sys/vfs/procfs/procfs_vnops.c

   1 /*
   2  * Copyright (c) 1993, 1995 Jan-Simon Pendry
   3  * Copyright (c) 1993, 1995
   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. All advertising materials mentioning features or use of this software
  18  *    must display the following acknowledgement:
  19  *      This product includes software developed by the University of
  20  *      California, Berkeley and its contributors.
  21  * 4. Neither the name of the University nor the names of its contributors
  22  *    may be used to endorse or promote products derived from this software
  23  *    without specific prior written permission.
  24  *
  25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  35  * SUCH DAMAGE.
  36  *
  37  *      @(#)procfs_vnops.c      8.18 (Berkeley) 5/21/95
  38  *
  39  * $FreeBSD: src/sys/miscfs/procfs/procfs_vnops.c,v 1.76.2.7 2002/01/22 17:22:59 nectar Exp $
  40  * $DragonFly: src/sys/vfs/procfs/procfs_vnops.c,v 1.46 2007/11/20 21:03:50 dillon Exp $
  41  */
  42
  43 /*
  44  * procfs vnode interface
  45  */
  46
  47 #include <sys/param.h>
  48 #include <sys/systm.h>
  49 #include <sys/time.h>
  50 #include <sys/kernel.h>
  51 #include <sys/lock.h>
  52 #include <sys/fcntl.h>
  53 #include <sys/proc.h>
  54 #include <sys/signalvar.h>
  55 #include <sys/vnode.h>
  56 #include <sys/uio.h>
  57 #include <sys/mount.h>
  58 #include <sys/namei.h>
  59 #include <sys/dirent.h>
  60 #include <sys/malloc.h>
  61 #include <sys/reg.h>
  62 #include <vm/vm_zone.h>
  63 #include <vfs/procfs/procfs.h>
  64 #include <sys/pioctl.h>
  65
  66 #include <machine/limits.h>
  67
  68 static int      procfs_access (struct vop_access_args *);
  69 static int      procfs_badop (struct vop_generic_args *);
  70 static int      procfs_bmap (struct vop_bmap_args *);
  71 static int      procfs_close (struct vop_close_args *);
  72 static int      procfs_getattr (struct vop_getattr_args *);
  73 static int      procfs_inactive (struct vop_inactive_args *);
  74 static int      procfs_ioctl (struct vop_ioctl_args *);
  75 static int      procfs_lookup (struct vop_old_lookup_args *);
  76 static int      procfs_open (struct vop_open_args *);
  77 static int      procfs_print (struct vop_print_args *);
  78 static int      procfs_readdir (struct vop_readdir_args *);
  79 static int      procfs_readlink (struct vop_readlink_args *);
  80 static int      procfs_reclaim (struct vop_reclaim_args *);
  81 static int      procfs_setattr (struct vop_setattr_args *);
  82
  83 static int      procfs_readdir_proc(struct vop_readdir_args *);
  84 static int      procfs_readdir_root(struct vop_readdir_args *);
  85
  86 /*
  87  * procfs vnode operations.
  88  */
  89 struct vop_ops procfs_vnode_vops = {
  90         .vop_default =          vop_defaultop,
  91         .vop_access =           procfs_access,
  92         .vop_advlock =          (void *)procfs_badop,
  93         .vop_bmap =             procfs_bmap,
  94         .vop_close =            procfs_close,
  95         .vop_old_create =       (void *)procfs_badop,
  96         .vop_getattr =          procfs_getattr,
  97         .vop_inactive =         procfs_inactive,
  98         .vop_old_link =         (void *)procfs_badop,
  99         .vop_old_lookup =       procfs_lookup,
 100         .vop_old_mkdir =        (void *)procfs_badop,
 101         .vop_old_mknod =        (void *)procfs_badop,
 102         .vop_open =             procfs_open,
 103         .vop_pathconf =         vop_stdpathconf,
 104         .vop_print =            procfs_print,
 105         .vop_read =             procfs_rw,
 106         .vop_readdir =          procfs_readdir,
 107         .vop_readlink =         procfs_readlink,
 108         .vop_reclaim =          procfs_reclaim,
 109         .vop_old_remove =       (void *)procfs_badop,
 110         .vop_old_rename =       (void *)procfs_badop,
 111         .vop_old_rmdir =        (void *)procfs_badop,
 112         .vop_setattr =          procfs_setattr,
 113         .vop_old_symlink =      (void *)procfs_badop,
 114         .vop_write =            (void *)procfs_rw,
 115         .vop_ioctl =            procfs_ioctl
 116 };
 117
 118
 119 /*
 120  * This is a list of the valid names in the
 121  * process-specific sub-directories.  It is
 122  * used in procfs_lookup and procfs_readdir
 123  */
 124 static struct proc_target {
 125         u_char  pt_type;
 126         u_char  pt_namlen;
 127         char    *pt_name;
 128         pfstype pt_pfstype;
 129         int     (*pt_valid) (struct lwp *p);
 130 } proc_targets[] = {
 131 #define N(s) sizeof(s)-1, s
 132         /*        name          type            validp */
 133         { DT_DIR, N("."),       Pproc,          NULL },
 134         { DT_DIR, N(".."),      Proot,          NULL },
 135         { DT_REG, N("mem"),     Pmem,           NULL },
 136         { DT_REG, N("regs"),    Pregs,          procfs_validregs },
 137         { DT_REG, N("fpregs"),  Pfpregs,        procfs_validfpregs },
 138         { DT_REG, N("dbregs"),  Pdbregs,        procfs_validdbregs },
 139         { DT_REG, N("ctl"),     Pctl,           NULL },
 140         { DT_REG, N("status"),  Pstatus,        NULL },
 141         { DT_REG, N("note"),    Pnote,          NULL },
 142         { DT_REG, N("notepg"),  Pnotepg,        NULL },
 143         { DT_REG, N("map"),     Pmap,           procfs_validmap },
 144         { DT_REG, N("etype"),   Ptype,          procfs_validtype },
 145         { DT_REG, N("cmdline"), Pcmdline,       NULL },
 146         { DT_REG, N("rlimit"),  Prlimit,        NULL },
 147         { DT_LNK, N("file"),    Pfile,          NULL },
 148 #undef N
 149 };
 150 static const int nproc_targets = sizeof(proc_targets) / sizeof(proc_targets[0]);
 151
 152 static pid_t atopid (const char *, u_int);
 153
 154 /*
 155  * set things up for doing i/o on
 156  * the pfsnode (vp).  (vp) is locked
 157  * on entry, and should be left locked
 158  * on exit.
 159  *
 160  * for procfs we don't need to do anything
 161  * in particular for i/o.  all that is done
 162  * is to support exclusive open on process
 163  * memory images.
 164  *
 165  * procfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
 166  *             struct file *a_fp)
 167  */
 168 static int
 169 procfs_open(struct vop_open_args *ap)
 170 {
 171         struct pfsnode *pfs = VTOPFS(ap->a_vp);
 172         struct proc *p1, *p2;
 173
 174         p2 = PFIND(pfs->pfs_pid);
 175         if (p2 == NULL)
 176                 return (ENOENT);
 177         if (pfs->pfs_pid && !PRISON_CHECK(ap->a_cred, p2->p_ucred))
 178                 return (ENOENT);
 179
 180         switch (pfs->pfs_type) {
 181         case Pmem:
 182                 if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) ||
 183                     ((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE)))
 184                         return (EBUSY);
 185
 186                 p1 = curproc;
 187                 KKASSERT(p1);
 188                 /* Can't trace a process that's currently exec'ing. */
 189                 if ((p2->p_flag & P_INEXEC) != 0)
 190                         return EAGAIN;
 191                 if (!CHECKIO(p1, p2) || p_trespass(ap->a_cred, p2->p_ucred))
 192                         return (EPERM);
 193
 194                 if (ap->a_mode & FWRITE)
 195                         pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
 196
 197                 break;
 198
 199         default:
 200                 break;
 201         }
 202
 203         return (vop_stdopen(ap));
 204 }
 205
 206 /*
 207  * close the pfsnode (vp) after doing i/o.
 208  * (vp) is not locked on entry or exit.
 209  *
 210  * nothing to do for procfs other than undo
 211  * any exclusive open flag (see _open above).
 212  *
 213  * procfs_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred)
 214  */
 215 static int
 216 procfs_close(struct vop_close_args *ap)
 217 {
 218         struct pfsnode *pfs = VTOPFS(ap->a_vp);
 219         struct proc *p;
 220
 221         switch (pfs->pfs_type) {
 222         case Pmem:
 223                 if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
 224                         pfs->pfs_flags &= ~(FWRITE|O_EXCL);
 225                 /*
 226                  * v_opencount determines the last real close on the vnode.
 227                  *
 228                  * If this is the last close, then it checks to see if
 229                  * the target process has PF_LINGER set in p_pfsflags,
 230                  * if this is *not* the case, then the process' stop flags
 231                  * are cleared, and the process is woken up.  This is
 232                  * to help prevent the case where a process has been
 233                  * told to stop on an event, but then the requesting process
 234                  * has gone away or forgotten about it.
 235                  */
 236                 if ((ap->a_vp->v_opencount < 2)
 237                     && (p = pfind(pfs->pfs_pid))
 238                     && !(p->p_pfsflags & PF_LINGER)) {
 239                         p->p_stops = 0;
 240                         p->p_step = 0;
 241                         wakeup(&p->p_step);
 242                 }
 243                 break;
 244         default:
 245                 break;
 246         }
 247
 248         return (vop_stdclose(ap));
 249 }
 250
 251 /*
 252  * do an ioctl operation on a pfsnode (vp).
 253  * (vp) is not locked on entry or exit.
 254  */
 255 static int
 256 procfs_ioctl(struct vop_ioctl_args *ap)
 257 {
 258         struct pfsnode *pfs = VTOPFS(ap->a_vp);
 259         struct proc *procp;
 260         struct proc *p;
 261         int error;
 262         int signo;
 263         struct procfs_status *psp;
 264         unsigned char flags;
 265
 266         procp = pfind(pfs->pfs_pid);
 267         if (procp == NULL)
 268                 return ENOTTY;
 269         p = curproc;
 270         if (p == NULL)
 271                 return EINVAL;
 272
 273         /* Can't trace a process that's currently exec'ing. */
 274         if ((procp->p_flag & P_INEXEC) != 0)
 275                 return EAGAIN;
 276         if (!CHECKIO(p, procp) || p_trespass(ap->a_cred, procp->p_ucred))
 277                 return EPERM;
 278
 279         switch (ap->a_command) {
 280         case PIOCBIS:
 281           procp->p_stops |= *(unsigned int*)ap->a_data;
 282           break;
 283         case PIOCBIC:
 284           procp->p_stops &= ~*(unsigned int*)ap->a_data;
 285           break;
 286         case PIOCSFL:
 287           /*
 288            * NFLAGS is "non-suser_xxx flags" -- currently, only
 289            * PFS_ISUGID ("ignore set u/g id");
 290            */
 291 #define NFLAGS  (PF_ISUGID)
 292           flags = (unsigned char)*(unsigned int*)ap->a_data;
 293           if (flags & NFLAGS && (error = suser_cred(ap->a_cred, 0)))
 294             return error;
 295           procp->p_pfsflags = flags;
 296           break;
 297         case PIOCGFL:
 298           *(unsigned int*)ap->a_data = (unsigned int)procp->p_pfsflags;
 299           break;
 300         case PIOCSTATUS:
 301           psp = (struct procfs_status *)ap->a_data;
 302           psp->state = (procp->p_step == 0);
 303           psp->flags = procp->p_pfsflags;
 304           psp->events = procp->p_stops;
 305           if (procp->p_step) {
 306             psp->why = procp->p_stype;
 307             psp->val = procp->p_xstat;
 308           } else {
 309             psp->why = psp->val = 0;    /* Not defined values */
 310           }
 311           break;
 312         case PIOCWAIT:
 313           psp = (struct procfs_status *)ap->a_data;
 314           if (procp->p_step == 0) {
 315             error = tsleep(&procp->p_stype, PCATCH, "piocwait", 0);
 316             if (error)
 317               return error;
 318           }
 319           psp->state = 1;       /* It stopped */
 320           psp->flags = procp->p_pfsflags;
 321           psp->events = procp->p_stops;
 322           psp->why = procp->p_stype;    /* why it stopped */
 323           psp->val = procp->p_xstat;    /* any extra info */
 324           break;
 325         case PIOCCONT:  /* Restart a proc */
 326           if (procp->p_step == 0)
 327             return EINVAL;      /* Can only start a stopped process */
 328           if ((signo = *(int*)ap->a_data) != 0) {
 329             if (signo >= NSIG || signo <= 0)
 330               return EINVAL;
 331             ksignal(procp, signo);
 332           }
 333           procp->p_step = 0;
 334           wakeup(&procp->p_step);
 335           break;
 336         default:
 337           return (ENOTTY);
 338         }
 339         return 0;
 340 }
 341
 342 /*
 343  * do block mapping for pfsnode (vp).
 344  * since we don't use the buffer cache
 345  * for procfs this function should never
 346  * be called.  in any case, it's not clear
 347  * what part of the kernel ever makes use
 348  * of this function.  for sanity, this is the
 349  * usual no-op bmap, although returning
 350  * (EIO) would be a reasonable alternative.
 351  *
 352  * XXX mmap assumes buffer cache operation
 353  *
 354  * procfs_bmap(struct vnode *a_vp, off_t a_loffset,
 355  *              off_t *a_doffsetp, int *a_runp, int *a_runb)
 356  */
 357 static int
 358 procfs_bmap(struct vop_bmap_args *ap)
 359 {
 360         if (ap->a_doffsetp != NULL)
 361                 *ap->a_doffsetp = ap->a_loffset;
 362         if (ap->a_runp != NULL)
 363                 *ap->a_runp = 0;
 364         if (ap->a_runb != NULL)
 365                 *ap->a_runb = 0;
 366         return (0);
 367 }
 368
 369 /*
 370  * procfs_inactive is called when the pfsnode
 371  * is vrele'd and the reference count goes
 372  * to zero.  (vp) will be on the vnode free
 373  * list, so to get it back vget() must be
 374  * used.
 375  *
 376  * (vp) is locked on entry, but must be unlocked on exit.
 377  *
 378  * procfs_inactive(struct vnode *a_vp)
 379  */
 380 static int
 381 procfs_inactive(struct vop_inactive_args *ap)
 382 {
 383         /*struct vnode *vp = ap->a_vp;*/
 384
 385         return (0);
 386 }
 387
 388 /*
 389  * _reclaim is called when getnewvnode()
 390  * wants to make use of an entry on the vnode
 391  * free list.  at this time the filesystem needs
 392  * to free any private data and remove the node
 393  * from any private lists.
 394  *
 395  * procfs_reclaim(struct vnode *a_vp)
 396  */
 397 static int
 398 procfs_reclaim(struct vop_reclaim_args *ap)
 399 {
 400         return (procfs_freevp(ap->a_vp));
 401 }
 402
 403 /*
 404  * _print is used for debugging.
 405  * just print a readable description
 406  * of (vp).
 407  *
 408  * procfs_print(struct vnode *a_vp)
 409  */
 410 static int
 411 procfs_print(struct vop_print_args *ap)
 412 {
 413         struct pfsnode *pfs = VTOPFS(ap->a_vp);
 414
 415         kprintf("tag VT_PROCFS, type %d, pid %ld, mode %x, flags %lx\n",
 416             pfs->pfs_type, (long)pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
 417         return (0);
 418 }
 419
 420 /*
 421  * generic entry point for unsupported operations
 422  */
 423 static int
 424 procfs_badop(struct vop_generic_args *ap)
 425 {
 426         return (EIO);
 427 }
 428
 429 /*
 430  * Invent attributes for pfsnode (vp) and store
 431  * them in (vap).
 432  * Directories lengths are returned as zero since
 433  * any real length would require the genuine size
 434  * to be computed, and nothing cares anyway.
 435  *
 436  * this is relatively minimal for procfs.
 437  *
 438  * procfs_getattr(struct vnode *a_vp, struct vattr *a_vap)
 439  */
 440 static int
 441 procfs_getattr(struct vop_getattr_args *ap)
 442 {
 443         struct pfsnode *pfs = VTOPFS(ap->a_vp);
 444         struct vattr *vap = ap->a_vap;
 445         struct proc *procp;
 446         int error;
 447
 448         /*
 449          * First make sure that the process and its credentials
 450          * still exist.
 451          */
 452         switch (pfs->pfs_type) {
 453         case Proot:
 454         case Pcurproc:
 455                 procp = 0;
 456                 break;
 457
 458         default:
 459                 procp = PFIND(pfs->pfs_pid);
 460                 if (procp == NULL || procp->p_ucred == NULL)
 461                         return (ENOENT);
 462         }
 463
 464         error = 0;
 465
 466         /* start by zeroing out the attributes */
 467         VATTR_NULL(vap);
 468
 469         /* next do all the common fields */
 470         vap->va_type = ap->a_vp->v_type;
 471         vap->va_mode = pfs->pfs_mode;
 472         vap->va_fileid = pfs->pfs_fileno;
 473         vap->va_flags = 0;
 474         vap->va_blocksize = PAGE_SIZE;
 475         vap->va_bytes = vap->va_size = 0;
 476         vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
 477
 478         /*
 479          * Make all times be current TOD.
 480          * It would be possible to get the process start
 481          * time from the p_stat structure, but there's
 482          * no "file creation" time stamp anyway, and the
 483          * p_stat structure is not addressible if u. gets
 484          * swapped out for that process.
 485          */
 486         nanotime(&vap->va_ctime);
 487         vap->va_atime = vap->va_mtime = vap->va_ctime;
 488
 489         /*
 490          * If the process has exercised some setuid or setgid
 491          * privilege, then rip away read/write permission so
 492          * that only root can gain access.
 493          */
 494         switch (pfs->pfs_type) {
 495         case Pctl:
 496         case Pregs:
 497         case Pfpregs:
 498         case Pdbregs:
 499         case Pmem:
 500                 if (procp->p_flag & P_SUGID)
 501                         vap->va_mode &= ~((VREAD|VWRITE)|
 502                                           ((VREAD|VWRITE)>>3)|
 503                                           ((VREAD|VWRITE)>>6));
 504                 break;
 505         default:
 506                 break;
 507         }
 508
 509         /*
 510          * now do the object specific fields
 511          *
 512          * The size could be set from struct reg, but it's hardly
 513          * worth the trouble, and it puts some (potentially) machine
 514          * dependent data into this machine-independent code.  If it
 515          * becomes important then this function should break out into
 516          * a per-file stat function in the corresponding .c file.
 517          */
 518
 519         vap->va_nlink = 1;
 520         if (procp) {
 521                 vap->va_uid = procp->p_ucred->cr_uid;
 522                 vap->va_gid = procp->p_ucred->cr_gid;
 523         }
 524
 525         switch (pfs->pfs_type) {
 526         case Proot:
 527                 /*
 528                  * Set nlink to 1 to tell fts(3) we don't actually know.
 529                  */
 530                 vap->va_nlink = 1;
 531                 vap->va_uid = 0;
 532                 vap->va_gid = 0;
 533                 vap->va_size = vap->va_bytes = DEV_BSIZE;
 534                 break;
 535
 536         case Pcurproc: {
 537                 char buf[16];           /* should be enough */
 538                 vap->va_uid = 0;
 539                 vap->va_gid = 0;
 540                 vap->va_size = vap->va_bytes =
 541                     ksnprintf(buf, sizeof(buf), "%ld", (long)curproc->p_pid);
 542                 break;
 543         }
 544
 545         case Pproc:
 546                 vap->va_nlink = nproc_targets;
 547                 vap->va_size = vap->va_bytes = DEV_BSIZE;
 548                 break;
 549
 550         case Pfile: {
 551                 char *fullpath, *freepath;
 552                 error = vn_fullpath(procp, NULL, &fullpath, &freepath);
 553                 if (error == 0) {
 554                         vap->va_size = strlen(fullpath);
 555                         kfree(freepath, M_TEMP);
 556                 } else {
 557                         vap->va_size = sizeof("unknown") - 1;
 558                         error = 0;
 559                 }
 560                 vap->va_bytes = vap->va_size;
 561                 break;
 562         }
 563
 564         case Pmem:
 565                 /*
 566                  * If we denied owner access earlier, then we have to
 567                  * change the owner to root - otherwise 'ps' and friends
 568                  * will break even though they are setgid kmem. *SIGH*
 569                  */
 570                 if (procp->p_flag & P_SUGID)
 571                         vap->va_uid = 0;
 572                 else
 573                         vap->va_uid = procp->p_ucred->cr_uid;
 574                 break;
 575
 576         case Pregs:
 577                 vap->va_bytes = vap->va_size = sizeof(struct reg);
 578                 break;
 579
 580         case Pfpregs:
 581                 vap->va_bytes = vap->va_size = sizeof(struct fpreg);
 582                 break;
 583
 584         case Pdbregs:
 585                 vap->va_bytes = vap->va_size = sizeof(struct dbreg);
 586                 break;
 587
 588         case Ptype:
 589         case Pmap:
 590         case Pctl:
 591         case Pstatus:
 592         case Pnote:
 593         case Pnotepg:
 594         case Pcmdline:
 595         case Prlimit:
 596                 break;
 597
 598         default:
 599                 panic("procfs_getattr");
 600         }
 601
 602         return (error);
 603 }
 604
 605 /*
 606  * procfs_setattr(struct vnode *a_vp, struct vattr *a_vap,
 607  *                struct ucred *a_cred)
 608  */
 609 static int
 610 procfs_setattr(struct vop_setattr_args *ap)
 611 {
 612         if (ap->a_vap->va_flags != VNOVAL)
 613                 return (EOPNOTSUPP);
 614
 615         /*
 616          * just fake out attribute setting
 617          * it's not good to generate an error
 618          * return, otherwise things like creat()
 619          * will fail when they try to set the
 620          * file length to 0.  worse, this means
 621          * that echo $note > /proc/$pid/note will fail.
 622          */
 623
 624         return (0);
 625 }
 626
 627 /*
 628  * implement access checking.
 629  *
 630  * something very similar to this code is duplicated
 631  * throughout the 4bsd kernel and should be moved
 632  * into kern/vfs_subr.c sometime.
 633  *
 634  * actually, the check for super-user is slightly
 635  * broken since it will allow read access to write-only
 636  * objects.  this doesn't cause any particular trouble
 637  * but does mean that the i/o entry points need to check
 638  * that the operation really does make sense.
 639  *
 640  * procfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
 641  */
 642 static int
 643 procfs_access(struct vop_access_args *ap)
 644 {
 645         struct vattr *vap;
 646         struct vattr vattr;
 647         int error;
 648
 649         /*
 650          * If you're the super-user,
 651          * you always get access.
 652          */
 653         if (ap->a_cred->cr_uid == 0)
 654                 return (0);
 655
 656         vap = &vattr;
 657         error = VOP_GETATTR(ap->a_vp, vap);
 658         if (error)
 659                 return (error);
 660
 661         /*
 662          * Access check is based on only one of owner, group, public.
 663          * If not owner, then check group. If not a member of the
 664          * group, then check public access.
 665          */
 666         if (ap->a_cred->cr_uid != vap->va_uid) {
 667                 gid_t *gp;
 668                 int i;
 669
 670                 ap->a_mode >>= 3;
 671                 gp = ap->a_cred->cr_groups;
 672                 for (i = 0; i < ap->a_cred->cr_ngroups; i++, gp++)
 673                         if (vap->va_gid == *gp)
 674                                 goto found;
 675                 ap->a_mode >>= 3;
 676 found:
 677                 ;
 678         }
 679
 680         if ((vap->va_mode & ap->a_mode) == ap->a_mode)
 681                 return (0);
 682
 683         return (EACCES);
 684 }
 685
 686 /*
 687  * lookup.  this is incredibly complicated in the general case, however
 688  * for most pseudo-filesystems very little needs to be done.
 689  *
 690  * procfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
 691  *               struct componentname *a_cnp)
 692  */
 693 static int
 694 procfs_lookup(struct vop_old_lookup_args *ap)
 695 {
 696         struct componentname *cnp = ap->a_cnp;
 697         struct vnode **vpp = ap->a_vpp;
 698         struct vnode *dvp = ap->a_dvp;
 699         char *pname = cnp->cn_nameptr;
 700         /* struct proc *curp = cnp->cn_proc; */
 701         struct proc_target *pt;
 702         pid_t pid;
 703         struct pfsnode *pfs;
 704         struct proc *p;
 705         struct lwp *lp;
 706         int i;
 707         int error;
 708
 709         *vpp = NULL;
 710
 711         if (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)
 712                 return (EROFS);
 713
 714         error = 0;
 715         if (cnp->cn_namelen == 1 && *pname == '.') {
 716                 *vpp = dvp;
 717                 vref(*vpp);
 718                 goto out;
 719         }
 720
 721         pfs = VTOPFS(dvp);
 722         switch (pfs->pfs_type) {
 723         case Proot:
 724                 if (cnp->cn_flags & CNP_ISDOTDOT)
 725                         return (EIO);
 726
 727                 if (CNEQ(cnp, "curproc", 7)) {
 728                         error = procfs_allocvp(dvp->v_mount, vpp, 0, Pcurproc);
 729                         goto out;
 730                 }
 731
 732                 pid = atopid(pname, cnp->cn_namelen);
 733                 if (pid == NO_PID)
 734                         break;
 735
 736                 p = PFIND(pid);
 737                 if (p == NULL)
 738                         break;
 739
 740                 if (!PRISON_CHECK(ap->a_cnp->cn_cred, p->p_ucred))
 741                         break;
 742
 743                 if (ps_showallprocs == 0 && ap->a_cnp->cn_cred->cr_uid != 0 &&
 744                     ap->a_cnp->cn_cred->cr_uid != p->p_ucred->cr_uid)
 745                         break;
 746
 747                 error = procfs_allocvp(dvp->v_mount, vpp, pid, Pproc);
 748                 goto out;
 749
 750         case Pproc:
 751                 if (cnp->cn_flags & CNP_ISDOTDOT) {
 752                         error = procfs_root(dvp->v_mount, vpp);
 753                         goto out;
 754                 }
 755
 756                 p = PFIND(pfs->pfs_pid);
 757                 if (p == NULL)
 758                         break;
 759                 /* XXX lwp */
 760                 lp = FIRST_LWP_IN_PROC(p);
 761
 762                 if (!PRISON_CHECK(ap->a_cnp->cn_cred, p->p_ucred))
 763                         break;
 764
 765                 if (ps_showallprocs == 0 && ap->a_cnp->cn_cred->cr_uid != 0 &&
 766                     ap->a_cnp->cn_cred->cr_uid != p->p_ucred->cr_uid)
 767                         break;
 768
 769                 for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
 770                         if (cnp->cn_namelen == pt->pt_namlen &&
 771                             bcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
 772                             (pt->pt_valid == NULL || (*pt->pt_valid)(lp)))
 773                                 goto found;
 774                 }
 775                 break;
 776         found:
 777                 error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
 778                                         pt->pt_pfstype);
 779                 goto out;
 780
 781         default:
 782                 error = ENOTDIR;
 783                 goto out;
 784         }
 785         if (cnp->cn_nameiop == NAMEI_LOOKUP)
 786                 error = ENOENT;
 787         else
 788                 error = EROFS;
 789         /*
 790          * If no error occured *vpp will hold a referenced locked vnode.
 791          * dvp was passed to us locked and *vpp must be returned locked.
 792          * If *vpp != dvp then we should unlock dvp if (1) this is not the
 793          * last component or (2) CNP_LOCKPARENT is not set.
 794          */
 795 out:
 796         if (error == 0 && *vpp != dvp) {
 797                 if ((cnp->cn_flags & CNP_LOCKPARENT) == 0) {
 798                         cnp->cn_flags |= CNP_PDIRUNLOCK;
 799                         vn_unlock(dvp);
 800                 }
 801         }
 802         return (error);
 803 }
 804
 805 /*
 806  * Does this process have a text file?
 807  */
 808 int
 809 procfs_validfile(struct lwp *lp)
 810 {
 811         return (procfs_findtextvp(lp->lwp_proc) != NULLVP);
 812 }
 813
 814 /*
 815  * readdir() returns directory entries from pfsnode (vp).
 816  *
 817  * We generate just one directory entry at a time, as it would probably
 818  * not pay off to buffer several entries locally to save uiomove calls.
 819  *
 820  * procfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred,
 821  *                int *a_eofflag, int *a_ncookies, off_t **a_cookies)
 822  */
 823 static int
 824 procfs_readdir(struct vop_readdir_args *ap)
 825 {
 826         struct pfsnode *pfs;
 827         int error;
 828
 829         if (ap->a_uio->uio_offset < 0 || ap->a_uio->uio_offset > INT_MAX)
 830                 return (EINVAL);
 831         if ((error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
 832                 return (error);
 833         pfs = VTOPFS(ap->a_vp);
 834
 835         switch (pfs->pfs_type) {
 836         case Pproc:
 837                 /*
 838                  * this is for the process-specific sub-directories.
 839                  * all that is needed to is copy out all the entries
 840                  * from the procent[] table (top of this file).
 841                  */
 842                 error = procfs_readdir_proc(ap);
 843                 break;
 844         case Proot:
 845                 /*
 846                  * this is for the root of the procfs filesystem
 847                  * what is needed is a special entry for "curproc"
 848                  * followed by an entry for each process on allproc
 849                  */
 850                 error = procfs_readdir_root(ap);
 851                 break;
 852         default:
 853                 error = ENOTDIR;
 854                 break;
 855         }
 856
 857         vn_unlock(ap->a_vp);
 858         return (error);
 859 }
 860
 861 static int
 862 procfs_readdir_proc(struct vop_readdir_args *ap)
 863 {
 864         struct pfsnode *pfs;
 865         int error, i, retval;
 866         struct proc *p;
 867         struct lwp *lp;
 868         struct proc_target *pt;
 869         struct uio *uio = ap->a_uio;
 870
 871         pfs = VTOPFS(ap->a_vp);
 872         p = PFIND(pfs->pfs_pid);
 873         if (p == NULL)
 874                 return(0);
 875         if (!PRISON_CHECK(ap->a_cred, p->p_ucred))
 876                 return(0);
 877         /* XXX lwp */
 878         lp = FIRST_LWP_IN_PROC(p);
 879
 880         error = 0;
 881         i = (int)uio->uio_offset;
 882         if (i < 0)
 883                 return (EINVAL);
 884
 885         for (pt = &proc_targets[i];
 886              !error && uio->uio_resid > 0 && i < nproc_targets; pt++, i++) {
 887                 if (pt->pt_valid && (*pt->pt_valid)(lp) == 0)
 888                         continue;
 889
 890                 retval = vop_write_dirent(&error, uio,
 891                     PROCFS_FILENO(pfs->pfs_pid, pt->pt_pfstype), pt->pt_type,
 892                     pt->pt_namlen, pt->pt_name);
 893                 if (retval)
 894                         break;
 895         }
 896
 897         uio->uio_offset = (off_t)i;
 898
 899         return(0);
 900 }
 901
 902 struct procfs_readdir_root_info {
 903         int error;
 904         int i;
 905         int pcnt;
 906         struct uio *uio;
 907         struct ucred *cred;
 908 };
 909
 910 static int procfs_readdir_root_callback(struct proc *p, void *data);
 911
 912 static int
 913 procfs_readdir_root(struct vop_readdir_args *ap)
 914 {
 915         struct procfs_readdir_root_info info;
 916         struct uio *uio = ap->a_uio;
 917         int res;
 918
 919         info.error = 0;
 920         info.i = (int)uio->uio_offset;
 921
 922         if (info.i < 0)
 923                 return (EINVAL);
 924
 925         info.pcnt = 0;
 926         info.uio = uio;
 927         info.cred = ap->a_cred;
 928         while (info.pcnt < 3) {
 929                 res = procfs_readdir_root_callback(NULL, &info);
 930                 if (res < 0)
 931                         break;
 932         }
 933         if (res >= 0)
 934                 allproc_scan(procfs_readdir_root_callback, &info);
 935         uio->uio_offset = (off_t)info.i;
 936
 937         return (info.error);
 938 }
 939
 940 static int
 941 procfs_readdir_root_callback(struct proc *p, void *data)
 942 {
 943         struct procfs_readdir_root_info *info = data;
 944         struct uio *uio;
 945         int retval;
 946         ino_t d_ino;
 947         const char *d_name;
 948         char d_name_pid[20];
 949         size_t d_namlen;
 950         uint8_t d_type;
 951
 952         uio = info->uio;
 953
 954         if (uio->uio_resid <= 0 || info->error)
 955                 return(-1);
 956
 957         switch (info->pcnt) {
 958         case 0:         /* `.' */
 959                 d_ino = PROCFS_FILENO(0, Proot);
 960                 d_name = ".";
 961                 d_namlen = 1;
 962                 d_type = DT_DIR;
 963                 break;
 964         case 1:         /* `..' */
 965                 d_ino = PROCFS_FILENO(0, Proot);
 966                 d_name = "..";
 967                 d_namlen = 2;
 968                 d_type = DT_DIR;
 969                 break;
 970
 971         case 2:
 972                 d_ino = PROCFS_FILENO(0, Pcurproc);
 973                 d_namlen = 7;
 974                 d_name = "curproc";
 975                 d_type = DT_LNK;
 976                 break;
 977
 978
 979         default:
 980                 if (!PRISON_CHECK(info->cred, p->p_ucred))
 981                         return(0);
 982                 if (ps_showallprocs == 0 &&
 983                     info->cred->cr_uid != 0 &&
 984                     info->cred->cr_uid != p->p_ucred->cr_uid) {
 985                         return(0);
 986                 }
 987
 988                 /*
 989                  * Skip entries we have already returned (optimization)
 990                  */
 991                 if (info->pcnt < info->i) {
 992                         ++info->pcnt;
 993                         return(0);
 994                 }
 995
 996                 d_ino = PROCFS_FILENO(p->p_pid, Pproc);
 997                 d_namlen = ksnprintf(d_name_pid, sizeof(d_name_pid),
 998                     "%ld", (long)p->p_pid);
 999                 d_name = d_name_pid;
1000                 d_type = DT_DIR;
1001                 break;
1002         }
1003
1004         /*
1005          * Skip entries we have already returned (optimization)
1006          */
1007         if (info->pcnt < info->i) {
1008                 ++info->pcnt;
1009                 return(0);
1010         }
1011
1012         retval = vop_write_dirent(&info->error, uio,
1013                                   d_ino, d_type, d_namlen, d_name);
1014         if (retval)
1015                 return(-1);
1016         ++info->pcnt;
1017         ++info->i;
1018         return(0);
1019 }
1020
1021 /*
1022  * readlink reads the link of `curproc' or `file'
1023  */
1024 static int
1025 procfs_readlink(struct vop_readlink_args *ap)
1026 {
1027         char buf[16];           /* should be enough */
1028         struct proc *procp;
1029         struct vnode *vp = ap->a_vp;
1030         struct pfsnode *pfs = VTOPFS(vp);
1031         char *fullpath, *freepath;
1032         int error, len;
1033
1034         switch (pfs->pfs_type) {
1035         case Pcurproc:
1036                 if (pfs->pfs_fileno != PROCFS_FILENO(0, Pcurproc))
1037                         return (EINVAL);
1038
1039                 len = ksnprintf(buf, sizeof(buf), "%ld", (long)curproc->p_pid);
1040
1041                 return (uiomove(buf, len, ap->a_uio));
1042         /*
1043          * There _should_ be no way for an entire process to disappear
1044          * from under us...
1045          */
1046         case Pfile:
1047                 procp = PFIND(pfs->pfs_pid);
1048                 if (procp == NULL || procp->p_ucred == NULL) {
1049                         kprintf("procfs_readlink: pid %d disappeared\n",
1050                             pfs->pfs_pid);
1051                         return (uiomove("unknown", sizeof("unknown") - 1,
1052                             ap->a_uio));
1053                 }
1054                 error = vn_fullpath(procp, NULL, &fullpath, &freepath);
1055                 if (error != 0)
1056                         return (uiomove("unknown", sizeof("unknown") - 1,
1057                             ap->a_uio));
1058                 error = uiomove(fullpath, strlen(fullpath), ap->a_uio);
1059                 kfree(freepath, M_TEMP);
1060                 return (error);
1061         default:
1062                 return (EINVAL);
1063         }
1064 }
1065
1066 /*
1067  * convert decimal ascii to pid_t
1068  */
1069 static pid_t
1070 atopid(const char *b, u_int len)
1071 {
1072         pid_t p = 0;
1073
1074         while (len--) {
1075                 char c = *b++;
1076                 if (c < '0' || c > '9')
1077                         return (NO_PID);
1078                 p = 10 * p + (c - '0');
1079                 if (p > PID_MAX)
1080                         return (NO_PID);
1081         }
1082
1083         return (p);
1084 }
1085