create_srq_resp resp;
2977 struct ib_usrq_object *obj;
2980 struct ib_uobject *uninitialized_var(xrcd_uobj);
2981 struct ib_srq_init_attr attr;
2984 obj = kmalloc(sizeof *obj, GFP_KERNEL);
2988 init_uobj(&obj->uevent.uobject, cmd->user_handle, file->ucontext, &srq_lock_class);
2989 down_write(&obj->uevent.uobject.mutex);
2991 if (cmd->srq_type == IB_SRQT_XRC) {
2992 attr.ext.xrc.xrcd = idr_read_xrcd(cmd->xrcd_handle, file->ucontext, &xrcd_uobj);
2993 if (!attr.ext.xrc.xrcd) {
2998 obj->uxrcd = container_of(xrcd_uobj, struct ib_uxrcd_object, uobject);
2999 atomic_inc(&obj->uxrcd->refcnt);
3001 attr.ext.xrc.cq = idr_read_cq(cmd->cq_handle, file->ucontext, 0);
3002 if (!attr.ext.xrc.cq) {
3008 pd = idr_read_pd(cmd->pd_handle, file->ucontext);
3014 attr.event_handler = ib_uverbs_srq_event_handler;
3015 attr.srq_context = file;
3016 attr.srq_type = cmd->srq_type;
3017 attr.attr.max_wr = cmd->max_wr;
3018 attr.attr.max_sge = cmd->max_sge;
3019 attr.attr.srq_limit = cmd->srq_limit;
3021 obj->uevent.events_reported = 0;
3022 INIT_LIST_HEAD(&obj->uevent.event_list);
3024 srq = pd->device->create_srq(pd, &attr, udata);
3030 srq->device = pd->device;
3032 srq->srq_type = cmd->srq_type;
3033 srq->uobject = &obj->uevent.uobject;
3034 srq->event_handler = attr.event_handler;
3035 srq->srq_context = attr.srq_context;
3037 if (cmd->srq_type == IB_SRQT_XRC) {
3038 srq->ext.xrc.cq = attr.ext.xrc.cq;
3039 srq->ext.xrc.xrcd = attr.ext.xrc.xrcd;
3040 atomic_inc(&attr.ext.xrc.cq->usecnt);
3041 atomic_inc(&attr.ext.xrc.xrcd->usecnt);
3044 atomic_inc(&pd->usecnt);
3045 atomic_set(&srq->usecnt, 0);
3047 obj->uevent.uobject.object = srq;
3048 ret = idr_add_uobj(&ib_uverbs_srq_idr, &obj->uevent.uobject);
3052 memset(&resp, 0, sizeof resp);
3053 resp.srq_handle = obj->uevent.uobject.id;
3054 resp.max_wr = attr.attr.max_wr;
3055 resp.max_sge = attr.attr.max_sge;
3056 if (cmd->srq_type == IB_SRQT_XRC)
3057 resp.srqn = srq->ext.xrc.srq_num;
3059 if (copy_to_user((void __user *) (unsigned long) cmd->response,
3060 &resp, sizeof resp)) {
3065 if (cmd->srq_type == IB_SRQT_XRC) {
3066 put_uobj_read(xrcd_uobj);
3067 put_cq_read(attr.ext.xrc.cq);
3071 mutex_lock(&file->mutex);
3072 list_add_tail(&obj->uevent.uobject.list, &file->ucontext->srq_list);
3073 mutex_unlock(&file->mutex);
3075 obj->uevent.uobject.live = 1;
3077 up_write(&obj->uevent.uobject.mutex);
3082 idr_remove_uobj(&ib_uverbs_srq_idr, &obj->uevent.uobject);
3085 ib_destroy_srq(srq);
3091 if (cmd->srq_type == IB_SRQT_XRC)
3092 put_cq_read(attr.ext.xrc.cq);
3095 if (cmd->srq_type == IB_SRQT_XRC) {
3096 atomic_dec(&obj->uxrcd->refcnt);
3097 put_uobj_read(xrcd_uobj);
3101 put_uobj_write(&obj->uevent.uobject);
3105 ssize_t ib_uverbs_create_srq(struct ib_uverbs_file *file,
3106 const char __user *buf, int in_len,
3109 struct ib_uverbs_create_srq cmd;
3110 struct ib_uverbs_create_xsrq xcmd;
3111 struct ib_uverbs_create_srq_resp resp;
3112 struct ib_udata udata;
3115 if (out_len < sizeof resp)
3118 if (copy_from_user(&cmd, buf, sizeof cmd))
3121 xcmd.response = cmd.response;
3122 xcmd.user_handle = cmd.user_handle;
3123 xcmd.srq_type = IB_SRQT_BASIC;
3124 xcmd.pd_handle = cmd.pd_handle;
3125 xcmd.max_wr = cmd.max_wr;
3126 xcmd.max_sge = cmd.max_sge;
3127 xcmd.srq_limit = cmd.srq_limit;
3129 INIT_UDATA(&udata, buf + sizeof cmd,
3130 (unsigned long) cmd.response + sizeof resp,
3131 in_len - sizeof cmd, out_len - sizeof resp);
3133 ret = __uverbs_create_xsrq(file, &xcmd, &udata);
3140 ssize_t ib_uverbs_create_xsrq(struct ib_uverbs_file *file,
3141 const char __user *buf, int in_len, int out_len)
3143 struct ib_uverbs_create_xsrq cmd;
3144 struct ib_uverbs_create_srq_resp resp;
3145 struct ib_udata udata;
3148 if (out_len < sizeof resp)
3151 if (copy_from_user(&cmd, buf, sizeof cmd))
3154 INIT_UDATA(&udata, buf + sizeof cmd,
3155 (unsigned long) cmd.response + sizeof resp,
3156 in_len - sizeof cmd, out_len - sizeof resp);
3158 ret = __uverbs_create_xsrq(file, &cmd, &udata);
3165 ssize_t ib_uverbs_modify_srq(struct ib_uverbs_file *file,
3166 const char __user *buf, int in_len,
3169 struct ib_uverbs_modify_srq cmd;
3170 struct ib_udata udata;
3172 struct ib_srq_attr attr;
3175 if (copy_from_user(&cmd, buf, sizeof cmd))
3178 INIT_UDATA(&udata, buf + sizeof cmd, NULL, in_len - sizeof cmd,
3181 srq = idr_read_srq(cmd.srq_handle, file->ucontext);
3185 attr.max_wr = cmd.max_wr;
3186 attr.srq_limit = cmd.srq_limit;
3188 ret = srq->device->modify_srq(srq, &attr, cmd.attr_mask, &udata);
3192 return ret ? ret : in_len;
3195 ssize_t ib_uverbs_query_srq(struct ib_uverbs_file *file,
3196 const char __user *buf,
3197 int in_len, int out_len)
3199 struct ib_uverbs_query_srq cmd;
3200 struct ib_uverbs_query_srq_resp resp;
3201 struct ib_srq_attr attr;
3205 if (out_len < sizeof resp)
3208 if (copy_from_user(&cmd, buf, sizeof cmd))
3211 srq = idr_read_srq(cmd.srq_handle, file->ucontext);
3215 ret = ib_query_srq(srq, &attr);
3222 memset(&resp, 0, sizeof resp);
3224 resp.max_wr = attr.max_wr;
3225 resp.max_sge = attr.max_sge;
3226 resp.srq_limit = attr.srq_limit;
3228 if (copy_to_user((void __user *) (unsigned long) cmd.response,
3229 &resp, sizeof resp))
3235 ssize_t ib_uverbs_destroy_srq(struct ib_uverbs_file *file,
3236 const char __user *buf, int in_len,
3239 struct ib_uverbs_destroy_srq cmd;
3240 struct ib_uverbs_destroy_srq_resp resp;
3241 struct ib_uobject *uobj;
3243 struct ib_uevent_object *obj;
3245 struct ib_usrq_object *us;
3246 enum ib_srq_type srq_type;
3248 if (copy_from_user(&cmd, buf, sizeof cmd))
3251 uobj = idr_write_uobj(&ib_uverbs_srq_idr, cmd.srq_handle, file->ucontext);
3255 obj = container_of(uobj, struct ib_uevent_object, uobject);
3256 srq_type = srq->srq_type;
3258 ret = ib_destroy_srq(srq);
3262 put_uobj_write(uobj);
3267 if (srq_type == IB_SRQT_XRC) {
3268 us = container_of(obj, struct ib_usrq_object, uevent);
3269 atomic_dec(&us->uxrcd->refcnt);
3272 idr_remove_uobj(&ib_uverbs_srq_idr, uobj);
3274 mutex_lock(&file->mutex);
3275 list_del(&uobj->list);
3276 mutex_unlock(&file->mutex);
3278 ib_uverbs_release_uevent(file, obj);
3280 memset(&resp, 0, sizeof resp);
3281 resp.events_reported = obj->events_reported;
3285 if (copy_to_user((void __user *) (unsigned long) cmd.response,
3286 &resp, sizeof resp))
3289 return ret ? ret : in_len;