fss.c revision 1.30
1 /* $NetBSD: fss.c,v 1.30 2007/01/19 14:49:09 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 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the NetBSD 21 * Foundation, Inc. and its contributors. 22 * 4. Neither the name of The NetBSD Foundation nor the names of its 23 * contributors may be used to endorse or promote products derived 24 * from this software without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 36 * POSSIBILITY OF SUCH DAMAGE. 37 */ 38 39 /* 40 * File system snapshot disk driver. 41 * 42 * Block/character interface to the snapshot of a mounted file system. 43 */ 44 45 #include <sys/cdefs.h> 46 __KERNEL_RCSID(0, "$NetBSD: fss.c,v 1.30 2007/01/19 14:49:09 hannken Exp $"); 47 48 #include "fss.h" 49 50 #include <sys/param.h> 51 #include <sys/systm.h> 52 #include <sys/namei.h> 53 #include <sys/proc.h> 54 #include <sys/errno.h> 55 #include <sys/buf.h> 56 #include <sys/malloc.h> 57 #include <sys/ioctl.h> 58 #include <sys/disklabel.h> 59 #include <sys/device.h> 60 #include <sys/disk.h> 61 #include <sys/stat.h> 62 #include <sys/mount.h> 63 #include <sys/vnode.h> 64 #include <sys/file.h> 65 #include <sys/uio.h> 66 #include <sys/conf.h> 67 #include <sys/kthread.h> 68 #include <sys/fstrans.h> 69 70 #include <miscfs/specfs/specdev.h> 71 72 #include <dev/fssvar.h> 73 74 #include <machine/stdarg.h> 75 76 #ifdef DEBUG 77 #define FSS_STATISTICS 78 #endif 79 80 #ifdef FSS_STATISTICS 81 struct fss_stat { 82 u_int64_t cow_calls; 83 u_int64_t cow_copied; 84 u_int64_t cow_cache_full; 85 u_int64_t indir_read; 86 u_int64_t indir_write; 87 }; 88 89 static struct fss_stat fss_stat[NFSS]; 90 91 #define FSS_STAT_INC(sc, field) \ 92 do { \ 93 fss_stat[sc->sc_unit].field++; \ 94 } while (0) 95 #define FSS_STAT_SET(sc, field, value) \ 96 do { \ 97 fss_stat[sc->sc_unit].field = value; \ 98 } while (0) 99 #define FSS_STAT_ADD(sc, field, value) \ 100 do { \ 101 fss_stat[sc->sc_unit].field += value; \ 102 } while (0) 103 #define FSS_STAT_VAL(sc, field) fss_stat[sc->sc_unit].field 104 #define FSS_STAT_CLEAR(sc) \ 105 do { \ 106 memset(&fss_stat[sc->sc_unit], 0, \ 107 sizeof(struct fss_stat)); \ 108 } while (0) 109 #else /* FSS_STATISTICS */ 110 #define FSS_STAT_INC(sc, field) 111 #define FSS_STAT_SET(sc, field, value) 112 #define FSS_STAT_ADD(sc, field, value) 113 #define FSS_STAT_CLEAR(sc) 114 #endif /* FSS_STATISTICS */ 115 116 static struct fss_softc fss_softc[NFSS]; 117 118 void fssattach(int); 119 120 dev_type_open(fss_open); 121 dev_type_close(fss_close); 122 dev_type_read(fss_read); 123 dev_type_write(fss_write); 124 dev_type_ioctl(fss_ioctl); 125 dev_type_strategy(fss_strategy); 126 dev_type_dump(fss_dump); 127 dev_type_size(fss_size); 128 129 static int fss_copy_on_write(void *, struct buf *); 130 static inline void fss_error(struct fss_softc *, const char *, ...); 131 static int fss_create_files(struct fss_softc *, struct fss_set *, 132 off_t *, struct lwp *); 133 static int fss_create_snapshot(struct fss_softc *, struct fss_set *, 134 struct lwp *); 135 static int fss_delete_snapshot(struct fss_softc *, struct lwp *); 136 static int fss_softc_alloc(struct fss_softc *); 137 static void fss_softc_free(struct fss_softc *); 138 static void fss_cluster_iodone(struct buf *); 139 static void fss_read_cluster(struct fss_softc *, u_int32_t); 140 static void fss_bs_thread(void *); 141 static int fss_bs_io(struct fss_softc *, fss_io_type, 142 u_int32_t, off_t, int, caddr_t); 143 static u_int32_t *fss_bs_indir(struct fss_softc *, u_int32_t); 144 145 const struct bdevsw fss_bdevsw = { 146 fss_open, fss_close, fss_strategy, fss_ioctl, 147 fss_dump, fss_size, D_DISK 148 }; 149 150 const struct cdevsw fss_cdevsw = { 151 fss_open, fss_close, fss_read, fss_write, fss_ioctl, 152 nostop, notty, nopoll, nommap, nokqfilter, D_DISK 153 }; 154 155 void 156 fssattach(int num) 157 { 158 int i; 159 struct fss_softc *sc; 160 161 for (i = 0; i < NFSS; i++) { 162 sc = &fss_softc[i]; 163 sc->sc_unit = i; 164 sc->sc_bdev = NODEV; 165 simple_lock_init(&sc->sc_slock); 166 lockinit(&sc->sc_lock, PRIBIO, "fsslock", 0, 0); 167 bufq_alloc(&sc->sc_bufq, "fcfs", 0); 168 } 169 } 170 171 int 172 fss_open(dev_t dev, int flags, int mode, struct lwp *l) 173 { 174 int s, mflag; 175 struct fss_softc *sc; 176 177 mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN); 178 179 if ((sc = FSS_DEV_TO_SOFTC(dev)) == NULL) 180 return ENODEV; 181 182 FSS_LOCK(sc, s); 183 184 sc->sc_flags |= mflag; 185 186 FSS_UNLOCK(sc, s); 187 188 return 0; 189 } 190 191 int 192 fss_close(dev_t dev, int flags, int mode, struct lwp *l) 193 { 194 int s, mflag, error; 195 struct fss_softc *sc; 196 197 mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN); 198 199 if ((sc = FSS_DEV_TO_SOFTC(dev)) == NULL) 200 return ENODEV; 201 202 FSS_LOCK(sc, s); 203 204 if ((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) == mflag) { 205 if ((sc->sc_uflags & FSS_UNCONFIG_ON_CLOSE) != 0 && 206 (sc->sc_flags & FSS_ACTIVE) != 0) { 207 FSS_UNLOCK(sc, s); 208 error = fss_ioctl(dev, FSSIOCCLR, NULL, FWRITE, l); 209 if (error) 210 return error; 211 FSS_LOCK(sc, s); 212 } 213 sc->sc_uflags &= ~FSS_UNCONFIG_ON_CLOSE; 214 } 215 216 sc->sc_flags &= ~mflag; 217 218 FSS_UNLOCK(sc, s); 219 220 return 0; 221 } 222 223 void 224 fss_strategy(struct buf *bp) 225 { 226 int s; 227 struct fss_softc *sc; 228 229 sc = FSS_DEV_TO_SOFTC(bp->b_dev); 230 231 FSS_LOCK(sc, s); 232 233 if ((bp->b_flags & B_READ) != B_READ || 234 sc == NULL || !FSS_ISVALID(sc)) { 235 236 FSS_UNLOCK(sc, s); 237 238 bp->b_error = (sc == NULL ? ENODEV : EROFS); 239 bp->b_flags |= B_ERROR; 240 bp->b_resid = bp->b_bcount; 241 biodone(bp); 242 return; 243 } 244 245 bp->b_rawblkno = bp->b_blkno; 246 BUFQ_PUT(sc->sc_bufq, bp); 247 wakeup(&sc->sc_bs_proc); 248 249 FSS_UNLOCK(sc, s); 250 } 251 252 int 253 fss_read(dev_t dev, struct uio *uio, int flags) 254 { 255 return physio(fss_strategy, NULL, dev, B_READ, minphys, uio); 256 } 257 258 int 259 fss_write(dev_t dev, struct uio *uio, int flags) 260 { 261 return physio(fss_strategy, NULL, dev, B_WRITE, minphys, uio); 262 } 263 264 int 265 fss_ioctl(dev_t dev, u_long cmd, caddr_t data, int flag, struct lwp *l) 266 { 267 int error; 268 struct fss_softc *sc; 269 struct fss_set *fss = (struct fss_set *)data; 270 struct fss_get *fsg = (struct fss_get *)data; 271 272 if ((sc = FSS_DEV_TO_SOFTC(dev)) == NULL) 273 return ENODEV; 274 275 switch (cmd) { 276 case FSSIOCSET: 277 lockmgr(&sc->sc_lock, LK_EXCLUSIVE, NULL); 278 if ((flag & FWRITE) == 0) 279 error = EPERM; 280 else if ((sc->sc_flags & FSS_ACTIVE) != 0) 281 error = EBUSY; 282 else 283 error = fss_create_snapshot(sc, fss, l); 284 lockmgr(&sc->sc_lock, LK_RELEASE, NULL); 285 break; 286 287 case FSSIOCCLR: 288 lockmgr(&sc->sc_lock, LK_EXCLUSIVE, NULL); 289 if ((flag & FWRITE) == 0) 290 error = EPERM; 291 else if ((sc->sc_flags & FSS_ACTIVE) == 0) 292 error = ENXIO; 293 else 294 error = fss_delete_snapshot(sc, l); 295 lockmgr(&sc->sc_lock, LK_RELEASE, NULL); 296 break; 297 298 case FSSIOCGET: 299 lockmgr(&sc->sc_lock, LK_EXCLUSIVE, NULL); 300 switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) { 301 case FSS_ACTIVE: 302 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 303 fsg->fsg_csize = FSS_CLSIZE(sc); 304 fsg->fsg_time = sc->sc_time; 305 fsg->fsg_mount_size = sc->sc_clcount; 306 fsg->fsg_bs_size = sc->sc_clnext; 307 error = 0; 308 break; 309 case FSS_PERSISTENT | FSS_ACTIVE: 310 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 311 fsg->fsg_csize = 0; 312 fsg->fsg_time = sc->sc_time; 313 fsg->fsg_mount_size = 0; 314 fsg->fsg_bs_size = 0; 315 error = 0; 316 break; 317 default: 318 error = ENXIO; 319 break; 320 } 321 lockmgr(&sc->sc_lock, LK_RELEASE, NULL); 322 break; 323 324 case FSSIOFSET: 325 sc->sc_uflags = *(int *)data; 326 error = 0; 327 break; 328 329 case FSSIOFGET: 330 *(int *)data = sc->sc_uflags; 331 error = 0; 332 break; 333 334 default: 335 error = EINVAL; 336 break; 337 } 338 339 return error; 340 } 341 342 int 343 fss_size(dev_t dev) 344 { 345 return -1; 346 } 347 348 int 349 fss_dump(dev_t dev, daddr_t blkno, caddr_t va, 350 size_t size) 351 { 352 return EROFS; 353 } 354 355 /* 356 * An error occurred reading or writing the snapshot or backing store. 357 * If it is the first error log to console. 358 * The caller holds the simplelock. 359 */ 360 static inline void 361 fss_error(struct fss_softc *sc, const char *fmt, ...) 362 { 363 va_list ap; 364 365 if ((sc->sc_flags & (FSS_ACTIVE|FSS_ERROR)) == FSS_ACTIVE) { 366 va_start(ap, fmt); 367 printf("fss%d: snapshot invalid: ", sc->sc_unit); 368 vprintf(fmt, ap); 369 printf("\n"); 370 va_end(ap); 371 } 372 if ((sc->sc_flags & FSS_ACTIVE) == FSS_ACTIVE) 373 sc->sc_flags |= FSS_ERROR; 374 } 375 376 /* 377 * Allocate the variable sized parts of the softc and 378 * fork the kernel thread. 379 * 380 * The fields sc_clcount, sc_clshift, sc_cache_size and sc_indir_size 381 * must be initialized. 382 */ 383 static int 384 fss_softc_alloc(struct fss_softc *sc) 385 { 386 int i, len, error; 387 388 len = (sc->sc_clcount+NBBY-1)/NBBY; 389 sc->sc_copied = malloc(len, M_TEMP, M_ZERO|M_WAITOK|M_CANFAIL); 390 if (sc->sc_copied == NULL) 391 return(ENOMEM); 392 393 len = sc->sc_cache_size*sizeof(struct fss_cache); 394 sc->sc_cache = malloc(len, M_TEMP, M_ZERO|M_WAITOK|M_CANFAIL); 395 if (sc->sc_cache == NULL) 396 return(ENOMEM); 397 398 len = FSS_CLSIZE(sc); 399 for (i = 0; i < sc->sc_cache_size; i++) { 400 sc->sc_cache[i].fc_type = FSS_CACHE_FREE; 401 sc->sc_cache[i].fc_softc = sc; 402 sc->sc_cache[i].fc_xfercount = 0; 403 sc->sc_cache[i].fc_data = malloc(len, M_TEMP, 404 M_WAITOK|M_CANFAIL); 405 if (sc->sc_cache[i].fc_data == NULL) 406 return(ENOMEM); 407 } 408 409 len = (sc->sc_indir_size+NBBY-1)/NBBY; 410 sc->sc_indir_valid = malloc(len, M_TEMP, M_ZERO|M_WAITOK|M_CANFAIL); 411 if (sc->sc_indir_valid == NULL) 412 return(ENOMEM); 413 414 len = FSS_CLSIZE(sc); 415 sc->sc_indir_data = malloc(len, M_TEMP, M_ZERO|M_WAITOK|M_CANFAIL); 416 if (sc->sc_indir_data == NULL) 417 return(ENOMEM); 418 419 if ((error = kthread_create1(fss_bs_thread, sc, &sc->sc_bs_proc, 420 "fssbs%d", sc->sc_unit)) != 0) 421 return error; 422 423 sc->sc_flags |= FSS_BS_THREAD; 424 return 0; 425 } 426 427 /* 428 * Free the variable sized parts of the softc. 429 */ 430 static void 431 fss_softc_free(struct fss_softc *sc) 432 { 433 int s, i; 434 435 if ((sc->sc_flags & FSS_BS_THREAD) != 0) { 436 FSS_LOCK(sc, s); 437 sc->sc_flags &= ~FSS_BS_THREAD; 438 wakeup(&sc->sc_bs_proc); 439 while (sc->sc_bs_proc != NULL) 440 ltsleep(&sc->sc_bs_proc, PRIBIO, "fssthread", 0, 441 &sc->sc_slock); 442 FSS_UNLOCK(sc, s); 443 } 444 445 if (sc->sc_copied != NULL) 446 free(sc->sc_copied, M_TEMP); 447 sc->sc_copied = NULL; 448 449 if (sc->sc_cache != NULL) { 450 for (i = 0; i < sc->sc_cache_size; i++) 451 if (sc->sc_cache[i].fc_data != NULL) 452 free(sc->sc_cache[i].fc_data, M_TEMP); 453 free(sc->sc_cache, M_TEMP); 454 } 455 sc->sc_cache = NULL; 456 457 if (sc->sc_indir_valid != NULL) 458 free(sc->sc_indir_valid, M_TEMP); 459 sc->sc_indir_valid = NULL; 460 461 if (sc->sc_indir_data != NULL) 462 free(sc->sc_indir_data, M_TEMP); 463 sc->sc_indir_data = NULL; 464 } 465 466 /* 467 * Check if an unmount is ok. If forced, set this snapshot into ERROR state. 468 */ 469 int 470 fss_umount_hook(struct mount *mp, int forced) 471 { 472 int i, s; 473 474 for (i = 0; i < NFSS; i++) { 475 FSS_LOCK(&fss_softc[i], s); 476 if ((fss_softc[i].sc_flags & FSS_ACTIVE) != 0 && 477 fss_softc[i].sc_mount == mp) { 478 if (forced) 479 fss_error(&fss_softc[i], "forced unmount"); 480 else { 481 FSS_UNLOCK(&fss_softc[i], s); 482 return EBUSY; 483 } 484 } 485 FSS_UNLOCK(&fss_softc[i], s); 486 } 487 488 return 0; 489 } 490 491 /* 492 * A buffer is written to the snapshotted block device. Copy to 493 * backing store if needed. 494 */ 495 static int 496 fss_copy_on_write(void *v, struct buf *bp) 497 { 498 int s; 499 u_int32_t cl, ch, c; 500 struct fss_softc *sc = v; 501 502 FSS_LOCK(sc, s); 503 if (!FSS_ISVALID(sc)) { 504 FSS_UNLOCK(sc, s); 505 return 0; 506 } 507 508 FSS_UNLOCK(sc, s); 509 510 FSS_STAT_INC(sc, cow_calls); 511 512 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 513 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 514 515 for (c = cl; c <= ch; c++) 516 fss_read_cluster(sc, c); 517 518 return 0; 519 } 520 521 /* 522 * Lookup and open needed files. 523 * 524 * For file system internal snapshot initializes sc_mntname, sc_mount, 525 * sc_bs_vp and sc_time. 526 * 527 * Otherwise returns dev and size of the underlying block device. 528 * Initializes sc_mntname, sc_mount_vp, sc_bdev, sc_bs_vp and sc_mount 529 */ 530 static int 531 fss_create_files(struct fss_softc *sc, struct fss_set *fss, 532 off_t *bsize, struct lwp *l) 533 { 534 int error, bits, fsbsize; 535 struct timespec ts; 536 struct partinfo dpart; 537 struct vattr va; 538 struct nameidata nd; 539 540 /* 541 * Get the mounted file system. 542 */ 543 544 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_mount, l); 545 if ((error = namei(&nd)) != 0) 546 return error; 547 548 if ((nd.ni_vp->v_flag & VROOT) != VROOT) { 549 vrele(nd.ni_vp); 550 return EINVAL; 551 } 552 553 sc->sc_mount = nd.ni_vp->v_mount; 554 memcpy(sc->sc_mntname, sc->sc_mount->mnt_stat.f_mntonname, MNAMELEN); 555 556 vrele(nd.ni_vp); 557 558 /* 559 * Check for file system internal snapshot. 560 */ 561 562 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_bstore, l); 563 if ((error = namei(&nd)) != 0) 564 return error; 565 566 if (nd.ni_vp->v_type == VREG && nd.ni_vp->v_mount == sc->sc_mount) { 567 vrele(nd.ni_vp); 568 sc->sc_flags |= FSS_PERSISTENT; 569 570 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_bstore, l); 571 if ((error = vn_open(&nd, FREAD, 0)) != 0) 572 return error; 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 vrele(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, l); 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 error = VOP_IOCTL(nd.ni_vp, DIOCGPART, &dpart, FREAD, 613 l->l_cred, l); 614 if (error) { 615 vrele(nd.ni_vp); 616 return error; 617 } 618 619 sc->sc_mount_vp = nd.ni_vp; 620 sc->sc_bdev = nd.ni_vp->v_rdev; 621 *bsize = (off_t)dpart.disklab->d_secsize*dpart.part->p_size; 622 vrele(nd.ni_vp); 623 624 /* 625 * Get the backing store 626 */ 627 628 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_bstore, l); 629 if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) 630 return error; 631 VOP_UNLOCK(nd.ni_vp, 0); 632 633 sc->sc_bs_vp = nd.ni_vp; 634 635 if (nd.ni_vp->v_type != VREG && nd.ni_vp->v_type != VCHR) 636 return EINVAL; 637 638 if (sc->sc_bs_vp->v_type == VREG) { 639 error = VOP_GETATTR(sc->sc_bs_vp, &va, l->l_cred, l); 640 if (error != 0) 641 return error; 642 sc->sc_bs_size = va.va_size; 643 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 644 if (fsbsize & (fsbsize-1)) /* No power of two */ 645 return EINVAL; 646 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < 32; 647 sc->sc_bs_bshift++) 648 if (FSS_FSBSIZE(sc) == fsbsize) 649 break; 650 if (sc->sc_bs_bshift >= 32) 651 return EINVAL; 652 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 653 } else { 654 sc->sc_bs_bshift = DEV_BSHIFT; 655 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 656 } 657 658 /* 659 * As all IO to from/to the backing store goes through 660 * VOP_STRATEGY() clean the buffer cache to prevent 661 * cache incoherencies. 662 */ 663 if ((error = vinvalbuf(sc->sc_bs_vp, V_SAVE, l->l_cred, l, 0, 0)) != 0) 664 return error; 665 666 return 0; 667 } 668 669 /* 670 * Create a snapshot. 671 */ 672 static int 673 fss_create_snapshot(struct fss_softc *sc, struct fss_set *fss, struct lwp *l) 674 { 675 int len, error; 676 u_int32_t csize; 677 off_t bsize; 678 679 bsize = 0; /* XXX gcc */ 680 681 /* 682 * Open needed files. 683 */ 684 if ((error = fss_create_files(sc, fss, &bsize, l)) != 0) 685 goto bad; 686 687 if (sc->sc_flags & FSS_PERSISTENT) { 688 fss_softc_alloc(sc); 689 sc->sc_flags |= FSS_ACTIVE; 690 return 0; 691 } 692 693 /* 694 * Set cluster size. Must be a power of two and 695 * a multiple of backing store block size. 696 */ 697 if (fss->fss_csize <= 0) 698 csize = MAXPHYS; 699 else 700 csize = fss->fss_csize; 701 if (bsize/csize > FSS_CLUSTER_MAX) 702 csize = bsize/FSS_CLUSTER_MAX+1; 703 704 for (sc->sc_clshift = sc->sc_bs_bshift; sc->sc_clshift < 32; 705 sc->sc_clshift++) 706 if (FSS_CLSIZE(sc) >= csize) 707 break; 708 if (sc->sc_clshift >= 32) { 709 error = EINVAL; 710 goto bad; 711 } 712 sc->sc_clmask = FSS_CLSIZE(sc)-1; 713 714 /* 715 * Set number of cache slots. 716 */ 717 if (FSS_CLSIZE(sc) <= 8192) 718 sc->sc_cache_size = 32; 719 else if (FSS_CLSIZE(sc) <= 65536) 720 sc->sc_cache_size = 8; 721 else 722 sc->sc_cache_size = 4; 723 724 /* 725 * Set number of clusters and size of last cluster. 726 */ 727 sc->sc_clcount = FSS_BTOCL(sc, bsize-1)+1; 728 sc->sc_clresid = FSS_CLOFF(sc, bsize-1)+1; 729 730 /* 731 * Set size of indirect table. 732 */ 733 len = sc->sc_clcount*sizeof(u_int32_t); 734 sc->sc_indir_size = FSS_BTOCL(sc, len)+1; 735 sc->sc_clnext = sc->sc_indir_size; 736 sc->sc_indir_cur = 0; 737 738 if ((error = fss_softc_alloc(sc)) != 0) 739 goto bad; 740 741 /* 742 * Activate the snapshot. 743 */ 744 745 if ((error = vfs_suspend(sc->sc_mount, 0)) != 0) 746 goto bad; 747 748 microtime(&sc->sc_time); 749 750 if (error == 0) 751 error = vn_cow_establish(sc->sc_mount_vp, 752 fss_copy_on_write, sc); 753 if (error == 0) 754 sc->sc_flags |= FSS_ACTIVE; 755 756 vfs_resume(sc->sc_mount); 757 758 if (error != 0) 759 goto bad; 760 761 #ifdef DEBUG 762 printf("fss%d: %s snapshot active\n", sc->sc_unit, sc->sc_mntname); 763 printf("fss%d: %u clusters of %u, %u cache slots, %u indir clusters\n", 764 sc->sc_unit, sc->sc_clcount, FSS_CLSIZE(sc), 765 sc->sc_cache_size, sc->sc_indir_size); 766 #endif 767 768 return 0; 769 770 bad: 771 fss_softc_free(sc); 772 if (sc->sc_bs_vp != NULL) { 773 if (sc->sc_flags & FSS_PERSISTENT) 774 vn_close(sc->sc_bs_vp, FREAD, l->l_cred, l); 775 else 776 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred, l); 777 } 778 sc->sc_bs_vp = NULL; 779 780 return error; 781 } 782 783 /* 784 * Delete a snapshot. 785 */ 786 static int 787 fss_delete_snapshot(struct fss_softc *sc, struct lwp *l) 788 { 789 int s; 790 791 if ((sc->sc_flags & FSS_PERSISTENT) == 0) 792 vn_cow_disestablish(sc->sc_mount_vp, fss_copy_on_write, sc); 793 794 FSS_LOCK(sc, s); 795 sc->sc_flags &= ~(FSS_ACTIVE|FSS_ERROR); 796 sc->sc_mount = NULL; 797 sc->sc_bdev = NODEV; 798 FSS_UNLOCK(sc, s); 799 800 fss_softc_free(sc); 801 if (sc->sc_flags & FSS_PERSISTENT) 802 vn_close(sc->sc_bs_vp, FREAD, l->l_cred, l); 803 else 804 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred, l); 805 sc->sc_bs_vp = NULL; 806 sc->sc_flags &= ~FSS_PERSISTENT; 807 808 FSS_STAT_CLEAR(sc); 809 810 return 0; 811 } 812 813 /* 814 * A read from the snapshotted block device has completed. 815 */ 816 static void 817 fss_cluster_iodone(struct buf *bp) 818 { 819 int s; 820 struct fss_cache *scp = bp->b_private; 821 822 KASSERT(bp->b_vp == NULL); 823 824 FSS_LOCK(scp->fc_softc, s); 825 826 if (bp->b_flags & B_ERROR) 827 fss_error(scp->fc_softc, "fs read error %d", bp->b_error); 828 829 if (--scp->fc_xfercount == 0) 830 wakeup(&scp->fc_data); 831 832 FSS_UNLOCK(scp->fc_softc, s); 833 834 putiobuf(bp); 835 } 836 837 /* 838 * Read a cluster from the snapshotted block device to the cache. 839 */ 840 static void 841 fss_read_cluster(struct fss_softc *sc, u_int32_t cl) 842 { 843 int s, todo, len; 844 caddr_t addr; 845 daddr_t dblk; 846 struct buf *bp; 847 struct fss_cache *scp, *scl; 848 849 /* 850 * Get a free cache slot. 851 */ 852 scl = sc->sc_cache+sc->sc_cache_size; 853 854 FSS_LOCK(sc, s); 855 856 restart: 857 if (isset(sc->sc_copied, cl) || !FSS_ISVALID(sc)) { 858 FSS_UNLOCK(sc, s); 859 return; 860 } 861 862 for (scp = sc->sc_cache; scp < scl; scp++) 863 if (scp->fc_type != FSS_CACHE_FREE && 864 scp->fc_cluster == cl) { 865 ltsleep(&scp->fc_type, PRIBIO, "cowwait2", 0, 866 &sc->sc_slock); 867 goto restart; 868 } 869 870 for (scp = sc->sc_cache; scp < scl; scp++) 871 if (scp->fc_type == FSS_CACHE_FREE) { 872 scp->fc_type = FSS_CACHE_BUSY; 873 scp->fc_cluster = cl; 874 break; 875 } 876 if (scp >= scl) { 877 FSS_STAT_INC(sc, cow_cache_full); 878 ltsleep(&sc->sc_cache, PRIBIO, "cowwait3", 0, &sc->sc_slock); 879 goto restart; 880 } 881 882 FSS_UNLOCK(sc, s); 883 884 /* 885 * Start the read. 886 */ 887 FSS_STAT_INC(sc, cow_copied); 888 889 dblk = btodb(FSS_CLTOB(sc, cl)); 890 addr = scp->fc_data; 891 if (cl == sc->sc_clcount-1) { 892 todo = sc->sc_clresid; 893 memset(addr+todo, 0, FSS_CLSIZE(sc)-todo); 894 } else 895 todo = FSS_CLSIZE(sc); 896 while (todo > 0) { 897 len = todo; 898 if (len > MAXPHYS) 899 len = MAXPHYS; 900 901 bp = getiobuf(); 902 bp->b_flags = B_READ|B_CALL; 903 bp->b_bcount = len; 904 bp->b_bufsize = bp->b_bcount; 905 bp->b_error = 0; 906 bp->b_data = addr; 907 bp->b_blkno = dblk; 908 bp->b_proc = NULL; 909 bp->b_dev = sc->sc_bdev; 910 bp->b_vp = NULLVP; 911 bp->b_private = scp; 912 bp->b_iodone = fss_cluster_iodone; 913 914 DEV_STRATEGY(bp); 915 916 FSS_LOCK(sc, s); 917 scp->fc_xfercount++; 918 FSS_UNLOCK(sc, s); 919 920 dblk += btodb(len); 921 addr += len; 922 todo -= len; 923 } 924 925 /* 926 * Wait for all read requests to complete. 927 */ 928 FSS_LOCK(sc, s); 929 while (scp->fc_xfercount > 0) 930 ltsleep(&scp->fc_data, PRIBIO, "cowwait", 0, &sc->sc_slock); 931 932 scp->fc_type = FSS_CACHE_VALID; 933 setbit(sc->sc_copied, scp->fc_cluster); 934 FSS_UNLOCK(sc, s); 935 936 wakeup(&sc->sc_bs_proc); 937 } 938 939 /* 940 * Read/write clusters from/to backing store. 941 * For persistent snapshots must be called with cl == 0. off is the 942 * offset into the snapshot. 943 */ 944 static int 945 fss_bs_io(struct fss_softc *sc, fss_io_type rw, 946 u_int32_t cl, off_t off, int len, caddr_t data) 947 { 948 int error; 949 950 off += FSS_CLTOB(sc, cl); 951 952 vn_lock(sc->sc_bs_vp, LK_EXCLUSIVE|LK_RETRY); 953 954 error = vn_rdwr((rw == FSS_READ ? UIO_READ : UIO_WRITE), sc->sc_bs_vp, 955 data, len, off, UIO_SYSSPACE, IO_UNIT|IO_NODELOCKED, 956 sc->sc_bs_proc->p_cred, NULL, NULL); 957 if (error == 0) { 958 simple_lock(&sc->sc_bs_vp->v_interlock); 959 error = VOP_PUTPAGES(sc->sc_bs_vp, trunc_page(off), 960 round_page(off+len), PGO_CLEANIT|PGO_SYNCIO|PGO_FREE); 961 } 962 963 VOP_UNLOCK(sc->sc_bs_vp, 0); 964 965 return error; 966 } 967 968 /* 969 * Get a pointer to the indirect slot for this cluster. 970 */ 971 static u_int32_t * 972 fss_bs_indir(struct fss_softc *sc, u_int32_t cl) 973 { 974 u_int32_t icl; 975 int ioff; 976 977 icl = cl/(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 978 ioff = cl%(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 979 980 if (sc->sc_indir_cur == icl) 981 return &sc->sc_indir_data[ioff]; 982 983 if (sc->sc_indir_dirty) { 984 FSS_STAT_INC(sc, indir_write); 985 if (fss_bs_io(sc, FSS_WRITE, sc->sc_indir_cur, 0, 986 FSS_CLSIZE(sc), (caddr_t)sc->sc_indir_data) != 0) 987 return NULL; 988 setbit(sc->sc_indir_valid, sc->sc_indir_cur); 989 } 990 991 sc->sc_indir_dirty = 0; 992 sc->sc_indir_cur = icl; 993 994 if (isset(sc->sc_indir_valid, sc->sc_indir_cur)) { 995 FSS_STAT_INC(sc, indir_read); 996 if (fss_bs_io(sc, FSS_READ, sc->sc_indir_cur, 0, 997 FSS_CLSIZE(sc), (caddr_t)sc->sc_indir_data) != 0) 998 return NULL; 999 } else 1000 memset(sc->sc_indir_data, 0, FSS_CLSIZE(sc)); 1001 1002 return &sc->sc_indir_data[ioff]; 1003 } 1004 1005 /* 1006 * The kernel thread (one for every active snapshot). 1007 * 1008 * After wakeup it cleans the cache and runs the I/O requests. 1009 */ 1010 static void 1011 fss_bs_thread(void *arg) 1012 { 1013 int error, len, nfreed, nio, s; 1014 long off; 1015 caddr_t addr; 1016 u_int32_t c, cl, ch, *indirp; 1017 struct buf *bp, *nbp; 1018 struct fss_softc *sc; 1019 struct fss_cache *scp, *scl; 1020 1021 sc = arg; 1022 1023 scl = sc->sc_cache+sc->sc_cache_size; 1024 1025 nbp = getiobuf(); 1026 1027 nfreed = nio = 1; /* Dont sleep the first time */ 1028 1029 FSS_LOCK(sc, s); 1030 1031 for (;;) { 1032 if (nfreed == 0 && nio == 0) 1033 ltsleep(&sc->sc_bs_proc, PVM-1, "fssbs", 0, 1034 &sc->sc_slock); 1035 1036 if ((sc->sc_flags & FSS_BS_THREAD) == 0) { 1037 sc->sc_bs_proc = NULL; 1038 wakeup(&sc->sc_bs_proc); 1039 1040 FSS_UNLOCK(sc, s); 1041 1042 putiobuf(nbp); 1043 #ifdef FSS_STATISTICS 1044 if ((sc->sc_flags & FSS_PERSISTENT) == 0) { 1045 printf("fss%d: cow called %" PRId64 " times," 1046 " copied %" PRId64 " clusters," 1047 " cache full %" PRId64 " times\n", 1048 sc->sc_unit, 1049 FSS_STAT_VAL(sc, cow_calls), 1050 FSS_STAT_VAL(sc, cow_copied), 1051 FSS_STAT_VAL(sc, cow_cache_full)); 1052 printf("fss%d: %" PRId64 " indir reads," 1053 " %" PRId64 " indir writes\n", 1054 sc->sc_unit, 1055 FSS_STAT_VAL(sc, indir_read), 1056 FSS_STAT_VAL(sc, indir_write)); 1057 } 1058 #endif /* FSS_STATISTICS */ 1059 kthread_exit(0); 1060 } 1061 1062 /* 1063 * Process I/O requests (persistent) 1064 */ 1065 1066 if (sc->sc_flags & FSS_PERSISTENT) { 1067 nfreed = nio = 0; 1068 1069 if ((bp = BUFQ_GET(sc->sc_bufq)) == NULL) 1070 continue; 1071 1072 nio++; 1073 1074 if (FSS_ISVALID(sc)) { 1075 FSS_UNLOCK(sc, s); 1076 1077 error = fss_bs_io(sc, FSS_READ, 0, 1078 dbtob(bp->b_blkno), bp->b_bcount, 1079 bp->b_data); 1080 1081 FSS_LOCK(sc, s); 1082 } else 1083 error = ENXIO; 1084 1085 if (error) { 1086 bp->b_error = error; 1087 bp->b_flags |= B_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_flags |= B_ERROR; 1143 bp->b_resid = bp->b_bcount; 1144 biodone(bp); 1145 continue; 1146 } 1147 1148 /* 1149 * First read from the snapshotted block device. 1150 * XXX Split to only read those parts that have not 1151 * been saved to backing store? 1152 */ 1153 1154 FSS_UNLOCK(sc, s); 1155 1156 BUF_INIT(nbp); 1157 nbp->b_flags = B_READ; 1158 nbp->b_bcount = bp->b_bcount; 1159 nbp->b_bufsize = bp->b_bcount; 1160 nbp->b_error = 0; 1161 nbp->b_data = bp->b_data; 1162 nbp->b_blkno = bp->b_blkno; 1163 nbp->b_proc = bp->b_proc; 1164 nbp->b_dev = sc->sc_bdev; 1165 nbp->b_vp = NULLVP; 1166 1167 DEV_STRATEGY(nbp); 1168 1169 if (biowait(nbp) != 0) { 1170 bp->b_resid = bp->b_bcount; 1171 bp->b_error = nbp->b_error; 1172 bp->b_flags |= B_ERROR; 1173 biodone(bp); 1174 FSS_LOCK(sc, s); 1175 continue; 1176 } 1177 1178 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 1179 off = FSS_CLOFF(sc, dbtob(bp->b_blkno)); 1180 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 1181 bp->b_resid = bp->b_bcount; 1182 addr = bp->b_data; 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, 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 bp->b_flags |= B_ERROR; 1231 FSS_LOCK(sc, s); 1232 break; 1233 } 1234 1235 FSS_LOCK(sc, s); 1236 1237 } 1238 1239 biodone(bp); 1240 } 1241 } 1242
Indexes created Wed Nov 11 00:26:22 UTC 2026