sys/kern/kern_alq.c

   1 /*-
   2  * Copyright (c) 2002, Jeffrey Roberson <jeff@freebsd.org>
   3  * All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice unmodified, this list of conditions, and the following
  10  *    disclaimer.
  11  * 2. Redistributions in binary form must reproduce the above copyright
  12  *    notice, this list of conditions and the following disclaimer in the
  13  *    documentation and/or other materials provided with the distribution.
  14  *
  15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  18  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  19  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  20  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  21  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  22  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  23  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  24  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  25  */
  26
  27 #include <sys/cdefs.h>
  28 __FBSDID("$FreeBSD$");
  29
  30 #include "opt_mac.h"
  31
  32 #include <sys/param.h>
  33 #include <sys/systm.h>
  34 #include <sys/kernel.h>
  35 #include <sys/kthread.h>
  36 #include <sys/lock.h>
  37 #include <sys/mount.h>
  38 #include <sys/mutex.h>
  39 #include <sys/namei.h>
  40 #include <sys/proc.h>
  41 #include <sys/vnode.h>
  42 #include <sys/alq.h>
  43 #include <sys/malloc.h>
  44 #include <sys/unistd.h>
  45 #include <sys/fcntl.h>
  46 #include <sys/eventhandler.h>
  47
  48 #include <security/mac/mac_framework.h>
  49
  50 /* Async. Logging Queue */
  51 struct alq {
  52         int     aq_entmax;              /* Max entries */
  53         int     aq_entlen;              /* Entry length */
  54         char    *aq_entbuf;             /* Buffer for stored entries */
  55         int     aq_flags;               /* Queue flags */
  56         struct mtx      aq_mtx;         /* Queue lock */
  57         struct vnode    *aq_vp;         /* Open vnode handle */
  58         struct ucred    *aq_cred;       /* Credentials of the opening thread */
  59         struct ale      *aq_first;      /* First ent */
  60         struct ale      *aq_entfree;    /* First free ent */
  61         struct ale      *aq_entvalid;   /* First ent valid for writing */
  62         LIST_ENTRY(alq) aq_act;         /* List of active queues */
  63         LIST_ENTRY(alq) aq_link;        /* List of all queues */
  64 };
  65
  66 #define AQ_WANTED       0x0001          /* Wakeup sleeper when io is done */
  67 #define AQ_ACTIVE       0x0002          /* on the active list */
  68 #define AQ_FLUSHING     0x0004          /* doing IO */
  69 #define AQ_SHUTDOWN     0x0008          /* Queue no longer valid */
  70
  71 #define ALQ_LOCK(alq)   mtx_lock_spin(&(alq)->aq_mtx)
  72 #define ALQ_UNLOCK(alq) mtx_unlock_spin(&(alq)->aq_mtx)
  73
  74 static MALLOC_DEFINE(M_ALD, "ALD", "ALD");
  75
  76 /*
  77  * The ald_mtx protects the ald_queues list and the ald_active list.
  78  */
  79 static struct mtx ald_mtx;
  80 static LIST_HEAD(, alq) ald_queues;
  81 static LIST_HEAD(, alq) ald_active;
  82 static int ald_shutingdown = 0;
  83 struct thread *ald_thread;
  84 static struct proc *ald_proc;
  85
  86 #define ALD_LOCK()      mtx_lock(&ald_mtx)
  87 #define ALD_UNLOCK()    mtx_unlock(&ald_mtx)
  88
  89 /* Daemon functions */
  90 static int ald_add(struct alq *);
  91 static int ald_rem(struct alq *);
  92 static void ald_startup(void *);
  93 static void ald_daemon(void);
  94 static void ald_shutdown(void *, int);
  95 static void ald_activate(struct alq *);
  96 static void ald_deactivate(struct alq *);
  97
  98 /* Internal queue functions */
  99 static void alq_shutdown(struct alq *);
 100 static int alq_doio(struct alq *);
 101
 102
 103 /*
 104  * Add a new queue to the global list.  Fail if we're shutting down.
 105  */
 106 static int
 107 ald_add(struct alq *alq)
 108 {
 109         int error;
 110
 111         error = 0;
 112
 113         ALD_LOCK();
 114         if (ald_shutingdown) {
 115                 error = EBUSY;
 116                 goto done;
 117         }
 118         LIST_INSERT_HEAD(&ald_queues, alq, aq_link);
 119 done:
 120         ALD_UNLOCK();
 121         return (error);
 122 }
 123
 124 /*
 125  * Remove a queue from the global list unless we're shutting down.  If so,
 126  * the ald will take care of cleaning up it's resources.
 127  */
 128 static int
 129 ald_rem(struct alq *alq)
 130 {
 131         int error;
 132
 133         error = 0;
 134
 135         ALD_LOCK();
 136         if (ald_shutingdown) {
 137                 error = EBUSY;
 138                 goto done;
 139         }
 140         LIST_REMOVE(alq, aq_link);
 141 done:
 142         ALD_UNLOCK();
 143         return (error);
 144 }
 145
 146 /*
 147  * Put a queue on the active list.  This will schedule it for writing.
 148  */
 149 static void
 150 ald_activate(struct alq *alq)
 151 {
 152         LIST_INSERT_HEAD(&ald_active, alq, aq_act);
 153         wakeup(&ald_active);
 154 }
 155
 156 static void
 157 ald_deactivate(struct alq *alq)
 158 {
 159         LIST_REMOVE(alq, aq_act);
 160         alq->aq_flags &= ~AQ_ACTIVE;
 161 }
 162
 163 static void
 164 ald_startup(void *unused)
 165 {
 166         mtx_init(&ald_mtx, "ALDmtx", NULL, MTX_DEF|MTX_QUIET);
 167         LIST_INIT(&ald_queues);
 168         LIST_INIT(&ald_active);
 169 }
 170
 171 static void
 172 ald_daemon(void)
 173 {
 174         int needwakeup;
 175         struct alq *alq;
 176
 177         ald_thread = FIRST_THREAD_IN_PROC(ald_proc);
 178
 179         EVENTHANDLER_REGISTER(shutdown_pre_sync, ald_shutdown, NULL,
 180             SHUTDOWN_PRI_FIRST);
 181
 182         ALD_LOCK();
 183
 184         for (;;) {
 185                 while ((alq = LIST_FIRST(&ald_active)) == NULL)
 186                         msleep(&ald_active, &ald_mtx, PWAIT, "aldslp", 0);
 187
 188                 ALQ_LOCK(alq);
 189                 ald_deactivate(alq);
 190                 ALD_UNLOCK();
 191                 needwakeup = alq_doio(alq);
 192                 ALQ_UNLOCK(alq);
 193                 if (needwakeup)
 194                         wakeup(alq);
 195                 ALD_LOCK();
 196         }
 197 }
 198
 199 static void
 200 ald_shutdown(void *arg, int howto)
 201 {
 202         struct alq *alq;
 203
 204         ALD_LOCK();
 205         ald_shutingdown = 1;
 206
 207         while ((alq = LIST_FIRST(&ald_queues)) != NULL) {
 208                 LIST_REMOVE(alq, aq_link);
 209                 ALD_UNLOCK();
 210                 alq_shutdown(alq);
 211                 ALD_LOCK();
 212         }
 213         ALD_UNLOCK();
 214 }
 215
 216 static void
 217 alq_shutdown(struct alq *alq)
 218 {
 219         ALQ_LOCK(alq);
 220
 221         /* Stop any new writers. */
 222         alq->aq_flags |= AQ_SHUTDOWN;
 223
 224         /* Drain IO */
 225         while (alq->aq_flags & (AQ_FLUSHING|AQ_ACTIVE)) {
 226                 alq->aq_flags |= AQ_WANTED;
 227                 msleep_spin(alq, &alq->aq_mtx, "aldclose", 0);
 228         }
 229         ALQ_UNLOCK(alq);
 230
 231         vn_close(alq->aq_vp, FWRITE, alq->aq_cred,
 232             curthread);
 233         crfree(alq->aq_cred);
 234 }
 235
 236 /*
 237  * Flush all pending data to disk.  This operation will block.
 238  */
 239 static int
 240 alq_doio(struct alq *alq)
 241 {
 242         struct thread *td;
 243         struct mount *mp;
 244         struct vnode *vp;
 245         struct uio auio;
 246         struct iovec aiov[2];
 247         struct ale *ale;
 248         struct ale *alstart;
 249         int totlen;
 250         int iov;
 251         int vfslocked;
 252
 253         vp = alq->aq_vp;
 254         td = curthread;
 255         totlen = 0;
 256         iov = 0;
 257
 258         alstart = ale = alq->aq_entvalid;
 259         alq->aq_entvalid = NULL;
 260
 261         bzero(&aiov, sizeof(aiov));
 262         bzero(&auio, sizeof(auio));
 263
 264         do {
 265                 if (aiov[iov].iov_base == NULL)
 266                         aiov[iov].iov_base = ale->ae_data;
 267                 aiov[iov].iov_len += alq->aq_entlen;
 268                 totlen += alq->aq_entlen;
 269                 /* Check to see if we're wrapping the buffer */
 270                 if (ale->ae_data + alq->aq_entlen != ale->ae_next->ae_data)
 271                         iov++;
 272                 ale->ae_flags &= ~AE_VALID;
 273                 ale = ale->ae_next;
 274         } while (ale->ae_flags & AE_VALID);
 275
 276         alq->aq_flags |= AQ_FLUSHING;
 277         ALQ_UNLOCK(alq);
 278
 279         if (iov == 2 || aiov[iov].iov_base == NULL)
 280                 iov--;
 281
 282         auio.uio_iov = &aiov[0];
 283         auio.uio_offset = 0;
 284         auio.uio_segflg = UIO_SYSSPACE;
 285         auio.uio_rw = UIO_WRITE;
 286         auio.uio_iovcnt = iov + 1;
 287         auio.uio_resid = totlen;
 288         auio.uio_td = td;
 289
 290         /*
 291          * Do all of the junk required to write now.
 292          */
 293         vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 294         vn_start_write(vp, &mp, V_WAIT);
 295         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 296         VOP_LEASE(vp, td, alq->aq_cred, LEASE_WRITE);
 297         /*
 298          * XXX: VOP_WRITE error checks are ignored.
 299          */
 300 #ifdef MAC
 301         if (mac_vnode_check_write(alq->aq_cred, NOCRED, vp) == 0)
 302 #endif
 303                 VOP_WRITE(vp, &auio, IO_UNIT | IO_APPEND, alq->aq_cred);
 304         VOP_UNLOCK(vp, 0);
 305         vn_finished_write(mp);
 306         VFS_UNLOCK_GIANT(vfslocked);
 307
 308         ALQ_LOCK(alq);
 309         alq->aq_flags &= ~AQ_FLUSHING;
 310
 311         if (alq->aq_entfree == NULL)
 312                 alq->aq_entfree = alstart;
 313
 314         if (alq->aq_flags & AQ_WANTED) {
 315                 alq->aq_flags &= ~AQ_WANTED;
 316                 return (1);
 317         }
 318
 319         return(0);
 320 }
 321
 322 static struct kproc_desc ald_kp = {
 323         "ALQ Daemon",
 324         ald_daemon,
 325         &ald_proc
 326 };
 327
 328 SYSINIT(aldthread, SI_SUB_KTHREAD_IDLE, SI_ORDER_ANY, kproc_start, &ald_kp);
 329 SYSINIT(ald, SI_SUB_LOCK, SI_ORDER_ANY, ald_startup, NULL);
 330
 331
 332 /* User visible queue functions */
 333
 334 /*
 335  * Create the queue data structure, allocate the buffer, and open the file.
 336  */
 337 int
 338 alq_open(struct alq **alqp, const char *file, struct ucred *cred, int cmode,
 339     int size, int count)
 340 {
 341         struct thread *td;
 342         struct nameidata nd;
 343         struct ale *ale;
 344         struct ale *alp;
 345         struct alq *alq;
 346         char *bufp;
 347         int flags;
 348         int error;
 349         int i, vfslocked;
 350
 351         *alqp = NULL;
 352         td = curthread;
 353
 354         NDINIT(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_SYSSPACE, file, td);
 355         flags = FWRITE | O_NOFOLLOW | O_CREAT;
 356
 357         error = vn_open_cred(&nd, &flags, cmode, cred, NULL);
 358         if (error)
 359                 return (error);
 360
 361         vfslocked = NDHASGIANT(&nd);
 362         NDFREE(&nd, NDF_ONLY_PNBUF);
 363         /* We just unlock so we hold a reference */
 364         VOP_UNLOCK(nd.ni_vp, 0);
 365         VFS_UNLOCK_GIANT(vfslocked);
 366
 367         alq = malloc(sizeof(*alq), M_ALD, M_WAITOK|M_ZERO);
 368         alq->aq_entbuf = malloc(count * size, M_ALD, M_WAITOK|M_ZERO);
 369         alq->aq_first = malloc(sizeof(*ale) * count, M_ALD, M_WAITOK|M_ZERO);
 370         alq->aq_vp = nd.ni_vp;
 371         alq->aq_cred = crhold(cred);
 372         alq->aq_entmax = count;
 373         alq->aq_entlen = size;
 374         alq->aq_entfree = alq->aq_first;
 375
 376         mtx_init(&alq->aq_mtx, "ALD Queue", NULL, MTX_SPIN|MTX_QUIET);
 377
 378         bufp = alq->aq_entbuf;
 379         ale = alq->aq_first;
 380         alp = NULL;
 381
 382         /* Match up entries with buffers */
 383         for (i = 0; i < count; i++) {
 384                 if (alp)
 385                         alp->ae_next = ale;
 386                 ale->ae_data = bufp;
 387                 alp = ale;
 388                 ale++;
 389                 bufp += size;
 390         }
 391
 392         alp->ae_next = alq->aq_first;
 393
 394         if ((error = ald_add(alq)) != 0)
 395                 return (error);
 396         *alqp = alq;
 397
 398         return (0);
 399 }
 400
 401 /*
 402  * Copy a new entry into the queue.  If the operation would block either
 403  * wait or return an error depending on the value of waitok.
 404  */
 405 int
 406 alq_write(struct alq *alq, void *data, int waitok)
 407 {
 408         struct ale *ale;
 409
 410         if ((ale = alq_get(alq, waitok)) == NULL)
 411                 return (EWOULDBLOCK);
 412
 413         bcopy(data, ale->ae_data, alq->aq_entlen);
 414         alq_post(alq, ale);
 415
 416         return (0);
 417 }
 418
 419 struct ale *
 420 alq_get(struct alq *alq, int waitok)
 421 {
 422         struct ale *ale;
 423         struct ale *aln;
 424
 425         ale = NULL;
 426
 427         ALQ_LOCK(alq);
 428
 429         /* Loop until we get an entry or we're shutting down */
 430         while ((alq->aq_flags & AQ_SHUTDOWN) == 0 &&
 431             (ale = alq->aq_entfree) == NULL &&
 432             (waitok & ALQ_WAITOK)) {
 433                 alq->aq_flags |= AQ_WANTED;
 434                 msleep_spin(alq, &alq->aq_mtx, "alqget", 0);
 435         }
 436
 437         if (ale != NULL) {
 438                 aln = ale->ae_next;
 439                 if ((aln->ae_flags & AE_VALID) == 0)
 440                         alq->aq_entfree = aln;
 441                 else
 442                         alq->aq_entfree = NULL;
 443         } else
 444                 ALQ_UNLOCK(alq);
 445
 446
 447         return (ale);
 448 }
 449
 450 void
 451 alq_post(struct alq *alq, struct ale *ale)
 452 {
 453         int activate;
 454
 455         ale->ae_flags |= AE_VALID;
 456
 457         if (alq->aq_entvalid == NULL)
 458                 alq->aq_entvalid = ale;
 459
 460         if ((alq->aq_flags & AQ_ACTIVE) == 0) {
 461                 alq->aq_flags |= AQ_ACTIVE;
 462                 activate = 1;
 463         } else
 464                 activate = 0;
 465
 466         ALQ_UNLOCK(alq);
 467         if (activate) {
 468                 ALD_LOCK();
 469                 ald_activate(alq);
 470                 ALD_UNLOCK();
 471         }
 472 }
 473
 474 void
 475 alq_flush(struct alq *alq)
 476 {
 477         int needwakeup = 0;
 478
 479         ALD_LOCK();
 480         ALQ_LOCK(alq);
 481         if (alq->aq_flags & AQ_ACTIVE) {
 482                 ald_deactivate(alq);
 483                 ALD_UNLOCK();
 484                 needwakeup = alq_doio(alq);
 485         } else
 486                 ALD_UNLOCK();
 487         ALQ_UNLOCK(alq);
 488
 489         if (needwakeup)
 490                 wakeup(alq);
 491 }
 492
 493 /*
 494  * Flush remaining data, close the file and free all resources.
 495  */
 496 void
 497 alq_close(struct alq *alq)
 498 {
 499         /*
 500          * If we're already shuting down someone else will flush and close
 501          * the vnode.
 502          */
 503         if (ald_rem(alq) != 0)
 504                 return;
 505
 506         /*
 507          * Drain all pending IO.
 508          */
 509         alq_shutdown(alq);
 510
 511         mtx_destroy(&alq->aq_mtx);
 512         free(alq->aq_first, M_ALD);
 513         free(alq->aq_entbuf, M_ALD);
 514         free(alq, M_ALD);
 515 }