1 /*
2 * Copyright (c) 1993 Jan-Simon Pendry
3 * Copyright (c) 1993
4 * The Regents of the University of California. All rights reserved.
5 *
6 * This code is derived from software contributed to Berkeley by
7 * Jan-Simon Pendry.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 *
33 * @(#)procfs_subr.c 8.6 (Berkeley) 5/14/95
34 *
35 * $FreeBSD: src/sys/miscfs/procfs/procfs_subr.c,v 1.26.2.3 2002/02/18 21:28:04 des Exp $
36 */
37
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/sysctl.h>
41 #include <sys/uio.h>
42 #include <sys/proc.h>
43 #include <sys/mount.h>
44 #include <sys/vnode.h>
45 #include <sys/malloc.h>
46 #include <sys/spinlock.h>
47
48 #include <sys/spinlock2.h>
49
50 #include <vfs/procfs/procfs.h>
51
52 #define PFS_HSIZE 1031
53
54 struct pfshead {
55 struct spinlock spin;
56 struct pfsnode *first;
57 } __cachealign;
58
59 static struct pfshead pfshead[PFS_HSIZE];
60 static struct lock procfslk = LOCK_INITIALIZER("pvplk", 0, 0);
61
62 MALLOC_DEFINE(M_PROCFS, "procfs", "procfs v_data");
63
64 #define PFSHASH(pid) &pfshead[((pid) & ~PFS_DEAD) % PFS_HSIZE]
65
66 /*
67 * Allocate a pfsnode/vnode pair. If no error occurs the returned vnode
68 * will be referenced and exclusively locked.
69 *
70 * The pid, pfs_type, and mount point uniquely identify a pfsnode.
71 * The mount point is needed because someone might mount this filesystem
72 * twice.
73 *
74 * All pfsnodes are maintained on a singly-linked list. new nodes are
75 * only allocated when they cannot be found on this list. entries on
76 * the list are removed when the vfs reclaim entry is called.
77 *
78 * A single lock is kept for the entire list. this is needed because the
79 * getnewvnode() function can block waiting for a vnode to become free,
80 * in which case there may be more than one process trying to get the same
81 * vnode. this lock is only taken if we are going to call getnewvnode,
82 * since the kernel itself is single-threaded.
83 *
84 * If an entry is found on the list, then call vget() to take a reference
85 * and obtain the lock. This will properly re-reference the vnode if it
86 * had gotten onto the free list.
87 */
88 int
procfs_allocvp(struct mount * mp,struct vnode ** vpp,long pid,pfstype pfs_type)89 procfs_allocvp(struct mount *mp, struct vnode **vpp, long pid, pfstype pfs_type)
90 {
91 struct pfsnode *pfs;
92 struct vnode *vp;
93 struct pfshead *ph;
94 int error;
95
96 ph = PFSHASH(pid);
97 loop:
98 spin_lock(&ph->spin);
99 for (pfs = ph->first; pfs; pfs = pfs->pfs_next) {
100 if (pfs->pfs_pid == pid && pfs->pfs_type == pfs_type &&
101 PFSTOV(pfs)->v_mount == mp) {
102 vp = PFSTOV(pfs);
103 vhold(vp);
104 spin_unlock(&ph->spin);
105 if (vget(vp, LK_EXCLUSIVE)) {
106 vdrop(vp);
107 goto loop;
108 }
109 vdrop(vp);
110
111 /*
112 * Make sure the vnode is still in the cache after
113 * getting the interlock to avoid racing a free.
114 */
115 spin_lock(&ph->spin);
116 for (pfs = ph->first; pfs; pfs = pfs->pfs_next) {
117 if (PFSTOV(pfs) == vp &&
118 pfs->pfs_pid == pid &&
119 pfs->pfs_type == pfs_type &&
120 PFSTOV(pfs)->v_mount == mp) {
121 break;
122 }
123 }
124 if (pfs == NULL || PFSTOV(pfs) != vp) {
125 spin_unlock(&ph->spin);
126 vput(vp);
127 goto loop;
128
129 }
130 spin_unlock(&ph->spin);
131 *vpp = vp;
132 return (0);
133 }
134 }
135 spin_unlock(&ph->spin);
136
137 /*
138 * otherwise lock the vp list while we call getnewvnode
139 * since that can block.
140 */
141 if (lockmgr(&procfslk, LK_EXCLUSIVE|LK_SLEEPFAIL))
142 goto loop;
143
144 /*
145 * Do the MALLOC before the getnewvnode since doing so afterward
146 * might cause a bogus v_data pointer to get dereferenced
147 * elsewhere if MALLOC should block.
148 *
149 * XXX this may not matter anymore since getnewvnode now returns
150 * a VX locked vnode.
151 */
152 pfs = kmalloc(sizeof(struct pfsnode), M_PROCFS, M_WAITOK);
153
154 error = getnewvnode(VT_PROCFS, mp, vpp, 0, 0);
155 if (error) {
156 kfree(pfs, M_PROCFS);
157 goto out;
158 }
159 vp = *vpp;
160
161 vp->v_data = pfs;
162
163 pfs->pfs_next = 0;
164 pfs->pfs_pid = (pid_t) pid;
165 pfs->pfs_type = pfs_type;
166 pfs->pfs_vnode = vp;
167 pfs->pfs_flags = 0;
168 pfs->pfs_fileno = PROCFS_FILENO(pid, pfs_type);
169 lockinit(&pfs->pfs_lock, "pfslk", 0, 0);
170
171 switch (pfs_type) {
172 case Proot: /* /proc = dr-xr-xr-x */
173 pfs->pfs_mode = (VREAD|VEXEC) |
174 (VREAD|VEXEC) >> 3 |
175 (VREAD|VEXEC) >> 6;
176 vp->v_type = VDIR;
177 vp->v_flag = VROOT;
178 break;
179
180 case Pcurproc: /* /proc/curproc = lr--r--r-- */
181 pfs->pfs_mode = (VREAD) |
182 (VREAD >> 3) |
183 (VREAD >> 6);
184 vp->v_type = VLNK;
185 break;
186
187 case Pproc:
188 pfs->pfs_mode = (VREAD|VEXEC) |
189 (VREAD|VEXEC) >> 3 |
190 (VREAD|VEXEC) >> 6;
191 vp->v_type = VDIR;
192 break;
193
194 case Pfile:
195 pfs->pfs_mode = (VREAD|VEXEC) |
196 (VREAD|VEXEC) >> 3 |
197 (VREAD|VEXEC) >> 6;
198 vp->v_type = VLNK;
199 break;
200
201 case Pmem:
202 pfs->pfs_mode = (VREAD|VWRITE);
203 vp->v_type = VREG;
204 break;
205
206 case Pregs:
207 case Pfpregs:
208 case Pdbregs:
209 pfs->pfs_mode = (VREAD|VWRITE);
210 vp->v_type = VREG;
211 break;
212
213 case Pctl:
214 case Pnote:
215 case Pnotepg:
216 pfs->pfs_mode = (VWRITE);
217 vp->v_type = VREG;
218 break;
219
220 case Ptype:
221 case Pmap:
222 case Pstatus:
223 case Pcmdline:
224 case Prlimit:
225 pfs->pfs_mode = (VREAD) |
226 (VREAD >> 3) |
227 (VREAD >> 6);
228 vp->v_type = VREG;
229 break;
230
231 default:
232 panic("procfs_allocvp");
233 }
234
235 /* add to procfs vnode list */
236 spin_lock(&ph->spin);
237 pfs->pfs_next = ph->first;
238 ph->first = pfs;
239 spin_unlock(&ph->spin);
240 vx_downgrade(vp);
241
242 out:
243 lockmgr(&procfslk, LK_RELEASE);
244
245 return (error);
246 }
247
248 int
procfs_freevp(struct vnode * vp)249 procfs_freevp(struct vnode *vp)
250 {
251 struct pfshead *ph;
252 struct pfsnode **pp;
253 struct pfsnode *pfs;
254
255 pfs = VTOPFS(vp);
256 vp->v_data = NULL;
257 ph = PFSHASH(pfs->pfs_pid);
258
259 spin_lock(&ph->spin);
260 pp = &ph->first;
261 while (*pp != pfs) {
262 KKASSERT(*pp != NULL);
263 pp = &(*pp)->pfs_next;
264 }
265 *pp = pfs->pfs_next;
266 spin_unlock(&ph->spin);
267
268 pfs->pfs_next = NULL;
269 pfs->pfs_vnode = NULL;
270 kfree(pfs, M_PROCFS);
271
272 return (0);
273 }
274
275 /*
276 * Try to find the calling pid. Note that pfind()
277 * now references the proc structure to be returned
278 * and needs to be released later with PRELE().
279 */
280 struct proc *
pfs_pfind(pid_t pfs_pid)281 pfs_pfind(pid_t pfs_pid)
282 {
283 struct proc *p = NULL;
284
285 if (pfs_pid == 0) {
286 p = &proc0;
287 PHOLD(p);
288 } else {
289 p = pfind(pfs_pid);
290 }
291
292 /*
293 * Make sure the process is not in the middle of exiting (where
294 * a lot of its structural members may wind up being NULL). If it
295 * is we give up on it.
296 */
297 if (p) {
298 lwkt_gettoken(&p->p_token);
299 if (p->p_flags & P_POSTEXIT) {
300 lwkt_reltoken(&p->p_token);
301 PRELE(p);
302 p = NULL;
303 }
304 }
305 return p;
306 }
307
308 struct proc *
pfs_zpfind(pid_t pfs_pid)309 pfs_zpfind(pid_t pfs_pid)
310 {
311 struct proc *p = NULL;
312
313 if (pfs_pid == 0) {
314 p = &proc0;
315 PHOLD(p);
316 } else {
317 p = zpfind(pfs_pid);
318 }
319
320 /*
321 * Make sure the process is not in the middle of exiting (where
322 * a lot of its structural members may wind up being NULL). If it
323 * is we give up on it.
324 */
325 if (p) {
326 lwkt_gettoken(&p->p_token);
327 if (p->p_flags & P_POSTEXIT) {
328 lwkt_reltoken(&p->p_token);
329 PRELE(p);
330 p = NULL;
331 }
332 }
333 return p;
334 }
335
336 void
pfs_pdone(struct proc * p)337 pfs_pdone(struct proc *p)
338 {
339 if (p) {
340 lwkt_reltoken(&p->p_token);
341 PRELE(p);
342 }
343 }
344
345 int
procfs_rw(struct vop_read_args * ap)346 procfs_rw(struct vop_read_args *ap)
347 {
348 struct vnode *vp = ap->a_vp;
349 struct uio *uio = ap->a_uio;
350 struct thread *curtd = uio->uio_td;
351 struct proc *curp;
352 struct pfsnode *pfs = VTOPFS(vp);
353 struct proc *p;
354 struct lwp *lp;
355 int rtval;
356
357 if (curtd == NULL)
358 return (EINVAL);
359 if ((curp = curtd->td_proc) == NULL) /* XXX */
360 return (EINVAL);
361
362 p = pfs_pfind(pfs->pfs_pid);
363 if (p == NULL) {
364 rtval = EINVAL;
365 goto out;
366 }
367 if (p->p_pid == 1 && securelevel > 0 && uio->uio_rw == UIO_WRITE) {
368 rtval = EACCES;
369 goto out;
370 }
371
372 /*
373 * XXX lwp
374 */
375 lp = FIRST_LWP_IN_PROC(p);
376 if (lp == NULL) {
377 rtval = EINVAL;
378 goto out;
379 }
380 LWPHOLD(lp);
381
382 lockmgr(&pfs->pfs_lock, LK_EXCLUSIVE);
383
384 switch (pfs->pfs_type) {
385 case Pnote:
386 case Pnotepg:
387 rtval = procfs_donote(curp, lp, pfs, uio);
388 break;
389
390 case Pregs:
391 rtval = procfs_doregs(curp, lp, pfs, uio);
392 break;
393
394 case Pfpregs:
395 rtval = procfs_dofpregs(curp, lp, pfs, uio);
396 break;
397
398 case Pdbregs:
399 rtval = procfs_dodbregs(curp, lp, pfs, uio);
400 break;
401
402 case Pctl:
403 rtval = procfs_doctl(curp, lp, pfs, uio);
404 break;
405
406 case Pstatus:
407 rtval = procfs_dostatus(curp, lp, pfs, uio);
408 break;
409
410 case Pmap:
411 rtval = procfs_domap(curp, lp, pfs, uio);
412 break;
413
414 case Pmem:
415 rtval = procfs_domem(curp, lp, pfs, uio);
416 break;
417
418 case Ptype:
419 rtval = procfs_dotype(curp, lp, pfs, uio);
420 break;
421
422 case Pcmdline:
423 rtval = procfs_docmdline(curp, lp, pfs, uio);
424 break;
425
426 case Prlimit:
427 rtval = procfs_dorlimit(curp, lp, pfs, uio);
428 break;
429
430 default:
431 rtval = EOPNOTSUPP;
432 break;
433 }
434 LWPRELE(lp);
435
436 lockmgr(&pfs->pfs_lock, LK_RELEASE);
437
438 if (uio->uio_rw == UIO_WRITE && rtval == 0)
439 KNOTE(&PFSTOV(pfs)->v_pollinfo.vpi_kqinfo.ki_note, NOTE_WRITE);
440
441 out:
442 pfs_pdone(p);
443
444 return rtval;
445 }
446
447 /*
448 * Get a string from userland into (buf). Strip a trailing
449 * nl character (to allow easy access from the shell).
450 * The buffer should be *buflenp + 1 chars long. vfs_getuserstr
451 * will automatically add a nul char at the end.
452 *
453 * Returns 0 on success or the following errors
454 *
455 * EINVAL: file offset is non-zero.
456 * EMSGSIZE: message is longer than kernel buffer
457 * EFAULT: user i/o buffer is not addressable
458 */
459 int
vfs_getuserstr(struct uio * uio,char * buf,int * buflenp)460 vfs_getuserstr(struct uio *uio, char *buf, int *buflenp)
461 {
462 int xlen;
463 int error;
464
465 if (uio->uio_offset != 0)
466 return (EINVAL);
467
468 xlen = *buflenp;
469
470 /* must be able to read the whole string in one go */
471 if (xlen < uio->uio_resid)
472 return (EMSGSIZE);
473 xlen = uio->uio_resid;
474
475 if ((error = uiomove(buf, xlen, uio)) != 0)
476 return (error);
477
478 /* allow multiple writes without seeks */
479 uio->uio_offset = 0;
480
481 /* cleanup string and remove trailing newline */
482 buf[xlen] = '\0';
483 xlen = strlen(buf);
484 if (xlen > 0 && buf[xlen-1] == '\n')
485 buf[--xlen] = '\0';
486 *buflenp = xlen;
487
488 return (0);
489 }
490
491 vfs_namemap_t *
vfs_findname(vfs_namemap_t * nm,char * buf,int buflen)492 vfs_findname(vfs_namemap_t *nm, char *buf, int buflen)
493 {
494
495 for (; nm->nm_name; nm++)
496 if (bcmp(buf, nm->nm_name, buflen+1) == 0)
497 return (nm);
498
499 return (0);
500 }
501
502 void
procfs_exit(struct thread * td)503 procfs_exit(struct thread *td)
504 {
505 struct pfshead *ph;
506 struct pfsnode *pfs;
507 struct vnode *vp;
508 pid_t pid;
509
510 KKASSERT(td->td_proc);
511 pid = td->td_proc->p_pid;
512
513 /*
514 * NOTE: We can't just vgone() the vnode any more, not while
515 * it may potentially still be active. This will clean
516 * the vp and clear the mount and cause the new VOP subsystem
517 * to assert or panic when someone tries to do an operation
518 * on an open (exited) procfs descriptor.
519 *
520 * Prevent further operations on this pid by setting pfs_pid to -1.
521 * Note that a pfs_pid of 0 is used for nodes which do not track
522 * any particular pid.
523 *
524 * Use vx_get() to properly ref/lock a vp which may not have any
525 * refs and which may or may not already be reclaimed. vx_put()
526 * will then properly deactivate it and cause it to be recycled.
527 *
528 * The hash table can also get ripped out from under us when
529 * we block so take the easy way out and restart the scan.
530 */
531 for (;;) {
532 ph = PFSHASH(pid);
533 spin_lock(&ph->spin);
534 for (pfs = ph->first; pfs; pfs = pfs->pfs_next) {
535 if (pfs->pfs_pid == pid)
536 break;
537 }
538 if (pfs == NULL) {
539 spin_unlock(&ph->spin);
540 break;
541 }
542 vp = PFSTOV(pfs);
543 vhold(vp);
544 spin_unlock(&ph->spin);
545 vx_get(vp);
546 pfs->pfs_pid |= PFS_DEAD; /* does not effect hash */
547 vx_put(vp);
548 vdrop(vp);
549 }
550 }
551