sys/kern/subr_uio.c

   1 /*-
   2  * Copyright (c) 1982, 1986, 1991, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  * (c) UNIX System Laboratories, Inc.
   5  * All or some portions of this file are derived from material licensed
   6  * to the University of California by American Telephone and Telegraph
   7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
   8  * the permission of UNIX System Laboratories, Inc.
   9  *
  10  * Copyright (c) 2014 The FreeBSD Foundation
  11  *
  12  * Portions of this software were developed by Konstantin Belousov
  13  * under sponsorship from the FreeBSD Foundation.
  14  *
  15  * Redistribution and use in source and binary forms, with or without
  16  * modification, are permitted provided that the following conditions
  17  * are met:
  18  * 1. Redistributions of source code must retain the above copyright
  19  *    notice, this list of conditions and the following disclaimer.
  20  * 2. Redistributions in binary form must reproduce the above copyright
  21  *    notice, this list of conditions and the following disclaimer in the
  22  *    documentation and/or other materials provided with the distribution.
  23  * 4. Neither the name of the University nor the names of its contributors
  24  *    may be used to endorse or promote products derived from this software
  25  *    without specific prior written permission.
  26  *
  27  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  28  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  29  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  30  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  31  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  35  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  36  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  37  * SUCH DAMAGE.
  38  *
  39  *      @(#)kern_subr.c 8.3 (Berkeley) 1/21/94
  40  */
  41
  42 #include <sys/cdefs.h>
  43 __FBSDID("$FreeBSD$");
  44
  45 #include <sys/param.h>
  46 #include <sys/systm.h>
  47 #include <sys/kernel.h>
  48 #include <sys/limits.h>
  49 #include <sys/lock.h>
  50 #include <sys/mman.h>
  51 #include <sys/proc.h>
  52 #include <sys/resourcevar.h>
  53 #include <sys/rwlock.h>
  54 #include <sys/sched.h>
  55 #include <sys/sysctl.h>
  56 #include <sys/vnode.h>
  57
  58 #include <vm/vm.h>
  59 #include <vm/vm_param.h>
  60 #include <vm/vm_extern.h>
  61 #include <vm/vm_page.h>
  62 #include <vm/vm_pageout.h>
  63 #include <vm/vm_map.h>
  64
  65 #include <machine/bus.h>
  66
  67 SYSCTL_INT(_kern, KERN_IOV_MAX, iov_max, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, UIO_MAXIOV,
  68         "Maximum number of elements in an I/O vector; sysconf(_SC_IOV_MAX)");
  69
  70 static int uiomove_faultflag(void *cp, int n, struct uio *uio, int nofault);
  71
  72 int
  73 copyin_nofault(const void *udaddr, void *kaddr, size_t len)
  74 {
  75         int error, save;
  76
  77         save = vm_fault_disable_pagefaults();
  78         error = copyin(udaddr, kaddr, len);
  79         vm_fault_enable_pagefaults(save);
  80         return (error);
  81 }
  82
  83 int
  84 copyout_nofault(const void *kaddr, void *udaddr, size_t len)
  85 {
  86         int error, save;
  87
  88         save = vm_fault_disable_pagefaults();
  89         error = copyout(kaddr, udaddr, len);
  90         vm_fault_enable_pagefaults(save);
  91         return (error);
  92 }
  93
  94 #define PHYS_PAGE_COUNT(len)    (howmany(len, PAGE_SIZE) + 1)
  95
  96 int
  97 physcopyin(void *src, vm_paddr_t dst, size_t len)
  98 {
  99         vm_page_t m[PHYS_PAGE_COUNT(len)];
 100         struct iovec iov[1];
 101         struct uio uio;
 102         int i;
 103
 104         iov[0].iov_base = src;
 105         iov[0].iov_len = len;
 106         uio.uio_iov = iov;
 107         uio.uio_iovcnt = 1;
 108         uio.uio_offset = 0;
 109         uio.uio_resid = len;
 110         uio.uio_segflg = UIO_SYSSPACE;
 111         uio.uio_rw = UIO_WRITE;
 112         for (i = 0; i < PHYS_PAGE_COUNT(len); i++, dst += PAGE_SIZE)
 113                 m[i] = PHYS_TO_VM_PAGE(dst);
 114         return (uiomove_fromphys(m, dst & PAGE_MASK, len, &uio));
 115 }
 116
 117 int
 118 physcopyout(vm_paddr_t src, void *dst, size_t len)
 119 {
 120         vm_page_t m[PHYS_PAGE_COUNT(len)];
 121         struct iovec iov[1];
 122         struct uio uio;
 123         int i;
 124
 125         iov[0].iov_base = dst;
 126         iov[0].iov_len = len;
 127         uio.uio_iov = iov;
 128         uio.uio_iovcnt = 1;
 129         uio.uio_offset = 0;
 130         uio.uio_resid = len;
 131         uio.uio_segflg = UIO_SYSSPACE;
 132         uio.uio_rw = UIO_READ;
 133         for (i = 0; i < PHYS_PAGE_COUNT(len); i++, src += PAGE_SIZE)
 134                 m[i] = PHYS_TO_VM_PAGE(src);
 135         return (uiomove_fromphys(m, src & PAGE_MASK, len, &uio));
 136 }
 137
 138 #undef PHYS_PAGE_COUNT
 139
 140 int
 141 physcopyin_vlist(bus_dma_segment_t *src, off_t offset, vm_paddr_t dst,
 142     size_t len)
 143 {
 144         size_t seg_len;
 145         int error;
 146
 147         error = 0;
 148         while (offset >= src->ds_len) {
 149                 offset -= src->ds_len;
 150                 src++;
 151         }
 152
 153         while (len > 0 && error == 0) {
 154                 seg_len = MIN(src->ds_len - offset, len);
 155                 error = physcopyin((void *)(uintptr_t)(src->ds_addr + offset),
 156                     dst, seg_len);
 157                 offset = 0;
 158                 src++;
 159                 len -= seg_len;
 160                 dst += seg_len;
 161         }
 162
 163         return (error);
 164 }
 165
 166 int
 167 physcopyout_vlist(vm_paddr_t src, bus_dma_segment_t *dst, off_t offset,
 168     size_t len)
 169 {
 170         size_t seg_len;
 171         int error;
 172
 173         error = 0;
 174         while (offset >= dst->ds_len) {
 175                 offset -= dst->ds_len;
 176                 dst++;
 177         }
 178
 179         while (len > 0 && error == 0) {
 180                 seg_len = MIN(dst->ds_len - offset, len);
 181                 error = physcopyout(src, (void *)(uintptr_t)(dst->ds_addr +
 182                     offset), seg_len);
 183                 offset = 0;
 184                 dst++;
 185                 len -= seg_len;
 186                 src += seg_len;
 187         }
 188
 189         return (error);
 190 }
 191
 192 int
 193 uiomove(void *cp, int n, struct uio *uio)
 194 {
 195
 196         return (uiomove_faultflag(cp, n, uio, 0));
 197 }
 198
 199 int
 200 uiomove_nofault(void *cp, int n, struct uio *uio)
 201 {
 202
 203         return (uiomove_faultflag(cp, n, uio, 1));
 204 }
 205
 206 static int
 207 uiomove_faultflag(void *cp, int n, struct uio *uio, int nofault)
 208 {
 209         struct thread *td;
 210         struct iovec *iov;
 211         size_t cnt;
 212         int error, newflags, save;
 213
 214         td = curthread;
 215         error = 0;
 216
 217         KASSERT(uio->uio_rw == UIO_READ || uio->uio_rw == UIO_WRITE,
 218             ("uiomove: mode"));
 219         KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == td,
 220             ("uiomove proc"));
 221         if (!nofault)
 222                 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
 223                     "Calling uiomove()");
 224
 225         /* XXX does it make a sense to set TDP_DEADLKTREAT for UIO_SYSSPACE ? */
 226         newflags = TDP_DEADLKTREAT;
 227         if (uio->uio_segflg == UIO_USERSPACE && nofault) {
 228                 /*
 229                  * Fail if a non-spurious page fault occurs.
 230                  */
 231                 newflags |= TDP_NOFAULTING | TDP_RESETSPUR;
 232         }
 233         save = curthread_pflags_set(newflags);
 234
 235         while (n > 0 && uio->uio_resid) {
 236                 iov = uio->uio_iov;
 237                 cnt = iov->iov_len;
 238                 if (cnt == 0) {
 239                         uio->uio_iov++;
 240                         uio->uio_iovcnt--;
 241                         continue;
 242                 }
 243                 if (cnt > n)
 244                         cnt = n;
 245
 246                 switch (uio->uio_segflg) {
 247
 248                 case UIO_USERSPACE:
 249                         maybe_yield();
 250                         if (uio->uio_rw == UIO_READ)
 251                                 error = copyout(cp, iov->iov_base, cnt);
 252                         else
 253                                 error = copyin(iov->iov_base, cp, cnt);
 254                         if (error)
 255                                 goto out;
 256                         break;
 257
 258                 case UIO_SYSSPACE:
 259                         if (uio->uio_rw == UIO_READ)
 260                                 bcopy(cp, iov->iov_base, cnt);
 261                         else
 262                                 bcopy(iov->iov_base, cp, cnt);
 263                         break;
 264                 case UIO_NOCOPY:
 265                         break;
 266                 }
 267                 iov->iov_base = (char *)iov->iov_base + cnt;
 268                 iov->iov_len -= cnt;
 269                 uio->uio_resid -= cnt;
 270                 uio->uio_offset += cnt;
 271                 cp = (char *)cp + cnt;
 272                 n -= cnt;
 273         }
 274 out:
 275         curthread_pflags_restore(save);
 276         return (error);
 277 }
 278
 279 /*
 280  * Wrapper for uiomove() that validates the arguments against a known-good
 281  * kernel buffer.  Currently, uiomove accepts a signed (n) argument, which
 282  * is almost definitely a bad thing, so we catch that here as well.  We
 283  * return a runtime failure, but it might be desirable to generate a runtime
 284  * assertion failure instead.
 285  */
 286 int
 287 uiomove_frombuf(void *buf, int buflen, struct uio *uio)
 288 {
 289         size_t offset, n;
 290
 291         if (uio->uio_offset < 0 || uio->uio_resid < 0 ||
 292             (offset = uio->uio_offset) != uio->uio_offset)
 293                 return (EINVAL);
 294         if (buflen <= 0 || offset >= buflen)
 295                 return (0);
 296         if ((n = buflen - offset) > IOSIZE_MAX)
 297                 return (EINVAL);
 298         return (uiomove((char *)buf + offset, n, uio));
 299 }
 300
 301 /*
 302  * Give next character to user as result of read.
 303  */
 304 int
 305 ureadc(int c, struct uio *uio)
 306 {
 307         struct iovec *iov;
 308         char *iov_base;
 309
 310         WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
 311             "Calling ureadc()");
 312
 313 again:
 314         if (uio->uio_iovcnt == 0 || uio->uio_resid == 0)
 315                 panic("ureadc");
 316         iov = uio->uio_iov;
 317         if (iov->iov_len == 0) {
 318                 uio->uio_iovcnt--;
 319                 uio->uio_iov++;
 320                 goto again;
 321         }
 322         switch (uio->uio_segflg) {
 323
 324         case UIO_USERSPACE:
 325                 if (subyte(iov->iov_base, c) < 0)
 326                         return (EFAULT);
 327                 break;
 328
 329         case UIO_SYSSPACE:
 330                 iov_base = iov->iov_base;
 331                 *iov_base = c;
 332                 break;
 333
 334         case UIO_NOCOPY:
 335                 break;
 336         }
 337         iov->iov_base = (char *)iov->iov_base + 1;
 338         iov->iov_len--;
 339         uio->uio_resid--;
 340         uio->uio_offset++;
 341         return (0);
 342 }
 343
 344 int
 345 copyinfrom(const void * __restrict src, void * __restrict dst, size_t len,
 346     int seg)
 347 {
 348         int error = 0;
 349
 350         switch (seg) {
 351         case UIO_USERSPACE:
 352                 error = copyin(src, dst, len);
 353                 break;
 354         case UIO_SYSSPACE:
 355                 bcopy(src, dst, len);
 356                 break;
 357         default:
 358                 panic("copyinfrom: bad seg %d\n", seg);
 359         }
 360         return (error);
 361 }
 362
 363 int
 364 copyinstrfrom(const void * __restrict src, void * __restrict dst, size_t len,
 365     size_t * __restrict copied, int seg)
 366 {
 367         int error = 0;
 368
 369         switch (seg) {
 370         case UIO_USERSPACE:
 371                 error = copyinstr(src, dst, len, copied);
 372                 break;
 373         case UIO_SYSSPACE:
 374                 error = copystr(src, dst, len, copied);
 375                 break;
 376         default:
 377                 panic("copyinstrfrom: bad seg %d\n", seg);
 378         }
 379         return (error);
 380 }
 381
 382 int
 383 copyiniov(const struct iovec *iovp, u_int iovcnt, struct iovec **iov, int error)
 384 {
 385         u_int iovlen;
 386
 387         *iov = NULL;
 388         if (iovcnt > UIO_MAXIOV)
 389                 return (error);
 390         iovlen = iovcnt * sizeof (struct iovec);
 391         *iov = malloc(iovlen, M_IOV, M_WAITOK);
 392         error = copyin(iovp, *iov, iovlen);
 393         if (error) {
 394                 free(*iov, M_IOV);
 395                 *iov = NULL;
 396         }
 397         return (error);
 398 }
 399
 400 int
 401 copyinuio(const struct iovec *iovp, u_int iovcnt, struct uio **uiop)
 402 {
 403         struct iovec *iov;
 404         struct uio *uio;
 405         u_int iovlen;
 406         int error, i;
 407
 408         *uiop = NULL;
 409         if (iovcnt > UIO_MAXIOV)
 410                 return (EINVAL);
 411         iovlen = iovcnt * sizeof (struct iovec);
 412         uio = malloc(iovlen + sizeof *uio, M_IOV, M_WAITOK);
 413         iov = (struct iovec *)(uio + 1);
 414         error = copyin(iovp, iov, iovlen);
 415         if (error) {
 416                 free(uio, M_IOV);
 417                 return (error);
 418         }
 419         uio->uio_iov = iov;
 420         uio->uio_iovcnt = iovcnt;
 421         uio->uio_segflg = UIO_USERSPACE;
 422         uio->uio_offset = -1;
 423         uio->uio_resid = 0;
 424         for (i = 0; i < iovcnt; i++) {
 425                 if (iov->iov_len > IOSIZE_MAX - uio->uio_resid) {
 426                         free(uio, M_IOV);
 427                         return (EINVAL);
 428                 }
 429                 uio->uio_resid += iov->iov_len;
 430                 iov++;
 431         }
 432         *uiop = uio;
 433         return (0);
 434 }
 435
 436 struct uio *
 437 cloneuio(struct uio *uiop)
 438 {
 439         struct uio *uio;
 440         int iovlen;
 441
 442         iovlen = uiop->uio_iovcnt * sizeof (struct iovec);
 443         uio = malloc(iovlen + sizeof *uio, M_IOV, M_WAITOK);
 444         *uio = *uiop;
 445         uio->uio_iov = (struct iovec *)(uio + 1);
 446         bcopy(uiop->uio_iov, uio->uio_iov, iovlen);
 447         return (uio);
 448 }
 449
 450 /*
 451  * Map some anonymous memory in user space of size sz, rounded up to the page
 452  * boundary.
 453  */
 454 int
 455 copyout_map(struct thread *td, vm_offset_t *addr, size_t sz)
 456 {
 457         struct vmspace *vms;
 458         int error;
 459         vm_size_t size;
 460
 461         vms = td->td_proc->p_vmspace;
 462
 463         /*
 464          * Map somewhere after heap in process memory.
 465          */
 466         *addr = round_page((vm_offset_t)vms->vm_daddr +
 467             lim_max(td, RLIMIT_DATA));
 468
 469         /* round size up to page boundary */
 470         size = (vm_size_t)round_page(sz);
 471
 472         error = vm_mmap(&vms->vm_map, addr, size, VM_PROT_READ | VM_PROT_WRITE,
 473             VM_PROT_ALL, MAP_PRIVATE | MAP_ANON, OBJT_DEFAULT, NULL, 0);
 474
 475         return (error);
 476 }
 477
 478 /*
 479  * Unmap memory in user space.
 480  */
 481 int
 482 copyout_unmap(struct thread *td, vm_offset_t addr, size_t sz)
 483 {
 484         vm_map_t map;
 485         vm_size_t size;
 486
 487         if (sz == 0)
 488                 return (0);
 489
 490         map = &td->td_proc->p_vmspace->vm_map;
 491         size = (vm_size_t)round_page(sz);
 492
 493         if (vm_map_remove(map, addr, addr + size) != KERN_SUCCESS)
 494                 return (EINVAL);
 495
 496         return (0);
 497 }
 498
 499 #ifdef NO_FUEWORD
 500 /*
 501  * XXXKIB The temporal implementation of fue*() functions which do not
 502  * handle usermode -1 properly, mixing it with the fault code.  Keep
 503  * this until MD code is written.  Currently sparc64 and mips do not
 504  * have proper implementation.
 505  */
 506
 507 int
 508 fueword(volatile const void *base, long *val)
 509 {
 510         long res;
 511
 512         res = fuword(base);
 513         if (res == -1)
 514                 return (-1);
 515         *val = res;
 516         return (0);
 517 }
 518
 519 int
 520 fueword32(volatile const void *base, int32_t *val)
 521 {
 522         int32_t res;
 523
 524         res = fuword32(base);
 525         if (res == -1)
 526                 return (-1);
 527         *val = res;
 528         return (0);
 529 }
 530
 531 #ifdef _LP64
 532 int
 533 fueword64(volatile const void *base, int64_t *val)
 534 {
 535         int32_t res;
 536
 537         res = fuword64(base);
 538         if (res == -1)
 539                 return (-1);
 540         *val = res;
 541         return (0);
 542 }
 543 #endif
 544
 545 int
 546 casueword32(volatile uint32_t *base, uint32_t oldval, uint32_t *oldvalp,
 547     uint32_t newval)
 548 {
 549         int32_t ov;
 550
 551         ov = casuword32(base, oldval, newval);
 552         if (ov == -1)
 553                 return (-1);
 554         *oldvalp = ov;
 555         return (0);
 556 }
 557
 558 int
 559 casueword(volatile u_long *p, u_long oldval, u_long *oldvalp, u_long newval)
 560 {
 561         u_long ov;
 562
 563         ov = casuword(p, oldval, newval);
 564         if (ov == -1)
 565                 return (-1);
 566         *oldvalp = ov;
 567         return (0);
 568 }
 569 #else /* NO_FUEWORD */
 570 int32_t
 571 fuword32(volatile const void *addr)
 572 {
 573         int rv;
 574         int32_t val;
 575
 576         rv = fueword32(addr, &val);
 577         return (rv == -1 ? -1 : val);
 578 }
 579
 580 #ifdef _LP64
 581 int64_t
 582 fuword64(volatile const void *addr)
 583 {
 584         int rv;
 585         int64_t val;
 586
 587         rv = fueword64(addr, &val);
 588         return (rv == -1 ? -1 : val);
 589 }
 590 #endif /* _LP64 */
 591
 592 long
 593 fuword(volatile const void *addr)
 594 {
 595         long val;
 596         int rv;
 597
 598         rv = fueword(addr, &val);
 599         return (rv == -1 ? -1 : val);
 600 }
 601
 602 uint32_t
 603 casuword32(volatile uint32_t *addr, uint32_t old, uint32_t new)
 604 {
 605         int rv;
 606         uint32_t val;
 607
 608         rv = casueword32(addr, old, &val, new);
 609         return (rv == -1 ? -1 : val);
 610 }
 611
 612 u_long
 613 casuword(volatile u_long *addr, u_long old, u_long new)
 614 {
 615         int rv;
 616         u_long val;
 617
 618         rv = casueword(addr, old, &val, new);
 619         return (rv == -1 ? -1 : val);
 620 }
 621
 622 #endif /* NO_FUEWORD */