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