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 <linux/uaccess.h>
43 #include <rdma/uverbs_types.h>
44 #include <rdma/uverbs_std_types.h>
45 #include "rdma_core.h"
48 #include "core_priv.h"
50 static struct ib_uverbs_completion_event_file *
51 _ib_uverbs_lookup_comp_file(s32 fd, struct ib_uverbs_file *ufile)
53 struct ib_uobject *uobj = ufd_get_read(UVERBS_OBJECT_COMP_CHANNEL,
59 uverbs_uobject_get(uobj);
62 return container_of(uobj, struct ib_uverbs_completion_event_file,
65 #define ib_uverbs_lookup_comp_file(_fd, _ufile) \
66 _ib_uverbs_lookup_comp_file((_fd)*typecheck(s32, _fd), _ufile)
68 ssize_t ib_uverbs_get_context(struct ib_uverbs_file *file,
69 const char __user *buf,
70 int in_len, int out_len)
72 struct ib_uverbs_get_context cmd;
73 struct ib_uverbs_get_context_resp resp;
74 struct ib_udata udata;
75 struct ib_ucontext *ucontext;
77 struct ib_rdmacg_object cg_obj;
78 struct ib_device *ib_dev;
81 if (out_len < sizeof resp)
84 if (copy_from_user(&cmd, buf, sizeof cmd))
87 mutex_lock(&file->ucontext_lock);
88 ib_dev = srcu_dereference(file->device->ib_dev,
89 &file->device->disassociate_srcu);
100 ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
101 u64_to_user_ptr(cmd.response) + sizeof(resp),
102 in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
103 out_len - sizeof(resp));
105 ret = ib_rdmacg_try_charge(&cg_obj, ib_dev, RDMACG_RESOURCE_HCA_HANDLE);
109 ucontext = ib_dev->alloc_ucontext(ib_dev, &udata);
110 if (IS_ERR(ucontext)) {
111 ret = PTR_ERR(ucontext);
115 ucontext->device = ib_dev;
116 ucontext->cg_obj = cg_obj;
117 /* ufile is required when some objects are released */
118 ucontext->ufile = file;
121 ucontext->tgid = get_task_pid(current->group_leader, PIDTYPE_PID);
123 ucontext->closing = 0;
124 ucontext->cleanup_retryable = false;
126 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
127 ucontext->umem_tree = RB_ROOT_CACHED;
128 init_rwsem(&ucontext->umem_rwsem);
129 ucontext->odp_mrs_count = 0;
130 INIT_LIST_HEAD(&ucontext->no_private_counters);
132 if (!(ib_dev->attrs.device_cap_flags & IB_DEVICE_ON_DEMAND_PAGING))
133 ucontext->invalidate_range = NULL;
137 resp.num_comp_vectors = file->device->num_comp_vectors;
139 ret = get_unused_fd_flags(O_CLOEXEC);
144 filp = ib_uverbs_alloc_async_event_file(file, ib_dev);
150 if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp)) {
155 fd_install(resp.async_fd, filp);
158 * Make sure that ib_uverbs_get_ucontext() sees the pointer update
159 * only after all writes to setup the ucontext have completed
161 smp_store_release(&file->ucontext, ucontext);
163 mutex_unlock(&file->ucontext_lock);
168 ib_uverbs_free_async_event_file(file);
172 put_unused_fd(resp.async_fd);
175 put_pid(ucontext->tgid);
176 ib_dev->dealloc_ucontext(ucontext);
179 ib_rdmacg_uncharge(&cg_obj, ib_dev, RDMACG_RESOURCE_HCA_HANDLE);
182 mutex_unlock(&file->ucontext_lock);
186 static void copy_query_dev_fields(struct ib_ucontext *ucontext,
187 struct ib_uverbs_query_device_resp *resp,
188 struct ib_device_attr *attr)
190 struct ib_device *ib_dev = ucontext->device;
192 resp->fw_ver = attr->fw_ver;
193 resp->node_guid = ib_dev->node_guid;
194 resp->sys_image_guid = attr->sys_image_guid;
195 resp->max_mr_size = attr->max_mr_size;
196 resp->page_size_cap = attr->page_size_cap;
197 resp->vendor_id = attr->vendor_id;
198 resp->vendor_part_id = attr->vendor_part_id;
199 resp->hw_ver = attr->hw_ver;
200 resp->max_qp = attr->max_qp;
201 resp->max_qp_wr = attr->max_qp_wr;
202 resp->device_cap_flags = lower_32_bits(attr->device_cap_flags);
203 resp->max_sge = min(attr->max_send_sge, attr->max_recv_sge);
204 resp->max_sge_rd = attr->max_sge_rd;
205 resp->max_cq = attr->max_cq;
206 resp->max_cqe = attr->max_cqe;
207 resp->max_mr = attr->max_mr;
208 resp->max_pd = attr->max_pd;
209 resp->max_qp_rd_atom = attr->max_qp_rd_atom;
210 resp->max_ee_rd_atom = attr->max_ee_rd_atom;
211 resp->max_res_rd_atom = attr->max_res_rd_atom;
212 resp->max_qp_init_rd_atom = attr->max_qp_init_rd_atom;
213 resp->max_ee_init_rd_atom = attr->max_ee_init_rd_atom;
214 resp->atomic_cap = attr->atomic_cap;
215 resp->max_ee = attr->max_ee;
216 resp->max_rdd = attr->max_rdd;
217 resp->max_mw = attr->max_mw;
218 resp->max_raw_ipv6_qp = attr->max_raw_ipv6_qp;
219 resp->max_raw_ethy_qp = attr->max_raw_ethy_qp;
220 resp->max_mcast_grp = attr->max_mcast_grp;
221 resp->max_mcast_qp_attach = attr->max_mcast_qp_attach;
222 resp->max_total_mcast_qp_attach = attr->max_total_mcast_qp_attach;
223 resp->max_ah = attr->max_ah;
224 resp->max_fmr = attr->max_fmr;
225 resp->max_map_per_fmr = attr->max_map_per_fmr;
226 resp->max_srq = attr->max_srq;
227 resp->max_srq_wr = attr->max_srq_wr;
228 resp->max_srq_sge = attr->max_srq_sge;
229 resp->max_pkeys = attr->max_pkeys;
230 resp->local_ca_ack_delay = attr->local_ca_ack_delay;
231 resp->phys_port_cnt = ib_dev->phys_port_cnt;
234 ssize_t ib_uverbs_query_device(struct ib_uverbs_file *file,
235 const char __user *buf,
236 int in_len, int out_len)
238 struct ib_uverbs_query_device cmd;
239 struct ib_uverbs_query_device_resp resp;
240 struct ib_ucontext *ucontext;
242 ucontext = ib_uverbs_get_ucontext(file);
243 if (IS_ERR(ucontext))
244 return PTR_ERR(ucontext);
246 if (out_len < sizeof resp)
249 if (copy_from_user(&cmd, buf, sizeof cmd))
252 memset(&resp, 0, sizeof resp);
253 copy_query_dev_fields(ucontext, &resp, &ucontext->device->attrs);
255 if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp))
262 * ib_uverbs_query_port_resp.port_cap_flags started out as just a copy of the
263 * PortInfo CapabilityMask, but was extended with unique bits.
265 static u32 make_port_cap_flags(const struct ib_port_attr *attr)
269 /* All IBA CapabilityMask bits are passed through here, except bit 26,
270 * which is overridden with IP_BASED_GIDS. This is due to a historical
271 * mistake in the implementation of IP_BASED_GIDS. Otherwise all other
272 * bits match the IBA definition across all kernel versions.
274 res = attr->port_cap_flags & ~(u32)IB_UVERBS_PCF_IP_BASED_GIDS;
277 res |= IB_UVERBS_PCF_IP_BASED_GIDS;
282 ssize_t ib_uverbs_query_port(struct ib_uverbs_file *file,
283 const char __user *buf,
284 int in_len, int out_len)
286 struct ib_uverbs_query_port cmd;
287 struct ib_uverbs_query_port_resp resp;
288 struct ib_port_attr attr;
290 struct ib_ucontext *ucontext;
291 struct ib_device *ib_dev;
293 ucontext = ib_uverbs_get_ucontext(file);
294 if (IS_ERR(ucontext))
295 return PTR_ERR(ucontext);
296 ib_dev = ucontext->device;
298 if (out_len < sizeof resp)
301 if (copy_from_user(&cmd, buf, sizeof cmd))
304 ret = ib_query_port(ib_dev, cmd.port_num, &attr);
308 memset(&resp, 0, sizeof resp);
310 resp.state = attr.state;
311 resp.max_mtu = attr.max_mtu;
312 resp.active_mtu = attr.active_mtu;
313 resp.gid_tbl_len = attr.gid_tbl_len;
314 resp.port_cap_flags = make_port_cap_flags(&attr);
315 resp.max_msg_sz = attr.max_msg_sz;
316 resp.bad_pkey_cntr = attr.bad_pkey_cntr;
317 resp.qkey_viol_cntr = attr.qkey_viol_cntr;
318 resp.pkey_tbl_len = attr.pkey_tbl_len;
320 if (rdma_is_grh_required(ib_dev, cmd.port_num))
321 resp.flags |= IB_UVERBS_QPF_GRH_REQUIRED;
323 if (rdma_cap_opa_ah(ib_dev, cmd.port_num)) {
324 resp.lid = OPA_TO_IB_UCAST_LID(attr.lid);
325 resp.sm_lid = OPA_TO_IB_UCAST_LID(attr.sm_lid);
327 resp.lid = ib_lid_cpu16(attr.lid);
328 resp.sm_lid = ib_lid_cpu16(attr.sm_lid);
331 resp.max_vl_num = attr.max_vl_num;
332 resp.sm_sl = attr.sm_sl;
333 resp.subnet_timeout = attr.subnet_timeout;
334 resp.init_type_reply = attr.init_type_reply;
335 resp.active_width = attr.active_width;
336 resp.active_speed = attr.active_speed;
337 resp.phys_state = attr.phys_state;
338 resp.link_layer = rdma_port_get_link_layer(ib_dev,
341 if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp))
347 ssize_t ib_uverbs_alloc_pd(struct ib_uverbs_file *file,
348 const char __user *buf,
349 int in_len, int out_len)
351 struct ib_uverbs_alloc_pd cmd;
352 struct ib_uverbs_alloc_pd_resp resp;
353 struct ib_udata udata;
354 struct ib_uobject *uobj;
357 struct ib_device *ib_dev;
359 if (out_len < sizeof resp)
362 if (copy_from_user(&cmd, buf, sizeof cmd))
365 ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
366 u64_to_user_ptr(cmd.response) + sizeof(resp),
367 in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
368 out_len - sizeof(resp));
370 uobj = uobj_alloc(UVERBS_OBJECT_PD, file, &ib_dev);
372 return PTR_ERR(uobj);
374 pd = ib_dev->alloc_pd(ib_dev, uobj->context, &udata);
382 pd->__internal_mr = NULL;
383 atomic_set(&pd->usecnt, 0);
386 memset(&resp, 0, sizeof resp);
387 resp.pd_handle = uobj->id;
388 pd->res.type = RDMA_RESTRACK_PD;
389 rdma_restrack_add(&pd->res);
391 if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp)) {
396 return uobj_alloc_commit(uobj, in_len);
402 uobj_alloc_abort(uobj);
406 ssize_t ib_uverbs_dealloc_pd(struct ib_uverbs_file *file,
407 const char __user *buf,
408 int in_len, int out_len)
410 struct ib_uverbs_dealloc_pd cmd;
412 if (copy_from_user(&cmd, buf, sizeof cmd))
415 return uobj_perform_destroy(UVERBS_OBJECT_PD, cmd.pd_handle, file,
419 struct xrcd_table_entry {
421 struct ib_xrcd *xrcd;
425 static int xrcd_table_insert(struct ib_uverbs_device *dev,
427 struct ib_xrcd *xrcd)
429 struct xrcd_table_entry *entry, *scan;
430 struct rb_node **p = &dev->xrcd_tree.rb_node;
431 struct rb_node *parent = NULL;
433 entry = kmalloc(sizeof *entry, GFP_KERNEL);
438 entry->inode = inode;
442 scan = rb_entry(parent, struct xrcd_table_entry, node);
444 if (inode < scan->inode) {
446 } else if (inode > scan->inode) {
454 rb_link_node(&entry->node, parent, p);
455 rb_insert_color(&entry->node, &dev->xrcd_tree);
460 static struct xrcd_table_entry *xrcd_table_search(struct ib_uverbs_device *dev,
463 struct xrcd_table_entry *entry;
464 struct rb_node *p = dev->xrcd_tree.rb_node;
467 entry = rb_entry(p, struct xrcd_table_entry, node);
469 if (inode < entry->inode)
471 else if (inode > entry->inode)
480 static struct ib_xrcd *find_xrcd(struct ib_uverbs_device *dev, struct inode *inode)
482 struct xrcd_table_entry *entry;
484 entry = xrcd_table_search(dev, inode);
491 static void xrcd_table_delete(struct ib_uverbs_device *dev,
494 struct xrcd_table_entry *entry;
496 entry = xrcd_table_search(dev, inode);
499 rb_erase(&entry->node, &dev->xrcd_tree);
504 ssize_t ib_uverbs_open_xrcd(struct ib_uverbs_file *file,
505 const char __user *buf, int in_len,
508 struct ib_uverbs_open_xrcd cmd;
509 struct ib_uverbs_open_xrcd_resp resp;
510 struct ib_udata udata;
511 struct ib_uxrcd_object *obj;
512 struct ib_xrcd *xrcd = NULL;
513 struct fd f = {NULL, 0};
514 struct inode *inode = NULL;
517 struct ib_device *ib_dev;
519 if (out_len < sizeof resp)
522 if (copy_from_user(&cmd, buf, sizeof cmd))
525 ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
526 u64_to_user_ptr(cmd.response) + sizeof(resp),
527 in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
528 out_len - sizeof(resp));
530 mutex_lock(&file->device->xrcd_tree_mutex);
533 /* search for file descriptor */
537 goto err_tree_mutex_unlock;
540 inode = file_inode(f.file);
541 xrcd = find_xrcd(file->device, inode);
542 if (!xrcd && !(cmd.oflags & O_CREAT)) {
543 /* no file descriptor. Need CREATE flag */
545 goto err_tree_mutex_unlock;
548 if (xrcd && cmd.oflags & O_EXCL) {
550 goto err_tree_mutex_unlock;
554 obj = (struct ib_uxrcd_object *)uobj_alloc(UVERBS_OBJECT_XRCD, file,
558 goto err_tree_mutex_unlock;
562 xrcd = ib_dev->alloc_xrcd(ib_dev, obj->uobject.context, &udata);
569 xrcd->device = ib_dev;
570 atomic_set(&xrcd->usecnt, 0);
571 mutex_init(&xrcd->tgt_qp_mutex);
572 INIT_LIST_HEAD(&xrcd->tgt_qp_list);
576 atomic_set(&obj->refcnt, 0);
577 obj->uobject.object = xrcd;
578 memset(&resp, 0, sizeof resp);
579 resp.xrcd_handle = obj->uobject.id;
583 /* create new inode/xrcd table entry */
584 ret = xrcd_table_insert(file->device, inode, xrcd);
586 goto err_dealloc_xrcd;
588 atomic_inc(&xrcd->usecnt);
591 if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp)) {
599 mutex_unlock(&file->device->xrcd_tree_mutex);
601 return uobj_alloc_commit(&obj->uobject, in_len);
606 xrcd_table_delete(file->device, inode);
607 atomic_dec(&xrcd->usecnt);
611 ib_dealloc_xrcd(xrcd);
614 uobj_alloc_abort(&obj->uobject);
616 err_tree_mutex_unlock:
620 mutex_unlock(&file->device->xrcd_tree_mutex);
625 ssize_t ib_uverbs_close_xrcd(struct ib_uverbs_file *file,
626 const char __user *buf, int in_len,
629 struct ib_uverbs_close_xrcd cmd;
631 if (copy_from_user(&cmd, buf, sizeof cmd))
634 return uobj_perform_destroy(UVERBS_OBJECT_XRCD, cmd.xrcd_handle, file,
638 int ib_uverbs_dealloc_xrcd(struct ib_uobject *uobject,
639 struct ib_xrcd *xrcd,
640 enum rdma_remove_reason why)
644 struct ib_uverbs_device *dev = uobject->context->ufile->device;
647 if (inode && !atomic_dec_and_test(&xrcd->usecnt))
650 ret = ib_dealloc_xrcd(xrcd);
652 if (ib_is_destroy_retryable(ret, why, uobject)) {
653 atomic_inc(&xrcd->usecnt);
658 xrcd_table_delete(dev, inode);
663 ssize_t ib_uverbs_reg_mr(struct ib_uverbs_file *file,
664 const char __user *buf, int in_len,
667 struct ib_uverbs_reg_mr cmd;
668 struct ib_uverbs_reg_mr_resp resp;
669 struct ib_udata udata;
670 struct ib_uobject *uobj;
674 struct ib_device *ib_dev;
676 if (out_len < sizeof resp)
679 if (copy_from_user(&cmd, buf, sizeof cmd))
682 ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
683 u64_to_user_ptr(cmd.response) + sizeof(resp),
684 in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
685 out_len - sizeof(resp));
687 if ((cmd.start & ~PAGE_MASK) != (cmd.hca_va & ~PAGE_MASK))
690 ret = ib_check_mr_access(cmd.access_flags);
694 uobj = uobj_alloc(UVERBS_OBJECT_MR, file, &ib_dev);
696 return PTR_ERR(uobj);
698 pd = uobj_get_obj_read(pd, UVERBS_OBJECT_PD, cmd.pd_handle, file);
704 if (cmd.access_flags & IB_ACCESS_ON_DEMAND) {
705 if (!(pd->device->attrs.device_cap_flags &
706 IB_DEVICE_ON_DEMAND_PAGING)) {
707 pr_debug("ODP support not available\n");
713 mr = pd->device->reg_user_mr(pd, cmd.start, cmd.length, cmd.hca_va,
714 cmd.access_flags, &udata);
720 mr->device = pd->device;
724 atomic_inc(&pd->usecnt);
725 mr->res.type = RDMA_RESTRACK_MR;
726 rdma_restrack_add(&mr->res);
730 memset(&resp, 0, sizeof resp);
731 resp.lkey = mr->lkey;
732 resp.rkey = mr->rkey;
733 resp.mr_handle = uobj->id;
735 if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp)) {
740 uobj_put_obj_read(pd);
742 return uobj_alloc_commit(uobj, in_len);
748 uobj_put_obj_read(pd);
751 uobj_alloc_abort(uobj);
755 ssize_t ib_uverbs_rereg_mr(struct ib_uverbs_file *file,
756 const char __user *buf, int in_len,
759 struct ib_uverbs_rereg_mr cmd;
760 struct ib_uverbs_rereg_mr_resp resp;
761 struct ib_udata udata;
762 struct ib_pd *pd = NULL;
764 struct ib_pd *old_pd;
766 struct ib_uobject *uobj;
768 if (out_len < sizeof(resp))
771 if (copy_from_user(&cmd, buf, sizeof(cmd)))
774 ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
775 u64_to_user_ptr(cmd.response) + sizeof(resp),
776 in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
777 out_len - sizeof(resp));
779 if (cmd.flags & ~IB_MR_REREG_SUPPORTED || !cmd.flags)
782 if ((cmd.flags & IB_MR_REREG_TRANS) &&
783 (!cmd.start || !cmd.hca_va || 0 >= cmd.length ||
784 (cmd.start & ~PAGE_MASK) != (cmd.hca_va & ~PAGE_MASK)))
787 uobj = uobj_get_write(UVERBS_OBJECT_MR, cmd.mr_handle, file);
789 return PTR_ERR(uobj);
798 if (cmd.flags & IB_MR_REREG_ACCESS) {
799 ret = ib_check_mr_access(cmd.access_flags);
804 if (cmd.flags & IB_MR_REREG_PD) {
805 pd = uobj_get_obj_read(pd, UVERBS_OBJECT_PD, cmd.pd_handle,
814 ret = mr->device->rereg_user_mr(mr, cmd.flags, cmd.start,
815 cmd.length, cmd.hca_va,
816 cmd.access_flags, pd, &udata);
818 if (cmd.flags & IB_MR_REREG_PD) {
819 atomic_inc(&pd->usecnt);
821 atomic_dec(&old_pd->usecnt);
827 memset(&resp, 0, sizeof(resp));
828 resp.lkey = mr->lkey;
829 resp.rkey = mr->rkey;
831 if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof(resp)))
837 if (cmd.flags & IB_MR_REREG_PD)
838 uobj_put_obj_read(pd);
841 uobj_put_write(uobj);
846 ssize_t ib_uverbs_dereg_mr(struct ib_uverbs_file *file,
847 const char __user *buf, int in_len,
850 struct ib_uverbs_dereg_mr cmd;
852 if (copy_from_user(&cmd, buf, sizeof cmd))
855 return uobj_perform_destroy(UVERBS_OBJECT_MR, cmd.mr_handle, file,
859 ssize_t ib_uverbs_alloc_mw(struct ib_uverbs_file *file,
860 const char __user *buf, int in_len,
863 struct ib_uverbs_alloc_mw cmd;
864 struct ib_uverbs_alloc_mw_resp resp;
865 struct ib_uobject *uobj;
868 struct ib_udata udata;
870 struct ib_device *ib_dev;
872 if (out_len < sizeof(resp))
875 if (copy_from_user(&cmd, buf, sizeof(cmd)))
878 uobj = uobj_alloc(UVERBS_OBJECT_MW, file, &ib_dev);
880 return PTR_ERR(uobj);
882 pd = uobj_get_obj_read(pd, UVERBS_OBJECT_PD, cmd.pd_handle, file);
888 ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
889 u64_to_user_ptr(cmd.response) + sizeof(resp),
890 in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
891 out_len - sizeof(resp));
893 mw = pd->device->alloc_mw(pd, cmd.mw_type, &udata);
899 mw->device = pd->device;
902 atomic_inc(&pd->usecnt);
906 memset(&resp, 0, sizeof(resp));
907 resp.rkey = mw->rkey;
908 resp.mw_handle = uobj->id;
910 if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof(resp))) {
915 uobj_put_obj_read(pd);
916 return uobj_alloc_commit(uobj, in_len);
919 uverbs_dealloc_mw(mw);
921 uobj_put_obj_read(pd);
923 uobj_alloc_abort(uobj);
927 ssize_t ib_uverbs_dealloc_mw(struct ib_uverbs_file *file,
928 const char __user *buf, int in_len,
931 struct ib_uverbs_dealloc_mw cmd;
933 if (copy_from_user(&cmd, buf, sizeof(cmd)))
936 return uobj_perform_destroy(UVERBS_OBJECT_MW, cmd.mw_handle, file,
940 ssize_t ib_uverbs_create_comp_channel(struct ib_uverbs_file *file,
941 const char __user *buf, int in_len,
944 struct ib_uverbs_create_comp_channel cmd;
945 struct ib_uverbs_create_comp_channel_resp resp;
946 struct ib_uobject *uobj;
947 struct ib_uverbs_completion_event_file *ev_file;
948 struct ib_device *ib_dev;
950 if (out_len < sizeof resp)
953 if (copy_from_user(&cmd, buf, sizeof cmd))
956 uobj = uobj_alloc(UVERBS_OBJECT_COMP_CHANNEL, file, &ib_dev);
958 return PTR_ERR(uobj);
962 ev_file = container_of(uobj, struct ib_uverbs_completion_event_file,
964 ib_uverbs_init_event_queue(&ev_file->ev_queue);
966 if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp)) {
967 uobj_alloc_abort(uobj);
971 return uobj_alloc_commit(uobj, in_len);
974 static struct ib_ucq_object *create_cq(struct ib_uverbs_file *file,
975 struct ib_udata *ucore,
976 struct ib_udata *uhw,
977 struct ib_uverbs_ex_create_cq *cmd,
979 int (*cb)(struct ib_uverbs_file *file,
980 struct ib_ucq_object *obj,
981 struct ib_uverbs_ex_create_cq_resp *resp,
982 struct ib_udata *udata,
986 struct ib_ucq_object *obj;
987 struct ib_uverbs_completion_event_file *ev_file = NULL;
990 struct ib_uverbs_ex_create_cq_resp resp;
991 struct ib_cq_init_attr attr = {};
992 struct ib_device *ib_dev;
994 if (cmd->comp_vector >= file->device->num_comp_vectors)
995 return ERR_PTR(-EINVAL);
997 obj = (struct ib_ucq_object *)uobj_alloc(UVERBS_OBJECT_CQ, file,
1002 if (!ib_dev->create_cq) {
1007 if (cmd->comp_channel >= 0) {
1008 ev_file = ib_uverbs_lookup_comp_file(cmd->comp_channel, file);
1009 if (IS_ERR(ev_file)) {
1010 ret = PTR_ERR(ev_file);
1015 obj->uobject.user_handle = cmd->user_handle;
1016 obj->comp_events_reported = 0;
1017 obj->async_events_reported = 0;
1018 INIT_LIST_HEAD(&obj->comp_list);
1019 INIT_LIST_HEAD(&obj->async_list);
1021 attr.cqe = cmd->cqe;
1022 attr.comp_vector = cmd->comp_vector;
1024 if (cmd_sz > offsetof(typeof(*cmd), flags) + sizeof(cmd->flags))
1025 attr.flags = cmd->flags;
1027 cq = ib_dev->create_cq(ib_dev, &attr, obj->uobject.context, uhw);
1033 cq->device = ib_dev;
1034 cq->uobject = &obj->uobject;
1035 cq->comp_handler = ib_uverbs_comp_handler;
1036 cq->event_handler = ib_uverbs_cq_event_handler;
1037 cq->cq_context = ev_file ? &ev_file->ev_queue : NULL;
1038 atomic_set(&cq->usecnt, 0);
1040 obj->uobject.object = cq;
1041 memset(&resp, 0, sizeof resp);
1042 resp.base.cq_handle = obj->uobject.id;
1043 resp.base.cqe = cq->cqe;
1045 resp.response_length = offsetof(typeof(resp), response_length) +
1046 sizeof(resp.response_length);
1048 cq->res.type = RDMA_RESTRACK_CQ;
1049 rdma_restrack_add(&cq->res);
1051 ret = cb(file, obj, &resp, ucore, context);
1055 ret = uobj_alloc_commit(&obj->uobject, 0);
1057 return ERR_PTR(ret);
1065 ib_uverbs_release_ucq(file, ev_file, obj);
1068 uobj_alloc_abort(&obj->uobject);
1070 return ERR_PTR(ret);
1073 static int ib_uverbs_create_cq_cb(struct ib_uverbs_file *file,
1074 struct ib_ucq_object *obj,
1075 struct ib_uverbs_ex_create_cq_resp *resp,
1076 struct ib_udata *ucore, void *context)
1078 if (ib_copy_to_udata(ucore, &resp->base, sizeof(resp->base)))
1084 ssize_t ib_uverbs_create_cq(struct ib_uverbs_file *file,
1085 const char __user *buf, int in_len,
1088 struct ib_uverbs_create_cq cmd;
1089 struct ib_uverbs_ex_create_cq cmd_ex;
1090 struct ib_uverbs_create_cq_resp resp;
1091 struct ib_udata ucore;
1092 struct ib_udata uhw;
1093 struct ib_ucq_object *obj;
1095 if (out_len < sizeof(resp))
1098 if (copy_from_user(&cmd, buf, sizeof(cmd)))
1101 ib_uverbs_init_udata(&ucore, buf, u64_to_user_ptr(cmd.response),
1102 sizeof(cmd), sizeof(resp));
1104 ib_uverbs_init_udata(&uhw, buf + sizeof(cmd),
1105 u64_to_user_ptr(cmd.response) + sizeof(resp),
1106 in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
1107 out_len - sizeof(resp));
1109 memset(&cmd_ex, 0, sizeof(cmd_ex));
1110 cmd_ex.user_handle = cmd.user_handle;
1111 cmd_ex.cqe = cmd.cqe;
1112 cmd_ex.comp_vector = cmd.comp_vector;
1113 cmd_ex.comp_channel = cmd.comp_channel;
1115 obj = create_cq(file, &ucore, &uhw, &cmd_ex,
1116 offsetof(typeof(cmd_ex), comp_channel) +
1117 sizeof(cmd.comp_channel), ib_uverbs_create_cq_cb,
1121 return PTR_ERR(obj);
1126 static int ib_uverbs_ex_create_cq_cb(struct ib_uverbs_file *file,
1127 struct ib_ucq_object *obj,
1128 struct ib_uverbs_ex_create_cq_resp *resp,
1129 struct ib_udata *ucore, void *context)
1131 if (ib_copy_to_udata(ucore, resp, resp->response_length))
1137 int ib_uverbs_ex_create_cq(struct ib_uverbs_file *file,
1138 struct ib_udata *ucore,
1139 struct ib_udata *uhw)
1141 struct ib_uverbs_ex_create_cq_resp resp;
1142 struct ib_uverbs_ex_create_cq cmd;
1143 struct ib_ucq_object *obj;
1146 if (ucore->inlen < sizeof(cmd))
1149 err = ib_copy_from_udata(&cmd, ucore, sizeof(cmd));
1159 if (ucore->outlen < (offsetof(typeof(resp), response_length) +
1160 sizeof(resp.response_length)))
1163 obj = create_cq(file, ucore, uhw, &cmd,
1164 min(ucore->inlen, sizeof(cmd)),
1165 ib_uverbs_ex_create_cq_cb, NULL);
1167 return PTR_ERR_OR_ZERO(obj);
1170 ssize_t ib_uverbs_resize_cq(struct ib_uverbs_file *file,
1171 const char __user *buf, int in_len,
1174 struct ib_uverbs_resize_cq cmd;
1175 struct ib_uverbs_resize_cq_resp resp = {};
1176 struct ib_udata udata;
1180 if (copy_from_user(&cmd, buf, sizeof cmd))
1183 ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
1184 u64_to_user_ptr(cmd.response) + sizeof(resp),
1185 in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
1186 out_len - sizeof(resp));
1188 cq = uobj_get_obj_read(cq, UVERBS_OBJECT_CQ, cmd.cq_handle, file);
1192 ret = cq->device->resize_cq(cq, cmd.cqe, &udata);
1198 if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp.cqe))
1202 uobj_put_obj_read(cq);
1204 return ret ? ret : in_len;
1207 static int copy_wc_to_user(struct ib_device *ib_dev, void __user *dest,
1210 struct ib_uverbs_wc tmp;
1212 tmp.wr_id = wc->wr_id;
1213 tmp.status = wc->status;
1214 tmp.opcode = wc->opcode;
1215 tmp.vendor_err = wc->vendor_err;
1216 tmp.byte_len = wc->byte_len;
1217 tmp.ex.imm_data = wc->ex.imm_data;
1218 tmp.qp_num = wc->qp->qp_num;
1219 tmp.src_qp = wc->src_qp;
1220 tmp.wc_flags = wc->wc_flags;
1221 tmp.pkey_index = wc->pkey_index;
1222 if (rdma_cap_opa_ah(ib_dev, wc->port_num))
1223 tmp.slid = OPA_TO_IB_UCAST_LID(wc->slid);
1225 tmp.slid = ib_lid_cpu16(wc->slid);
1227 tmp.dlid_path_bits = wc->dlid_path_bits;
1228 tmp.port_num = wc->port_num;
1231 if (copy_to_user(dest, &tmp, sizeof tmp))
1237 ssize_t ib_uverbs_poll_cq(struct ib_uverbs_file *file,
1238 const char __user *buf, int in_len,
1241 struct ib_uverbs_poll_cq cmd;
1242 struct ib_uverbs_poll_cq_resp resp;
1243 u8 __user *header_ptr;
1244 u8 __user *data_ptr;
1249 if (copy_from_user(&cmd, buf, sizeof cmd))
1252 cq = uobj_get_obj_read(cq, UVERBS_OBJECT_CQ, cmd.cq_handle, file);
1256 /* we copy a struct ib_uverbs_poll_cq_resp to user space */
1257 header_ptr = u64_to_user_ptr(cmd.response);
1258 data_ptr = header_ptr + sizeof resp;
1260 memset(&resp, 0, sizeof resp);
1261 while (resp.count < cmd.ne) {
1262 ret = ib_poll_cq(cq, 1, &wc);
1268 ret = copy_wc_to_user(cq->device, data_ptr, &wc);
1272 data_ptr += sizeof(struct ib_uverbs_wc);
1276 if (copy_to_user(header_ptr, &resp, sizeof resp)) {
1284 uobj_put_obj_read(cq);
1288 ssize_t ib_uverbs_req_notify_cq(struct ib_uverbs_file *file,
1289 const char __user *buf, int in_len,
1292 struct ib_uverbs_req_notify_cq cmd;
1295 if (copy_from_user(&cmd, buf, sizeof cmd))
1298 cq = uobj_get_obj_read(cq, UVERBS_OBJECT_CQ, cmd.cq_handle, file);
1302 ib_req_notify_cq(cq, cmd.solicited_only ?
1303 IB_CQ_SOLICITED : IB_CQ_NEXT_COMP);
1305 uobj_put_obj_read(cq);
1310 ssize_t ib_uverbs_destroy_cq(struct ib_uverbs_file *file,
1311 const char __user *buf, int in_len,
1314 struct ib_uverbs_destroy_cq cmd;
1315 struct ib_uverbs_destroy_cq_resp resp;
1316 struct ib_uobject *uobj;
1317 struct ib_ucq_object *obj;
1319 if (copy_from_user(&cmd, buf, sizeof cmd))
1322 uobj = uobj_get_destroy(UVERBS_OBJECT_CQ, cmd.cq_handle, file);
1324 return PTR_ERR(uobj);
1326 obj = container_of(uobj, struct ib_ucq_object, uobject);
1327 memset(&resp, 0, sizeof(resp));
1328 resp.comp_events_reported = obj->comp_events_reported;
1329 resp.async_events_reported = obj->async_events_reported;
1331 uobj_put_destroy(uobj);
1333 if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp))
1339 static int create_qp(struct ib_uverbs_file *file,
1340 struct ib_udata *ucore,
1341 struct ib_udata *uhw,
1342 struct ib_uverbs_ex_create_qp *cmd,
1344 int (*cb)(struct ib_uverbs_file *file,
1345 struct ib_uverbs_ex_create_qp_resp *resp,
1346 struct ib_udata *udata),
1349 struct ib_uqp_object *obj;
1350 struct ib_device *device;
1351 struct ib_pd *pd = NULL;
1352 struct ib_xrcd *xrcd = NULL;
1353 struct ib_uobject *xrcd_uobj = ERR_PTR(-ENOENT);
1354 struct ib_cq *scq = NULL, *rcq = NULL;
1355 struct ib_srq *srq = NULL;
1358 struct ib_qp_init_attr attr = {};
1359 struct ib_uverbs_ex_create_qp_resp resp;
1361 struct ib_rwq_ind_table *ind_tbl = NULL;
1363 struct ib_device *ib_dev;
1365 if (cmd->qp_type == IB_QPT_RAW_PACKET && !capable(CAP_NET_RAW))
1368 obj = (struct ib_uqp_object *)uobj_alloc(UVERBS_OBJECT_QP, file,
1371 return PTR_ERR(obj);
1373 obj->uevent.uobject.user_handle = cmd->user_handle;
1374 mutex_init(&obj->mcast_lock);
1376 if (cmd_sz >= offsetof(typeof(*cmd), rwq_ind_tbl_handle) +
1377 sizeof(cmd->rwq_ind_tbl_handle) &&
1378 (cmd->comp_mask & IB_UVERBS_CREATE_QP_MASK_IND_TABLE)) {
1379 ind_tbl = uobj_get_obj_read(rwq_ind_table,
1380 UVERBS_OBJECT_RWQ_IND_TBL,
1381 cmd->rwq_ind_tbl_handle, file);
1387 attr.rwq_ind_tbl = ind_tbl;
1390 if (cmd_sz > sizeof(*cmd) &&
1391 !ib_is_udata_cleared(ucore, sizeof(*cmd),
1392 cmd_sz - sizeof(*cmd))) {
1397 if (ind_tbl && (cmd->max_recv_wr || cmd->max_recv_sge || cmd->is_srq)) {
1402 if (ind_tbl && !cmd->max_send_wr)
1405 if (cmd->qp_type == IB_QPT_XRC_TGT) {
1406 xrcd_uobj = uobj_get_read(UVERBS_OBJECT_XRCD, cmd->pd_handle,
1409 if (IS_ERR(xrcd_uobj)) {
1414 xrcd = (struct ib_xrcd *)xrcd_uobj->object;
1419 device = xrcd->device;
1421 if (cmd->qp_type == IB_QPT_XRC_INI) {
1422 cmd->max_recv_wr = 0;
1423 cmd->max_recv_sge = 0;
1426 srq = uobj_get_obj_read(srq, UVERBS_OBJECT_SRQ,
1427 cmd->srq_handle, file);
1428 if (!srq || srq->srq_type == IB_SRQT_XRC) {
1435 if (cmd->recv_cq_handle != cmd->send_cq_handle) {
1436 rcq = uobj_get_obj_read(
1437 cq, UVERBS_OBJECT_CQ,
1438 cmd->recv_cq_handle, file);
1448 scq = uobj_get_obj_read(cq, UVERBS_OBJECT_CQ,
1449 cmd->send_cq_handle, file);
1452 pd = uobj_get_obj_read(pd, UVERBS_OBJECT_PD, cmd->pd_handle,
1454 if (!pd || (!scq && has_sq)) {
1459 device = pd->device;
1462 attr.event_handler = ib_uverbs_qp_event_handler;
1463 attr.qp_context = file;
1468 attr.sq_sig_type = cmd->sq_sig_all ? IB_SIGNAL_ALL_WR :
1470 attr.qp_type = cmd->qp_type;
1471 attr.create_flags = 0;
1473 attr.cap.max_send_wr = cmd->max_send_wr;
1474 attr.cap.max_recv_wr = cmd->max_recv_wr;
1475 attr.cap.max_send_sge = cmd->max_send_sge;
1476 attr.cap.max_recv_sge = cmd->max_recv_sge;
1477 attr.cap.max_inline_data = cmd->max_inline_data;
1479 obj->uevent.events_reported = 0;
1480 INIT_LIST_HEAD(&obj->uevent.event_list);
1481 INIT_LIST_HEAD(&obj->mcast_list);
1483 if (cmd_sz >= offsetof(typeof(*cmd), create_flags) +
1484 sizeof(cmd->create_flags))
1485 attr.create_flags = cmd->create_flags;
1487 if (attr.create_flags & ~(IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK |
1488 IB_QP_CREATE_CROSS_CHANNEL |
1489 IB_QP_CREATE_MANAGED_SEND |
1490 IB_QP_CREATE_MANAGED_RECV |
1491 IB_QP_CREATE_SCATTER_FCS |
1492 IB_QP_CREATE_CVLAN_STRIPPING |
1493 IB_QP_CREATE_SOURCE_QPN |
1494 IB_QP_CREATE_PCI_WRITE_END_PADDING)) {
1499 if (attr.create_flags & IB_QP_CREATE_SOURCE_QPN) {
1500 if (!capable(CAP_NET_RAW)) {
1505 attr.source_qpn = cmd->source_qpn;
1508 buf = (void *)cmd + sizeof(*cmd);
1509 if (cmd_sz > sizeof(*cmd))
1510 if (!(buf[0] == 0 && !memcmp(buf, buf + 1,
1511 cmd_sz - sizeof(*cmd) - 1))) {
1516 if (cmd->qp_type == IB_QPT_XRC_TGT)
1517 qp = ib_create_qp(pd, &attr);
1519 qp = _ib_create_qp(device, pd, &attr, uhw,
1520 &obj->uevent.uobject);
1527 if (cmd->qp_type != IB_QPT_XRC_TGT) {
1528 ret = ib_create_qp_security(qp, device);
1534 qp->send_cq = attr.send_cq;
1535 qp->recv_cq = attr.recv_cq;
1537 qp->rwq_ind_tbl = ind_tbl;
1538 qp->event_handler = attr.event_handler;
1539 qp->qp_context = attr.qp_context;
1540 qp->qp_type = attr.qp_type;
1541 atomic_set(&qp->usecnt, 0);
1542 atomic_inc(&pd->usecnt);
1545 atomic_inc(&attr.send_cq->usecnt);
1547 atomic_inc(&attr.recv_cq->usecnt);
1549 atomic_inc(&attr.srq->usecnt);
1551 atomic_inc(&ind_tbl->usecnt);
1553 /* It is done in _ib_create_qp for other QP types */
1554 qp->uobject = &obj->uevent.uobject;
1557 obj->uevent.uobject.object = qp;
1559 memset(&resp, 0, sizeof resp);
1560 resp.base.qpn = qp->qp_num;
1561 resp.base.qp_handle = obj->uevent.uobject.id;
1562 resp.base.max_recv_sge = attr.cap.max_recv_sge;
1563 resp.base.max_send_sge = attr.cap.max_send_sge;
1564 resp.base.max_recv_wr = attr.cap.max_recv_wr;
1565 resp.base.max_send_wr = attr.cap.max_send_wr;
1566 resp.base.max_inline_data = attr.cap.max_inline_data;
1568 resp.response_length = offsetof(typeof(resp), response_length) +
1569 sizeof(resp.response_length);
1571 ret = cb(file, &resp, ucore);
1576 obj->uxrcd = container_of(xrcd_uobj, struct ib_uxrcd_object,
1578 atomic_inc(&obj->uxrcd->refcnt);
1579 uobj_put_read(xrcd_uobj);
1583 uobj_put_obj_read(pd);
1585 uobj_put_obj_read(scq);
1586 if (rcq && rcq != scq)
1587 uobj_put_obj_read(rcq);
1589 uobj_put_obj_read(srq);
1591 uobj_put_obj_read(ind_tbl);
1593 return uobj_alloc_commit(&obj->uevent.uobject, 0);
1598 if (!IS_ERR(xrcd_uobj))
1599 uobj_put_read(xrcd_uobj);
1601 uobj_put_obj_read(pd);
1603 uobj_put_obj_read(scq);
1604 if (rcq && rcq != scq)
1605 uobj_put_obj_read(rcq);
1607 uobj_put_obj_read(srq);
1609 uobj_put_obj_read(ind_tbl);
1611 uobj_alloc_abort(&obj->uevent.uobject);
1615 static int ib_uverbs_create_qp_cb(struct ib_uverbs_file *file,
1616 struct ib_uverbs_ex_create_qp_resp *resp,
1617 struct ib_udata *ucore)
1619 if (ib_copy_to_udata(ucore, &resp->base, sizeof(resp->base)))
1625 ssize_t ib_uverbs_create_qp(struct ib_uverbs_file *file,
1626 const char __user *buf, int in_len,
1629 struct ib_uverbs_create_qp cmd;
1630 struct ib_uverbs_ex_create_qp cmd_ex;
1631 struct ib_udata ucore;
1632 struct ib_udata uhw;
1633 ssize_t resp_size = sizeof(struct ib_uverbs_create_qp_resp);
1636 if (out_len < resp_size)
1639 if (copy_from_user(&cmd, buf, sizeof(cmd)))
1642 ib_uverbs_init_udata(&ucore, buf, u64_to_user_ptr(cmd.response),
1643 sizeof(cmd), resp_size);
1644 ib_uverbs_init_udata(&uhw, buf + sizeof(cmd),
1645 u64_to_user_ptr(cmd.response) + resp_size,
1646 in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
1647 out_len - resp_size);
1649 memset(&cmd_ex, 0, sizeof(cmd_ex));
1650 cmd_ex.user_handle = cmd.user_handle;
1651 cmd_ex.pd_handle = cmd.pd_handle;
1652 cmd_ex.send_cq_handle = cmd.send_cq_handle;
1653 cmd_ex.recv_cq_handle = cmd.recv_cq_handle;
1654 cmd_ex.srq_handle = cmd.srq_handle;
1655 cmd_ex.max_send_wr = cmd.max_send_wr;
1656 cmd_ex.max_recv_wr = cmd.max_recv_wr;
1657 cmd_ex.max_send_sge = cmd.max_send_sge;
1658 cmd_ex.max_recv_sge = cmd.max_recv_sge;
1659 cmd_ex.max_inline_data = cmd.max_inline_data;
1660 cmd_ex.sq_sig_all = cmd.sq_sig_all;
1661 cmd_ex.qp_type = cmd.qp_type;
1662 cmd_ex.is_srq = cmd.is_srq;
1664 err = create_qp(file, &ucore, &uhw, &cmd_ex,
1665 offsetof(typeof(cmd_ex), is_srq) +
1666 sizeof(cmd.is_srq), ib_uverbs_create_qp_cb,
1675 static int ib_uverbs_ex_create_qp_cb(struct ib_uverbs_file *file,
1676 struct ib_uverbs_ex_create_qp_resp *resp,
1677 struct ib_udata *ucore)
1679 if (ib_copy_to_udata(ucore, resp, resp->response_length))
1685 int ib_uverbs_ex_create_qp(struct ib_uverbs_file *file,
1686 struct ib_udata *ucore,
1687 struct ib_udata *uhw)
1689 struct ib_uverbs_ex_create_qp_resp resp;
1690 struct ib_uverbs_ex_create_qp cmd = {0};
1693 if (ucore->inlen < (offsetof(typeof(cmd), comp_mask) +
1694 sizeof(cmd.comp_mask)))
1697 err = ib_copy_from_udata(&cmd, ucore, min(sizeof(cmd), ucore->inlen));
1701 if (cmd.comp_mask & ~IB_UVERBS_CREATE_QP_SUP_COMP_MASK)
1707 if (ucore->outlen < (offsetof(typeof(resp), response_length) +
1708 sizeof(resp.response_length)))
1711 err = create_qp(file, ucore, uhw, &cmd,
1712 min(ucore->inlen, sizeof(cmd)),
1713 ib_uverbs_ex_create_qp_cb, NULL);
1721 ssize_t ib_uverbs_open_qp(struct ib_uverbs_file *file,
1722 const char __user *buf, int in_len, int out_len)
1724 struct ib_uverbs_open_qp cmd;
1725 struct ib_uverbs_create_qp_resp resp;
1726 struct ib_udata udata;
1727 struct ib_uqp_object *obj;
1728 struct ib_xrcd *xrcd;
1729 struct ib_uobject *uninitialized_var(xrcd_uobj);
1731 struct ib_qp_open_attr attr;
1733 struct ib_device *ib_dev;
1735 if (out_len < sizeof resp)
1738 if (copy_from_user(&cmd, buf, sizeof cmd))
1741 ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
1742 u64_to_user_ptr(cmd.response) + sizeof(resp),
1743 in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
1744 out_len - sizeof(resp));
1746 obj = (struct ib_uqp_object *)uobj_alloc(UVERBS_OBJECT_QP, file,
1749 return PTR_ERR(obj);
1751 xrcd_uobj = uobj_get_read(UVERBS_OBJECT_XRCD, cmd.pd_handle, file);
1752 if (IS_ERR(xrcd_uobj)) {
1757 xrcd = (struct ib_xrcd *)xrcd_uobj->object;
1763 attr.event_handler = ib_uverbs_qp_event_handler;
1764 attr.qp_context = file;
1765 attr.qp_num = cmd.qpn;
1766 attr.qp_type = cmd.qp_type;
1768 obj->uevent.events_reported = 0;
1769 INIT_LIST_HEAD(&obj->uevent.event_list);
1770 INIT_LIST_HEAD(&obj->mcast_list);
1772 qp = ib_open_qp(xrcd, &attr);
1778 obj->uevent.uobject.object = qp;
1779 obj->uevent.uobject.user_handle = cmd.user_handle;
1781 memset(&resp, 0, sizeof resp);
1782 resp.qpn = qp->qp_num;
1783 resp.qp_handle = obj->uevent.uobject.id;
1785 if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp)) {
1790 obj->uxrcd = container_of(xrcd_uobj, struct ib_uxrcd_object, uobject);
1791 atomic_inc(&obj->uxrcd->refcnt);
1792 qp->uobject = &obj->uevent.uobject;
1793 uobj_put_read(xrcd_uobj);
1795 return uobj_alloc_commit(&obj->uevent.uobject, in_len);
1800 uobj_put_read(xrcd_uobj);
1802 uobj_alloc_abort(&obj->uevent.uobject);
1806 static void copy_ah_attr_to_uverbs(struct ib_uverbs_qp_dest *uverb_attr,
1807 struct rdma_ah_attr *rdma_attr)
1809 const struct ib_global_route *grh;
1811 uverb_attr->dlid = rdma_ah_get_dlid(rdma_attr);
1812 uverb_attr->sl = rdma_ah_get_sl(rdma_attr);
1813 uverb_attr->src_path_bits = rdma_ah_get_path_bits(rdma_attr);
1814 uverb_attr->static_rate = rdma_ah_get_static_rate(rdma_attr);
1815 uverb_attr->is_global = !!(rdma_ah_get_ah_flags(rdma_attr) &
1817 if (uverb_attr->is_global) {
1818 grh = rdma_ah_read_grh(rdma_attr);
1819 memcpy(uverb_attr->dgid, grh->dgid.raw, 16);
1820 uverb_attr->flow_label = grh->flow_label;
1821 uverb_attr->sgid_index = grh->sgid_index;
1822 uverb_attr->hop_limit = grh->hop_limit;
1823 uverb_attr->traffic_class = grh->traffic_class;
1825 uverb_attr->port_num = rdma_ah_get_port_num(rdma_attr);
1828 ssize_t ib_uverbs_query_qp(struct ib_uverbs_file *file,
1829 const char __user *buf, int in_len,
1832 struct ib_uverbs_query_qp cmd;
1833 struct ib_uverbs_query_qp_resp resp;
1835 struct ib_qp_attr *attr;
1836 struct ib_qp_init_attr *init_attr;
1839 if (copy_from_user(&cmd, buf, sizeof cmd))
1842 attr = kmalloc(sizeof *attr, GFP_KERNEL);
1843 init_attr = kmalloc(sizeof *init_attr, GFP_KERNEL);
1844 if (!attr || !init_attr) {
1849 qp = uobj_get_obj_read(qp, UVERBS_OBJECT_QP, cmd.qp_handle, file);
1855 ret = ib_query_qp(qp, attr, cmd.attr_mask, init_attr);
1857 uobj_put_obj_read(qp);
1862 memset(&resp, 0, sizeof resp);
1864 resp.qp_state = attr->qp_state;
1865 resp.cur_qp_state = attr->cur_qp_state;
1866 resp.path_mtu = attr->path_mtu;
1867 resp.path_mig_state = attr->path_mig_state;
1868 resp.qkey = attr->qkey;
1869 resp.rq_psn = attr->rq_psn;
1870 resp.sq_psn = attr->sq_psn;
1871 resp.dest_qp_num = attr->dest_qp_num;
1872 resp.qp_access_flags = attr->qp_access_flags;
1873 resp.pkey_index = attr->pkey_index;
1874 resp.alt_pkey_index = attr->alt_pkey_index;
1875 resp.sq_draining = attr->sq_draining;
1876 resp.max_rd_atomic = attr->max_rd_atomic;
1877 resp.max_dest_rd_atomic = attr->max_dest_rd_atomic;
1878 resp.min_rnr_timer = attr->min_rnr_timer;
1879 resp.port_num = attr->port_num;
1880 resp.timeout = attr->timeout;
1881 resp.retry_cnt = attr->retry_cnt;
1882 resp.rnr_retry = attr->rnr_retry;
1883 resp.alt_port_num = attr->alt_port_num;
1884 resp.alt_timeout = attr->alt_timeout;
1886 copy_ah_attr_to_uverbs(&resp.dest, &attr->ah_attr);
1887 copy_ah_attr_to_uverbs(&resp.alt_dest, &attr->alt_ah_attr);
1889 resp.max_send_wr = init_attr->cap.max_send_wr;
1890 resp.max_recv_wr = init_attr->cap.max_recv_wr;
1891 resp.max_send_sge = init_attr->cap.max_send_sge;
1892 resp.max_recv_sge = init_attr->cap.max_recv_sge;
1893 resp.max_inline_data = init_attr->cap.max_inline_data;
1894 resp.sq_sig_all = init_attr->sq_sig_type == IB_SIGNAL_ALL_WR;
1896 if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp))
1903 return ret ? ret : in_len;
1906 /* Remove ignored fields set in the attribute mask */
1907 static int modify_qp_mask(enum ib_qp_type qp_type, int mask)
1910 case IB_QPT_XRC_INI:
1911 return mask & ~(IB_QP_MAX_DEST_RD_ATOMIC | IB_QP_MIN_RNR_TIMER);
1912 case IB_QPT_XRC_TGT:
1913 return mask & ~(IB_QP_MAX_QP_RD_ATOMIC | IB_QP_RETRY_CNT |
1920 static void copy_ah_attr_from_uverbs(struct ib_device *dev,
1921 struct rdma_ah_attr *rdma_attr,
1922 struct ib_uverbs_qp_dest *uverb_attr)
1924 rdma_attr->type = rdma_ah_find_type(dev, uverb_attr->port_num);
1925 if (uverb_attr->is_global) {
1926 rdma_ah_set_grh(rdma_attr, NULL,
1927 uverb_attr->flow_label,
1928 uverb_attr->sgid_index,
1929 uverb_attr->hop_limit,
1930 uverb_attr->traffic_class);
1931 rdma_ah_set_dgid_raw(rdma_attr, uverb_attr->dgid);
1933 rdma_ah_set_ah_flags(rdma_attr, 0);
1935 rdma_ah_set_dlid(rdma_attr, uverb_attr->dlid);
1936 rdma_ah_set_sl(rdma_attr, uverb_attr->sl);
1937 rdma_ah_set_path_bits(rdma_attr, uverb_attr->src_path_bits);
1938 rdma_ah_set_static_rate(rdma_attr, uverb_attr->static_rate);
1939 rdma_ah_set_port_num(rdma_attr, uverb_attr->port_num);
1940 rdma_ah_set_make_grd(rdma_attr, false);
1943 static int modify_qp(struct ib_uverbs_file *file,
1944 struct ib_uverbs_ex_modify_qp *cmd, struct ib_udata *udata)
1946 struct ib_qp_attr *attr;
1950 attr = kzalloc(sizeof(*attr), GFP_KERNEL);
1954 qp = uobj_get_obj_read(qp, UVERBS_OBJECT_QP, cmd->base.qp_handle, file);
1960 if ((cmd->base.attr_mask & IB_QP_PORT) &&
1961 !rdma_is_port_valid(qp->device, cmd->base.port_num)) {
1966 if ((cmd->base.attr_mask & IB_QP_AV) &&
1967 !rdma_is_port_valid(qp->device, cmd->base.dest.port_num)) {
1972 if ((cmd->base.attr_mask & IB_QP_ALT_PATH) &&
1973 (!rdma_is_port_valid(qp->device, cmd->base.alt_port_num) ||
1974 !rdma_is_port_valid(qp->device, cmd->base.alt_dest.port_num))) {
1979 if ((cmd->base.attr_mask & IB_QP_CUR_STATE &&
1980 cmd->base.cur_qp_state > IB_QPS_ERR) ||
1981 cmd->base.qp_state > IB_QPS_ERR) {
1986 attr->qp_state = cmd->base.qp_state;
1987 attr->cur_qp_state = cmd->base.cur_qp_state;
1988 attr->path_mtu = cmd->base.path_mtu;
1989 attr->path_mig_state = cmd->base.path_mig_state;
1990 attr->qkey = cmd->base.qkey;
1991 attr->rq_psn = cmd->base.rq_psn;
1992 attr->sq_psn = cmd->base.sq_psn;
1993 attr->dest_qp_num = cmd->base.dest_qp_num;
1994 attr->qp_access_flags = cmd->base.qp_access_flags;
1995 attr->pkey_index = cmd->base.pkey_index;
1996 attr->alt_pkey_index = cmd->base.alt_pkey_index;
1997 attr->en_sqd_async_notify = cmd->base.en_sqd_async_notify;
1998 attr->max_rd_atomic = cmd->base.max_rd_atomic;
1999 attr->max_dest_rd_atomic = cmd->base.max_dest_rd_atomic;
2000 attr->min_rnr_timer = cmd->base.min_rnr_timer;
2001 attr->port_num = cmd->base.port_num;
2002 attr->timeout = cmd->base.timeout;
2003 attr->retry_cnt = cmd->base.retry_cnt;
2004 attr->rnr_retry = cmd->base.rnr_retry;
2005 attr->alt_port_num = cmd->base.alt_port_num;
2006 attr->alt_timeout = cmd->base.alt_timeout;
2007 attr->rate_limit = cmd->rate_limit;
2009 if (cmd->base.attr_mask & IB_QP_AV)
2010 copy_ah_attr_from_uverbs(qp->device, &attr->ah_attr,
2013 if (cmd->base.attr_mask & IB_QP_ALT_PATH)
2014 copy_ah_attr_from_uverbs(qp->device, &attr->alt_ah_attr,
2015 &cmd->base.alt_dest);
2017 ret = ib_modify_qp_with_udata(qp, attr,
2018 modify_qp_mask(qp->qp_type,
2019 cmd->base.attr_mask),
2023 uobj_put_obj_read(qp);
2030 ssize_t ib_uverbs_modify_qp(struct ib_uverbs_file *file,
2031 const char __user *buf, int in_len,
2034 struct ib_uverbs_ex_modify_qp cmd = {};
2035 struct ib_udata udata;
2038 if (copy_from_user(&cmd.base, buf, sizeof(cmd.base)))
2041 if (cmd.base.attr_mask &
2042 ~((IB_USER_LEGACY_LAST_QP_ATTR_MASK << 1) - 1))
2045 ib_uverbs_init_udata(&udata, buf + sizeof(cmd.base), NULL,
2046 in_len - sizeof(cmd.base) - sizeof(struct ib_uverbs_cmd_hdr),
2049 ret = modify_qp(file, &cmd, &udata);
2056 int ib_uverbs_ex_modify_qp(struct ib_uverbs_file *file,
2057 struct ib_udata *ucore,
2058 struct ib_udata *uhw)
2060 struct ib_uverbs_ex_modify_qp cmd = {};
2064 * Last bit is reserved for extending the attr_mask by
2065 * using another field.
2067 BUILD_BUG_ON(IB_USER_LAST_QP_ATTR_MASK == (1 << 31));
2069 if (ucore->inlen < sizeof(cmd.base))
2072 ret = ib_copy_from_udata(&cmd, ucore, min(sizeof(cmd), ucore->inlen));
2076 if (cmd.base.attr_mask &
2077 ~((IB_USER_LAST_QP_ATTR_MASK << 1) - 1))
2080 if (ucore->inlen > sizeof(cmd)) {
2081 if (!ib_is_udata_cleared(ucore, sizeof(cmd),
2082 ucore->inlen - sizeof(cmd)))
2086 ret = modify_qp(file, &cmd, uhw);
2091 ssize_t ib_uverbs_destroy_qp(struct ib_uverbs_file *file,
2092 const char __user *buf, int in_len,
2095 struct ib_uverbs_destroy_qp cmd;
2096 struct ib_uverbs_destroy_qp_resp resp;
2097 struct ib_uobject *uobj;
2098 struct ib_uqp_object *obj;
2100 if (copy_from_user(&cmd, buf, sizeof cmd))
2103 uobj = uobj_get_destroy(UVERBS_OBJECT_QP, cmd.qp_handle, file);
2105 return PTR_ERR(uobj);
2107 obj = container_of(uobj, struct ib_uqp_object, uevent.uobject);
2108 memset(&resp, 0, sizeof(resp));
2109 resp.events_reported = obj->uevent.events_reported;
2111 uobj_put_destroy(uobj);
2113 if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp))
2119 static void *alloc_wr(size_t wr_size, __u32 num_sge)
2121 if (num_sge >= (U32_MAX - ALIGN(wr_size, sizeof (struct ib_sge))) /
2122 sizeof (struct ib_sge))
2125 return kmalloc(ALIGN(wr_size, sizeof (struct ib_sge)) +
2126 num_sge * sizeof (struct ib_sge), GFP_KERNEL);
2129 ssize_t ib_uverbs_post_send(struct ib_uverbs_file *file,
2130 const char __user *buf, int in_len,
2133 struct ib_uverbs_post_send cmd;
2134 struct ib_uverbs_post_send_resp resp;
2135 struct ib_uverbs_send_wr *user_wr;
2136 struct ib_send_wr *wr = NULL, *last, *next;
2137 const struct ib_send_wr *bad_wr;
2141 ssize_t ret = -EINVAL;
2144 if (copy_from_user(&cmd, buf, sizeof cmd))
2147 if (in_len < sizeof cmd + cmd.wqe_size * cmd.wr_count +
2148 cmd.sge_count * sizeof (struct ib_uverbs_sge))
2151 if (cmd.wqe_size < sizeof (struct ib_uverbs_send_wr))
2154 user_wr = kmalloc(cmd.wqe_size, GFP_KERNEL);
2158 qp = uobj_get_obj_read(qp, UVERBS_OBJECT_QP, cmd.qp_handle, file);
2162 is_ud = qp->qp_type == IB_QPT_UD;
2165 for (i = 0; i < cmd.wr_count; ++i) {
2166 if (copy_from_user(user_wr,
2167 buf + sizeof cmd + i * cmd.wqe_size,
2173 if (user_wr->num_sge + sg_ind > cmd.sge_count) {
2179 struct ib_ud_wr *ud;
2181 if (user_wr->opcode != IB_WR_SEND &&
2182 user_wr->opcode != IB_WR_SEND_WITH_IMM) {
2187 next_size = sizeof(*ud);
2188 ud = alloc_wr(next_size, user_wr->num_sge);
2194 ud->ah = uobj_get_obj_read(ah, UVERBS_OBJECT_AH,
2195 user_wr->wr.ud.ah, file);
2201 ud->remote_qpn = user_wr->wr.ud.remote_qpn;
2202 ud->remote_qkey = user_wr->wr.ud.remote_qkey;
2205 } else if (user_wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM ||
2206 user_wr->opcode == IB_WR_RDMA_WRITE ||
2207 user_wr->opcode == IB_WR_RDMA_READ) {
2208 struct ib_rdma_wr *rdma;
2210 next_size = sizeof(*rdma);
2211 rdma = alloc_wr(next_size, user_wr->num_sge);
2217 rdma->remote_addr = user_wr->wr.rdma.remote_addr;
2218 rdma->rkey = user_wr->wr.rdma.rkey;
2221 } else if (user_wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
2222 user_wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD) {
2223 struct ib_atomic_wr *atomic;
2225 next_size = sizeof(*atomic);
2226 atomic = alloc_wr(next_size, user_wr->num_sge);
2232 atomic->remote_addr = user_wr->wr.atomic.remote_addr;
2233 atomic->compare_add = user_wr->wr.atomic.compare_add;
2234 atomic->swap = user_wr->wr.atomic.swap;
2235 atomic->rkey = user_wr->wr.atomic.rkey;
2238 } else if (user_wr->opcode == IB_WR_SEND ||
2239 user_wr->opcode == IB_WR_SEND_WITH_IMM ||
2240 user_wr->opcode == IB_WR_SEND_WITH_INV) {
2241 next_size = sizeof(*next);
2242 next = alloc_wr(next_size, user_wr->num_sge);
2252 if (user_wr->opcode == IB_WR_SEND_WITH_IMM ||
2253 user_wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM) {
2255 (__be32 __force) user_wr->ex.imm_data;
2256 } else if (user_wr->opcode == IB_WR_SEND_WITH_INV) {
2257 next->ex.invalidate_rkey = user_wr->ex.invalidate_rkey;
2267 next->wr_id = user_wr->wr_id;
2268 next->num_sge = user_wr->num_sge;
2269 next->opcode = user_wr->opcode;
2270 next->send_flags = user_wr->send_flags;
2272 if (next->num_sge) {
2273 next->sg_list = (void *) next +
2274 ALIGN(next_size, sizeof(struct ib_sge));
2275 if (copy_from_user(next->sg_list,
2277 cmd.wr_count * cmd.wqe_size +
2278 sg_ind * sizeof (struct ib_sge),
2279 next->num_sge * sizeof (struct ib_sge))) {
2283 sg_ind += next->num_sge;
2285 next->sg_list = NULL;
2289 ret = qp->device->post_send(qp->real_qp, wr, &bad_wr);
2291 for (next = wr; next; next = next->next) {
2297 if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp))
2301 uobj_put_obj_read(qp);
2304 if (is_ud && ud_wr(wr)->ah)
2305 uobj_put_obj_read(ud_wr(wr)->ah);
2314 return ret ? ret : in_len;
2317 static struct ib_recv_wr *ib_uverbs_unmarshall_recv(const char __user *buf,
2323 struct ib_uverbs_recv_wr *user_wr;
2324 struct ib_recv_wr *wr = NULL, *last, *next;
2329 if (in_len < wqe_size * wr_count +
2330 sge_count * sizeof (struct ib_uverbs_sge))
2331 return ERR_PTR(-EINVAL);
2333 if (wqe_size < sizeof (struct ib_uverbs_recv_wr))
2334 return ERR_PTR(-EINVAL);
2336 user_wr = kmalloc(wqe_size, GFP_KERNEL);
2338 return ERR_PTR(-ENOMEM);
2342 for (i = 0; i < wr_count; ++i) {
2343 if (copy_from_user(user_wr, buf + i * wqe_size,
2349 if (user_wr->num_sge + sg_ind > sge_count) {
2354 if (user_wr->num_sge >=
2355 (U32_MAX - ALIGN(sizeof *next, sizeof (struct ib_sge))) /
2356 sizeof (struct ib_sge)) {
2361 next = kmalloc(ALIGN(sizeof *next, sizeof (struct ib_sge)) +
2362 user_wr->num_sge * sizeof (struct ib_sge),
2376 next->wr_id = user_wr->wr_id;
2377 next->num_sge = user_wr->num_sge;
2379 if (next->num_sge) {
2380 next->sg_list = (void *) next +
2381 ALIGN(sizeof *next, sizeof (struct ib_sge));
2382 if (copy_from_user(next->sg_list,
2383 buf + wr_count * wqe_size +
2384 sg_ind * sizeof (struct ib_sge),
2385 next->num_sge * sizeof (struct ib_sge))) {
2389 sg_ind += next->num_sge;
2391 next->sg_list = NULL;
2406 return ERR_PTR(ret);
2409 ssize_t ib_uverbs_post_recv(struct ib_uverbs_file *file,
2410 const char __user *buf, int in_len,
2413 struct ib_uverbs_post_recv cmd;
2414 struct ib_uverbs_post_recv_resp resp;
2415 struct ib_recv_wr *wr, *next;
2416 const struct ib_recv_wr *bad_wr;
2418 ssize_t ret = -EINVAL;
2420 if (copy_from_user(&cmd, buf, sizeof cmd))
2423 wr = ib_uverbs_unmarshall_recv(buf + sizeof cmd,
2424 in_len - sizeof cmd, cmd.wr_count,
2425 cmd.sge_count, cmd.wqe_size);
2429 qp = uobj_get_obj_read(qp, UVERBS_OBJECT_QP, cmd.qp_handle, file);
2434 ret = qp->device->post_recv(qp->real_qp, wr, &bad_wr);
2436 uobj_put_obj_read(qp);
2438 for (next = wr; next; next = next->next) {
2445 if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp))
2455 return ret ? ret : in_len;
2458 ssize_t ib_uverbs_post_srq_recv(struct ib_uverbs_file *file,
2459 const char __user *buf, int in_len,
2462 struct ib_uverbs_post_srq_recv cmd;
2463 struct ib_uverbs_post_srq_recv_resp resp;
2464 struct ib_recv_wr *wr, *next;
2465 const struct ib_recv_wr *bad_wr;
2467 ssize_t ret = -EINVAL;
2469 if (copy_from_user(&cmd, buf, sizeof cmd))
2472 wr = ib_uverbs_unmarshall_recv(buf + sizeof cmd,
2473 in_len - sizeof cmd, cmd.wr_count,
2474 cmd.sge_count, cmd.wqe_size);
2478 srq = uobj_get_obj_read(srq, UVERBS_OBJECT_SRQ, cmd.srq_handle, file);
2483 ret = srq->device->post_srq_recv ?
2484 srq->device->post_srq_recv(srq, wr, &bad_wr) : -EOPNOTSUPP;
2486 uobj_put_obj_read(srq);
2489 for (next = wr; next; next = next->next) {
2495 if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp))
2505 return ret ? ret : in_len;
2508 ssize_t ib_uverbs_create_ah(struct ib_uverbs_file *file,
2509 const char __user *buf, int in_len,
2512 struct ib_uverbs_create_ah cmd;
2513 struct ib_uverbs_create_ah_resp resp;
2514 struct ib_uobject *uobj;
2517 struct rdma_ah_attr attr = {};
2519 struct ib_udata udata;
2520 struct ib_device *ib_dev;
2522 if (out_len < sizeof resp)
2525 if (copy_from_user(&cmd, buf, sizeof cmd))
2528 ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
2529 u64_to_user_ptr(cmd.response) + sizeof(resp),
2530 in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
2531 out_len - sizeof(resp));
2533 uobj = uobj_alloc(UVERBS_OBJECT_AH, file, &ib_dev);
2535 return PTR_ERR(uobj);
2537 if (!rdma_is_port_valid(ib_dev, cmd.attr.port_num)) {
2542 pd = uobj_get_obj_read(pd, UVERBS_OBJECT_PD, cmd.pd_handle, file);
2548 attr.type = rdma_ah_find_type(ib_dev, cmd.attr.port_num);
2549 rdma_ah_set_make_grd(&attr, false);
2550 rdma_ah_set_dlid(&attr, cmd.attr.dlid);
2551 rdma_ah_set_sl(&attr, cmd.attr.sl);
2552 rdma_ah_set_path_bits(&attr, cmd.attr.src_path_bits);
2553 rdma_ah_set_static_rate(&attr, cmd.attr.static_rate);
2554 rdma_ah_set_port_num(&attr, cmd.attr.port_num);
2556 if (cmd.attr.is_global) {
2557 rdma_ah_set_grh(&attr, NULL, cmd.attr.grh.flow_label,
2558 cmd.attr.grh.sgid_index,
2559 cmd.attr.grh.hop_limit,
2560 cmd.attr.grh.traffic_class);
2561 rdma_ah_set_dgid_raw(&attr, cmd.attr.grh.dgid);
2563 rdma_ah_set_ah_flags(&attr, 0);
2566 ah = rdma_create_user_ah(pd, &attr, &udata);
2573 uobj->user_handle = cmd.user_handle;
2576 resp.ah_handle = uobj->id;
2578 if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp)) {
2583 uobj_put_obj_read(pd);
2584 return uobj_alloc_commit(uobj, in_len);
2587 rdma_destroy_ah(ah);
2590 uobj_put_obj_read(pd);
2593 uobj_alloc_abort(uobj);
2597 ssize_t ib_uverbs_destroy_ah(struct ib_uverbs_file *file,
2598 const char __user *buf, int in_len, int out_len)
2600 struct ib_uverbs_destroy_ah cmd;
2602 if (copy_from_user(&cmd, buf, sizeof cmd))
2605 return uobj_perform_destroy(UVERBS_OBJECT_AH, cmd.ah_handle, file,
2609 ssize_t ib_uverbs_attach_mcast(struct ib_uverbs_file *file,
2610 const char __user *buf, int in_len,
2613 struct ib_uverbs_attach_mcast cmd;
2615 struct ib_uqp_object *obj;
2616 struct ib_uverbs_mcast_entry *mcast;
2619 if (copy_from_user(&cmd, buf, sizeof cmd))
2622 qp = uobj_get_obj_read(qp, UVERBS_OBJECT_QP, cmd.qp_handle, file);
2626 obj = container_of(qp->uobject, struct ib_uqp_object, uevent.uobject);
2628 mutex_lock(&obj->mcast_lock);
2629 list_for_each_entry(mcast, &obj->mcast_list, list)
2630 if (cmd.mlid == mcast->lid &&
2631 !memcmp(cmd.gid, mcast->gid.raw, sizeof mcast->gid.raw)) {
2636 mcast = kmalloc(sizeof *mcast, GFP_KERNEL);
2642 mcast->lid = cmd.mlid;
2643 memcpy(mcast->gid.raw, cmd.gid, sizeof mcast->gid.raw);
2645 ret = ib_attach_mcast(qp, &mcast->gid, cmd.mlid);
2647 list_add_tail(&mcast->list, &obj->mcast_list);
2652 mutex_unlock(&obj->mcast_lock);
2653 uobj_put_obj_read(qp);
2655 return ret ? ret : in_len;
2658 ssize_t ib_uverbs_detach_mcast(struct ib_uverbs_file *file,
2659 const char __user *buf, int in_len,
2662 struct ib_uverbs_detach_mcast cmd;
2663 struct ib_uqp_object *obj;
2665 struct ib_uverbs_mcast_entry *mcast;
2669 if (copy_from_user(&cmd, buf, sizeof cmd))
2672 qp = uobj_get_obj_read(qp, UVERBS_OBJECT_QP, cmd.qp_handle, file);
2676 obj = container_of(qp->uobject, struct ib_uqp_object, uevent.uobject);
2677 mutex_lock(&obj->mcast_lock);
2679 list_for_each_entry(mcast, &obj->mcast_list, list)
2680 if (cmd.mlid == mcast->lid &&
2681 !memcmp(cmd.gid, mcast->gid.raw, sizeof mcast->gid.raw)) {
2682 list_del(&mcast->list);
2693 ret = ib_detach_mcast(qp, (union ib_gid *)cmd.gid, cmd.mlid);
2696 mutex_unlock(&obj->mcast_lock);
2697 uobj_put_obj_read(qp);
2698 return ret ? ret : in_len;
2701 struct ib_uflow_resources {
2704 size_t collection_num;
2705 size_t counters_num;
2706 struct ib_counters **counters;
2707 struct ib_flow_action **collection;
2710 static struct ib_uflow_resources *flow_resources_alloc(size_t num_specs)
2712 struct ib_uflow_resources *resources;
2714 resources = kzalloc(sizeof(*resources), GFP_KERNEL);
2722 resources->counters =
2723 kcalloc(num_specs, sizeof(*resources->counters), GFP_KERNEL);
2724 resources->collection =
2725 kcalloc(num_specs, sizeof(*resources->collection), GFP_KERNEL);
2727 if (!resources->counters || !resources->collection)
2731 resources->max = num_specs;
2735 kfree(resources->counters);
2741 void ib_uverbs_flow_resources_free(struct ib_uflow_resources *uflow_res)
2748 for (i = 0; i < uflow_res->collection_num; i++)
2749 atomic_dec(&uflow_res->collection[i]->usecnt);
2751 for (i = 0; i < uflow_res->counters_num; i++)
2752 atomic_dec(&uflow_res->counters[i]->usecnt);
2754 kfree(uflow_res->collection);
2755 kfree(uflow_res->counters);
2759 static void flow_resources_add(struct ib_uflow_resources *uflow_res,
2760 enum ib_flow_spec_type type,
2763 WARN_ON(uflow_res->num >= uflow_res->max);
2766 case IB_FLOW_SPEC_ACTION_HANDLE:
2767 atomic_inc(&((struct ib_flow_action *)ibobj)->usecnt);
2768 uflow_res->collection[uflow_res->collection_num++] =
2769 (struct ib_flow_action *)ibobj;
2771 case IB_FLOW_SPEC_ACTION_COUNT:
2772 atomic_inc(&((struct ib_counters *)ibobj)->usecnt);
2773 uflow_res->counters[uflow_res->counters_num++] =
2774 (struct ib_counters *)ibobj;
2783 static int kern_spec_to_ib_spec_action(struct ib_uverbs_file *ufile,
2784 struct ib_uverbs_flow_spec *kern_spec,
2785 union ib_flow_spec *ib_spec,
2786 struct ib_uflow_resources *uflow_res)
2788 ib_spec->type = kern_spec->type;
2789 switch (ib_spec->type) {
2790 case IB_FLOW_SPEC_ACTION_TAG:
2791 if (kern_spec->flow_tag.size !=
2792 sizeof(struct ib_uverbs_flow_spec_action_tag))
2795 ib_spec->flow_tag.size = sizeof(struct ib_flow_spec_action_tag);
2796 ib_spec->flow_tag.tag_id = kern_spec->flow_tag.tag_id;
2798 case IB_FLOW_SPEC_ACTION_DROP:
2799 if (kern_spec->drop.size !=
2800 sizeof(struct ib_uverbs_flow_spec_action_drop))
2803 ib_spec->drop.size = sizeof(struct ib_flow_spec_action_drop);
2805 case IB_FLOW_SPEC_ACTION_HANDLE:
2806 if (kern_spec->action.size !=
2807 sizeof(struct ib_uverbs_flow_spec_action_handle))
2809 ib_spec->action.act = uobj_get_obj_read(flow_action,
2810 UVERBS_OBJECT_FLOW_ACTION,
2811 kern_spec->action.handle,
2813 if (!ib_spec->action.act)
2815 ib_spec->action.size =
2816 sizeof(struct ib_flow_spec_action_handle);
2817 flow_resources_add(uflow_res,
2818 IB_FLOW_SPEC_ACTION_HANDLE,
2819 ib_spec->action.act);
2820 uobj_put_obj_read(ib_spec->action.act);
2822 case IB_FLOW_SPEC_ACTION_COUNT:
2823 if (kern_spec->flow_count.size !=
2824 sizeof(struct ib_uverbs_flow_spec_action_count))
2826 ib_spec->flow_count.counters =
2827 uobj_get_obj_read(counters,
2828 UVERBS_OBJECT_COUNTERS,
2829 kern_spec->flow_count.handle,
2831 if (!ib_spec->flow_count.counters)
2833 ib_spec->flow_count.size =
2834 sizeof(struct ib_flow_spec_action_count);
2835 flow_resources_add(uflow_res,
2836 IB_FLOW_SPEC_ACTION_COUNT,
2837 ib_spec->flow_count.counters);
2838 uobj_put_obj_read(ib_spec->flow_count.counters);
2846 static size_t kern_spec_filter_sz(const struct ib_uverbs_flow_spec_hdr *spec)
2848 /* Returns user space filter size, includes padding */
2849 return (spec->size - sizeof(struct ib_uverbs_flow_spec_hdr)) / 2;
2852 static ssize_t spec_filter_size(const void *kern_spec_filter, u16 kern_filter_size,
2853 u16 ib_real_filter_sz)
2856 * User space filter structures must be 64 bit aligned, otherwise this
2857 * may pass, but we won't handle additional new attributes.
2860 if (kern_filter_size > ib_real_filter_sz) {
2861 if (memchr_inv(kern_spec_filter +
2862 ib_real_filter_sz, 0,
2863 kern_filter_size - ib_real_filter_sz))
2865 return ib_real_filter_sz;
2867 return kern_filter_size;
2870 int ib_uverbs_kern_spec_to_ib_spec_filter(enum ib_flow_spec_type type,
2871 const void *kern_spec_mask,
2872 const void *kern_spec_val,
2873 size_t kern_filter_sz,
2874 union ib_flow_spec *ib_spec)
2876 ssize_t actual_filter_sz;
2877 ssize_t ib_filter_sz;
2879 /* User flow spec size must be aligned to 4 bytes */
2880 if (kern_filter_sz != ALIGN(kern_filter_sz, 4))
2883 ib_spec->type = type;
2885 if (ib_spec->type == (IB_FLOW_SPEC_INNER | IB_FLOW_SPEC_VXLAN_TUNNEL))
2888 switch (ib_spec->type & ~IB_FLOW_SPEC_INNER) {
2889 case IB_FLOW_SPEC_ETH:
2890 ib_filter_sz = offsetof(struct ib_flow_eth_filter, real_sz);
2891 actual_filter_sz = spec_filter_size(kern_spec_mask,
2894 if (actual_filter_sz <= 0)
2896 ib_spec->size = sizeof(struct ib_flow_spec_eth);
2897 memcpy(&ib_spec->eth.val, kern_spec_val, actual_filter_sz);
2898 memcpy(&ib_spec->eth.mask, kern_spec_mask, actual_filter_sz);
2900 case IB_FLOW_SPEC_IPV4:
2901 ib_filter_sz = offsetof(struct ib_flow_ipv4_filter, real_sz);
2902 actual_filter_sz = spec_filter_size(kern_spec_mask,
2905 if (actual_filter_sz <= 0)
2907 ib_spec->size = sizeof(struct ib_flow_spec_ipv4);
2908 memcpy(&ib_spec->ipv4.val, kern_spec_val, actual_filter_sz);
2909 memcpy(&ib_spec->ipv4.mask, kern_spec_mask, actual_filter_sz);
2911 case IB_FLOW_SPEC_IPV6:
2912 ib_filter_sz = offsetof(struct ib_flow_ipv6_filter, real_sz);
2913 actual_filter_sz = spec_filter_size(kern_spec_mask,
2916 if (actual_filter_sz <= 0)
2918 ib_spec->size = sizeof(struct ib_flow_spec_ipv6);
2919 memcpy(&ib_spec->ipv6.val, kern_spec_val, actual_filter_sz);
2920 memcpy(&ib_spec->ipv6.mask, kern_spec_mask, actual_filter_sz);
2922 if ((ntohl(ib_spec->ipv6.mask.flow_label)) >= BIT(20) ||
2923 (ntohl(ib_spec->ipv6.val.flow_label)) >= BIT(20))
2926 case IB_FLOW_SPEC_TCP:
2927 case IB_FLOW_SPEC_UDP:
2928 ib_filter_sz = offsetof(struct ib_flow_tcp_udp_filter, real_sz);
2929 actual_filter_sz = spec_filter_size(kern_spec_mask,
2932 if (actual_filter_sz <= 0)
2934 ib_spec->size = sizeof(struct ib_flow_spec_tcp_udp);
2935 memcpy(&ib_spec->tcp_udp.val, kern_spec_val, actual_filter_sz);
2936 memcpy(&ib_spec->tcp_udp.mask, kern_spec_mask, actual_filter_sz);
2938 case IB_FLOW_SPEC_VXLAN_TUNNEL:
2939 ib_filter_sz = offsetof(struct ib_flow_tunnel_filter, real_sz);
2940 actual_filter_sz = spec_filter_size(kern_spec_mask,
2943 if (actual_filter_sz <= 0)
2945 ib_spec->tunnel.size = sizeof(struct ib_flow_spec_tunnel);
2946 memcpy(&ib_spec->tunnel.val, kern_spec_val, actual_filter_sz);
2947 memcpy(&ib_spec->tunnel.mask, kern_spec_mask, actual_filter_sz);
2949 if ((ntohl(ib_spec->tunnel.mask.tunnel_id)) >= BIT(24) ||
2950 (ntohl(ib_spec->tunnel.val.tunnel_id)) >= BIT(24))
2953 case IB_FLOW_SPEC_ESP:
2954 ib_filter_sz = offsetof(struct ib_flow_esp_filter, real_sz);
2955 actual_filter_sz = spec_filter_size(kern_spec_mask,
2958 if (actual_filter_sz <= 0)
2960 ib_spec->esp.size = sizeof(struct ib_flow_spec_esp);
2961 memcpy(&ib_spec->esp.val, kern_spec_val, actual_filter_sz);
2962 memcpy(&ib_spec->esp.mask, kern_spec_mask, actual_filter_sz);
2964 case IB_FLOW_SPEC_GRE:
2965 ib_filter_sz = offsetof(struct ib_flow_gre_filter, real_sz);
2966 actual_filter_sz = spec_filter_size(kern_spec_mask,
2969 if (actual_filter_sz <= 0)
2971 ib_spec->gre.size = sizeof(struct ib_flow_spec_gre);
2972 memcpy(&ib_spec->gre.val, kern_spec_val, actual_filter_sz);
2973 memcpy(&ib_spec->gre.mask, kern_spec_mask, actual_filter_sz);
2975 case IB_FLOW_SPEC_MPLS:
2976 ib_filter_sz = offsetof(struct ib_flow_mpls_filter, real_sz);
2977 actual_filter_sz = spec_filter_size(kern_spec_mask,
2980 if (actual_filter_sz <= 0)
2982 ib_spec->mpls.size = sizeof(struct ib_flow_spec_mpls);
2983 memcpy(&ib_spec->mpls.val, kern_spec_val, actual_filter_sz);
2984 memcpy(&ib_spec->mpls.mask, kern_spec_mask, actual_filter_sz);
2992 static int kern_spec_to_ib_spec_filter(struct ib_uverbs_flow_spec *kern_spec,
2993 union ib_flow_spec *ib_spec)
2995 ssize_t kern_filter_sz;
2996 void *kern_spec_mask;
2997 void *kern_spec_val;
2999 kern_filter_sz = kern_spec_filter_sz(&kern_spec->hdr);
3001 kern_spec_val = (void *)kern_spec +
3002 sizeof(struct ib_uverbs_flow_spec_hdr);
3003 kern_spec_mask = kern_spec_val + kern_filter_sz;
3005 return ib_uverbs_kern_spec_to_ib_spec_filter(kern_spec->type,
3008 kern_filter_sz, ib_spec);
3011 static int kern_spec_to_ib_spec(struct ib_uverbs_file *ufile,
3012 struct ib_uverbs_flow_spec *kern_spec,
3013 union ib_flow_spec *ib_spec,
3014 struct ib_uflow_resources *uflow_res)
3016 if (kern_spec->reserved)
3019 if (kern_spec->type >= IB_FLOW_SPEC_ACTION_TAG)
3020 return kern_spec_to_ib_spec_action(ufile, kern_spec, ib_spec,
3023 return kern_spec_to_ib_spec_filter(kern_spec, ib_spec);
3026 int ib_uverbs_ex_create_wq(struct ib_uverbs_file *file,
3027 struct ib_udata *ucore,
3028 struct ib_udata *uhw)
3030 struct ib_uverbs_ex_create_wq cmd = {};
3031 struct ib_uverbs_ex_create_wq_resp resp = {};
3032 struct ib_uwq_object *obj;
3037 struct ib_wq_init_attr wq_init_attr = {};
3038 size_t required_cmd_sz;
3039 size_t required_resp_len;
3040 struct ib_device *ib_dev;
3042 required_cmd_sz = offsetof(typeof(cmd), max_sge) + sizeof(cmd.max_sge);
3043 required_resp_len = offsetof(typeof(resp), wqn) + sizeof(resp.wqn);
3045 if (ucore->inlen < required_cmd_sz)
3048 if (ucore->outlen < required_resp_len)
3051 if (ucore->inlen > sizeof(cmd) &&
3052 !ib_is_udata_cleared(ucore, sizeof(cmd),
3053 ucore->inlen - sizeof(cmd)))
3056 err = ib_copy_from_udata(&cmd, ucore, min(sizeof(cmd), ucore->inlen));
3063 obj = (struct ib_uwq_object *)uobj_alloc(UVERBS_OBJECT_WQ, file,
3066 return PTR_ERR(obj);
3068 pd = uobj_get_obj_read(pd, UVERBS_OBJECT_PD, cmd.pd_handle, file);
3074 cq = uobj_get_obj_read(cq, UVERBS_OBJECT_CQ, cmd.cq_handle, file);
3080 wq_init_attr.cq = cq;
3081 wq_init_attr.max_sge = cmd.max_sge;
3082 wq_init_attr.max_wr = cmd.max_wr;
3083 wq_init_attr.wq_context = file;
3084 wq_init_attr.wq_type = cmd.wq_type;
3085 wq_init_attr.event_handler = ib_uverbs_wq_event_handler;
3086 if (ucore->inlen >= (offsetof(typeof(cmd), create_flags) +
3087 sizeof(cmd.create_flags)))
3088 wq_init_attr.create_flags = cmd.create_flags;
3089 obj->uevent.events_reported = 0;
3090 INIT_LIST_HEAD(&obj->uevent.event_list);
3092 if (!pd->device->create_wq) {
3096 wq = pd->device->create_wq(pd, &wq_init_attr, uhw);
3102 wq->uobject = &obj->uevent.uobject;
3103 obj->uevent.uobject.object = wq;
3104 wq->wq_type = wq_init_attr.wq_type;
3107 wq->device = pd->device;
3108 wq->wq_context = wq_init_attr.wq_context;
3109 atomic_set(&wq->usecnt, 0);
3110 atomic_inc(&pd->usecnt);
3111 atomic_inc(&cq->usecnt);
3112 wq->uobject = &obj->uevent.uobject;
3113 obj->uevent.uobject.object = wq;
3115 memset(&resp, 0, sizeof(resp));
3116 resp.wq_handle = obj->uevent.uobject.id;
3117 resp.max_sge = wq_init_attr.max_sge;
3118 resp.max_wr = wq_init_attr.max_wr;
3119 resp.wqn = wq->wq_num;
3120 resp.response_length = required_resp_len;
3121 err = ib_copy_to_udata(ucore,
3122 &resp, resp.response_length);
3126 uobj_put_obj_read(pd);
3127 uobj_put_obj_read(cq);
3128 return uobj_alloc_commit(&obj->uevent.uobject, 0);
3133 uobj_put_obj_read(cq);
3135 uobj_put_obj_read(pd);
3137 uobj_alloc_abort(&obj->uevent.uobject);
3142 int ib_uverbs_ex_destroy_wq(struct ib_uverbs_file *file,
3143 struct ib_udata *ucore,
3144 struct ib_udata *uhw)
3146 struct ib_uverbs_ex_destroy_wq cmd = {};
3147 struct ib_uverbs_ex_destroy_wq_resp resp = {};
3148 struct ib_uobject *uobj;
3149 struct ib_uwq_object *obj;
3150 size_t required_cmd_sz;
3151 size_t required_resp_len;
3154 required_cmd_sz = offsetof(typeof(cmd), wq_handle) + sizeof(cmd.wq_handle);
3155 required_resp_len = offsetof(typeof(resp), reserved) + sizeof(resp.reserved);
3157 if (ucore->inlen < required_cmd_sz)
3160 if (ucore->outlen < required_resp_len)
3163 if (ucore->inlen > sizeof(cmd) &&
3164 !ib_is_udata_cleared(ucore, sizeof(cmd),
3165 ucore->inlen - sizeof(cmd)))
3168 ret = ib_copy_from_udata(&cmd, ucore, min(sizeof(cmd), ucore->inlen));
3175 resp.response_length = required_resp_len;
3176 uobj = uobj_get_destroy(UVERBS_OBJECT_WQ, cmd.wq_handle, file);
3178 return PTR_ERR(uobj);
3180 obj = container_of(uobj, struct ib_uwq_object, uevent.uobject);
3181 resp.events_reported = obj->uevent.events_reported;
3183 uobj_put_destroy(uobj);
3185 return ib_copy_to_udata(ucore, &resp, resp.response_length);
3188 int ib_uverbs_ex_modify_wq(struct ib_uverbs_file *file,
3189 struct ib_udata *ucore,
3190 struct ib_udata *uhw)
3192 struct ib_uverbs_ex_modify_wq cmd = {};
3194 struct ib_wq_attr wq_attr = {};
3195 size_t required_cmd_sz;
3198 required_cmd_sz = offsetof(typeof(cmd), curr_wq_state) + sizeof(cmd.curr_wq_state);
3199 if (ucore->inlen < required_cmd_sz)
3202 if (ucore->inlen > sizeof(cmd) &&
3203 !ib_is_udata_cleared(ucore, sizeof(cmd),
3204 ucore->inlen - sizeof(cmd)))
3207 ret = ib_copy_from_udata(&cmd, ucore, min(sizeof(cmd), ucore->inlen));
3214 if (cmd.attr_mask > (IB_WQ_STATE | IB_WQ_CUR_STATE | IB_WQ_FLAGS))
3217 wq = uobj_get_obj_read(wq, UVERBS_OBJECT_WQ, cmd.wq_handle, file);
3221 wq_attr.curr_wq_state = cmd.curr_wq_state;
3222 wq_attr.wq_state = cmd.wq_state;
3223 if (cmd.attr_mask & IB_WQ_FLAGS) {
3224 wq_attr.flags = cmd.flags;
3225 wq_attr.flags_mask = cmd.flags_mask;
3227 if (!wq->device->modify_wq) {
3231 ret = wq->device->modify_wq(wq, &wq_attr, cmd.attr_mask, uhw);
3233 uobj_put_obj_read(wq);
3237 int ib_uverbs_ex_create_rwq_ind_table(struct ib_uverbs_file *file,
3238 struct ib_udata *ucore,
3239 struct ib_udata *uhw)
3241 struct ib_uverbs_ex_create_rwq_ind_table cmd = {};
3242 struct ib_uverbs_ex_create_rwq_ind_table_resp resp = {};
3243 struct ib_uobject *uobj;
3245 struct ib_rwq_ind_table_init_attr init_attr = {};
3246 struct ib_rwq_ind_table *rwq_ind_tbl;
3247 struct ib_wq **wqs = NULL;
3248 u32 *wqs_handles = NULL;
3249 struct ib_wq *wq = NULL;
3250 int i, j, num_read_wqs;
3252 u32 expected_in_size;
3253 size_t required_cmd_sz_header;
3254 size_t required_resp_len;
3255 struct ib_device *ib_dev;
3257 required_cmd_sz_header = offsetof(typeof(cmd), log_ind_tbl_size) + sizeof(cmd.log_ind_tbl_size);
3258 required_resp_len = offsetof(typeof(resp), ind_tbl_num) + sizeof(resp.ind_tbl_num);
3260 if (ucore->inlen < required_cmd_sz_header)
3263 if (ucore->outlen < required_resp_len)
3266 err = ib_copy_from_udata(&cmd, ucore, required_cmd_sz_header);
3270 ucore->inbuf += required_cmd_sz_header;
3271 ucore->inlen -= required_cmd_sz_header;
3276 if (cmd.log_ind_tbl_size > IB_USER_VERBS_MAX_LOG_IND_TBL_SIZE)
3279 num_wq_handles = 1 << cmd.log_ind_tbl_size;
3280 expected_in_size = num_wq_handles * sizeof(__u32);
3281 if (num_wq_handles == 1)
3282 /* input size for wq handles is u64 aligned */
3283 expected_in_size += sizeof(__u32);
3285 if (ucore->inlen < expected_in_size)
3288 if (ucore->inlen > expected_in_size &&
3289 !ib_is_udata_cleared(ucore, expected_in_size,
3290 ucore->inlen - expected_in_size))
3293 wqs_handles = kcalloc(num_wq_handles, sizeof(*wqs_handles),
3298 err = ib_copy_from_udata(wqs_handles, ucore,
3299 num_wq_handles * sizeof(__u32));
3303 wqs = kcalloc(num_wq_handles, sizeof(*wqs), GFP_KERNEL);
3309 for (num_read_wqs = 0; num_read_wqs < num_wq_handles;
3311 wq = uobj_get_obj_read(wq, UVERBS_OBJECT_WQ,
3312 wqs_handles[num_read_wqs], file);
3318 wqs[num_read_wqs] = wq;
3321 uobj = uobj_alloc(UVERBS_OBJECT_RWQ_IND_TBL, file, &ib_dev);
3323 err = PTR_ERR(uobj);
3327 init_attr.log_ind_tbl_size = cmd.log_ind_tbl_size;
3328 init_attr.ind_tbl = wqs;
3330 if (!ib_dev->create_rwq_ind_table) {
3334 rwq_ind_tbl = ib_dev->create_rwq_ind_table(ib_dev, &init_attr, uhw);
3336 if (IS_ERR(rwq_ind_tbl)) {
3337 err = PTR_ERR(rwq_ind_tbl);
3341 rwq_ind_tbl->ind_tbl = wqs;
3342 rwq_ind_tbl->log_ind_tbl_size = init_attr.log_ind_tbl_size;
3343 rwq_ind_tbl->uobject = uobj;
3344 uobj->object = rwq_ind_tbl;
3345 rwq_ind_tbl->device = ib_dev;
3346 atomic_set(&rwq_ind_tbl->usecnt, 0);
3348 for (i = 0; i < num_wq_handles; i++)
3349 atomic_inc(&wqs[i]->usecnt);
3351 resp.ind_tbl_handle = uobj->id;
3352 resp.ind_tbl_num = rwq_ind_tbl->ind_tbl_num;
3353 resp.response_length = required_resp_len;
3355 err = ib_copy_to_udata(ucore,
3356 &resp, resp.response_length);
3362 for (j = 0; j < num_read_wqs; j++)
3363 uobj_put_obj_read(wqs[j]);
3365 return uobj_alloc_commit(uobj, 0);
3368 ib_destroy_rwq_ind_table(rwq_ind_tbl);
3370 uobj_alloc_abort(uobj);
3372 for (j = 0; j < num_read_wqs; j++)
3373 uobj_put_obj_read(wqs[j]);
3380 int ib_uverbs_ex_destroy_rwq_ind_table(struct ib_uverbs_file *file,
3381 struct ib_udata *ucore,
3382 struct ib_udata *uhw)
3384 struct ib_uverbs_ex_destroy_rwq_ind_table cmd = {};
3386 size_t required_cmd_sz;
3388 required_cmd_sz = offsetof(typeof(cmd), ind_tbl_handle) + sizeof(cmd.ind_tbl_handle);
3390 if (ucore->inlen < required_cmd_sz)
3393 if (ucore->inlen > sizeof(cmd) &&
3394 !ib_is_udata_cleared(ucore, sizeof(cmd),
3395 ucore->inlen - sizeof(cmd)))
3398 ret = ib_copy_from_udata(&cmd, ucore, min(sizeof(cmd), ucore->inlen));
3405 return uobj_perform_destroy(UVERBS_OBJECT_RWQ_IND_TBL,
3406 cmd.ind_tbl_handle, file, 0);
3409 int ib_uverbs_ex_create_flow(struct ib_uverbs_file *file,
3410 struct ib_udata *ucore,
3411 struct ib_udata *uhw)
3413 struct ib_uverbs_create_flow cmd;
3414 struct ib_uverbs_create_flow_resp resp;
3415 struct ib_uobject *uobj;
3416 struct ib_uflow_object *uflow;
3417 struct ib_flow *flow_id;
3418 struct ib_uverbs_flow_attr *kern_flow_attr;
3419 struct ib_flow_attr *flow_attr;
3421 struct ib_uflow_resources *uflow_res;
3426 struct ib_device *ib_dev;
3428 if (ucore->inlen < sizeof(cmd))
3431 if (ucore->outlen < sizeof(resp))
3434 err = ib_copy_from_udata(&cmd, ucore, sizeof(cmd));
3438 ucore->inbuf += sizeof(cmd);
3439 ucore->inlen -= sizeof(cmd);
3444 if (!capable(CAP_NET_RAW))
3447 if (cmd.flow_attr.flags >= IB_FLOW_ATTR_FLAGS_RESERVED)
3450 if ((cmd.flow_attr.flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
3451 ((cmd.flow_attr.type == IB_FLOW_ATTR_ALL_DEFAULT) ||
3452 (cmd.flow_attr.type == IB_FLOW_ATTR_MC_DEFAULT)))
3455 if (cmd.flow_attr.num_of_specs > IB_FLOW_SPEC_SUPPORT_LAYERS)
3458 if (cmd.flow_attr.size > ucore->inlen ||
3459 cmd.flow_attr.size >
3460 (cmd.flow_attr.num_of_specs * sizeof(struct ib_uverbs_flow_spec)))
3463 if (cmd.flow_attr.reserved[0] ||
3464 cmd.flow_attr.reserved[1])
3467 if (cmd.flow_attr.num_of_specs) {
3468 kern_flow_attr = kmalloc(sizeof(*kern_flow_attr) + cmd.flow_attr.size,
3470 if (!kern_flow_attr)
3473 memcpy(kern_flow_attr, &cmd.flow_attr, sizeof(*kern_flow_attr));
3474 err = ib_copy_from_udata(kern_flow_attr + 1, ucore,
3475 cmd.flow_attr.size);
3479 kern_flow_attr = &cmd.flow_attr;
3482 uobj = uobj_alloc(UVERBS_OBJECT_FLOW, file, &ib_dev);
3484 err = PTR_ERR(uobj);
3488 qp = uobj_get_obj_read(qp, UVERBS_OBJECT_QP, cmd.qp_handle, file);
3494 if (!qp->device->create_flow) {
3499 flow_attr = kzalloc(struct_size(flow_attr, flows,
3500 cmd.flow_attr.num_of_specs), GFP_KERNEL);
3505 uflow_res = flow_resources_alloc(cmd.flow_attr.num_of_specs);
3508 goto err_free_flow_attr;
3511 flow_attr->type = kern_flow_attr->type;
3512 flow_attr->priority = kern_flow_attr->priority;
3513 flow_attr->num_of_specs = kern_flow_attr->num_of_specs;
3514 flow_attr->port = kern_flow_attr->port;
3515 flow_attr->flags = kern_flow_attr->flags;
3516 flow_attr->size = sizeof(*flow_attr);
3518 kern_spec = kern_flow_attr + 1;
3519 ib_spec = flow_attr + 1;
3520 for (i = 0; i < flow_attr->num_of_specs &&
3521 cmd.flow_attr.size > offsetof(struct ib_uverbs_flow_spec, reserved) &&
3522 cmd.flow_attr.size >=
3523 ((struct ib_uverbs_flow_spec *)kern_spec)->size; i++) {
3524 err = kern_spec_to_ib_spec(file, kern_spec, ib_spec, uflow_res);
3529 ((union ib_flow_spec *) ib_spec)->size;
3530 cmd.flow_attr.size -= ((struct ib_uverbs_flow_spec *)kern_spec)->size;
3531 kern_spec += ((struct ib_uverbs_flow_spec *) kern_spec)->size;
3532 ib_spec += ((union ib_flow_spec *) ib_spec)->size;
3534 if (cmd.flow_attr.size || (i != flow_attr->num_of_specs)) {
3535 pr_warn("create flow failed, flow %d: %d bytes left from uverb cmd\n",
3536 i, cmd.flow_attr.size);
3541 flow_id = qp->device->create_flow(qp, flow_attr,
3542 IB_FLOW_DOMAIN_USER, uhw);
3544 if (IS_ERR(flow_id)) {
3545 err = PTR_ERR(flow_id);
3548 atomic_inc(&qp->usecnt);
3550 flow_id->device = qp->device;
3551 flow_id->uobject = uobj;
3552 uobj->object = flow_id;
3553 uflow = container_of(uobj, typeof(*uflow), uobject);
3554 uflow->resources = uflow_res;
3556 memset(&resp, 0, sizeof(resp));
3557 resp.flow_handle = uobj->id;
3559 err = ib_copy_to_udata(ucore,
3560 &resp, sizeof(resp));
3564 uobj_put_obj_read(qp);
3566 if (cmd.flow_attr.num_of_specs)
3567 kfree(kern_flow_attr);
3568 return uobj_alloc_commit(uobj, 0);
3570 if (!qp->device->destroy_flow(flow_id))
3571 atomic_dec(&qp->usecnt);
3573 ib_uverbs_flow_resources_free(uflow_res);
3577 uobj_put_obj_read(qp);
3579 uobj_alloc_abort(uobj);
3581 if (cmd.flow_attr.num_of_specs)
3582 kfree(kern_flow_attr);
3586 int ib_uverbs_ex_destroy_flow(struct ib_uverbs_file *file,
3587 struct ib_udata *ucore,
3588 struct ib_udata *uhw)
3590 struct ib_uverbs_destroy_flow cmd;
3593 if (ucore->inlen < sizeof(cmd))
3596 ret = ib_copy_from_udata(&cmd, ucore, sizeof(cmd));
3603 return uobj_perform_destroy(UVERBS_OBJECT_FLOW, cmd.flow_handle, file,
3607 static int __uverbs_create_xsrq(struct ib_uverbs_file *file,
3608 struct ib_uverbs_create_xsrq *cmd,
3609 struct ib_udata *udata)
3611 struct ib_uverbs_create_srq_resp resp;
3612 struct ib_usrq_object *obj;
3615 struct ib_uobject *uninitialized_var(xrcd_uobj);
3616 struct ib_srq_init_attr attr;
3618 struct ib_device *ib_dev;
3620 obj = (struct ib_usrq_object *)uobj_alloc(UVERBS_OBJECT_SRQ, file,
3623 return PTR_ERR(obj);
3625 if (cmd->srq_type == IB_SRQT_TM)
3626 attr.ext.tag_matching.max_num_tags = cmd->max_num_tags;
3628 if (cmd->srq_type == IB_SRQT_XRC) {
3629 xrcd_uobj = uobj_get_read(UVERBS_OBJECT_XRCD, cmd->xrcd_handle,
3631 if (IS_ERR(xrcd_uobj)) {
3636 attr.ext.xrc.xrcd = (struct ib_xrcd *)xrcd_uobj->object;
3637 if (!attr.ext.xrc.xrcd) {
3642 obj->uxrcd = container_of(xrcd_uobj, struct ib_uxrcd_object, uobject);
3643 atomic_inc(&obj->uxrcd->refcnt);
3646 if (ib_srq_has_cq(cmd->srq_type)) {
3647 attr.ext.cq = uobj_get_obj_read(cq, UVERBS_OBJECT_CQ,
3648 cmd->cq_handle, file);
3655 pd = uobj_get_obj_read(pd, UVERBS_OBJECT_PD, cmd->pd_handle, file);
3661 attr.event_handler = ib_uverbs_srq_event_handler;
3662 attr.srq_context = file;
3663 attr.srq_type = cmd->srq_type;
3664 attr.attr.max_wr = cmd->max_wr;
3665 attr.attr.max_sge = cmd->max_sge;
3666 attr.attr.srq_limit = cmd->srq_limit;
3668 obj->uevent.events_reported = 0;
3669 INIT_LIST_HEAD(&obj->uevent.event_list);
3671 srq = pd->device->create_srq(pd, &attr, udata);
3677 srq->device = pd->device;
3679 srq->srq_type = cmd->srq_type;
3680 srq->uobject = &obj->uevent.uobject;
3681 srq->event_handler = attr.event_handler;
3682 srq->srq_context = attr.srq_context;
3684 if (ib_srq_has_cq(cmd->srq_type)) {
3685 srq->ext.cq = attr.ext.cq;
3686 atomic_inc(&attr.ext.cq->usecnt);
3689 if (cmd->srq_type == IB_SRQT_XRC) {
3690 srq->ext.xrc.xrcd = attr.ext.xrc.xrcd;
3691 atomic_inc(&attr.ext.xrc.xrcd->usecnt);
3694 atomic_inc(&pd->usecnt);
3695 atomic_set(&srq->usecnt, 0);
3697 obj->uevent.uobject.object = srq;
3698 obj->uevent.uobject.user_handle = cmd->user_handle;
3700 memset(&resp, 0, sizeof resp);
3701 resp.srq_handle = obj->uevent.uobject.id;
3702 resp.max_wr = attr.attr.max_wr;
3703 resp.max_sge = attr.attr.max_sge;
3704 if (cmd->srq_type == IB_SRQT_XRC)
3705 resp.srqn = srq->ext.xrc.srq_num;
3707 if (copy_to_user(u64_to_user_ptr(cmd->response),
3708 &resp, sizeof resp)) {
3713 if (cmd->srq_type == IB_SRQT_XRC)
3714 uobj_put_read(xrcd_uobj);
3716 if (ib_srq_has_cq(cmd->srq_type))
3717 uobj_put_obj_read(attr.ext.cq);
3719 uobj_put_obj_read(pd);
3720 return uobj_alloc_commit(&obj->uevent.uobject, 0);
3723 ib_destroy_srq(srq);
3726 uobj_put_obj_read(pd);
3729 if (ib_srq_has_cq(cmd->srq_type))
3730 uobj_put_obj_read(attr.ext.cq);
3733 if (cmd->srq_type == IB_SRQT_XRC) {
3734 atomic_dec(&obj->uxrcd->refcnt);
3735 uobj_put_read(xrcd_uobj);
3739 uobj_alloc_abort(&obj->uevent.uobject);
3743 ssize_t ib_uverbs_create_srq(struct ib_uverbs_file *file,
3744 const char __user *buf, int in_len,
3747 struct ib_uverbs_create_srq cmd;
3748 struct ib_uverbs_create_xsrq xcmd;
3749 struct ib_uverbs_create_srq_resp resp;
3750 struct ib_udata udata;
3753 if (out_len < sizeof resp)
3756 if (copy_from_user(&cmd, buf, sizeof cmd))
3759 memset(&xcmd, 0, sizeof(xcmd));
3760 xcmd.response = cmd.response;
3761 xcmd.user_handle = cmd.user_handle;
3762 xcmd.srq_type = IB_SRQT_BASIC;
3763 xcmd.pd_handle = cmd.pd_handle;
3764 xcmd.max_wr = cmd.max_wr;
3765 xcmd.max_sge = cmd.max_sge;
3766 xcmd.srq_limit = cmd.srq_limit;
3768 ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
3769 u64_to_user_ptr(cmd.response) + sizeof(resp),
3770 in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
3771 out_len - sizeof(resp));
3773 ret = __uverbs_create_xsrq(file, &xcmd, &udata);
3780 ssize_t ib_uverbs_create_xsrq(struct ib_uverbs_file *file,
3781 const char __user *buf, int in_len, int out_len)
3783 struct ib_uverbs_create_xsrq cmd;
3784 struct ib_uverbs_create_srq_resp resp;
3785 struct ib_udata udata;
3788 if (out_len < sizeof resp)
3791 if (copy_from_user(&cmd, buf, sizeof cmd))
3794 ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
3795 u64_to_user_ptr(cmd.response) + sizeof(resp),
3796 in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
3797 out_len - sizeof(resp));
3799 ret = __uverbs_create_xsrq(file, &cmd, &udata);
3806 ssize_t ib_uverbs_modify_srq(struct ib_uverbs_file *file,
3807 const char __user *buf, int in_len,
3810 struct ib_uverbs_modify_srq cmd;
3811 struct ib_udata udata;
3813 struct ib_srq_attr attr;
3816 if (copy_from_user(&cmd, buf, sizeof cmd))
3819 ib_uverbs_init_udata(&udata, buf + sizeof cmd, NULL, in_len - sizeof cmd,
3822 srq = uobj_get_obj_read(srq, UVERBS_OBJECT_SRQ, cmd.srq_handle, file);
3826 attr.max_wr = cmd.max_wr;
3827 attr.srq_limit = cmd.srq_limit;
3829 ret = srq->device->modify_srq(srq, &attr, cmd.attr_mask, &udata);
3831 uobj_put_obj_read(srq);
3833 return ret ? ret : in_len;
3836 ssize_t ib_uverbs_query_srq(struct ib_uverbs_file *file,
3837 const char __user *buf,
3838 int in_len, int out_len)
3840 struct ib_uverbs_query_srq cmd;
3841 struct ib_uverbs_query_srq_resp resp;
3842 struct ib_srq_attr attr;
3846 if (out_len < sizeof resp)
3849 if (copy_from_user(&cmd, buf, sizeof cmd))
3852 srq = uobj_get_obj_read(srq, UVERBS_OBJECT_SRQ, cmd.srq_handle, file);
3856 ret = ib_query_srq(srq, &attr);
3858 uobj_put_obj_read(srq);
3863 memset(&resp, 0, sizeof resp);
3865 resp.max_wr = attr.max_wr;
3866 resp.max_sge = attr.max_sge;
3867 resp.srq_limit = attr.srq_limit;
3869 if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp))
3875 ssize_t ib_uverbs_destroy_srq(struct ib_uverbs_file *file,
3876 const char __user *buf, int in_len,
3879 struct ib_uverbs_destroy_srq cmd;
3880 struct ib_uverbs_destroy_srq_resp resp;
3881 struct ib_uobject *uobj;
3882 struct ib_uevent_object *obj;
3884 if (copy_from_user(&cmd, buf, sizeof cmd))
3887 uobj = uobj_get_destroy(UVERBS_OBJECT_SRQ, cmd.srq_handle, file);
3889 return PTR_ERR(uobj);
3891 obj = container_of(uobj, struct ib_uevent_object, uobject);
3892 memset(&resp, 0, sizeof(resp));
3893 resp.events_reported = obj->events_reported;
3895 uobj_put_destroy(uobj);
3897 if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof(resp)))
3903 int ib_uverbs_ex_query_device(struct ib_uverbs_file *file,
3904 struct ib_udata *ucore,
3905 struct ib_udata *uhw)
3907 struct ib_uverbs_ex_query_device_resp resp = { {0} };
3908 struct ib_uverbs_ex_query_device cmd;
3909 struct ib_device_attr attr = {0};
3910 struct ib_ucontext *ucontext;
3911 struct ib_device *ib_dev;
3914 ucontext = ib_uverbs_get_ucontext(file);
3915 if (IS_ERR(ucontext))
3916 return PTR_ERR(ucontext);
3917 ib_dev = ucontext->device;
3919 if (!ib_dev->query_device)
3922 if (ucore->inlen < sizeof(cmd))
3925 err = ib_copy_from_udata(&cmd, ucore, sizeof(cmd));
3935 resp.response_length = offsetof(typeof(resp), odp_caps);
3937 if (ucore->outlen < resp.response_length)
3940 err = ib_dev->query_device(ib_dev, &attr, uhw);
3944 copy_query_dev_fields(ucontext, &resp.base, &attr);
3946 if (ucore->outlen < resp.response_length + sizeof(resp.odp_caps))
3949 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
3950 resp.odp_caps.general_caps = attr.odp_caps.general_caps;
3951 resp.odp_caps.per_transport_caps.rc_odp_caps =
3952 attr.odp_caps.per_transport_caps.rc_odp_caps;
3953 resp.odp_caps.per_transport_caps.uc_odp_caps =
3954 attr.odp_caps.per_transport_caps.uc_odp_caps;
3955 resp.odp_caps.per_transport_caps.ud_odp_caps =
3956 attr.odp_caps.per_transport_caps.ud_odp_caps;
3958 resp.response_length += sizeof(resp.odp_caps);
3960 if (ucore->outlen < resp.response_length + sizeof(resp.timestamp_mask))
3963 resp.timestamp_mask = attr.timestamp_mask;
3964 resp.response_length += sizeof(resp.timestamp_mask);
3966 if (ucore->outlen < resp.response_length + sizeof(resp.hca_core_clock))
3969 resp.hca_core_clock = attr.hca_core_clock;
3970 resp.response_length += sizeof(resp.hca_core_clock);
3972 if (ucore->outlen < resp.response_length + sizeof(resp.device_cap_flags_ex))
3975 resp.device_cap_flags_ex = attr.device_cap_flags;
3976 resp.response_length += sizeof(resp.device_cap_flags_ex);
3978 if (ucore->outlen < resp.response_length + sizeof(resp.rss_caps))
3981 resp.rss_caps.supported_qpts = attr.rss_caps.supported_qpts;
3982 resp.rss_caps.max_rwq_indirection_tables =
3983 attr.rss_caps.max_rwq_indirection_tables;
3984 resp.rss_caps.max_rwq_indirection_table_size =
3985 attr.rss_caps.max_rwq_indirection_table_size;
3987 resp.response_length += sizeof(resp.rss_caps);
3989 if (ucore->outlen < resp.response_length + sizeof(resp.max_wq_type_rq))
3992 resp.max_wq_type_rq = attr.max_wq_type_rq;
3993 resp.response_length += sizeof(resp.max_wq_type_rq);
3995 if (ucore->outlen < resp.response_length + sizeof(resp.raw_packet_caps))
3998 resp.raw_packet_caps = attr.raw_packet_caps;
3999 resp.response_length += sizeof(resp.raw_packet_caps);
4001 if (ucore->outlen < resp.response_length + sizeof(resp.tm_caps))
4004 resp.tm_caps.max_rndv_hdr_size = attr.tm_caps.max_rndv_hdr_size;
4005 resp.tm_caps.max_num_tags = attr.tm_caps.max_num_tags;
4006 resp.tm_caps.max_ops = attr.tm_caps.max_ops;
4007 resp.tm_caps.max_sge = attr.tm_caps.max_sge;
4008 resp.tm_caps.flags = attr.tm_caps.flags;
4009 resp.response_length += sizeof(resp.tm_caps);
4011 if (ucore->outlen < resp.response_length + sizeof(resp.cq_moderation_caps))
4014 resp.cq_moderation_caps.max_cq_moderation_count =
4015 attr.cq_caps.max_cq_moderation_count;
4016 resp.cq_moderation_caps.max_cq_moderation_period =
4017 attr.cq_caps.max_cq_moderation_period;
4018 resp.response_length += sizeof(resp.cq_moderation_caps);
4020 if (ucore->outlen < resp.response_length + sizeof(resp.max_dm_size))
4023 resp.max_dm_size = attr.max_dm_size;
4024 resp.response_length += sizeof(resp.max_dm_size);
4026 err = ib_copy_to_udata(ucore, &resp, resp.response_length);
4030 int ib_uverbs_ex_modify_cq(struct ib_uverbs_file *file,
4031 struct ib_udata *ucore,
4032 struct ib_udata *uhw)
4034 struct ib_uverbs_ex_modify_cq cmd = {};
4036 size_t required_cmd_sz;
4039 required_cmd_sz = offsetof(typeof(cmd), reserved) +
4040 sizeof(cmd.reserved);
4041 if (ucore->inlen < required_cmd_sz)
4045 if (ucore->inlen > sizeof(cmd) &&
4046 !ib_is_udata_cleared(ucore, sizeof(cmd),
4047 ucore->inlen - sizeof(cmd)))
4050 ret = ib_copy_from_udata(&cmd, ucore, min(sizeof(cmd), ucore->inlen));
4054 if (!cmd.attr_mask || cmd.reserved)
4057 if (cmd.attr_mask > IB_CQ_MODERATE)
4060 cq = uobj_get_obj_read(cq, UVERBS_OBJECT_CQ, cmd.cq_handle, file);
4064 ret = rdma_set_cq_moderation(cq, cmd.attr.cq_count, cmd.attr.cq_period);
4066 uobj_put_obj_read(cq);