Linux Audio

Check our new training course

Loading...
v5.4
   1// SPDX-License-Identifier: GPL-2.0-only
   2/******************************************************************************
   3 * privcmd.c
   4 *
   5 * Interface to privileged domain-0 commands.
   6 *
   7 * Copyright (c) 2002-2004, K A Fraser, B Dragovic
   8 */
   9
  10#define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
  11
  12#include <linux/kernel.h>
  13#include <linux/module.h>
  14#include <linux/sched.h>
  15#include <linux/slab.h>
  16#include <linux/string.h>
  17#include <linux/errno.h>
  18#include <linux/mm.h>
  19#include <linux/mman.h>
  20#include <linux/uaccess.h>
  21#include <linux/swap.h>
  22#include <linux/highmem.h>
  23#include <linux/pagemap.h>
  24#include <linux/seq_file.h>
  25#include <linux/miscdevice.h>
  26#include <linux/moduleparam.h>
  27
  28#include <asm/pgalloc.h>
  29#include <asm/pgtable.h>
  30#include <asm/tlb.h>
  31#include <asm/xen/hypervisor.h>
  32#include <asm/xen/hypercall.h>
  33
  34#include <xen/xen.h>
  35#include <xen/privcmd.h>
  36#include <xen/interface/xen.h>
  37#include <xen/interface/memory.h>
  38#include <xen/interface/hvm/dm_op.h>
  39#include <xen/features.h>
  40#include <xen/page.h>
  41#include <xen/xen-ops.h>
  42#include <xen/balloon.h>
  43
  44#include "privcmd.h"
  45
  46MODULE_LICENSE("GPL");
  47
  48#define PRIV_VMA_LOCKED ((void *)1)
  49
  50static unsigned int privcmd_dm_op_max_num = 16;
  51module_param_named(dm_op_max_nr_bufs, privcmd_dm_op_max_num, uint, 0644);
  52MODULE_PARM_DESC(dm_op_max_nr_bufs,
  53		 "Maximum number of buffers per dm_op hypercall");
  54
  55static unsigned int privcmd_dm_op_buf_max_size = 4096;
  56module_param_named(dm_op_buf_max_size, privcmd_dm_op_buf_max_size, uint,
  57		   0644);
  58MODULE_PARM_DESC(dm_op_buf_max_size,
  59		 "Maximum size of a dm_op hypercall buffer");
  60
  61struct privcmd_data {
  62	domid_t domid;
  63};
  64
  65static int privcmd_vma_range_is_mapped(
  66               struct vm_area_struct *vma,
  67               unsigned long addr,
  68               unsigned long nr_pages);
  69
  70static long privcmd_ioctl_hypercall(struct file *file, void __user *udata)
  71{
  72	struct privcmd_data *data = file->private_data;
  73	struct privcmd_hypercall hypercall;
  74	long ret;
  75
  76	/* Disallow arbitrary hypercalls if restricted */
  77	if (data->domid != DOMID_INVALID)
  78		return -EPERM;
  79
  80	if (copy_from_user(&hypercall, udata, sizeof(hypercall)))
  81		return -EFAULT;
  82
  83	xen_preemptible_hcall_begin();
  84	ret = privcmd_call(hypercall.op,
  85			   hypercall.arg[0], hypercall.arg[1],
  86			   hypercall.arg[2], hypercall.arg[3],
  87			   hypercall.arg[4]);
  88	xen_preemptible_hcall_end();
  89
  90	return ret;
  91}
  92
  93static void free_page_list(struct list_head *pages)
  94{
  95	struct page *p, *n;
  96
  97	list_for_each_entry_safe(p, n, pages, lru)
  98		__free_page(p);
  99
 100	INIT_LIST_HEAD(pages);
 101}
 102
 103/*
 104 * Given an array of items in userspace, return a list of pages
 105 * containing the data.  If copying fails, either because of memory
 106 * allocation failure or a problem reading user memory, return an
 107 * error code; its up to the caller to dispose of any partial list.
 108 */
 109static int gather_array(struct list_head *pagelist,
 110			unsigned nelem, size_t size,
 111			const void __user *data)
 112{
 113	unsigned pageidx;
 114	void *pagedata;
 115	int ret;
 116
 117	if (size > PAGE_SIZE)
 118		return 0;
 119
 120	pageidx = PAGE_SIZE;
 121	pagedata = NULL;	/* quiet, gcc */
 122	while (nelem--) {
 123		if (pageidx > PAGE_SIZE-size) {
 124			struct page *page = alloc_page(GFP_KERNEL);
 125
 126			ret = -ENOMEM;
 127			if (page == NULL)
 128				goto fail;
 129
 130			pagedata = page_address(page);
 131
 132			list_add_tail(&page->lru, pagelist);
 133			pageidx = 0;
 134		}
 135
 136		ret = -EFAULT;
 137		if (copy_from_user(pagedata + pageidx, data, size))
 138			goto fail;
 139
 140		data += size;
 141		pageidx += size;
 142	}
 143
 144	ret = 0;
 145
 146fail:
 147	return ret;
 148}
 149
 150/*
 151 * Call function "fn" on each element of the array fragmented
 152 * over a list of pages.
 153 */
 154static int traverse_pages(unsigned nelem, size_t size,
 155			  struct list_head *pos,
 156			  int (*fn)(void *data, void *state),
 157			  void *state)
 158{
 159	void *pagedata;
 160	unsigned pageidx;
 161	int ret = 0;
 162
 163	BUG_ON(size > PAGE_SIZE);
 164
 165	pageidx = PAGE_SIZE;
 166	pagedata = NULL;	/* hush, gcc */
 167
 168	while (nelem--) {
 169		if (pageidx > PAGE_SIZE-size) {
 170			struct page *page;
 171			pos = pos->next;
 172			page = list_entry(pos, struct page, lru);
 173			pagedata = page_address(page);
 174			pageidx = 0;
 175		}
 176
 177		ret = (*fn)(pagedata + pageidx, state);
 178		if (ret)
 179			break;
 180		pageidx += size;
 181	}
 182
 183	return ret;
 184}
 185
 186/*
 187 * Similar to traverse_pages, but use each page as a "block" of
 188 * data to be processed as one unit.
 189 */
 190static int traverse_pages_block(unsigned nelem, size_t size,
 191				struct list_head *pos,
 192				int (*fn)(void *data, int nr, void *state),
 193				void *state)
 194{
 195	void *pagedata;
 
 196	int ret = 0;
 197
 198	BUG_ON(size > PAGE_SIZE);
 199
 
 
 200	while (nelem) {
 201		int nr = (PAGE_SIZE/size);
 202		struct page *page;
 203		if (nr > nelem)
 204			nr = nelem;
 205		pos = pos->next;
 206		page = list_entry(pos, struct page, lru);
 207		pagedata = page_address(page);
 208		ret = (*fn)(pagedata, nr, state);
 209		if (ret)
 210			break;
 211		nelem -= nr;
 212	}
 213
 214	return ret;
 215}
 216
 217struct mmap_gfn_state {
 218	unsigned long va;
 219	struct vm_area_struct *vma;
 220	domid_t domain;
 221};
 222
 223static int mmap_gfn_range(void *data, void *state)
 224{
 225	struct privcmd_mmap_entry *msg = data;
 226	struct mmap_gfn_state *st = state;
 227	struct vm_area_struct *vma = st->vma;
 228	int rc;
 229
 230	/* Do not allow range to wrap the address space. */
 231	if ((msg->npages > (LONG_MAX >> PAGE_SHIFT)) ||
 232	    ((unsigned long)(msg->npages << PAGE_SHIFT) >= -st->va))
 233		return -EINVAL;
 234
 235	/* Range chunks must be contiguous in va space. */
 236	if ((msg->va != st->va) ||
 237	    ((msg->va+(msg->npages<<PAGE_SHIFT)) > vma->vm_end))
 238		return -EINVAL;
 239
 240	rc = xen_remap_domain_gfn_range(vma,
 241					msg->va & PAGE_MASK,
 242					msg->mfn, msg->npages,
 243					vma->vm_page_prot,
 244					st->domain, NULL);
 245	if (rc < 0)
 246		return rc;
 247
 248	st->va += msg->npages << PAGE_SHIFT;
 249
 250	return 0;
 251}
 252
 253static long privcmd_ioctl_mmap(struct file *file, void __user *udata)
 254{
 255	struct privcmd_data *data = file->private_data;
 256	struct privcmd_mmap mmapcmd;
 257	struct mm_struct *mm = current->mm;
 258	struct vm_area_struct *vma;
 259	int rc;
 260	LIST_HEAD(pagelist);
 261	struct mmap_gfn_state state;
 262
 263	/* We only support privcmd_ioctl_mmap_batch for auto translated. */
 264	if (xen_feature(XENFEAT_auto_translated_physmap))
 265		return -ENOSYS;
 266
 267	if (copy_from_user(&mmapcmd, udata, sizeof(mmapcmd)))
 268		return -EFAULT;
 269
 270	/* If restriction is in place, check the domid matches */
 271	if (data->domid != DOMID_INVALID && data->domid != mmapcmd.dom)
 272		return -EPERM;
 273
 274	rc = gather_array(&pagelist,
 275			  mmapcmd.num, sizeof(struct privcmd_mmap_entry),
 276			  mmapcmd.entry);
 277
 278	if (rc || list_empty(&pagelist))
 279		goto out;
 280
 281	down_write(&mm->mmap_sem);
 282
 283	{
 284		struct page *page = list_first_entry(&pagelist,
 285						     struct page, lru);
 286		struct privcmd_mmap_entry *msg = page_address(page);
 287
 288		vma = find_vma(mm, msg->va);
 289		rc = -EINVAL;
 290
 291		if (!vma || (msg->va != vma->vm_start) || vma->vm_private_data)
 292			goto out_up;
 293		vma->vm_private_data = PRIV_VMA_LOCKED;
 294	}
 295
 296	state.va = vma->vm_start;
 297	state.vma = vma;
 298	state.domain = mmapcmd.dom;
 299
 300	rc = traverse_pages(mmapcmd.num, sizeof(struct privcmd_mmap_entry),
 301			    &pagelist,
 302			    mmap_gfn_range, &state);
 303
 304
 305out_up:
 306	up_write(&mm->mmap_sem);
 307
 308out:
 309	free_page_list(&pagelist);
 310
 311	return rc;
 312}
 313
 314struct mmap_batch_state {
 315	domid_t domain;
 316	unsigned long va;
 317	struct vm_area_struct *vma;
 318	int index;
 319	/* A tristate:
 320	 *      0 for no errors
 321	 *      1 if at least one error has happened (and no
 322	 *          -ENOENT errors have happened)
 323	 *      -ENOENT if at least 1 -ENOENT has happened.
 324	 */
 325	int global_error;
 326	int version;
 327
 328	/* User-space gfn array to store errors in the second pass for V1. */
 329	xen_pfn_t __user *user_gfn;
 330	/* User-space int array to store errors in the second pass for V2. */
 331	int __user *user_err;
 332};
 333
 334/* auto translated dom0 note: if domU being created is PV, then gfn is
 335 * mfn(addr on bus). If it's auto xlated, then gfn is pfn (input to HAP).
 336 */
 337static int mmap_batch_fn(void *data, int nr, void *state)
 338{
 339	xen_pfn_t *gfnp = data;
 340	struct mmap_batch_state *st = state;
 341	struct vm_area_struct *vma = st->vma;
 342	struct page **pages = vma->vm_private_data;
 343	struct page **cur_pages = NULL;
 344	int ret;
 345
 346	if (xen_feature(XENFEAT_auto_translated_physmap))
 347		cur_pages = &pages[st->index];
 348
 349	BUG_ON(nr < 0);
 350	ret = xen_remap_domain_gfn_array(st->vma, st->va & PAGE_MASK, gfnp, nr,
 351					 (int *)gfnp, st->vma->vm_page_prot,
 352					 st->domain, cur_pages);
 353
 354	/* Adjust the global_error? */
 355	if (ret != nr) {
 356		if (ret == -ENOENT)
 357			st->global_error = -ENOENT;
 358		else {
 359			/* Record that at least one error has happened. */
 360			if (st->global_error == 0)
 361				st->global_error = 1;
 362		}
 363	}
 364	st->va += XEN_PAGE_SIZE * nr;
 365	st->index += nr / XEN_PFN_PER_PAGE;
 366
 367	return 0;
 368}
 369
 370static int mmap_return_error(int err, struct mmap_batch_state *st)
 371{
 372	int ret;
 373
 374	if (st->version == 1) {
 375		if (err) {
 376			xen_pfn_t gfn;
 377
 378			ret = get_user(gfn, st->user_gfn);
 379			if (ret < 0)
 380				return ret;
 381			/*
 382			 * V1 encodes the error codes in the 32bit top
 383			 * nibble of the gfn (with its known
 384			 * limitations vis-a-vis 64 bit callers).
 385			 */
 386			gfn |= (err == -ENOENT) ?
 387				PRIVCMD_MMAPBATCH_PAGED_ERROR :
 388				PRIVCMD_MMAPBATCH_MFN_ERROR;
 389			return __put_user(gfn, st->user_gfn++);
 390		} else
 391			st->user_gfn++;
 392	} else { /* st->version == 2 */
 393		if (err)
 394			return __put_user(err, st->user_err++);
 395		else
 396			st->user_err++;
 397	}
 398
 399	return 0;
 400}
 401
 402static int mmap_return_errors(void *data, int nr, void *state)
 403{
 404	struct mmap_batch_state *st = state;
 405	int *errs = data;
 406	int i;
 407	int ret;
 408
 409	for (i = 0; i < nr; i++) {
 410		ret = mmap_return_error(errs[i], st);
 411		if (ret < 0)
 412			return ret;
 413	}
 414	return 0;
 415}
 416
 417/* Allocate pfns that are then mapped with gfns from foreign domid. Update
 418 * the vma with the page info to use later.
 419 * Returns: 0 if success, otherwise -errno
 420 */
 421static int alloc_empty_pages(struct vm_area_struct *vma, int numpgs)
 422{
 423	int rc;
 424	struct page **pages;
 425
 426	pages = kcalloc(numpgs, sizeof(pages[0]), GFP_KERNEL);
 427	if (pages == NULL)
 428		return -ENOMEM;
 429
 430	rc = alloc_xenballooned_pages(numpgs, pages);
 431	if (rc != 0) {
 432		pr_warn("%s Could not alloc %d pfns rc:%d\n", __func__,
 433			numpgs, rc);
 434		kfree(pages);
 435		return -ENOMEM;
 436	}
 437	BUG_ON(vma->vm_private_data != NULL);
 438	vma->vm_private_data = pages;
 439
 440	return 0;
 441}
 442
 443static const struct vm_operations_struct privcmd_vm_ops;
 444
 445static long privcmd_ioctl_mmap_batch(
 446	struct file *file, void __user *udata, int version)
 447{
 448	struct privcmd_data *data = file->private_data;
 449	int ret;
 450	struct privcmd_mmapbatch_v2 m;
 451	struct mm_struct *mm = current->mm;
 452	struct vm_area_struct *vma;
 453	unsigned long nr_pages;
 454	LIST_HEAD(pagelist);
 455	struct mmap_batch_state state;
 456
 457	switch (version) {
 458	case 1:
 459		if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch)))
 460			return -EFAULT;
 461		/* Returns per-frame error in m.arr. */
 462		m.err = NULL;
 463		if (!access_ok(m.arr, m.num * sizeof(*m.arr)))
 464			return -EFAULT;
 465		break;
 466	case 2:
 467		if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch_v2)))
 468			return -EFAULT;
 469		/* Returns per-frame error code in m.err. */
 470		if (!access_ok(m.err, m.num * (sizeof(*m.err))))
 471			return -EFAULT;
 472		break;
 473	default:
 474		return -EINVAL;
 475	}
 476
 477	/* If restriction is in place, check the domid matches */
 478	if (data->domid != DOMID_INVALID && data->domid != m.dom)
 479		return -EPERM;
 480
 481	nr_pages = DIV_ROUND_UP(m.num, XEN_PFN_PER_PAGE);
 482	if ((m.num <= 0) || (nr_pages > (LONG_MAX >> PAGE_SHIFT)))
 483		return -EINVAL;
 484
 485	ret = gather_array(&pagelist, m.num, sizeof(xen_pfn_t), m.arr);
 486
 487	if (ret)
 488		goto out;
 489	if (list_empty(&pagelist)) {
 490		ret = -EINVAL;
 491		goto out;
 492	}
 493
 494	if (version == 2) {
 495		/* Zero error array now to only copy back actual errors. */
 496		if (clear_user(m.err, sizeof(int) * m.num)) {
 497			ret = -EFAULT;
 498			goto out;
 499		}
 500	}
 501
 502	down_write(&mm->mmap_sem);
 503
 504	vma = find_vma(mm, m.addr);
 505	if (!vma ||
 506	    vma->vm_ops != &privcmd_vm_ops) {
 507		ret = -EINVAL;
 508		goto out_unlock;
 509	}
 510
 511	/*
 512	 * Caller must either:
 513	 *
 514	 * Map the whole VMA range, which will also allocate all the
 515	 * pages required for the auto_translated_physmap case.
 516	 *
 517	 * Or
 518	 *
 519	 * Map unmapped holes left from a previous map attempt (e.g.,
 520	 * because those foreign frames were previously paged out).
 521	 */
 522	if (vma->vm_private_data == NULL) {
 523		if (m.addr != vma->vm_start ||
 524		    m.addr + (nr_pages << PAGE_SHIFT) != vma->vm_end) {
 525			ret = -EINVAL;
 526			goto out_unlock;
 527		}
 528		if (xen_feature(XENFEAT_auto_translated_physmap)) {
 529			ret = alloc_empty_pages(vma, nr_pages);
 530			if (ret < 0)
 531				goto out_unlock;
 532		} else
 533			vma->vm_private_data = PRIV_VMA_LOCKED;
 534	} else {
 535		if (m.addr < vma->vm_start ||
 536		    m.addr + (nr_pages << PAGE_SHIFT) > vma->vm_end) {
 537			ret = -EINVAL;
 538			goto out_unlock;
 539		}
 540		if (privcmd_vma_range_is_mapped(vma, m.addr, nr_pages)) {
 541			ret = -EINVAL;
 542			goto out_unlock;
 543		}
 544	}
 545
 546	state.domain        = m.dom;
 547	state.vma           = vma;
 548	state.va            = m.addr;
 549	state.index         = 0;
 550	state.global_error  = 0;
 551	state.version       = version;
 552
 553	BUILD_BUG_ON(((PAGE_SIZE / sizeof(xen_pfn_t)) % XEN_PFN_PER_PAGE) != 0);
 554	/* mmap_batch_fn guarantees ret == 0 */
 555	BUG_ON(traverse_pages_block(m.num, sizeof(xen_pfn_t),
 556				    &pagelist, mmap_batch_fn, &state));
 557
 558	up_write(&mm->mmap_sem);
 559
 560	if (state.global_error) {
 561		/* Write back errors in second pass. */
 562		state.user_gfn = (xen_pfn_t *)m.arr;
 563		state.user_err = m.err;
 564		ret = traverse_pages_block(m.num, sizeof(xen_pfn_t),
 565					   &pagelist, mmap_return_errors, &state);
 566	} else
 567		ret = 0;
 568
 569	/* If we have not had any EFAULT-like global errors then set the global
 570	 * error to -ENOENT if necessary. */
 571	if ((ret == 0) && (state.global_error == -ENOENT))
 572		ret = -ENOENT;
 573
 574out:
 575	free_page_list(&pagelist);
 576	return ret;
 577
 578out_unlock:
 579	up_write(&mm->mmap_sem);
 580	goto out;
 581}
 582
 583static int lock_pages(
 584	struct privcmd_dm_op_buf kbufs[], unsigned int num,
 585	struct page *pages[], unsigned int nr_pages)
 586{
 587	unsigned int i;
 588
 589	for (i = 0; i < num; i++) {
 590		unsigned int requested;
 591		int pinned;
 592
 593		requested = DIV_ROUND_UP(
 594			offset_in_page(kbufs[i].uptr) + kbufs[i].size,
 595			PAGE_SIZE);
 596		if (requested > nr_pages)
 597			return -ENOSPC;
 598
 599		pinned = get_user_pages_fast(
 600			(unsigned long) kbufs[i].uptr,
 601			requested, FOLL_WRITE, pages);
 602		if (pinned < 0)
 603			return pinned;
 604
 605		nr_pages -= pinned;
 606		pages += pinned;
 607	}
 608
 609	return 0;
 610}
 611
 612static void unlock_pages(struct page *pages[], unsigned int nr_pages)
 613{
 614	unsigned int i;
 615
 616	if (!pages)
 617		return;
 618
 619	for (i = 0; i < nr_pages; i++) {
 620		if (pages[i])
 621			put_page(pages[i]);
 622	}
 623}
 624
 625static long privcmd_ioctl_dm_op(struct file *file, void __user *udata)
 626{
 627	struct privcmd_data *data = file->private_data;
 628	struct privcmd_dm_op kdata;
 629	struct privcmd_dm_op_buf *kbufs;
 630	unsigned int nr_pages = 0;
 631	struct page **pages = NULL;
 632	struct xen_dm_op_buf *xbufs = NULL;
 633	unsigned int i;
 634	long rc;
 635
 636	if (copy_from_user(&kdata, udata, sizeof(kdata)))
 637		return -EFAULT;
 638
 639	/* If restriction is in place, check the domid matches */
 640	if (data->domid != DOMID_INVALID && data->domid != kdata.dom)
 641		return -EPERM;
 642
 643	if (kdata.num == 0)
 644		return 0;
 645
 646	if (kdata.num > privcmd_dm_op_max_num)
 647		return -E2BIG;
 648
 649	kbufs = kcalloc(kdata.num, sizeof(*kbufs), GFP_KERNEL);
 650	if (!kbufs)
 651		return -ENOMEM;
 652
 653	if (copy_from_user(kbufs, kdata.ubufs,
 654			   sizeof(*kbufs) * kdata.num)) {
 655		rc = -EFAULT;
 656		goto out;
 657	}
 658
 659	for (i = 0; i < kdata.num; i++) {
 660		if (kbufs[i].size > privcmd_dm_op_buf_max_size) {
 661			rc = -E2BIG;
 662			goto out;
 663		}
 664
 665		if (!access_ok(kbufs[i].uptr,
 666			       kbufs[i].size)) {
 667			rc = -EFAULT;
 668			goto out;
 669		}
 670
 671		nr_pages += DIV_ROUND_UP(
 672			offset_in_page(kbufs[i].uptr) + kbufs[i].size,
 673			PAGE_SIZE);
 674	}
 675
 676	pages = kcalloc(nr_pages, sizeof(*pages), GFP_KERNEL);
 677	if (!pages) {
 678		rc = -ENOMEM;
 679		goto out;
 680	}
 681
 682	xbufs = kcalloc(kdata.num, sizeof(*xbufs), GFP_KERNEL);
 683	if (!xbufs) {
 684		rc = -ENOMEM;
 685		goto out;
 686	}
 687
 688	rc = lock_pages(kbufs, kdata.num, pages, nr_pages);
 689	if (rc)
 690		goto out;
 691
 692	for (i = 0; i < kdata.num; i++) {
 693		set_xen_guest_handle(xbufs[i].h, kbufs[i].uptr);
 694		xbufs[i].size = kbufs[i].size;
 695	}
 696
 697	xen_preemptible_hcall_begin();
 698	rc = HYPERVISOR_dm_op(kdata.dom, kdata.num, xbufs);
 699	xen_preemptible_hcall_end();
 700
 701out:
 702	unlock_pages(pages, nr_pages);
 703	kfree(xbufs);
 704	kfree(pages);
 705	kfree(kbufs);
 706
 707	return rc;
 708}
 709
 710static long privcmd_ioctl_restrict(struct file *file, void __user *udata)
 711{
 712	struct privcmd_data *data = file->private_data;
 713	domid_t dom;
 714
 715	if (copy_from_user(&dom, udata, sizeof(dom)))
 716		return -EFAULT;
 717
 718	/* Set restriction to the specified domain, or check it matches */
 719	if (data->domid == DOMID_INVALID)
 720		data->domid = dom;
 721	else if (data->domid != dom)
 722		return -EINVAL;
 723
 724	return 0;
 725}
 726
 727static long privcmd_ioctl_mmap_resource(struct file *file, void __user *udata)
 728{
 729	struct privcmd_data *data = file->private_data;
 730	struct mm_struct *mm = current->mm;
 731	struct vm_area_struct *vma;
 732	struct privcmd_mmap_resource kdata;
 733	xen_pfn_t *pfns = NULL;
 734	struct xen_mem_acquire_resource xdata;
 735	int rc;
 736
 737	if (copy_from_user(&kdata, udata, sizeof(kdata)))
 738		return -EFAULT;
 739
 740	/* If restriction is in place, check the domid matches */
 741	if (data->domid != DOMID_INVALID && data->domid != kdata.dom)
 742		return -EPERM;
 743
 744	down_write(&mm->mmap_sem);
 745
 746	vma = find_vma(mm, kdata.addr);
 747	if (!vma || vma->vm_ops != &privcmd_vm_ops) {
 748		rc = -EINVAL;
 749		goto out;
 750	}
 751
 752	pfns = kcalloc(kdata.num, sizeof(*pfns), GFP_KERNEL);
 753	if (!pfns) {
 754		rc = -ENOMEM;
 755		goto out;
 756	}
 757
 758	if (IS_ENABLED(CONFIG_XEN_AUTO_XLATE) &&
 759	    xen_feature(XENFEAT_auto_translated_physmap)) {
 760		unsigned int nr = DIV_ROUND_UP(kdata.num, XEN_PFN_PER_PAGE);
 761		struct page **pages;
 762		unsigned int i;
 763
 764		rc = alloc_empty_pages(vma, nr);
 765		if (rc < 0)
 766			goto out;
 767
 768		pages = vma->vm_private_data;
 769		for (i = 0; i < kdata.num; i++) {
 770			xen_pfn_t pfn =
 771				page_to_xen_pfn(pages[i / XEN_PFN_PER_PAGE]);
 772
 773			pfns[i] = pfn + (i % XEN_PFN_PER_PAGE);
 774		}
 775	} else
 776		vma->vm_private_data = PRIV_VMA_LOCKED;
 777
 778	memset(&xdata, 0, sizeof(xdata));
 779	xdata.domid = kdata.dom;
 780	xdata.type = kdata.type;
 781	xdata.id = kdata.id;
 782	xdata.frame = kdata.idx;
 783	xdata.nr_frames = kdata.num;
 784	set_xen_guest_handle(xdata.frame_list, pfns);
 785
 786	xen_preemptible_hcall_begin();
 787	rc = HYPERVISOR_memory_op(XENMEM_acquire_resource, &xdata);
 788	xen_preemptible_hcall_end();
 789
 790	if (rc)
 791		goto out;
 792
 793	if (IS_ENABLED(CONFIG_XEN_AUTO_XLATE) &&
 794	    xen_feature(XENFEAT_auto_translated_physmap)) {
 795		rc = xen_remap_vma_range(vma, kdata.addr, kdata.num << PAGE_SHIFT);
 796	} else {
 797		unsigned int domid =
 798			(xdata.flags & XENMEM_rsrc_acq_caller_owned) ?
 799			DOMID_SELF : kdata.dom;
 800		int num;
 801
 802		num = xen_remap_domain_mfn_array(vma,
 803						 kdata.addr & PAGE_MASK,
 804						 pfns, kdata.num, (int *)pfns,
 805						 vma->vm_page_prot,
 806						 domid,
 807						 vma->vm_private_data);
 808		if (num < 0)
 809			rc = num;
 810		else if (num != kdata.num) {
 811			unsigned int i;
 812
 813			for (i = 0; i < num; i++) {
 814				rc = pfns[i];
 815				if (rc < 0)
 816					break;
 817			}
 818		} else
 819			rc = 0;
 820	}
 821
 822out:
 823	up_write(&mm->mmap_sem);
 824	kfree(pfns);
 825
 826	return rc;
 827}
 828
 829static long privcmd_ioctl(struct file *file,
 830			  unsigned int cmd, unsigned long data)
 831{
 832	int ret = -ENOTTY;
 833	void __user *udata = (void __user *) data;
 834
 835	switch (cmd) {
 836	case IOCTL_PRIVCMD_HYPERCALL:
 837		ret = privcmd_ioctl_hypercall(file, udata);
 838		break;
 839
 840	case IOCTL_PRIVCMD_MMAP:
 841		ret = privcmd_ioctl_mmap(file, udata);
 842		break;
 843
 844	case IOCTL_PRIVCMD_MMAPBATCH:
 845		ret = privcmd_ioctl_mmap_batch(file, udata, 1);
 846		break;
 847
 848	case IOCTL_PRIVCMD_MMAPBATCH_V2:
 849		ret = privcmd_ioctl_mmap_batch(file, udata, 2);
 850		break;
 851
 852	case IOCTL_PRIVCMD_DM_OP:
 853		ret = privcmd_ioctl_dm_op(file, udata);
 854		break;
 855
 856	case IOCTL_PRIVCMD_RESTRICT:
 857		ret = privcmd_ioctl_restrict(file, udata);
 858		break;
 859
 860	case IOCTL_PRIVCMD_MMAP_RESOURCE:
 861		ret = privcmd_ioctl_mmap_resource(file, udata);
 862		break;
 863
 864	default:
 
 865		break;
 866	}
 867
 868	return ret;
 869}
 870
 871static int privcmd_open(struct inode *ino, struct file *file)
 872{
 873	struct privcmd_data *data = kzalloc(sizeof(*data), GFP_KERNEL);
 874
 875	if (!data)
 876		return -ENOMEM;
 877
 878	/* DOMID_INVALID implies no restriction */
 879	data->domid = DOMID_INVALID;
 880
 881	file->private_data = data;
 882	return 0;
 883}
 884
 885static int privcmd_release(struct inode *ino, struct file *file)
 886{
 887	struct privcmd_data *data = file->private_data;
 888
 889	kfree(data);
 890	return 0;
 891}
 892
 893static void privcmd_close(struct vm_area_struct *vma)
 894{
 895	struct page **pages = vma->vm_private_data;
 896	int numpgs = vma_pages(vma);
 897	int numgfns = (vma->vm_end - vma->vm_start) >> XEN_PAGE_SHIFT;
 898	int rc;
 899
 900	if (!xen_feature(XENFEAT_auto_translated_physmap) || !numpgs || !pages)
 901		return;
 902
 903	rc = xen_unmap_domain_gfn_range(vma, numgfns, pages);
 904	if (rc == 0)
 905		free_xenballooned_pages(numpgs, pages);
 906	else
 907		pr_crit("unable to unmap MFN range: leaking %d pages. rc=%d\n",
 908			numpgs, rc);
 909	kfree(pages);
 910}
 911
 912static vm_fault_t privcmd_fault(struct vm_fault *vmf)
 913{
 914	printk(KERN_DEBUG "privcmd_fault: vma=%p %lx-%lx, pgoff=%lx, uv=%p\n",
 915	       vmf->vma, vmf->vma->vm_start, vmf->vma->vm_end,
 916	       vmf->pgoff, (void *)vmf->address);
 917
 918	return VM_FAULT_SIGBUS;
 919}
 920
 921static const struct vm_operations_struct privcmd_vm_ops = {
 922	.close = privcmd_close,
 923	.fault = privcmd_fault
 924};
 925
 926static int privcmd_mmap(struct file *file, struct vm_area_struct *vma)
 927{
 928	/* DONTCOPY is essential for Xen because copy_page_range doesn't know
 929	 * how to recreate these mappings */
 930	vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTCOPY |
 931			 VM_DONTEXPAND | VM_DONTDUMP;
 932	vma->vm_ops = &privcmd_vm_ops;
 933	vma->vm_private_data = NULL;
 934
 935	return 0;
 936}
 937
 938/*
 939 * For MMAPBATCH*. This allows asserting the singleshot mapping
 940 * on a per pfn/pte basis. Mapping calls that fail with ENOENT
 941 * can be then retried until success.
 942 */
 943static int is_mapped_fn(pte_t *pte, unsigned long addr, void *data)
 
 944{
 945	return pte_none(*pte) ? 0 : -EBUSY;
 946}
 947
 948static int privcmd_vma_range_is_mapped(
 949	           struct vm_area_struct *vma,
 950	           unsigned long addr,
 951	           unsigned long nr_pages)
 952{
 953	return apply_to_page_range(vma->vm_mm, addr, nr_pages << PAGE_SHIFT,
 954				   is_mapped_fn, NULL) != 0;
 955}
 956
 957const struct file_operations xen_privcmd_fops = {
 958	.owner = THIS_MODULE,
 959	.unlocked_ioctl = privcmd_ioctl,
 960	.open = privcmd_open,
 961	.release = privcmd_release,
 962	.mmap = privcmd_mmap,
 963};
 964EXPORT_SYMBOL_GPL(xen_privcmd_fops);
 965
 966static struct miscdevice privcmd_dev = {
 967	.minor = MISC_DYNAMIC_MINOR,
 968	.name = "xen/privcmd",
 969	.fops = &xen_privcmd_fops,
 970};
 971
 972static int __init privcmd_init(void)
 973{
 974	int err;
 975
 976	if (!xen_domain())
 977		return -ENODEV;
 978
 979	err = misc_register(&privcmd_dev);
 980	if (err != 0) {
 981		pr_err("Could not register Xen privcmd device\n");
 982		return err;
 983	}
 984
 985	err = misc_register(&xen_privcmdbuf_dev);
 986	if (err != 0) {
 987		pr_err("Could not register Xen hypercall-buf device\n");
 988		misc_deregister(&privcmd_dev);
 989		return err;
 990	}
 991
 992	return 0;
 993}
 994
 995static void __exit privcmd_exit(void)
 996{
 997	misc_deregister(&privcmd_dev);
 998	misc_deregister(&xen_privcmdbuf_dev);
 999}
1000
1001module_init(privcmd_init);
1002module_exit(privcmd_exit);
v4.6
 
  1/******************************************************************************
  2 * privcmd.c
  3 *
  4 * Interface to privileged domain-0 commands.
  5 *
  6 * Copyright (c) 2002-2004, K A Fraser, B Dragovic
  7 */
  8
  9#define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
 10
 11#include <linux/kernel.h>
 12#include <linux/module.h>
 13#include <linux/sched.h>
 14#include <linux/slab.h>
 15#include <linux/string.h>
 16#include <linux/errno.h>
 17#include <linux/mm.h>
 18#include <linux/mman.h>
 19#include <linux/uaccess.h>
 20#include <linux/swap.h>
 21#include <linux/highmem.h>
 22#include <linux/pagemap.h>
 23#include <linux/seq_file.h>
 24#include <linux/miscdevice.h>
 
 25
 26#include <asm/pgalloc.h>
 27#include <asm/pgtable.h>
 28#include <asm/tlb.h>
 29#include <asm/xen/hypervisor.h>
 30#include <asm/xen/hypercall.h>
 31
 32#include <xen/xen.h>
 33#include <xen/privcmd.h>
 34#include <xen/interface/xen.h>
 
 
 35#include <xen/features.h>
 36#include <xen/page.h>
 37#include <xen/xen-ops.h>
 38#include <xen/balloon.h>
 39
 40#include "privcmd.h"
 41
 42MODULE_LICENSE("GPL");
 43
 44#define PRIV_VMA_LOCKED ((void *)1)
 45
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 46static int privcmd_vma_range_is_mapped(
 47               struct vm_area_struct *vma,
 48               unsigned long addr,
 49               unsigned long nr_pages);
 50
 51static long privcmd_ioctl_hypercall(void __user *udata)
 52{
 
 53	struct privcmd_hypercall hypercall;
 54	long ret;
 55
 
 
 
 
 56	if (copy_from_user(&hypercall, udata, sizeof(hypercall)))
 57		return -EFAULT;
 58
 59	xen_preemptible_hcall_begin();
 60	ret = privcmd_call(hypercall.op,
 61			   hypercall.arg[0], hypercall.arg[1],
 62			   hypercall.arg[2], hypercall.arg[3],
 63			   hypercall.arg[4]);
 64	xen_preemptible_hcall_end();
 65
 66	return ret;
 67}
 68
 69static void free_page_list(struct list_head *pages)
 70{
 71	struct page *p, *n;
 72
 73	list_for_each_entry_safe(p, n, pages, lru)
 74		__free_page(p);
 75
 76	INIT_LIST_HEAD(pages);
 77}
 78
 79/*
 80 * Given an array of items in userspace, return a list of pages
 81 * containing the data.  If copying fails, either because of memory
 82 * allocation failure or a problem reading user memory, return an
 83 * error code; its up to the caller to dispose of any partial list.
 84 */
 85static int gather_array(struct list_head *pagelist,
 86			unsigned nelem, size_t size,
 87			const void __user *data)
 88{
 89	unsigned pageidx;
 90	void *pagedata;
 91	int ret;
 92
 93	if (size > PAGE_SIZE)
 94		return 0;
 95
 96	pageidx = PAGE_SIZE;
 97	pagedata = NULL;	/* quiet, gcc */
 98	while (nelem--) {
 99		if (pageidx > PAGE_SIZE-size) {
100			struct page *page = alloc_page(GFP_KERNEL);
101
102			ret = -ENOMEM;
103			if (page == NULL)
104				goto fail;
105
106			pagedata = page_address(page);
107
108			list_add_tail(&page->lru, pagelist);
109			pageidx = 0;
110		}
111
112		ret = -EFAULT;
113		if (copy_from_user(pagedata + pageidx, data, size))
114			goto fail;
115
116		data += size;
117		pageidx += size;
118	}
119
120	ret = 0;
121
122fail:
123	return ret;
124}
125
126/*
127 * Call function "fn" on each element of the array fragmented
128 * over a list of pages.
129 */
130static int traverse_pages(unsigned nelem, size_t size,
131			  struct list_head *pos,
132			  int (*fn)(void *data, void *state),
133			  void *state)
134{
135	void *pagedata;
136	unsigned pageidx;
137	int ret = 0;
138
139	BUG_ON(size > PAGE_SIZE);
140
141	pageidx = PAGE_SIZE;
142	pagedata = NULL;	/* hush, gcc */
143
144	while (nelem--) {
145		if (pageidx > PAGE_SIZE-size) {
146			struct page *page;
147			pos = pos->next;
148			page = list_entry(pos, struct page, lru);
149			pagedata = page_address(page);
150			pageidx = 0;
151		}
152
153		ret = (*fn)(pagedata + pageidx, state);
154		if (ret)
155			break;
156		pageidx += size;
157	}
158
159	return ret;
160}
161
162/*
163 * Similar to traverse_pages, but use each page as a "block" of
164 * data to be processed as one unit.
165 */
166static int traverse_pages_block(unsigned nelem, size_t size,
167				struct list_head *pos,
168				int (*fn)(void *data, int nr, void *state),
169				void *state)
170{
171	void *pagedata;
172	unsigned pageidx;
173	int ret = 0;
174
175	BUG_ON(size > PAGE_SIZE);
176
177	pageidx = PAGE_SIZE;
178
179	while (nelem) {
180		int nr = (PAGE_SIZE/size);
181		struct page *page;
182		if (nr > nelem)
183			nr = nelem;
184		pos = pos->next;
185		page = list_entry(pos, struct page, lru);
186		pagedata = page_address(page);
187		ret = (*fn)(pagedata, nr, state);
188		if (ret)
189			break;
190		nelem -= nr;
191	}
192
193	return ret;
194}
195
196struct mmap_gfn_state {
197	unsigned long va;
198	struct vm_area_struct *vma;
199	domid_t domain;
200};
201
202static int mmap_gfn_range(void *data, void *state)
203{
204	struct privcmd_mmap_entry *msg = data;
205	struct mmap_gfn_state *st = state;
206	struct vm_area_struct *vma = st->vma;
207	int rc;
208
209	/* Do not allow range to wrap the address space. */
210	if ((msg->npages > (LONG_MAX >> PAGE_SHIFT)) ||
211	    ((unsigned long)(msg->npages << PAGE_SHIFT) >= -st->va))
212		return -EINVAL;
213
214	/* Range chunks must be contiguous in va space. */
215	if ((msg->va != st->va) ||
216	    ((msg->va+(msg->npages<<PAGE_SHIFT)) > vma->vm_end))
217		return -EINVAL;
218
219	rc = xen_remap_domain_gfn_range(vma,
220					msg->va & PAGE_MASK,
221					msg->mfn, msg->npages,
222					vma->vm_page_prot,
223					st->domain, NULL);
224	if (rc < 0)
225		return rc;
226
227	st->va += msg->npages << PAGE_SHIFT;
228
229	return 0;
230}
231
232static long privcmd_ioctl_mmap(void __user *udata)
233{
 
234	struct privcmd_mmap mmapcmd;
235	struct mm_struct *mm = current->mm;
236	struct vm_area_struct *vma;
237	int rc;
238	LIST_HEAD(pagelist);
239	struct mmap_gfn_state state;
240
241	/* We only support privcmd_ioctl_mmap_batch for auto translated. */
242	if (xen_feature(XENFEAT_auto_translated_physmap))
243		return -ENOSYS;
244
245	if (copy_from_user(&mmapcmd, udata, sizeof(mmapcmd)))
246		return -EFAULT;
247
 
 
 
 
248	rc = gather_array(&pagelist,
249			  mmapcmd.num, sizeof(struct privcmd_mmap_entry),
250			  mmapcmd.entry);
251
252	if (rc || list_empty(&pagelist))
253		goto out;
254
255	down_write(&mm->mmap_sem);
256
257	{
258		struct page *page = list_first_entry(&pagelist,
259						     struct page, lru);
260		struct privcmd_mmap_entry *msg = page_address(page);
261
262		vma = find_vma(mm, msg->va);
263		rc = -EINVAL;
264
265		if (!vma || (msg->va != vma->vm_start) || vma->vm_private_data)
266			goto out_up;
267		vma->vm_private_data = PRIV_VMA_LOCKED;
268	}
269
270	state.va = vma->vm_start;
271	state.vma = vma;
272	state.domain = mmapcmd.dom;
273
274	rc = traverse_pages(mmapcmd.num, sizeof(struct privcmd_mmap_entry),
275			    &pagelist,
276			    mmap_gfn_range, &state);
277
278
279out_up:
280	up_write(&mm->mmap_sem);
281
282out:
283	free_page_list(&pagelist);
284
285	return rc;
286}
287
288struct mmap_batch_state {
289	domid_t domain;
290	unsigned long va;
291	struct vm_area_struct *vma;
292	int index;
293	/* A tristate:
294	 *      0 for no errors
295	 *      1 if at least one error has happened (and no
296	 *          -ENOENT errors have happened)
297	 *      -ENOENT if at least 1 -ENOENT has happened.
298	 */
299	int global_error;
300	int version;
301
302	/* User-space gfn array to store errors in the second pass for V1. */
303	xen_pfn_t __user *user_gfn;
304	/* User-space int array to store errors in the second pass for V2. */
305	int __user *user_err;
306};
307
308/* auto translated dom0 note: if domU being created is PV, then gfn is
309 * mfn(addr on bus). If it's auto xlated, then gfn is pfn (input to HAP).
310 */
311static int mmap_batch_fn(void *data, int nr, void *state)
312{
313	xen_pfn_t *gfnp = data;
314	struct mmap_batch_state *st = state;
315	struct vm_area_struct *vma = st->vma;
316	struct page **pages = vma->vm_private_data;
317	struct page **cur_pages = NULL;
318	int ret;
319
320	if (xen_feature(XENFEAT_auto_translated_physmap))
321		cur_pages = &pages[st->index];
322
323	BUG_ON(nr < 0);
324	ret = xen_remap_domain_gfn_array(st->vma, st->va & PAGE_MASK, gfnp, nr,
325					 (int *)gfnp, st->vma->vm_page_prot,
326					 st->domain, cur_pages);
327
328	/* Adjust the global_error? */
329	if (ret != nr) {
330		if (ret == -ENOENT)
331			st->global_error = -ENOENT;
332		else {
333			/* Record that at least one error has happened. */
334			if (st->global_error == 0)
335				st->global_error = 1;
336		}
337	}
338	st->va += PAGE_SIZE * nr;
339	st->index += nr;
340
341	return 0;
342}
343
344static int mmap_return_error(int err, struct mmap_batch_state *st)
345{
346	int ret;
347
348	if (st->version == 1) {
349		if (err) {
350			xen_pfn_t gfn;
351
352			ret = get_user(gfn, st->user_gfn);
353			if (ret < 0)
354				return ret;
355			/*
356			 * V1 encodes the error codes in the 32bit top
357			 * nibble of the gfn (with its known
358			 * limitations vis-a-vis 64 bit callers).
359			 */
360			gfn |= (err == -ENOENT) ?
361				PRIVCMD_MMAPBATCH_PAGED_ERROR :
362				PRIVCMD_MMAPBATCH_MFN_ERROR;
363			return __put_user(gfn, st->user_gfn++);
364		} else
365			st->user_gfn++;
366	} else { /* st->version == 2 */
367		if (err)
368			return __put_user(err, st->user_err++);
369		else
370			st->user_err++;
371	}
372
373	return 0;
374}
375
376static int mmap_return_errors(void *data, int nr, void *state)
377{
378	struct mmap_batch_state *st = state;
379	int *errs = data;
380	int i;
381	int ret;
382
383	for (i = 0; i < nr; i++) {
384		ret = mmap_return_error(errs[i], st);
385		if (ret < 0)
386			return ret;
387	}
388	return 0;
389}
390
391/* Allocate pfns that are then mapped with gfns from foreign domid. Update
392 * the vma with the page info to use later.
393 * Returns: 0 if success, otherwise -errno
394 */
395static int alloc_empty_pages(struct vm_area_struct *vma, int numpgs)
396{
397	int rc;
398	struct page **pages;
399
400	pages = kcalloc(numpgs, sizeof(pages[0]), GFP_KERNEL);
401	if (pages == NULL)
402		return -ENOMEM;
403
404	rc = alloc_xenballooned_pages(numpgs, pages);
405	if (rc != 0) {
406		pr_warn("%s Could not alloc %d pfns rc:%d\n", __func__,
407			numpgs, rc);
408		kfree(pages);
409		return -ENOMEM;
410	}
411	BUG_ON(vma->vm_private_data != NULL);
412	vma->vm_private_data = pages;
413
414	return 0;
415}
416
417static const struct vm_operations_struct privcmd_vm_ops;
418
419static long privcmd_ioctl_mmap_batch(void __user *udata, int version)
 
420{
 
421	int ret;
422	struct privcmd_mmapbatch_v2 m;
423	struct mm_struct *mm = current->mm;
424	struct vm_area_struct *vma;
425	unsigned long nr_pages;
426	LIST_HEAD(pagelist);
427	struct mmap_batch_state state;
428
429	switch (version) {
430	case 1:
431		if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch)))
432			return -EFAULT;
433		/* Returns per-frame error in m.arr. */
434		m.err = NULL;
435		if (!access_ok(VERIFY_WRITE, m.arr, m.num * sizeof(*m.arr)))
436			return -EFAULT;
437		break;
438	case 2:
439		if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch_v2)))
440			return -EFAULT;
441		/* Returns per-frame error code in m.err. */
442		if (!access_ok(VERIFY_WRITE, m.err, m.num * (sizeof(*m.err))))
443			return -EFAULT;
444		break;
445	default:
446		return -EINVAL;
447	}
448
 
 
 
 
449	nr_pages = DIV_ROUND_UP(m.num, XEN_PFN_PER_PAGE);
450	if ((m.num <= 0) || (nr_pages > (LONG_MAX >> PAGE_SHIFT)))
451		return -EINVAL;
452
453	ret = gather_array(&pagelist, m.num, sizeof(xen_pfn_t), m.arr);
454
455	if (ret)
456		goto out;
457	if (list_empty(&pagelist)) {
458		ret = -EINVAL;
459		goto out;
460	}
461
462	if (version == 2) {
463		/* Zero error array now to only copy back actual errors. */
464		if (clear_user(m.err, sizeof(int) * m.num)) {
465			ret = -EFAULT;
466			goto out;
467		}
468	}
469
470	down_write(&mm->mmap_sem);
471
472	vma = find_vma(mm, m.addr);
473	if (!vma ||
474	    vma->vm_ops != &privcmd_vm_ops) {
475		ret = -EINVAL;
476		goto out_unlock;
477	}
478
479	/*
480	 * Caller must either:
481	 *
482	 * Map the whole VMA range, which will also allocate all the
483	 * pages required for the auto_translated_physmap case.
484	 *
485	 * Or
486	 *
487	 * Map unmapped holes left from a previous map attempt (e.g.,
488	 * because those foreign frames were previously paged out).
489	 */
490	if (vma->vm_private_data == NULL) {
491		if (m.addr != vma->vm_start ||
492		    m.addr + (nr_pages << PAGE_SHIFT) != vma->vm_end) {
493			ret = -EINVAL;
494			goto out_unlock;
495		}
496		if (xen_feature(XENFEAT_auto_translated_physmap)) {
497			ret = alloc_empty_pages(vma, nr_pages);
498			if (ret < 0)
499				goto out_unlock;
500		} else
501			vma->vm_private_data = PRIV_VMA_LOCKED;
502	} else {
503		if (m.addr < vma->vm_start ||
504		    m.addr + (nr_pages << PAGE_SHIFT) > vma->vm_end) {
505			ret = -EINVAL;
506			goto out_unlock;
507		}
508		if (privcmd_vma_range_is_mapped(vma, m.addr, nr_pages)) {
509			ret = -EINVAL;
510			goto out_unlock;
511		}
512	}
513
514	state.domain        = m.dom;
515	state.vma           = vma;
516	state.va            = m.addr;
517	state.index         = 0;
518	state.global_error  = 0;
519	state.version       = version;
520
521	BUILD_BUG_ON(((PAGE_SIZE / sizeof(xen_pfn_t)) % XEN_PFN_PER_PAGE) != 0);
522	/* mmap_batch_fn guarantees ret == 0 */
523	BUG_ON(traverse_pages_block(m.num, sizeof(xen_pfn_t),
524				    &pagelist, mmap_batch_fn, &state));
525
526	up_write(&mm->mmap_sem);
527
528	if (state.global_error) {
529		/* Write back errors in second pass. */
530		state.user_gfn = (xen_pfn_t *)m.arr;
531		state.user_err = m.err;
532		ret = traverse_pages_block(m.num, sizeof(xen_pfn_t),
533					   &pagelist, mmap_return_errors, &state);
534	} else
535		ret = 0;
536
537	/* If we have not had any EFAULT-like global errors then set the global
538	 * error to -ENOENT if necessary. */
539	if ((ret == 0) && (state.global_error == -ENOENT))
540		ret = -ENOENT;
541
542out:
543	free_page_list(&pagelist);
544	return ret;
545
546out_unlock:
547	up_write(&mm->mmap_sem);
548	goto out;
549}
550
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
551static long privcmd_ioctl(struct file *file,
552			  unsigned int cmd, unsigned long data)
553{
554	int ret = -ENOSYS;
555	void __user *udata = (void __user *) data;
556
557	switch (cmd) {
558	case IOCTL_PRIVCMD_HYPERCALL:
559		ret = privcmd_ioctl_hypercall(udata);
560		break;
561
562	case IOCTL_PRIVCMD_MMAP:
563		ret = privcmd_ioctl_mmap(udata);
564		break;
565
566	case IOCTL_PRIVCMD_MMAPBATCH:
567		ret = privcmd_ioctl_mmap_batch(udata, 1);
568		break;
569
570	case IOCTL_PRIVCMD_MMAPBATCH_V2:
571		ret = privcmd_ioctl_mmap_batch(udata, 2);
 
 
 
 
 
 
 
 
 
 
 
 
572		break;
573
574	default:
575		ret = -EINVAL;
576		break;
577	}
578
579	return ret;
580}
581
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
582static void privcmd_close(struct vm_area_struct *vma)
583{
584	struct page **pages = vma->vm_private_data;
585	int numpgs = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
586	int numgfns = (vma->vm_end - vma->vm_start) >> XEN_PAGE_SHIFT;
587	int rc;
588
589	if (!xen_feature(XENFEAT_auto_translated_physmap) || !numpgs || !pages)
590		return;
591
592	rc = xen_unmap_domain_gfn_range(vma, numgfns, pages);
593	if (rc == 0)
594		free_xenballooned_pages(numpgs, pages);
595	else
596		pr_crit("unable to unmap MFN range: leaking %d pages. rc=%d\n",
597			numpgs, rc);
598	kfree(pages);
599}
600
601static int privcmd_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
602{
603	printk(KERN_DEBUG "privcmd_fault: vma=%p %lx-%lx, pgoff=%lx, uv=%p\n",
604	       vma, vma->vm_start, vma->vm_end,
605	       vmf->pgoff, vmf->virtual_address);
606
607	return VM_FAULT_SIGBUS;
608}
609
610static const struct vm_operations_struct privcmd_vm_ops = {
611	.close = privcmd_close,
612	.fault = privcmd_fault
613};
614
615static int privcmd_mmap(struct file *file, struct vm_area_struct *vma)
616{
617	/* DONTCOPY is essential for Xen because copy_page_range doesn't know
618	 * how to recreate these mappings */
619	vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTCOPY |
620			 VM_DONTEXPAND | VM_DONTDUMP;
621	vma->vm_ops = &privcmd_vm_ops;
622	vma->vm_private_data = NULL;
623
624	return 0;
625}
626
627/*
628 * For MMAPBATCH*. This allows asserting the singleshot mapping
629 * on a per pfn/pte basis. Mapping calls that fail with ENOENT
630 * can be then retried until success.
631 */
632static int is_mapped_fn(pte_t *pte, struct page *pmd_page,
633	                unsigned long addr, void *data)
634{
635	return pte_none(*pte) ? 0 : -EBUSY;
636}
637
638static int privcmd_vma_range_is_mapped(
639	           struct vm_area_struct *vma,
640	           unsigned long addr,
641	           unsigned long nr_pages)
642{
643	return apply_to_page_range(vma->vm_mm, addr, nr_pages << PAGE_SHIFT,
644				   is_mapped_fn, NULL) != 0;
645}
646
647const struct file_operations xen_privcmd_fops = {
648	.owner = THIS_MODULE,
649	.unlocked_ioctl = privcmd_ioctl,
 
 
650	.mmap = privcmd_mmap,
651};
652EXPORT_SYMBOL_GPL(xen_privcmd_fops);
653
654static struct miscdevice privcmd_dev = {
655	.minor = MISC_DYNAMIC_MINOR,
656	.name = "xen/privcmd",
657	.fops = &xen_privcmd_fops,
658};
659
660static int __init privcmd_init(void)
661{
662	int err;
663
664	if (!xen_domain())
665		return -ENODEV;
666
667	err = misc_register(&privcmd_dev);
668	if (err != 0) {
669		pr_err("Could not register Xen privcmd device\n");
670		return err;
671	}
 
 
 
 
 
 
 
 
672	return 0;
673}
674
675static void __exit privcmd_exit(void)
676{
677	misc_deregister(&privcmd_dev);
 
678}
679
680module_init(privcmd_init);
681module_exit(privcmd_exit);