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