Disallow writes to filesystems mounted read-only via NULLFS. In this case
[dragonfly.git] / sys / kern / kern_fp.c
CommitLineData
39f91578 1/*
8c10bfcf
MD
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 *
39f91578
MD
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
8c10bfcf 10 *
39f91578
MD
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
8c10bfcf
MD
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
39f91578 32 * SUCH DAMAGE.
8c10bfcf 33 *
468bb1f9 34 * $DragonFly: src/sys/kern/kern_fp.c,v 1.18 2006/09/18 17:42:27 dillon Exp $
39f91578
MD
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>
fad57d0e 57#include <sys/nlookup.h>
39f91578
MD
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_page.h>
78#include <vm/vm_kern.h>
79
80#include <sys/file2.h>
81#include <machine/limits.h>
82
83typedef struct file *file_t;
84
85/*
86 * fp_open:
87 *
88 * Open a file as specified. Use O_* flags for flags.
89 *
a235f7bb 90 * NOTE! O_ROOTCRED not quite working yet, vn_open() asserts that the
4a8a7971
MD
91 * cred must match the process's cred. XXX
92 *
93 * NOTE! when fp_open() is called from a pure thread, root creds are
94 * used.
39f91578
MD
95 */
96int
97fp_open(const char *path, int flags, int mode, file_t *fpp)
98{
fad57d0e 99 struct nlookupdata nd;
39f91578
MD
100 struct thread *td;
101 struct file *fp;
102 int error;
103
104 if ((error = falloc(NULL, fpp, NULL)) != 0)
105 return (error);
106 fp = *fpp;
107 td = curthread;
4a8a7971 108 if (td->td_proc) {
fad57d0e 109 if ((flags & O_ROOTCRED) == 0)
4a8a7971 110 fsetcred(fp, td->td_proc->p_ucred);
4a8a7971 111 }
fad57d0e 112 error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_LOCKVP);
39f91578 113 flags = FFLAGS(flags);
fad57d0e
MD
114 if (error == 0)
115 error = vn_open(&nd, fp, flags, mode);
116 nlookup_done(&nd);
117 if (error) {
9f87144f 118 fdrop(fp);
a235f7bb 119 *fpp = NULL;
39f91578
MD
120 }
121 return(error);
122}
123
351b7b6d
MD
124
125/*
fad57d0e
MD
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.
128 *
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.
131 *
132 * On failure the vnode remains locked and refd and the caller is responsible
133 * for vput()ing it.
351b7b6d
MD
134 */
135int
136fp_vpopen(struct vnode *vp, int flags, file_t *fpp)
137{
138 struct thread *td;
139 struct file *fp;
140 int vmode;
141 int error;
142
351b7b6d
MD
143 td = curthread;
144
145 /*
146 * Vnode checks (from vn_open())
147 */
148 if (vp->v_type == VLNK) {
149 error = EMLINK;
fad57d0e 150 goto bad2;
351b7b6d
MD
151 }
152 if (vp->v_type == VSOCK) {
153 error = EOPNOTSUPP;
fad57d0e 154 goto bad2;
351b7b6d
MD
155 }
156 flags = FFLAGS(flags);
157 vmode = 0;
158 if (flags & (FWRITE | O_TRUNC)) {
159 if (vp->v_type == VDIR) {
160 error = EISDIR;
fad57d0e 161 goto bad2;
351b7b6d 162 }
468bb1f9 163 error = vn_writechk(vp, NULL);
351b7b6d 164 if (error)
fad57d0e 165 goto bad2;
351b7b6d
MD
166 vmode |= VWRITE;
167 }
168 if (flags & FREAD)
169 vmode |= VREAD;
170 if (vmode) {
87de5057 171 error = VOP_ACCESS(vp, vmode, td->td_proc->p_ucred);
351b7b6d 172 if (error)
fad57d0e 173 goto bad2;
351b7b6d
MD
174 }
175
176 /*
177 * File pointer setup
178 */
179 if ((error = falloc(NULL, fpp, NULL)) != 0)
fad57d0e 180 goto bad2;
351b7b6d
MD
181 fp = *fpp;
182 if ((flags & O_ROOTCRED) == 0 && td->td_proc)
183 fsetcred(fp, td->td_proc->p_ucred);
351b7b6d 184
87de5057 185 error = VOP_OPEN(vp, flags, td->td_proc->p_ucred, fp);
fad57d0e
MD
186 if (error)
187 goto bad1;
188
8ddc6004 189 vput(vp);
fad57d0e
MD
190 return (0);
191bad1:
192 fp->f_ops = &badfileops; /* open failed, don't close */
193 fp->f_data = NULL;
9f87144f 194 fdrop(fp);
fad57d0e
MD
195 /* leave the vnode intact, but fall through and unlock it anyway */
196bad2:
197 *fpp = NULL;
351b7b6d
MD
198 return (error);
199}
200
731100e5
MD
201/*
202 * fp_*read() is meant to operate like the normal descriptor based syscalls
203 * would. Note that if 'buf' points to user memory a UIO_USERSPACE
204 * transfer will be used.
205 */
39f91578 206int
731100e5 207fp_pread(file_t fp, void *buf, size_t nbytes, off_t offset, ssize_t *res)
39f91578
MD
208{
209 struct uio auio;
210 struct iovec aiov;
211 size_t count;
212 int error;
213
214 if (res)
215 *res = 0;
216 if (nbytes > INT_MAX)
217 return (EINVAL);
218 bzero(&auio, sizeof(auio));
219 aiov.iov_base = (caddr_t)buf;
220 aiov.iov_len = nbytes;
221 auio.uio_iov = &aiov;
222 auio.uio_iovcnt = 1;
223 auio.uio_offset = offset;
224 auio.uio_resid = nbytes;
225 auio.uio_rw = UIO_READ;
731100e5
MD
226 if ((vm_offset_t)buf < VM_MAXUSER_ADDRESS)
227 auio.uio_segflg = UIO_USERSPACE;
228 else
229 auio.uio_segflg = UIO_SYSSPACE;
39f91578
MD
230 auio.uio_td = curthread;
231
232 count = nbytes;
9ba76b73 233 error = fo_read(fp, &auio, fp->f_cred, O_FOFFSET);
39f91578
MD
234 if (error) {
235 if (auio.uio_resid != nbytes && (error == ERESTART || error == EINTR ||
236 error == EWOULDBLOCK)
237 ) {
238 error = 0;
239 }
240 }
241 count -= auio.uio_resid;
242 if (res)
243 *res = count;
244 return(error);
245}
246
247int
ba9c8ec1 248fp_read(file_t fp, void *buf, size_t nbytes, ssize_t *res, int all)
731100e5
MD
249{
250 struct uio auio;
251 struct iovec aiov;
731100e5 252 int error;
ba9c8ec1 253 int lastresid;
731100e5
MD
254
255 if (res)
256 *res = 0;
257 if (nbytes > INT_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 if ((vm_offset_t)buf < VM_MAXUSER_ADDRESS)
268 auio.uio_segflg = UIO_USERSPACE;
269 else
270 auio.uio_segflg = UIO_SYSSPACE;
271 auio.uio_td = curthread;
272
ba9c8ec1
MD
273 /*
274 * If all is false call fo_read() once.
275 * If all is true we attempt to read the entire request. We have to
276 * break out of the loop if an unrecoverable error or EOF occurs.
277 */
278 do {
279 lastresid = auio.uio_resid;
87de5057 280 error = fo_read(fp, &auio, fp->f_cred, 0);
ba9c8ec1
MD
281 } while (all && auio.uio_resid &&
282 ((error == 0 && auio.uio_resid != lastresid) ||
283 error == ERESTART || error == EINTR));
284 if (all && error == 0 && auio.uio_resid)
285 error = ESPIPE;
286
287 /*
288 * If an error occured but some data was read, silently forget the
289 * error. However, if this is a non-blocking descriptor and 'all'
290 * was specified, return an error even if some data was read (this
291 * is considered a bug in the caller for using an illegal combination
292 * of 'all' and a non-blocking descriptor).
293 */
731100e5 294 if (error) {
ba9c8ec1
MD
295 if (auio.uio_resid != nbytes) {
296 if (error == ERESTART || error == EINTR)
297 error = 0;
298 if (error == EWOULDBLOCK && all == 0)
299 error = 0;
731100e5
MD
300 }
301 }
731100e5 302 if (res)
ba9c8ec1 303 *res = nbytes - auio.uio_resid;
731100e5
MD
304 return(error);
305}
306
307int
308fp_pwrite(file_t fp, void *buf, size_t nbytes, off_t offset, ssize_t *res)
39f91578
MD
309{
310 struct uio auio;
311 struct iovec aiov;
312 size_t count;
313 int error;
314
315 if (res)
316 *res = 0;
317 if (nbytes > INT_MAX)
318 return (EINVAL);
319 bzero(&auio, sizeof(auio));
320 aiov.iov_base = (caddr_t)buf;
321 aiov.iov_len = nbytes;
322 auio.uio_iov = &aiov;
323 auio.uio_iovcnt = 1;
324 auio.uio_offset = offset;
325 auio.uio_resid = nbytes;
326 auio.uio_rw = UIO_WRITE;
731100e5
MD
327 if ((vm_offset_t)buf < VM_MAXUSER_ADDRESS)
328 auio.uio_segflg = UIO_USERSPACE;
329 else
330 auio.uio_segflg = UIO_SYSSPACE;
39f91578
MD
331 auio.uio_td = curthread;
332
333 count = nbytes;
9ba76b73 334 error = fo_write(fp, &auio, fp->f_cred, O_FOFFSET);
39f91578
MD
335 if (error) {
336 if (auio.uio_resid != nbytes && (error == ERESTART || error == EINTR ||
337 error == EWOULDBLOCK)
338 ) {
339 error = 0;
340 }
341 }
342 count -= auio.uio_resid;
343 if (res)
344 *res = count;
345 return(error);
346}
347
731100e5
MD
348
349int
350fp_write(file_t fp, void *buf, size_t nbytes, ssize_t *res)
351{
352 struct uio auio;
353 struct iovec aiov;
354 size_t count;
355 int error;
356
357 if (res)
358 *res = 0;
359 if (nbytes > INT_MAX)
360 return (EINVAL);
361 bzero(&auio, sizeof(auio));
362 aiov.iov_base = (caddr_t)buf;
363 aiov.iov_len = nbytes;
364 auio.uio_iov = &aiov;
365 auio.uio_iovcnt = 1;
366 auio.uio_offset = 0;
367 auio.uio_resid = nbytes;
368 auio.uio_rw = UIO_WRITE;
369 if ((vm_offset_t)buf < VM_MAXUSER_ADDRESS)
370 auio.uio_segflg = UIO_USERSPACE;
371 else
372 auio.uio_segflg = UIO_SYSSPACE;
373 auio.uio_td = curthread;
374
375 count = nbytes;
87de5057 376 error = fo_write(fp, &auio, fp->f_cred, 0);
731100e5
MD
377 if (error) {
378 if (auio.uio_resid != nbytes && (error == ERESTART || error == EINTR ||
379 error == EWOULDBLOCK)
380 ) {
381 error = 0;
382 }
383 }
384 count -= auio.uio_resid;
385 if (res)
386 *res = count;
387 return(error);
388}
389
39f91578
MD
390int
391fp_stat(file_t fp, struct stat *ub)
392{
393 int error;
394
87de5057 395 error = fo_stat(fp, ub, fp->f_cred);
39f91578
MD
396 return(error);
397}
398
399/*
400 * non-anonymous, non-stack descriptor mappings only!
401 *
402 * This routine mostly snarfed from vm/vm_mmap.c
403 */
404int
405fp_mmap(void *addr_arg, size_t size, int prot, int flags, struct file *fp,
406 off_t pos, void **resp)
407{
408 struct thread *td = curthread;
409 struct proc *p = td->td_proc;
410 vm_size_t pageoff;
411 vm_prot_t maxprot;
412 vm_offset_t addr;
413 void *handle;
414 int error;
415 vm_object_t obj;
416 struct vmspace *vms = p->p_vmspace;
417 struct vnode *vp;
418 int disablexworkaround;
419
420 prot &= VM_PROT_ALL;
421
422 if ((ssize_t)size < 0 || (flags & MAP_ANON))
423 return(EINVAL);
424
425 pageoff = (pos & PAGE_MASK);
426 pos -= pageoff;
427
428 /* Adjust size for rounding (on both ends). */
429 size += pageoff; /* low end... */
430 size = (vm_size_t)round_page(size); /* hi end */
431 addr = (vm_offset_t)addr_arg;
432
433 /*
434 * Check for illegal addresses. Watch out for address wrap... Note
435 * that VM_*_ADDRESS are not constants due to casts (argh).
436 */
437 if (flags & MAP_FIXED) {
438 /*
439 * The specified address must have the same remainder
440 * as the file offset taken modulo PAGE_SIZE, so it
441 * should be aligned after adjustment by pageoff.
442 */
443 addr -= pageoff;
444 if (addr & PAGE_MASK)
445 return (EINVAL);
446 /* Address range must be all in user VM space. */
447 if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
448 return (EINVAL);
449#ifndef i386
450 if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS)
451 return (EINVAL);
452#endif
453 if (addr + size < addr)
454 return (EINVAL);
455 } else if (addr == 0 ||
456 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
457 addr < round_page((vm_offset_t)vms->vm_daddr + maxdsiz))
458 ) {
459 /*
460 * XXX for non-fixed mappings where no hint is provided or
461 * the hint would fall in the potential heap space,
462 * place it after the end of the largest possible heap.
463 *
464 * There should really be a pmap call to determine a reasonable
465 * location.
466 */
467 addr = round_page((vm_offset_t)vms->vm_daddr + maxdsiz);
468 }
469
470 /*
471 * Mapping file, get fp for validation. Obtain vnode and make
472 * sure it is of appropriate type.
473 */
474 if (fp->f_type != DTYPE_VNODE)
475 return (EINVAL);
476
477 /*
478 * POSIX shared-memory objects are defined to have
479 * kernel persistence, and are not defined to support
480 * read(2)/write(2) -- or even open(2). Thus, we can
481 * use MAP_ASYNC to trade on-disk coherence for speed.
482 * The shm_open(3) library routine turns on the FPOSIXSHM
483 * flag to request this behavior.
484 */
485 if (fp->f_flag & FPOSIXSHM)
486 flags |= MAP_NOSYNC;
487 vp = (struct vnode *) fp->f_data;
488 if (vp->v_type != VREG && vp->v_type != VCHR)
489 return (EINVAL);
490
491 /*
492 * Get the proper underlying object
493 */
494 if (vp->v_type == VREG) {
7540ab49 495 if ((obj = vp->v_object) == NULL)
39f91578 496 return (EINVAL);
7540ab49 497 KKASSERT(vp == (struct vnode *)obj->handle);
39f91578
MD
498 }
499
500 /*
501 * XXX hack to handle use of /dev/zero to map anon memory (ala
502 * SunOS).
503 */
504 if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) {
505 handle = NULL;
506 maxprot = VM_PROT_ALL;
507 flags |= MAP_ANON;
508 pos = 0;
509 } else {
510 /*
511 * cdevs does not provide private mappings of any kind.
512 */
513 /*
514 * However, for XIG X server to continue to work,
515 * we should allow the superuser to do it anyway.
516 * We only allow it at securelevel < 1.
517 * (Because the XIG X server writes directly to video
518 * memory via /dev/mem, it should never work at any
519 * other securelevel.
520 * XXX this will have to go
521 */
522 if (securelevel >= 1)
523 disablexworkaround = 1;
524 else
525 disablexworkaround = suser(td);
526 if (vp->v_type == VCHR && disablexworkaround &&
527 (flags & (MAP_PRIVATE|MAP_COPY))) {
528 error = EINVAL;
529 goto done;
530 }
531 /*
532 * Ensure that file and memory protections are
533 * compatible. Note that we only worry about
534 * writability if mapping is shared; in this case,
535 * current and max prot are dictated by the open file.
536 * XXX use the vnode instead? Problem is: what
537 * credentials do we use for determination? What if
538 * proc does a setuid?
539 */
540 maxprot = VM_PROT_EXECUTE; /* ??? */
541 if (fp->f_flag & FREAD) {
542 maxprot |= VM_PROT_READ;
543 } else if (prot & PROT_READ) {
544 error = EACCES;
545 goto done;
546 }
547 /*
548 * If we are sharing potential changes (either via
549 * MAP_SHARED or via the implicit sharing of character
550 * device mappings), and we are trying to get write
551 * permission although we opened it without asking
552 * for it, bail out. Check for superuser, only if
553 * we're at securelevel < 1, to allow the XIG X server
554 * to continue to work.
555 */
556
557 if ((flags & MAP_SHARED) != 0 ||
558 (vp->v_type == VCHR && disablexworkaround)
559 ) {
560 if ((fp->f_flag & FWRITE) != 0) {
561 struct vattr va;
87de5057 562 if ((error = VOP_GETATTR(vp, &va))) {
39f91578
MD
563 goto done;
564 }
565 if ((va.va_flags & (IMMUTABLE|APPEND)) == 0) {
566 maxprot |= VM_PROT_WRITE;
567 } else if (prot & PROT_WRITE) {
568 error = EPERM;
569 goto done;
570 }
571 } else if ((prot & PROT_WRITE) != 0) {
572 error = EACCES;
573 goto done;
574 }
575 } else {
576 maxprot |= VM_PROT_WRITE;
577 }
578 handle = (void *)vp;
579 }
580 error = vm_mmap(&vms->vm_map, &addr, size, prot,
581 maxprot, flags, handle, pos);
582 if (error == 0 && addr_arg)
583 *resp = (void *)addr;
584done:
585 return (error);
586}
587
588int
589fp_close(file_t fp)
590{
9f87144f 591 return(fdrop(fp));
39f91578
MD
592}
593
004d2de5
MD
594int
595fp_shutdown(file_t fp, int how)
596{
87de5057 597 return(fo_shutdown(fp, how));
004d2de5
MD
598}
599