sys/kern/kern_fp.c

   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
  35 /*
  36  * Direct file pointer API functions for in-kernel operations on files.  These
  37  * functions provide a open/read/write/close like interface within the kernel
  38  * for operating on files that are not necessarily associated with processes
  39  * and which do not (typically) have descriptors.
  40  *
  41  * FUTURE: file handle conversion routines to support checkpointing,
  42  * and additional file operations (ioctl, fcntl).
  43  */
  44
  45 #include <sys/param.h>
  46 #include <sys/kernel.h>
  47 #include <sys/systm.h>
  48 #include <sys/uio.h>
  49 #include <sys/malloc.h>
  50 #include <sys/sysmsg.h>
  51 #include <sys/conf.h>
  52 #include <sys/filedesc.h>
  53 #include <sys/sysctl.h>
  54 #include <sys/vnode.h>
  55 #include <sys/proc.h>
  56 #include <sys/caps.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_kern.h>
  78
  79 #include <sys/file2.h>
  80 #include <machine/limits.h>
  81
  82 typedef struct file *file_t;
  83
  84 /*
  85  * fp_open:
  86  *
  87  *      Open a file as specified.  Use O_* flags for flags.
  88  *
  89  *      vn_open() asserts that the cred must match the process's cred.
  90  *
  91  *      NOTE! when fp_open() is called from a pure thread, root creds are
  92  *      used.
  93  */
  94 int
  95 fp_open(const char *path, int flags, int mode, file_t *fpp)
  96 {
  97     struct nlookupdata nd;
  98     struct thread *td;
  99     int error;
 100
 101     if ((error = falloc(NULL, fpp, NULL)) != 0)
 102         return (error);
 103     td = curthread;
 104     if (td->td_proc)
 105         fsetcred(*fpp, td->td_proc->p_ucred);
 106     error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_LOCKVP);
 107     flags = FFLAGS(flags);
 108     if (error == 0)
 109         error = vn_open(&nd, fpp, flags, mode);
 110     nlookup_done(&nd);
 111     if (error) {
 112         fdrop(*fpp);
 113         *fpp = NULL;
 114     }
 115     return(error);
 116 }
 117
 118
 119 /*
 120  * fp_vpopen(): convert a vnode to a file pointer, call VOP_OPEN() on the
 121  * the vnode.  The vnode must be refd and locked.
 122  *
 123  * On success the vnode's ref is inherited by the file pointer and the caller
 124  * should not vrele() it, and the vnode is unlocked.
 125  *
 126  * On failure the vnode remains locked and refd and the caller is responsible
 127  * for vput()ing it.
 128  */
 129 int
 130 fp_vpopen(struct vnode *vp, int flags, file_t *fpp)
 131 {
 132     struct thread *td;
 133     struct file *fp;
 134     int vmode;
 135     int error;
 136
 137     td = curthread;
 138
 139     /*
 140      * Vnode checks (from vn_open())
 141      */
 142     if (vp->v_type == VLNK) {
 143         error = EMLINK;
 144         goto bad2;
 145     }
 146     if (vp->v_type == VSOCK) {
 147         error = EOPNOTSUPP;
 148         goto bad2;
 149     }
 150     flags = FFLAGS(flags);
 151     vmode = 0;
 152     if (flags & (FWRITE | O_TRUNC)) {
 153         if (vp->v_type == VDIR) {
 154             error = EISDIR;
 155             goto bad2;
 156         }
 157         error = vn_writechk(vp);
 158         if (error)
 159             goto bad2;
 160         vmode |= VWRITE;
 161     }
 162     if (flags & FREAD)
 163         vmode |= VREAD;
 164     if (vmode) {
 165         error = VOP_ACCESS(vp, vmode, td->td_proc->p_ucred);
 166         if (error)
 167             goto bad2;
 168     }
 169
 170     /*
 171      * File pointer setup
 172      */
 173     if ((error = falloc(NULL, fpp, NULL)) != 0)
 174         goto bad2;
 175     if (td->td_proc)
 176         fsetcred(*fpp, td->td_proc->p_ucred);
 177
 178     error = VOP_OPEN(vp, flags, td->td_proc->p_ucred, fpp);
 179
 180     if (error)
 181         goto bad1;
 182
 183     vput(vp);
 184     return (0);
 185
 186 bad1:
 187     fp = *fpp;
 188     fp->f_ops = &badfileops;    /* open failed, don't close */
 189     fp->f_data = NULL;
 190     fdrop(fp);
 191     /* leave the vnode intact, but fall through and unlock it anyway */
 192 bad2:
 193     *fpp = NULL;
 194
 195     return (error);
 196 }
 197
 198 /*
 199  * fp_*read() is meant to operate like the normal descriptor based syscalls
 200  * would.  Note that if 'buf' points to user memory a UIO_USERSPACE
 201  * transfer will be used.
 202  */
 203 int
 204 fp_pread(file_t fp, void *buf, size_t nbytes, off_t offset, ssize_t *res,
 205          enum uio_seg seg)
 206 {
 207     struct uio auio;
 208     struct iovec aiov;
 209     size_t count;
 210     int error;
 211
 212     if (res)
 213         *res = 0;
 214     if (nbytes > LONG_MAX)
 215         return (EINVAL);
 216     bzero(&auio, sizeof(auio));
 217     aiov.iov_base = (caddr_t)buf;
 218     aiov.iov_len = nbytes;
 219     auio.uio_iov = &aiov;
 220     auio.uio_iovcnt = 1;
 221     auio.uio_offset = offset;
 222     auio.uio_resid = nbytes;
 223     auio.uio_rw = UIO_READ;
 224     auio.uio_segflg = seg;
 225     auio.uio_td = curthread;
 226
 227     count = nbytes;
 228     error = fo_read(fp, &auio, fp->f_cred, O_FOFFSET);
 229     if (error) {
 230         if (auio.uio_resid != nbytes && (error == ERESTART || error == EINTR ||
 231             error == EWOULDBLOCK)
 232         ) {
 233             error = 0;
 234         }
 235     }
 236     count -= auio.uio_resid;
 237     if (res)
 238         *res = count;
 239     return(error);
 240 }
 241
 242 int
 243 fp_read(file_t fp, void *buf, size_t nbytes, ssize_t *res, int all,
 244         enum uio_seg seg)
 245 {
 246     struct uio auio;
 247     struct iovec aiov;
 248     int error;
 249     int lastresid;
 250
 251     if (res)
 252         *res = 0;
 253     if (nbytes > LONG_MAX)
 254         return (EINVAL);
 255     bzero(&auio, sizeof(auio));
 256     aiov.iov_base = (caddr_t)buf;
 257     aiov.iov_len = nbytes;
 258     auio.uio_iov = &aiov;
 259     auio.uio_iovcnt = 1;
 260     auio.uio_offset = 0;
 261     auio.uio_resid = nbytes;
 262     auio.uio_rw = UIO_READ;
 263     auio.uio_segflg = seg;
 264     auio.uio_td = curthread;
 265
 266     /*
 267      * If all is false call fo_read() once.
 268      * If all is true we attempt to read the entire request.  We have to
 269      * break out of the loop if an unrecoverable error or EOF occurs.
 270      */
 271     do {
 272         lastresid = auio.uio_resid;
 273         error = fo_read(fp, &auio, fp->f_cred, 0);
 274     } while (all && auio.uio_resid &&
 275              ((error == 0 && auio.uio_resid != lastresid) ||
 276              error == ERESTART || error == EINTR));
 277     if (all && error == 0 && auio.uio_resid)
 278         error = ESPIPE;
 279
 280     /*
 281      * If an error occured but some data was read, silently forget the
 282      * error.  However, if this is a non-blocking descriptor and 'all'
 283      * was specified, return an error even if some data was read (this
 284      * is considered a bug in the caller for using an illegal combination
 285      * of 'all' and a non-blocking descriptor).
 286      */
 287     if (error) {
 288         if (auio.uio_resid != nbytes) {
 289             if (error == ERESTART || error == EINTR)
 290                 error = 0;
 291             if (error == EWOULDBLOCK && all == 0)
 292                 error = 0;
 293         }
 294     }
 295     if (res)
 296         *res = nbytes - auio.uio_resid;
 297     return(error);
 298 }
 299
 300 int
 301 fp_pwrite(file_t fp, void *buf, size_t nbytes, off_t offset, ssize_t *res,
 302           enum uio_seg seg)
 303 {
 304     struct uio auio;
 305     struct iovec aiov;
 306     size_t count;
 307     int error;
 308
 309     if (res)
 310         *res = 0;
 311     if (nbytes > LONG_MAX)
 312         return (EINVAL);
 313     bzero(&auio, sizeof(auio));
 314     aiov.iov_base = (caddr_t)buf;
 315     aiov.iov_len = nbytes;
 316     auio.uio_iov = &aiov;
 317     auio.uio_iovcnt = 1;
 318     auio.uio_offset = offset;
 319     auio.uio_resid = nbytes;
 320     auio.uio_rw = UIO_WRITE;
 321     auio.uio_segflg = seg;
 322     auio.uio_td = curthread;
 323
 324     count = nbytes;
 325     error = fo_write(fp, &auio, fp->f_cred, O_FOFFSET);
 326     if (error) {
 327         if (auio.uio_resid != nbytes && (error == ERESTART || error == EINTR ||
 328             error == EWOULDBLOCK)
 329         ) {
 330             error = 0;
 331         }
 332     }
 333     count -= auio.uio_resid;
 334     if (res)
 335         *res = count;
 336     return(error);
 337 }
 338
 339
 340 int
 341 fp_write(file_t fp, void *buf, size_t nbytes, ssize_t *res, enum uio_seg seg)
 342 {
 343     struct uio auio;
 344     struct iovec aiov;
 345     size_t count;
 346     int error;
 347
 348     if (res)
 349         *res = 0;
 350     if (nbytes > LONG_MAX)
 351         return (EINVAL);
 352     bzero(&auio, sizeof(auio));
 353     aiov.iov_base = (caddr_t)buf;
 354     aiov.iov_len = nbytes;
 355     auio.uio_iov = &aiov;
 356     auio.uio_iovcnt = 1;
 357     auio.uio_offset = 0;
 358     auio.uio_resid = nbytes;
 359     auio.uio_rw = UIO_WRITE;
 360     auio.uio_segflg = seg;
 361     auio.uio_td = curthread;
 362
 363     count = nbytes;
 364     error = fo_write(fp, &auio, fp->f_cred, 0);
 365     if (error) {
 366         if (auio.uio_resid != nbytes && (error == ERESTART || error == EINTR ||
 367             error == EWOULDBLOCK)
 368         ) {
 369             error = 0;
 370         }
 371     }
 372     count -= auio.uio_resid;
 373     if (res)
 374         *res = count;
 375     return(error);
 376 }
 377
 378 int
 379 fp_stat(file_t fp, struct stat *ub)
 380 {
 381     int error;
 382
 383     error = fo_stat(fp, ub, fp->f_cred);
 384     return(error);
 385 }
 386
 387 /*
 388  * non-anonymous, non-stack descriptor mappings only!
 389  *
 390  * This routine mostly snarfed from vm/vm_mmap.c
 391  */
 392 int
 393 fp_mmap(void *addr_arg, size_t size, int prot, int flags, struct file *fp,
 394     off_t pos, void **resp)
 395 {
 396     struct thread *td = curthread;
 397     struct proc *p = td->td_proc;
 398     vm_size_t pageoff;
 399     vm_prot_t maxprot;
 400     vm_offset_t addr;
 401     void *handle;
 402     int error;
 403     vm_object_t obj;
 404     struct vmspace *vms = p->p_vmspace;
 405     struct vnode *vp;
 406
 407     prot &= VM_PROT_ALL;
 408
 409     if ((ssize_t)size < 0 || (flags & MAP_ANON))
 410         return(EINVAL);
 411
 412     pageoff = (pos & PAGE_MASK);
 413     pos -= pageoff;
 414
 415     /* Adjust size for rounding (on both ends). */
 416     size += pageoff;                            /* low end... */
 417     size = (vm_size_t)round_page(size);         /* hi end */
 418     addr = (vm_offset_t)addr_arg;
 419
 420     /*
 421      * Check for illegal addresses.  Watch out for address wrap... Note
 422      * that VM_*_ADDRESS are not constants due to casts (argh).
 423      */
 424     if (flags & MAP_FIXED) {
 425         /*
 426          * The specified address must have the same remainder
 427          * as the file offset taken modulo PAGE_SIZE, so it
 428          * should be aligned after adjustment by pageoff.
 429          */
 430         addr -= pageoff;
 431         if (addr & PAGE_MASK)
 432             return (EINVAL);
 433         /* Address range must be all in user VM space. */
 434         if (VM_MAX_USER_ADDRESS > 0 && addr + size > VM_MAX_USER_ADDRESS)
 435             return (EINVAL);
 436         if (VM_MIN_USER_ADDRESS > 0 && addr < VM_MIN_USER_ADDRESS)
 437             return (EINVAL);
 438         if (addr + size < addr)
 439             return (EINVAL);
 440     } else if (addr == 0 ||
 441         (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
 442          addr < round_page((vm_offset_t)vms->vm_daddr + maxdsiz))
 443     ) {
 444         /*
 445          * XXX for non-fixed mappings where no hint is provided or
 446          * the hint would fall in the potential heap space,
 447          * place it after the end of the largest possible heap.
 448          *
 449          * There should really be a pmap call to determine a reasonable
 450          * location.
 451          */
 452         addr = round_page((vm_offset_t)vms->vm_daddr + maxdsiz);
 453     }
 454
 455     /*
 456      * Mapping file, get fp for validation. Obtain vnode and make
 457      * sure it is of appropriate type.
 458      */
 459     if (fp->f_type != DTYPE_VNODE)
 460         return (EINVAL);
 461
 462     /*
 463      * POSIX shared-memory objects are defined to have
 464      * kernel persistence, and are not defined to support
 465      * read(2)/write(2) -- or even open(2).  Thus, we can
 466      * use MAP_ASYNC to trade on-disk coherence for speed.
 467      * The shm_open(3) library routine turns on the FPOSIXSHM
 468      * flag to request this behavior.
 469      */
 470     if (fp->f_flag & FPOSIXSHM)
 471         flags |= MAP_NOSYNC;
 472     vp = (struct vnode *) fp->f_data;
 473     if (vp->v_type != VREG && vp->v_type != VCHR)
 474         return (EINVAL);
 475
 476     /*
 477      * Get the proper underlying object
 478      */
 479     if (vp->v_type == VREG) {
 480         if ((obj = vp->v_object) == NULL)
 481             return (EINVAL);
 482         KKASSERT(vp == (struct vnode *)obj->handle);
 483     }
 484
 485     /*
 486      * XXX hack to handle use of /dev/zero to map anon memory (ala
 487      * SunOS).
 488      */
 489     if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) {
 490         handle = NULL;
 491         maxprot = VM_PROT_ALL;
 492         flags |= MAP_ANON;
 493         pos = 0;
 494     } else {
 495         /*
 496          * cdevs does not provide private mappings of any kind.
 497          */
 498         if (vp->v_type == VCHR &&
 499             (flags & (MAP_PRIVATE|MAP_COPY))) {
 500                 error = EINVAL;
 501                 goto done;
 502         }
 503         /*
 504          * Ensure that file and memory protections are
 505          * compatible.  Note that we only worry about
 506          * writability if mapping is shared; in this case,
 507          * current and max prot are dictated by the open file.
 508          * XXX use the vnode instead?  Problem is: what
 509          * credentials do we use for determination? What if
 510          * proc does a setuid?
 511          */
 512         maxprot = VM_PROT_EXECUTE;      /* ??? */
 513         if (fp->f_flag & FREAD) {
 514             maxprot |= VM_PROT_READ;
 515         } else if (prot & PROT_READ) {
 516             error = EACCES;
 517             goto done;
 518         }
 519         /*
 520          * If we are sharing potential changes (either via
 521          * MAP_SHARED or via the implicit sharing of character
 522          * device mappings), and we are trying to get write
 523          * permission although we opened it without asking
 524          * for it, bail out.
 525          */
 526
 527         if ((flags & MAP_SHARED) != 0 ||
 528             (vp->v_type == VCHR)
 529         ) {
 530             if ((fp->f_flag & FWRITE) != 0) {
 531                 struct vattr va;
 532                 if ((error = VOP_GETATTR_FP(vp, &va, fp))) {
 533                     goto done;
 534                 }
 535                 if ((va.va_flags & (IMMUTABLE|APPEND)) == 0) {
 536                     maxprot |= VM_PROT_WRITE;
 537                 } else if (prot & PROT_WRITE) {
 538                     error = EPERM;
 539                     goto done;
 540                 }
 541             } else if ((prot & PROT_WRITE) != 0) {
 542                 error = EACCES;
 543                 goto done;
 544             }
 545         } else {
 546             maxprot |= VM_PROT_WRITE;
 547         }
 548         handle = (void *)vp;
 549     }
 550     error = vm_mmap(&vms->vm_map, &addr, size, prot,
 551                     maxprot, flags, handle, pos, fp);
 552     if (error == 0 && addr_arg)
 553         *resp = (void *)addr;
 554 done:
 555     return (error);
 556 }
 557
 558 int
 559 fp_close(file_t fp)
 560 {
 561     return(fdrop(fp));
 562 }
 563
 564 int
 565 fp_shutdown(file_t fp, int how)
 566 {
 567     return(fo_shutdown(fp, how));
 568 }
 569