1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 2001 Dag-Erling Coïdan Smørgrav
5 * All rights reserved.
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 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer
12 * in this position and unchanged.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. The name of the author may not be used to endorse or promote products
17 * derived from this software without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29 */
30
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD: stable/12/sys/fs/pseudofs/pseudofs_vnops.c 369700 2021-05-01 00:41:25Z kib $");
33
34 #include "opt_pseudofs.h"
35
36 #include <sys/param.h>
37 #include <sys/kernel.h>
38 #include <sys/systm.h>
39 #include <sys/ctype.h>
40 #include <sys/dirent.h>
41 #include <sys/fcntl.h>
42 #include <sys/limits.h>
43 #include <sys/lock.h>
44 #include <sys/malloc.h>
45 #include <sys/mount.h>
46 #include <sys/mutex.h>
47 #include <sys/namei.h>
48 #include <sys/proc.h>
49 #include <sys/sbuf.h>
50 #include <sys/sx.h>
51 #include <sys/sysctl.h>
52 #include <sys/vnode.h>
53
54 #include <fs/pseudofs/pseudofs.h>
55 #include <fs/pseudofs/pseudofs_internal.h>
56
57 #define KASSERT_PN_IS_DIR(pn) \
58 KASSERT((pn)->pn_type == pfstype_root || \
59 (pn)->pn_type == pfstype_dir || \
60 (pn)->pn_type == pfstype_procdir, \
61 ("%s(): VDIR vnode refers to non-directory pfs_node", __func__))
62
63 #define KASSERT_PN_IS_FILE(pn) \
64 KASSERT((pn)->pn_type == pfstype_file, \
65 ("%s(): VREG vnode refers to non-file pfs_node", __func__))
66
67 #define KASSERT_PN_IS_LINK(pn) \
68 KASSERT((pn)->pn_type == pfstype_symlink, \
69 ("%s(): VLNK vnode refers to non-link pfs_node", __func__))
70
71 #define PFS_MAXBUFSIZ 1024 * 1024
72
73 /*
74 * Returns the fileno, adjusted for target pid
75 */
76 static uint32_t
pn_fileno(struct pfs_node * pn,pid_t pid)77 pn_fileno(struct pfs_node *pn, pid_t pid)
78 {
79
80 KASSERT(pn->pn_fileno > 0,
81 ("%s(): no fileno allocated", __func__));
82 if (pid != NO_PID)
83 return (pn->pn_fileno * NO_PID + pid);
84 return (pn->pn_fileno);
85 }
86
87 /*
88 * Returns non-zero if given file is visible to given thread.
89 */
90 static int
pfs_visible_proc(struct thread * td,struct pfs_node * pn,struct proc * proc)91 pfs_visible_proc(struct thread *td, struct pfs_node *pn, struct proc *proc)
92 {
93 int visible;
94
95 if (proc == NULL)
96 return (0);
97
98 PROC_LOCK_ASSERT(proc, MA_OWNED);
99
100 visible = ((proc->p_flag & P_WEXIT) == 0);
101 if (visible)
102 visible = (p_cansee(td, proc) == 0);
103 if (visible && pn->pn_vis != NULL)
104 visible = pn_vis(td, proc, pn);
105 if (!visible)
106 return (0);
107 return (1);
108 }
109
110 static int
pfs_visible(struct thread * td,struct pfs_node * pn,pid_t pid,bool allproc_locked,struct proc ** p)111 pfs_visible(struct thread *td, struct pfs_node *pn, pid_t pid,
112 bool allproc_locked, struct proc **p)
113 {
114 struct proc *proc;
115
116 PFS_TRACE(("%s (pid: %d, req: %d)",
117 pn->pn_name, pid, td->td_proc->p_pid));
118
119 if (p)
120 *p = NULL;
121 if (pid == NO_PID)
122 PFS_RETURN (1);
123 proc = allproc_locked ? pfind_locked(pid) : pfind(pid);
124 if (proc == NULL)
125 PFS_RETURN (0);
126 if (pfs_visible_proc(td, pn, proc)) {
127 if (p)
128 *p = proc;
129 else
130 PROC_UNLOCK(proc);
131 PFS_RETURN (1);
132 }
133 PROC_UNLOCK(proc);
134 PFS_RETURN (0);
135 }
136
137 /*
138 * Verify permissions
139 */
140 static int
pfs_access(struct vop_access_args * va)141 pfs_access(struct vop_access_args *va)
142 {
143 struct vnode *vn = va->a_vp;
144 struct pfs_vdata *pvd = vn->v_data;
145 struct vattr vattr;
146 int error;
147
148 PFS_TRACE(("%s", pvd->pvd_pn->pn_name));
149 (void)pvd;
150
151 error = VOP_GETATTR(vn, &vattr, va->a_cred);
152 if (error)
153 PFS_RETURN (error);
154 error = vaccess(vn->v_type, vattr.va_mode, vattr.va_uid,
155 vattr.va_gid, va->a_accmode, va->a_cred, NULL);
156 PFS_RETURN (error);
157 }
158
159 /*
160 * Close a file or directory
161 */
162 static int
pfs_close(struct vop_close_args * va)163 pfs_close(struct vop_close_args *va)
164 {
165 struct vnode *vn = va->a_vp;
166 struct pfs_vdata *pvd = vn->v_data;
167 struct pfs_node *pn = pvd->pvd_pn;
168 struct proc *proc;
169 int error;
170
171 PFS_TRACE(("%s", pn->pn_name));
172 pfs_assert_not_owned(pn);
173
174 /*
175 * Do nothing unless this is the last close and the node has a
176 * last-close handler.
177 */
178 if (vrefcnt(vn) > 1 || pn->pn_close == NULL)
179 PFS_RETURN (0);
180
181 if (pvd->pvd_pid != NO_PID) {
182 proc = pfind(pvd->pvd_pid);
183 } else {
184 proc = NULL;
185 }
186
187 error = pn_close(va->a_td, proc, pn);
188
189 if (proc != NULL)
190 PROC_UNLOCK(proc);
191
192 PFS_RETURN (error);
193 }
194
195 /*
196 * Get file attributes
197 */
198 static int
pfs_getattr(struct vop_getattr_args * va)199 pfs_getattr(struct vop_getattr_args *va)
200 {
201 struct vnode *vn = va->a_vp;
202 struct pfs_vdata *pvd = vn->v_data;
203 struct pfs_node *pn = pvd->pvd_pn;
204 struct vattr *vap = va->a_vap;
205 struct proc *proc;
206 int error = 0;
207
208 PFS_TRACE(("%s", pn->pn_name));
209 pfs_assert_not_owned(pn);
210
211 if (!pfs_visible(curthread, pn, pvd->pvd_pid, false, &proc))
212 PFS_RETURN (ENOENT);
213
214 vap->va_type = vn->v_type;
215 vap->va_fileid = pn_fileno(pn, pvd->pvd_pid);
216 vap->va_flags = 0;
217 vap->va_blocksize = PAGE_SIZE;
218 vap->va_bytes = vap->va_size = 0;
219 vap->va_filerev = 0;
220 vap->va_fsid = vn->v_mount->mnt_stat.f_fsid.val[0];
221 vap->va_nlink = 1;
222 nanotime(&vap->va_ctime);
223 vap->va_atime = vap->va_mtime = vap->va_ctime;
224
225 switch (pn->pn_type) {
226 case pfstype_procdir:
227 case pfstype_root:
228 case pfstype_dir:
229 #if 0
230 pfs_lock(pn);
231 /* compute link count */
232 pfs_unlock(pn);
233 #endif
234 vap->va_mode = 0555;
235 break;
236 case pfstype_file:
237 case pfstype_symlink:
238 vap->va_mode = 0444;
239 break;
240 default:
241 printf("shouldn't be here!\n");
242 vap->va_mode = 0;
243 break;
244 }
245
246 if (proc != NULL) {
247 vap->va_uid = proc->p_ucred->cr_ruid;
248 vap->va_gid = proc->p_ucred->cr_rgid;
249 } else {
250 vap->va_uid = 0;
251 vap->va_gid = 0;
252 }
253
254 if (pn->pn_attr != NULL)
255 error = pn_attr(curthread, proc, pn, vap);
256
257 if(proc != NULL)
258 PROC_UNLOCK(proc);
259
260 PFS_RETURN (error);
261 }
262
263 /*
264 * Perform an ioctl
265 */
266 static int
pfs_ioctl(struct vop_ioctl_args * va)267 pfs_ioctl(struct vop_ioctl_args *va)
268 {
269 struct vnode *vn;
270 struct pfs_vdata *pvd;
271 struct pfs_node *pn;
272 struct proc *proc;
273 int error;
274
275 vn = va->a_vp;
276 vn_lock(vn, LK_SHARED | LK_RETRY);
277 if (vn->v_iflag & VI_DOOMED) {
278 VOP_UNLOCK(vn, 0);
279 return (EBADF);
280 }
281 pvd = vn->v_data;
282 pn = pvd->pvd_pn;
283
284 PFS_TRACE(("%s: %lx", pn->pn_name, va->a_command));
285 pfs_assert_not_owned(pn);
286
287 if (vn->v_type != VREG) {
288 VOP_UNLOCK(vn, 0);
289 PFS_RETURN (EINVAL);
290 }
291 KASSERT_PN_IS_FILE(pn);
292
293 if (pn->pn_ioctl == NULL) {
294 VOP_UNLOCK(vn, 0);
295 PFS_RETURN (ENOTTY);
296 }
297
298 /*
299 * This is necessary because process' privileges may
300 * have changed since the open() call.
301 */
302 if (!pfs_visible(curthread, pn, pvd->pvd_pid, false, &proc)) {
303 VOP_UNLOCK(vn, 0);
304 PFS_RETURN (EIO);
305 }
306
307 error = pn_ioctl(curthread, proc, pn, va->a_command, va->a_data);
308
309 if (proc != NULL)
310 PROC_UNLOCK(proc);
311
312 VOP_UNLOCK(vn, 0);
313 PFS_RETURN (error);
314 }
315
316 /*
317 * Perform getextattr
318 */
319 static int
pfs_getextattr(struct vop_getextattr_args * va)320 pfs_getextattr(struct vop_getextattr_args *va)
321 {
322 struct vnode *vn = va->a_vp;
323 struct pfs_vdata *pvd = vn->v_data;
324 struct pfs_node *pn = pvd->pvd_pn;
325 struct proc *proc;
326 int error;
327
328 PFS_TRACE(("%s", pn->pn_name));
329 pfs_assert_not_owned(pn);
330
331 /*
332 * This is necessary because either process' privileges may
333 * have changed since the open() call.
334 */
335 if (!pfs_visible(curthread, pn, pvd->pvd_pid, false, &proc))
336 PFS_RETURN (EIO);
337
338 if (pn->pn_getextattr == NULL)
339 error = EOPNOTSUPP;
340 else
341 error = pn_getextattr(curthread, proc, pn,
342 va->a_attrnamespace, va->a_name, va->a_uio,
343 va->a_size, va->a_cred);
344
345 if (proc != NULL)
346 PROC_UNLOCK(proc);
347
348 PFS_RETURN (error);
349 }
350
351 /*
352 * Convert a vnode to its component name
353 */
354 static int
pfs_vptocnp(struct vop_vptocnp_args * ap)355 pfs_vptocnp(struct vop_vptocnp_args *ap)
356 {
357 struct vnode *vp = ap->a_vp;
358 struct vnode **dvp = ap->a_vpp;
359 struct pfs_vdata *pvd = vp->v_data;
360 struct pfs_node *pd = pvd->pvd_pn;
361 struct pfs_node *pn;
362 struct mount *mp;
363 char *buf = ap->a_buf;
364 int *buflen = ap->a_buflen;
365 char pidbuf[PFS_NAMELEN];
366 pid_t pid = pvd->pvd_pid;
367 int len, i, error, locked;
368
369 i = *buflen;
370 error = 0;
371
372 pfs_lock(pd);
373
374 if (vp->v_type == VDIR && pd->pn_type == pfstype_root) {
375 *dvp = vp;
376 vhold(*dvp);
377 pfs_unlock(pd);
378 PFS_RETURN (0);
379 } else if (vp->v_type == VDIR && pd->pn_type == pfstype_procdir) {
380 len = snprintf(pidbuf, sizeof(pidbuf), "%d", pid);
381 i -= len;
382 if (i < 0) {
383 error = ENOMEM;
384 goto failed;
385 }
386 bcopy(pidbuf, buf + i, len);
387 } else {
388 len = strlen(pd->pn_name);
389 i -= len;
390 if (i < 0) {
391 error = ENOMEM;
392 goto failed;
393 }
394 bcopy(pd->pn_name, buf + i, len);
395 }
396
397 pn = pd->pn_parent;
398 pfs_unlock(pd);
399
400 mp = vp->v_mount;
401 error = vfs_busy(mp, 0);
402 if (error)
403 return (error);
404
405 /*
406 * vp is held by caller.
407 */
408 locked = VOP_ISLOCKED(vp);
409 VOP_UNLOCK(vp, 0);
410
411 error = pfs_vncache_alloc(mp, dvp, pn, pid);
412 if (error) {
413 vn_lock(vp, locked | LK_RETRY);
414 vfs_unbusy(mp);
415 PFS_RETURN(error);
416 }
417
418 *buflen = i;
419 VOP_UNLOCK(*dvp, 0);
420 vn_lock(vp, locked | LK_RETRY);
421 vfs_unbusy(mp);
422
423 PFS_RETURN (0);
424 failed:
425 pfs_unlock(pd);
426 PFS_RETURN(error);
427 }
428
429 /*
430 * Look up a file or directory
431 */
432 static int
pfs_lookup(struct vop_cachedlookup_args * va)433 pfs_lookup(struct vop_cachedlookup_args *va)
434 {
435 struct vnode *vn = va->a_dvp;
436 struct vnode **vpp = va->a_vpp;
437 struct componentname *cnp = va->a_cnp;
438 struct pfs_vdata *pvd = vn->v_data;
439 struct pfs_node *pd = pvd->pvd_pn;
440 struct pfs_node *pn, *pdn = NULL;
441 struct mount *mp;
442 pid_t pid = pvd->pvd_pid;
443 char *pname;
444 int error, i, namelen, visible;
445
446 PFS_TRACE(("%.*s", (int)cnp->cn_namelen, cnp->cn_nameptr));
447 pfs_assert_not_owned(pd);
448
449 if (vn->v_type != VDIR)
450 PFS_RETURN (ENOTDIR);
451 KASSERT_PN_IS_DIR(pd);
452
453 /*
454 * Don't support DELETE or RENAME. CREATE is supported so
455 * that O_CREAT will work, but the lookup will still fail if
456 * the file does not exist.
457 */
458 if ((cnp->cn_flags & ISLASTCN) &&
459 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
460 PFS_RETURN (EOPNOTSUPP);
461
462 /* shortcut: check if the name is too long */
463 if (cnp->cn_namelen >= PFS_NAMELEN)
464 PFS_RETURN (ENOENT);
465
466 /* check that parent directory is visible... */
467 if (!pfs_visible(curthread, pd, pvd->pvd_pid, false, NULL))
468 PFS_RETURN (ENOENT);
469
470 /* self */
471 namelen = cnp->cn_namelen;
472 pname = cnp->cn_nameptr;
473 if (namelen == 1 && pname[0] == '.') {
474 pn = pd;
475 *vpp = vn;
476 VREF(vn);
477 PFS_RETURN (0);
478 }
479
480 mp = vn->v_mount;
481
482 /* parent */
483 if (cnp->cn_flags & ISDOTDOT) {
484 if (pd->pn_type == pfstype_root)
485 PFS_RETURN (EIO);
486 error = vfs_busy(mp, MBF_NOWAIT);
487 if (error != 0) {
488 vfs_ref(mp);
489 VOP_UNLOCK(vn, 0);
490 error = vfs_busy(mp, 0);
491 vn_lock(vn, LK_EXCLUSIVE | LK_RETRY);
492 vfs_rel(mp);
493 if (error != 0)
494 PFS_RETURN(ENOENT);
495 if (vn->v_iflag & VI_DOOMED) {
496 vfs_unbusy(mp);
497 PFS_RETURN(ENOENT);
498 }
499 }
500 VOP_UNLOCK(vn, 0);
501 KASSERT(pd->pn_parent != NULL,
502 ("%s(): non-root directory has no parent", __func__));
503 /*
504 * This one is tricky. Descendents of procdir nodes
505 * inherit their parent's process affinity, but
506 * there's no easy reverse mapping. For simplicity,
507 * we assume that if this node is a procdir, its
508 * parent isn't (which is correct as long as
509 * descendents of procdir nodes are never procdir
510 * nodes themselves)
511 */
512 if (pd->pn_type == pfstype_procdir)
513 pid = NO_PID;
514 pfs_lock(pd);
515 pn = pd->pn_parent;
516 pfs_unlock(pd);
517 goto got_pnode;
518 }
519
520 pfs_lock(pd);
521
522 /* named node */
523 for (pn = pd->pn_nodes; pn != NULL; pn = pn->pn_next)
524 if (pn->pn_type == pfstype_procdir)
525 pdn = pn;
526 else if (pn->pn_name[namelen] == '\0' &&
527 bcmp(pname, pn->pn_name, namelen) == 0) {
528 pfs_unlock(pd);
529 goto got_pnode;
530 }
531
532 /* process dependent node */
533 if ((pn = pdn) != NULL) {
534 pid = 0;
535 for (pid = 0, i = 0; i < namelen && isdigit(pname[i]); ++i)
536 if ((pid = pid * 10 + pname[i] - '0') > PID_MAX)
537 break;
538 if (i == cnp->cn_namelen) {
539 pfs_unlock(pd);
540 goto got_pnode;
541 }
542 }
543
544 pfs_unlock(pd);
545
546 PFS_RETURN (ENOENT);
547
548 got_pnode:
549 pfs_assert_not_owned(pd);
550 pfs_assert_not_owned(pn);
551 visible = pfs_visible(curthread, pn, pid, false, NULL);
552 if (!visible) {
553 error = ENOENT;
554 goto failed;
555 }
556
557 error = pfs_vncache_alloc(mp, vpp, pn, pid);
558 if (error)
559 goto failed;
560
561 if (cnp->cn_flags & ISDOTDOT) {
562 vfs_unbusy(mp);
563 vn_lock(vn, LK_EXCLUSIVE | LK_RETRY);
564 if (vn->v_iflag & VI_DOOMED) {
565 vput(*vpp);
566 *vpp = NULL;
567 PFS_RETURN(ENOENT);
568 }
569 }
570 if (cnp->cn_flags & MAKEENTRY && !(vn->v_iflag & VI_DOOMED))
571 cache_enter(vn, *vpp, cnp);
572 PFS_RETURN (0);
573 failed:
574 if (cnp->cn_flags & ISDOTDOT) {
575 vfs_unbusy(mp);
576 vn_lock(vn, LK_EXCLUSIVE | LK_RETRY);
577 *vpp = NULL;
578 }
579 PFS_RETURN(error);
580 }
581
582 /*
583 * Open a file or directory.
584 */
585 static int
pfs_open(struct vop_open_args * va)586 pfs_open(struct vop_open_args *va)
587 {
588 struct vnode *vn = va->a_vp;
589 struct pfs_vdata *pvd = vn->v_data;
590 struct pfs_node *pn = pvd->pvd_pn;
591 int mode = va->a_mode;
592
593 PFS_TRACE(("%s (mode 0x%x)", pn->pn_name, mode));
594 pfs_assert_not_owned(pn);
595
596 /* check if the requested mode is permitted */
597 if (((mode & FREAD) && !(mode & PFS_RD)) ||
598 ((mode & FWRITE) && !(mode & PFS_WR)))
599 PFS_RETURN (EPERM);
600
601 /* we don't support locking */
602 if ((mode & O_SHLOCK) || (mode & O_EXLOCK))
603 PFS_RETURN (EOPNOTSUPP);
604
605 PFS_RETURN (0);
606 }
607
608 /*
609 * Read from a file
610 */
611 static int
pfs_read(struct vop_read_args * va)612 pfs_read(struct vop_read_args *va)
613 {
614 struct vnode *vn = va->a_vp;
615 struct pfs_vdata *pvd = vn->v_data;
616 struct pfs_node *pn = pvd->pvd_pn;
617 struct uio *uio = va->a_uio;
618 struct proc *proc;
619 struct sbuf *sb = NULL;
620 int error, locked;
621 off_t buflen;
622
623 PFS_TRACE(("%s", pn->pn_name));
624 pfs_assert_not_owned(pn);
625
626 if (vn->v_type != VREG)
627 PFS_RETURN (EINVAL);
628 KASSERT_PN_IS_FILE(pn);
629
630 if (!(pn->pn_flags & PFS_RD))
631 PFS_RETURN (EBADF);
632
633 if (pn->pn_fill == NULL)
634 PFS_RETURN (EIO);
635
636 /*
637 * This is necessary because either process' privileges may
638 * have changed since the open() call.
639 */
640 if (!pfs_visible(curthread, pn, pvd->pvd_pid, false, &proc))
641 PFS_RETURN (EIO);
642 if (proc != NULL) {
643 _PHOLD(proc);
644 PROC_UNLOCK(proc);
645 }
646
647 vhold(vn);
648 locked = VOP_ISLOCKED(vn);
649 VOP_UNLOCK(vn, 0);
650
651 if (pn->pn_flags & PFS_RAWRD) {
652 PFS_TRACE(("%zd resid", uio->uio_resid));
653 error = pn_fill(curthread, proc, pn, NULL, uio);
654 PFS_TRACE(("%zd resid", uio->uio_resid));
655 goto ret;
656 }
657
658 if (uio->uio_resid < 0 || uio->uio_offset < 0 ||
659 uio->uio_resid > OFF_MAX - uio->uio_offset) {
660 error = EINVAL;
661 goto ret;
662 }
663 buflen = uio->uio_offset + uio->uio_resid;
664 if (buflen > PFS_MAXBUFSIZ)
665 buflen = PFS_MAXBUFSIZ;
666
667 sb = sbuf_new(sb, NULL, buflen + 1, 0);
668 if (sb == NULL) {
669 error = EIO;
670 goto ret;
671 }
672
673 error = pn_fill(curthread, proc, pn, sb, uio);
674
675 if (error) {
676 sbuf_delete(sb);
677 goto ret;
678 }
679
680 /*
681 * XXX: If the buffer overflowed, sbuf_len() will not return
682 * the data length. Then just use the full length because an
683 * overflowed sbuf must be full.
684 */
685 if (sbuf_finish(sb) == 0)
686 buflen = sbuf_len(sb);
687 error = uiomove_frombuf(sbuf_data(sb), buflen, uio);
688 sbuf_delete(sb);
689 ret:
690 vn_lock(vn, locked | LK_RETRY);
691 vdrop(vn);
692 if (proc != NULL)
693 PRELE(proc);
694 PFS_RETURN (error);
695 }
696
697 /*
698 * Iterate through directory entries
699 */
700 static int
pfs_iterate(struct thread * td,struct proc * proc,struct pfs_node * pd,struct pfs_node ** pn,struct proc ** p)701 pfs_iterate(struct thread *td, struct proc *proc, struct pfs_node *pd,
702 struct pfs_node **pn, struct proc **p)
703 {
704 int visible;
705
706 sx_assert(&allproc_lock, SX_SLOCKED);
707 pfs_assert_owned(pd);
708 again:
709 if (*pn == NULL) {
710 /* first node */
711 *pn = pd->pn_nodes;
712 } else if ((*pn)->pn_type != pfstype_procdir) {
713 /* next node */
714 *pn = (*pn)->pn_next;
715 }
716 if (*pn != NULL && (*pn)->pn_type == pfstype_procdir) {
717 /* next process */
718 if (*p == NULL)
719 *p = LIST_FIRST(&allproc);
720 else
721 *p = LIST_NEXT(*p, p_list);
722 /* out of processes: next node */
723 if (*p == NULL)
724 *pn = (*pn)->pn_next;
725 else
726 PROC_LOCK(*p);
727 }
728
729 if ((*pn) == NULL)
730 return (-1);
731
732 if (*p != NULL) {
733 visible = pfs_visible_proc(td, *pn, *p);
734 PROC_UNLOCK(*p);
735 } else if (proc != NULL) {
736 visible = pfs_visible_proc(td, *pn, proc);
737 } else {
738 visible = 1;
739 }
740 if (!visible)
741 goto again;
742
743 return (0);
744 }
745
746 /* Directory entry list */
747 struct pfsentry {
748 STAILQ_ENTRY(pfsentry) link;
749 struct dirent entry;
750 };
751 STAILQ_HEAD(pfsdirentlist, pfsentry);
752
753 /*
754 * Return directory entries.
755 */
756 static int
pfs_readdir(struct vop_readdir_args * va)757 pfs_readdir(struct vop_readdir_args *va)
758 {
759 struct vnode *vn = va->a_vp;
760 struct pfs_vdata *pvd = vn->v_data;
761 struct pfs_node *pd = pvd->pvd_pn;
762 pid_t pid = pvd->pvd_pid;
763 struct proc *p, *proc;
764 struct pfs_node *pn;
765 struct uio *uio;
766 struct pfsentry *pfsent, *pfsent2;
767 struct pfsdirentlist lst;
768 off_t offset;
769 int error, i, resid;
770
771 STAILQ_INIT(&lst);
772 error = 0;
773 KASSERT(pd->pn_info == vn->v_mount->mnt_data,
774 ("%s(): pn_info does not match mountpoint", __func__));
775 PFS_TRACE(("%s pid %lu", pd->pn_name, (unsigned long)pid));
776 pfs_assert_not_owned(pd);
777
778 if (vn->v_type != VDIR)
779 PFS_RETURN (ENOTDIR);
780 KASSERT_PN_IS_DIR(pd);
781 uio = va->a_uio;
782
783 /* only allow reading entire entries */
784 offset = uio->uio_offset;
785 resid = uio->uio_resid;
786 if (offset < 0 || offset % PFS_DELEN != 0 ||
787 (resid && resid < PFS_DELEN))
788 PFS_RETURN (EINVAL);
789 if (resid == 0)
790 PFS_RETURN (0);
791
792 sx_slock(&allproc_lock);
793 pfs_lock(pd);
794
795 /* check if the directory is visible to the caller */
796 if (!pfs_visible(curthread, pd, pid, true, &proc)) {
797 sx_sunlock(&allproc_lock);
798 pfs_unlock(pd);
799 PFS_RETURN (ENOENT);
800 }
801 KASSERT(pid == NO_PID || proc != NULL,
802 ("%s(): no process for pid %lu", __func__, (unsigned long)pid));
803
804 /* skip unwanted entries */
805 for (pn = NULL, p = NULL; offset > 0; offset -= PFS_DELEN) {
806 if (pfs_iterate(curthread, proc, pd, &pn, &p) == -1) {
807 /* nothing left... */
808 if (proc != NULL)
809 PROC_UNLOCK(proc);
810 pfs_unlock(pd);
811 sx_sunlock(&allproc_lock);
812 PFS_RETURN (0);
813 }
814 }
815
816 /* fill in entries */
817 while (pfs_iterate(curthread, proc, pd, &pn, &p) != -1 &&
818 resid >= PFS_DELEN) {
819 if ((pfsent = malloc(sizeof(struct pfsentry), M_IOV,
820 M_NOWAIT | M_ZERO)) == NULL) {
821 error = ENOMEM;
822 break;
823 }
824 pfsent->entry.d_reclen = PFS_DELEN;
825 pfsent->entry.d_fileno = pn_fileno(pn, pid);
826 /* PFS_DELEN was picked to fit PFS_NAMLEN */
827 for (i = 0; i < PFS_NAMELEN - 1 && pn->pn_name[i] != '\0'; ++i)
828 pfsent->entry.d_name[i] = pn->pn_name[i];
829 pfsent->entry.d_namlen = i;
830 /* NOTE: d_off is the offset of the *next* entry. */
831 pfsent->entry.d_off = offset + PFS_DELEN;
832 switch (pn->pn_type) {
833 case pfstype_procdir:
834 KASSERT(p != NULL,
835 ("reached procdir node with p == NULL"));
836 pfsent->entry.d_namlen = snprintf(pfsent->entry.d_name,
837 PFS_NAMELEN, "%d", p->p_pid);
838 /* fall through */
839 case pfstype_root:
840 case pfstype_dir:
841 case pfstype_this:
842 case pfstype_parent:
843 pfsent->entry.d_type = DT_DIR;
844 break;
845 case pfstype_file:
846 pfsent->entry.d_type = DT_REG;
847 break;
848 case pfstype_symlink:
849 pfsent->entry.d_type = DT_LNK;
850 break;
851 default:
852 panic("%s has unexpected node type: %d", pn->pn_name, pn->pn_type);
853 }
854 PFS_TRACE(("%s", pfsent->entry.d_name));
855 dirent_terminate(&pfsent->entry);
856 STAILQ_INSERT_TAIL(&lst, pfsent, link);
857 offset += PFS_DELEN;
858 resid -= PFS_DELEN;
859 }
860 if (proc != NULL)
861 PROC_UNLOCK(proc);
862 pfs_unlock(pd);
863 sx_sunlock(&allproc_lock);
864 i = 0;
865 STAILQ_FOREACH_SAFE(pfsent, &lst, link, pfsent2) {
866 if (error == 0)
867 error = uiomove(&pfsent->entry, PFS_DELEN, uio);
868 free(pfsent, M_IOV);
869 i++;
870 }
871 PFS_TRACE(("%ju bytes", (uintmax_t)(i * PFS_DELEN)));
872 PFS_RETURN (error);
873 }
874
875 /*
876 * Read a symbolic link
877 */
878 static int
pfs_readlink(struct vop_readlink_args * va)879 pfs_readlink(struct vop_readlink_args *va)
880 {
881 struct vnode *vn = va->a_vp;
882 struct pfs_vdata *pvd = vn->v_data;
883 struct pfs_node *pn = pvd->pvd_pn;
884 struct uio *uio = va->a_uio;
885 struct proc *proc = NULL;
886 char buf[PATH_MAX];
887 struct sbuf sb;
888 int error, locked;
889
890 PFS_TRACE(("%s", pn->pn_name));
891 pfs_assert_not_owned(pn);
892
893 if (vn->v_type != VLNK)
894 PFS_RETURN (EINVAL);
895 KASSERT_PN_IS_LINK(pn);
896
897 if (pn->pn_fill == NULL)
898 PFS_RETURN (EIO);
899
900 if (pvd->pvd_pid != NO_PID) {
901 if ((proc = pfind(pvd->pvd_pid)) == NULL)
902 PFS_RETURN (EIO);
903 if (proc->p_flag & P_WEXIT) {
904 PROC_UNLOCK(proc);
905 PFS_RETURN (EIO);
906 }
907 _PHOLD(proc);
908 PROC_UNLOCK(proc);
909 }
910 vhold(vn);
911 locked = VOP_ISLOCKED(vn);
912 VOP_UNLOCK(vn, 0);
913
914 /* sbuf_new() can't fail with a static buffer */
915 sbuf_new(&sb, buf, sizeof buf, 0);
916
917 error = pn_fill(curthread, proc, pn, &sb, NULL);
918
919 if (proc != NULL)
920 PRELE(proc);
921 vn_lock(vn, locked | LK_RETRY);
922 vdrop(vn);
923
924 if (error) {
925 sbuf_delete(&sb);
926 PFS_RETURN (error);
927 }
928
929 if (sbuf_finish(&sb) != 0) {
930 sbuf_delete(&sb);
931 PFS_RETURN (ENAMETOOLONG);
932 }
933
934 error = uiomove_frombuf(sbuf_data(&sb), sbuf_len(&sb), uio);
935 sbuf_delete(&sb);
936 PFS_RETURN (error);
937 }
938
939 /*
940 * Reclaim a vnode
941 */
942 static int
pfs_reclaim(struct vop_reclaim_args * va)943 pfs_reclaim(struct vop_reclaim_args *va)
944 {
945 struct vnode *vn = va->a_vp;
946 struct pfs_vdata *pvd = vn->v_data;
947 struct pfs_node *pn = pvd->pvd_pn;
948
949 PFS_TRACE(("%s", pn->pn_name));
950 pfs_assert_not_owned(pn);
951
952 return (pfs_vncache_free(va->a_vp));
953 }
954
955 /*
956 * Set attributes
957 */
958 static int
pfs_setattr(struct vop_setattr_args * va)959 pfs_setattr(struct vop_setattr_args *va)
960 {
961 struct vnode *vn = va->a_vp;
962 struct pfs_vdata *pvd = vn->v_data;
963 struct pfs_node *pn = pvd->pvd_pn;
964
965 PFS_TRACE(("%s", pn->pn_name));
966 pfs_assert_not_owned(pn);
967
968 /* Silently ignore unchangeable attributes. */
969 PFS_RETURN (0);
970 }
971
972 /*
973 * Write to a file
974 */
975 static int
pfs_write(struct vop_write_args * va)976 pfs_write(struct vop_write_args *va)
977 {
978 struct vnode *vn = va->a_vp;
979 struct pfs_vdata *pvd = vn->v_data;
980 struct pfs_node *pn = pvd->pvd_pn;
981 struct uio *uio = va->a_uio;
982 struct proc *proc;
983 struct sbuf sb;
984 int error;
985
986 PFS_TRACE(("%s", pn->pn_name));
987 pfs_assert_not_owned(pn);
988
989 if (vn->v_type != VREG)
990 PFS_RETURN (EINVAL);
991 KASSERT_PN_IS_FILE(pn);
992
993 if (!(pn->pn_flags & PFS_WR))
994 PFS_RETURN (EBADF);
995
996 if (pn->pn_fill == NULL)
997 PFS_RETURN (EIO);
998
999 if (uio->uio_resid > PFS_MAXBUFSIZ)
1000 PFS_RETURN (EIO);
1001
1002 /*
1003 * This is necessary because either process' privileges may
1004 * have changed since the open() call.
1005 */
1006 if (!pfs_visible(curthread, pn, pvd->pvd_pid, false, &proc))
1007 PFS_RETURN (EIO);
1008 if (proc != NULL) {
1009 _PHOLD(proc);
1010 PROC_UNLOCK(proc);
1011 }
1012
1013 if (pn->pn_flags & PFS_RAWWR) {
1014 error = pn_fill(curthread, proc, pn, NULL, uio);
1015 if (proc != NULL)
1016 PRELE(proc);
1017 PFS_RETURN (error);
1018 }
1019
1020 sbuf_uionew(&sb, uio, &error);
1021 if (error) {
1022 if (proc != NULL)
1023 PRELE(proc);
1024 PFS_RETURN (error);
1025 }
1026
1027 error = pn_fill(curthread, proc, pn, &sb, uio);
1028
1029 sbuf_delete(&sb);
1030 if (proc != NULL)
1031 PRELE(proc);
1032 PFS_RETURN (error);
1033 }
1034
1035 /*
1036 * Vnode operations
1037 */
1038 struct vop_vector pfs_vnodeops = {
1039 .vop_default = &default_vnodeops,
1040
1041 .vop_access = pfs_access,
1042 .vop_cachedlookup = pfs_lookup,
1043 .vop_close = pfs_close,
1044 .vop_create = VOP_EOPNOTSUPP,
1045 .vop_getattr = pfs_getattr,
1046 .vop_getextattr = pfs_getextattr,
1047 .vop_ioctl = pfs_ioctl,
1048 .vop_link = VOP_EOPNOTSUPP,
1049 .vop_lookup = vfs_cache_lookup,
1050 .vop_mkdir = VOP_EOPNOTSUPP,
1051 .vop_mknod = VOP_EOPNOTSUPP,
1052 .vop_open = pfs_open,
1053 .vop_read = pfs_read,
1054 .vop_readdir = pfs_readdir,
1055 .vop_readlink = pfs_readlink,
1056 .vop_reclaim = pfs_reclaim,
1057 .vop_remove = VOP_EOPNOTSUPP,
1058 .vop_rename = VOP_EOPNOTSUPP,
1059 .vop_rmdir = VOP_EOPNOTSUPP,
1060 .vop_setattr = pfs_setattr,
1061 .vop_symlink = VOP_EOPNOTSUPP,
1062 .vop_vptocnp = pfs_vptocnp,
1063 .vop_write = pfs_write,
1064 /* XXX I've probably forgotten a few that need VOP_EOPNOTSUPP */
1065 };
1066