kernel/os/acct.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 1992, 2010, Oracle and/or its affiliates. All rights reserved.
  23  */
  24
  25 /*      Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */
  26 /*        All Rights Reserved   */
  27
  28
  29 #include <sys/types.h>
  30 #include <sys/sysmacros.h>
  31 #include <sys/param.h>
  32 #include <sys/systm.h>
  33 #include <sys/acct.h>
  34 #include <sys/cred.h>
  35 #include <sys/user.h>
  36 #include <sys/errno.h>
  37 #include <sys/file.h>
  38 #include <sys/vnode.h>
  39 #include <sys/debug.h>
  40 #include <sys/proc.h>
  41 #include <sys/resource.h>
  42 #include <sys/session.h>
  43 #include <sys/modctl.h>
  44 #include <sys/syscall.h>
  45 #include <sys/policy.h>
  46 #include <sys/list.h>
  47 #include <sys/time.h>
  48 #include <sys/msacct.h>
  49 #include <sys/zone.h>
  50
  51 /*
  52  * Each zone has its own accounting settings (on or off) and associated
  53  * file.  The global zone is not special in this aspect; it will only
  54  * generate records for processes that ran in the global zone.  We could
  55  * allow the global zone to record all activity on the system, but there
  56  * would be no way of knowing the zone in which the processes executed.
  57  * sysacct() is thus virtualized to only act on the caller's zone.
  58  */
  59 struct acct_globals {
  60         struct acct     acctbuf;
  61         kmutex_t        aclock;
  62         struct vnode    *acctvp;
  63         list_node_t     aclink;
  64 };
  65
  66 /*
  67  * We need a list of all accounting settings for all zones, so we can
  68  * accurately determine if a file is in use for accounting (possibly by
  69  * another zone).
  70  */
  71 static zone_key_t acct_zone_key;
  72 static list_t acct_list;
  73 kmutex_t acct_list_lock;
  74
  75 static struct sysent acctsysent = {
  76         1,
  77         SE_NOUNLOAD | SE_ARGC | SE_32RVAL1,
  78         sysacct
  79 };
  80
  81 static struct modlsys modlsys = {
  82         &mod_syscallops, "acct(2) syscall", &acctsysent
  83 };
  84
  85 #ifdef _SYSCALL32_IMPL
  86 static struct modlsys modlsys32 = {
  87         &mod_syscallops32, "32-bit acct(2) syscall", &acctsysent
  88 };
  89 #endif
  90
  91 static struct modlinkage modlinkage = {
  92         MODREV_1,
  93         &modlsys,
  94 #ifdef _SYSCALL32_IMPL
  95         &modlsys32,
  96 #endif
  97         NULL
  98 };
  99
 100 /*ARGSUSED*/
 101 static void *
 102 acct_init(zoneid_t zoneid)
 103 {
 104         struct acct_globals *ag;
 105
 106         ag = kmem_alloc(sizeof (*ag), KM_SLEEP);
 107         bzero(&ag->acctbuf, sizeof (ag->acctbuf));
 108         mutex_init(&ag->aclock, NULL, MUTEX_DEFAULT, NULL);
 109         ag->acctvp = NULL;
 110
 111         mutex_enter(&acct_list_lock);
 112         list_insert_tail(&acct_list, ag);
 113         mutex_exit(&acct_list_lock);
 114         return (ag);
 115 }
 116
 117 /* ARGSUSED */
 118 static void
 119 acct_shutdown(zoneid_t zoneid, void *arg)
 120 {
 121         struct acct_globals *ag = arg;
 122
 123         mutex_enter(&ag->aclock);
 124         if (ag->acctvp) {
 125                 /*
 126                  * This needs to be done as a shutdown callback, otherwise this
 127                  * held vnode may cause filesystems to be busy, and the zone
 128                  * shutdown operation to fail.
 129                  */
 130                 (void) fop_close(ag->acctvp, FWRITE, 1, (offset_t)0, kcred,
 131                     NULL);
 132                 VN_RELE(ag->acctvp);
 133         }
 134         ag->acctvp = NULL;
 135         mutex_exit(&ag->aclock);
 136 }
 137
 138 /*ARGSUSED*/
 139 static void
 140 acct_fini(zoneid_t zoneid, void *arg)
 141 {
 142         struct acct_globals *ag = arg;
 143
 144         mutex_enter(&acct_list_lock);
 145         list_remove(&acct_list, ag);
 146         mutex_exit(&acct_list_lock);
 147
 148         mutex_destroy(&ag->aclock);
 149         kmem_free(ag, sizeof (*ag));
 150 }
 151
 152 int
 153 _init(void)
 154 {
 155         int error;
 156
 157         mutex_init(&acct_list_lock, NULL, MUTEX_DEFAULT, NULL);
 158         list_create(&acct_list, sizeof (struct acct_globals),
 159             offsetof(struct acct_globals, aclink));
 160         /*
 161          * Using an initializer here wastes a bit of memory for zones that
 162          * don't use accounting, but vastly simplifies the locking.
 163          */
 164         zone_key_create(&acct_zone_key, acct_init, acct_shutdown, acct_fini);
 165         if ((error = mod_install(&modlinkage)) != 0) {
 166                 (void) zone_key_delete(acct_zone_key);
 167                 list_destroy(&acct_list);
 168                 mutex_destroy(&acct_list_lock);
 169         }
 170         return (error);
 171 }
 172
 173 int
 174 _info(struct modinfo *modinfop)
 175 {
 176         return (mod_info(&modlinkage, modinfop));
 177 }
 178
 179 /*
 180  * acct() is a "weak stub" routine called from exit().
 181  * Once this module has been loaded, we refuse to allow
 182  * it to unload - otherwise accounting would quietly
 183  * cease.  See 1211661.  It's possible to make this module
 184  * unloadable but it's substantially safer not to bother.
 185  */
 186 int
 187 _fini(void)
 188 {
 189         return (EBUSY);
 190 }
 191
 192 /*
 193  * See if vp is in use by the accounting system on any zone.  This does a deep
 194  * comparison of vnodes such that a file and a lofs "shadow" node of it will
 195  * appear to be the same.
 196  *
 197  * If 'compare_vfs' is true, the function will do a comparison of vfs_t's
 198  * instead (ie, is the vfs_t on which the vnode resides in use by the
 199  * accounting system in any zone).
 200  *
 201  * Returns 1 if found (in use), 0 otherwise.
 202  */
 203 static int
 204 acct_find(vnode_t *vp, boolean_t compare_vfs)
 205 {
 206         struct acct_globals *ag;
 207         vnode_t *realvp;
 208
 209         ASSERT(MUTEX_HELD(&acct_list_lock));
 210         ASSERT(vp != NULL);
 211
 212         if (fop_realvp(vp, &realvp, NULL))
 213                 realvp = vp;
 214         for (ag = list_head(&acct_list); ag != NULL;
 215             ag = list_next(&acct_list, ag)) {
 216                 vnode_t *racctvp;
 217                 boolean_t found = B_FALSE;
 218
 219                 mutex_enter(&ag->aclock);
 220                 if (ag->acctvp == NULL) {
 221                         mutex_exit(&ag->aclock);
 222                         continue;
 223                 }
 224                 if (fop_realvp(ag->acctvp, &racctvp, NULL))
 225                         racctvp = ag->acctvp;
 226                 if (compare_vfs) {
 227                         if (racctvp->v_vfsp == realvp->v_vfsp)
 228                                 found = B_TRUE;
 229                 } else {
 230                         if (VN_CMP(realvp, racctvp))
 231                                 found = B_TRUE;
 232                 }
 233                 mutex_exit(&ag->aclock);
 234                 if (found)
 235                         return (1);
 236         }
 237         return (0);
 238 }
 239
 240 /*
 241  * Returns 1 if the vfs that vnode resides on is in use for the accounting
 242  * subsystem, 0 otherwise.
 243  */
 244 int
 245 acct_fs_in_use(vnode_t *vp)
 246 {
 247         int found;
 248
 249         if (vp == NULL)
 250                 return (0);
 251         mutex_enter(&acct_list_lock);
 252         found = acct_find(vp, B_TRUE);
 253         mutex_exit(&acct_list_lock);
 254         return (found);
 255 }
 256
 257 /*
 258  * Perform process accounting functions.
 259  */
 260 int
 261 sysacct(char *fname)
 262 {
 263         struct acct_globals *ag;
 264         struct vnode *vp;
 265         int error = 0;
 266
 267         if (secpolicy_acct(CRED()) != 0)
 268                 return (set_errno(EPERM));
 269
 270         ag = zone_getspecific(acct_zone_key, curproc->p_zone);
 271         ASSERT(ag != NULL);
 272
 273         if (fname == NULL) {
 274                 /*
 275                  * Close the file and stop accounting.
 276                  */
 277                 mutex_enter(&ag->aclock);
 278                 vp = ag->acctvp;
 279                 ag->acctvp = NULL;
 280                 mutex_exit(&ag->aclock);
 281                 if (vp) {
 282                         error = fop_close(vp, FWRITE, 1, (offset_t)0, CRED(),
 283                             NULL);
 284                         VN_RELE(vp);
 285                 }
 286                 return (error == 0 ? 0 : set_errno(error));
 287         }
 288
 289         /*
 290          * Either (a) open a new file and begin accounting -or- (b)
 291          * switch accounting from an old to a new file.
 292          *
 293          * (Open the file without holding aclock in case it
 294          * sleeps (holding the lock prevents process exit).)
 295          */
 296         if ((error = vn_open(fname, UIO_USERSPACE, FWRITE,
 297             0, &vp, (enum create)0, 0)) != 0) {
 298                 /* SVID  compliance */
 299                 if (error == EISDIR)
 300                         error = EACCES;
 301                 return (set_errno(error));
 302         }
 303
 304         if (vp->v_type != VREG) {
 305                 error = EACCES;
 306         } else {
 307                 mutex_enter(&acct_list_lock);
 308                 if (acct_find(vp, B_FALSE)) {
 309                         error = EBUSY;
 310                 } else {
 311                         mutex_enter(&ag->aclock);
 312                         if (ag->acctvp) {
 313                                 vnode_t *oldvp;
 314
 315                                 /*
 316                                  * close old acctvp, and point acct()
 317                                  * at new file by swapping vp and acctvp
 318                                  */
 319                                 oldvp = ag->acctvp;
 320                                 ag->acctvp = vp;
 321                                 vp = oldvp;
 322                         } else {
 323                                 /*
 324                                  * no existing file, start accounting ..
 325                                  */
 326                                 ag->acctvp = vp;
 327                                 vp = NULL;
 328                         }
 329                         mutex_exit(&ag->aclock);
 330                 }
 331                 mutex_exit(&acct_list_lock);
 332         }
 333
 334         if (vp) {
 335                 (void) fop_close(vp, FWRITE, 1, (offset_t)0, CRED(), NULL);
 336                 VN_RELE(vp);
 337         }
 338         return (error == 0 ? 0 : set_errno(error));
 339 }
 340
 341 /*
 342  * Produce a pseudo-floating point representation
 343  * with 3 bits base-8 exponent, 13 bits fraction.
 344  */
 345 static comp_t
 346 acct_compress(ulong_t t)
 347 {
 348         int exp = 0, round = 0;
 349
 350         while (t >= 8192) {
 351                 exp++;
 352                 round = t & 04;
 353                 t >>= 3;
 354         }
 355         if (round) {
 356                 t++;
 357                 if (t >= 8192) {
 358                         t >>= 3;
 359                         exp++;
 360                 }
 361         }
 362 #ifdef _LP64
 363         if (exp > 7) {
 364                 /* prevent wraparound */
 365                 t = 8191;
 366                 exp = 7;
 367         }
 368 #endif
 369         return ((exp << 13) + t);
 370 }
 371
 372 /*
 373  * On exit, write a record on the accounting file.
 374  */
 375 void
 376 acct(char st)
 377 {
 378         struct vnode *vp;
 379         struct cred *cr;
 380         struct proc *p;
 381         user_t *ua;
 382         struct vattr va;
 383         ssize_t resid = 0;
 384         int error;
 385         struct acct_globals *ag;
 386
 387         /*
 388          * If sysacct module is loaded when zone is in down state then
 389          * the following function can return NULL.
 390          */
 391         ag = zone_getspecific(acct_zone_key, curproc->p_zone);
 392         if (ag == NULL)
 393                 return;
 394
 395         mutex_enter(&ag->aclock);
 396         if ((vp = ag->acctvp) == NULL) {
 397                 mutex_exit(&ag->aclock);
 398                 return;
 399         }
 400
 401         /*
 402          * This only gets called from exit after all lwp's have exited so no
 403          * cred locking is needed.
 404          */
 405         p = curproc;
 406         ua = PTOU(p);
 407         bcopy(ua->u_comm, ag->acctbuf.ac_comm, sizeof (ag->acctbuf.ac_comm));
 408         ag->acctbuf.ac_btime = ua->u_start.tv_sec;
 409         ag->acctbuf.ac_utime = acct_compress(NSEC_TO_TICK(p->p_acct[LMS_USER]));
 410         ag->acctbuf.ac_stime = acct_compress(
 411             NSEC_TO_TICK(p->p_acct[LMS_SYSTEM] + p->p_acct[LMS_TRAP]));
 412         ag->acctbuf.ac_etime = acct_compress(ddi_get_lbolt() - ua->u_ticks);
 413         ag->acctbuf.ac_mem = acct_compress((ulong_t)ua->u_mem);
 414         ag->acctbuf.ac_io = acct_compress((ulong_t)p->p_ru.ioch);
 415         ag->acctbuf.ac_rw = acct_compress((ulong_t)(p->p_ru.inblock +
 416             p->p_ru.oublock));
 417         cr = CRED();
 418         ag->acctbuf.ac_uid = crgetruid(cr);
 419         ag->acctbuf.ac_gid = crgetrgid(cr);
 420         (void) cmpldev(&ag->acctbuf.ac_tty, cttydev(p));
 421         ag->acctbuf.ac_stat = st;
 422         ag->acctbuf.ac_flag = (ua->u_acflag | AEXPND);
 423
 424         /*
 425          * Save the size. If the write fails, reset the size to avoid
 426          * corrupted acct files.
 427          *
 428          * Large Files: We deliberately prevent accounting files from
 429          * exceeding the 2GB limit as none of the accounting commands are
 430          * currently large file aware.
 431          */
 432         va.va_mask = AT_SIZE;
 433         if (fop_getattr(vp, &va, 0, kcred, NULL) == 0) {
 434                 error = vn_rdwr(UIO_WRITE, vp, (caddr_t)&ag->acctbuf,
 435                     sizeof (ag->acctbuf), 0LL, UIO_SYSSPACE, FAPPEND,
 436                     (rlim64_t)MAXOFF32_T, kcred, &resid);
 437                 if (error || resid)
 438                         (void) fop_setattr(vp, &va, 0, kcred, NULL);
 439         }
 440         mutex_exit(&ag->aclock);
 441 }