kern_alq.c revision 145142
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: head/sys/kern/kern_alq.c 145142 2005-04-16 12:12:27Z rwatson $"); 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/mac.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/* Async. Logging Queue */ 49struct alq { 50 int aq_entmax; /* Max entries */ 51 int aq_entlen; /* Entry length */ 52 char *aq_entbuf; /* Buffer for stored entries */ 53 int aq_flags; /* Queue flags */ 54 struct mtx aq_mtx; /* Queue lock */ 55 struct vnode *aq_vp; /* Open vnode handle */ 56 struct ucred *aq_cred; /* Credentials of the opening thread */ 57 struct ale *aq_first; /* First ent */ 58 struct ale *aq_entfree; /* First free ent */ 59 struct ale *aq_entvalid; /* First ent valid for writing */ 60 LIST_ENTRY(alq) aq_act; /* List of active queues */ 61 LIST_ENTRY(alq) aq_link; /* List of all queues */ 62}; 63 64#define AQ_WANTED 0x0001 /* Wakeup sleeper when io is done */ 65#define AQ_ACTIVE 0x0002 /* on the active list */ 66#define AQ_FLUSHING 0x0004 /* doing IO */ 67#define AQ_SHUTDOWN 0x0008 /* Queue no longer valid */ 68 69#define ALQ_LOCK(alq) mtx_lock_spin(&(alq)->aq_mtx) 70#define ALQ_UNLOCK(alq) mtx_unlock_spin(&(alq)->aq_mtx) 71 72static MALLOC_DEFINE(M_ALD, "ALD", "ALD"); 73 74/* 75 * The ald_mtx protects the ald_queues list and the ald_active list. 76 */ 77static struct mtx ald_mtx; 78static LIST_HEAD(, alq) ald_queues; 79static LIST_HEAD(, alq) ald_active; 80static int ald_shutingdown = 0; 81struct thread *ald_thread; 82static struct proc *ald_proc; 83 84#define ALD_LOCK() mtx_lock(&ald_mtx) 85#define ALD_UNLOCK() mtx_unlock(&ald_mtx) 86 87/* Daemon functions */ 88static int ald_add(struct alq *); 89static int ald_rem(struct alq *); 90static void ald_startup(void *); 91static void ald_daemon(void); 92static void ald_shutdown(void *, int); 93static void ald_activate(struct alq *); 94static void ald_deactivate(struct alq *); 95 96/* Internal queue functions */ 97static void alq_shutdown(struct alq *); 98static int alq_doio(struct alq *); 99 100 101/* 102 * Add a new queue to the global list. Fail if we're shutting down. 103 */ 104static int 105ald_add(struct alq *alq) 106{ 107 int error; 108 109 error = 0; 110 111 ALD_LOCK(); 112 if (ald_shutingdown) { 113 error = EBUSY; 114 goto done; 115 } 116 LIST_INSERT_HEAD(&ald_queues, alq, aq_link); 117done: 118 ALD_UNLOCK(); 119 return (error); 120} 121 122/* 123 * Remove a queue from the global list unless we're shutting down. If so, 124 * the ald will take care of cleaning up it's resources. 125 */ 126static int 127ald_rem(struct alq *alq) 128{ 129 int error; 130 131 error = 0; 132 133 ALD_LOCK(); 134 if (ald_shutingdown) { 135 error = EBUSY; 136 goto done; 137 } 138 LIST_REMOVE(alq, aq_link); 139done: 140 ALD_UNLOCK(); 141 return (error); 142} 143 144/* 145 * Put a queue on the active list. This will schedule it for writing. 146 */ 147static void 148ald_activate(struct alq *alq) 149{ 150 LIST_INSERT_HEAD(&ald_active, alq, aq_act); 151 wakeup(&ald_active); 152} 153 154static void 155ald_deactivate(struct alq *alq) 156{ 157 LIST_REMOVE(alq, aq_act); 158 alq->aq_flags &= ~AQ_ACTIVE; 159} 160 161static void 162ald_startup(void *unused) 163{ 164 mtx_init(&ald_mtx, "ALDmtx", NULL, MTX_DEF|MTX_QUIET); 165 LIST_INIT(&ald_queues); 166 LIST_INIT(&ald_active); 167} 168 169static void 170ald_daemon(void) 171{ 172 int needwakeup; 173 struct alq *alq; 174 175 mtx_lock(&Giant); 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 199static void 200ald_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 216static void 217alq_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 ALQ_UNLOCK(alq); 228 tsleep(alq, PWAIT, "aldclose", 0); 229 ALQ_LOCK(alq); 230 } 231 ALQ_UNLOCK(alq); 232 233 vn_close(alq->aq_vp, FWRITE, alq->aq_cred, 234 curthread); 235 crfree(alq->aq_cred); 236} 237 238/* 239 * Flush all pending data to disk. This operation will block. 240 */ 241static int 242alq_doio(struct alq *alq) 243{ 244 struct thread *td; 245 struct mount *mp; 246 struct vnode *vp; 247 struct uio auio; 248 struct iovec aiov[2]; 249 struct ale *ale; 250 struct ale *alstart; 251 int totlen; 252 int iov; 253 254 vp = alq->aq_vp; 255 td = curthread; 256 totlen = 0; 257 iov = 0; 258 259 alstart = ale = alq->aq_entvalid; 260 alq->aq_entvalid = NULL; 261 262 bzero(&aiov, sizeof(aiov)); 263 bzero(&auio, sizeof(auio)); 264 265 do { 266 if (aiov[iov].iov_base == NULL) 267 aiov[iov].iov_base = ale->ae_data; 268 aiov[iov].iov_len += alq->aq_entlen; 269 totlen += alq->aq_entlen; 270 /* Check to see if we're wrapping the buffer */ 271 if (ale->ae_data + alq->aq_entlen != ale->ae_next->ae_data) 272 iov++; 273 ale->ae_flags &= ~AE_VALID; 274 ale = ale->ae_next; 275 } while (ale->ae_flags & AE_VALID); 276 277 alq->aq_flags |= AQ_FLUSHING; 278 ALQ_UNLOCK(alq); 279 280 if (iov == 2 || aiov[iov].iov_base == NULL) 281 iov--; 282 283 auio.uio_iov = &aiov[0]; 284 auio.uio_offset = 0; 285 auio.uio_segflg = UIO_SYSSPACE; 286 auio.uio_rw = UIO_WRITE; 287 auio.uio_iovcnt = iov + 1; 288 auio.uio_resid = totlen; 289 auio.uio_td = td; 290 291 /* 292 * Do all of the junk required to write now. 293 */ 294 vn_start_write(vp, &mp, V_WAIT); 295 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 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_check_vnode_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, td); 305 vn_finished_write(mp); 306 307 ALQ_LOCK(alq); 308 alq->aq_flags &= ~AQ_FLUSHING; 309 310 if (alq->aq_entfree == NULL) 311 alq->aq_entfree = alstart; 312 313 if (alq->aq_flags & AQ_WANTED) { 314 alq->aq_flags &= ~AQ_WANTED; 315 return (1); 316 } 317 318 return(0); 319} 320 321static struct kproc_desc ald_kp = { 322 "ALQ Daemon", 323 ald_daemon, 324 &ald_proc 325}; 326 327SYSINIT(aldthread, SI_SUB_KTHREAD_IDLE, SI_ORDER_ANY, kproc_start, &ald_kp) 328SYSINIT(ald, SI_SUB_LOCK, SI_ORDER_ANY, ald_startup, NULL) 329 330 331/* User visible queue functions */ 332 333/* 334 * Create the queue data structure, allocate the buffer, and open the file. 335 */ 336int 337alq_open(struct alq **alqp, const char *file, struct ucred *cred, int cmode, 338 int size, int count) 339{ 340 struct thread *td; 341 struct nameidata nd; 342 struct ale *ale; 343 struct ale *alp; 344 struct alq *alq; 345 char *bufp; 346 int flags; 347 int error; 348 int i; 349 350 *alqp = NULL; 351 td = curthread; 352 353 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, file, td); 354 flags = FWRITE | O_NOFOLLOW | O_CREAT; 355 356 error = vn_open_cred(&nd, &flags, cmode, cred, -1); 357 if (error) 358 return (error); 359 360 NDFREE(&nd, NDF_ONLY_PNBUF); 361 /* We just unlock so we hold a reference */ 362 VOP_UNLOCK(nd.ni_vp, 0, td); 363 364 alq = malloc(sizeof(*alq), M_ALD, M_WAITOK|M_ZERO); 365 alq->aq_entbuf = malloc(count * size, M_ALD, M_WAITOK|M_ZERO); 366 alq->aq_first = malloc(sizeof(*ale) * count, M_ALD, M_WAITOK|M_ZERO); 367 alq->aq_vp = nd.ni_vp; 368 alq->aq_cred = crhold(cred); 369 alq->aq_entmax = count; 370 alq->aq_entlen = size; 371 alq->aq_entfree = alq->aq_first; 372 373 mtx_init(&alq->aq_mtx, "ALD Queue", NULL, MTX_SPIN|MTX_QUIET); 374 375 bufp = alq->aq_entbuf; 376 ale = alq->aq_first; 377 alp = NULL; 378 379 /* Match up entries with buffers */ 380 for (i = 0; i < count; i++) { 381 if (alp) 382 alp->ae_next = ale; 383 ale->ae_data = bufp; 384 alp = ale; 385 ale++; 386 bufp += size; 387 } 388 389 alp->ae_next = alq->aq_first; 390 391 if ((error = ald_add(alq)) != 0) 392 return (error); 393 *alqp = alq; 394 395 return (0); 396} 397 398/* 399 * Copy a new entry into the queue. If the operation would block either 400 * wait or return an error depending on the value of waitok. 401 */ 402int 403alq_write(struct alq *alq, void *data, int waitok) 404{ 405 struct ale *ale; 406 407 if ((ale = alq_get(alq, waitok)) == NULL) 408 return (EWOULDBLOCK); 409 410 bcopy(data, ale->ae_data, alq->aq_entlen); 411 alq_post(alq, ale); 412 413 return (0); 414} 415 416struct ale * 417alq_get(struct alq *alq, int waitok) 418{ 419 struct ale *ale; 420 struct ale *aln; 421 422 ale = NULL; 423 424 ALQ_LOCK(alq); 425 426 /* Loop until we get an entry or we're shutting down */ 427 while ((alq->aq_flags & AQ_SHUTDOWN) == 0 && 428 (ale = alq->aq_entfree) == NULL && 429 (waitok & ALQ_WAITOK)) { 430 alq->aq_flags |= AQ_WANTED; 431 ALQ_UNLOCK(alq); 432 tsleep(alq, PWAIT, "alqget", 0); 433 ALQ_LOCK(alq); 434 } 435 436 if (ale != NULL) { 437 aln = ale->ae_next; 438 if ((aln->ae_flags & AE_VALID) == 0) 439 alq->aq_entfree = aln; 440 else 441 alq->aq_entfree = NULL; 442 } else 443 ALQ_UNLOCK(alq); 444 445 446 return (ale); 447} 448 449void 450alq_post(struct alq *alq, struct ale *ale) 451{ 452 int activate; 453 454 ale->ae_flags |= AE_VALID; 455 456 if (alq->aq_entvalid == NULL) 457 alq->aq_entvalid = ale; 458 459 if ((alq->aq_flags & AQ_ACTIVE) == 0) { 460 alq->aq_flags |= AQ_ACTIVE; 461 activate = 1; 462 } else 463 activate = 0; 464 465 ALQ_UNLOCK(alq); 466 if (activate) { 467 ALD_LOCK(); 468 ald_activate(alq); 469 ALD_UNLOCK(); 470 } 471} 472 473void 474alq_flush(struct alq *alq) 475{ 476 int needwakeup = 0; 477 478 ALD_LOCK(); 479 ALQ_LOCK(alq); 480 if (alq->aq_flags & AQ_ACTIVE) { 481 ald_deactivate(alq); 482 ALD_UNLOCK(); 483 needwakeup = alq_doio(alq); 484 } else 485 ALD_UNLOCK(); 486 ALQ_UNLOCK(alq); 487 488 if (needwakeup) 489 wakeup(alq); 490} 491 492/* 493 * Flush remaining data, close the file and free all resources. 494 */ 495void 496alq_close(struct alq *alq) 497{ 498 /* 499 * If we're already shuting down someone else will flush and close 500 * the vnode. 501 */ 502 if (ald_rem(alq) != 0) 503 return; 504 505 /* 506 * Drain all pending IO. 507 */ 508 alq_shutdown(alq); 509 510 mtx_destroy(&alq->aq_mtx); 511 free(alq->aq_first, M_ALD); 512 free(alq->aq_entbuf, M_ALD); 513 free(alq, M_ALD); 514} 515