fss.c revision 1.53
1 /* $NetBSD: fss.c,v 1.53 2008/09/11 09:37:53 hannken Exp $ */ 2 3 /*- 4 * Copyright (c) 2003 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Juergen Hannken-Illjes. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 /* 33 * File system snapshot disk driver. 34 * 35 * Block/character interface to the snapshot of a mounted file system. 36 */ 37 38 #include <sys/cdefs.h> 39 __KERNEL_RCSID(0, "$NetBSD: fss.c,v 1.53 2008/09/11 09:37:53 hannken Exp $"); 40 41 #include "fss.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/namei.h> 46 #include <sys/proc.h> 47 #include <sys/errno.h> 48 #include <sys/buf.h> 49 #include <sys/malloc.h> 50 #include <sys/ioctl.h> 51 #include <sys/disklabel.h> 52 #include <sys/device.h> 53 #include <sys/disk.h> 54 #include <sys/stat.h> 55 #include <sys/mount.h> 56 #include <sys/vnode.h> 57 #include <sys/file.h> 58 #include <sys/uio.h> 59 #include <sys/conf.h> 60 #include <sys/kthread.h> 61 #include <sys/fstrans.h> 62 #include <sys/simplelock.h> 63 64 #include <miscfs/specfs/specdev.h> 65 66 #include <dev/fssvar.h> 67 68 #include <uvm/uvm.h> 69 70 #include <machine/stdarg.h> 71 72 #ifdef DEBUG 73 #define FSS_STATISTICS 74 #endif 75 76 #ifdef FSS_STATISTICS 77 struct fss_stat { 78 u_int64_t cow_calls; 79 u_int64_t cow_copied; 80 u_int64_t cow_cache_full; 81 u_int64_t indir_read; 82 u_int64_t indir_write; 83 }; 84 85 static struct fss_stat fss_stat[NFSS]; 86 87 #define FSS_STAT_INC(sc, field) \ 88 do { \ 89 fss_stat[sc->sc_unit].field++; \ 90 } while (0) 91 #define FSS_STAT_SET(sc, field, value) \ 92 do { \ 93 fss_stat[sc->sc_unit].field = value; \ 94 } while (0) 95 #define FSS_STAT_ADD(sc, field, value) \ 96 do { \ 97 fss_stat[sc->sc_unit].field += value; \ 98 } while (0) 99 #define FSS_STAT_VAL(sc, field) fss_stat[sc->sc_unit].field 100 #define FSS_STAT_CLEAR(sc) \ 101 do { \ 102 memset(&fss_stat[sc->sc_unit], 0, \ 103 sizeof(struct fss_stat)); \ 104 } while (0) 105 #else /* FSS_STATISTICS */ 106 #define FSS_STAT_INC(sc, field) 107 #define FSS_STAT_SET(sc, field, value) 108 #define FSS_STAT_ADD(sc, field, value) 109 #define FSS_STAT_CLEAR(sc) 110 #endif /* FSS_STATISTICS */ 111 112 static struct fss_softc fss_softc[NFSS]; 113 114 void fssattach(int); 115 116 dev_type_open(fss_open); 117 dev_type_close(fss_close); 118 dev_type_read(fss_read); 119 dev_type_write(fss_write); 120 dev_type_ioctl(fss_ioctl); 121 dev_type_strategy(fss_strategy); 122 dev_type_dump(fss_dump); 123 dev_type_size(fss_size); 124 125 static int fss_copy_on_write(void *, struct buf *, bool); 126 static inline void fss_error(struct fss_softc *, const char *, ...); 127 static int fss_create_files(struct fss_softc *, struct fss_set *, 128 off_t *, struct lwp *); 129 static int fss_create_snapshot(struct fss_softc *, struct fss_set *, 130 struct lwp *); 131 static int fss_delete_snapshot(struct fss_softc *, struct lwp *); 132 static int fss_softc_alloc(struct fss_softc *); 133 static void fss_softc_free(struct fss_softc *); 134 static int fss_read_cluster(struct fss_softc *, u_int32_t); 135 static void fss_bs_thread(void *); 136 static int fss_bs_io(struct fss_softc *, fss_io_type, 137 u_int32_t, off_t, int, void *); 138 static u_int32_t *fss_bs_indir(struct fss_softc *, u_int32_t); 139 140 const struct bdevsw fss_bdevsw = { 141 fss_open, fss_close, fss_strategy, fss_ioctl, 142 fss_dump, fss_size, D_DISK 143 }; 144 145 const struct cdevsw fss_cdevsw = { 146 fss_open, fss_close, fss_read, fss_write, fss_ioctl, 147 nostop, notty, nopoll, nommap, nokqfilter, D_DISK 148 }; 149 150 void 151 fssattach(int num) 152 { 153 int i; 154 struct fss_softc *sc; 155 156 for (i = 0; i < NFSS; i++) { 157 sc = &fss_softc[i]; 158 sc->sc_unit = i; 159 sc->sc_bdev = NODEV; 160 simple_lock_init(&sc->sc_slock); 161 mutex_init(&sc->sc_lock, MUTEX_DEFAULT, IPL_NONE); 162 bufq_alloc(&sc->sc_bufq, "fcfs", 0); 163 } 164 } 165 166 int 167 fss_open(dev_t dev, int flags, int mode, struct lwp *l) 168 { 169 int s, mflag; 170 struct fss_softc *sc; 171 172 mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN); 173 174 if ((sc = FSS_DEV_TO_SOFTC(dev)) == NULL) 175 return ENODEV; 176 177 FSS_LOCK(sc, s); 178 179 sc->sc_flags |= mflag; 180 181 FSS_UNLOCK(sc, s); 182 183 return 0; 184 } 185 186 int 187 fss_close(dev_t dev, int flags, int mode, struct lwp *l) 188 { 189 int s, mflag, error; 190 struct fss_softc *sc; 191 192 mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN); 193 194 if ((sc = FSS_DEV_TO_SOFTC(dev)) == NULL) 195 return ENODEV; 196 197 FSS_LOCK(sc, s); 198 199 if ((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) == mflag) { 200 if ((sc->sc_uflags & FSS_UNCONFIG_ON_CLOSE) != 0 && 201 (sc->sc_flags & FSS_ACTIVE) != 0) { 202 FSS_UNLOCK(sc, s); 203 error = fss_ioctl(dev, FSSIOCCLR, NULL, FWRITE, l); 204 if (error) 205 return error; 206 FSS_LOCK(sc, s); 207 } 208 sc->sc_uflags &= ~FSS_UNCONFIG_ON_CLOSE; 209 } 210 211 sc->sc_flags &= ~mflag; 212 213 FSS_UNLOCK(sc, s); 214 215 return 0; 216 } 217 218 void 219 fss_strategy(struct buf *bp) 220 { 221 int s; 222 struct fss_softc *sc; 223 224 sc = FSS_DEV_TO_SOFTC(bp->b_dev); 225 226 FSS_LOCK(sc, s); 227 228 if ((bp->b_flags & B_READ) != B_READ || 229 sc == NULL || !FSS_ISVALID(sc)) { 230 231 FSS_UNLOCK(sc, s); 232 233 bp->b_error = (sc == NULL ? ENODEV : EROFS); 234 bp->b_resid = bp->b_bcount; 235 biodone(bp); 236 return; 237 } 238 239 bp->b_rawblkno = bp->b_blkno; 240 BUFQ_PUT(sc->sc_bufq, bp); 241 wakeup(&sc->sc_bs_lwp); 242 243 FSS_UNLOCK(sc, s); 244 } 245 246 int 247 fss_read(dev_t dev, struct uio *uio, int flags) 248 { 249 return physio(fss_strategy, NULL, dev, B_READ, minphys, uio); 250 } 251 252 int 253 fss_write(dev_t dev, struct uio *uio, int flags) 254 { 255 return physio(fss_strategy, NULL, dev, B_WRITE, minphys, uio); 256 } 257 258 int 259 fss_ioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 260 { 261 int error; 262 struct fss_softc *sc; 263 struct fss_set *fss = (struct fss_set *)data; 264 struct fss_get *fsg = (struct fss_get *)data; 265 266 if ((sc = FSS_DEV_TO_SOFTC(dev)) == NULL) 267 return ENODEV; 268 269 switch (cmd) { 270 case FSSIOCSET: 271 mutex_enter(&sc->sc_lock); 272 if ((flag & FWRITE) == 0) 273 error = EPERM; 274 else if ((sc->sc_flags & FSS_ACTIVE) != 0) 275 error = EBUSY; 276 else 277 error = fss_create_snapshot(sc, fss, l); 278 mutex_exit(&sc->sc_lock); 279 break; 280 281 case FSSIOCCLR: 282 mutex_enter(&sc->sc_lock); 283 if ((flag & FWRITE) == 0) 284 error = EPERM; 285 else if ((sc->sc_flags & FSS_ACTIVE) == 0) 286 error = ENXIO; 287 else 288 error = fss_delete_snapshot(sc, l); 289 mutex_exit(&sc->sc_lock); 290 break; 291 292 case FSSIOCGET: 293 mutex_enter(&sc->sc_lock); 294 switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) { 295 case FSS_ACTIVE: 296 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 297 fsg->fsg_csize = FSS_CLSIZE(sc); 298 fsg->fsg_time = sc->sc_time; 299 fsg->fsg_mount_size = sc->sc_clcount; 300 fsg->fsg_bs_size = sc->sc_clnext; 301 error = 0; 302 break; 303 case FSS_PERSISTENT | FSS_ACTIVE: 304 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 305 fsg->fsg_csize = 0; 306 fsg->fsg_time = sc->sc_time; 307 fsg->fsg_mount_size = 0; 308 fsg->fsg_bs_size = 0; 309 error = 0; 310 break; 311 default: 312 error = ENXIO; 313 break; 314 } 315 mutex_exit(&sc->sc_lock); 316 break; 317 318 case FSSIOFSET: 319 sc->sc_uflags = *(int *)data; 320 error = 0; 321 break; 322 323 case FSSIOFGET: 324 *(int *)data = sc->sc_uflags; 325 error = 0; 326 break; 327 328 default: 329 error = EINVAL; 330 break; 331 } 332 333 return error; 334 } 335 336 int 337 fss_size(dev_t dev) 338 { 339 return -1; 340 } 341 342 int 343 fss_dump(dev_t dev, daddr_t blkno, void *va, 344 size_t size) 345 { 346 return EROFS; 347 } 348 349 /* 350 * An error occurred reading or writing the snapshot or backing store. 351 * If it is the first error log to console. 352 * The caller holds the simplelock. 353 */ 354 static inline void 355 fss_error(struct fss_softc *sc, const char *fmt, ...) 356 { 357 va_list ap; 358 359 if ((sc->sc_flags & (FSS_ACTIVE|FSS_ERROR)) == FSS_ACTIVE) { 360 va_start(ap, fmt); 361 printf("fss%d: snapshot invalid: ", sc->sc_unit); 362 vprintf(fmt, ap); 363 printf("\n"); 364 va_end(ap); 365 } 366 if ((sc->sc_flags & FSS_ACTIVE) == FSS_ACTIVE) 367 sc->sc_flags |= FSS_ERROR; 368 } 369 370 /* 371 * Allocate the variable sized parts of the softc and 372 * fork the kernel thread. 373 * 374 * The fields sc_clcount, sc_clshift, sc_cache_size and sc_indir_size 375 * must be initialized. 376 */ 377 static int 378 fss_softc_alloc(struct fss_softc *sc) 379 { 380 int i, len, error; 381 382 len = (sc->sc_clcount+NBBY-1)/NBBY; 383 sc->sc_copied = malloc(len, M_TEMP, M_ZERO|M_WAITOK|M_CANFAIL); 384 if (sc->sc_copied == NULL) 385 return(ENOMEM); 386 387 len = sc->sc_cache_size*sizeof(struct fss_cache); 388 sc->sc_cache = malloc(len, M_TEMP, M_ZERO|M_WAITOK|M_CANFAIL); 389 if (sc->sc_cache == NULL) 390 return(ENOMEM); 391 392 len = FSS_CLSIZE(sc); 393 for (i = 0; i < sc->sc_cache_size; i++) { 394 sc->sc_cache[i].fc_type = FSS_CACHE_FREE; 395 sc->sc_cache[i].fc_data = malloc(len, M_TEMP, 396 M_WAITOK|M_CANFAIL); 397 if (sc->sc_cache[i].fc_data == NULL) 398 return(ENOMEM); 399 } 400 401 len = (sc->sc_indir_size+NBBY-1)/NBBY; 402 sc->sc_indir_valid = malloc(len, M_TEMP, M_ZERO|M_WAITOK|M_CANFAIL); 403 if (sc->sc_indir_valid == NULL) 404 return(ENOMEM); 405 406 len = FSS_CLSIZE(sc); 407 sc->sc_indir_data = malloc(len, M_TEMP, M_ZERO|M_WAITOK|M_CANFAIL); 408 if (sc->sc_indir_data == NULL) 409 return(ENOMEM); 410 411 if ((error = kthread_create(PRI_BIO, 0, NULL, fss_bs_thread, sc, 412 &sc->sc_bs_lwp, "fssbs%d", sc->sc_unit)) != 0) 413 return error; 414 415 sc->sc_flags |= FSS_BS_THREAD; 416 return 0; 417 } 418 419 /* 420 * Free the variable sized parts of the softc. 421 */ 422 static void 423 fss_softc_free(struct fss_softc *sc) 424 { 425 int s, i; 426 427 if ((sc->sc_flags & FSS_BS_THREAD) != 0) { 428 FSS_LOCK(sc, s); 429 sc->sc_flags &= ~FSS_BS_THREAD; 430 wakeup(&sc->sc_bs_lwp); 431 while (sc->sc_bs_lwp != NULL) 432 ltsleep(&sc->sc_bs_lwp, PRIBIO, "fssthread", 0, 433 &sc->sc_slock); 434 FSS_UNLOCK(sc, s); 435 } 436 437 if (sc->sc_copied != NULL) 438 free(sc->sc_copied, M_TEMP); 439 sc->sc_copied = NULL; 440 441 if (sc->sc_cache != NULL) { 442 for (i = 0; i < sc->sc_cache_size; i++) 443 if (sc->sc_cache[i].fc_data != NULL) 444 free(sc->sc_cache[i].fc_data, M_TEMP); 445 free(sc->sc_cache, M_TEMP); 446 } 447 sc->sc_cache = NULL; 448 449 if (sc->sc_indir_valid != NULL) 450 free(sc->sc_indir_valid, M_TEMP); 451 sc->sc_indir_valid = NULL; 452 453 if (sc->sc_indir_data != NULL) 454 free(sc->sc_indir_data, M_TEMP); 455 sc->sc_indir_data = NULL; 456 } 457 458 /* 459 * Check if an unmount is ok. If forced, set this snapshot into ERROR state. 460 */ 461 int 462 fss_umount_hook(struct mount *mp, int forced) 463 { 464 int i, s; 465 466 for (i = 0; i < NFSS; i++) { 467 FSS_LOCK(&fss_softc[i], s); 468 if ((fss_softc[i].sc_flags & FSS_ACTIVE) != 0 && 469 fss_softc[i].sc_mount == mp) { 470 if (forced) 471 fss_error(&fss_softc[i], "forced unmount"); 472 else { 473 FSS_UNLOCK(&fss_softc[i], s); 474 return EBUSY; 475 } 476 } 477 FSS_UNLOCK(&fss_softc[i], s); 478 } 479 480 return 0; 481 } 482 483 /* 484 * A buffer is written to the snapshotted block device. Copy to 485 * backing store if needed. 486 */ 487 static int 488 fss_copy_on_write(void *v, struct buf *bp, bool data_valid) 489 { 490 int error, s; 491 u_int32_t cl, ch, c; 492 struct fss_softc *sc = v; 493 494 FSS_LOCK(sc, s); 495 if (!FSS_ISVALID(sc)) { 496 FSS_UNLOCK(sc, s); 497 return 0; 498 } 499 500 FSS_STAT_INC(sc, cow_calls); 501 502 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 503 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 504 error = 0; 505 if (curlwp == uvm.pagedaemon_lwp) { 506 for (c = cl; c <= ch; c++) 507 if (isclr(sc->sc_copied, c)) { 508 error = ENOMEM; 509 break; 510 } 511 } 512 FSS_UNLOCK(sc, s); 513 514 if (error == 0) 515 for (c = cl; c <= ch; c++) { 516 error = fss_read_cluster(sc, c); 517 if (error) 518 break; 519 } 520 521 return error; 522 } 523 524 /* 525 * Lookup and open needed files. 526 * 527 * For file system internal snapshot initializes sc_mntname, sc_mount, 528 * sc_bs_vp and sc_time. 529 * 530 * Otherwise returns dev and size of the underlying block device. 531 * Initializes sc_mntname, sc_mount, sc_bdev, sc_bs_vp and sc_mount 532 */ 533 static int 534 fss_create_files(struct fss_softc *sc, struct fss_set *fss, 535 off_t *bsize, struct lwp *l) 536 { 537 int error, bits, fsbsize; 538 const struct bdevsw *bdev; 539 struct timespec ts; 540 struct partinfo dpart; 541 struct vattr va; 542 struct nameidata nd; 543 544 /* 545 * Get the mounted file system. 546 */ 547 548 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_mount); 549 if ((error = namei(&nd)) != 0) 550 return error; 551 552 if ((nd.ni_vp->v_vflag & VV_ROOT) != VV_ROOT) { 553 vrele(nd.ni_vp); 554 return EINVAL; 555 } 556 557 sc->sc_mount = nd.ni_vp->v_mount; 558 memcpy(sc->sc_mntname, sc->sc_mount->mnt_stat.f_mntonname, MNAMELEN); 559 560 vrele(nd.ni_vp); 561 562 /* 563 * Check for file system internal snapshot. 564 */ 565 566 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_USERSPACE, fss->fss_bstore); 567 if ((error = namei(&nd)) != 0) 568 return error; 569 570 if (nd.ni_vp->v_type == VREG && nd.ni_vp->v_mount == sc->sc_mount) { 571 sc->sc_flags |= FSS_PERSISTENT; 572 sc->sc_bs_vp = nd.ni_vp; 573 574 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 575 bits = sizeof(sc->sc_bs_bshift)*NBBY; 576 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < bits; 577 sc->sc_bs_bshift++) 578 if (FSS_FSBSIZE(sc) == fsbsize) 579 break; 580 if (sc->sc_bs_bshift >= bits) { 581 VOP_UNLOCK(sc->sc_bs_vp, 0); 582 return EINVAL; 583 } 584 585 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 586 sc->sc_clshift = 0; 587 588 error = VFS_SNAPSHOT(sc->sc_mount, sc->sc_bs_vp, &ts); 589 TIMESPEC_TO_TIMEVAL(&sc->sc_time, &ts); 590 591 VOP_UNLOCK(sc->sc_bs_vp, 0); 592 593 return error; 594 } 595 vput(nd.ni_vp); 596 597 /* 598 * Get the block device it is mounted on. 599 */ 600 601 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, 602 sc->sc_mount->mnt_stat.f_mntfromname); 603 if ((error = namei(&nd)) != 0) 604 return error; 605 606 if (nd.ni_vp->v_type != VBLK) { 607 vrele(nd.ni_vp); 608 return EINVAL; 609 } 610 611 sc->sc_bdev = nd.ni_vp->v_rdev; 612 vrele(nd.ni_vp); 613 614 /* 615 * Get the block device size. 616 */ 617 618 bdev = bdevsw_lookup(sc->sc_bdev); 619 if (bdev == NULL) 620 return ENXIO; 621 error = (*bdev->d_ioctl)(sc->sc_bdev, DIOCGPART, &dpart, FREAD, l); 622 if (error) 623 return error; 624 625 *bsize = (off_t)dpart.disklab->d_secsize*dpart.part->p_size; 626 627 /* 628 * Get the backing store 629 */ 630 631 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_bstore); 632 if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) 633 return error; 634 VOP_UNLOCK(nd.ni_vp, 0); 635 636 sc->sc_bs_vp = nd.ni_vp; 637 638 if (nd.ni_vp->v_type != VREG && nd.ni_vp->v_type != VCHR) 639 return EINVAL; 640 641 if (sc->sc_bs_vp->v_type == VREG) { 642 error = VOP_GETATTR(sc->sc_bs_vp, &va, l->l_cred); 643 if (error != 0) 644 return error; 645 sc->sc_bs_size = va.va_size; 646 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 647 if (fsbsize & (fsbsize-1)) /* No power of two */ 648 return EINVAL; 649 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < 32; 650 sc->sc_bs_bshift++) 651 if (FSS_FSBSIZE(sc) == fsbsize) 652 break; 653 if (sc->sc_bs_bshift >= 32) 654 return EINVAL; 655 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 656 } else { 657 sc->sc_bs_bshift = DEV_BSHIFT; 658 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 659 } 660 661 /* 662 * As all IO to from/to the backing store goes through 663 * VOP_STRATEGY() clean the buffer cache to prevent 664 * cache incoherencies. 665 */ 666 if ((error = vinvalbuf(sc->sc_bs_vp, V_SAVE, l->l_cred, l, 0, 0)) != 0) 667 return error; 668 669 return 0; 670 } 671 672 /* 673 * Create a snapshot. 674 */ 675 static int 676 fss_create_snapshot(struct fss_softc *sc, struct fss_set *fss, struct lwp *l) 677 { 678 int len, error; 679 u_int32_t csize; 680 off_t bsize; 681 682 bsize = 0; /* XXX gcc */ 683 684 /* 685 * Open needed files. 686 */ 687 if ((error = fss_create_files(sc, fss, &bsize, l)) != 0) 688 goto bad; 689 690 if (sc->sc_flags & FSS_PERSISTENT) { 691 fss_softc_alloc(sc); 692 sc->sc_flags |= FSS_ACTIVE; 693 return 0; 694 } 695 696 /* 697 * Set cluster size. Must be a power of two and 698 * a multiple of backing store block size. 699 */ 700 if (fss->fss_csize <= 0) 701 csize = MAXPHYS; 702 else 703 csize = fss->fss_csize; 704 if (bsize/csize > FSS_CLUSTER_MAX) 705 csize = bsize/FSS_CLUSTER_MAX+1; 706 707 for (sc->sc_clshift = sc->sc_bs_bshift; sc->sc_clshift < 32; 708 sc->sc_clshift++) 709 if (FSS_CLSIZE(sc) >= csize) 710 break; 711 if (sc->sc_clshift >= 32) { 712 error = EINVAL; 713 goto bad; 714 } 715 sc->sc_clmask = FSS_CLSIZE(sc)-1; 716 717 /* 718 * Set number of cache slots. 719 */ 720 if (FSS_CLSIZE(sc) <= 8192) 721 sc->sc_cache_size = 32; 722 else if (FSS_CLSIZE(sc) <= 65536) 723 sc->sc_cache_size = 8; 724 else 725 sc->sc_cache_size = 4; 726 727 /* 728 * Set number of clusters and size of last cluster. 729 */ 730 sc->sc_clcount = FSS_BTOCL(sc, bsize-1)+1; 731 sc->sc_clresid = FSS_CLOFF(sc, bsize-1)+1; 732 733 /* 734 * Set size of indirect table. 735 */ 736 len = sc->sc_clcount*sizeof(u_int32_t); 737 sc->sc_indir_size = FSS_BTOCL(sc, len)+1; 738 sc->sc_clnext = sc->sc_indir_size; 739 sc->sc_indir_cur = 0; 740 741 if ((error = fss_softc_alloc(sc)) != 0) 742 goto bad; 743 744 /* 745 * Activate the snapshot. 746 */ 747 748 if ((error = vfs_suspend(sc->sc_mount, 0)) != 0) 749 goto bad; 750 751 microtime(&sc->sc_time); 752 753 if (error == 0) 754 error = fscow_establish(sc->sc_mount, 755 fss_copy_on_write, sc); 756 if (error == 0) 757 sc->sc_flags |= FSS_ACTIVE; 758 759 vfs_resume(sc->sc_mount); 760 761 if (error != 0) 762 goto bad; 763 764 #ifdef DEBUG 765 printf("fss%d: %s snapshot active\n", sc->sc_unit, sc->sc_mntname); 766 printf("fss%d: %u clusters of %u, %u cache slots, %u indir clusters\n", 767 sc->sc_unit, sc->sc_clcount, FSS_CLSIZE(sc), 768 sc->sc_cache_size, sc->sc_indir_size); 769 #endif 770 771 return 0; 772 773 bad: 774 fss_softc_free(sc); 775 if (sc->sc_bs_vp != NULL) { 776 if (sc->sc_flags & FSS_PERSISTENT) 777 vn_close(sc->sc_bs_vp, FREAD, l->l_cred); 778 else 779 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 780 } 781 sc->sc_bs_vp = NULL; 782 783 return error; 784 } 785 786 /* 787 * Delete a snapshot. 788 */ 789 static int 790 fss_delete_snapshot(struct fss_softc *sc, struct lwp *l) 791 { 792 int s; 793 794 if ((sc->sc_flags & FSS_PERSISTENT) == 0) 795 fscow_disestablish(sc->sc_mount, fss_copy_on_write, sc); 796 797 FSS_LOCK(sc, s); 798 sc->sc_flags &= ~(FSS_ACTIVE|FSS_ERROR); 799 sc->sc_mount = NULL; 800 sc->sc_bdev = NODEV; 801 FSS_UNLOCK(sc, s); 802 803 fss_softc_free(sc); 804 if (sc->sc_flags & FSS_PERSISTENT) 805 vn_close(sc->sc_bs_vp, FREAD, l->l_cred); 806 else 807 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 808 sc->sc_bs_vp = NULL; 809 sc->sc_flags &= ~FSS_PERSISTENT; 810 811 FSS_STAT_CLEAR(sc); 812 813 return 0; 814 } 815 816 /* 817 * Read a cluster from the snapshotted block device to the cache. 818 */ 819 static int 820 fss_read_cluster(struct fss_softc *sc, u_int32_t cl) 821 { 822 int error, s, todo, offset, len; 823 daddr_t dblk; 824 struct buf *bp, *mbp; 825 struct fss_cache *scp, *scl; 826 827 /* 828 * Get a free cache slot. 829 */ 830 scl = sc->sc_cache+sc->sc_cache_size; 831 832 FSS_LOCK(sc, s); 833 834 restart: 835 if (isset(sc->sc_copied, cl) || !FSS_ISVALID(sc)) { 836 FSS_UNLOCK(sc, s); 837 return 0; 838 } 839 840 for (scp = sc->sc_cache; scp < scl; scp++) 841 if (scp->fc_type != FSS_CACHE_FREE && 842 scp->fc_cluster == cl) { 843 ltsleep(&scp->fc_type, PRIBIO, "cowwait2", 0, 844 &sc->sc_slock); 845 goto restart; 846 } 847 848 for (scp = sc->sc_cache; scp < scl; scp++) 849 if (scp->fc_type == FSS_CACHE_FREE) { 850 scp->fc_type = FSS_CACHE_BUSY; 851 scp->fc_cluster = cl; 852 break; 853 } 854 if (scp >= scl) { 855 FSS_STAT_INC(sc, cow_cache_full); 856 ltsleep(&sc->sc_cache, PRIBIO, "cowwait3", 0, &sc->sc_slock); 857 goto restart; 858 } 859 860 FSS_UNLOCK(sc, s); 861 862 /* 863 * Start the read. 864 */ 865 FSS_STAT_INC(sc, cow_copied); 866 867 dblk = btodb(FSS_CLTOB(sc, cl)); 868 if (cl == sc->sc_clcount-1) { 869 todo = sc->sc_clresid; 870 memset((char *)scp->fc_data + todo, 0, FSS_CLSIZE(sc) - todo); 871 } else 872 todo = FSS_CLSIZE(sc); 873 offset = 0; 874 mbp = getiobuf(NULL, true); 875 mbp->b_bufsize = todo; 876 mbp->b_data = scp->fc_data; 877 mbp->b_resid = mbp->b_bcount = todo; 878 mbp->b_flags = B_READ; 879 mbp->b_cflags = BC_BUSY; 880 mbp->b_dev = sc->sc_bdev; 881 while (todo > 0) { 882 len = todo; 883 if (len > MAXPHYS) 884 len = MAXPHYS; 885 if (btodb(FSS_CLTOB(sc, cl)) == dblk && len == todo) 886 bp = mbp; 887 else { 888 bp = getiobuf(NULL, true); 889 nestiobuf_setup(mbp, bp, offset, len); 890 } 891 bp->b_lblkno = 0; 892 bp->b_blkno = dblk; 893 bdev_strategy(bp); 894 dblk += btodb(len); 895 offset += len; 896 todo -= len; 897 } 898 error = biowait(mbp); 899 putiobuf(mbp); 900 901 FSS_LOCK(sc, s); 902 scp->fc_type = (error ? FSS_CACHE_FREE : FSS_CACHE_VALID); 903 if (error) 904 wakeup(&scp->fc_type); 905 else { 906 setbit(sc->sc_copied, scp->fc_cluster); 907 wakeup(&sc->sc_bs_lwp); 908 } 909 FSS_UNLOCK(sc, s); 910 911 return error; 912 } 913 914 /* 915 * Read/write clusters from/to backing store. 916 * For persistent snapshots must be called with cl == 0. off is the 917 * offset into the snapshot. 918 */ 919 static int 920 fss_bs_io(struct fss_softc *sc, fss_io_type rw, 921 u_int32_t cl, off_t off, int len, void *data) 922 { 923 int error; 924 925 off += FSS_CLTOB(sc, cl); 926 927 vn_lock(sc->sc_bs_vp, LK_EXCLUSIVE|LK_RETRY); 928 929 error = vn_rdwr((rw == FSS_READ ? UIO_READ : UIO_WRITE), sc->sc_bs_vp, 930 data, len, off, UIO_SYSSPACE, IO_UNIT|IO_NODELOCKED, 931 sc->sc_bs_lwp->l_cred, NULL, NULL); 932 if (error == 0) { 933 mutex_enter(&sc->sc_bs_vp->v_interlock); 934 error = VOP_PUTPAGES(sc->sc_bs_vp, trunc_page(off), 935 round_page(off+len), PGO_CLEANIT|PGO_SYNCIO|PGO_FREE); 936 } 937 938 VOP_UNLOCK(sc->sc_bs_vp, 0); 939 940 return error; 941 } 942 943 /* 944 * Get a pointer to the indirect slot for this cluster. 945 */ 946 static u_int32_t * 947 fss_bs_indir(struct fss_softc *sc, u_int32_t cl) 948 { 949 u_int32_t icl; 950 int ioff; 951 952 icl = cl/(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 953 ioff = cl%(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 954 955 if (sc->sc_indir_cur == icl) 956 return &sc->sc_indir_data[ioff]; 957 958 if (sc->sc_indir_dirty) { 959 FSS_STAT_INC(sc, indir_write); 960 if (fss_bs_io(sc, FSS_WRITE, sc->sc_indir_cur, 0, 961 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 962 return NULL; 963 setbit(sc->sc_indir_valid, sc->sc_indir_cur); 964 } 965 966 sc->sc_indir_dirty = 0; 967 sc->sc_indir_cur = icl; 968 969 if (isset(sc->sc_indir_valid, sc->sc_indir_cur)) { 970 FSS_STAT_INC(sc, indir_read); 971 if (fss_bs_io(sc, FSS_READ, sc->sc_indir_cur, 0, 972 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 973 return NULL; 974 } else 975 memset(sc->sc_indir_data, 0, FSS_CLSIZE(sc)); 976 977 return &sc->sc_indir_data[ioff]; 978 } 979 980 /* 981 * The kernel thread (one for every active snapshot). 982 * 983 * After wakeup it cleans the cache and runs the I/O requests. 984 */ 985 static void 986 fss_bs_thread(void *arg) 987 { 988 int error, len, nfreed, nio, s; 989 long off; 990 char *addr; 991 u_int32_t c, cl, ch, *indirp; 992 struct buf *bp, *nbp; 993 struct fss_softc *sc; 994 struct fss_cache *scp, *scl; 995 996 sc = arg; 997 998 scl = sc->sc_cache+sc->sc_cache_size; 999 1000 nfreed = nio = 1; /* Dont sleep the first time */ 1001 1002 FSS_LOCK(sc, s); 1003 1004 for (;;) { 1005 if (nfreed == 0 && nio == 0) 1006 ltsleep(&sc->sc_bs_lwp, PVM-1, "fssbs", 0, 1007 &sc->sc_slock); 1008 1009 if ((sc->sc_flags & FSS_BS_THREAD) == 0) { 1010 sc->sc_bs_lwp = NULL; 1011 wakeup(&sc->sc_bs_lwp); 1012 1013 FSS_UNLOCK(sc, s); 1014 1015 #ifdef FSS_STATISTICS 1016 if ((sc->sc_flags & FSS_PERSISTENT) == 0) { 1017 printf("fss%d: cow called %" PRId64 " times," 1018 " copied %" PRId64 " clusters," 1019 " cache full %" PRId64 " times\n", 1020 sc->sc_unit, 1021 FSS_STAT_VAL(sc, cow_calls), 1022 FSS_STAT_VAL(sc, cow_copied), 1023 FSS_STAT_VAL(sc, cow_cache_full)); 1024 printf("fss%d: %" PRId64 " indir reads," 1025 " %" PRId64 " indir writes\n", 1026 sc->sc_unit, 1027 FSS_STAT_VAL(sc, indir_read), 1028 FSS_STAT_VAL(sc, indir_write)); 1029 } 1030 #endif /* FSS_STATISTICS */ 1031 kthread_exit(0); 1032 } 1033 1034 /* 1035 * Process I/O requests (persistent) 1036 */ 1037 1038 if (sc->sc_flags & FSS_PERSISTENT) { 1039 nfreed = nio = 0; 1040 1041 if ((bp = BUFQ_GET(sc->sc_bufq)) == NULL) 1042 continue; 1043 1044 nio++; 1045 1046 if (FSS_ISVALID(sc)) { 1047 FSS_UNLOCK(sc, s); 1048 1049 error = fss_bs_io(sc, FSS_READ, 0, 1050 dbtob(bp->b_blkno), bp->b_bcount, 1051 bp->b_data); 1052 1053 FSS_LOCK(sc, s); 1054 } else 1055 error = ENXIO; 1056 1057 if (error) { 1058 bp->b_error = error; 1059 bp->b_resid = bp->b_bcount; 1060 } else 1061 bp->b_resid = 0; 1062 1063 biodone(bp); 1064 1065 continue; 1066 } 1067 1068 /* 1069 * Clean the cache 1070 */ 1071 nfreed = 0; 1072 for (scp = sc->sc_cache; scp < scl; scp++) { 1073 if (scp->fc_type != FSS_CACHE_VALID) 1074 continue; 1075 1076 FSS_UNLOCK(sc, s); 1077 1078 indirp = fss_bs_indir(sc, scp->fc_cluster); 1079 if (indirp != NULL) { 1080 error = fss_bs_io(sc, FSS_WRITE, sc->sc_clnext, 1081 0, FSS_CLSIZE(sc), scp->fc_data); 1082 } else 1083 error = EIO; 1084 1085 FSS_LOCK(sc, s); 1086 1087 if (error == 0) { 1088 *indirp = sc->sc_clnext++; 1089 sc->sc_indir_dirty = 1; 1090 } else 1091 fss_error(sc, "write bs error %d", error); 1092 1093 scp->fc_type = FSS_CACHE_FREE; 1094 nfreed++; 1095 wakeup(&scp->fc_type); 1096 } 1097 1098 if (nfreed) 1099 wakeup(&sc->sc_cache); 1100 1101 /* 1102 * Process I/O requests 1103 */ 1104 nio = 0; 1105 1106 if ((bp = BUFQ_GET(sc->sc_bufq)) == NULL) 1107 continue; 1108 1109 nio++; 1110 1111 if (!FSS_ISVALID(sc)) { 1112 bp->b_error = ENXIO; 1113 bp->b_resid = bp->b_bcount; 1114 biodone(bp); 1115 continue; 1116 } 1117 1118 /* 1119 * First read from the snapshotted block device unless 1120 * this request is completely covered by backing store. 1121 */ 1122 1123 FSS_UNLOCK(sc, s); 1124 1125 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 1126 off = FSS_CLOFF(sc, dbtob(bp->b_blkno)); 1127 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 1128 1129 error = 0; 1130 for (c = cl; c <= ch; c++) { 1131 if (isset(sc->sc_copied, c)) 1132 continue; 1133 /* Not on backing store, read from device. */ 1134 nbp = getiobuf(NULL, true); 1135 nbp->b_flags = B_READ; 1136 nbp->b_resid = nbp->b_bcount = bp->b_bcount; 1137 nbp->b_bufsize = bp->b_bcount; 1138 nbp->b_data = bp->b_data; 1139 nbp->b_blkno = bp->b_blkno; 1140 nbp->b_lblkno = 0; 1141 nbp->b_dev = sc->sc_bdev; 1142 SET(nbp->b_cflags, BC_BUSY); /* mark buffer busy */ 1143 1144 bdev_strategy(nbp); 1145 1146 error = biowait(nbp); 1147 if (error != 0) { 1148 bp->b_resid = bp->b_bcount; 1149 bp->b_error = nbp->b_error; 1150 biodone(bp); 1151 } 1152 putiobuf(nbp); 1153 break; 1154 } 1155 if (error) { 1156 FSS_LOCK(sc, s); 1157 continue; 1158 } 1159 1160 FSS_LOCK(sc, s); 1161 1162 /* 1163 * Replace those parts that have been saved to backing store. 1164 */ 1165 1166 addr = bp->b_data; 1167 bp->b_resid = bp->b_bcount; 1168 for (c = cl; c <= ch; 1169 c++, off = 0, bp->b_resid -= len, addr += len) { 1170 len = FSS_CLSIZE(sc)-off; 1171 if (len > bp->b_resid) 1172 len = bp->b_resid; 1173 1174 if (isclr(sc->sc_copied, c)) 1175 continue; 1176 1177 FSS_UNLOCK(sc, s); 1178 1179 indirp = fss_bs_indir(sc, c); 1180 1181 FSS_LOCK(sc, s); 1182 1183 if (indirp == NULL || *indirp == 0) { 1184 /* 1185 * Not on backing store. Either in cache 1186 * or hole in the snapshotted block device. 1187 */ 1188 for (scp = sc->sc_cache; scp < scl; scp++) 1189 if (scp->fc_type == FSS_CACHE_VALID && 1190 scp->fc_cluster == c) 1191 break; 1192 if (scp < scl) 1193 memcpy(addr, (char *)scp->fc_data+off, len); 1194 else 1195 memset(addr, 0, len); 1196 continue; 1197 } 1198 /* 1199 * Read from backing store. 1200 */ 1201 1202 FSS_UNLOCK(sc, s); 1203 1204 if ((error = fss_bs_io(sc, FSS_READ, *indirp, 1205 off, len, addr)) != 0) { 1206 bp->b_resid = bp->b_bcount; 1207 bp->b_error = error; 1208 FSS_LOCK(sc, s); 1209 break; 1210 } 1211 1212 FSS_LOCK(sc, s); 1213 1214 } 1215 1216 biodone(bp); 1217 } 1218 } 1219
Indexes created Wed Nov 11 00:26:22 UTC 2026