Commit | Line | Data |
---|---|---|
eb8ffbfe RD |
1 | /* |
2 | * Copyright (c) 2005 Topspin Communications. All rights reserved. | |
3 | * Copyright (c) 2005 Cisco Systems. All rights reserved. | |
2a1d9b7f | 4 | * Copyright (c) 2005 Mellanox Technologies. All rights reserved. |
eb8ffbfe RD |
5 | * |
6 | * This software is available to you under a choice of one of two | |
7 | * licenses. You may choose to be licensed under the terms of the GNU | |
8 | * General Public License (GPL) Version 2, available from the file | |
9 | * COPYING in the main directory of this source tree, or the | |
10 | * OpenIB.org BSD license below: | |
11 | * | |
12 | * Redistribution and use in source and binary forms, with or | |
13 | * without modification, are permitted provided that the following | |
14 | * conditions are met: | |
15 | * | |
16 | * - Redistributions of source code must retain the above | |
17 | * copyright notice, this list of conditions and the following | |
18 | * disclaimer. | |
19 | * | |
20 | * - Redistributions in binary form must reproduce the above | |
21 | * copyright notice, this list of conditions and the following | |
22 | * disclaimer in the documentation and/or other materials | |
23 | * provided with the distribution. | |
24 | * | |
25 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, | |
26 | * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF | |
27 | * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND | |
28 | * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS | |
29 | * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN | |
30 | * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN | |
31 | * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE | |
32 | * SOFTWARE. | |
eb8ffbfe RD |
33 | */ |
34 | ||
35 | #include <linux/mm.h> | |
36 | #include <linux/dma-mapping.h> | |
3f07c014 | 37 | #include <linux/sched/signal.h> |
6e84f315 | 38 | #include <linux/sched/mm.h> |
b108d976 | 39 | #include <linux/export.h> |
c8d8beea | 40 | #include <linux/hugetlb.h> |
5a0e3ad6 | 41 | #include <linux/slab.h> |
8ada2c1c | 42 | #include <rdma/ib_umem_odp.h> |
eb8ffbfe RD |
43 | |
44 | #include "uverbs.h" | |
45 | ||
92ddc447 | 46 | |
eb8ffbfe RD |
47 | static void __ib_umem_release(struct ib_device *dev, struct ib_umem *umem, int dirty) |
48 | { | |
eeb8461e YH |
49 | struct scatterlist *sg; |
50 | struct page *page; | |
eb8ffbfe RD |
51 | int i; |
52 | ||
eeb8461e YH |
53 | if (umem->nmap > 0) |
54 | ib_dma_unmap_sg(dev, umem->sg_head.sgl, | |
17069d32 | 55 | umem->npages, |
eeb8461e | 56 | DMA_BIDIRECTIONAL); |
45711f1a | 57 | |
eeb8461e | 58 | for_each_sg(umem->sg_head.sgl, sg, umem->npages, i) { |
eb8ffbfe | 59 | |
eeb8461e | 60 | page = sg_page(sg); |
53376fed | 61 | if (!PageDirty(page) && umem->writable && dirty) |
eeb8461e YH |
62 | set_page_dirty_lock(page); |
63 | put_page(page); | |
eb8ffbfe | 64 | } |
eeb8461e YH |
65 | |
66 | sg_free_table(&umem->sg_head); | |
67 | return; | |
68 | ||
eb8ffbfe RD |
69 | } |
70 | ||
f7c6a7b5 RD |
71 | /** |
72 | * ib_umem_get - Pin and DMA map userspace memory. | |
8ada2c1c SR |
73 | * |
74 | * If access flags indicate ODP memory, avoid pinning. Instead, stores | |
882214e2 | 75 | * the mm for future page fault handling in conjunction with MMU notifiers. |
8ada2c1c | 76 | * |
f7c6a7b5 RD |
77 | * @context: userspace context to pin memory for |
78 | * @addr: userspace virtual address to start at | |
79 | * @size: length of region to pin | |
80 | * @access: IB_ACCESS_xxx flags for memory being pinned | |
cb9fbc5c | 81 | * @dmasync: flush in-flight DMA when the memory region is written |
f7c6a7b5 RD |
82 | */ |
83 | struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr, | |
cb9fbc5c | 84 | size_t size, int access, int dmasync) |
eb8ffbfe | 85 | { |
f7c6a7b5 | 86 | struct ib_umem *umem; |
eb8ffbfe | 87 | struct page **page_list; |
c8d8beea | 88 | struct vm_area_struct **vma_list; |
eb8ffbfe RD |
89 | unsigned long locked; |
90 | unsigned long lock_limit; | |
91 | unsigned long cur_base; | |
92 | unsigned long npages; | |
f7c6a7b5 | 93 | int ret; |
eb8ffbfe | 94 | int i; |
00085f1e | 95 | unsigned long dma_attrs = 0; |
eeb8461e YH |
96 | struct scatterlist *sg, *sg_list_start; |
97 | int need_release = 0; | |
768ae309 | 98 | unsigned int gup_flags = FOLL_WRITE; |
cb9fbc5c AK |
99 | |
100 | if (dmasync) | |
00085f1e | 101 | dma_attrs |= DMA_ATTR_WRITE_BARRIER; |
eb8ffbfe | 102 | |
8494057a SR |
103 | /* |
104 | * If the combination of the addr and size requested for this memory | |
105 | * region causes an integer overflow, return error. | |
106 | */ | |
66578b0b YD |
107 | if (((addr + size) < addr) || |
108 | PAGE_ALIGN(addr + size) < (addr + size)) | |
8494057a SR |
109 | return ERR_PTR(-EINVAL); |
110 | ||
eb8ffbfe | 111 | if (!can_do_mlock()) |
f7c6a7b5 | 112 | return ERR_PTR(-EPERM); |
eb8ffbfe | 113 | |
eeb8461e | 114 | umem = kzalloc(sizeof *umem, GFP_KERNEL); |
f7c6a7b5 RD |
115 | if (!umem) |
116 | return ERR_PTR(-ENOMEM); | |
eb8ffbfe | 117 | |
3e7e1193 AK |
118 | umem->context = context; |
119 | umem->length = size; | |
120 | umem->address = addr; | |
121 | umem->page_shift = PAGE_SHIFT; | |
f7c6a7b5 | 122 | /* |
860f10a7 SG |
123 | * We ask for writable memory if any of the following |
124 | * access flags are set. "Local write" and "remote write" | |
f7c6a7b5 RD |
125 | * obviously require write access. "Remote atomic" can do |
126 | * things like fetch and add, which will modify memory, and | |
127 | * "MW bind" can change permissions by binding a window. | |
128 | */ | |
860f10a7 SG |
129 | umem->writable = !!(access & |
130 | (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE | | |
131 | IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND)); | |
eb8ffbfe | 132 | |
8ada2c1c | 133 | if (access & IB_ACCESS_ON_DEMAND) { |
0008b84e | 134 | ret = ib_umem_odp_get(context, umem, access); |
8ada2c1c SR |
135 | if (ret) { |
136 | kfree(umem); | |
137 | return ERR_PTR(ret); | |
138 | } | |
139 | return umem; | |
140 | } | |
141 | ||
142 | umem->odp_data = NULL; | |
143 | ||
c8d8beea JF |
144 | /* We assume the memory is from hugetlb until proved otherwise */ |
145 | umem->hugetlb = 1; | |
146 | ||
f7c6a7b5 RD |
147 | page_list = (struct page **) __get_free_page(GFP_KERNEL); |
148 | if (!page_list) { | |
149 | kfree(umem); | |
150 | return ERR_PTR(-ENOMEM); | |
151 | } | |
eb8ffbfe | 152 | |
c8d8beea JF |
153 | /* |
154 | * if we can't alloc the vma_list, it's not so bad; | |
155 | * just assume the memory is not hugetlb memory | |
156 | */ | |
157 | vma_list = (struct vm_area_struct **) __get_free_page(GFP_KERNEL); | |
158 | if (!vma_list) | |
159 | umem->hugetlb = 0; | |
160 | ||
406f9e5f | 161 | npages = ib_umem_num_pages(umem); |
eb8ffbfe RD |
162 | |
163 | down_write(¤t->mm->mmap_sem); | |
164 | ||
bc3e53f6 | 165 | locked = npages + current->mm->pinned_vm; |
ccbe9f0b | 166 | lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; |
eb8ffbfe RD |
167 | |
168 | if ((locked > lock_limit) && !capable(CAP_IPC_LOCK)) { | |
169 | ret = -ENOMEM; | |
170 | goto out; | |
171 | } | |
172 | ||
f7c6a7b5 | 173 | cur_base = addr & PAGE_MASK; |
eb8ffbfe | 174 | |
3c7ba576 | 175 | if (npages == 0 || npages > UINT_MAX) { |
eeb8461e YH |
176 | ret = -EINVAL; |
177 | goto out; | |
178 | } | |
179 | ||
180 | ret = sg_alloc_table(&umem->sg_head, npages, GFP_KERNEL); | |
181 | if (ret) | |
182 | goto out; | |
183 | ||
768ae309 LS |
184 | if (!umem->writable) |
185 | gup_flags |= FOLL_FORCE; | |
186 | ||
eeb8461e YH |
187 | need_release = 1; |
188 | sg_list_start = umem->sg_head.sgl; | |
189 | ||
eb8ffbfe | 190 | while (npages) { |
5f1d43de | 191 | ret = get_user_pages_longterm(cur_base, |
8079ffa0 | 192 | min_t(unsigned long, npages, |
eb8ffbfe | 193 | PAGE_SIZE / sizeof (struct page *)), |
768ae309 | 194 | gup_flags, page_list, vma_list); |
eb8ffbfe RD |
195 | |
196 | if (ret < 0) | |
197 | goto out; | |
198 | ||
eeb8461e | 199 | umem->npages += ret; |
eb8ffbfe RD |
200 | cur_base += ret * PAGE_SIZE; |
201 | npages -= ret; | |
202 | ||
eeb8461e YH |
203 | for_each_sg(sg_list_start, sg, ret, i) { |
204 | if (vma_list && !is_vm_hugetlb_page(vma_list[i])) | |
205 | umem->hugetlb = 0; | |
206 | ||
207 | sg_set_page(sg, page_list[i], PAGE_SIZE, 0); | |
eb8ffbfe RD |
208 | } |
209 | ||
eeb8461e YH |
210 | /* preparing for next loop */ |
211 | sg_list_start = sg; | |
eb8ffbfe RD |
212 | } |
213 | ||
eeb8461e YH |
214 | umem->nmap = ib_dma_map_sg_attrs(context->device, |
215 | umem->sg_head.sgl, | |
216 | umem->npages, | |
217 | DMA_BIDIRECTIONAL, | |
00085f1e | 218 | dma_attrs); |
eeb8461e YH |
219 | |
220 | if (umem->nmap <= 0) { | |
221 | ret = -ENOMEM; | |
222 | goto out; | |
223 | } | |
224 | ||
225 | ret = 0; | |
226 | ||
eb8ffbfe | 227 | out: |
f7c6a7b5 | 228 | if (ret < 0) { |
eeb8461e YH |
229 | if (need_release) |
230 | __ib_umem_release(context->device, umem, 0); | |
f7c6a7b5 RD |
231 | kfree(umem); |
232 | } else | |
bc3e53f6 | 233 | current->mm->pinned_vm = locked; |
eb8ffbfe RD |
234 | |
235 | up_write(¤t->mm->mmap_sem); | |
c8d8beea JF |
236 | if (vma_list) |
237 | free_page((unsigned long) vma_list); | |
eb8ffbfe RD |
238 | free_page((unsigned long) page_list); |
239 | ||
f7c6a7b5 | 240 | return ret < 0 ? ERR_PTR(ret) : umem; |
eb8ffbfe | 241 | } |
f7c6a7b5 | 242 | EXPORT_SYMBOL(ib_umem_get); |
eb8ffbfe | 243 | |
1bf66a30 | 244 | static void ib_umem_account(struct work_struct *work) |
eb8ffbfe | 245 | { |
1bf66a30 RD |
246 | struct ib_umem *umem = container_of(work, struct ib_umem, work); |
247 | ||
248 | down_write(&umem->mm->mmap_sem); | |
bc3e53f6 | 249 | umem->mm->pinned_vm -= umem->diff; |
1bf66a30 RD |
250 | up_write(&umem->mm->mmap_sem); |
251 | mmput(umem->mm); | |
252 | kfree(umem); | |
eb8ffbfe RD |
253 | } |
254 | ||
f7c6a7b5 RD |
255 | /** |
256 | * ib_umem_release - release memory pinned with ib_umem_get | |
257 | * @umem: umem struct to release | |
258 | */ | |
259 | void ib_umem_release(struct ib_umem *umem) | |
eb8ffbfe | 260 | { |
f7c6a7b5 | 261 | struct ib_ucontext *context = umem->context; |
eb8ffbfe | 262 | struct mm_struct *mm; |
87773dd5 | 263 | struct task_struct *task; |
f7c6a7b5 | 264 | unsigned long diff; |
eb8ffbfe | 265 | |
8ada2c1c SR |
266 | if (umem->odp_data) { |
267 | ib_umem_odp_release(umem); | |
268 | return; | |
269 | } | |
270 | ||
f7c6a7b5 | 271 | __ib_umem_release(umem->context->device, umem, 1); |
eb8ffbfe | 272 | |
8e907ed4 | 273 | task = get_pid_task(umem->context->tgid, PIDTYPE_PID); |
87773dd5 SB |
274 | if (!task) |
275 | goto out; | |
276 | mm = get_task_mm(task); | |
277 | put_task_struct(task); | |
278 | if (!mm) | |
279 | goto out; | |
eb8ffbfe | 280 | |
406f9e5f | 281 | diff = ib_umem_num_pages(umem); |
f7c6a7b5 | 282 | |
eb8ffbfe RD |
283 | /* |
284 | * We may be called with the mm's mmap_sem already held. This | |
285 | * can happen when a userspace munmap() is the call that drops | |
286 | * the last reference to our file and calls our release | |
287 | * method. If there are memory regions to destroy, we'll end | |
f7c6a7b5 RD |
288 | * up here and not be able to take the mmap_sem. In that case |
289 | * we defer the vm_locked accounting to the system workqueue. | |
eb8ffbfe | 290 | */ |
24bce508 RD |
291 | if (context->closing) { |
292 | if (!down_write_trylock(&mm->mmap_sem)) { | |
293 | INIT_WORK(&umem->work, ib_umem_account); | |
294 | umem->mm = mm; | |
295 | umem->diff = diff; | |
296 | ||
f0626710 | 297 | queue_work(ib_wq, &umem->work); |
24bce508 RD |
298 | return; |
299 | } | |
f7c6a7b5 RD |
300 | } else |
301 | down_write(&mm->mmap_sem); | |
302 | ||
87773dd5 | 303 | mm->pinned_vm -= diff; |
f7c6a7b5 RD |
304 | up_write(&mm->mmap_sem); |
305 | mmput(mm); | |
87773dd5 | 306 | out: |
1bf66a30 | 307 | kfree(umem); |
f7c6a7b5 RD |
308 | } |
309 | EXPORT_SYMBOL(ib_umem_release); | |
310 | ||
311 | int ib_umem_page_count(struct ib_umem *umem) | |
312 | { | |
f7c6a7b5 RD |
313 | int i; |
314 | int n; | |
eeb8461e | 315 | struct scatterlist *sg; |
f7c6a7b5 | 316 | |
8ada2c1c SR |
317 | if (umem->odp_data) |
318 | return ib_umem_num_pages(umem); | |
319 | ||
f7c6a7b5 | 320 | n = 0; |
eeb8461e | 321 | for_each_sg(umem->sg_head.sgl, sg, umem->nmap, i) |
3e7e1193 | 322 | n += sg_dma_len(sg) >> umem->page_shift; |
eb8ffbfe | 323 | |
f7c6a7b5 | 324 | return n; |
eb8ffbfe | 325 | } |
f7c6a7b5 | 326 | EXPORT_SYMBOL(ib_umem_page_count); |
c5d76f13 HE |
327 | |
328 | /* | |
329 | * Copy from the given ib_umem's pages to the given buffer. | |
330 | * | |
331 | * umem - the umem to copy from | |
332 | * offset - offset to start copying from | |
333 | * dst - destination buffer | |
334 | * length - buffer length | |
335 | * | |
336 | * Returns 0 on success, or an error code. | |
337 | */ | |
338 | int ib_umem_copy_from(void *dst, struct ib_umem *umem, size_t offset, | |
339 | size_t length) | |
340 | { | |
341 | size_t end = offset + length; | |
342 | int ret; | |
343 | ||
344 | if (offset > umem->length || length > umem->length - offset) { | |
345 | pr_err("ib_umem_copy_from not in range. offset: %zd umem length: %zd end: %zd\n", | |
346 | offset, umem->length, end); | |
347 | return -EINVAL; | |
348 | } | |
349 | ||
edf1a84f | 350 | ret = sg_pcopy_to_buffer(umem->sg_head.sgl, umem->npages, dst, length, |
c5d76f13 HE |
351 | offset + ib_umem_offset(umem)); |
352 | ||
353 | if (ret < 0) | |
354 | return ret; | |
355 | else if (ret != length) | |
356 | return -EINVAL; | |
357 | else | |
358 | return 0; | |
359 | } | |
360 | EXPORT_SYMBOL(ib_umem_copy_from); |