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