fss.c revision 1.52
1 /* $NetBSD: fss.c,v 1.52 2008/08/26 13:04:55 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 1.52 2008/08/26 13:04:55 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 void fss_cluster_iodone(struct buf *); 135 static void fss_read_cluster(struct fss_softc *, u_int32_t); 136 static void fss_bs_thread(void *); 137 static int fss_bs_io(struct fss_softc *, fss_io_type, 138 u_int32_t, off_t, int, void *); 139 static u_int32_t *fss_bs_indir(struct fss_softc *, u_int32_t); 140 141 const struct bdevsw fss_bdevsw = { 142 fss_open, fss_close, fss_strategy, fss_ioctl, 143 fss_dump, fss_size, D_DISK 144 }; 145 146 const struct cdevsw fss_cdevsw = { 147 fss_open, fss_close, fss_read, fss_write, fss_ioctl, 148 nostop, notty, nopoll, nommap, nokqfilter, D_DISK 149 }; 150 151 void 152 fssattach(int num) 153 { 154 int i; 155 struct fss_softc *sc; 156 157 for (i = 0; i < NFSS; i++) { 158 sc = &fss_softc[i]; 159 sc->sc_unit = i; 160 sc->sc_bdev = NODEV; 161 simple_lock_init(&sc->sc_slock); 162 mutex_init(&sc->sc_lock, MUTEX_DEFAULT, IPL_NONE); 163 bufq_alloc(&sc->sc_bufq, "fcfs", 0); 164 } 165 } 166 167 int 168 fss_open(dev_t dev, int flags, int mode, struct lwp *l) 169 { 170 int s, mflag; 171 struct fss_softc *sc; 172 173 mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN); 174 175 if ((sc = FSS_DEV_TO_SOFTC(dev)) == NULL) 176 return ENODEV; 177 178 FSS_LOCK(sc, s); 179 180 sc->sc_flags |= mflag; 181 182 FSS_UNLOCK(sc, s); 183 184 return 0; 185 } 186 187 int 188 fss_close(dev_t dev, int flags, int mode, struct lwp *l) 189 { 190 int s, mflag, error; 191 struct fss_softc *sc; 192 193 mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN); 194 195 if ((sc = FSS_DEV_TO_SOFTC(dev)) == NULL) 196 return ENODEV; 197 198 FSS_LOCK(sc, s); 199 200 if ((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) == mflag) { 201 if ((sc->sc_uflags & FSS_UNCONFIG_ON_CLOSE) != 0 && 202 (sc->sc_flags & FSS_ACTIVE) != 0) { 203 FSS_UNLOCK(sc, s); 204 error = fss_ioctl(dev, FSSIOCCLR, NULL, FWRITE, l); 205 if (error) 206 return error; 207 FSS_LOCK(sc, s); 208 } 209 sc->sc_uflags &= ~FSS_UNCONFIG_ON_CLOSE; 210 } 211 212 sc->sc_flags &= ~mflag; 213 214 FSS_UNLOCK(sc, s); 215 216 return 0; 217 } 218 219 void 220 fss_strategy(struct buf *bp) 221 { 222 int s; 223 struct fss_softc *sc; 224 225 sc = FSS_DEV_TO_SOFTC(bp->b_dev); 226 227 FSS_LOCK(sc, s); 228 229 if ((bp->b_flags & B_READ) != B_READ || 230 sc == NULL || !FSS_ISVALID(sc)) { 231 232 FSS_UNLOCK(sc, s); 233 234 bp->b_error = (sc == NULL ? ENODEV : EROFS); 235 bp->b_resid = bp->b_bcount; 236 biodone(bp); 237 return; 238 } 239 240 bp->b_rawblkno = bp->b_blkno; 241 BUFQ_PUT(sc->sc_bufq, bp); 242 wakeup(&sc->sc_bs_lwp); 243 244 FSS_UNLOCK(sc, s); 245 } 246 247 int 248 fss_read(dev_t dev, struct uio *uio, int flags) 249 { 250 return physio(fss_strategy, NULL, dev, B_READ, minphys, uio); 251 } 252 253 int 254 fss_write(dev_t dev, struct uio *uio, int flags) 255 { 256 return physio(fss_strategy, NULL, dev, B_WRITE, minphys, uio); 257 } 258 259 int 260 fss_ioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 261 { 262 int error; 263 struct fss_softc *sc; 264 struct fss_set *fss = (struct fss_set *)data; 265 struct fss_get *fsg = (struct fss_get *)data; 266 267 if ((sc = FSS_DEV_TO_SOFTC(dev)) == NULL) 268 return ENODEV; 269 270 switch (cmd) { 271 case FSSIOCSET: 272 mutex_enter(&sc->sc_lock); 273 if ((flag & FWRITE) == 0) 274 error = EPERM; 275 else if ((sc->sc_flags & FSS_ACTIVE) != 0) 276 error = EBUSY; 277 else 278 error = fss_create_snapshot(sc, fss, l); 279 mutex_exit(&sc->sc_lock); 280 break; 281 282 case FSSIOCCLR: 283 mutex_enter(&sc->sc_lock); 284 if ((flag & FWRITE) == 0) 285 error = EPERM; 286 else if ((sc->sc_flags & FSS_ACTIVE) == 0) 287 error = ENXIO; 288 else 289 error = fss_delete_snapshot(sc, l); 290 mutex_exit(&sc->sc_lock); 291 break; 292 293 case FSSIOCGET: 294 mutex_enter(&sc->sc_lock); 295 switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) { 296 case FSS_ACTIVE: 297 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 298 fsg->fsg_csize = FSS_CLSIZE(sc); 299 fsg->fsg_time = sc->sc_time; 300 fsg->fsg_mount_size = sc->sc_clcount; 301 fsg->fsg_bs_size = sc->sc_clnext; 302 error = 0; 303 break; 304 case FSS_PERSISTENT | FSS_ACTIVE: 305 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 306 fsg->fsg_csize = 0; 307 fsg->fsg_time = sc->sc_time; 308 fsg->fsg_mount_size = 0; 309 fsg->fsg_bs_size = 0; 310 error = 0; 311 break; 312 default: 313 error = ENXIO; 314 break; 315 } 316 mutex_exit(&sc->sc_lock); 317 break; 318 319 case FSSIOFSET: 320 sc->sc_uflags = *(int *)data; 321 error = 0; 322 break; 323 324 case FSSIOFGET: 325 *(int *)data = sc->sc_uflags; 326 error = 0; 327 break; 328 329 default: 330 error = EINVAL; 331 break; 332 } 333 334 return error; 335 } 336 337 int 338 fss_size(dev_t dev) 339 { 340 return -1; 341 } 342 343 int 344 fss_dump(dev_t dev, daddr_t blkno, void *va, 345 size_t size) 346 { 347 return EROFS; 348 } 349 350 /* 351 * An error occurred reading or writing the snapshot or backing store. 352 * If it is the first error log to console. 353 * The caller holds the simplelock. 354 */ 355 static inline void 356 fss_error(struct fss_softc *sc, const char *fmt, ...) 357 { 358 va_list ap; 359 360 if ((sc->sc_flags & (FSS_ACTIVE|FSS_ERROR)) == FSS_ACTIVE) { 361 va_start(ap, fmt); 362 printf("fss%d: snapshot invalid: ", sc->sc_unit); 363 vprintf(fmt, ap); 364 printf("\n"); 365 va_end(ap); 366 } 367 if ((sc->sc_flags & FSS_ACTIVE) == FSS_ACTIVE) 368 sc->sc_flags |= FSS_ERROR; 369 } 370 371 /* 372 * Allocate the variable sized parts of the softc and 373 * fork the kernel thread. 374 * 375 * The fields sc_clcount, sc_clshift, sc_cache_size and sc_indir_size 376 * must be initialized. 377 */ 378 static int 379 fss_softc_alloc(struct fss_softc *sc) 380 { 381 int i, len, error; 382 383 len = (sc->sc_clcount+NBBY-1)/NBBY; 384 sc->sc_copied = malloc(len, M_TEMP, M_ZERO|M_WAITOK|M_CANFAIL); 385 if (sc->sc_copied == NULL) 386 return(ENOMEM); 387 388 len = sc->sc_cache_size*sizeof(struct fss_cache); 389 sc->sc_cache = malloc(len, M_TEMP, M_ZERO|M_WAITOK|M_CANFAIL); 390 if (sc->sc_cache == NULL) 391 return(ENOMEM); 392 393 len = FSS_CLSIZE(sc); 394 for (i = 0; i < sc->sc_cache_size; i++) { 395 sc->sc_cache[i].fc_type = FSS_CACHE_FREE; 396 sc->sc_cache[i].fc_softc = sc; 397 sc->sc_cache[i].fc_xfercount = 0; 398 sc->sc_cache[i].fc_data = malloc(len, M_TEMP, 399 M_WAITOK|M_CANFAIL); 400 if (sc->sc_cache[i].fc_data == NULL) 401 return(ENOMEM); 402 } 403 404 len = (sc->sc_indir_size+NBBY-1)/NBBY; 405 sc->sc_indir_valid = malloc(len, M_TEMP, M_ZERO|M_WAITOK|M_CANFAIL); 406 if (sc->sc_indir_valid == NULL) 407 return(ENOMEM); 408 409 len = FSS_CLSIZE(sc); 410 sc->sc_indir_data = malloc(len, M_TEMP, M_ZERO|M_WAITOK|M_CANFAIL); 411 if (sc->sc_indir_data == NULL) 412 return(ENOMEM); 413 414 if ((error = kthread_create(PRI_BIO, 0, NULL, fss_bs_thread, sc, 415 &sc->sc_bs_lwp, "fssbs%d", sc->sc_unit)) != 0) 416 return error; 417 418 sc->sc_flags |= FSS_BS_THREAD; 419 return 0; 420 } 421 422 /* 423 * Free the variable sized parts of the softc. 424 */ 425 static void 426 fss_softc_free(struct fss_softc *sc) 427 { 428 int s, i; 429 430 if ((sc->sc_flags & FSS_BS_THREAD) != 0) { 431 FSS_LOCK(sc, s); 432 sc->sc_flags &= ~FSS_BS_THREAD; 433 wakeup(&sc->sc_bs_lwp); 434 while (sc->sc_bs_lwp != NULL) 435 ltsleep(&sc->sc_bs_lwp, PRIBIO, "fssthread", 0, 436 &sc->sc_slock); 437 FSS_UNLOCK(sc, s); 438 } 439 440 if (sc->sc_copied != NULL) 441 free(sc->sc_copied, M_TEMP); 442 sc->sc_copied = NULL; 443 444 if (sc->sc_cache != NULL) { 445 for (i = 0; i < sc->sc_cache_size; i++) 446 if (sc->sc_cache[i].fc_data != NULL) 447 free(sc->sc_cache[i].fc_data, M_TEMP); 448 free(sc->sc_cache, M_TEMP); 449 } 450 sc->sc_cache = NULL; 451 452 if (sc->sc_indir_valid != NULL) 453 free(sc->sc_indir_valid, M_TEMP); 454 sc->sc_indir_valid = NULL; 455 456 if (sc->sc_indir_data != NULL) 457 free(sc->sc_indir_data, M_TEMP); 458 sc->sc_indir_data = NULL; 459 } 460 461 /* 462 * Check if an unmount is ok. If forced, set this snapshot into ERROR state. 463 */ 464 int 465 fss_umount_hook(struct mount *mp, int forced) 466 { 467 int i, s; 468 469 for (i = 0; i < NFSS; i++) { 470 FSS_LOCK(&fss_softc[i], s); 471 if ((fss_softc[i].sc_flags & FSS_ACTIVE) != 0 && 472 fss_softc[i].sc_mount == mp) { 473 if (forced) 474 fss_error(&fss_softc[i], "forced unmount"); 475 else { 476 FSS_UNLOCK(&fss_softc[i], s); 477 return EBUSY; 478 } 479 } 480 FSS_UNLOCK(&fss_softc[i], s); 481 } 482 483 return 0; 484 } 485 486 /* 487 * A buffer is written to the snapshotted block device. Copy to 488 * backing store if needed. 489 */ 490 static int 491 fss_copy_on_write(void *v, struct buf *bp, bool data_valid) 492 { 493 int s; 494 u_int32_t cl, ch, c; 495 struct fss_softc *sc = v; 496 497 FSS_LOCK(sc, s); 498 if (!FSS_ISVALID(sc)) { 499 FSS_UNLOCK(sc, s); 500 return 0; 501 } 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 if (curlwp == uvm.pagedaemon_lwp) { 509 for (c = cl; c <= ch; c++) 510 if (isclr(sc->sc_copied, c)) { 511 FSS_UNLOCK(sc, s); 512 513 return ENOMEM; 514 } 515 } 516 517 FSS_UNLOCK(sc, s); 518 519 for (c = cl; c <= ch; c++) 520 fss_read_cluster(sc, c); 521 522 return 0; 523 } 524 525 /* 526 * Lookup and open needed files. 527 * 528 * For file system internal snapshot initializes sc_mntname, sc_mount, 529 * sc_bs_vp and sc_time. 530 * 531 * Otherwise returns dev and size of the underlying block device. 532 * Initializes sc_mntname, sc_mount, sc_bdev, sc_bs_vp and sc_mount 533 */ 534 static int 535 fss_create_files(struct fss_softc *sc, struct fss_set *fss, 536 off_t *bsize, struct lwp *l) 537 { 538 int error, bits, fsbsize; 539 const struct bdevsw *bdev; 540 struct timespec ts; 541 struct partinfo dpart; 542 struct vattr va; 543 struct nameidata nd; 544 545 /* 546 * Get the mounted file system. 547 */ 548 549 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_mount); 550 if ((error = namei(&nd)) != 0) 551 return error; 552 553 if ((nd.ni_vp->v_vflag & VV_ROOT) != VV_ROOT) { 554 vrele(nd.ni_vp); 555 return EINVAL; 556 } 557 558 sc->sc_mount = nd.ni_vp->v_mount; 559 memcpy(sc->sc_mntname, sc->sc_mount->mnt_stat.f_mntonname, MNAMELEN); 560 561 vrele(nd.ni_vp); 562 563 /* 564 * Check for file system internal snapshot. 565 */ 566 567 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_USERSPACE, fss->fss_bstore); 568 if ((error = namei(&nd)) != 0) 569 return error; 570 571 if (nd.ni_vp->v_type == VREG && nd.ni_vp->v_mount == sc->sc_mount) { 572 sc->sc_flags |= FSS_PERSISTENT; 573 sc->sc_bs_vp = nd.ni_vp; 574 575 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 576 bits = sizeof(sc->sc_bs_bshift)*NBBY; 577 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < bits; 578 sc->sc_bs_bshift++) 579 if (FSS_FSBSIZE(sc) == fsbsize) 580 break; 581 if (sc->sc_bs_bshift >= bits) { 582 VOP_UNLOCK(sc->sc_bs_vp, 0); 583 return EINVAL; 584 } 585 586 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 587 sc->sc_clshift = 0; 588 589 error = VFS_SNAPSHOT(sc->sc_mount, sc->sc_bs_vp, &ts); 590 TIMESPEC_TO_TIMEVAL(&sc->sc_time, &ts); 591 592 VOP_UNLOCK(sc->sc_bs_vp, 0); 593 594 return error; 595 } 596 vput(nd.ni_vp); 597 598 /* 599 * Get the block device it is mounted on. 600 */ 601 602 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, 603 sc->sc_mount->mnt_stat.f_mntfromname); 604 if ((error = namei(&nd)) != 0) 605 return error; 606 607 if (nd.ni_vp->v_type != VBLK) { 608 vrele(nd.ni_vp); 609 return EINVAL; 610 } 611 612 sc->sc_bdev = nd.ni_vp->v_rdev; 613 vrele(nd.ni_vp); 614 615 /* 616 * Get the block device size. 617 */ 618 619 bdev = bdevsw_lookup(sc->sc_bdev); 620 if (bdev == NULL) 621 return ENXIO; 622 error = (*bdev->d_ioctl)(sc->sc_bdev, DIOCGPART, &dpart, FREAD, l); 623 if (error) 624 return error; 625 626 *bsize = (off_t)dpart.disklab->d_secsize*dpart.part->p_size; 627 628 /* 629 * Get the backing store 630 */ 631 632 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_bstore); 633 if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) 634 return error; 635 VOP_UNLOCK(nd.ni_vp, 0); 636 637 sc->sc_bs_vp = nd.ni_vp; 638 639 if (nd.ni_vp->v_type != VREG && nd.ni_vp->v_type != VCHR) 640 return EINVAL; 641 642 if (sc->sc_bs_vp->v_type == VREG) { 643 error = VOP_GETATTR(sc->sc_bs_vp, &va, l->l_cred); 644 if (error != 0) 645 return error; 646 sc->sc_bs_size = va.va_size; 647 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 648 if (fsbsize & (fsbsize-1)) /* No power of two */ 649 return EINVAL; 650 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < 32; 651 sc->sc_bs_bshift++) 652 if (FSS_FSBSIZE(sc) == fsbsize) 653 break; 654 if (sc->sc_bs_bshift >= 32) 655 return EINVAL; 656 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 657 } else { 658 sc->sc_bs_bshift = DEV_BSHIFT; 659 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 660 } 661 662 /* 663 * As all IO to from/to the backing store goes through 664 * VOP_STRATEGY() clean the buffer cache to prevent 665 * cache incoherencies. 666 */ 667 if ((error = vinvalbuf(sc->sc_bs_vp, V_SAVE, l->l_cred, l, 0, 0)) != 0) 668 return error; 669 670 return 0; 671 } 672 673 /* 674 * Create a snapshot. 675 */ 676 static int 677 fss_create_snapshot(struct fss_softc *sc, struct fss_set *fss, struct lwp *l) 678 { 679 int len, error; 680 u_int32_t csize; 681 off_t bsize; 682 683 bsize = 0; /* XXX gcc */ 684 685 /* 686 * Open needed files. 687 */ 688 if ((error = fss_create_files(sc, fss, &bsize, l)) != 0) 689 goto bad; 690 691 if (sc->sc_flags & FSS_PERSISTENT) { 692 fss_softc_alloc(sc); 693 sc->sc_flags |= FSS_ACTIVE; 694 return 0; 695 } 696 697 /* 698 * Set cluster size. Must be a power of two and 699 * a multiple of backing store block size. 700 */ 701 if (fss->fss_csize <= 0) 702 csize = MAXPHYS; 703 else 704 csize = fss->fss_csize; 705 if (bsize/csize > FSS_CLUSTER_MAX) 706 csize = bsize/FSS_CLUSTER_MAX+1; 707 708 for (sc->sc_clshift = sc->sc_bs_bshift; sc->sc_clshift < 32; 709 sc->sc_clshift++) 710 if (FSS_CLSIZE(sc) >= csize) 711 break; 712 if (sc->sc_clshift >= 32) { 713 error = EINVAL; 714 goto bad; 715 } 716 sc->sc_clmask = FSS_CLSIZE(sc)-1; 717 718 /* 719 * Set number of cache slots. 720 */ 721 if (FSS_CLSIZE(sc) <= 8192) 722 sc->sc_cache_size = 32; 723 else if (FSS_CLSIZE(sc) <= 65536) 724 sc->sc_cache_size = 8; 725 else 726 sc->sc_cache_size = 4; 727 728 /* 729 * Set number of clusters and size of last cluster. 730 */ 731 sc->sc_clcount = FSS_BTOCL(sc, bsize-1)+1; 732 sc->sc_clresid = FSS_CLOFF(sc, bsize-1)+1; 733 734 /* 735 * Set size of indirect table. 736 */ 737 len = sc->sc_clcount*sizeof(u_int32_t); 738 sc->sc_indir_size = FSS_BTOCL(sc, len)+1; 739 sc->sc_clnext = sc->sc_indir_size; 740 sc->sc_indir_cur = 0; 741 742 if ((error = fss_softc_alloc(sc)) != 0) 743 goto bad; 744 745 /* 746 * Activate the snapshot. 747 */ 748 749 if ((error = vfs_suspend(sc->sc_mount, 0)) != 0) 750 goto bad; 751 752 microtime(&sc->sc_time); 753 754 if (error == 0) 755 error = fscow_establish(sc->sc_mount, 756 fss_copy_on_write, sc); 757 if (error == 0) 758 sc->sc_flags |= FSS_ACTIVE; 759 760 vfs_resume(sc->sc_mount); 761 762 if (error != 0) 763 goto bad; 764 765 #ifdef DEBUG 766 printf("fss%d: %s snapshot active\n", sc->sc_unit, sc->sc_mntname); 767 printf("fss%d: %u clusters of %u, %u cache slots, %u indir clusters\n", 768 sc->sc_unit, sc->sc_clcount, FSS_CLSIZE(sc), 769 sc->sc_cache_size, sc->sc_indir_size); 770 #endif 771 772 return 0; 773 774 bad: 775 fss_softc_free(sc); 776 if (sc->sc_bs_vp != NULL) { 777 if (sc->sc_flags & FSS_PERSISTENT) 778 vn_close(sc->sc_bs_vp, FREAD, l->l_cred); 779 else 780 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 781 } 782 sc->sc_bs_vp = NULL; 783 784 return error; 785 } 786 787 /* 788 * Delete a snapshot. 789 */ 790 static int 791 fss_delete_snapshot(struct fss_softc *sc, struct lwp *l) 792 { 793 int s; 794 795 if ((sc->sc_flags & FSS_PERSISTENT) == 0) 796 fscow_disestablish(sc->sc_mount, fss_copy_on_write, sc); 797 798 FSS_LOCK(sc, s); 799 sc->sc_flags &= ~(FSS_ACTIVE|FSS_ERROR); 800 sc->sc_mount = NULL; 801 sc->sc_bdev = NODEV; 802 FSS_UNLOCK(sc, s); 803 804 fss_softc_free(sc); 805 if (sc->sc_flags & FSS_PERSISTENT) 806 vn_close(sc->sc_bs_vp, FREAD, l->l_cred); 807 else 808 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 809 sc->sc_bs_vp = NULL; 810 sc->sc_flags &= ~FSS_PERSISTENT; 811 812 FSS_STAT_CLEAR(sc); 813 814 return 0; 815 } 816 817 /* 818 * A read from the snapshotted block device has completed. 819 */ 820 static void 821 fss_cluster_iodone(struct buf *bp) 822 { 823 int s; 824 struct fss_cache *scp = bp->b_private; 825 826 KASSERT(bp->b_vp == NULL); 827 828 FSS_LOCK(scp->fc_softc, s); 829 830 if (bp->b_error != 0) 831 fss_error(scp->fc_softc, "fs read error %d", bp->b_error); 832 833 if (--scp->fc_xfercount == 0) 834 wakeup(&scp->fc_data); 835 836 FSS_UNLOCK(scp->fc_softc, s); 837 838 putiobuf(bp); 839 } 840 841 /* 842 * Read a cluster from the snapshotted block device to the cache. 843 */ 844 static void 845 fss_read_cluster(struct fss_softc *sc, u_int32_t cl) 846 { 847 int s, todo, len; 848 char *addr; 849 daddr_t dblk; 850 struct buf *bp; 851 struct fss_cache *scp, *scl; 852 853 /* 854 * Get a free cache slot. 855 */ 856 scl = sc->sc_cache+sc->sc_cache_size; 857 858 FSS_LOCK(sc, s); 859 860 restart: 861 if (isset(sc->sc_copied, cl) || !FSS_ISVALID(sc)) { 862 FSS_UNLOCK(sc, s); 863 return; 864 } 865 866 for (scp = sc->sc_cache; scp < scl; scp++) 867 if (scp->fc_type != FSS_CACHE_FREE && 868 scp->fc_cluster == cl) { 869 ltsleep(&scp->fc_type, PRIBIO, "cowwait2", 0, 870 &sc->sc_slock); 871 goto restart; 872 } 873 874 for (scp = sc->sc_cache; scp < scl; scp++) 875 if (scp->fc_type == FSS_CACHE_FREE) { 876 scp->fc_type = FSS_CACHE_BUSY; 877 scp->fc_cluster = cl; 878 break; 879 } 880 if (scp >= scl) { 881 FSS_STAT_INC(sc, cow_cache_full); 882 ltsleep(&sc->sc_cache, PRIBIO, "cowwait3", 0, &sc->sc_slock); 883 goto restart; 884 } 885 886 FSS_UNLOCK(sc, s); 887 888 /* 889 * Start the read. 890 */ 891 FSS_STAT_INC(sc, cow_copied); 892 893 dblk = btodb(FSS_CLTOB(sc, cl)); 894 addr = scp->fc_data; 895 if (cl == sc->sc_clcount-1) { 896 todo = sc->sc_clresid; 897 memset((char *)addr + todo, 0, FSS_CLSIZE(sc) - todo); 898 } else 899 todo = FSS_CLSIZE(sc); 900 while (todo > 0) { 901 len = todo; 902 if (len > MAXPHYS) 903 len = MAXPHYS; 904 905 bp = getiobuf(NULL, true); 906 bp->b_flags = B_READ; 907 bp->b_bcount = len; 908 bp->b_bufsize = bp->b_bcount; 909 bp->b_error = 0; 910 bp->b_data = addr; 911 bp->b_blkno = dblk; 912 bp->b_proc = NULL; 913 bp->b_dev = sc->sc_bdev; 914 bp->b_private = scp; 915 bp->b_iodone = fss_cluster_iodone; 916 SET(bp->b_cflags, BC_BUSY); /* mark buffer busy */ 917 918 bdev_strategy(bp); 919 920 FSS_LOCK(sc, s); 921 scp->fc_xfercount++; 922 FSS_UNLOCK(sc, s); 923 924 dblk += btodb(len); 925 addr += len; 926 todo -= len; 927 } 928 929 /* 930 * Wait for all read requests to complete. 931 */ 932 FSS_LOCK(sc, s); 933 while (scp->fc_xfercount > 0) 934 ltsleep(&scp->fc_data, PRIBIO, "cowwait", 0, &sc->sc_slock); 935 936 scp->fc_type = FSS_CACHE_VALID; 937 setbit(sc->sc_copied, scp->fc_cluster); 938 FSS_UNLOCK(sc, s); 939 940 wakeup(&sc->sc_bs_lwp); 941 } 942 943 /* 944 * Read/write clusters from/to backing store. 945 * For persistent snapshots must be called with cl == 0. off is the 946 * offset into the snapshot. 947 */ 948 static int 949 fss_bs_io(struct fss_softc *sc, fss_io_type rw, 950 u_int32_t cl, off_t off, int len, void *data) 951 { 952 int error; 953 954 off += FSS_CLTOB(sc, cl); 955 956 vn_lock(sc->sc_bs_vp, LK_EXCLUSIVE|LK_RETRY); 957 958 error = vn_rdwr((rw == FSS_READ ? UIO_READ : UIO_WRITE), sc->sc_bs_vp, 959 data, len, off, UIO_SYSSPACE, IO_UNIT|IO_NODELOCKED, 960 sc->sc_bs_lwp->l_cred, NULL, NULL); 961 if (error == 0) { 962 mutex_enter(&sc->sc_bs_vp->v_interlock); 963 error = VOP_PUTPAGES(sc->sc_bs_vp, trunc_page(off), 964 round_page(off+len), PGO_CLEANIT|PGO_SYNCIO|PGO_FREE); 965 } 966 967 VOP_UNLOCK(sc->sc_bs_vp, 0); 968 969 return error; 970 } 971 972 /* 973 * Get a pointer to the indirect slot for this cluster. 974 */ 975 static u_int32_t * 976 fss_bs_indir(struct fss_softc *sc, u_int32_t cl) 977 { 978 u_int32_t icl; 979 int ioff; 980 981 icl = cl/(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 982 ioff = cl%(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 983 984 if (sc->sc_indir_cur == icl) 985 return &sc->sc_indir_data[ioff]; 986 987 if (sc->sc_indir_dirty) { 988 FSS_STAT_INC(sc, indir_write); 989 if (fss_bs_io(sc, FSS_WRITE, sc->sc_indir_cur, 0, 990 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 991 return NULL; 992 setbit(sc->sc_indir_valid, sc->sc_indir_cur); 993 } 994 995 sc->sc_indir_dirty = 0; 996 sc->sc_indir_cur = icl; 997 998 if (isset(sc->sc_indir_valid, sc->sc_indir_cur)) { 999 FSS_STAT_INC(sc, indir_read); 1000 if (fss_bs_io(sc, FSS_READ, sc->sc_indir_cur, 0, 1001 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 1002 return NULL; 1003 } else 1004 memset(sc->sc_indir_data, 0, FSS_CLSIZE(sc)); 1005 1006 return &sc->sc_indir_data[ioff]; 1007 } 1008 1009 /* 1010 * The kernel thread (one for every active snapshot). 1011 * 1012 * After wakeup it cleans the cache and runs the I/O requests. 1013 */ 1014 static void 1015 fss_bs_thread(void *arg) 1016 { 1017 int error, len, nfreed, nio, s; 1018 long off; 1019 char *addr; 1020 u_int32_t c, cl, ch, *indirp; 1021 struct buf *bp, *nbp; 1022 struct fss_softc *sc; 1023 struct fss_cache *scp, *scl; 1024 1025 sc = arg; 1026 1027 scl = sc->sc_cache+sc->sc_cache_size; 1028 1029 nfreed = nio = 1; /* Dont sleep the first time */ 1030 1031 FSS_LOCK(sc, s); 1032 1033 for (;;) { 1034 if (nfreed == 0 && nio == 0) 1035 ltsleep(&sc->sc_bs_lwp, PVM-1, "fssbs", 0, 1036 &sc->sc_slock); 1037 1038 if ((sc->sc_flags & FSS_BS_THREAD) == 0) { 1039 sc->sc_bs_lwp = NULL; 1040 wakeup(&sc->sc_bs_lwp); 1041 1042 FSS_UNLOCK(sc, s); 1043 1044 #ifdef FSS_STATISTICS 1045 if ((sc->sc_flags & FSS_PERSISTENT) == 0) { 1046 printf("fss%d: cow called %" PRId64 " times," 1047 " copied %" PRId64 " clusters," 1048 " cache full %" PRId64 " times\n", 1049 sc->sc_unit, 1050 FSS_STAT_VAL(sc, cow_calls), 1051 FSS_STAT_VAL(sc, cow_copied), 1052 FSS_STAT_VAL(sc, cow_cache_full)); 1053 printf("fss%d: %" PRId64 " indir reads," 1054 " %" PRId64 " indir writes\n", 1055 sc->sc_unit, 1056 FSS_STAT_VAL(sc, indir_read), 1057 FSS_STAT_VAL(sc, indir_write)); 1058 } 1059 #endif /* FSS_STATISTICS */ 1060 kthread_exit(0); 1061 } 1062 1063 /* 1064 * Process I/O requests (persistent) 1065 */ 1066 1067 if (sc->sc_flags & FSS_PERSISTENT) { 1068 nfreed = nio = 0; 1069 1070 if ((bp = BUFQ_GET(sc->sc_bufq)) == NULL) 1071 continue; 1072 1073 nio++; 1074 1075 if (FSS_ISVALID(sc)) { 1076 FSS_UNLOCK(sc, s); 1077 1078 error = fss_bs_io(sc, FSS_READ, 0, 1079 dbtob(bp->b_blkno), bp->b_bcount, 1080 bp->b_data); 1081 1082 FSS_LOCK(sc, s); 1083 } else 1084 error = ENXIO; 1085 1086 if (error) { 1087 bp->b_error = error; 1088 bp->b_resid = bp->b_bcount; 1089 } else 1090 bp->b_resid = 0; 1091 1092 biodone(bp); 1093 1094 continue; 1095 } 1096 1097 /* 1098 * Clean the cache 1099 */ 1100 nfreed = 0; 1101 for (scp = sc->sc_cache; scp < scl; scp++) { 1102 if (scp->fc_type != FSS_CACHE_VALID) 1103 continue; 1104 1105 FSS_UNLOCK(sc, s); 1106 1107 indirp = fss_bs_indir(sc, scp->fc_cluster); 1108 if (indirp != NULL) { 1109 error = fss_bs_io(sc, FSS_WRITE, sc->sc_clnext, 1110 0, FSS_CLSIZE(sc), scp->fc_data); 1111 } else 1112 error = EIO; 1113 1114 FSS_LOCK(sc, s); 1115 1116 if (error == 0) { 1117 *indirp = sc->sc_clnext++; 1118 sc->sc_indir_dirty = 1; 1119 } else 1120 fss_error(sc, "write bs error %d", error); 1121 1122 scp->fc_type = FSS_CACHE_FREE; 1123 nfreed++; 1124 wakeup(&scp->fc_type); 1125 } 1126 1127 if (nfreed) 1128 wakeup(&sc->sc_cache); 1129 1130 /* 1131 * Process I/O requests 1132 */ 1133 nio = 0; 1134 1135 if ((bp = BUFQ_GET(sc->sc_bufq)) == NULL) 1136 continue; 1137 1138 nio++; 1139 1140 if (!FSS_ISVALID(sc)) { 1141 bp->b_error = ENXIO; 1142 bp->b_resid = bp->b_bcount; 1143 biodone(bp); 1144 continue; 1145 } 1146 1147 /* 1148 * First read from the snapshotted block device. 1149 * XXX Split to only read those parts that have not 1150 * been saved to backing store? 1151 */ 1152 1153 FSS_UNLOCK(sc, s); 1154 1155 nbp = getiobuf(NULL, true); 1156 nbp->b_flags = B_READ; 1157 nbp->b_bcount = bp->b_bcount; 1158 nbp->b_bufsize = bp->b_bcount; 1159 nbp->b_error = 0; 1160 nbp->b_data = bp->b_data; 1161 nbp->b_blkno = bp->b_blkno; 1162 nbp->b_proc = bp->b_proc; 1163 nbp->b_dev = sc->sc_bdev; 1164 SET(nbp->b_cflags, BC_BUSY); /* mark buffer busy */ 1165 1166 bdev_strategy(nbp); 1167 1168 if (biowait(nbp) != 0) { 1169 bp->b_resid = bp->b_bcount; 1170 bp->b_error = nbp->b_error; 1171 biodone(bp); 1172 putiobuf(nbp); 1173 FSS_LOCK(sc, s); 1174 continue; 1175 } 1176 1177 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 1178 off = FSS_CLOFF(sc, dbtob(bp->b_blkno)); 1179 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 1180 bp->b_resid = bp->b_bcount; 1181 addr = bp->b_data; 1182 putiobuf(nbp); 1183 1184 FSS_LOCK(sc, s); 1185 1186 /* 1187 * Replace those parts that have been saved to backing store. 1188 */ 1189 1190 for (c = cl; c <= ch; 1191 c++, off = 0, bp->b_resid -= len, addr += len) { 1192 len = FSS_CLSIZE(sc)-off; 1193 if (len > bp->b_resid) 1194 len = bp->b_resid; 1195 1196 if (isclr(sc->sc_copied, c)) 1197 continue; 1198 1199 FSS_UNLOCK(sc, s); 1200 1201 indirp = fss_bs_indir(sc, c); 1202 1203 FSS_LOCK(sc, s); 1204 1205 if (indirp == NULL || *indirp == 0) { 1206 /* 1207 * Not on backing store. Either in cache 1208 * or hole in the snapshotted block device. 1209 */ 1210 for (scp = sc->sc_cache; scp < scl; scp++) 1211 if (scp->fc_type == FSS_CACHE_VALID && 1212 scp->fc_cluster == c) 1213 break; 1214 if (scp < scl) 1215 memcpy(addr, (char *)scp->fc_data+off, len); 1216 else 1217 memset(addr, 0, len); 1218 continue; 1219 } 1220 /* 1221 * Read from backing store. 1222 */ 1223 1224 FSS_UNLOCK(sc, s); 1225 1226 if ((error = fss_bs_io(sc, FSS_READ, *indirp, 1227 off, len, addr)) != 0) { 1228 bp->b_resid = bp->b_bcount; 1229 bp->b_error = error; 1230 FSS_LOCK(sc, s); 1231 break; 1232 } 1233 1234 FSS_LOCK(sc, s); 1235 1236 } 1237 1238 biodone(bp); 1239 } 1240 } 1241
Indexes created Wed Nov 11 00:26:22 UTC 2026