1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660
|
// Copyright 2018 The gVisor Authors.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package mm
import (
"gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/errors/linuxerr"
"gvisor.dev/gvisor/pkg/hostarch"
"gvisor.dev/gvisor/pkg/safemem"
"gvisor.dev/gvisor/pkg/sentry/platform"
"gvisor.dev/gvisor/pkg/usermem"
)
// There are two supported ways to copy data to/from application virtual
// memory:
//
// 1. Internally-mapped copying: Determine the platform.File that backs the
// copied-to/from virtual address, obtain a mapping of its pages, and read or
// write to the mapping.
//
// 2. AddressSpace copying: If platform.Platform.SupportsAddressSpaceIO() is
// true, AddressSpace permissions are applicable, and an AddressSpace is
// available, copy directly through the AddressSpace, handling faults as
// needed.
//
// (Given that internally-mapped copying requires that backing memory is always
// implemented using a host file descriptor, we could also preadv/pwritev to it
// instead. But this would incur a host syscall for each use of the mapped
// page, whereas mmap is a one-time cost.)
//
// The fixed overhead of internally-mapped copying is expected to be higher
// than that of AddressSpace copying since the former always needs to translate
// addresses, whereas the latter only needs to do so when faults occur.
// However, the throughput of internally-mapped copying is expected to be
// somewhat higher than that of AddressSpace copying due to the high cost of
// page faults and because implementations of the latter usually rely on
// safecopy, which doesn't use AVX registers. So we prefer to use AddressSpace
// copying (when available) for smaller copies, and switch to internally-mapped
// copying once a size threshold is exceeded.
const (
// copyMapMinBytes is the size threshold for switching to internally-mapped
// copying in CopyOut, CopyIn, and ZeroOut.
copyMapMinBytes = 32 << 10 // 32 KB
// rwMapMinBytes is the size threshold for switching to internally-mapped
// copying in CopyOutFrom and CopyInTo. It's lower than copyMapMinBytes
// since AddressSpace copying in this case requires additional buffering;
// see CopyOutFrom for details.
rwMapMinBytes = 512
)
// CheckIORange is similar to hostarch.Addr.ToRange, but applies bounds checks
// consistent with Linux's arch/x86/include/asm/uaccess.h:access_ok().
//
// Preconditions: length >= 0.
func (mm *MemoryManager) CheckIORange(addr hostarch.Addr, length int64) (hostarch.AddrRange, bool) {
// Note that access_ok() constrains end even if length == 0.
ar, ok := addr.ToRange(uint64(length))
return ar, (ok && ar.End <= mm.layout.MaxAddr)
}
// checkIOVec applies bound checks consistent with Linux's
// arch/x86/include/asm/uaccess.h:access_ok() to ars.
func (mm *MemoryManager) checkIOVec(ars hostarch.AddrRangeSeq) bool {
for !ars.IsEmpty() {
ar := ars.Head()
if _, ok := mm.CheckIORange(ar.Start, int64(ar.Length())); !ok {
return false
}
ars = ars.Tail()
}
return true
}
func (mm *MemoryManager) asioEnabled(opts usermem.IOOpts) bool {
return mm.haveASIO && !opts.IgnorePermissions && opts.AddressSpaceActive
}
// translateIOError converts errors to EFAULT, as is usually reported for all
// I/O errors originating from MM in Linux.
func translateIOError(ctx context.Context, err error) error {
if err == nil {
return nil
}
if logIOErrors {
ctx.Debugf("MM I/O error: %v", err)
}
return linuxerr.EFAULT
}
// CopyOut implements usermem.IO.CopyOut.
func (mm *MemoryManager) CopyOut(ctx context.Context, addr hostarch.Addr, src []byte, opts usermem.IOOpts) (int, error) {
ar, ok := mm.CheckIORange(addr, int64(len(src)))
if !ok {
return 0, linuxerr.EFAULT
}
if len(src) == 0 {
return 0, nil
}
// Do AddressSpace IO if applicable.
if mm.asioEnabled(opts) && len(src) < copyMapMinBytes {
return mm.asCopyOut(ctx, addr, src)
}
// Go through internal mappings.
n64, err := mm.withInternalMappings(ctx, ar, hostarch.Write, opts.IgnorePermissions, func(ims safemem.BlockSeq) (uint64, error) {
n, err := safemem.CopySeq(ims, safemem.BlockSeqOf(safemem.BlockFromSafeSlice(src)))
return n, translateIOError(ctx, err)
})
return int(n64), err
}
func (mm *MemoryManager) asCopyOut(ctx context.Context, addr hostarch.Addr, src []byte) (int, error) {
var done int
for {
n, err := mm.as.CopyOut(addr+hostarch.Addr(done), src[done:])
done += n
if err == nil {
return done, nil
}
if f, ok := err.(platform.SegmentationFault); ok {
ar, _ := addr.ToRange(uint64(len(src)))
if err := mm.handleASIOFault(ctx, f.Addr, ar, hostarch.Write); err != nil {
return done, err
}
continue
}
return done, translateIOError(ctx, err)
}
}
// CopyIn implements usermem.IO.CopyIn.
func (mm *MemoryManager) CopyIn(ctx context.Context, addr hostarch.Addr, dst []byte, opts usermem.IOOpts) (int, error) {
ar, ok := mm.CheckIORange(addr, int64(len(dst)))
if !ok {
return 0, linuxerr.EFAULT
}
if len(dst) == 0 {
return 0, nil
}
// Do AddressSpace IO if applicable.
if mm.asioEnabled(opts) && len(dst) < copyMapMinBytes {
return mm.asCopyIn(ctx, addr, dst)
}
// Go through internal mappings.
n64, err := mm.withInternalMappings(ctx, ar, hostarch.Read, opts.IgnorePermissions, func(ims safemem.BlockSeq) (uint64, error) {
n, err := safemem.CopySeq(safemem.BlockSeqOf(safemem.BlockFromSafeSlice(dst)), ims)
return n, translateIOError(ctx, err)
})
return int(n64), err
}
func (mm *MemoryManager) asCopyIn(ctx context.Context, addr hostarch.Addr, dst []byte) (int, error) {
var done int
for {
n, err := mm.as.CopyIn(addr+hostarch.Addr(done), dst[done:])
done += n
if err == nil {
return done, nil
}
if f, ok := err.(platform.SegmentationFault); ok {
ar, _ := addr.ToRange(uint64(len(dst)))
if err := mm.handleASIOFault(ctx, f.Addr, ar, hostarch.Read); err != nil {
return done, err
}
continue
}
return done, translateIOError(ctx, err)
}
}
// ZeroOut implements usermem.IO.ZeroOut.
func (mm *MemoryManager) ZeroOut(ctx context.Context, addr hostarch.Addr, toZero int64, opts usermem.IOOpts) (int64, error) {
ar, ok := mm.CheckIORange(addr, toZero)
if !ok {
return 0, linuxerr.EFAULT
}
if toZero == 0 {
return 0, nil
}
// Do AddressSpace IO if applicable.
if mm.asioEnabled(opts) && toZero < copyMapMinBytes {
return mm.asZeroOut(ctx, addr, toZero)
}
// Go through internal mappings.
return mm.withInternalMappings(ctx, ar, hostarch.Write, opts.IgnorePermissions, func(dsts safemem.BlockSeq) (uint64, error) {
n, err := safemem.ZeroSeq(dsts)
return n, translateIOError(ctx, err)
})
}
func (mm *MemoryManager) asZeroOut(ctx context.Context, addr hostarch.Addr, toZero int64) (int64, error) {
var done int64
for {
n, err := mm.as.ZeroOut(addr+hostarch.Addr(done), uintptr(toZero-done))
done += int64(n)
if err == nil {
return done, nil
}
if f, ok := err.(platform.SegmentationFault); ok {
ar, _ := addr.ToRange(uint64(toZero))
if err := mm.handleASIOFault(ctx, f.Addr, ar, hostarch.Write); err != nil {
return done, err
}
continue
}
return done, translateIOError(ctx, err)
}
}
// CopyOutFrom implements usermem.IO.CopyOutFrom.
func (mm *MemoryManager) CopyOutFrom(ctx context.Context, ars hostarch.AddrRangeSeq, src safemem.Reader, opts usermem.IOOpts) (int64, error) {
if !mm.checkIOVec(ars) {
return 0, linuxerr.EFAULT
}
if ars.NumBytes() == 0 {
return 0, nil
}
// Do AddressSpace IO if applicable.
if mm.asioEnabled(opts) && ars.NumBytes() < rwMapMinBytes {
// We have to introduce a buffered copy, instead of just passing a
// safemem.BlockSeq representing addresses in the AddressSpace to src.
// This is because usermem.IO.CopyOutFrom() guarantees that it calls
// src.ReadToBlocks() at most once, which is incompatible with handling
// faults between calls. In the future, this is probably best resolved
// by introducing a CopyOutFrom variant or option that allows it to
// call src.ReadToBlocks() any number of times.
//
// This issue applies to CopyInTo as well.
buf := make([]byte, int(ars.NumBytes()))
bufN, bufErr := src.ReadToBlocks(safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf)))
var done int64
for done < int64(bufN) {
ar := ars.Head()
cplen := int64(ar.Length())
if cplen > int64(bufN)-done {
cplen = int64(bufN) - done
}
n, err := mm.asCopyOut(ctx, ar.Start, buf[int(done):int(done+cplen)])
done += int64(n)
if err != nil {
return done, err
}
ars = ars.Tail()
}
// Do not convert errors returned by src to EFAULT.
return done, bufErr
}
// Go through internal mappings.
return mm.withVecInternalMappings(ctx, ars, hostarch.Write, opts.IgnorePermissions, src.ReadToBlocks)
}
// CopyInTo implements usermem.IO.CopyInTo.
func (mm *MemoryManager) CopyInTo(ctx context.Context, ars hostarch.AddrRangeSeq, dst safemem.Writer, opts usermem.IOOpts) (int64, error) {
if !mm.checkIOVec(ars) {
return 0, linuxerr.EFAULT
}
if ars.NumBytes() == 0 {
return 0, nil
}
// Do AddressSpace IO if applicable.
if mm.asioEnabled(opts) && ars.NumBytes() < rwMapMinBytes {
buf := make([]byte, int(ars.NumBytes()))
var done int
var bufErr error
for !ars.IsEmpty() {
ar := ars.Head()
var n int
n, bufErr = mm.asCopyIn(ctx, ar.Start, buf[done:done+int(ar.Length())])
done += n
if bufErr != nil {
break
}
ars = ars.Tail()
}
n, err := dst.WriteFromBlocks(safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf[:done])))
if err != nil {
return int64(n), err
}
// Do not convert errors returned by dst to EFAULT.
return int64(n), bufErr
}
// Go through internal mappings.
return mm.withVecInternalMappings(ctx, ars, hostarch.Read, opts.IgnorePermissions, dst.WriteFromBlocks)
}
// EnsurePMAsExist attempts to ensure that PMAs exist for the given addr with the
// requested length. It returns the length to which it was able to either
// initialize PMAs for, or ascertain that PMAs exist for. If this length is
// smaller than the requested length it returns an error explaining why.
func (mm *MemoryManager) EnsurePMAsExist(ctx context.Context, addr hostarch.Addr, length int64, opts usermem.IOOpts) (int64, error) {
ar, ok := mm.CheckIORange(addr, length)
if !ok {
return 0, linuxerr.EFAULT
}
n64, err := mm.withInternalMappings(ctx, ar, hostarch.Write, opts.IgnorePermissions, func(ims safemem.BlockSeq) (uint64, error) {
return uint64(ims.NumBytes()), nil
})
return int64(n64), err
}
// SwapUint32 implements usermem.IO.SwapUint32.
func (mm *MemoryManager) SwapUint32(ctx context.Context, addr hostarch.Addr, new uint32, opts usermem.IOOpts) (uint32, error) {
ar, ok := mm.CheckIORange(addr, 4)
if !ok {
return 0, linuxerr.EFAULT
}
// Do AddressSpace IO if applicable.
if mm.haveASIO && opts.AddressSpaceActive && !opts.IgnorePermissions {
for {
old, err := mm.as.SwapUint32(addr, new)
if err == nil {
return old, nil
}
if f, ok := err.(platform.SegmentationFault); ok {
if err := mm.handleASIOFault(ctx, f.Addr, ar, hostarch.ReadWrite); err != nil {
return 0, err
}
continue
}
return 0, translateIOError(ctx, err)
}
}
// Go through internal mappings.
var old uint32
_, err := mm.withInternalMappings(ctx, ar, hostarch.ReadWrite, opts.IgnorePermissions, func(ims safemem.BlockSeq) (uint64, error) {
if ims.NumBlocks() != 1 || ims.NumBytes() != 4 {
// Atomicity is unachievable across mappings.
return 0, linuxerr.EFAULT
}
im := ims.Head()
var err error
old, err = safemem.SwapUint32(im, new)
if err != nil {
return 0, translateIOError(ctx, err)
}
// Return the number of bytes read.
return 4, nil
})
return old, err
}
// CompareAndSwapUint32 implements usermem.IO.CompareAndSwapUint32.
func (mm *MemoryManager) CompareAndSwapUint32(ctx context.Context, addr hostarch.Addr, old, new uint32, opts usermem.IOOpts) (uint32, error) {
ar, ok := mm.CheckIORange(addr, 4)
if !ok {
return 0, linuxerr.EFAULT
}
// Do AddressSpace IO if applicable.
if mm.haveASIO && opts.AddressSpaceActive && !opts.IgnorePermissions {
for {
prev, err := mm.as.CompareAndSwapUint32(addr, old, new)
if err == nil {
return prev, nil
}
if f, ok := err.(platform.SegmentationFault); ok {
if err := mm.handleASIOFault(ctx, f.Addr, ar, hostarch.ReadWrite); err != nil {
return 0, err
}
continue
}
return 0, translateIOError(ctx, err)
}
}
// Go through internal mappings.
var prev uint32
_, err := mm.withInternalMappings(ctx, ar, hostarch.ReadWrite, opts.IgnorePermissions, func(ims safemem.BlockSeq) (uint64, error) {
if ims.NumBlocks() != 1 || ims.NumBytes() != 4 {
// Atomicity is unachievable across mappings.
return 0, linuxerr.EFAULT
}
im := ims.Head()
var err error
prev, err = safemem.CompareAndSwapUint32(im, old, new)
if err != nil {
return 0, translateIOError(ctx, err)
}
// Return the number of bytes read.
return 4, nil
})
return prev, err
}
// LoadUint32 implements usermem.IO.LoadUint32.
func (mm *MemoryManager) LoadUint32(ctx context.Context, addr hostarch.Addr, opts usermem.IOOpts) (uint32, error) {
ar, ok := mm.CheckIORange(addr, 4)
if !ok {
return 0, linuxerr.EFAULT
}
// Do AddressSpace IO if applicable.
if mm.haveASIO && opts.AddressSpaceActive && !opts.IgnorePermissions {
for {
val, err := mm.as.LoadUint32(addr)
if err == nil {
return val, nil
}
if f, ok := err.(platform.SegmentationFault); ok {
if err := mm.handleASIOFault(ctx, f.Addr, ar, hostarch.Read); err != nil {
return 0, err
}
continue
}
return 0, translateIOError(ctx, err)
}
}
// Go through internal mappings.
var val uint32
_, err := mm.withInternalMappings(ctx, ar, hostarch.Read, opts.IgnorePermissions, func(ims safemem.BlockSeq) (uint64, error) {
if ims.NumBlocks() != 1 || ims.NumBytes() != 4 {
// Atomicity is unachievable across mappings.
return 0, linuxerr.EFAULT
}
im := ims.Head()
var err error
val, err = safemem.LoadUint32(im)
if err != nil {
return 0, translateIOError(ctx, err)
}
// Return the number of bytes read.
return 4, nil
})
return val, err
}
// handleASIOFault handles a page fault at address addr for an AddressSpaceIO
// operation spanning ioar.
//
// Preconditions:
// - mm.as != nil.
// - ioar.Length() != 0.
// - ioar.Contains(addr).
func (mm *MemoryManager) handleASIOFault(ctx context.Context, addr hostarch.Addr, ioar hostarch.AddrRange, at hostarch.AccessType) error {
// Try to map all remaining pages in the I/O operation. This RoundUp can't
// overflow because otherwise it would have been caught by CheckIORange.
end, _ := ioar.End.RoundUp()
ar := hostarch.AddrRange{addr.RoundDown(), end}
// Don't bother trying existingPMAsLocked; in most cases, if we did have
// existing pmas, we wouldn't have faulted.
// Ensure that we have usable vmas. Here and below, only return early if we
// can't map the first (faulting) page; failure to map later pages are
// silently ignored. This maximizes partial success.
mm.mappingMu.RLock()
vseg, vend, err := mm.getVMAsLocked(ctx, ar, at, false)
if vendaddr := vend.Start(); vendaddr < ar.End {
if vendaddr <= ar.Start {
mm.mappingMu.RUnlock()
return translateIOError(ctx, err)
}
ar.End = vendaddr
}
// Ensure that we have usable pmas.
mm.activeMu.Lock()
pseg, pend, err := mm.getPMAsLocked(ctx, vseg, ar, at)
mm.mappingMu.RUnlock()
if pendaddr := pend.Start(); pendaddr < ar.End {
if pendaddr <= ar.Start {
mm.activeMu.Unlock()
return translateIOError(ctx, err)
}
ar.End = pendaddr
}
// Downgrade to a read-lock on activeMu since we don't need to mutate pmas
// anymore.
mm.activeMu.DowngradeLock()
err = mm.mapASLocked(pseg, ar, false)
mm.activeMu.RUnlock()
return translateIOError(ctx, err)
}
// withInternalMappings ensures that pmas exist for all addresses in ar,
// support access of type (at, ignorePermissions), and have internal mappings
// cached. It then calls f with mm.activeMu locked for reading, passing
// internal mappings for the subrange of ar for which this property holds.
//
// withInternalMappings takes a function returning uint64 since many safemem
// functions have this property, but returns an int64 since this is usually
// more useful for usermem.IO methods.
//
// Preconditions: 0 < ar.Length() <= math.MaxInt64.
func (mm *MemoryManager) withInternalMappings(ctx context.Context, ar hostarch.AddrRange, at hostarch.AccessType, ignorePermissions bool, f func(safemem.BlockSeq) (uint64, error)) (int64, error) {
// If pmas are already available, we can do IO without touching mm.vmas or
// mm.mappingMu.
mm.activeMu.RLock()
if pseg := mm.existingPMAsLocked(ar, at, ignorePermissions, true /* needInternalMappings */); pseg.Ok() {
n, err := f(mm.internalMappingsLocked(pseg, ar))
mm.activeMu.RUnlock()
// Do not convert errors returned by f to EFAULT.
return int64(n), err
}
mm.activeMu.RUnlock()
// Ensure that we have usable vmas.
mm.mappingMu.RLock()
vseg, vend, verr := mm.getVMAsLocked(ctx, ar, at, ignorePermissions)
if vendaddr := vend.Start(); vendaddr < ar.End {
if vendaddr <= ar.Start {
mm.mappingMu.RUnlock()
return 0, translateIOError(ctx, verr)
}
ar.End = vendaddr
}
// Ensure that we have usable pmas.
mm.activeMu.Lock()
pseg, pend, perr := mm.getPMAsLocked(ctx, vseg, ar, at)
mm.mappingMu.RUnlock()
if pendaddr := pend.Start(); pendaddr < ar.End {
if pendaddr <= ar.Start {
mm.activeMu.Unlock()
return 0, translateIOError(ctx, perr)
}
ar.End = pendaddr
}
imend, imerr := mm.getPMAInternalMappingsLocked(pseg, ar)
mm.activeMu.DowngradeLock()
if imendaddr := imend.Start(); imendaddr < ar.End {
if imendaddr <= ar.Start {
mm.activeMu.RUnlock()
return 0, translateIOError(ctx, imerr)
}
ar.End = imendaddr
}
// Do I/O.
un, err := f(mm.internalMappingsLocked(pseg, ar))
mm.activeMu.RUnlock()
n := int64(un)
// Return the first error in order of progress through ar.
if err != nil {
// Do not convert errors returned by f to EFAULT.
return n, err
}
if imerr != nil {
return n, translateIOError(ctx, imerr)
}
if perr != nil {
return n, translateIOError(ctx, perr)
}
return n, translateIOError(ctx, verr)
}
// withVecInternalMappings ensures that pmas exist for all addresses in ars,
// support access of type (at, ignorePermissions), and have internal mappings
// cached. It then calls f with mm.activeMu locked for reading, passing
// internal mappings for the subset of ars for which this property holds.
//
// Preconditions: !ars.IsEmpty().
func (mm *MemoryManager) withVecInternalMappings(ctx context.Context, ars hostarch.AddrRangeSeq, at hostarch.AccessType, ignorePermissions bool, f func(safemem.BlockSeq) (uint64, error)) (int64, error) {
// withInternalMappings is faster than withVecInternalMappings because of
// iterator plumbing (this isn't generally practical in the vector case due
// to iterator invalidation between AddrRanges). Use it if possible.
if ars.NumRanges() == 1 {
return mm.withInternalMappings(ctx, ars.Head(), at, ignorePermissions, f)
}
// If pmas are already available, we can do IO without touching mm.vmas or
// mm.mappingMu.
mm.activeMu.RLock()
if mm.existingVecPMAsLocked(ars, at, ignorePermissions, true /* needInternalMappings */) {
n, err := f(mm.vecInternalMappingsLocked(ars))
mm.activeMu.RUnlock()
// Do not convert errors returned by f to EFAULT.
return int64(n), err
}
mm.activeMu.RUnlock()
// Ensure that we have usable vmas.
mm.mappingMu.RLock()
vars, verr := mm.getVecVMAsLocked(ctx, ars, at, ignorePermissions)
if vars.NumBytes() == 0 {
mm.mappingMu.RUnlock()
return 0, translateIOError(ctx, verr)
}
// Ensure that we have usable pmas.
mm.activeMu.Lock()
pars, perr := mm.getVecPMAsLocked(ctx, vars, at)
mm.mappingMu.RUnlock()
if pars.NumBytes() == 0 {
mm.activeMu.Unlock()
return 0, translateIOError(ctx, perr)
}
imars, imerr := mm.getVecPMAInternalMappingsLocked(pars)
mm.activeMu.DowngradeLock()
if imars.NumBytes() == 0 {
mm.activeMu.RUnlock()
return 0, translateIOError(ctx, imerr)
}
// Do I/O.
un, err := f(mm.vecInternalMappingsLocked(imars))
mm.activeMu.RUnlock()
n := int64(un)
// Return the first error in order of progress through ars.
if err != nil {
// Do not convert errors from f to EFAULT.
return n, err
}
if imerr != nil {
return n, translateIOError(ctx, imerr)
}
if perr != nil {
return n, translateIOError(ctx, perr)
}
return n, translateIOError(ctx, verr)
}
// truncatedAddrRangeSeq returns a copy of ars, but with the end truncated to
// at most address end on AddrRange arsit.Head(). It is used in vector I/O paths to
// truncate hostarch.AddrRangeSeq when errors occur.
//
// Preconditions:
// - !arsit.IsEmpty().
// - end <= arsit.Head().End.
func truncatedAddrRangeSeq(ars, arsit hostarch.AddrRangeSeq, end hostarch.Addr) hostarch.AddrRangeSeq {
ar := arsit.Head()
if end <= ar.Start {
return ars.TakeFirst64(ars.NumBytes() - arsit.NumBytes())
}
return ars.TakeFirst64(ars.NumBytes() - arsit.NumBytes() + int64(end-ar.Start))
}
|