Linux Audio

Check our new training course

Loading...
v5.9
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Helpers for the host side of a virtio ring.
   4 *
   5 * Since these may be in userspace, we use (inline) accessors.
   6 */
   7#include <linux/compiler.h>
   8#include <linux/module.h>
   9#include <linux/vringh.h>
  10#include <linux/virtio_ring.h>
  11#include <linux/kernel.h>
  12#include <linux/ratelimit.h>
  13#include <linux/uaccess.h>
  14#include <linux/slab.h>
  15#include <linux/export.h>
  16#if IS_REACHABLE(CONFIG_VHOST_IOTLB)
  17#include <linux/bvec.h>
  18#include <linux/highmem.h>
  19#include <linux/vhost_iotlb.h>
  20#endif
  21#include <uapi/linux/virtio_config.h>
  22
  23static __printf(1,2) __cold void vringh_bad(const char *fmt, ...)
  24{
  25	static DEFINE_RATELIMIT_STATE(vringh_rs,
  26				      DEFAULT_RATELIMIT_INTERVAL,
  27				      DEFAULT_RATELIMIT_BURST);
  28	if (__ratelimit(&vringh_rs)) {
  29		va_list ap;
  30		va_start(ap, fmt);
  31		printk(KERN_NOTICE "vringh:");
  32		vprintk(fmt, ap);
  33		va_end(ap);
  34	}
  35}
  36
  37/* Returns vring->num if empty, -ve on error. */
  38static inline int __vringh_get_head(const struct vringh *vrh,
  39				    int (*getu16)(const struct vringh *vrh,
  40						  u16 *val, const __virtio16 *p),
  41				    u16 *last_avail_idx)
  42{
  43	u16 avail_idx, i, head;
  44	int err;
  45
  46	err = getu16(vrh, &avail_idx, &vrh->vring.avail->idx);
  47	if (err) {
  48		vringh_bad("Failed to access avail idx at %p",
  49			   &vrh->vring.avail->idx);
  50		return err;
  51	}
  52
  53	if (*last_avail_idx == avail_idx)
  54		return vrh->vring.num;
  55
  56	/* Only get avail ring entries after they have been exposed by guest. */
  57	virtio_rmb(vrh->weak_barriers);
  58
  59	i = *last_avail_idx & (vrh->vring.num - 1);
  60
  61	err = getu16(vrh, &head, &vrh->vring.avail->ring[i]);
  62	if (err) {
  63		vringh_bad("Failed to read head: idx %d address %p",
  64			   *last_avail_idx, &vrh->vring.avail->ring[i]);
  65		return err;
  66	}
  67
  68	if (head >= vrh->vring.num) {
  69		vringh_bad("Guest says index %u > %u is available",
  70			   head, vrh->vring.num);
  71		return -EINVAL;
  72	}
  73
  74	(*last_avail_idx)++;
  75	return head;
  76}
  77
  78/* Copy some bytes to/from the iovec.  Returns num copied. */
  79static inline ssize_t vringh_iov_xfer(struct vringh *vrh,
  80				      struct vringh_kiov *iov,
  81				      void *ptr, size_t len,
  82				      int (*xfer)(const struct vringh *vrh,
  83						  void *addr, void *ptr,
  84						  size_t len))
  85{
  86	int err, done = 0;
  87
  88	while (len && iov->i < iov->used) {
  89		size_t partlen;
  90
  91		partlen = min(iov->iov[iov->i].iov_len, len);
  92		err = xfer(vrh, iov->iov[iov->i].iov_base, ptr, partlen);
  93		if (err)
  94			return err;
  95		done += partlen;
  96		len -= partlen;
  97		ptr += partlen;
  98		iov->consumed += partlen;
  99		iov->iov[iov->i].iov_len -= partlen;
 100		iov->iov[iov->i].iov_base += partlen;
 101
 102		if (!iov->iov[iov->i].iov_len) {
 103			/* Fix up old iov element then increment. */
 104			iov->iov[iov->i].iov_len = iov->consumed;
 105			iov->iov[iov->i].iov_base -= iov->consumed;
 106
 107			
 108			iov->consumed = 0;
 109			iov->i++;
 110		}
 111	}
 112	return done;
 113}
 114
 115/* May reduce *len if range is shorter. */
 116static inline bool range_check(struct vringh *vrh, u64 addr, size_t *len,
 117			       struct vringh_range *range,
 118			       bool (*getrange)(struct vringh *,
 119						u64, struct vringh_range *))
 120{
 121	if (addr < range->start || addr > range->end_incl) {
 122		if (!getrange(vrh, addr, range))
 123			return false;
 124	}
 125	BUG_ON(addr < range->start || addr > range->end_incl);
 126
 127	/* To end of memory? */
 128	if (unlikely(addr + *len == 0)) {
 129		if (range->end_incl == -1ULL)
 130			return true;
 131		goto truncate;
 132	}
 133
 134	/* Otherwise, don't wrap. */
 135	if (addr + *len < addr) {
 136		vringh_bad("Wrapping descriptor %zu@0x%llx",
 137			   *len, (unsigned long long)addr);
 138		return false;
 139	}
 140
 141	if (unlikely(addr + *len - 1 > range->end_incl))
 142		goto truncate;
 143	return true;
 144
 145truncate:
 146	*len = range->end_incl + 1 - addr;
 147	return true;
 148}
 149
 150static inline bool no_range_check(struct vringh *vrh, u64 addr, size_t *len,
 151				  struct vringh_range *range,
 152				  bool (*getrange)(struct vringh *,
 153						   u64, struct vringh_range *))
 154{
 155	return true;
 156}
 157
 158/* No reason for this code to be inline. */
 159static int move_to_indirect(const struct vringh *vrh,
 160			    int *up_next, u16 *i, void *addr,
 161			    const struct vring_desc *desc,
 162			    struct vring_desc **descs, int *desc_max)
 163{
 164	u32 len;
 165
 166	/* Indirect tables can't have indirect. */
 167	if (*up_next != -1) {
 168		vringh_bad("Multilevel indirect %u->%u", *up_next, *i);
 169		return -EINVAL;
 170	}
 171
 172	len = vringh32_to_cpu(vrh, desc->len);
 173	if (unlikely(len % sizeof(struct vring_desc))) {
 174		vringh_bad("Strange indirect len %u", desc->len);
 175		return -EINVAL;
 176	}
 177
 178	/* We will check this when we follow it! */
 179	if (desc->flags & cpu_to_vringh16(vrh, VRING_DESC_F_NEXT))
 180		*up_next = vringh16_to_cpu(vrh, desc->next);
 181	else
 182		*up_next = -2;
 183	*descs = addr;
 184	*desc_max = len / sizeof(struct vring_desc);
 185
 186	/* Now, start at the first indirect. */
 187	*i = 0;
 188	return 0;
 189}
 190
 191static int resize_iovec(struct vringh_kiov *iov, gfp_t gfp)
 192{
 193	struct kvec *new;
 194	unsigned int flag, new_num = (iov->max_num & ~VRINGH_IOV_ALLOCATED) * 2;
 195
 196	if (new_num < 8)
 197		new_num = 8;
 198
 199	flag = (iov->max_num & VRINGH_IOV_ALLOCATED);
 200	if (flag)
 201		new = krealloc(iov->iov, new_num * sizeof(struct iovec), gfp);
 202	else {
 203		new = kmalloc_array(new_num, sizeof(struct iovec), gfp);
 204		if (new) {
 205			memcpy(new, iov->iov,
 206			       iov->max_num * sizeof(struct iovec));
 207			flag = VRINGH_IOV_ALLOCATED;
 208		}
 209	}
 210	if (!new)
 211		return -ENOMEM;
 212	iov->iov = new;
 213	iov->max_num = (new_num | flag);
 214	return 0;
 215}
 216
 217static u16 __cold return_from_indirect(const struct vringh *vrh, int *up_next,
 218				       struct vring_desc **descs, int *desc_max)
 219{
 220	u16 i = *up_next;
 221
 222	*up_next = -1;
 223	*descs = vrh->vring.desc;
 224	*desc_max = vrh->vring.num;
 225	return i;
 226}
 227
 228static int slow_copy(struct vringh *vrh, void *dst, const void *src,
 229		     bool (*rcheck)(struct vringh *vrh, u64 addr, size_t *len,
 230				    struct vringh_range *range,
 231				    bool (*getrange)(struct vringh *vrh,
 232						     u64,
 233						     struct vringh_range *)),
 234		     bool (*getrange)(struct vringh *vrh,
 235				      u64 addr,
 236				      struct vringh_range *r),
 237		     struct vringh_range *range,
 238		     int (*copy)(const struct vringh *vrh,
 239				 void *dst, const void *src, size_t len))
 240{
 241	size_t part, len = sizeof(struct vring_desc);
 242
 243	do {
 244		u64 addr;
 245		int err;
 246
 247		part = len;
 248		addr = (u64)(unsigned long)src - range->offset;
 249
 250		if (!rcheck(vrh, addr, &part, range, getrange))
 251			return -EINVAL;
 252
 253		err = copy(vrh, dst, src, part);
 254		if (err)
 255			return err;
 256
 257		dst += part;
 258		src += part;
 259		len -= part;
 260	} while (len);
 261	return 0;
 262}
 263
 264static inline int
 265__vringh_iov(struct vringh *vrh, u16 i,
 266	     struct vringh_kiov *riov,
 267	     struct vringh_kiov *wiov,
 268	     bool (*rcheck)(struct vringh *vrh, u64 addr, size_t *len,
 269			    struct vringh_range *range,
 270			    bool (*getrange)(struct vringh *, u64,
 271					     struct vringh_range *)),
 272	     bool (*getrange)(struct vringh *, u64, struct vringh_range *),
 273	     gfp_t gfp,
 274	     int (*copy)(const struct vringh *vrh,
 275			 void *dst, const void *src, size_t len))
 276{
 277	int err, count = 0, up_next, desc_max;
 278	struct vring_desc desc, *descs;
 279	struct vringh_range range = { -1ULL, 0 }, slowrange;
 280	bool slow = false;
 281
 282	/* We start traversing vring's descriptor table. */
 283	descs = vrh->vring.desc;
 284	desc_max = vrh->vring.num;
 285	up_next = -1;
 286
 287	if (riov)
 288		riov->i = riov->used = 0;
 289	else if (wiov)
 290		wiov->i = wiov->used = 0;
 291	else
 292		/* You must want something! */
 293		BUG();
 294
 295	for (;;) {
 296		void *addr;
 297		struct vringh_kiov *iov;
 298		size_t len;
 299
 300		if (unlikely(slow))
 301			err = slow_copy(vrh, &desc, &descs[i], rcheck, getrange,
 302					&slowrange, copy);
 303		else
 304			err = copy(vrh, &desc, &descs[i], sizeof(desc));
 305		if (unlikely(err))
 306			goto fail;
 307
 308		if (unlikely(desc.flags &
 309			     cpu_to_vringh16(vrh, VRING_DESC_F_INDIRECT))) {
 310			u64 a = vringh64_to_cpu(vrh, desc.addr);
 311
 312			/* Make sure it's OK, and get offset. */
 313			len = vringh32_to_cpu(vrh, desc.len);
 314			if (!rcheck(vrh, a, &len, &range, getrange)) {
 315				err = -EINVAL;
 316				goto fail;
 317			}
 318
 319			if (unlikely(len != vringh32_to_cpu(vrh, desc.len))) {
 320				slow = true;
 321				/* We need to save this range to use offset */
 322				slowrange = range;
 323			}
 324
 325			addr = (void *)(long)(a + range.offset);
 326			err = move_to_indirect(vrh, &up_next, &i, addr, &desc,
 327					       &descs, &desc_max);
 328			if (err)
 329				goto fail;
 330			continue;
 331		}
 332
 333		if (count++ == vrh->vring.num) {
 334			vringh_bad("Descriptor loop in %p", descs);
 335			err = -ELOOP;
 336			goto fail;
 337		}
 338
 339		if (desc.flags & cpu_to_vringh16(vrh, VRING_DESC_F_WRITE))
 340			iov = wiov;
 341		else {
 342			iov = riov;
 343			if (unlikely(wiov && wiov->i)) {
 344				vringh_bad("Readable desc %p after writable",
 345					   &descs[i]);
 346				err = -EINVAL;
 347				goto fail;
 348			}
 349		}
 350
 351		if (!iov) {
 352			vringh_bad("Unexpected %s desc",
 353				   !wiov ? "writable" : "readable");
 354			err = -EPROTO;
 355			goto fail;
 356		}
 357
 358	again:
 359		/* Make sure it's OK, and get offset. */
 360		len = vringh32_to_cpu(vrh, desc.len);
 361		if (!rcheck(vrh, vringh64_to_cpu(vrh, desc.addr), &len, &range,
 362			    getrange)) {
 363			err = -EINVAL;
 364			goto fail;
 365		}
 366		addr = (void *)(unsigned long)(vringh64_to_cpu(vrh, desc.addr) +
 367					       range.offset);
 368
 369		if (unlikely(iov->used == (iov->max_num & ~VRINGH_IOV_ALLOCATED))) {
 370			err = resize_iovec(iov, gfp);
 371			if (err)
 372				goto fail;
 373		}
 374
 375		iov->iov[iov->used].iov_base = addr;
 376		iov->iov[iov->used].iov_len = len;
 377		iov->used++;
 378
 379		if (unlikely(len != vringh32_to_cpu(vrh, desc.len))) {
 380			desc.len = cpu_to_vringh32(vrh,
 381				   vringh32_to_cpu(vrh, desc.len) - len);
 382			desc.addr = cpu_to_vringh64(vrh,
 383				    vringh64_to_cpu(vrh, desc.addr) + len);
 384			goto again;
 385		}
 386
 387		if (desc.flags & cpu_to_vringh16(vrh, VRING_DESC_F_NEXT)) {
 388			i = vringh16_to_cpu(vrh, desc.next);
 389		} else {
 390			/* Just in case we need to finish traversing above. */
 391			if (unlikely(up_next > 0)) {
 392				i = return_from_indirect(vrh, &up_next,
 393							 &descs, &desc_max);
 394				slow = false;
 395			} else
 396				break;
 397		}
 398
 399		if (i >= desc_max) {
 400			vringh_bad("Chained index %u > %u", i, desc_max);
 401			err = -EINVAL;
 402			goto fail;
 403		}
 404	}
 405
 406	return 0;
 407
 408fail:
 409	return err;
 410}
 411
 412static inline int __vringh_complete(struct vringh *vrh,
 413				    const struct vring_used_elem *used,
 414				    unsigned int num_used,
 415				    int (*putu16)(const struct vringh *vrh,
 416						  __virtio16 *p, u16 val),
 417				    int (*putused)(const struct vringh *vrh,
 418						   struct vring_used_elem *dst,
 419						   const struct vring_used_elem
 420						   *src, unsigned num))
 421{
 422	struct vring_used *used_ring;
 423	int err;
 424	u16 used_idx, off;
 425
 426	used_ring = vrh->vring.used;
 427	used_idx = vrh->last_used_idx + vrh->completed;
 428
 429	off = used_idx % vrh->vring.num;
 430
 431	/* Compiler knows num_used == 1 sometimes, hence extra check */
 432	if (num_used > 1 && unlikely(off + num_used >= vrh->vring.num)) {
 433		u16 part = vrh->vring.num - off;
 434		err = putused(vrh, &used_ring->ring[off], used, part);
 435		if (!err)
 436			err = putused(vrh, &used_ring->ring[0], used + part,
 437				      num_used - part);
 438	} else
 439		err = putused(vrh, &used_ring->ring[off], used, num_used);
 440
 441	if (err) {
 442		vringh_bad("Failed to write %u used entries %u at %p",
 443			   num_used, off, &used_ring->ring[off]);
 444		return err;
 445	}
 446
 447	/* Make sure buffer is written before we update index. */
 448	virtio_wmb(vrh->weak_barriers);
 449
 450	err = putu16(vrh, &vrh->vring.used->idx, used_idx + num_used);
 451	if (err) {
 452		vringh_bad("Failed to update used index at %p",
 453			   &vrh->vring.used->idx);
 454		return err;
 455	}
 456
 457	vrh->completed += num_used;
 458	return 0;
 459}
 460
 461
 462static inline int __vringh_need_notify(struct vringh *vrh,
 463				       int (*getu16)(const struct vringh *vrh,
 464						     u16 *val,
 465						     const __virtio16 *p))
 466{
 467	bool notify;
 468	u16 used_event;
 469	int err;
 470
 471	/* Flush out used index update. This is paired with the
 472	 * barrier that the Guest executes when enabling
 473	 * interrupts. */
 474	virtio_mb(vrh->weak_barriers);
 475
 476	/* Old-style, without event indices. */
 477	if (!vrh->event_indices) {
 478		u16 flags;
 479		err = getu16(vrh, &flags, &vrh->vring.avail->flags);
 480		if (err) {
 481			vringh_bad("Failed to get flags at %p",
 482				   &vrh->vring.avail->flags);
 483			return err;
 484		}
 485		return (!(flags & VRING_AVAIL_F_NO_INTERRUPT));
 486	}
 487
 488	/* Modern: we know when other side wants to know. */
 489	err = getu16(vrh, &used_event, &vring_used_event(&vrh->vring));
 490	if (err) {
 491		vringh_bad("Failed to get used event idx at %p",
 492			   &vring_used_event(&vrh->vring));
 493		return err;
 494	}
 495
 496	/* Just in case we added so many that we wrap. */
 497	if (unlikely(vrh->completed > 0xffff))
 498		notify = true;
 499	else
 500		notify = vring_need_event(used_event,
 501					  vrh->last_used_idx + vrh->completed,
 502					  vrh->last_used_idx);
 503
 504	vrh->last_used_idx += vrh->completed;
 505	vrh->completed = 0;
 506	return notify;
 507}
 508
 509static inline bool __vringh_notify_enable(struct vringh *vrh,
 510					  int (*getu16)(const struct vringh *vrh,
 511							u16 *val, const __virtio16 *p),
 512					  int (*putu16)(const struct vringh *vrh,
 513							__virtio16 *p, u16 val))
 514{
 515	u16 avail;
 516
 517	if (!vrh->event_indices) {
 518		/* Old-school; update flags. */
 519		if (putu16(vrh, &vrh->vring.used->flags, 0) != 0) {
 520			vringh_bad("Clearing used flags %p",
 521				   &vrh->vring.used->flags);
 522			return true;
 523		}
 524	} else {
 525		if (putu16(vrh, &vring_avail_event(&vrh->vring),
 526			   vrh->last_avail_idx) != 0) {
 527			vringh_bad("Updating avail event index %p",
 528				   &vring_avail_event(&vrh->vring));
 529			return true;
 530		}
 531	}
 532
 533	/* They could have slipped one in as we were doing that: make
 534	 * sure it's written, then check again. */
 535	virtio_mb(vrh->weak_barriers);
 536
 537	if (getu16(vrh, &avail, &vrh->vring.avail->idx) != 0) {
 538		vringh_bad("Failed to check avail idx at %p",
 539			   &vrh->vring.avail->idx);
 540		return true;
 541	}
 542
 543	/* This is unlikely, so we just leave notifications enabled
 544	 * (if we're using event_indices, we'll only get one
 545	 * notification anyway). */
 546	return avail == vrh->last_avail_idx;
 547}
 548
 549static inline void __vringh_notify_disable(struct vringh *vrh,
 550					   int (*putu16)(const struct vringh *vrh,
 551							 __virtio16 *p, u16 val))
 552{
 553	if (!vrh->event_indices) {
 554		/* Old-school; update flags. */
 555		if (putu16(vrh, &vrh->vring.used->flags,
 556			   VRING_USED_F_NO_NOTIFY)) {
 557			vringh_bad("Setting used flags %p",
 558				   &vrh->vring.used->flags);
 559		}
 560	}
 561}
 562
 563/* Userspace access helpers: in this case, addresses are really userspace. */
 564static inline int getu16_user(const struct vringh *vrh, u16 *val, const __virtio16 *p)
 565{
 566	__virtio16 v = 0;
 567	int rc = get_user(v, (__force __virtio16 __user *)p);
 568	*val = vringh16_to_cpu(vrh, v);
 569	return rc;
 570}
 571
 572static inline int putu16_user(const struct vringh *vrh, __virtio16 *p, u16 val)
 573{
 574	__virtio16 v = cpu_to_vringh16(vrh, val);
 575	return put_user(v, (__force __virtio16 __user *)p);
 576}
 577
 578static inline int copydesc_user(const struct vringh *vrh,
 579				void *dst, const void *src, size_t len)
 580{
 581	return copy_from_user(dst, (__force void __user *)src, len) ?
 582		-EFAULT : 0;
 583}
 584
 585static inline int putused_user(const struct vringh *vrh,
 586			       struct vring_used_elem *dst,
 587			       const struct vring_used_elem *src,
 588			       unsigned int num)
 589{
 590	return copy_to_user((__force void __user *)dst, src,
 591			    sizeof(*dst) * num) ? -EFAULT : 0;
 592}
 593
 594static inline int xfer_from_user(const struct vringh *vrh, void *src,
 595				 void *dst, size_t len)
 596{
 597	return copy_from_user(dst, (__force void __user *)src, len) ?
 598		-EFAULT : 0;
 599}
 600
 601static inline int xfer_to_user(const struct vringh *vrh,
 602			       void *dst, void *src, size_t len)
 603{
 604	return copy_to_user((__force void __user *)dst, src, len) ?
 605		-EFAULT : 0;
 606}
 607
 608/**
 609 * vringh_init_user - initialize a vringh for a userspace vring.
 610 * @vrh: the vringh to initialize.
 611 * @features: the feature bits for this ring.
 612 * @num: the number of elements.
 613 * @weak_barriers: true if we only need memory barriers, not I/O.
 614 * @desc: the userpace descriptor pointer.
 615 * @avail: the userpace avail pointer.
 616 * @used: the userpace used pointer.
 617 *
 618 * Returns an error if num is invalid: you should check pointers
 619 * yourself!
 620 */
 621int vringh_init_user(struct vringh *vrh, u64 features,
 622		     unsigned int num, bool weak_barriers,
 623		     vring_desc_t __user *desc,
 624		     vring_avail_t __user *avail,
 625		     vring_used_t __user *used)
 626{
 627	/* Sane power of 2 please! */
 628	if (!num || num > 0xffff || (num & (num - 1))) {
 629		vringh_bad("Bad ring size %u", num);
 630		return -EINVAL;
 631	}
 632
 633	vrh->little_endian = (features & (1ULL << VIRTIO_F_VERSION_1));
 634	vrh->event_indices = (features & (1 << VIRTIO_RING_F_EVENT_IDX));
 635	vrh->weak_barriers = weak_barriers;
 636	vrh->completed = 0;
 637	vrh->last_avail_idx = 0;
 638	vrh->last_used_idx = 0;
 639	vrh->vring.num = num;
 640	/* vring expects kernel addresses, but only used via accessors. */
 641	vrh->vring.desc = (__force struct vring_desc *)desc;
 642	vrh->vring.avail = (__force struct vring_avail *)avail;
 643	vrh->vring.used = (__force struct vring_used *)used;
 644	return 0;
 645}
 646EXPORT_SYMBOL(vringh_init_user);
 647
 648/**
 649 * vringh_getdesc_user - get next available descriptor from userspace ring.
 650 * @vrh: the userspace vring.
 651 * @riov: where to put the readable descriptors (or NULL)
 652 * @wiov: where to put the writable descriptors (or NULL)
 653 * @getrange: function to call to check ranges.
 654 * @head: head index we received, for passing to vringh_complete_user().
 655 *
 656 * Returns 0 if there was no descriptor, 1 if there was, or -errno.
 657 *
 658 * Note that on error return, you can tell the difference between an
 659 * invalid ring and a single invalid descriptor: in the former case,
 660 * *head will be vrh->vring.num.  You may be able to ignore an invalid
 661 * descriptor, but there's not much you can do with an invalid ring.
 662 *
 663 * Note that you may need to clean up riov and wiov, even on error!
 664 */
 665int vringh_getdesc_user(struct vringh *vrh,
 666			struct vringh_iov *riov,
 667			struct vringh_iov *wiov,
 668			bool (*getrange)(struct vringh *vrh,
 669					 u64 addr, struct vringh_range *r),
 670			u16 *head)
 671{
 672	int err;
 673
 674	*head = vrh->vring.num;
 675	err = __vringh_get_head(vrh, getu16_user, &vrh->last_avail_idx);
 676	if (err < 0)
 677		return err;
 678
 679	/* Empty... */
 680	if (err == vrh->vring.num)
 681		return 0;
 682
 683	/* We need the layouts to be the identical for this to work */
 684	BUILD_BUG_ON(sizeof(struct vringh_kiov) != sizeof(struct vringh_iov));
 685	BUILD_BUG_ON(offsetof(struct vringh_kiov, iov) !=
 686		     offsetof(struct vringh_iov, iov));
 687	BUILD_BUG_ON(offsetof(struct vringh_kiov, i) !=
 688		     offsetof(struct vringh_iov, i));
 689	BUILD_BUG_ON(offsetof(struct vringh_kiov, used) !=
 690		     offsetof(struct vringh_iov, used));
 691	BUILD_BUG_ON(offsetof(struct vringh_kiov, max_num) !=
 692		     offsetof(struct vringh_iov, max_num));
 693	BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
 694	BUILD_BUG_ON(offsetof(struct iovec, iov_base) !=
 695		     offsetof(struct kvec, iov_base));
 696	BUILD_BUG_ON(offsetof(struct iovec, iov_len) !=
 697		     offsetof(struct kvec, iov_len));
 698	BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_base)
 699		     != sizeof(((struct kvec *)NULL)->iov_base));
 700	BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_len)
 701		     != sizeof(((struct kvec *)NULL)->iov_len));
 702
 703	*head = err;
 704	err = __vringh_iov(vrh, *head, (struct vringh_kiov *)riov,
 705			   (struct vringh_kiov *)wiov,
 706			   range_check, getrange, GFP_KERNEL, copydesc_user);
 707	if (err)
 708		return err;
 709
 710	return 1;
 711}
 712EXPORT_SYMBOL(vringh_getdesc_user);
 713
 714/**
 715 * vringh_iov_pull_user - copy bytes from vring_iov.
 716 * @riov: the riov as passed to vringh_getdesc_user() (updated as we consume)
 717 * @dst: the place to copy.
 718 * @len: the maximum length to copy.
 719 *
 720 * Returns the bytes copied <= len or a negative errno.
 721 */
 722ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len)
 723{
 724	return vringh_iov_xfer(NULL, (struct vringh_kiov *)riov,
 725			       dst, len, xfer_from_user);
 726}
 727EXPORT_SYMBOL(vringh_iov_pull_user);
 728
 729/**
 730 * vringh_iov_push_user - copy bytes into vring_iov.
 731 * @wiov: the wiov as passed to vringh_getdesc_user() (updated as we consume)
 732 * @dst: the place to copy.
 733 * @len: the maximum length to copy.
 734 *
 735 * Returns the bytes copied <= len or a negative errno.
 736 */
 737ssize_t vringh_iov_push_user(struct vringh_iov *wiov,
 738			     const void *src, size_t len)
 739{
 740	return vringh_iov_xfer(NULL, (struct vringh_kiov *)wiov,
 741			       (void *)src, len, xfer_to_user);
 742}
 743EXPORT_SYMBOL(vringh_iov_push_user);
 744
 745/**
 746 * vringh_abandon_user - we've decided not to handle the descriptor(s).
 747 * @vrh: the vring.
 748 * @num: the number of descriptors to put back (ie. num
 749 *	 vringh_get_user() to undo).
 750 *
 751 * The next vringh_get_user() will return the old descriptor(s) again.
 752 */
 753void vringh_abandon_user(struct vringh *vrh, unsigned int num)
 754{
 755	/* We only update vring_avail_event(vr) when we want to be notified,
 756	 * so we haven't changed that yet. */
 757	vrh->last_avail_idx -= num;
 758}
 759EXPORT_SYMBOL(vringh_abandon_user);
 760
 761/**
 762 * vringh_complete_user - we've finished with descriptor, publish it.
 763 * @vrh: the vring.
 764 * @head: the head as filled in by vringh_getdesc_user.
 765 * @len: the length of data we have written.
 766 *
 767 * You should check vringh_need_notify_user() after one or more calls
 768 * to this function.
 769 */
 770int vringh_complete_user(struct vringh *vrh, u16 head, u32 len)
 771{
 772	struct vring_used_elem used;
 773
 774	used.id = cpu_to_vringh32(vrh, head);
 775	used.len = cpu_to_vringh32(vrh, len);
 776	return __vringh_complete(vrh, &used, 1, putu16_user, putused_user);
 777}
 778EXPORT_SYMBOL(vringh_complete_user);
 779
 780/**
 781 * vringh_complete_multi_user - we've finished with many descriptors.
 782 * @vrh: the vring.
 783 * @used: the head, length pairs.
 784 * @num_used: the number of used elements.
 785 *
 786 * You should check vringh_need_notify_user() after one or more calls
 787 * to this function.
 788 */
 789int vringh_complete_multi_user(struct vringh *vrh,
 790			       const struct vring_used_elem used[],
 791			       unsigned num_used)
 792{
 793	return __vringh_complete(vrh, used, num_used,
 794				 putu16_user, putused_user);
 795}
 796EXPORT_SYMBOL(vringh_complete_multi_user);
 797
 798/**
 799 * vringh_notify_enable_user - we want to know if something changes.
 800 * @vrh: the vring.
 801 *
 802 * This always enables notifications, but returns false if there are
 803 * now more buffers available in the vring.
 804 */
 805bool vringh_notify_enable_user(struct vringh *vrh)
 806{
 807	return __vringh_notify_enable(vrh, getu16_user, putu16_user);
 808}
 809EXPORT_SYMBOL(vringh_notify_enable_user);
 810
 811/**
 812 * vringh_notify_disable_user - don't tell us if something changes.
 813 * @vrh: the vring.
 814 *
 815 * This is our normal running state: we disable and then only enable when
 816 * we're going to sleep.
 817 */
 818void vringh_notify_disable_user(struct vringh *vrh)
 819{
 820	__vringh_notify_disable(vrh, putu16_user);
 821}
 822EXPORT_SYMBOL(vringh_notify_disable_user);
 823
 824/**
 825 * vringh_need_notify_user - must we tell the other side about used buffers?
 826 * @vrh: the vring we've called vringh_complete_user() on.
 827 *
 828 * Returns -errno or 0 if we don't need to tell the other side, 1 if we do.
 829 */
 830int vringh_need_notify_user(struct vringh *vrh)
 831{
 832	return __vringh_need_notify(vrh, getu16_user);
 833}
 834EXPORT_SYMBOL(vringh_need_notify_user);
 835
 836/* Kernelspace access helpers. */
 837static inline int getu16_kern(const struct vringh *vrh,
 838			      u16 *val, const __virtio16 *p)
 839{
 840	*val = vringh16_to_cpu(vrh, READ_ONCE(*p));
 841	return 0;
 842}
 843
 844static inline int putu16_kern(const struct vringh *vrh, __virtio16 *p, u16 val)
 845{
 846	WRITE_ONCE(*p, cpu_to_vringh16(vrh, val));
 847	return 0;
 848}
 849
 850static inline int copydesc_kern(const struct vringh *vrh,
 851				void *dst, const void *src, size_t len)
 852{
 853	memcpy(dst, src, len);
 854	return 0;
 855}
 856
 857static inline int putused_kern(const struct vringh *vrh,
 858			       struct vring_used_elem *dst,
 859			       const struct vring_used_elem *src,
 860			       unsigned int num)
 861{
 862	memcpy(dst, src, num * sizeof(*dst));
 863	return 0;
 864}
 865
 866static inline int xfer_kern(const struct vringh *vrh, void *src,
 867			    void *dst, size_t len)
 868{
 869	memcpy(dst, src, len);
 870	return 0;
 871}
 872
 873static inline int kern_xfer(const struct vringh *vrh, void *dst,
 874			    void *src, size_t len)
 875{
 876	memcpy(dst, src, len);
 877	return 0;
 878}
 879
 880/**
 881 * vringh_init_kern - initialize a vringh for a kernelspace vring.
 882 * @vrh: the vringh to initialize.
 883 * @features: the feature bits for this ring.
 884 * @num: the number of elements.
 885 * @weak_barriers: true if we only need memory barriers, not I/O.
 886 * @desc: the userpace descriptor pointer.
 887 * @avail: the userpace avail pointer.
 888 * @used: the userpace used pointer.
 889 *
 890 * Returns an error if num is invalid.
 891 */
 892int vringh_init_kern(struct vringh *vrh, u64 features,
 893		     unsigned int num, bool weak_barriers,
 894		     struct vring_desc *desc,
 895		     struct vring_avail *avail,
 896		     struct vring_used *used)
 897{
 898	/* Sane power of 2 please! */
 899	if (!num || num > 0xffff || (num & (num - 1))) {
 900		vringh_bad("Bad ring size %u", num);
 901		return -EINVAL;
 902	}
 903
 904	vrh->little_endian = (features & (1ULL << VIRTIO_F_VERSION_1));
 905	vrh->event_indices = (features & (1 << VIRTIO_RING_F_EVENT_IDX));
 906	vrh->weak_barriers = weak_barriers;
 907	vrh->completed = 0;
 908	vrh->last_avail_idx = 0;
 909	vrh->last_used_idx = 0;
 910	vrh->vring.num = num;
 911	vrh->vring.desc = desc;
 912	vrh->vring.avail = avail;
 913	vrh->vring.used = used;
 914	return 0;
 915}
 916EXPORT_SYMBOL(vringh_init_kern);
 917
 918/**
 919 * vringh_getdesc_kern - get next available descriptor from kernelspace ring.
 920 * @vrh: the kernelspace vring.
 921 * @riov: where to put the readable descriptors (or NULL)
 922 * @wiov: where to put the writable descriptors (or NULL)
 923 * @head: head index we received, for passing to vringh_complete_kern().
 924 * @gfp: flags for allocating larger riov/wiov.
 925 *
 926 * Returns 0 if there was no descriptor, 1 if there was, or -errno.
 927 *
 928 * Note that on error return, you can tell the difference between an
 929 * invalid ring and a single invalid descriptor: in the former case,
 930 * *head will be vrh->vring.num.  You may be able to ignore an invalid
 931 * descriptor, but there's not much you can do with an invalid ring.
 932 *
 933 * Note that you may need to clean up riov and wiov, even on error!
 934 */
 935int vringh_getdesc_kern(struct vringh *vrh,
 936			struct vringh_kiov *riov,
 937			struct vringh_kiov *wiov,
 938			u16 *head,
 939			gfp_t gfp)
 940{
 941	int err;
 942
 943	err = __vringh_get_head(vrh, getu16_kern, &vrh->last_avail_idx);
 944	if (err < 0)
 945		return err;
 946
 947	/* Empty... */
 948	if (err == vrh->vring.num)
 949		return 0;
 950
 951	*head = err;
 952	err = __vringh_iov(vrh, *head, riov, wiov, no_range_check, NULL,
 953			   gfp, copydesc_kern);
 954	if (err)
 955		return err;
 956
 957	return 1;
 958}
 959EXPORT_SYMBOL(vringh_getdesc_kern);
 960
 961/**
 962 * vringh_iov_pull_kern - copy bytes from vring_iov.
 963 * @riov: the riov as passed to vringh_getdesc_kern() (updated as we consume)
 964 * @dst: the place to copy.
 965 * @len: the maximum length to copy.
 966 *
 967 * Returns the bytes copied <= len or a negative errno.
 968 */
 969ssize_t vringh_iov_pull_kern(struct vringh_kiov *riov, void *dst, size_t len)
 970{
 971	return vringh_iov_xfer(NULL, riov, dst, len, xfer_kern);
 972}
 973EXPORT_SYMBOL(vringh_iov_pull_kern);
 974
 975/**
 976 * vringh_iov_push_kern - copy bytes into vring_iov.
 977 * @wiov: the wiov as passed to vringh_getdesc_kern() (updated as we consume)
 978 * @dst: the place to copy.
 979 * @len: the maximum length to copy.
 980 *
 981 * Returns the bytes copied <= len or a negative errno.
 982 */
 983ssize_t vringh_iov_push_kern(struct vringh_kiov *wiov,
 984			     const void *src, size_t len)
 985{
 986	return vringh_iov_xfer(NULL, wiov, (void *)src, len, kern_xfer);
 987}
 988EXPORT_SYMBOL(vringh_iov_push_kern);
 989
 990/**
 991 * vringh_abandon_kern - we've decided not to handle the descriptor(s).
 992 * @vrh: the vring.
 993 * @num: the number of descriptors to put back (ie. num
 994 *	 vringh_get_kern() to undo).
 995 *
 996 * The next vringh_get_kern() will return the old descriptor(s) again.
 997 */
 998void vringh_abandon_kern(struct vringh *vrh, unsigned int num)
 999{
1000	/* We only update vring_avail_event(vr) when we want to be notified,
1001	 * so we haven't changed that yet. */
1002	vrh->last_avail_idx -= num;
1003}
1004EXPORT_SYMBOL(vringh_abandon_kern);
1005
1006/**
1007 * vringh_complete_kern - we've finished with descriptor, publish it.
1008 * @vrh: the vring.
1009 * @head: the head as filled in by vringh_getdesc_kern.
1010 * @len: the length of data we have written.
1011 *
1012 * You should check vringh_need_notify_kern() after one or more calls
1013 * to this function.
1014 */
1015int vringh_complete_kern(struct vringh *vrh, u16 head, u32 len)
1016{
1017	struct vring_used_elem used;
1018
1019	used.id = cpu_to_vringh32(vrh, head);
1020	used.len = cpu_to_vringh32(vrh, len);
1021
1022	return __vringh_complete(vrh, &used, 1, putu16_kern, putused_kern);
1023}
1024EXPORT_SYMBOL(vringh_complete_kern);
1025
1026/**
1027 * vringh_notify_enable_kern - we want to know if something changes.
1028 * @vrh: the vring.
1029 *
1030 * This always enables notifications, but returns false if there are
1031 * now more buffers available in the vring.
1032 */
1033bool vringh_notify_enable_kern(struct vringh *vrh)
1034{
1035	return __vringh_notify_enable(vrh, getu16_kern, putu16_kern);
1036}
1037EXPORT_SYMBOL(vringh_notify_enable_kern);
1038
1039/**
1040 * vringh_notify_disable_kern - don't tell us if something changes.
1041 * @vrh: the vring.
1042 *
1043 * This is our normal running state: we disable and then only enable when
1044 * we're going to sleep.
1045 */
1046void vringh_notify_disable_kern(struct vringh *vrh)
1047{
1048	__vringh_notify_disable(vrh, putu16_kern);
1049}
1050EXPORT_SYMBOL(vringh_notify_disable_kern);
1051
1052/**
1053 * vringh_need_notify_kern - must we tell the other side about used buffers?
1054 * @vrh: the vring we've called vringh_complete_kern() on.
1055 *
1056 * Returns -errno or 0 if we don't need to tell the other side, 1 if we do.
1057 */
1058int vringh_need_notify_kern(struct vringh *vrh)
1059{
1060	return __vringh_need_notify(vrh, getu16_kern);
1061}
1062EXPORT_SYMBOL(vringh_need_notify_kern);
1063
1064#if IS_REACHABLE(CONFIG_VHOST_IOTLB)
1065
1066static int iotlb_translate(const struct vringh *vrh,
1067			   u64 addr, u64 len, struct bio_vec iov[],
1068			   int iov_size, u32 perm)
1069{
1070	struct vhost_iotlb_map *map;
1071	struct vhost_iotlb *iotlb = vrh->iotlb;
1072	int ret = 0;
1073	u64 s = 0;
1074
1075	while (len > s) {
1076		u64 size, pa, pfn;
1077
1078		if (unlikely(ret >= iov_size)) {
1079			ret = -ENOBUFS;
1080			break;
1081		}
1082
1083		map = vhost_iotlb_itree_first(iotlb, addr,
1084					      addr + len - 1);
1085		if (!map || map->start > addr) {
1086			ret = -EINVAL;
1087			break;
1088		} else if (!(map->perm & perm)) {
1089			ret = -EPERM;
1090			break;
1091		}
1092
1093		size = map->size - addr + map->start;
1094		pa = map->addr + addr - map->start;
1095		pfn = pa >> PAGE_SHIFT;
1096		iov[ret].bv_page = pfn_to_page(pfn);
1097		iov[ret].bv_len = min(len - s, size);
1098		iov[ret].bv_offset = pa & (PAGE_SIZE - 1);
1099		s += size;
1100		addr += size;
1101		++ret;
1102	}
1103
1104	return ret;
1105}
1106
1107static inline int copy_from_iotlb(const struct vringh *vrh, void *dst,
1108				  void *src, size_t len)
1109{
1110	struct iov_iter iter;
1111	struct bio_vec iov[16];
1112	int ret;
1113
1114	ret = iotlb_translate(vrh, (u64)(uintptr_t)src,
1115			      len, iov, 16, VHOST_MAP_RO);
1116	if (ret < 0)
1117		return ret;
1118
1119	iov_iter_bvec(&iter, READ, iov, ret, len);
1120
1121	ret = copy_from_iter(dst, len, &iter);
1122
1123	return ret;
1124}
1125
1126static inline int copy_to_iotlb(const struct vringh *vrh, void *dst,
1127				void *src, size_t len)
1128{
1129	struct iov_iter iter;
1130	struct bio_vec iov[16];
1131	int ret;
1132
1133	ret = iotlb_translate(vrh, (u64)(uintptr_t)dst,
1134			      len, iov, 16, VHOST_MAP_WO);
1135	if (ret < 0)
1136		return ret;
1137
1138	iov_iter_bvec(&iter, WRITE, iov, ret, len);
1139
1140	return copy_to_iter(src, len, &iter);
1141}
1142
1143static inline int getu16_iotlb(const struct vringh *vrh,
1144			       u16 *val, const __virtio16 *p)
1145{
1146	struct bio_vec iov;
1147	void *kaddr, *from;
1148	int ret;
1149
1150	/* Atomic read is needed for getu16 */
1151	ret = iotlb_translate(vrh, (u64)(uintptr_t)p, sizeof(*p),
1152			      &iov, 1, VHOST_MAP_RO);
1153	if (ret < 0)
1154		return ret;
1155
1156	kaddr = kmap_atomic(iov.bv_page);
1157	from = kaddr + iov.bv_offset;
1158	*val = vringh16_to_cpu(vrh, READ_ONCE(*(__virtio16 *)from));
1159	kunmap_atomic(kaddr);
1160
1161	return 0;
1162}
1163
1164static inline int putu16_iotlb(const struct vringh *vrh,
1165			       __virtio16 *p, u16 val)
1166{
1167	struct bio_vec iov;
1168	void *kaddr, *to;
1169	int ret;
1170
1171	/* Atomic write is needed for putu16 */
1172	ret = iotlb_translate(vrh, (u64)(uintptr_t)p, sizeof(*p),
1173			      &iov, 1, VHOST_MAP_WO);
1174	if (ret < 0)
1175		return ret;
1176
1177	kaddr = kmap_atomic(iov.bv_page);
1178	to = kaddr + iov.bv_offset;
1179	WRITE_ONCE(*(__virtio16 *)to, cpu_to_vringh16(vrh, val));
1180	kunmap_atomic(kaddr);
1181
1182	return 0;
1183}
1184
1185static inline int copydesc_iotlb(const struct vringh *vrh,
1186				 void *dst, const void *src, size_t len)
1187{
1188	int ret;
1189
1190	ret = copy_from_iotlb(vrh, dst, (void *)src, len);
1191	if (ret != len)
1192		return -EFAULT;
1193
1194	return 0;
1195}
1196
1197static inline int xfer_from_iotlb(const struct vringh *vrh, void *src,
1198				  void *dst, size_t len)
1199{
1200	int ret;
1201
1202	ret = copy_from_iotlb(vrh, dst, src, len);
1203	if (ret != len)
1204		return -EFAULT;
1205
1206	return 0;
1207}
1208
1209static inline int xfer_to_iotlb(const struct vringh *vrh,
1210			       void *dst, void *src, size_t len)
1211{
1212	int ret;
1213
1214	ret = copy_to_iotlb(vrh, dst, src, len);
1215	if (ret != len)
1216		return -EFAULT;
1217
1218	return 0;
1219}
1220
1221static inline int putused_iotlb(const struct vringh *vrh,
1222				struct vring_used_elem *dst,
1223				const struct vring_used_elem *src,
1224				unsigned int num)
1225{
1226	int size = num * sizeof(*dst);
1227	int ret;
1228
1229	ret = copy_to_iotlb(vrh, dst, (void *)src, num * sizeof(*dst));
1230	if (ret != size)
1231		return -EFAULT;
1232
1233	return 0;
1234}
1235
1236/**
1237 * vringh_init_iotlb - initialize a vringh for a ring with IOTLB.
1238 * @vrh: the vringh to initialize.
1239 * @features: the feature bits for this ring.
1240 * @num: the number of elements.
1241 * @weak_barriers: true if we only need memory barriers, not I/O.
1242 * @desc: the userpace descriptor pointer.
1243 * @avail: the userpace avail pointer.
1244 * @used: the userpace used pointer.
1245 *
1246 * Returns an error if num is invalid.
1247 */
1248int vringh_init_iotlb(struct vringh *vrh, u64 features,
1249		      unsigned int num, bool weak_barriers,
1250		      struct vring_desc *desc,
1251		      struct vring_avail *avail,
1252		      struct vring_used *used)
1253{
1254	return vringh_init_kern(vrh, features, num, weak_barriers,
1255				desc, avail, used);
1256}
1257EXPORT_SYMBOL(vringh_init_iotlb);
1258
1259/**
1260 * vringh_set_iotlb - initialize a vringh for a ring with IOTLB.
1261 * @vrh: the vring
1262 * @iotlb: iotlb associated with this vring
1263 */
1264void vringh_set_iotlb(struct vringh *vrh, struct vhost_iotlb *iotlb)
1265{
1266	vrh->iotlb = iotlb;
1267}
1268EXPORT_SYMBOL(vringh_set_iotlb);
1269
1270/**
1271 * vringh_getdesc_iotlb - get next available descriptor from ring with
1272 * IOTLB.
1273 * @vrh: the kernelspace vring.
1274 * @riov: where to put the readable descriptors (or NULL)
1275 * @wiov: where to put the writable descriptors (or NULL)
1276 * @head: head index we received, for passing to vringh_complete_iotlb().
1277 * @gfp: flags for allocating larger riov/wiov.
1278 *
1279 * Returns 0 if there was no descriptor, 1 if there was, or -errno.
1280 *
1281 * Note that on error return, you can tell the difference between an
1282 * invalid ring and a single invalid descriptor: in the former case,
1283 * *head will be vrh->vring.num.  You may be able to ignore an invalid
1284 * descriptor, but there's not much you can do with an invalid ring.
1285 *
1286 * Note that you may need to clean up riov and wiov, even on error!
1287 */
1288int vringh_getdesc_iotlb(struct vringh *vrh,
1289			 struct vringh_kiov *riov,
1290			 struct vringh_kiov *wiov,
1291			 u16 *head,
1292			 gfp_t gfp)
1293{
1294	int err;
1295
1296	err = __vringh_get_head(vrh, getu16_iotlb, &vrh->last_avail_idx);
1297	if (err < 0)
1298		return err;
1299
1300	/* Empty... */
1301	if (err == vrh->vring.num)
1302		return 0;
1303
1304	*head = err;
1305	err = __vringh_iov(vrh, *head, riov, wiov, no_range_check, NULL,
1306			   gfp, copydesc_iotlb);
1307	if (err)
1308		return err;
1309
1310	return 1;
1311}
1312EXPORT_SYMBOL(vringh_getdesc_iotlb);
1313
1314/**
1315 * vringh_iov_pull_iotlb - copy bytes from vring_iov.
1316 * @vrh: the vring.
1317 * @riov: the riov as passed to vringh_getdesc_iotlb() (updated as we consume)
1318 * @dst: the place to copy.
1319 * @len: the maximum length to copy.
1320 *
1321 * Returns the bytes copied <= len or a negative errno.
1322 */
1323ssize_t vringh_iov_pull_iotlb(struct vringh *vrh,
1324			      struct vringh_kiov *riov,
1325			      void *dst, size_t len)
1326{
1327	return vringh_iov_xfer(vrh, riov, dst, len, xfer_from_iotlb);
1328}
1329EXPORT_SYMBOL(vringh_iov_pull_iotlb);
1330
1331/**
1332 * vringh_iov_push_iotlb - copy bytes into vring_iov.
1333 * @vrh: the vring.
1334 * @wiov: the wiov as passed to vringh_getdesc_iotlb() (updated as we consume)
1335 * @dst: the place to copy.
1336 * @len: the maximum length to copy.
1337 *
1338 * Returns the bytes copied <= len or a negative errno.
1339 */
1340ssize_t vringh_iov_push_iotlb(struct vringh *vrh,
1341			      struct vringh_kiov *wiov,
1342			      const void *src, size_t len)
1343{
1344	return vringh_iov_xfer(vrh, wiov, (void *)src, len, xfer_to_iotlb);
1345}
1346EXPORT_SYMBOL(vringh_iov_push_iotlb);
1347
1348/**
1349 * vringh_abandon_iotlb - we've decided not to handle the descriptor(s).
1350 * @vrh: the vring.
1351 * @num: the number of descriptors to put back (ie. num
1352 *	 vringh_get_iotlb() to undo).
1353 *
1354 * The next vringh_get_iotlb() will return the old descriptor(s) again.
1355 */
1356void vringh_abandon_iotlb(struct vringh *vrh, unsigned int num)
1357{
1358	/* We only update vring_avail_event(vr) when we want to be notified,
1359	 * so we haven't changed that yet.
1360	 */
1361	vrh->last_avail_idx -= num;
1362}
1363EXPORT_SYMBOL(vringh_abandon_iotlb);
1364
1365/**
1366 * vringh_complete_iotlb - we've finished with descriptor, publish it.
1367 * @vrh: the vring.
1368 * @head: the head as filled in by vringh_getdesc_iotlb.
1369 * @len: the length of data we have written.
1370 *
1371 * You should check vringh_need_notify_iotlb() after one or more calls
1372 * to this function.
1373 */
1374int vringh_complete_iotlb(struct vringh *vrh, u16 head, u32 len)
1375{
1376	struct vring_used_elem used;
1377
1378	used.id = cpu_to_vringh32(vrh, head);
1379	used.len = cpu_to_vringh32(vrh, len);
1380
1381	return __vringh_complete(vrh, &used, 1, putu16_iotlb, putused_iotlb);
1382}
1383EXPORT_SYMBOL(vringh_complete_iotlb);
1384
1385/**
1386 * vringh_notify_enable_iotlb - we want to know if something changes.
1387 * @vrh: the vring.
1388 *
1389 * This always enables notifications, but returns false if there are
1390 * now more buffers available in the vring.
1391 */
1392bool vringh_notify_enable_iotlb(struct vringh *vrh)
1393{
1394	return __vringh_notify_enable(vrh, getu16_iotlb, putu16_iotlb);
1395}
1396EXPORT_SYMBOL(vringh_notify_enable_iotlb);
1397
1398/**
1399 * vringh_notify_disable_iotlb - don't tell us if something changes.
1400 * @vrh: the vring.
1401 *
1402 * This is our normal running state: we disable and then only enable when
1403 * we're going to sleep.
1404 */
1405void vringh_notify_disable_iotlb(struct vringh *vrh)
1406{
1407	__vringh_notify_disable(vrh, putu16_iotlb);
1408}
1409EXPORT_SYMBOL(vringh_notify_disable_iotlb);
1410
1411/**
1412 * vringh_need_notify_iotlb - must we tell the other side about used buffers?
1413 * @vrh: the vring we've called vringh_complete_iotlb() on.
1414 *
1415 * Returns -errno or 0 if we don't need to tell the other side, 1 if we do.
1416 */
1417int vringh_need_notify_iotlb(struct vringh *vrh)
1418{
1419	return __vringh_need_notify(vrh, getu16_iotlb);
1420}
1421EXPORT_SYMBOL(vringh_need_notify_iotlb);
1422
1423#endif
1424
1425MODULE_LICENSE("GPL");
v5.4
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Helpers for the host side of a virtio ring.
   4 *
   5 * Since these may be in userspace, we use (inline) accessors.
   6 */
   7#include <linux/compiler.h>
   8#include <linux/module.h>
   9#include <linux/vringh.h>
  10#include <linux/virtio_ring.h>
  11#include <linux/kernel.h>
  12#include <linux/ratelimit.h>
  13#include <linux/uaccess.h>
  14#include <linux/slab.h>
  15#include <linux/export.h>
 
 
 
 
 
  16#include <uapi/linux/virtio_config.h>
  17
  18static __printf(1,2) __cold void vringh_bad(const char *fmt, ...)
  19{
  20	static DEFINE_RATELIMIT_STATE(vringh_rs,
  21				      DEFAULT_RATELIMIT_INTERVAL,
  22				      DEFAULT_RATELIMIT_BURST);
  23	if (__ratelimit(&vringh_rs)) {
  24		va_list ap;
  25		va_start(ap, fmt);
  26		printk(KERN_NOTICE "vringh:");
  27		vprintk(fmt, ap);
  28		va_end(ap);
  29	}
  30}
  31
  32/* Returns vring->num if empty, -ve on error. */
  33static inline int __vringh_get_head(const struct vringh *vrh,
  34				    int (*getu16)(const struct vringh *vrh,
  35						  u16 *val, const __virtio16 *p),
  36				    u16 *last_avail_idx)
  37{
  38	u16 avail_idx, i, head;
  39	int err;
  40
  41	err = getu16(vrh, &avail_idx, &vrh->vring.avail->idx);
  42	if (err) {
  43		vringh_bad("Failed to access avail idx at %p",
  44			   &vrh->vring.avail->idx);
  45		return err;
  46	}
  47
  48	if (*last_avail_idx == avail_idx)
  49		return vrh->vring.num;
  50
  51	/* Only get avail ring entries after they have been exposed by guest. */
  52	virtio_rmb(vrh->weak_barriers);
  53
  54	i = *last_avail_idx & (vrh->vring.num - 1);
  55
  56	err = getu16(vrh, &head, &vrh->vring.avail->ring[i]);
  57	if (err) {
  58		vringh_bad("Failed to read head: idx %d address %p",
  59			   *last_avail_idx, &vrh->vring.avail->ring[i]);
  60		return err;
  61	}
  62
  63	if (head >= vrh->vring.num) {
  64		vringh_bad("Guest says index %u > %u is available",
  65			   head, vrh->vring.num);
  66		return -EINVAL;
  67	}
  68
  69	(*last_avail_idx)++;
  70	return head;
  71}
  72
  73/* Copy some bytes to/from the iovec.  Returns num copied. */
  74static inline ssize_t vringh_iov_xfer(struct vringh_kiov *iov,
 
  75				      void *ptr, size_t len,
  76				      int (*xfer)(void *addr, void *ptr,
 
  77						  size_t len))
  78{
  79	int err, done = 0;
  80
  81	while (len && iov->i < iov->used) {
  82		size_t partlen;
  83
  84		partlen = min(iov->iov[iov->i].iov_len, len);
  85		err = xfer(iov->iov[iov->i].iov_base, ptr, partlen);
  86		if (err)
  87			return err;
  88		done += partlen;
  89		len -= partlen;
  90		ptr += partlen;
  91		iov->consumed += partlen;
  92		iov->iov[iov->i].iov_len -= partlen;
  93		iov->iov[iov->i].iov_base += partlen;
  94
  95		if (!iov->iov[iov->i].iov_len) {
  96			/* Fix up old iov element then increment. */
  97			iov->iov[iov->i].iov_len = iov->consumed;
  98			iov->iov[iov->i].iov_base -= iov->consumed;
 
  99			
 100			iov->consumed = 0;
 101			iov->i++;
 102		}
 103	}
 104	return done;
 105}
 106
 107/* May reduce *len if range is shorter. */
 108static inline bool range_check(struct vringh *vrh, u64 addr, size_t *len,
 109			       struct vringh_range *range,
 110			       bool (*getrange)(struct vringh *,
 111						u64, struct vringh_range *))
 112{
 113	if (addr < range->start || addr > range->end_incl) {
 114		if (!getrange(vrh, addr, range))
 115			return false;
 116	}
 117	BUG_ON(addr < range->start || addr > range->end_incl);
 118
 119	/* To end of memory? */
 120	if (unlikely(addr + *len == 0)) {
 121		if (range->end_incl == -1ULL)
 122			return true;
 123		goto truncate;
 124	}
 125
 126	/* Otherwise, don't wrap. */
 127	if (addr + *len < addr) {
 128		vringh_bad("Wrapping descriptor %zu@0x%llx",
 129			   *len, (unsigned long long)addr);
 130		return false;
 131	}
 132
 133	if (unlikely(addr + *len - 1 > range->end_incl))
 134		goto truncate;
 135	return true;
 136
 137truncate:
 138	*len = range->end_incl + 1 - addr;
 139	return true;
 140}
 141
 142static inline bool no_range_check(struct vringh *vrh, u64 addr, size_t *len,
 143				  struct vringh_range *range,
 144				  bool (*getrange)(struct vringh *,
 145						   u64, struct vringh_range *))
 146{
 147	return true;
 148}
 149
 150/* No reason for this code to be inline. */
 151static int move_to_indirect(const struct vringh *vrh,
 152			    int *up_next, u16 *i, void *addr,
 153			    const struct vring_desc *desc,
 154			    struct vring_desc **descs, int *desc_max)
 155{
 156	u32 len;
 157
 158	/* Indirect tables can't have indirect. */
 159	if (*up_next != -1) {
 160		vringh_bad("Multilevel indirect %u->%u", *up_next, *i);
 161		return -EINVAL;
 162	}
 163
 164	len = vringh32_to_cpu(vrh, desc->len);
 165	if (unlikely(len % sizeof(struct vring_desc))) {
 166		vringh_bad("Strange indirect len %u", desc->len);
 167		return -EINVAL;
 168	}
 169
 170	/* We will check this when we follow it! */
 171	if (desc->flags & cpu_to_vringh16(vrh, VRING_DESC_F_NEXT))
 172		*up_next = vringh16_to_cpu(vrh, desc->next);
 173	else
 174		*up_next = -2;
 175	*descs = addr;
 176	*desc_max = len / sizeof(struct vring_desc);
 177
 178	/* Now, start at the first indirect. */
 179	*i = 0;
 180	return 0;
 181}
 182
 183static int resize_iovec(struct vringh_kiov *iov, gfp_t gfp)
 184{
 185	struct kvec *new;
 186	unsigned int flag, new_num = (iov->max_num & ~VRINGH_IOV_ALLOCATED) * 2;
 187
 188	if (new_num < 8)
 189		new_num = 8;
 190
 191	flag = (iov->max_num & VRINGH_IOV_ALLOCATED);
 192	if (flag)
 193		new = krealloc(iov->iov, new_num * sizeof(struct iovec), gfp);
 194	else {
 195		new = kmalloc_array(new_num, sizeof(struct iovec), gfp);
 196		if (new) {
 197			memcpy(new, iov->iov,
 198			       iov->max_num * sizeof(struct iovec));
 199			flag = VRINGH_IOV_ALLOCATED;
 200		}
 201	}
 202	if (!new)
 203		return -ENOMEM;
 204	iov->iov = new;
 205	iov->max_num = (new_num | flag);
 206	return 0;
 207}
 208
 209static u16 __cold return_from_indirect(const struct vringh *vrh, int *up_next,
 210				       struct vring_desc **descs, int *desc_max)
 211{
 212	u16 i = *up_next;
 213
 214	*up_next = -1;
 215	*descs = vrh->vring.desc;
 216	*desc_max = vrh->vring.num;
 217	return i;
 218}
 219
 220static int slow_copy(struct vringh *vrh, void *dst, const void *src,
 221		     bool (*rcheck)(struct vringh *vrh, u64 addr, size_t *len,
 222				    struct vringh_range *range,
 223				    bool (*getrange)(struct vringh *vrh,
 224						     u64,
 225						     struct vringh_range *)),
 226		     bool (*getrange)(struct vringh *vrh,
 227				      u64 addr,
 228				      struct vringh_range *r),
 229		     struct vringh_range *range,
 230		     int (*copy)(void *dst, const void *src, size_t len))
 
 231{
 232	size_t part, len = sizeof(struct vring_desc);
 233
 234	do {
 235		u64 addr;
 236		int err;
 237
 238		part = len;
 239		addr = (u64)(unsigned long)src - range->offset;
 240
 241		if (!rcheck(vrh, addr, &part, range, getrange))
 242			return -EINVAL;
 243
 244		err = copy(dst, src, part);
 245		if (err)
 246			return err;
 247
 248		dst += part;
 249		src += part;
 250		len -= part;
 251	} while (len);
 252	return 0;
 253}
 254
 255static inline int
 256__vringh_iov(struct vringh *vrh, u16 i,
 257	     struct vringh_kiov *riov,
 258	     struct vringh_kiov *wiov,
 259	     bool (*rcheck)(struct vringh *vrh, u64 addr, size_t *len,
 260			    struct vringh_range *range,
 261			    bool (*getrange)(struct vringh *, u64,
 262					     struct vringh_range *)),
 263	     bool (*getrange)(struct vringh *, u64, struct vringh_range *),
 264	     gfp_t gfp,
 265	     int (*copy)(void *dst, const void *src, size_t len))
 
 266{
 267	int err, count = 0, up_next, desc_max;
 268	struct vring_desc desc, *descs;
 269	struct vringh_range range = { -1ULL, 0 }, slowrange;
 270	bool slow = false;
 271
 272	/* We start traversing vring's descriptor table. */
 273	descs = vrh->vring.desc;
 274	desc_max = vrh->vring.num;
 275	up_next = -1;
 276
 277	if (riov)
 278		riov->i = riov->used = 0;
 279	else if (wiov)
 280		wiov->i = wiov->used = 0;
 281	else
 282		/* You must want something! */
 283		BUG();
 284
 285	for (;;) {
 286		void *addr;
 287		struct vringh_kiov *iov;
 288		size_t len;
 289
 290		if (unlikely(slow))
 291			err = slow_copy(vrh, &desc, &descs[i], rcheck, getrange,
 292					&slowrange, copy);
 293		else
 294			err = copy(&desc, &descs[i], sizeof(desc));
 295		if (unlikely(err))
 296			goto fail;
 297
 298		if (unlikely(desc.flags &
 299			     cpu_to_vringh16(vrh, VRING_DESC_F_INDIRECT))) {
 300			u64 a = vringh64_to_cpu(vrh, desc.addr);
 301
 302			/* Make sure it's OK, and get offset. */
 303			len = vringh32_to_cpu(vrh, desc.len);
 304			if (!rcheck(vrh, a, &len, &range, getrange)) {
 305				err = -EINVAL;
 306				goto fail;
 307			}
 308
 309			if (unlikely(len != vringh32_to_cpu(vrh, desc.len))) {
 310				slow = true;
 311				/* We need to save this range to use offset */
 312				slowrange = range;
 313			}
 314
 315			addr = (void *)(long)(a + range.offset);
 316			err = move_to_indirect(vrh, &up_next, &i, addr, &desc,
 317					       &descs, &desc_max);
 318			if (err)
 319				goto fail;
 320			continue;
 321		}
 322
 323		if (count++ == vrh->vring.num) {
 324			vringh_bad("Descriptor loop in %p", descs);
 325			err = -ELOOP;
 326			goto fail;
 327		}
 328
 329		if (desc.flags & cpu_to_vringh16(vrh, VRING_DESC_F_WRITE))
 330			iov = wiov;
 331		else {
 332			iov = riov;
 333			if (unlikely(wiov && wiov->i)) {
 334				vringh_bad("Readable desc %p after writable",
 335					   &descs[i]);
 336				err = -EINVAL;
 337				goto fail;
 338			}
 339		}
 340
 341		if (!iov) {
 342			vringh_bad("Unexpected %s desc",
 343				   !wiov ? "writable" : "readable");
 344			err = -EPROTO;
 345			goto fail;
 346		}
 347
 348	again:
 349		/* Make sure it's OK, and get offset. */
 350		len = vringh32_to_cpu(vrh, desc.len);
 351		if (!rcheck(vrh, vringh64_to_cpu(vrh, desc.addr), &len, &range,
 352			    getrange)) {
 353			err = -EINVAL;
 354			goto fail;
 355		}
 356		addr = (void *)(unsigned long)(vringh64_to_cpu(vrh, desc.addr) +
 357					       range.offset);
 358
 359		if (unlikely(iov->used == (iov->max_num & ~VRINGH_IOV_ALLOCATED))) {
 360			err = resize_iovec(iov, gfp);
 361			if (err)
 362				goto fail;
 363		}
 364
 365		iov->iov[iov->used].iov_base = addr;
 366		iov->iov[iov->used].iov_len = len;
 367		iov->used++;
 368
 369		if (unlikely(len != vringh32_to_cpu(vrh, desc.len))) {
 370			desc.len = cpu_to_vringh32(vrh,
 371				   vringh32_to_cpu(vrh, desc.len) - len);
 372			desc.addr = cpu_to_vringh64(vrh,
 373				    vringh64_to_cpu(vrh, desc.addr) + len);
 374			goto again;
 375		}
 376
 377		if (desc.flags & cpu_to_vringh16(vrh, VRING_DESC_F_NEXT)) {
 378			i = vringh16_to_cpu(vrh, desc.next);
 379		} else {
 380			/* Just in case we need to finish traversing above. */
 381			if (unlikely(up_next > 0)) {
 382				i = return_from_indirect(vrh, &up_next,
 383							 &descs, &desc_max);
 384				slow = false;
 385			} else
 386				break;
 387		}
 388
 389		if (i >= desc_max) {
 390			vringh_bad("Chained index %u > %u", i, desc_max);
 391			err = -EINVAL;
 392			goto fail;
 393		}
 394	}
 395
 396	return 0;
 397
 398fail:
 399	return err;
 400}
 401
 402static inline int __vringh_complete(struct vringh *vrh,
 403				    const struct vring_used_elem *used,
 404				    unsigned int num_used,
 405				    int (*putu16)(const struct vringh *vrh,
 406						  __virtio16 *p, u16 val),
 407				    int (*putused)(struct vring_used_elem *dst,
 
 408						   const struct vring_used_elem
 409						   *src, unsigned num))
 410{
 411	struct vring_used *used_ring;
 412	int err;
 413	u16 used_idx, off;
 414
 415	used_ring = vrh->vring.used;
 416	used_idx = vrh->last_used_idx + vrh->completed;
 417
 418	off = used_idx % vrh->vring.num;
 419
 420	/* Compiler knows num_used == 1 sometimes, hence extra check */
 421	if (num_used > 1 && unlikely(off + num_used >= vrh->vring.num)) {
 422		u16 part = vrh->vring.num - off;
 423		err = putused(&used_ring->ring[off], used, part);
 424		if (!err)
 425			err = putused(&used_ring->ring[0], used + part,
 426				      num_used - part);
 427	} else
 428		err = putused(&used_ring->ring[off], used, num_used);
 429
 430	if (err) {
 431		vringh_bad("Failed to write %u used entries %u at %p",
 432			   num_used, off, &used_ring->ring[off]);
 433		return err;
 434	}
 435
 436	/* Make sure buffer is written before we update index. */
 437	virtio_wmb(vrh->weak_barriers);
 438
 439	err = putu16(vrh, &vrh->vring.used->idx, used_idx + num_used);
 440	if (err) {
 441		vringh_bad("Failed to update used index at %p",
 442			   &vrh->vring.used->idx);
 443		return err;
 444	}
 445
 446	vrh->completed += num_used;
 447	return 0;
 448}
 449
 450
 451static inline int __vringh_need_notify(struct vringh *vrh,
 452				       int (*getu16)(const struct vringh *vrh,
 453						     u16 *val,
 454						     const __virtio16 *p))
 455{
 456	bool notify;
 457	u16 used_event;
 458	int err;
 459
 460	/* Flush out used index update. This is paired with the
 461	 * barrier that the Guest executes when enabling
 462	 * interrupts. */
 463	virtio_mb(vrh->weak_barriers);
 464
 465	/* Old-style, without event indices. */
 466	if (!vrh->event_indices) {
 467		u16 flags;
 468		err = getu16(vrh, &flags, &vrh->vring.avail->flags);
 469		if (err) {
 470			vringh_bad("Failed to get flags at %p",
 471				   &vrh->vring.avail->flags);
 472			return err;
 473		}
 474		return (!(flags & VRING_AVAIL_F_NO_INTERRUPT));
 475	}
 476
 477	/* Modern: we know when other side wants to know. */
 478	err = getu16(vrh, &used_event, &vring_used_event(&vrh->vring));
 479	if (err) {
 480		vringh_bad("Failed to get used event idx at %p",
 481			   &vring_used_event(&vrh->vring));
 482		return err;
 483	}
 484
 485	/* Just in case we added so many that we wrap. */
 486	if (unlikely(vrh->completed > 0xffff))
 487		notify = true;
 488	else
 489		notify = vring_need_event(used_event,
 490					  vrh->last_used_idx + vrh->completed,
 491					  vrh->last_used_idx);
 492
 493	vrh->last_used_idx += vrh->completed;
 494	vrh->completed = 0;
 495	return notify;
 496}
 497
 498static inline bool __vringh_notify_enable(struct vringh *vrh,
 499					  int (*getu16)(const struct vringh *vrh,
 500							u16 *val, const __virtio16 *p),
 501					  int (*putu16)(const struct vringh *vrh,
 502							__virtio16 *p, u16 val))
 503{
 504	u16 avail;
 505
 506	if (!vrh->event_indices) {
 507		/* Old-school; update flags. */
 508		if (putu16(vrh, &vrh->vring.used->flags, 0) != 0) {
 509			vringh_bad("Clearing used flags %p",
 510				   &vrh->vring.used->flags);
 511			return true;
 512		}
 513	} else {
 514		if (putu16(vrh, &vring_avail_event(&vrh->vring),
 515			   vrh->last_avail_idx) != 0) {
 516			vringh_bad("Updating avail event index %p",
 517				   &vring_avail_event(&vrh->vring));
 518			return true;
 519		}
 520	}
 521
 522	/* They could have slipped one in as we were doing that: make
 523	 * sure it's written, then check again. */
 524	virtio_mb(vrh->weak_barriers);
 525
 526	if (getu16(vrh, &avail, &vrh->vring.avail->idx) != 0) {
 527		vringh_bad("Failed to check avail idx at %p",
 528			   &vrh->vring.avail->idx);
 529		return true;
 530	}
 531
 532	/* This is unlikely, so we just leave notifications enabled
 533	 * (if we're using event_indices, we'll only get one
 534	 * notification anyway). */
 535	return avail == vrh->last_avail_idx;
 536}
 537
 538static inline void __vringh_notify_disable(struct vringh *vrh,
 539					   int (*putu16)(const struct vringh *vrh,
 540							 __virtio16 *p, u16 val))
 541{
 542	if (!vrh->event_indices) {
 543		/* Old-school; update flags. */
 544		if (putu16(vrh, &vrh->vring.used->flags,
 545			   VRING_USED_F_NO_NOTIFY)) {
 546			vringh_bad("Setting used flags %p",
 547				   &vrh->vring.used->flags);
 548		}
 549	}
 550}
 551
 552/* Userspace access helpers: in this case, addresses are really userspace. */
 553static inline int getu16_user(const struct vringh *vrh, u16 *val, const __virtio16 *p)
 554{
 555	__virtio16 v = 0;
 556	int rc = get_user(v, (__force __virtio16 __user *)p);
 557	*val = vringh16_to_cpu(vrh, v);
 558	return rc;
 559}
 560
 561static inline int putu16_user(const struct vringh *vrh, __virtio16 *p, u16 val)
 562{
 563	__virtio16 v = cpu_to_vringh16(vrh, val);
 564	return put_user(v, (__force __virtio16 __user *)p);
 565}
 566
 567static inline int copydesc_user(void *dst, const void *src, size_t len)
 
 568{
 569	return copy_from_user(dst, (__force void __user *)src, len) ?
 570		-EFAULT : 0;
 571}
 572
 573static inline int putused_user(struct vring_used_elem *dst,
 
 574			       const struct vring_used_elem *src,
 575			       unsigned int num)
 576{
 577	return copy_to_user((__force void __user *)dst, src,
 578			    sizeof(*dst) * num) ? -EFAULT : 0;
 579}
 580
 581static inline int xfer_from_user(void *src, void *dst, size_t len)
 
 582{
 583	return copy_from_user(dst, (__force void __user *)src, len) ?
 584		-EFAULT : 0;
 585}
 586
 587static inline int xfer_to_user(void *dst, void *src, size_t len)
 
 588{
 589	return copy_to_user((__force void __user *)dst, src, len) ?
 590		-EFAULT : 0;
 591}
 592
 593/**
 594 * vringh_init_user - initialize a vringh for a userspace vring.
 595 * @vrh: the vringh to initialize.
 596 * @features: the feature bits for this ring.
 597 * @num: the number of elements.
 598 * @weak_barriers: true if we only need memory barriers, not I/O.
 599 * @desc: the userpace descriptor pointer.
 600 * @avail: the userpace avail pointer.
 601 * @used: the userpace used pointer.
 602 *
 603 * Returns an error if num is invalid: you should check pointers
 604 * yourself!
 605 */
 606int vringh_init_user(struct vringh *vrh, u64 features,
 607		     unsigned int num, bool weak_barriers,
 608		     struct vring_desc __user *desc,
 609		     struct vring_avail __user *avail,
 610		     struct vring_used __user *used)
 611{
 612	/* Sane power of 2 please! */
 613	if (!num || num > 0xffff || (num & (num - 1))) {
 614		vringh_bad("Bad ring size %u", num);
 615		return -EINVAL;
 616	}
 617
 618	vrh->little_endian = (features & (1ULL << VIRTIO_F_VERSION_1));
 619	vrh->event_indices = (features & (1 << VIRTIO_RING_F_EVENT_IDX));
 620	vrh->weak_barriers = weak_barriers;
 621	vrh->completed = 0;
 622	vrh->last_avail_idx = 0;
 623	vrh->last_used_idx = 0;
 624	vrh->vring.num = num;
 625	/* vring expects kernel addresses, but only used via accessors. */
 626	vrh->vring.desc = (__force struct vring_desc *)desc;
 627	vrh->vring.avail = (__force struct vring_avail *)avail;
 628	vrh->vring.used = (__force struct vring_used *)used;
 629	return 0;
 630}
 631EXPORT_SYMBOL(vringh_init_user);
 632
 633/**
 634 * vringh_getdesc_user - get next available descriptor from userspace ring.
 635 * @vrh: the userspace vring.
 636 * @riov: where to put the readable descriptors (or NULL)
 637 * @wiov: where to put the writable descriptors (or NULL)
 638 * @getrange: function to call to check ranges.
 639 * @head: head index we received, for passing to vringh_complete_user().
 640 *
 641 * Returns 0 if there was no descriptor, 1 if there was, or -errno.
 642 *
 643 * Note that on error return, you can tell the difference between an
 644 * invalid ring and a single invalid descriptor: in the former case,
 645 * *head will be vrh->vring.num.  You may be able to ignore an invalid
 646 * descriptor, but there's not much you can do with an invalid ring.
 647 *
 648 * Note that you may need to clean up riov and wiov, even on error!
 649 */
 650int vringh_getdesc_user(struct vringh *vrh,
 651			struct vringh_iov *riov,
 652			struct vringh_iov *wiov,
 653			bool (*getrange)(struct vringh *vrh,
 654					 u64 addr, struct vringh_range *r),
 655			u16 *head)
 656{
 657	int err;
 658
 659	*head = vrh->vring.num;
 660	err = __vringh_get_head(vrh, getu16_user, &vrh->last_avail_idx);
 661	if (err < 0)
 662		return err;
 663
 664	/* Empty... */
 665	if (err == vrh->vring.num)
 666		return 0;
 667
 668	/* We need the layouts to be the identical for this to work */
 669	BUILD_BUG_ON(sizeof(struct vringh_kiov) != sizeof(struct vringh_iov));
 670	BUILD_BUG_ON(offsetof(struct vringh_kiov, iov) !=
 671		     offsetof(struct vringh_iov, iov));
 672	BUILD_BUG_ON(offsetof(struct vringh_kiov, i) !=
 673		     offsetof(struct vringh_iov, i));
 674	BUILD_BUG_ON(offsetof(struct vringh_kiov, used) !=
 675		     offsetof(struct vringh_iov, used));
 676	BUILD_BUG_ON(offsetof(struct vringh_kiov, max_num) !=
 677		     offsetof(struct vringh_iov, max_num));
 678	BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
 679	BUILD_BUG_ON(offsetof(struct iovec, iov_base) !=
 680		     offsetof(struct kvec, iov_base));
 681	BUILD_BUG_ON(offsetof(struct iovec, iov_len) !=
 682		     offsetof(struct kvec, iov_len));
 683	BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_base)
 684		     != sizeof(((struct kvec *)NULL)->iov_base));
 685	BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_len)
 686		     != sizeof(((struct kvec *)NULL)->iov_len));
 687
 688	*head = err;
 689	err = __vringh_iov(vrh, *head, (struct vringh_kiov *)riov,
 690			   (struct vringh_kiov *)wiov,
 691			   range_check, getrange, GFP_KERNEL, copydesc_user);
 692	if (err)
 693		return err;
 694
 695	return 1;
 696}
 697EXPORT_SYMBOL(vringh_getdesc_user);
 698
 699/**
 700 * vringh_iov_pull_user - copy bytes from vring_iov.
 701 * @riov: the riov as passed to vringh_getdesc_user() (updated as we consume)
 702 * @dst: the place to copy.
 703 * @len: the maximum length to copy.
 704 *
 705 * Returns the bytes copied <= len or a negative errno.
 706 */
 707ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len)
 708{
 709	return vringh_iov_xfer((struct vringh_kiov *)riov,
 710			       dst, len, xfer_from_user);
 711}
 712EXPORT_SYMBOL(vringh_iov_pull_user);
 713
 714/**
 715 * vringh_iov_push_user - copy bytes into vring_iov.
 716 * @wiov: the wiov as passed to vringh_getdesc_user() (updated as we consume)
 717 * @dst: the place to copy.
 718 * @len: the maximum length to copy.
 719 *
 720 * Returns the bytes copied <= len or a negative errno.
 721 */
 722ssize_t vringh_iov_push_user(struct vringh_iov *wiov,
 723			     const void *src, size_t len)
 724{
 725	return vringh_iov_xfer((struct vringh_kiov *)wiov,
 726			       (void *)src, len, xfer_to_user);
 727}
 728EXPORT_SYMBOL(vringh_iov_push_user);
 729
 730/**
 731 * vringh_abandon_user - we've decided not to handle the descriptor(s).
 732 * @vrh: the vring.
 733 * @num: the number of descriptors to put back (ie. num
 734 *	 vringh_get_user() to undo).
 735 *
 736 * The next vringh_get_user() will return the old descriptor(s) again.
 737 */
 738void vringh_abandon_user(struct vringh *vrh, unsigned int num)
 739{
 740	/* We only update vring_avail_event(vr) when we want to be notified,
 741	 * so we haven't changed that yet. */
 742	vrh->last_avail_idx -= num;
 743}
 744EXPORT_SYMBOL(vringh_abandon_user);
 745
 746/**
 747 * vringh_complete_user - we've finished with descriptor, publish it.
 748 * @vrh: the vring.
 749 * @head: the head as filled in by vringh_getdesc_user.
 750 * @len: the length of data we have written.
 751 *
 752 * You should check vringh_need_notify_user() after one or more calls
 753 * to this function.
 754 */
 755int vringh_complete_user(struct vringh *vrh, u16 head, u32 len)
 756{
 757	struct vring_used_elem used;
 758
 759	used.id = cpu_to_vringh32(vrh, head);
 760	used.len = cpu_to_vringh32(vrh, len);
 761	return __vringh_complete(vrh, &used, 1, putu16_user, putused_user);
 762}
 763EXPORT_SYMBOL(vringh_complete_user);
 764
 765/**
 766 * vringh_complete_multi_user - we've finished with many descriptors.
 767 * @vrh: the vring.
 768 * @used: the head, length pairs.
 769 * @num_used: the number of used elements.
 770 *
 771 * You should check vringh_need_notify_user() after one or more calls
 772 * to this function.
 773 */
 774int vringh_complete_multi_user(struct vringh *vrh,
 775			       const struct vring_used_elem used[],
 776			       unsigned num_used)
 777{
 778	return __vringh_complete(vrh, used, num_used,
 779				 putu16_user, putused_user);
 780}
 781EXPORT_SYMBOL(vringh_complete_multi_user);
 782
 783/**
 784 * vringh_notify_enable_user - we want to know if something changes.
 785 * @vrh: the vring.
 786 *
 787 * This always enables notifications, but returns false if there are
 788 * now more buffers available in the vring.
 789 */
 790bool vringh_notify_enable_user(struct vringh *vrh)
 791{
 792	return __vringh_notify_enable(vrh, getu16_user, putu16_user);
 793}
 794EXPORT_SYMBOL(vringh_notify_enable_user);
 795
 796/**
 797 * vringh_notify_disable_user - don't tell us if something changes.
 798 * @vrh: the vring.
 799 *
 800 * This is our normal running state: we disable and then only enable when
 801 * we're going to sleep.
 802 */
 803void vringh_notify_disable_user(struct vringh *vrh)
 804{
 805	__vringh_notify_disable(vrh, putu16_user);
 806}
 807EXPORT_SYMBOL(vringh_notify_disable_user);
 808
 809/**
 810 * vringh_need_notify_user - must we tell the other side about used buffers?
 811 * @vrh: the vring we've called vringh_complete_user() on.
 812 *
 813 * Returns -errno or 0 if we don't need to tell the other side, 1 if we do.
 814 */
 815int vringh_need_notify_user(struct vringh *vrh)
 816{
 817	return __vringh_need_notify(vrh, getu16_user);
 818}
 819EXPORT_SYMBOL(vringh_need_notify_user);
 820
 821/* Kernelspace access helpers. */
 822static inline int getu16_kern(const struct vringh *vrh,
 823			      u16 *val, const __virtio16 *p)
 824{
 825	*val = vringh16_to_cpu(vrh, READ_ONCE(*p));
 826	return 0;
 827}
 828
 829static inline int putu16_kern(const struct vringh *vrh, __virtio16 *p, u16 val)
 830{
 831	WRITE_ONCE(*p, cpu_to_vringh16(vrh, val));
 832	return 0;
 833}
 834
 835static inline int copydesc_kern(void *dst, const void *src, size_t len)
 
 836{
 837	memcpy(dst, src, len);
 838	return 0;
 839}
 840
 841static inline int putused_kern(struct vring_used_elem *dst,
 
 842			       const struct vring_used_elem *src,
 843			       unsigned int num)
 844{
 845	memcpy(dst, src, num * sizeof(*dst));
 846	return 0;
 847}
 848
 849static inline int xfer_kern(void *src, void *dst, size_t len)
 
 850{
 851	memcpy(dst, src, len);
 852	return 0;
 853}
 854
 855static inline int kern_xfer(void *dst, void *src, size_t len)
 
 856{
 857	memcpy(dst, src, len);
 858	return 0;
 859}
 860
 861/**
 862 * vringh_init_kern - initialize a vringh for a kernelspace vring.
 863 * @vrh: the vringh to initialize.
 864 * @features: the feature bits for this ring.
 865 * @num: the number of elements.
 866 * @weak_barriers: true if we only need memory barriers, not I/O.
 867 * @desc: the userpace descriptor pointer.
 868 * @avail: the userpace avail pointer.
 869 * @used: the userpace used pointer.
 870 *
 871 * Returns an error if num is invalid.
 872 */
 873int vringh_init_kern(struct vringh *vrh, u64 features,
 874		     unsigned int num, bool weak_barriers,
 875		     struct vring_desc *desc,
 876		     struct vring_avail *avail,
 877		     struct vring_used *used)
 878{
 879	/* Sane power of 2 please! */
 880	if (!num || num > 0xffff || (num & (num - 1))) {
 881		vringh_bad("Bad ring size %u", num);
 882		return -EINVAL;
 883	}
 884
 885	vrh->little_endian = (features & (1ULL << VIRTIO_F_VERSION_1));
 886	vrh->event_indices = (features & (1 << VIRTIO_RING_F_EVENT_IDX));
 887	vrh->weak_barriers = weak_barriers;
 888	vrh->completed = 0;
 889	vrh->last_avail_idx = 0;
 890	vrh->last_used_idx = 0;
 891	vrh->vring.num = num;
 892	vrh->vring.desc = desc;
 893	vrh->vring.avail = avail;
 894	vrh->vring.used = used;
 895	return 0;
 896}
 897EXPORT_SYMBOL(vringh_init_kern);
 898
 899/**
 900 * vringh_getdesc_kern - get next available descriptor from kernelspace ring.
 901 * @vrh: the kernelspace vring.
 902 * @riov: where to put the readable descriptors (or NULL)
 903 * @wiov: where to put the writable descriptors (or NULL)
 904 * @head: head index we received, for passing to vringh_complete_kern().
 905 * @gfp: flags for allocating larger riov/wiov.
 906 *
 907 * Returns 0 if there was no descriptor, 1 if there was, or -errno.
 908 *
 909 * Note that on error return, you can tell the difference between an
 910 * invalid ring and a single invalid descriptor: in the former case,
 911 * *head will be vrh->vring.num.  You may be able to ignore an invalid
 912 * descriptor, but there's not much you can do with an invalid ring.
 913 *
 914 * Note that you may need to clean up riov and wiov, even on error!
 915 */
 916int vringh_getdesc_kern(struct vringh *vrh,
 917			struct vringh_kiov *riov,
 918			struct vringh_kiov *wiov,
 919			u16 *head,
 920			gfp_t gfp)
 921{
 922	int err;
 923
 924	err = __vringh_get_head(vrh, getu16_kern, &vrh->last_avail_idx);
 925	if (err < 0)
 926		return err;
 927
 928	/* Empty... */
 929	if (err == vrh->vring.num)
 930		return 0;
 931
 932	*head = err;
 933	err = __vringh_iov(vrh, *head, riov, wiov, no_range_check, NULL,
 934			   gfp, copydesc_kern);
 935	if (err)
 936		return err;
 937
 938	return 1;
 939}
 940EXPORT_SYMBOL(vringh_getdesc_kern);
 941
 942/**
 943 * vringh_iov_pull_kern - copy bytes from vring_iov.
 944 * @riov: the riov as passed to vringh_getdesc_kern() (updated as we consume)
 945 * @dst: the place to copy.
 946 * @len: the maximum length to copy.
 947 *
 948 * Returns the bytes copied <= len or a negative errno.
 949 */
 950ssize_t vringh_iov_pull_kern(struct vringh_kiov *riov, void *dst, size_t len)
 951{
 952	return vringh_iov_xfer(riov, dst, len, xfer_kern);
 953}
 954EXPORT_SYMBOL(vringh_iov_pull_kern);
 955
 956/**
 957 * vringh_iov_push_kern - copy bytes into vring_iov.
 958 * @wiov: the wiov as passed to vringh_getdesc_kern() (updated as we consume)
 959 * @dst: the place to copy.
 960 * @len: the maximum length to copy.
 961 *
 962 * Returns the bytes copied <= len or a negative errno.
 963 */
 964ssize_t vringh_iov_push_kern(struct vringh_kiov *wiov,
 965			     const void *src, size_t len)
 966{
 967	return vringh_iov_xfer(wiov, (void *)src, len, kern_xfer);
 968}
 969EXPORT_SYMBOL(vringh_iov_push_kern);
 970
 971/**
 972 * vringh_abandon_kern - we've decided not to handle the descriptor(s).
 973 * @vrh: the vring.
 974 * @num: the number of descriptors to put back (ie. num
 975 *	 vringh_get_kern() to undo).
 976 *
 977 * The next vringh_get_kern() will return the old descriptor(s) again.
 978 */
 979void vringh_abandon_kern(struct vringh *vrh, unsigned int num)
 980{
 981	/* We only update vring_avail_event(vr) when we want to be notified,
 982	 * so we haven't changed that yet. */
 983	vrh->last_avail_idx -= num;
 984}
 985EXPORT_SYMBOL(vringh_abandon_kern);
 986
 987/**
 988 * vringh_complete_kern - we've finished with descriptor, publish it.
 989 * @vrh: the vring.
 990 * @head: the head as filled in by vringh_getdesc_kern.
 991 * @len: the length of data we have written.
 992 *
 993 * You should check vringh_need_notify_kern() after one or more calls
 994 * to this function.
 995 */
 996int vringh_complete_kern(struct vringh *vrh, u16 head, u32 len)
 997{
 998	struct vring_used_elem used;
 999
1000	used.id = cpu_to_vringh32(vrh, head);
1001	used.len = cpu_to_vringh32(vrh, len);
1002
1003	return __vringh_complete(vrh, &used, 1, putu16_kern, putused_kern);
1004}
1005EXPORT_SYMBOL(vringh_complete_kern);
1006
1007/**
1008 * vringh_notify_enable_kern - we want to know if something changes.
1009 * @vrh: the vring.
1010 *
1011 * This always enables notifications, but returns false if there are
1012 * now more buffers available in the vring.
1013 */
1014bool vringh_notify_enable_kern(struct vringh *vrh)
1015{
1016	return __vringh_notify_enable(vrh, getu16_kern, putu16_kern);
1017}
1018EXPORT_SYMBOL(vringh_notify_enable_kern);
1019
1020/**
1021 * vringh_notify_disable_kern - don't tell us if something changes.
1022 * @vrh: the vring.
1023 *
1024 * This is our normal running state: we disable and then only enable when
1025 * we're going to sleep.
1026 */
1027void vringh_notify_disable_kern(struct vringh *vrh)
1028{
1029	__vringh_notify_disable(vrh, putu16_kern);
1030}
1031EXPORT_SYMBOL(vringh_notify_disable_kern);
1032
1033/**
1034 * vringh_need_notify_kern - must we tell the other side about used buffers?
1035 * @vrh: the vring we've called vringh_complete_kern() on.
1036 *
1037 * Returns -errno or 0 if we don't need to tell the other side, 1 if we do.
1038 */
1039int vringh_need_notify_kern(struct vringh *vrh)
1040{
1041	return __vringh_need_notify(vrh, getu16_kern);
1042}
1043EXPORT_SYMBOL(vringh_need_notify_kern);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1044
1045MODULE_LICENSE("GPL");