fss.c revision 1.50
1 /* $NetBSD: fss.c,v 1.50 2008/08/12 10:14:37 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 1.50 2008/08/12 10:14:37 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 <machine/stdarg.h> 69 70 #ifdef DEBUG 71 #define FSS_STATISTICS 72 #endif 73 74 #ifdef FSS_STATISTICS 75 struct fss_stat { 76 u_int64_t cow_calls; 77 u_int64_t cow_copied; 78 u_int64_t cow_cache_full; 79 u_int64_t indir_read; 80 u_int64_t indir_write; 81 }; 82 83 static struct fss_stat fss_stat[NFSS]; 84 85 #define FSS_STAT_INC(sc, field) \ 86 do { \ 87 fss_stat[sc->sc_unit].field++; \ 88 } while (0) 89 #define FSS_STAT_SET(sc, field, value) \ 90 do { \ 91 fss_stat[sc->sc_unit].field = value; \ 92 } while (0) 93 #define FSS_STAT_ADD(sc, field, value) \ 94 do { \ 95 fss_stat[sc->sc_unit].field += value; \ 96 } while (0) 97 #define FSS_STAT_VAL(sc, field) fss_stat[sc->sc_unit].field 98 #define FSS_STAT_CLEAR(sc) \ 99 do { \ 100 memset(&fss_stat[sc->sc_unit], 0, \ 101 sizeof(struct fss_stat)); \ 102 } while (0) 103 #else /* FSS_STATISTICS */ 104 #define FSS_STAT_INC(sc, field) 105 #define FSS_STAT_SET(sc, field, value) 106 #define FSS_STAT_ADD(sc, field, value) 107 #define FSS_STAT_CLEAR(sc) 108 #endif /* FSS_STATISTICS */ 109 110 static struct fss_softc fss_softc[NFSS]; 111 112 void fssattach(int); 113 114 dev_type_open(fss_open); 115 dev_type_close(fss_close); 116 dev_type_read(fss_read); 117 dev_type_write(fss_write); 118 dev_type_ioctl(fss_ioctl); 119 dev_type_strategy(fss_strategy); 120 dev_type_dump(fss_dump); 121 dev_type_size(fss_size); 122 123 static int fss_copy_on_write(void *, struct buf *, bool); 124 static inline void fss_error(struct fss_softc *, const char *, ...); 125 static int fss_create_files(struct fss_softc *, struct fss_set *, 126 off_t *, struct lwp *); 127 static int fss_create_snapshot(struct fss_softc *, struct fss_set *, 128 struct lwp *); 129 static int fss_delete_snapshot(struct fss_softc *, struct lwp *); 130 static int fss_softc_alloc(struct fss_softc *); 131 static void fss_softc_free(struct fss_softc *); 132 static void fss_cluster_iodone(struct buf *); 133 static void fss_read_cluster(struct fss_softc *, u_int32_t); 134 static void fss_bs_thread(void *); 135 static int fss_bs_io(struct fss_softc *, fss_io_type, 136 u_int32_t, off_t, int, void *); 137 static u_int32_t *fss_bs_indir(struct fss_softc *, u_int32_t); 138 139 const struct bdevsw fss_bdevsw = { 140 fss_open, fss_close, fss_strategy, fss_ioctl, 141 fss_dump, fss_size, D_DISK 142 }; 143 144 const struct cdevsw fss_cdevsw = { 145 fss_open, fss_close, fss_read, fss_write, fss_ioctl, 146 nostop, notty, nopoll, nommap, nokqfilter, D_DISK 147 }; 148 149 void 150 fssattach(int num) 151 { 152 int i; 153 struct fss_softc *sc; 154 155 for (i = 0; i < NFSS; i++) { 156 sc = &fss_softc[i]; 157 sc->sc_unit = i; 158 sc->sc_bdev = NODEV; 159 simple_lock_init(&sc->sc_slock); 160 mutex_init(&sc->sc_lock, MUTEX_DEFAULT, IPL_NONE); 161 bufq_alloc(&sc->sc_bufq, "fcfs", 0); 162 } 163 } 164 165 int 166 fss_open(dev_t dev, int flags, int mode, struct lwp *l) 167 { 168 int s, mflag; 169 struct fss_softc *sc; 170 171 mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN); 172 173 if ((sc = FSS_DEV_TO_SOFTC(dev)) == NULL) 174 return ENODEV; 175 176 FSS_LOCK(sc, s); 177 178 sc->sc_flags |= mflag; 179 180 FSS_UNLOCK(sc, s); 181 182 return 0; 183 } 184 185 int 186 fss_close(dev_t dev, int flags, int mode, struct lwp *l) 187 { 188 int s, mflag, error; 189 struct fss_softc *sc; 190 191 mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN); 192 193 if ((sc = FSS_DEV_TO_SOFTC(dev)) == NULL) 194 return ENODEV; 195 196 FSS_LOCK(sc, s); 197 198 if ((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) == mflag) { 199 if ((sc->sc_uflags & FSS_UNCONFIG_ON_CLOSE) != 0 && 200 (sc->sc_flags & FSS_ACTIVE) != 0) { 201 FSS_UNLOCK(sc, s); 202 error = fss_ioctl(dev, FSSIOCCLR, NULL, FWRITE, l); 203 if (error) 204 return error; 205 FSS_LOCK(sc, s); 206 } 207 sc->sc_uflags &= ~FSS_UNCONFIG_ON_CLOSE; 208 } 209 210 sc->sc_flags &= ~mflag; 211 212 FSS_UNLOCK(sc, s); 213 214 return 0; 215 } 216 217 void 218 fss_strategy(struct buf *bp) 219 { 220 int s; 221 struct fss_softc *sc; 222 223 sc = FSS_DEV_TO_SOFTC(bp->b_dev); 224 225 FSS_LOCK(sc, s); 226 227 if ((bp->b_flags & B_READ) != B_READ || 228 sc == NULL || !FSS_ISVALID(sc)) { 229 230 FSS_UNLOCK(sc, s); 231 232 bp->b_error = (sc == NULL ? ENODEV : EROFS); 233 bp->b_resid = bp->b_bcount; 234 biodone(bp); 235 return; 236 } 237 238 bp->b_rawblkno = bp->b_blkno; 239 BUFQ_PUT(sc->sc_bufq, bp); 240 wakeup(&sc->sc_bs_lwp); 241 242 FSS_UNLOCK(sc, s); 243 } 244 245 int 246 fss_read(dev_t dev, struct uio *uio, int flags) 247 { 248 return physio(fss_strategy, NULL, dev, B_READ, minphys, uio); 249 } 250 251 int 252 fss_write(dev_t dev, struct uio *uio, int flags) 253 { 254 return physio(fss_strategy, NULL, dev, B_WRITE, minphys, uio); 255 } 256 257 int 258 fss_ioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 259 { 260 int error; 261 struct fss_softc *sc; 262 struct fss_set *fss = (struct fss_set *)data; 263 struct fss_get *fsg = (struct fss_get *)data; 264 265 if ((sc = FSS_DEV_TO_SOFTC(dev)) == NULL) 266 return ENODEV; 267 268 switch (cmd) { 269 case FSSIOCSET: 270 mutex_enter(&sc->sc_lock); 271 if ((flag & FWRITE) == 0) 272 error = EPERM; 273 else if ((sc->sc_flags & FSS_ACTIVE) != 0) 274 error = EBUSY; 275 else 276 error = fss_create_snapshot(sc, fss, l); 277 mutex_exit(&sc->sc_lock); 278 break; 279 280 case FSSIOCCLR: 281 mutex_enter(&sc->sc_lock); 282 if ((flag & FWRITE) == 0) 283 error = EPERM; 284 else if ((sc->sc_flags & FSS_ACTIVE) == 0) 285 error = ENXIO; 286 else 287 error = fss_delete_snapshot(sc, l); 288 mutex_exit(&sc->sc_lock); 289 break; 290 291 case FSSIOCGET: 292 mutex_enter(&sc->sc_lock); 293 switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) { 294 case FSS_ACTIVE: 295 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 296 fsg->fsg_csize = FSS_CLSIZE(sc); 297 fsg->fsg_time = sc->sc_time; 298 fsg->fsg_mount_size = sc->sc_clcount; 299 fsg->fsg_bs_size = sc->sc_clnext; 300 error = 0; 301 break; 302 case FSS_PERSISTENT | FSS_ACTIVE: 303 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 304 fsg->fsg_csize = 0; 305 fsg->fsg_time = sc->sc_time; 306 fsg->fsg_mount_size = 0; 307 fsg->fsg_bs_size = 0; 308 error = 0; 309 break; 310 default: 311 error = ENXIO; 312 break; 313 } 314 mutex_exit(&sc->sc_lock); 315 break; 316 317 case FSSIOFSET: 318 sc->sc_uflags = *(int *)data; 319 error = 0; 320 break; 321 322 case FSSIOFGET: 323 *(int *)data = sc->sc_uflags; 324 error = 0; 325 break; 326 327 default: 328 error = EINVAL; 329 break; 330 } 331 332 return error; 333 } 334 335 int 336 fss_size(dev_t dev) 337 { 338 return -1; 339 } 340 341 int 342 fss_dump(dev_t dev, daddr_t blkno, void *va, 343 size_t size) 344 { 345 return EROFS; 346 } 347 348 /* 349 * An error occurred reading or writing the snapshot or backing store. 350 * If it is the first error log to console. 351 * The caller holds the simplelock. 352 */ 353 static inline void 354 fss_error(struct fss_softc *sc, const char *fmt, ...) 355 { 356 va_list ap; 357 358 if ((sc->sc_flags & (FSS_ACTIVE|FSS_ERROR)) == FSS_ACTIVE) { 359 va_start(ap, fmt); 360 printf("fss%d: snapshot invalid: ", sc->sc_unit); 361 vprintf(fmt, ap); 362 printf("\n"); 363 va_end(ap); 364 } 365 if ((sc->sc_flags & FSS_ACTIVE) == FSS_ACTIVE) 366 sc->sc_flags |= FSS_ERROR; 367 } 368 369 /* 370 * Allocate the variable sized parts of the softc and 371 * fork the kernel thread. 372 * 373 * The fields sc_clcount, sc_clshift, sc_cache_size and sc_indir_size 374 * must be initialized. 375 */ 376 static int 377 fss_softc_alloc(struct fss_softc *sc) 378 { 379 int i, len, error; 380 381 len = (sc->sc_clcount+NBBY-1)/NBBY; 382 sc->sc_copied = malloc(len, M_TEMP, M_ZERO|M_WAITOK|M_CANFAIL); 383 if (sc->sc_copied == NULL) 384 return(ENOMEM); 385 386 len = sc->sc_cache_size*sizeof(struct fss_cache); 387 sc->sc_cache = malloc(len, M_TEMP, M_ZERO|M_WAITOK|M_CANFAIL); 388 if (sc->sc_cache == NULL) 389 return(ENOMEM); 390 391 len = FSS_CLSIZE(sc); 392 for (i = 0; i < sc->sc_cache_size; i++) { 393 sc->sc_cache[i].fc_type = FSS_CACHE_FREE; 394 sc->sc_cache[i].fc_softc = sc; 395 sc->sc_cache[i].fc_xfercount = 0; 396 sc->sc_cache[i].fc_data = malloc(len, M_TEMP, 397 M_WAITOK|M_CANFAIL); 398 if (sc->sc_cache[i].fc_data == NULL) 399 return(ENOMEM); 400 } 401 402 len = (sc->sc_indir_size+NBBY-1)/NBBY; 403 sc->sc_indir_valid = malloc(len, M_TEMP, M_ZERO|M_WAITOK|M_CANFAIL); 404 if (sc->sc_indir_valid == NULL) 405 return(ENOMEM); 406 407 len = FSS_CLSIZE(sc); 408 sc->sc_indir_data = malloc(len, M_TEMP, M_ZERO|M_WAITOK|M_CANFAIL); 409 if (sc->sc_indir_data == NULL) 410 return(ENOMEM); 411 412 if ((error = kthread_create(PRI_BIO, 0, NULL, fss_bs_thread, sc, 413 &sc->sc_bs_lwp, "fssbs%d", sc->sc_unit)) != 0) 414 return error; 415 416 sc->sc_flags |= FSS_BS_THREAD; 417 return 0; 418 } 419 420 /* 421 * Free the variable sized parts of the softc. 422 */ 423 static void 424 fss_softc_free(struct fss_softc *sc) 425 { 426 int s, i; 427 428 if ((sc->sc_flags & FSS_BS_THREAD) != 0) { 429 FSS_LOCK(sc, s); 430 sc->sc_flags &= ~FSS_BS_THREAD; 431 wakeup(&sc->sc_bs_lwp); 432 while (sc->sc_bs_lwp != NULL) 433 ltsleep(&sc->sc_bs_lwp, PRIBIO, "fssthread", 0, 434 &sc->sc_slock); 435 FSS_UNLOCK(sc, s); 436 } 437 438 if (sc->sc_copied != NULL) 439 free(sc->sc_copied, M_TEMP); 440 sc->sc_copied = NULL; 441 442 if (sc->sc_cache != NULL) { 443 for (i = 0; i < sc->sc_cache_size; i++) 444 if (sc->sc_cache[i].fc_data != NULL) 445 free(sc->sc_cache[i].fc_data, M_TEMP); 446 free(sc->sc_cache, M_TEMP); 447 } 448 sc->sc_cache = NULL; 449 450 if (sc->sc_indir_valid != NULL) 451 free(sc->sc_indir_valid, M_TEMP); 452 sc->sc_indir_valid = NULL; 453 454 if (sc->sc_indir_data != NULL) 455 free(sc->sc_indir_data, M_TEMP); 456 sc->sc_indir_data = NULL; 457 } 458 459 /* 460 * Check if an unmount is ok. If forced, set this snapshot into ERROR state. 461 */ 462 int 463 fss_umount_hook(struct mount *mp, int forced) 464 { 465 int i, s; 466 467 for (i = 0; i < NFSS; i++) { 468 FSS_LOCK(&fss_softc[i], s); 469 if ((fss_softc[i].sc_flags & FSS_ACTIVE) != 0 && 470 fss_softc[i].sc_mount == mp) { 471 if (forced) 472 fss_error(&fss_softc[i], "forced unmount"); 473 else { 474 FSS_UNLOCK(&fss_softc[i], s); 475 return EBUSY; 476 } 477 } 478 FSS_UNLOCK(&fss_softc[i], s); 479 } 480 481 return 0; 482 } 483 484 /* 485 * A buffer is written to the snapshotted block device. Copy to 486 * backing store if needed. 487 */ 488 static int 489 fss_copy_on_write(void *v, struct buf *bp, bool data_valid) 490 { 491 int s; 492 u_int32_t cl, ch, c; 493 struct fss_softc *sc = v; 494 495 FSS_LOCK(sc, s); 496 if (!FSS_ISVALID(sc)) { 497 FSS_UNLOCK(sc, s); 498 return 0; 499 } 500 501 FSS_UNLOCK(sc, s); 502 503 FSS_STAT_INC(sc, cow_calls); 504 505 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 506 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 507 508 for (c = cl; c <= ch; c++) 509 fss_read_cluster(sc, c); 510 511 return 0; 512 } 513 514 /* 515 * Lookup and open needed files. 516 * 517 * For file system internal snapshot initializes sc_mntname, sc_mount, 518 * sc_bs_vp and sc_time. 519 * 520 * Otherwise returns dev and size of the underlying block device. 521 * Initializes sc_mntname, sc_mount, sc_bdev, sc_bs_vp and sc_mount 522 */ 523 static int 524 fss_create_files(struct fss_softc *sc, struct fss_set *fss, 525 off_t *bsize, struct lwp *l) 526 { 527 int error, bits, fsbsize; 528 struct timespec ts; 529 struct partinfo dpart; 530 struct vattr va; 531 struct nameidata nd; 532 533 /* 534 * Get the mounted file system. 535 */ 536 537 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_mount); 538 if ((error = namei(&nd)) != 0) 539 return error; 540 541 if ((nd.ni_vp->v_vflag & VV_ROOT) != VV_ROOT) { 542 vrele(nd.ni_vp); 543 return EINVAL; 544 } 545 546 sc->sc_mount = nd.ni_vp->v_mount; 547 memcpy(sc->sc_mntname, sc->sc_mount->mnt_stat.f_mntonname, MNAMELEN); 548 549 vrele(nd.ni_vp); 550 551 /* 552 * Check for file system internal snapshot. 553 */ 554 555 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_USERSPACE, fss->fss_bstore); 556 if ((error = namei(&nd)) != 0) 557 return error; 558 559 if (nd.ni_vp->v_type == VREG && nd.ni_vp->v_mount == sc->sc_mount) { 560 sc->sc_flags |= FSS_PERSISTENT; 561 sc->sc_bs_vp = nd.ni_vp; 562 563 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 564 bits = sizeof(sc->sc_bs_bshift)*NBBY; 565 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < bits; 566 sc->sc_bs_bshift++) 567 if (FSS_FSBSIZE(sc) == fsbsize) 568 break; 569 if (sc->sc_bs_bshift >= bits) { 570 VOP_UNLOCK(sc->sc_bs_vp, 0); 571 return EINVAL; 572 } 573 574 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 575 sc->sc_clshift = 0; 576 577 error = VFS_SNAPSHOT(sc->sc_mount, sc->sc_bs_vp, &ts); 578 TIMESPEC_TO_TIMEVAL(&sc->sc_time, &ts); 579 580 VOP_UNLOCK(sc->sc_bs_vp, 0); 581 582 return error; 583 } 584 vput(nd.ni_vp); 585 586 /* 587 * Get the block device it is mounted on. 588 */ 589 590 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, 591 sc->sc_mount->mnt_stat.f_mntfromname); 592 if ((error = namei(&nd)) != 0) 593 return error; 594 595 if (nd.ni_vp->v_type != VBLK) { 596 vrele(nd.ni_vp); 597 return EINVAL; 598 } 599 600 error = VOP_IOCTL(nd.ni_vp, DIOCGPART, &dpart, FREAD, l->l_cred); 601 if (error) { 602 vrele(nd.ni_vp); 603 return error; 604 } 605 606 sc->sc_bdev = nd.ni_vp->v_rdev; 607 *bsize = (off_t)dpart.disklab->d_secsize*dpart.part->p_size; 608 vrele(nd.ni_vp); 609 610 /* 611 * Get the backing store 612 */ 613 614 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_bstore); 615 if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) 616 return error; 617 VOP_UNLOCK(nd.ni_vp, 0); 618 619 sc->sc_bs_vp = nd.ni_vp; 620 621 if (nd.ni_vp->v_type != VREG && nd.ni_vp->v_type != VCHR) 622 return EINVAL; 623 624 if (sc->sc_bs_vp->v_type == VREG) { 625 error = VOP_GETATTR(sc->sc_bs_vp, &va, l->l_cred); 626 if (error != 0) 627 return error; 628 sc->sc_bs_size = va.va_size; 629 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 630 if (fsbsize & (fsbsize-1)) /* No power of two */ 631 return EINVAL; 632 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < 32; 633 sc->sc_bs_bshift++) 634 if (FSS_FSBSIZE(sc) == fsbsize) 635 break; 636 if (sc->sc_bs_bshift >= 32) 637 return EINVAL; 638 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 639 } else { 640 sc->sc_bs_bshift = DEV_BSHIFT; 641 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 642 } 643 644 /* 645 * As all IO to from/to the backing store goes through 646 * VOP_STRATEGY() clean the buffer cache to prevent 647 * cache incoherencies. 648 */ 649 if ((error = vinvalbuf(sc->sc_bs_vp, V_SAVE, l->l_cred, l, 0, 0)) != 0) 650 return error; 651 652 return 0; 653 } 654 655 /* 656 * Create a snapshot. 657 */ 658 static int 659 fss_create_snapshot(struct fss_softc *sc, struct fss_set *fss, struct lwp *l) 660 { 661 int len, error; 662 u_int32_t csize; 663 off_t bsize; 664 665 bsize = 0; /* XXX gcc */ 666 667 /* 668 * Open needed files. 669 */ 670 if ((error = fss_create_files(sc, fss, &bsize, l)) != 0) 671 goto bad; 672 673 if (sc->sc_flags & FSS_PERSISTENT) { 674 fss_softc_alloc(sc); 675 sc->sc_flags |= FSS_ACTIVE; 676 return 0; 677 } 678 679 /* 680 * Set cluster size. Must be a power of two and 681 * a multiple of backing store block size. 682 */ 683 if (fss->fss_csize <= 0) 684 csize = MAXPHYS; 685 else 686 csize = fss->fss_csize; 687 if (bsize/csize > FSS_CLUSTER_MAX) 688 csize = bsize/FSS_CLUSTER_MAX+1; 689 690 for (sc->sc_clshift = sc->sc_bs_bshift; sc->sc_clshift < 32; 691 sc->sc_clshift++) 692 if (FSS_CLSIZE(sc) >= csize) 693 break; 694 if (sc->sc_clshift >= 32) { 695 error = EINVAL; 696 goto bad; 697 } 698 sc->sc_clmask = FSS_CLSIZE(sc)-1; 699 700 /* 701 * Set number of cache slots. 702 */ 703 if (FSS_CLSIZE(sc) <= 8192) 704 sc->sc_cache_size = 32; 705 else if (FSS_CLSIZE(sc) <= 65536) 706 sc->sc_cache_size = 8; 707 else 708 sc->sc_cache_size = 4; 709 710 /* 711 * Set number of clusters and size of last cluster. 712 */ 713 sc->sc_clcount = FSS_BTOCL(sc, bsize-1)+1; 714 sc->sc_clresid = FSS_CLOFF(sc, bsize-1)+1; 715 716 /* 717 * Set size of indirect table. 718 */ 719 len = sc->sc_clcount*sizeof(u_int32_t); 720 sc->sc_indir_size = FSS_BTOCL(sc, len)+1; 721 sc->sc_clnext = sc->sc_indir_size; 722 sc->sc_indir_cur = 0; 723 724 if ((error = fss_softc_alloc(sc)) != 0) 725 goto bad; 726 727 /* 728 * Activate the snapshot. 729 */ 730 731 if ((error = vfs_suspend(sc->sc_mount, 0)) != 0) 732 goto bad; 733 734 microtime(&sc->sc_time); 735 736 if (error == 0) 737 error = fscow_establish(sc->sc_mount, 738 fss_copy_on_write, sc); 739 if (error == 0) 740 sc->sc_flags |= FSS_ACTIVE; 741 742 vfs_resume(sc->sc_mount); 743 744 if (error != 0) 745 goto bad; 746 747 #ifdef DEBUG 748 printf("fss%d: %s snapshot active\n", sc->sc_unit, sc->sc_mntname); 749 printf("fss%d: %u clusters of %u, %u cache slots, %u indir clusters\n", 750 sc->sc_unit, sc->sc_clcount, FSS_CLSIZE(sc), 751 sc->sc_cache_size, sc->sc_indir_size); 752 #endif 753 754 return 0; 755 756 bad: 757 fss_softc_free(sc); 758 if (sc->sc_bs_vp != NULL) { 759 if (sc->sc_flags & FSS_PERSISTENT) 760 vn_close(sc->sc_bs_vp, FREAD, l->l_cred); 761 else 762 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 763 } 764 sc->sc_bs_vp = NULL; 765 766 return error; 767 } 768 769 /* 770 * Delete a snapshot. 771 */ 772 static int 773 fss_delete_snapshot(struct fss_softc *sc, struct lwp *l) 774 { 775 int s; 776 777 if ((sc->sc_flags & FSS_PERSISTENT) == 0) 778 fscow_disestablish(sc->sc_mount, fss_copy_on_write, sc); 779 780 FSS_LOCK(sc, s); 781 sc->sc_flags &= ~(FSS_ACTIVE|FSS_ERROR); 782 sc->sc_mount = NULL; 783 sc->sc_bdev = NODEV; 784 FSS_UNLOCK(sc, s); 785 786 fss_softc_free(sc); 787 if (sc->sc_flags & FSS_PERSISTENT) 788 vn_close(sc->sc_bs_vp, FREAD, l->l_cred); 789 else 790 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 791 sc->sc_bs_vp = NULL; 792 sc->sc_flags &= ~FSS_PERSISTENT; 793 794 FSS_STAT_CLEAR(sc); 795 796 return 0; 797 } 798 799 /* 800 * A read from the snapshotted block device has completed. 801 */ 802 static void 803 fss_cluster_iodone(struct buf *bp) 804 { 805 int s; 806 struct fss_cache *scp = bp->b_private; 807 808 KASSERT(bp->b_vp == NULL); 809 810 FSS_LOCK(scp->fc_softc, s); 811 812 if (bp->b_error != 0) 813 fss_error(scp->fc_softc, "fs read error %d", bp->b_error); 814 815 if (--scp->fc_xfercount == 0) 816 wakeup(&scp->fc_data); 817 818 FSS_UNLOCK(scp->fc_softc, s); 819 820 putiobuf(bp); 821 } 822 823 /* 824 * Read a cluster from the snapshotted block device to the cache. 825 */ 826 static void 827 fss_read_cluster(struct fss_softc *sc, u_int32_t cl) 828 { 829 int s, todo, len; 830 char *addr; 831 daddr_t dblk; 832 struct buf *bp; 833 struct fss_cache *scp, *scl; 834 835 /* 836 * Get a free cache slot. 837 */ 838 scl = sc->sc_cache+sc->sc_cache_size; 839 840 FSS_LOCK(sc, s); 841 842 restart: 843 if (isset(sc->sc_copied, cl) || !FSS_ISVALID(sc)) { 844 FSS_UNLOCK(sc, s); 845 return; 846 } 847 848 for (scp = sc->sc_cache; scp < scl; scp++) 849 if (scp->fc_type != FSS_CACHE_FREE && 850 scp->fc_cluster == cl) { 851 ltsleep(&scp->fc_type, PRIBIO, "cowwait2", 0, 852 &sc->sc_slock); 853 goto restart; 854 } 855 856 for (scp = sc->sc_cache; scp < scl; scp++) 857 if (scp->fc_type == FSS_CACHE_FREE) { 858 scp->fc_type = FSS_CACHE_BUSY; 859 scp->fc_cluster = cl; 860 break; 861 } 862 if (scp >= scl) { 863 FSS_STAT_INC(sc, cow_cache_full); 864 ltsleep(&sc->sc_cache, PRIBIO, "cowwait3", 0, &sc->sc_slock); 865 goto restart; 866 } 867 868 FSS_UNLOCK(sc, s); 869 870 /* 871 * Start the read. 872 */ 873 FSS_STAT_INC(sc, cow_copied); 874 875 dblk = btodb(FSS_CLTOB(sc, cl)); 876 addr = scp->fc_data; 877 if (cl == sc->sc_clcount-1) { 878 todo = sc->sc_clresid; 879 memset((char *)addr + todo, 0, FSS_CLSIZE(sc) - todo); 880 } else 881 todo = FSS_CLSIZE(sc); 882 while (todo > 0) { 883 len = todo; 884 if (len > MAXPHYS) 885 len = MAXPHYS; 886 887 bp = getiobuf(NULL, true); 888 bp->b_flags = B_READ; 889 bp->b_bcount = len; 890 bp->b_bufsize = bp->b_bcount; 891 bp->b_error = 0; 892 bp->b_data = addr; 893 bp->b_blkno = dblk; 894 bp->b_proc = NULL; 895 bp->b_dev = sc->sc_bdev; 896 bp->b_private = scp; 897 bp->b_iodone = fss_cluster_iodone; 898 SET(bp->b_cflags, BC_BUSY); /* mark buffer busy */ 899 900 bdev_strategy(bp); 901 902 FSS_LOCK(sc, s); 903 scp->fc_xfercount++; 904 FSS_UNLOCK(sc, s); 905 906 dblk += btodb(len); 907 addr += len; 908 todo -= len; 909 } 910 911 /* 912 * Wait for all read requests to complete. 913 */ 914 FSS_LOCK(sc, s); 915 while (scp->fc_xfercount > 0) 916 ltsleep(&scp->fc_data, PRIBIO, "cowwait", 0, &sc->sc_slock); 917 918 scp->fc_type = FSS_CACHE_VALID; 919 setbit(sc->sc_copied, scp->fc_cluster); 920 FSS_UNLOCK(sc, s); 921 922 wakeup(&sc->sc_bs_lwp); 923 } 924 925 /* 926 * Read/write clusters from/to backing store. 927 * For persistent snapshots must be called with cl == 0. off is the 928 * offset into the snapshot. 929 */ 930 static int 931 fss_bs_io(struct fss_softc *sc, fss_io_type rw, 932 u_int32_t cl, off_t off, int len, void *data) 933 { 934 int error; 935 936 off += FSS_CLTOB(sc, cl); 937 938 vn_lock(sc->sc_bs_vp, LK_EXCLUSIVE|LK_RETRY); 939 940 error = vn_rdwr((rw == FSS_READ ? UIO_READ : UIO_WRITE), sc->sc_bs_vp, 941 data, len, off, UIO_SYSSPACE, IO_UNIT|IO_NODELOCKED, 942 sc->sc_bs_lwp->l_cred, NULL, NULL); 943 if (error == 0) { 944 mutex_enter(&sc->sc_bs_vp->v_interlock); 945 error = VOP_PUTPAGES(sc->sc_bs_vp, trunc_page(off), 946 round_page(off+len), PGO_CLEANIT|PGO_SYNCIO|PGO_FREE); 947 } 948 949 VOP_UNLOCK(sc->sc_bs_vp, 0); 950 951 return error; 952 } 953 954 /* 955 * Get a pointer to the indirect slot for this cluster. 956 */ 957 static u_int32_t * 958 fss_bs_indir(struct fss_softc *sc, u_int32_t cl) 959 { 960 u_int32_t icl; 961 int ioff; 962 963 icl = cl/(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 964 ioff = cl%(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 965 966 if (sc->sc_indir_cur == icl) 967 return &sc->sc_indir_data[ioff]; 968 969 if (sc->sc_indir_dirty) { 970 FSS_STAT_INC(sc, indir_write); 971 if (fss_bs_io(sc, FSS_WRITE, sc->sc_indir_cur, 0, 972 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 973 return NULL; 974 setbit(sc->sc_indir_valid, sc->sc_indir_cur); 975 } 976 977 sc->sc_indir_dirty = 0; 978 sc->sc_indir_cur = icl; 979 980 if (isset(sc->sc_indir_valid, sc->sc_indir_cur)) { 981 FSS_STAT_INC(sc, indir_read); 982 if (fss_bs_io(sc, FSS_READ, sc->sc_indir_cur, 0, 983 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 984 return NULL; 985 } else 986 memset(sc->sc_indir_data, 0, FSS_CLSIZE(sc)); 987 988 return &sc->sc_indir_data[ioff]; 989 } 990 991 /* 992 * The kernel thread (one for every active snapshot). 993 * 994 * After wakeup it cleans the cache and runs the I/O requests. 995 */ 996 static void 997 fss_bs_thread(void *arg) 998 { 999 int error, len, nfreed, nio, s; 1000 long off; 1001 char *addr; 1002 u_int32_t c, cl, ch, *indirp; 1003 struct buf *bp, *nbp; 1004 struct fss_softc *sc; 1005 struct fss_cache *scp, *scl; 1006 1007 sc = arg; 1008 1009 scl = sc->sc_cache+sc->sc_cache_size; 1010 1011 nfreed = nio = 1; /* Dont sleep the first time */ 1012 1013 FSS_LOCK(sc, s); 1014 1015 for (;;) { 1016 if (nfreed == 0 && nio == 0) 1017 ltsleep(&sc->sc_bs_lwp, PVM-1, "fssbs", 0, 1018 &sc->sc_slock); 1019 1020 if ((sc->sc_flags & FSS_BS_THREAD) == 0) { 1021 sc->sc_bs_lwp = NULL; 1022 wakeup(&sc->sc_bs_lwp); 1023 1024 FSS_UNLOCK(sc, s); 1025 1026 #ifdef FSS_STATISTICS 1027 if ((sc->sc_flags & FSS_PERSISTENT) == 0) { 1028 printf("fss%d: cow called %" PRId64 " times," 1029 " copied %" PRId64 " clusters," 1030 " cache full %" PRId64 " times\n", 1031 sc->sc_unit, 1032 FSS_STAT_VAL(sc, cow_calls), 1033 FSS_STAT_VAL(sc, cow_copied), 1034 FSS_STAT_VAL(sc, cow_cache_full)); 1035 printf("fss%d: %" PRId64 " indir reads," 1036 " %" PRId64 " indir writes\n", 1037 sc->sc_unit, 1038 FSS_STAT_VAL(sc, indir_read), 1039 FSS_STAT_VAL(sc, indir_write)); 1040 } 1041 #endif /* FSS_STATISTICS */ 1042 kthread_exit(0); 1043 } 1044 1045 /* 1046 * Process I/O requests (persistent) 1047 */ 1048 1049 if (sc->sc_flags & FSS_PERSISTENT) { 1050 nfreed = nio = 0; 1051 1052 if ((bp = BUFQ_GET(sc->sc_bufq)) == NULL) 1053 continue; 1054 1055 nio++; 1056 1057 if (FSS_ISVALID(sc)) { 1058 FSS_UNLOCK(sc, s); 1059 1060 error = fss_bs_io(sc, FSS_READ, 0, 1061 dbtob(bp->b_blkno), bp->b_bcount, 1062 bp->b_data); 1063 1064 FSS_LOCK(sc, s); 1065 } else 1066 error = ENXIO; 1067 1068 if (error) { 1069 bp->b_error = error; 1070 bp->b_resid = bp->b_bcount; 1071 } else 1072 bp->b_resid = 0; 1073 1074 biodone(bp); 1075 1076 continue; 1077 } 1078 1079 /* 1080 * Clean the cache 1081 */ 1082 nfreed = 0; 1083 for (scp = sc->sc_cache; scp < scl; scp++) { 1084 if (scp->fc_type != FSS_CACHE_VALID) 1085 continue; 1086 1087 FSS_UNLOCK(sc, s); 1088 1089 indirp = fss_bs_indir(sc, scp->fc_cluster); 1090 if (indirp != NULL) { 1091 error = fss_bs_io(sc, FSS_WRITE, sc->sc_clnext, 1092 0, FSS_CLSIZE(sc), scp->fc_data); 1093 } else 1094 error = EIO; 1095 1096 FSS_LOCK(sc, s); 1097 1098 if (error == 0) { 1099 *indirp = sc->sc_clnext++; 1100 sc->sc_indir_dirty = 1; 1101 } else 1102 fss_error(sc, "write bs error %d", error); 1103 1104 scp->fc_type = FSS_CACHE_FREE; 1105 nfreed++; 1106 wakeup(&scp->fc_type); 1107 } 1108 1109 if (nfreed) 1110 wakeup(&sc->sc_cache); 1111 1112 /* 1113 * Process I/O requests 1114 */ 1115 nio = 0; 1116 1117 if ((bp = BUFQ_GET(sc->sc_bufq)) == NULL) 1118 continue; 1119 1120 nio++; 1121 1122 if (!FSS_ISVALID(sc)) { 1123 bp->b_error = ENXIO; 1124 bp->b_resid = bp->b_bcount; 1125 biodone(bp); 1126 continue; 1127 } 1128 1129 /* 1130 * First read from the snapshotted block device. 1131 * XXX Split to only read those parts that have not 1132 * been saved to backing store? 1133 */ 1134 1135 FSS_UNLOCK(sc, s); 1136 1137 nbp = getiobuf(NULL, true); 1138 nbp->b_flags = B_READ; 1139 nbp->b_bcount = bp->b_bcount; 1140 nbp->b_bufsize = bp->b_bcount; 1141 nbp->b_error = 0; 1142 nbp->b_data = bp->b_data; 1143 nbp->b_blkno = bp->b_blkno; 1144 nbp->b_proc = bp->b_proc; 1145 nbp->b_dev = sc->sc_bdev; 1146 SET(nbp->b_cflags, BC_BUSY); /* mark buffer busy */ 1147 1148 bdev_strategy(nbp); 1149 1150 if (biowait(nbp) != 0) { 1151 bp->b_resid = bp->b_bcount; 1152 bp->b_error = nbp->b_error; 1153 biodone(bp); 1154 putiobuf(nbp); 1155 FSS_LOCK(sc, s); 1156 continue; 1157 } 1158 1159 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 1160 off = FSS_CLOFF(sc, dbtob(bp->b_blkno)); 1161 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 1162 bp->b_resid = bp->b_bcount; 1163 addr = bp->b_data; 1164 putiobuf(nbp); 1165 1166 FSS_LOCK(sc, s); 1167 1168 /* 1169 * Replace those parts that have been saved to backing store. 1170 */ 1171 1172 for (c = cl; c <= ch; 1173 c++, off = 0, bp->b_resid -= len, addr += len) { 1174 len = FSS_CLSIZE(sc)-off; 1175 if (len > bp->b_resid) 1176 len = bp->b_resid; 1177 1178 if (isclr(sc->sc_copied, c)) 1179 continue; 1180 1181 FSS_UNLOCK(sc, s); 1182 1183 indirp = fss_bs_indir(sc, c); 1184 1185 FSS_LOCK(sc, s); 1186 1187 if (indirp == NULL || *indirp == 0) { 1188 /* 1189 * Not on backing store. Either in cache 1190 * or hole in the snapshotted block device. 1191 */ 1192 for (scp = sc->sc_cache; scp < scl; scp++) 1193 if (scp->fc_type == FSS_CACHE_VALID && 1194 scp->fc_cluster == c) 1195 break; 1196 if (scp < scl) 1197 memcpy(addr, (char *)scp->fc_data+off, len); 1198 else 1199 memset(addr, 0, len); 1200 continue; 1201 } 1202 /* 1203 * Read from backing store. 1204 */ 1205 1206 FSS_UNLOCK(sc, s); 1207 1208 if ((error = fss_bs_io(sc, FSS_READ, *indirp, 1209 off, len, addr)) != 0) { 1210 bp->b_resid = bp->b_bcount; 1211 bp->b_error = error; 1212 FSS_LOCK(sc, s); 1213 break; 1214 } 1215 1216 FSS_LOCK(sc, s); 1217 1218 } 1219 1220 biodone(bp); 1221 } 1222 } 1223
Indexes created Wed Nov 11 00:26:22 UTC 2026