2 * Copyright (c) 2003,2004 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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
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.
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
34 * $DragonFly: src/sys/kern/kern_fp.c,v 1.13 2006/03/29 18:44:50 dillon Exp $
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.
43 * FUTURE: file handle conversion routines to support checkpointing,
44 * and additional file operations (ioctl, fcntl).
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>
53 #include <sys/filedesc.h>
54 #include <sys/sysctl.h>
55 #include <sys/vnode.h>
57 #include <sys/nlookup.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>
68 #include <vm/vm_param.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>
80 #include <sys/file2.h>
81 #include <machine/limits.h>
83 typedef struct file *file_t;
88 * Open a file as specified. Use O_* flags for flags.
90 * NOTE! O_ROOTCRED not quite working yet, vn_open() asserts that the
91 * cred must match the process's cred. XXX
93 * NOTE! when fp_open() is called from a pure thread, root creds are
97 fp_open(const char *path, int flags, int mode, file_t *fpp)
99 struct nlookupdata nd;
104 if ((error = falloc(NULL, fpp, NULL)) != 0)
109 if ((flags & O_ROOTCRED) == 0)
110 fsetcred(fp, td->td_proc->p_ucred);
112 error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_LOCKVP);
113 flags = FFLAGS(flags);
115 error = vn_open(&nd, fp, flags, mode);
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.
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.
132 * On failure the vnode remains locked and refd and the caller is responsible
136 fp_vpopen(struct vnode *vp, int flags, file_t *fpp)
146 * Vnode checks (from vn_open())
148 if (vp->v_type == VLNK) {
152 if (vp->v_type == VSOCK) {
156 flags = FFLAGS(flags);
158 if (flags & (FWRITE | O_TRUNC)) {
159 if (vp->v_type == VDIR) {
163 error = vn_writechk(vp);
171 error = VOP_ACCESS(vp, vmode, td->td_proc->p_ucred, td);
179 if ((error = falloc(NULL, fpp, NULL)) != 0)
182 if ((flags & O_ROOTCRED) == 0 && td->td_proc)
183 fsetcred(fp, td->td_proc->p_ucred);
184 fp->f_type = DTYPE_VNODE;
186 fp->f_ops = &vnode_fileops;
189 error = VOP_OPEN(vp, flags, td->td_proc->p_ucred, fp, td);
194 * All done, update v_writecount now that no more errors can occur.
198 VOP_UNLOCK(vp, 0, td);
201 fp->f_ops = &badfileops; /* open failed, don't close */
204 /* leave the vnode intact, but fall through and unlock it anyway */
211 * fp_*read() is meant to operate like the normal descriptor based syscalls
212 * would. Note that if 'buf' points to user memory a UIO_USERSPACE
213 * transfer will be used.
216 fp_pread(file_t fp, void *buf, size_t nbytes, off_t offset, ssize_t *res)
225 if (nbytes > INT_MAX)
227 bzero(&auio, sizeof(auio));
228 aiov.iov_base = (caddr_t)buf;
229 aiov.iov_len = nbytes;
230 auio.uio_iov = &aiov;
232 auio.uio_offset = offset;
233 auio.uio_resid = nbytes;
234 auio.uio_rw = UIO_READ;
235 if ((vm_offset_t)buf < VM_MAXUSER_ADDRESS)
236 auio.uio_segflg = UIO_USERSPACE;
238 auio.uio_segflg = UIO_SYSSPACE;
239 auio.uio_td = curthread;
242 error = fo_read(fp, &auio, fp->f_cred, FOF_OFFSET, auio.uio_td);
244 if (auio.uio_resid != nbytes && (error == ERESTART || error == EINTR ||
245 error == EWOULDBLOCK)
250 count -= auio.uio_resid;
257 fp_read(file_t fp, void *buf, size_t nbytes, ssize_t *res, int all)
266 if (nbytes > INT_MAX)
268 bzero(&auio, sizeof(auio));
269 aiov.iov_base = (caddr_t)buf;
270 aiov.iov_len = nbytes;
271 auio.uio_iov = &aiov;
274 auio.uio_resid = nbytes;
275 auio.uio_rw = UIO_READ;
276 if ((vm_offset_t)buf < VM_MAXUSER_ADDRESS)
277 auio.uio_segflg = UIO_USERSPACE;
279 auio.uio_segflg = UIO_SYSSPACE;
280 auio.uio_td = curthread;
283 * If all is false call fo_read() once.
284 * If all is true we attempt to read the entire request. We have to
285 * break out of the loop if an unrecoverable error or EOF occurs.
288 lastresid = auio.uio_resid;
289 error = fo_read(fp, &auio, fp->f_cred, 0, auio.uio_td);
290 } while (all && auio.uio_resid &&
291 ((error == 0 && auio.uio_resid != lastresid) ||
292 error == ERESTART || error == EINTR));
293 if (all && error == 0 && auio.uio_resid)
297 * If an error occured but some data was read, silently forget the
298 * error. However, if this is a non-blocking descriptor and 'all'
299 * was specified, return an error even if some data was read (this
300 * is considered a bug in the caller for using an illegal combination
301 * of 'all' and a non-blocking descriptor).
304 if (auio.uio_resid != nbytes) {
305 if (error == ERESTART || error == EINTR)
307 if (error == EWOULDBLOCK && all == 0)
312 *res = nbytes - auio.uio_resid;
317 fp_pwrite(file_t fp, void *buf, size_t nbytes, off_t offset, ssize_t *res)
326 if (nbytes > INT_MAX)
328 bzero(&auio, sizeof(auio));
329 aiov.iov_base = (caddr_t)buf;
330 aiov.iov_len = nbytes;
331 auio.uio_iov = &aiov;
333 auio.uio_offset = offset;
334 auio.uio_resid = nbytes;
335 auio.uio_rw = UIO_WRITE;
336 if ((vm_offset_t)buf < VM_MAXUSER_ADDRESS)
337 auio.uio_segflg = UIO_USERSPACE;
339 auio.uio_segflg = UIO_SYSSPACE;
340 auio.uio_td = curthread;
343 error = fo_write(fp, &auio, fp->f_cred, FOF_OFFSET, auio.uio_td);
345 if (auio.uio_resid != nbytes && (error == ERESTART || error == EINTR ||
346 error == EWOULDBLOCK)
351 count -= auio.uio_resid;
359 fp_write(file_t fp, void *buf, size_t nbytes, ssize_t *res)
368 if (nbytes > INT_MAX)
370 bzero(&auio, sizeof(auio));
371 aiov.iov_base = (caddr_t)buf;
372 aiov.iov_len = nbytes;
373 auio.uio_iov = &aiov;
376 auio.uio_resid = nbytes;
377 auio.uio_rw = UIO_WRITE;
378 if ((vm_offset_t)buf < VM_MAXUSER_ADDRESS)
379 auio.uio_segflg = UIO_USERSPACE;
381 auio.uio_segflg = UIO_SYSSPACE;
382 auio.uio_td = curthread;
385 error = fo_write(fp, &auio, fp->f_cred, 0, auio.uio_td);
387 if (auio.uio_resid != nbytes && (error == ERESTART || error == EINTR ||
388 error == EWOULDBLOCK)
393 count -= auio.uio_resid;
400 fp_stat(file_t fp, struct stat *ub)
404 error = fo_stat(fp, ub, curthread);
409 * non-anonymous, non-stack descriptor mappings only!
411 * This routine mostly snarfed from vm/vm_mmap.c
414 fp_mmap(void *addr_arg, size_t size, int prot, int flags, struct file *fp,
415 off_t pos, void **resp)
417 struct thread *td = curthread;
418 struct proc *p = td->td_proc;
425 struct vmspace *vms = p->p_vmspace;
427 int disablexworkaround;
431 if ((ssize_t)size < 0 || (flags & MAP_ANON))
434 pageoff = (pos & PAGE_MASK);
437 /* Adjust size for rounding (on both ends). */
438 size += pageoff; /* low end... */
439 size = (vm_size_t)round_page(size); /* hi end */
440 addr = (vm_offset_t)addr_arg;
443 * Check for illegal addresses. Watch out for address wrap... Note
444 * that VM_*_ADDRESS are not constants due to casts (argh).
446 if (flags & MAP_FIXED) {
448 * The specified address must have the same remainder
449 * as the file offset taken modulo PAGE_SIZE, so it
450 * should be aligned after adjustment by pageoff.
453 if (addr & PAGE_MASK)
455 /* Address range must be all in user VM space. */
456 if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
459 if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS)
462 if (addr + size < addr)
464 } else if (addr == 0 ||
465 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
466 addr < round_page((vm_offset_t)vms->vm_daddr + maxdsiz))
469 * XXX for non-fixed mappings where no hint is provided or
470 * the hint would fall in the potential heap space,
471 * place it after the end of the largest possible heap.
473 * There should really be a pmap call to determine a reasonable
476 addr = round_page((vm_offset_t)vms->vm_daddr + maxdsiz);
480 * Mapping file, get fp for validation. Obtain vnode and make
481 * sure it is of appropriate type.
483 if (fp->f_type != DTYPE_VNODE)
487 * POSIX shared-memory objects are defined to have
488 * kernel persistence, and are not defined to support
489 * read(2)/write(2) -- or even open(2). Thus, we can
490 * use MAP_ASYNC to trade on-disk coherence for speed.
491 * The shm_open(3) library routine turns on the FPOSIXSHM
492 * flag to request this behavior.
494 if (fp->f_flag & FPOSIXSHM)
496 vp = (struct vnode *) fp->f_data;
497 if (vp->v_type != VREG && vp->v_type != VCHR)
501 * Get the proper underlying object
503 if (vp->v_type == VREG) {
504 if ((obj = vp->v_object) == NULL)
506 KKASSERT(vp == (struct vnode *)obj->handle);
510 * XXX hack to handle use of /dev/zero to map anon memory (ala
513 if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) {
515 maxprot = VM_PROT_ALL;
520 * cdevs does not provide private mappings of any kind.
523 * However, for XIG X server to continue to work,
524 * we should allow the superuser to do it anyway.
525 * We only allow it at securelevel < 1.
526 * (Because the XIG X server writes directly to video
527 * memory via /dev/mem, it should never work at any
529 * XXX this will have to go
531 if (securelevel >= 1)
532 disablexworkaround = 1;
534 disablexworkaround = suser(td);
535 if (vp->v_type == VCHR && disablexworkaround &&
536 (flags & (MAP_PRIVATE|MAP_COPY))) {
541 * Ensure that file and memory protections are
542 * compatible. Note that we only worry about
543 * writability if mapping is shared; in this case,
544 * current and max prot are dictated by the open file.
545 * XXX use the vnode instead? Problem is: what
546 * credentials do we use for determination? What if
547 * proc does a setuid?
549 maxprot = VM_PROT_EXECUTE; /* ??? */
550 if (fp->f_flag & FREAD) {
551 maxprot |= VM_PROT_READ;
552 } else if (prot & PROT_READ) {
557 * If we are sharing potential changes (either via
558 * MAP_SHARED or via the implicit sharing of character
559 * device mappings), and we are trying to get write
560 * permission although we opened it without asking
561 * for it, bail out. Check for superuser, only if
562 * we're at securelevel < 1, to allow the XIG X server
563 * to continue to work.
566 if ((flags & MAP_SHARED) != 0 ||
567 (vp->v_type == VCHR && disablexworkaround)
569 if ((fp->f_flag & FWRITE) != 0) {
571 if ((error = VOP_GETATTR(vp, &va, td))) {
574 if ((va.va_flags & (IMMUTABLE|APPEND)) == 0) {
575 maxprot |= VM_PROT_WRITE;
576 } else if (prot & PROT_WRITE) {
580 } else if ((prot & PROT_WRITE) != 0) {
585 maxprot |= VM_PROT_WRITE;
589 error = vm_mmap(&vms->vm_map, &addr, size, prot,
590 maxprot, flags, handle, pos);
591 if (error == 0 && addr_arg)
592 *resp = (void *)addr;
600 return(fdrop(fp, curthread));
604 fp_shutdown(file_t fp, int how)
606 return(fo_shutdown(fp, how, curthread));