1 /* 2 * Copyright (c) 2003,2004 The DragonFly Project. All rights reserved. 3 * 4 * This code is derived from software contributed to The DragonFly Project 5 * by Matthew Dillon <dillon@backplane.com> 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * 3. Neither the name of The DragonFly Project nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific, prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * $DragonFly: src/sys/kern/kern_fp.c,v 1.20 2007/01/12 06:06:57 dillon Exp $ 35 */ 36 37 /* 38 * Direct file pointer API functions for in-kernel operations on files. These 39 * functions provide a open/read/write/close like interface within the kernel 40 * for operating on files that are not necessarily associated with processes 41 * and which do not (typically) have descriptors. 42 * 43 * FUTURE: file handle conversion routines to support checkpointing, 44 * and additional file operations (ioctl, fcntl). 45 */ 46 47 #include <sys/param.h> 48 #include <sys/kernel.h> 49 #include <sys/systm.h> 50 #include <sys/malloc.h> 51 #include <sys/sysproto.h> 52 #include <sys/conf.h> 53 #include <sys/filedesc.h> 54 #include <sys/sysctl.h> 55 #include <sys/vnode.h> 56 #include <sys/proc.h> 57 #include <sys/nlookup.h> 58 #include <sys/file.h> 59 #include <sys/stat.h> 60 #include <sys/filio.h> 61 #include <sys/fcntl.h> 62 #include <sys/unistd.h> 63 #include <sys/resourcevar.h> 64 #include <sys/event.h> 65 #include <sys/mman.h> 66 67 #include <vm/vm.h> 68 #include <vm/vm_param.h> 69 #include <sys/lock.h> 70 #include <vm/pmap.h> 71 #include <vm/vm_map.h> 72 #include <vm/vm_object.h> 73 #include <vm/vm_page.h> 74 #include <vm/vm_pager.h> 75 #include <vm/vm_pageout.h> 76 #include <vm/vm_extern.h> 77 #include <vm/vm_page.h> 78 #include <vm/vm_kern.h> 79 80 #include <sys/file2.h> 81 #include <machine/limits.h> 82 83 typedef struct file *file_t; 84 85 /* 86 * fp_open: 87 * 88 * Open a file as specified. Use O_* flags for flags. 89 * 90 * NOTE! O_ROOTCRED not quite working yet, vn_open() asserts that the 91 * cred must match the process's cred. XXX 92 * 93 * NOTE! when fp_open() is called from a pure thread, root creds are 94 * used. 95 */ 96 int 97 fp_open(const char *path, int flags, int mode, file_t *fpp) 98 { 99 struct nlookupdata nd; 100 struct thread *td; 101 struct file *fp; 102 int error; 103 104 if ((error = falloc(NULL, fpp, NULL)) != 0) 105 return (error); 106 fp = *fpp; 107 td = curthread; 108 if (td->td_proc) { 109 if ((flags & O_ROOTCRED) == 0) 110 fsetcred(fp, td->td_proc->p_ucred); 111 } 112 error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_LOCKVP); 113 flags = FFLAGS(flags); 114 if (error == 0) 115 error = vn_open(&nd, fp, flags, mode); 116 nlookup_done(&nd); 117 if (error) { 118 fdrop(fp); 119 *fpp = NULL; 120 } 121 return(error); 122 } 123 124 125 /* 126 * fp_vpopen(): convert a vnode to a file pointer, call VOP_OPEN() on the 127 * the vnode. The vnode must be refd and locked. 128 * 129 * On success the vnode's ref is inherited by the file pointer and the caller 130 * should not vrele() it, and the vnode is unlocked. 131 * 132 * On failure the vnode remains locked and refd and the caller is responsible 133 * for vput()ing it. 134 */ 135 int 136 fp_vpopen(struct vnode *vp, int flags, file_t *fpp) 137 { 138 struct thread *td; 139 struct file *fp; 140 int vmode; 141 int error; 142 143 td = curthread; 144 145 /* 146 * Vnode checks (from vn_open()) 147 */ 148 if (vp->v_type == VLNK) { 149 error = EMLINK; 150 goto bad2; 151 } 152 if (vp->v_type == VSOCK) { 153 error = EOPNOTSUPP; 154 goto bad2; 155 } 156 flags = FFLAGS(flags); 157 vmode = 0; 158 if (flags & (FWRITE | O_TRUNC)) { 159 if (vp->v_type == VDIR) { 160 error = EISDIR; 161 goto bad2; 162 } 163 error = vn_writechk(vp, NULL); 164 if (error) 165 goto bad2; 166 vmode |= VWRITE; 167 } 168 if (flags & FREAD) 169 vmode |= VREAD; 170 if (vmode) { 171 error = VOP_ACCESS(vp, vmode, td->td_proc->p_ucred); 172 if (error) 173 goto bad2; 174 } 175 176 /* 177 * File pointer setup 178 */ 179 if ((error = falloc(NULL, fpp, NULL)) != 0) 180 goto bad2; 181 fp = *fpp; 182 if ((flags & O_ROOTCRED) == 0 && td->td_proc) 183 fsetcred(fp, td->td_proc->p_ucred); 184 185 error = VOP_OPEN(vp, flags, td->td_proc->p_ucred, fp); 186 if (error) 187 goto bad1; 188 189 vput(vp); 190 return (0); 191 bad1: 192 fp->f_ops = &badfileops; /* open failed, don't close */ 193 fp->f_data = NULL; 194 fdrop(fp); 195 /* leave the vnode intact, but fall through and unlock it anyway */ 196 bad2: 197 *fpp = NULL; 198 return (error); 199 } 200 201 /* 202 * fp_*read() is meant to operate like the normal descriptor based syscalls 203 * would. Note that if 'buf' points to user memory a UIO_USERSPACE 204 * transfer will be used. 205 */ 206 int 207 fp_pread(file_t fp, void *buf, size_t nbytes, off_t offset, ssize_t *res, 208 enum uio_seg seg) 209 { 210 struct uio auio; 211 struct iovec aiov; 212 size_t count; 213 int error; 214 215 if (res) 216 *res = 0; 217 if (nbytes > INT_MAX) 218 return (EINVAL); 219 bzero(&auio, sizeof(auio)); 220 aiov.iov_base = (caddr_t)buf; 221 aiov.iov_len = nbytes; 222 auio.uio_iov = &aiov; 223 auio.uio_iovcnt = 1; 224 auio.uio_offset = offset; 225 auio.uio_resid = nbytes; 226 auio.uio_rw = UIO_READ; 227 auio.uio_segflg = seg; 228 auio.uio_td = curthread; 229 230 count = nbytes; 231 error = fo_read(fp, &auio, fp->f_cred, O_FOFFSET); 232 if (error) { 233 if (auio.uio_resid != nbytes && (error == ERESTART || error == EINTR || 234 error == EWOULDBLOCK) 235 ) { 236 error = 0; 237 } 238 } 239 count -= auio.uio_resid; 240 if (res) 241 *res = count; 242 return(error); 243 } 244 245 int 246 fp_read(file_t fp, void *buf, size_t nbytes, ssize_t *res, int all, 247 enum uio_seg seg) 248 { 249 struct uio auio; 250 struct iovec aiov; 251 int error; 252 int lastresid; 253 254 if (res) 255 *res = 0; 256 if (nbytes > INT_MAX) 257 return (EINVAL); 258 bzero(&auio, sizeof(auio)); 259 aiov.iov_base = (caddr_t)buf; 260 aiov.iov_len = nbytes; 261 auio.uio_iov = &aiov; 262 auio.uio_iovcnt = 1; 263 auio.uio_offset = 0; 264 auio.uio_resid = nbytes; 265 auio.uio_rw = UIO_READ; 266 auio.uio_segflg = seg; 267 auio.uio_td = curthread; 268 269 /* 270 * If all is false call fo_read() once. 271 * If all is true we attempt to read the entire request. We have to 272 * break out of the loop if an unrecoverable error or EOF occurs. 273 */ 274 do { 275 lastresid = auio.uio_resid; 276 error = fo_read(fp, &auio, fp->f_cred, 0); 277 } while (all && auio.uio_resid && 278 ((error == 0 && auio.uio_resid != lastresid) || 279 error == ERESTART || error == EINTR)); 280 if (all && error == 0 && auio.uio_resid) 281 error = ESPIPE; 282 283 /* 284 * If an error occured but some data was read, silently forget the 285 * error. However, if this is a non-blocking descriptor and 'all' 286 * was specified, return an error even if some data was read (this 287 * is considered a bug in the caller for using an illegal combination 288 * of 'all' and a non-blocking descriptor). 289 */ 290 if (error) { 291 if (auio.uio_resid != nbytes) { 292 if (error == ERESTART || error == EINTR) 293 error = 0; 294 if (error == EWOULDBLOCK && all == 0) 295 error = 0; 296 } 297 } 298 if (res) 299 *res = nbytes - auio.uio_resid; 300 return(error); 301 } 302 303 int 304 fp_pwrite(file_t fp, void *buf, size_t nbytes, off_t offset, ssize_t *res, 305 enum uio_seg seg) 306 { 307 struct uio auio; 308 struct iovec aiov; 309 size_t count; 310 int error; 311 312 if (res) 313 *res = 0; 314 if (nbytes > INT_MAX) 315 return (EINVAL); 316 bzero(&auio, sizeof(auio)); 317 aiov.iov_base = (caddr_t)buf; 318 aiov.iov_len = nbytes; 319 auio.uio_iov = &aiov; 320 auio.uio_iovcnt = 1; 321 auio.uio_offset = offset; 322 auio.uio_resid = nbytes; 323 auio.uio_rw = UIO_WRITE; 324 auio.uio_segflg = seg; 325 auio.uio_td = curthread; 326 327 count = nbytes; 328 error = fo_write(fp, &auio, fp->f_cred, O_FOFFSET); 329 if (error) { 330 if (auio.uio_resid != nbytes && (error == ERESTART || error == EINTR || 331 error == EWOULDBLOCK) 332 ) { 333 error = 0; 334 } 335 } 336 count -= auio.uio_resid; 337 if (res) 338 *res = count; 339 return(error); 340 } 341 342 343 int 344 fp_write(file_t fp, void *buf, size_t nbytes, ssize_t *res, enum uio_seg seg) 345 { 346 struct uio auio; 347 struct iovec aiov; 348 size_t count; 349 int error; 350 351 if (res) 352 *res = 0; 353 if (nbytes > INT_MAX) 354 return (EINVAL); 355 bzero(&auio, sizeof(auio)); 356 aiov.iov_base = (caddr_t)buf; 357 aiov.iov_len = nbytes; 358 auio.uio_iov = &aiov; 359 auio.uio_iovcnt = 1; 360 auio.uio_offset = 0; 361 auio.uio_resid = nbytes; 362 auio.uio_rw = UIO_WRITE; 363 auio.uio_segflg = seg; 364 auio.uio_td = curthread; 365 366 count = nbytes; 367 error = fo_write(fp, &auio, fp->f_cred, 0); 368 if (error) { 369 if (auio.uio_resid != nbytes && (error == ERESTART || error == EINTR || 370 error == EWOULDBLOCK) 371 ) { 372 error = 0; 373 } 374 } 375 count -= auio.uio_resid; 376 if (res) 377 *res = count; 378 return(error); 379 } 380 381 int 382 fp_stat(file_t fp, struct stat *ub) 383 { 384 int error; 385 386 error = fo_stat(fp, ub, fp->f_cred); 387 return(error); 388 } 389 390 /* 391 * non-anonymous, non-stack descriptor mappings only! 392 * 393 * This routine mostly snarfed from vm/vm_mmap.c 394 */ 395 int 396 fp_mmap(void *addr_arg, size_t size, int prot, int flags, struct file *fp, 397 off_t pos, void **resp) 398 { 399 struct thread *td = curthread; 400 struct proc *p = td->td_proc; 401 vm_size_t pageoff; 402 vm_prot_t maxprot; 403 vm_offset_t addr; 404 void *handle; 405 int error; 406 vm_object_t obj; 407 struct vmspace *vms = p->p_vmspace; 408 struct vnode *vp; 409 int disablexworkaround; 410 411 prot &= VM_PROT_ALL; 412 413 if ((ssize_t)size < 0 || (flags & MAP_ANON)) 414 return(EINVAL); 415 416 pageoff = (pos & PAGE_MASK); 417 pos -= pageoff; 418 419 /* Adjust size for rounding (on both ends). */ 420 size += pageoff; /* low end... */ 421 size = (vm_size_t)round_page(size); /* hi end */ 422 addr = (vm_offset_t)addr_arg; 423 424 /* 425 * Check for illegal addresses. Watch out for address wrap... Note 426 * that VM_*_ADDRESS are not constants due to casts (argh). 427 */ 428 if (flags & MAP_FIXED) { 429 /* 430 * The specified address must have the same remainder 431 * as the file offset taken modulo PAGE_SIZE, so it 432 * should be aligned after adjustment by pageoff. 433 */ 434 addr -= pageoff; 435 if (addr & PAGE_MASK) 436 return (EINVAL); 437 /* Address range must be all in user VM space. */ 438 if (VM_MAX_USER_ADDRESS > 0 && addr + size > VM_MAX_USER_ADDRESS) 439 return (EINVAL); 440 if (VM_MIN_USER_ADDRESS > 0 && addr < VM_MIN_USER_ADDRESS) 441 return (EINVAL); 442 if (addr + size < addr) 443 return (EINVAL); 444 } else if (addr == 0 || 445 (addr >= round_page((vm_offset_t)vms->vm_taddr) && 446 addr < round_page((vm_offset_t)vms->vm_daddr + maxdsiz)) 447 ) { 448 /* 449 * XXX for non-fixed mappings where no hint is provided or 450 * the hint would fall in the potential heap space, 451 * place it after the end of the largest possible heap. 452 * 453 * There should really be a pmap call to determine a reasonable 454 * location. 455 */ 456 addr = round_page((vm_offset_t)vms->vm_daddr + maxdsiz); 457 } 458 459 /* 460 * Mapping file, get fp for validation. Obtain vnode and make 461 * sure it is of appropriate type. 462 */ 463 if (fp->f_type != DTYPE_VNODE) 464 return (EINVAL); 465 466 /* 467 * POSIX shared-memory objects are defined to have 468 * kernel persistence, and are not defined to support 469 * read(2)/write(2) -- or even open(2). Thus, we can 470 * use MAP_ASYNC to trade on-disk coherence for speed. 471 * The shm_open(3) library routine turns on the FPOSIXSHM 472 * flag to request this behavior. 473 */ 474 if (fp->f_flag & FPOSIXSHM) 475 flags |= MAP_NOSYNC; 476 vp = (struct vnode *) fp->f_data; 477 if (vp->v_type != VREG && vp->v_type != VCHR) 478 return (EINVAL); 479 480 /* 481 * Get the proper underlying object 482 */ 483 if (vp->v_type == VREG) { 484 if ((obj = vp->v_object) == NULL) 485 return (EINVAL); 486 KKASSERT(vp == (struct vnode *)obj->handle); 487 } 488 489 /* 490 * XXX hack to handle use of /dev/zero to map anon memory (ala 491 * SunOS). 492 */ 493 if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) { 494 handle = NULL; 495 maxprot = VM_PROT_ALL; 496 flags |= MAP_ANON; 497 pos = 0; 498 } else { 499 /* 500 * cdevs does not provide private mappings of any kind. 501 */ 502 /* 503 * However, for XIG X server to continue to work, 504 * we should allow the superuser to do it anyway. 505 * We only allow it at securelevel < 1. 506 * (Because the XIG X server writes directly to video 507 * memory via /dev/mem, it should never work at any 508 * other securelevel. 509 * XXX this will have to go 510 */ 511 if (securelevel >= 1) 512 disablexworkaround = 1; 513 else 514 disablexworkaround = suser(td); 515 if (vp->v_type == VCHR && disablexworkaround && 516 (flags & (MAP_PRIVATE|MAP_COPY))) { 517 error = EINVAL; 518 goto done; 519 } 520 /* 521 * Ensure that file and memory protections are 522 * compatible. Note that we only worry about 523 * writability if mapping is shared; in this case, 524 * current and max prot are dictated by the open file. 525 * XXX use the vnode instead? Problem is: what 526 * credentials do we use for determination? What if 527 * proc does a setuid? 528 */ 529 maxprot = VM_PROT_EXECUTE; /* ??? */ 530 if (fp->f_flag & FREAD) { 531 maxprot |= VM_PROT_READ; 532 } else if (prot & PROT_READ) { 533 error = EACCES; 534 goto done; 535 } 536 /* 537 * If we are sharing potential changes (either via 538 * MAP_SHARED or via the implicit sharing of character 539 * device mappings), and we are trying to get write 540 * permission although we opened it without asking 541 * for it, bail out. Check for superuser, only if 542 * we're at securelevel < 1, to allow the XIG X server 543 * to continue to work. 544 */ 545 546 if ((flags & MAP_SHARED) != 0 || 547 (vp->v_type == VCHR && disablexworkaround) 548 ) { 549 if ((fp->f_flag & FWRITE) != 0) { 550 struct vattr va; 551 if ((error = VOP_GETATTR(vp, &va))) { 552 goto done; 553 } 554 if ((va.va_flags & (IMMUTABLE|APPEND)) == 0) { 555 maxprot |= VM_PROT_WRITE; 556 } else if (prot & PROT_WRITE) { 557 error = EPERM; 558 goto done; 559 } 560 } else if ((prot & PROT_WRITE) != 0) { 561 error = EACCES; 562 goto done; 563 } 564 } else { 565 maxprot |= VM_PROT_WRITE; 566 } 567 handle = (void *)vp; 568 } 569 error = vm_mmap(&vms->vm_map, &addr, size, prot, 570 maxprot, flags, handle, pos); 571 if (error == 0 && addr_arg) 572 *resp = (void *)addr; 573 done: 574 return (error); 575 } 576 577 int 578 fp_close(file_t fp) 579 { 580 return(fdrop(fp)); 581 } 582 583 int 584 fp_shutdown(file_t fp, int how) 585 { 586 return(fo_shutdown(fp, how)); 587 } 588 589