fss.c revision 1.58
1 /* $NetBSD: fss.c,v 1.58 2008/09/18 10:52:14 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 1.58 2008/09/18 10:52:14 hannken Exp $"); 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/namei.h> 44 #include <sys/proc.h> 45 #include <sys/errno.h> 46 #include <sys/malloc.h> 47 #include <sys/buf.h> 48 #include <sys/ioctl.h> 49 #include <sys/disklabel.h> 50 #include <sys/device.h> 51 #include <sys/disk.h> 52 #include <sys/stat.h> 53 #include <sys/mount.h> 54 #include <sys/vnode.h> 55 #include <sys/file.h> 56 #include <sys/uio.h> 57 #include <sys/conf.h> 58 #include <sys/kthread.h> 59 #include <sys/fstrans.h> 60 #include <sys/simplelock.h> 61 62 #include <miscfs/specfs/specdev.h> 63 64 #include <dev/fssvar.h> 65 66 #include <uvm/uvm.h> 67 68 void fssattach(int); 69 70 dev_type_open(fss_open); 71 dev_type_close(fss_close); 72 dev_type_read(fss_read); 73 dev_type_write(fss_write); 74 dev_type_ioctl(fss_ioctl); 75 dev_type_strategy(fss_strategy); 76 dev_type_dump(fss_dump); 77 dev_type_size(fss_size); 78 79 static void fss_unmount_hook(struct mount *); 80 static int fss_copy_on_write(void *, struct buf *, bool); 81 static inline void fss_error(struct fss_softc *, const char *); 82 static int fss_create_files(struct fss_softc *, struct fss_set *, 83 off_t *, struct lwp *); 84 static int fss_create_snapshot(struct fss_softc *, struct fss_set *, 85 struct lwp *); 86 static int fss_delete_snapshot(struct fss_softc *, struct lwp *); 87 static int fss_softc_alloc(struct fss_softc *); 88 static void fss_softc_free(struct fss_softc *); 89 static int fss_read_cluster(struct fss_softc *, u_int32_t); 90 static void fss_bs_thread(void *); 91 static int fss_bs_io(struct fss_softc *, fss_io_type, 92 u_int32_t, off_t, int, void *); 93 static u_int32_t *fss_bs_indir(struct fss_softc *, u_int32_t); 94 95 static kmutex_t fss_device_lock; /* Protect fss_num_attached. */ 96 static int fss_num_attached = 0; /* Number of attached devices. */ 97 static struct vfs_hooks fss_vfs_hooks = { 98 .vh_unmount = fss_unmount_hook 99 }; 100 101 const struct bdevsw fss_bdevsw = { 102 fss_open, fss_close, fss_strategy, fss_ioctl, 103 fss_dump, fss_size, D_DISK 104 }; 105 106 const struct cdevsw fss_cdevsw = { 107 fss_open, fss_close, fss_read, fss_write, fss_ioctl, 108 nostop, notty, nopoll, nommap, nokqfilter, D_DISK 109 }; 110 111 static int fss_match(device_t, cfdata_t, void *); 112 static void fss_attach(device_t, device_t, void *); 113 static int fss_detach(device_t, int); 114 115 CFATTACH_DECL_NEW(fss, sizeof(struct fss_softc), 116 fss_match, fss_attach, fss_detach, NULL); 117 extern struct cfdriver fss_cd; 118 119 void 120 fssattach(int num) 121 { 122 123 mutex_init(&fss_device_lock, MUTEX_DEFAULT, IPL_NONE); 124 if (config_cfattach_attach(fss_cd.cd_name, &fss_ca)) 125 aprint_error("%s: unable to register\n", fss_cd.cd_name); 126 } 127 128 static int 129 fss_match(device_t self, cfdata_t cfdata, void *aux) 130 { 131 return 1; 132 } 133 134 static void 135 fss_attach(device_t parent, device_t self, void *aux) 136 { 137 struct fss_softc *sc = device_private(self); 138 139 sc->sc_dev = self; 140 sc->sc_bdev = NODEV; 141 mutex_init(&sc->sc_slock, MUTEX_DEFAULT, IPL_NONE); 142 mutex_init(&sc->sc_lock, MUTEX_DEFAULT, IPL_NONE); 143 cv_init(&sc->sc_work_cv, "fssbs"); 144 cv_init(&sc->sc_cache_cv, "cowwait"); 145 bufq_alloc(&sc->sc_bufq, "fcfs", 0); 146 sc->sc_dkdev = malloc(sizeof(*sc->sc_dkdev), M_DEVBUF, M_WAITOK); 147 sc->sc_dkdev->dk_info = NULL; 148 disk_init(sc->sc_dkdev, device_xname(self), NULL); 149 if (!pmf_device_register(self, NULL, NULL)) 150 aprint_error_dev(self, "couldn't establish power handler\n"); 151 152 mutex_enter(&fss_device_lock); 153 if (fss_num_attached++ == 0) 154 vfs_hooks_attach(&fss_vfs_hooks); 155 mutex_exit(&fss_device_lock); 156 } 157 158 static int 159 fss_detach(device_t self, int flags) 160 { 161 struct fss_softc *sc = device_private(self); 162 163 if (sc->sc_flags & FSS_ACTIVE) 164 return EBUSY; 165 166 mutex_enter(&fss_device_lock); 167 if (--fss_num_attached == 0) 168 vfs_hooks_detach(&fss_vfs_hooks); 169 mutex_exit(&fss_device_lock); 170 171 pmf_device_deregister(self); 172 mutex_destroy(&sc->sc_slock); 173 mutex_destroy(&sc->sc_lock); 174 cv_destroy(&sc->sc_work_cv); 175 cv_destroy(&sc->sc_cache_cv); 176 bufq_free(sc->sc_bufq); 177 disk_destroy(sc->sc_dkdev); 178 free(sc->sc_dkdev, M_DEVBUF); 179 180 return 0; 181 } 182 183 int 184 fss_open(dev_t dev, int flags, int mode, struct lwp *l) 185 { 186 int mflag; 187 cfdata_t cf; 188 struct fss_softc *sc; 189 190 mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN); 191 192 sc = device_lookup_private(&fss_cd, minor(dev)); 193 if (sc == NULL) { 194 cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK); 195 cf->cf_name = fss_cd.cd_name; 196 cf->cf_atname = fss_cd.cd_name; 197 cf->cf_unit = minor(dev); 198 cf->cf_fstate = FSTATE_STAR; 199 sc = device_private(config_attach_pseudo(cf)); 200 if (sc == NULL) 201 return ENOMEM; 202 } 203 204 mutex_enter(&sc->sc_slock); 205 206 sc->sc_flags |= mflag; 207 208 mutex_exit(&sc->sc_slock); 209 210 return 0; 211 } 212 213 int 214 fss_close(dev_t dev, int flags, int mode, struct lwp *l) 215 { 216 int mflag, error; 217 cfdata_t cf; 218 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(dev)); 219 220 mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN); 221 222 mutex_enter(&sc->sc_slock); 223 224 if ((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) == mflag) { 225 if ((sc->sc_uflags & FSS_UNCONFIG_ON_CLOSE) != 0 && 226 (sc->sc_flags & FSS_ACTIVE) != 0) { 227 mutex_exit(&sc->sc_slock); 228 error = fss_ioctl(dev, FSSIOCCLR, NULL, FWRITE, l); 229 if (error) 230 return error; 231 mutex_enter(&sc->sc_slock); 232 } 233 sc->sc_uflags &= ~FSS_UNCONFIG_ON_CLOSE; 234 } 235 236 sc->sc_flags &= ~mflag; 237 238 mutex_exit(&sc->sc_slock); 239 240 if ((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN|FSS_ACTIVE)) == 0) { 241 cf = device_cfdata(sc->sc_dev); 242 error = config_detach(sc->sc_dev, DETACH_QUIET); 243 if (error) 244 return error; 245 free(cf, M_DEVBUF); 246 } 247 248 return 0; 249 } 250 251 void 252 fss_strategy(struct buf *bp) 253 { 254 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(bp->b_dev)); 255 256 mutex_enter(&sc->sc_slock); 257 258 if ((bp->b_flags & B_READ) != B_READ || 259 sc == NULL || !FSS_ISVALID(sc)) { 260 261 mutex_exit(&sc->sc_slock); 262 263 bp->b_error = (sc == NULL ? ENODEV : EROFS); 264 bp->b_resid = bp->b_bcount; 265 biodone(bp); 266 return; 267 } 268 269 bp->b_rawblkno = bp->b_blkno; 270 BUFQ_PUT(sc->sc_bufq, bp); 271 cv_signal(&sc->sc_work_cv); 272 273 mutex_exit(&sc->sc_slock); 274 } 275 276 int 277 fss_read(dev_t dev, struct uio *uio, int flags) 278 { 279 return physio(fss_strategy, NULL, dev, B_READ, minphys, uio); 280 } 281 282 int 283 fss_write(dev_t dev, struct uio *uio, int flags) 284 { 285 return physio(fss_strategy, NULL, dev, B_WRITE, minphys, uio); 286 } 287 288 int 289 fss_ioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 290 { 291 int error; 292 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(dev)); 293 struct fss_set *fss = (struct fss_set *)data; 294 struct fss_get *fsg = (struct fss_get *)data; 295 296 switch (cmd) { 297 case FSSIOCSET: 298 mutex_enter(&sc->sc_lock); 299 if ((flag & FWRITE) == 0) 300 error = EPERM; 301 else if ((sc->sc_flags & FSS_ACTIVE) != 0) 302 error = EBUSY; 303 else 304 error = fss_create_snapshot(sc, fss, l); 305 mutex_exit(&sc->sc_lock); 306 break; 307 308 case FSSIOCCLR: 309 mutex_enter(&sc->sc_lock); 310 if ((flag & FWRITE) == 0) 311 error = EPERM; 312 else if ((sc->sc_flags & FSS_ACTIVE) == 0) 313 error = ENXIO; 314 else 315 error = fss_delete_snapshot(sc, l); 316 mutex_exit(&sc->sc_lock); 317 break; 318 319 case FSSIOCGET: 320 mutex_enter(&sc->sc_lock); 321 switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) { 322 case FSS_ACTIVE: 323 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 324 fsg->fsg_csize = FSS_CLSIZE(sc); 325 fsg->fsg_time = sc->sc_time; 326 fsg->fsg_mount_size = sc->sc_clcount; 327 fsg->fsg_bs_size = sc->sc_clnext; 328 error = 0; 329 break; 330 case FSS_PERSISTENT | FSS_ACTIVE: 331 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 332 fsg->fsg_csize = 0; 333 fsg->fsg_time = sc->sc_time; 334 fsg->fsg_mount_size = 0; 335 fsg->fsg_bs_size = 0; 336 error = 0; 337 break; 338 default: 339 error = ENXIO; 340 break; 341 } 342 mutex_exit(&sc->sc_lock); 343 break; 344 345 case FSSIOFSET: 346 sc->sc_uflags = *(int *)data; 347 error = 0; 348 break; 349 350 case FSSIOFGET: 351 *(int *)data = sc->sc_uflags; 352 error = 0; 353 break; 354 355 default: 356 error = EINVAL; 357 break; 358 } 359 360 return error; 361 } 362 363 int 364 fss_size(dev_t dev) 365 { 366 return -1; 367 } 368 369 int 370 fss_dump(dev_t dev, daddr_t blkno, void *va, 371 size_t size) 372 { 373 return EROFS; 374 } 375 376 /* 377 * An error occurred reading or writing the snapshot or backing store. 378 * If it is the first error log to console. 379 * The caller holds the mutex. 380 */ 381 static inline void 382 fss_error(struct fss_softc *sc, const char *msg) 383 { 384 385 if ((sc->sc_flags & (FSS_ACTIVE|FSS_ERROR)) == FSS_ACTIVE) 386 aprint_error_dev(sc->sc_dev, "snapshot invalid: %s\n", msg); 387 if ((sc->sc_flags & FSS_ACTIVE) == FSS_ACTIVE) 388 sc->sc_flags |= FSS_ERROR; 389 } 390 391 /* 392 * Allocate the variable sized parts of the softc and 393 * fork the kernel thread. 394 * 395 * The fields sc_clcount, sc_clshift, sc_cache_size and sc_indir_size 396 * must be initialized. 397 */ 398 static int 399 fss_softc_alloc(struct fss_softc *sc) 400 { 401 int i, error; 402 403 if ((sc->sc_flags & FSS_PERSISTENT) == 0) { 404 sc->sc_copied = 405 kmem_zalloc(howmany(sc->sc_clcount, NBBY), KM_SLEEP); 406 if (sc->sc_copied == NULL) 407 return(ENOMEM); 408 409 sc->sc_cache = kmem_alloc(sc->sc_cache_size * 410 sizeof(struct fss_cache), KM_SLEEP); 411 if (sc->sc_cache == NULL) 412 return(ENOMEM); 413 414 for (i = 0; i < sc->sc_cache_size; i++) { 415 sc->sc_cache[i].fc_type = FSS_CACHE_FREE; 416 sc->sc_cache[i].fc_data = 417 kmem_alloc(FSS_CLSIZE(sc), KM_SLEEP); 418 if (sc->sc_cache[i].fc_data == NULL) 419 return(ENOMEM); 420 cv_init(&sc->sc_cache[i].fc_state_cv, "cowwait1"); 421 } 422 423 sc->sc_indir_valid = 424 kmem_zalloc(howmany(sc->sc_indir_size, NBBY), KM_SLEEP); 425 if (sc->sc_indir_valid == NULL) 426 return(ENOMEM); 427 428 sc->sc_indir_data = kmem_zalloc(FSS_CLSIZE(sc), KM_SLEEP); 429 if (sc->sc_indir_data == NULL) 430 return(ENOMEM); 431 } else { 432 sc->sc_copied = NULL; 433 sc->sc_cache = NULL; 434 sc->sc_indir_valid = NULL; 435 sc->sc_indir_data = NULL; 436 } 437 438 if ((error = kthread_create(PRI_BIO, 0, NULL, fss_bs_thread, sc, 439 &sc->sc_bs_lwp, device_xname(sc->sc_dev))) != 0) 440 return error; 441 442 sc->sc_flags |= FSS_BS_THREAD; 443 444 disk_attach(sc->sc_dkdev); 445 446 return 0; 447 } 448 449 /* 450 * Free the variable sized parts of the softc. 451 */ 452 static void 453 fss_softc_free(struct fss_softc *sc) 454 { 455 int i; 456 457 if ((sc->sc_flags & FSS_BS_THREAD) != 0) { 458 mutex_enter(&sc->sc_slock); 459 sc->sc_flags &= ~FSS_BS_THREAD; 460 cv_signal(&sc->sc_work_cv); 461 while (sc->sc_bs_lwp != NULL) 462 kpause("fssdetach", false, 1, &sc->sc_slock); 463 mutex_exit(&sc->sc_slock); 464 } 465 466 disk_detach(sc->sc_dkdev); 467 468 if (sc->sc_copied != NULL) 469 kmem_free(sc->sc_copied, howmany(sc->sc_clcount, NBBY)); 470 sc->sc_copied = NULL; 471 472 if (sc->sc_cache != NULL) { 473 for (i = 0; i < sc->sc_cache_size; i++) 474 if (sc->sc_cache[i].fc_data != NULL) { 475 cv_destroy(&sc->sc_cache[i].fc_state_cv); 476 kmem_free(sc->sc_cache[i].fc_data, 477 FSS_CLSIZE(sc)); 478 } 479 kmem_free(sc->sc_cache, 480 sc->sc_cache_size*sizeof(struct fss_cache)); 481 } 482 sc->sc_cache = NULL; 483 484 if (sc->sc_indir_valid != NULL) 485 kmem_free(sc->sc_indir_valid, howmany(sc->sc_indir_size, NBBY)); 486 sc->sc_indir_valid = NULL; 487 488 if (sc->sc_indir_data != NULL) 489 kmem_free(sc->sc_indir_data, FSS_CLSIZE(sc)); 490 sc->sc_indir_data = NULL; 491 } 492 493 /* 494 * Set all active snapshots on this file system into ERROR state. 495 */ 496 static void 497 fss_unmount_hook(struct mount *mp) 498 { 499 int i; 500 struct fss_softc *sc; 501 502 for (i = 0; i < fss_cd.cd_ndevs; i++) { 503 if ((sc = device_lookup_private(&fss_cd, i)) == NULL) 504 continue; 505 mutex_enter(&sc->sc_slock); 506 if ((sc->sc_flags & FSS_ACTIVE) != 0 && 507 sc->sc_mount == mp) 508 fss_error(sc, "forced unmount"); 509 mutex_exit(&sc->sc_slock); 510 } 511 } 512 513 /* 514 * A buffer is written to the snapshotted block device. Copy to 515 * backing store if needed. 516 */ 517 static int 518 fss_copy_on_write(void *v, struct buf *bp, bool data_valid) 519 { 520 int error; 521 u_int32_t cl, ch, c; 522 struct fss_softc *sc = v; 523 524 mutex_enter(&sc->sc_slock); 525 if (!FSS_ISVALID(sc)) { 526 mutex_exit(&sc->sc_slock); 527 return 0; 528 } 529 530 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 531 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 532 error = 0; 533 if (curlwp == uvm.pagedaemon_lwp) { 534 for (c = cl; c <= ch; c++) 535 if (isclr(sc->sc_copied, c)) { 536 error = ENOMEM; 537 break; 538 } 539 } 540 mutex_exit(&sc->sc_slock); 541 542 if (error == 0) 543 for (c = cl; c <= ch; c++) { 544 error = fss_read_cluster(sc, c); 545 if (error) 546 break; 547 } 548 549 return error; 550 } 551 552 /* 553 * Lookup and open needed files. 554 * 555 * For file system internal snapshot initializes sc_mntname, sc_mount, 556 * sc_bs_vp and sc_time. 557 * 558 * Otherwise returns dev and size of the underlying block device. 559 * Initializes sc_mntname, sc_mount, sc_bdev, sc_bs_vp and sc_mount 560 */ 561 static int 562 fss_create_files(struct fss_softc *sc, struct fss_set *fss, 563 off_t *bsize, struct lwp *l) 564 { 565 int error, bits, fsbsize; 566 struct timespec ts; 567 struct partinfo dpart; 568 struct vattr va; 569 struct nameidata nd; 570 571 /* 572 * Get the mounted file system. 573 */ 574 575 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_mount); 576 if ((error = namei(&nd)) != 0) 577 return error; 578 579 if ((nd.ni_vp->v_vflag & VV_ROOT) != VV_ROOT) { 580 vrele(nd.ni_vp); 581 return EINVAL; 582 } 583 584 sc->sc_mount = nd.ni_vp->v_mount; 585 memcpy(sc->sc_mntname, sc->sc_mount->mnt_stat.f_mntonname, MNAMELEN); 586 587 vrele(nd.ni_vp); 588 589 /* 590 * Check for file system internal snapshot. 591 */ 592 593 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_USERSPACE, fss->fss_bstore); 594 if ((error = namei(&nd)) != 0) 595 return error; 596 597 if (nd.ni_vp->v_type == VREG && nd.ni_vp->v_mount == sc->sc_mount) { 598 sc->sc_flags |= FSS_PERSISTENT; 599 sc->sc_bs_vp = nd.ni_vp; 600 601 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 602 bits = sizeof(sc->sc_bs_bshift)*NBBY; 603 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < bits; 604 sc->sc_bs_bshift++) 605 if (FSS_FSBSIZE(sc) == fsbsize) 606 break; 607 if (sc->sc_bs_bshift >= bits) { 608 VOP_UNLOCK(sc->sc_bs_vp, 0); 609 return EINVAL; 610 } 611 612 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 613 sc->sc_clshift = 0; 614 615 error = VFS_SNAPSHOT(sc->sc_mount, sc->sc_bs_vp, &ts); 616 TIMESPEC_TO_TIMEVAL(&sc->sc_time, &ts); 617 618 VOP_UNLOCK(sc->sc_bs_vp, 0); 619 620 return error; 621 } 622 vput(nd.ni_vp); 623 624 /* 625 * Get the block device it is mounted on. 626 */ 627 628 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, 629 sc->sc_mount->mnt_stat.f_mntfromname); 630 if ((error = namei(&nd)) != 0) 631 return error; 632 633 if (nd.ni_vp->v_type != VBLK) { 634 vrele(nd.ni_vp); 635 return EINVAL; 636 } 637 638 sc->sc_bdev = nd.ni_vp->v_rdev; 639 vrele(nd.ni_vp); 640 641 /* 642 * Get the block device size. 643 */ 644 645 error = bdev_ioctl(sc->sc_bdev, DIOCGPART, &dpart, FREAD, l); 646 if (error) 647 return error; 648 649 *bsize = (off_t)dpart.disklab->d_secsize*dpart.part->p_size; 650 651 /* 652 * Get the backing store 653 */ 654 655 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_bstore); 656 if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) 657 return error; 658 VOP_UNLOCK(nd.ni_vp, 0); 659 660 sc->sc_bs_vp = nd.ni_vp; 661 662 if (nd.ni_vp->v_type != VREG && nd.ni_vp->v_type != VCHR) 663 return EINVAL; 664 665 if (sc->sc_bs_vp->v_type == VREG) { 666 error = VOP_GETATTR(sc->sc_bs_vp, &va, l->l_cred); 667 if (error != 0) 668 return error; 669 sc->sc_bs_size = va.va_size; 670 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 671 if (fsbsize & (fsbsize-1)) /* No power of two */ 672 return EINVAL; 673 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < 32; 674 sc->sc_bs_bshift++) 675 if (FSS_FSBSIZE(sc) == fsbsize) 676 break; 677 if (sc->sc_bs_bshift >= 32) 678 return EINVAL; 679 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 680 } else { 681 sc->sc_bs_bshift = DEV_BSHIFT; 682 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 683 } 684 685 /* 686 * As all IO to from/to the backing store goes through 687 * VOP_STRATEGY() clean the buffer cache to prevent 688 * cache incoherencies. 689 */ 690 if ((error = vinvalbuf(sc->sc_bs_vp, V_SAVE, l->l_cred, l, 0, 0)) != 0) 691 return error; 692 693 return 0; 694 } 695 696 /* 697 * Create a snapshot. 698 */ 699 static int 700 fss_create_snapshot(struct fss_softc *sc, struct fss_set *fss, struct lwp *l) 701 { 702 int len, error; 703 u_int32_t csize; 704 off_t bsize; 705 706 bsize = 0; /* XXX gcc */ 707 708 /* 709 * Open needed files. 710 */ 711 if ((error = fss_create_files(sc, fss, &bsize, l)) != 0) 712 goto bad; 713 714 if (sc->sc_flags & FSS_PERSISTENT) { 715 fss_softc_alloc(sc); 716 sc->sc_flags |= FSS_ACTIVE; 717 return 0; 718 } 719 720 /* 721 * Set cluster size. Must be a power of two and 722 * a multiple of backing store block size. 723 */ 724 if (fss->fss_csize <= 0) 725 csize = MAXPHYS; 726 else 727 csize = fss->fss_csize; 728 if (bsize/csize > FSS_CLUSTER_MAX) 729 csize = bsize/FSS_CLUSTER_MAX+1; 730 731 for (sc->sc_clshift = sc->sc_bs_bshift; sc->sc_clshift < 32; 732 sc->sc_clshift++) 733 if (FSS_CLSIZE(sc) >= csize) 734 break; 735 if (sc->sc_clshift >= 32) { 736 error = EINVAL; 737 goto bad; 738 } 739 sc->sc_clmask = FSS_CLSIZE(sc)-1; 740 741 /* 742 * Set number of cache slots. 743 */ 744 if (FSS_CLSIZE(sc) <= 8192) 745 sc->sc_cache_size = 32; 746 else if (FSS_CLSIZE(sc) <= 65536) 747 sc->sc_cache_size = 8; 748 else 749 sc->sc_cache_size = 4; 750 751 /* 752 * Set number of clusters and size of last cluster. 753 */ 754 sc->sc_clcount = FSS_BTOCL(sc, bsize-1)+1; 755 sc->sc_clresid = FSS_CLOFF(sc, bsize-1)+1; 756 757 /* 758 * Set size of indirect table. 759 */ 760 len = sc->sc_clcount*sizeof(u_int32_t); 761 sc->sc_indir_size = FSS_BTOCL(sc, len)+1; 762 sc->sc_clnext = sc->sc_indir_size; 763 sc->sc_indir_cur = 0; 764 765 if ((error = fss_softc_alloc(sc)) != 0) 766 goto bad; 767 768 /* 769 * Activate the snapshot. 770 */ 771 772 if ((error = vfs_suspend(sc->sc_mount, 0)) != 0) 773 goto bad; 774 775 microtime(&sc->sc_time); 776 777 if (error == 0) 778 error = fscow_establish(sc->sc_mount, 779 fss_copy_on_write, sc); 780 if (error == 0) 781 sc->sc_flags |= FSS_ACTIVE; 782 783 vfs_resume(sc->sc_mount); 784 785 if (error != 0) 786 goto bad; 787 788 aprint_debug_dev(sc->sc_dev, "%s snapshot active\n", sc->sc_mntname); 789 aprint_debug_dev(sc->sc_dev, 790 "%u clusters of %u, %u cache slots, %u indir clusters\n", 791 sc->sc_clcount, FSS_CLSIZE(sc), 792 sc->sc_cache_size, sc->sc_indir_size); 793 794 return 0; 795 796 bad: 797 fss_softc_free(sc); 798 if (sc->sc_bs_vp != NULL) { 799 if (sc->sc_flags & FSS_PERSISTENT) 800 vn_close(sc->sc_bs_vp, FREAD, l->l_cred); 801 else 802 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 803 } 804 sc->sc_bs_vp = NULL; 805 806 return error; 807 } 808 809 /* 810 * Delete a snapshot. 811 */ 812 static int 813 fss_delete_snapshot(struct fss_softc *sc, struct lwp *l) 814 { 815 816 if ((sc->sc_flags & FSS_PERSISTENT) == 0) 817 fscow_disestablish(sc->sc_mount, fss_copy_on_write, sc); 818 819 mutex_enter(&sc->sc_slock); 820 sc->sc_flags &= ~(FSS_ACTIVE|FSS_ERROR); 821 sc->sc_mount = NULL; 822 sc->sc_bdev = NODEV; 823 mutex_exit(&sc->sc_slock); 824 825 fss_softc_free(sc); 826 if (sc->sc_flags & FSS_PERSISTENT) 827 vn_close(sc->sc_bs_vp, FREAD, l->l_cred); 828 else 829 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 830 sc->sc_bs_vp = NULL; 831 sc->sc_flags &= ~FSS_PERSISTENT; 832 833 return 0; 834 } 835 836 /* 837 * Read a cluster from the snapshotted block device to the cache. 838 */ 839 static int 840 fss_read_cluster(struct fss_softc *sc, u_int32_t cl) 841 { 842 int error, todo, offset, len; 843 daddr_t dblk; 844 struct buf *bp, *mbp; 845 struct fss_cache *scp, *scl; 846 847 /* 848 * Get a free cache slot. 849 */ 850 scl = sc->sc_cache+sc->sc_cache_size; 851 852 mutex_enter(&sc->sc_slock); 853 854 restart: 855 if (isset(sc->sc_copied, cl) || !FSS_ISVALID(sc)) { 856 mutex_exit(&sc->sc_slock); 857 return 0; 858 } 859 860 for (scp = sc->sc_cache; scp < scl; scp++) 861 if (scp->fc_cluster == cl) { 862 if (scp->fc_type == FSS_CACHE_VALID) { 863 mutex_exit(&sc->sc_slock); 864 return 0; 865 } else if (scp->fc_type == FSS_CACHE_BUSY) { 866 cv_wait(&scp->fc_state_cv, &sc->sc_slock); 867 goto restart; 868 } 869 } 870 871 for (scp = sc->sc_cache; scp < scl; scp++) 872 if (scp->fc_type == FSS_CACHE_FREE) { 873 scp->fc_type = FSS_CACHE_BUSY; 874 scp->fc_cluster = cl; 875 break; 876 } 877 if (scp >= scl) { 878 cv_wait(&sc->sc_cache_cv, &sc->sc_slock); 879 goto restart; 880 } 881 882 mutex_exit(&sc->sc_slock); 883 884 /* 885 * Start the read. 886 */ 887 dblk = btodb(FSS_CLTOB(sc, cl)); 888 if (cl == sc->sc_clcount-1) { 889 todo = sc->sc_clresid; 890 memset((char *)scp->fc_data + todo, 0, FSS_CLSIZE(sc) - todo); 891 } else 892 todo = FSS_CLSIZE(sc); 893 offset = 0; 894 mbp = getiobuf(NULL, true); 895 mbp->b_bufsize = todo; 896 mbp->b_data = scp->fc_data; 897 mbp->b_resid = mbp->b_bcount = todo; 898 mbp->b_flags = B_READ; 899 mbp->b_cflags = BC_BUSY; 900 mbp->b_dev = sc->sc_bdev; 901 while (todo > 0) { 902 len = todo; 903 if (len > MAXPHYS) 904 len = MAXPHYS; 905 if (btodb(FSS_CLTOB(sc, cl)) == dblk && len == todo) 906 bp = mbp; 907 else { 908 bp = getiobuf(NULL, true); 909 nestiobuf_setup(mbp, bp, offset, len); 910 } 911 bp->b_lblkno = 0; 912 bp->b_blkno = dblk; 913 bdev_strategy(bp); 914 dblk += btodb(len); 915 offset += len; 916 todo -= len; 917 } 918 error = biowait(mbp); 919 putiobuf(mbp); 920 921 mutex_enter(&sc->sc_slock); 922 scp->fc_type = (error ? FSS_CACHE_FREE : FSS_CACHE_VALID); 923 cv_broadcast(&scp->fc_state_cv); 924 if (error == 0) { 925 setbit(sc->sc_copied, scp->fc_cluster); 926 cv_signal(&sc->sc_work_cv); 927 } 928 mutex_exit(&sc->sc_slock); 929 930 return error; 931 } 932 933 /* 934 * Read/write clusters from/to backing store. 935 * For persistent snapshots must be called with cl == 0. off is the 936 * offset into the snapshot. 937 */ 938 static int 939 fss_bs_io(struct fss_softc *sc, fss_io_type rw, 940 u_int32_t cl, off_t off, int len, void *data) 941 { 942 int error; 943 944 off += FSS_CLTOB(sc, cl); 945 946 vn_lock(sc->sc_bs_vp, LK_EXCLUSIVE|LK_RETRY); 947 948 error = vn_rdwr((rw == FSS_READ ? UIO_READ : UIO_WRITE), sc->sc_bs_vp, 949 data, len, off, UIO_SYSSPACE, IO_UNIT|IO_NODELOCKED, 950 sc->sc_bs_lwp->l_cred, NULL, NULL); 951 if (error == 0) { 952 mutex_enter(&sc->sc_bs_vp->v_interlock); 953 error = VOP_PUTPAGES(sc->sc_bs_vp, trunc_page(off), 954 round_page(off+len), PGO_CLEANIT|PGO_SYNCIO|PGO_FREE); 955 } 956 957 VOP_UNLOCK(sc->sc_bs_vp, 0); 958 959 return error; 960 } 961 962 /* 963 * Get a pointer to the indirect slot for this cluster. 964 */ 965 static u_int32_t * 966 fss_bs_indir(struct fss_softc *sc, u_int32_t cl) 967 { 968 u_int32_t icl; 969 int ioff; 970 971 icl = cl/(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 972 ioff = cl%(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 973 974 if (sc->sc_indir_cur == icl) 975 return &sc->sc_indir_data[ioff]; 976 977 if (sc->sc_indir_dirty) { 978 if (fss_bs_io(sc, FSS_WRITE, sc->sc_indir_cur, 0, 979 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 980 return NULL; 981 setbit(sc->sc_indir_valid, sc->sc_indir_cur); 982 } 983 984 sc->sc_indir_dirty = 0; 985 sc->sc_indir_cur = icl; 986 987 if (isset(sc->sc_indir_valid, sc->sc_indir_cur)) { 988 if (fss_bs_io(sc, FSS_READ, sc->sc_indir_cur, 0, 989 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 990 return NULL; 991 } else 992 memset(sc->sc_indir_data, 0, FSS_CLSIZE(sc)); 993 994 return &sc->sc_indir_data[ioff]; 995 } 996 997 /* 998 * The kernel thread (one for every active snapshot). 999 * 1000 * After wakeup it cleans the cache and runs the I/O requests. 1001 */ 1002 static void 1003 fss_bs_thread(void *arg) 1004 { 1005 bool thread_idle; 1006 int error, i, len, crotor; 1007 long off; 1008 char *addr; 1009 u_int32_t c, cl, ch, *indirp; 1010 struct buf *bp, *nbp; 1011 struct fss_softc *sc; 1012 struct fss_cache *scp, *scl; 1013 1014 sc = arg; 1015 scl = sc->sc_cache+sc->sc_cache_size; 1016 crotor = 0; 1017 thread_idle = false; 1018 1019 mutex_enter(&sc->sc_slock); 1020 1021 for (;;) { 1022 if (thread_idle) 1023 cv_wait(&sc->sc_work_cv, &sc->sc_slock); 1024 thread_idle = true; 1025 if ((sc->sc_flags & FSS_BS_THREAD) == 0) { 1026 sc->sc_bs_lwp = NULL; 1027 mutex_exit(&sc->sc_slock); 1028 kthread_exit(0); 1029 } 1030 1031 /* 1032 * Process I/O requests (persistent) 1033 */ 1034 1035 if (sc->sc_flags & FSS_PERSISTENT) { 1036 if ((bp = BUFQ_GET(sc->sc_bufq)) == NULL) 1037 continue; 1038 1039 thread_idle = false; 1040 1041 disk_busy(sc->sc_dkdev); 1042 1043 if (FSS_ISVALID(sc)) { 1044 mutex_exit(&sc->sc_slock); 1045 1046 error = fss_bs_io(sc, FSS_READ, 0, 1047 dbtob(bp->b_blkno), bp->b_bcount, 1048 bp->b_data); 1049 1050 mutex_enter(&sc->sc_slock); 1051 } else 1052 error = ENXIO; 1053 1054 if (error) { 1055 bp->b_error = error; 1056 bp->b_resid = bp->b_bcount; 1057 } else 1058 bp->b_resid = 0; 1059 1060 disk_unbusy(sc->sc_dkdev, bp->b_bcount - bp->b_resid, 1061 (bp->b_flags & B_READ)); 1062 1063 biodone(bp); 1064 1065 continue; 1066 } 1067 1068 /* 1069 * Clean the cache 1070 */ 1071 for (i = 0; i < sc->sc_cache_size; i++) { 1072 crotor = (crotor + 1) % sc->sc_cache_size; 1073 scp = sc->sc_cache + crotor; 1074 if (scp->fc_type != FSS_CACHE_VALID) 1075 continue; 1076 1077 thread_idle = false; 1078 1079 mutex_exit(&sc->sc_slock); 1080 1081 indirp = fss_bs_indir(sc, scp->fc_cluster); 1082 if (indirp != NULL) { 1083 error = fss_bs_io(sc, FSS_WRITE, sc->sc_clnext, 1084 0, FSS_CLSIZE(sc), scp->fc_data); 1085 } else 1086 error = EIO; 1087 1088 mutex_enter(&sc->sc_slock); 1089 1090 if (error == 0) { 1091 *indirp = sc->sc_clnext++; 1092 sc->sc_indir_dirty = 1; 1093 } else 1094 fss_error(sc, "write error on backing store"); 1095 1096 scp->fc_type = FSS_CACHE_FREE; 1097 cv_signal(&sc->sc_cache_cv); 1098 break; 1099 } 1100 1101 /* 1102 * Process I/O requests 1103 */ 1104 if ((bp = BUFQ_GET(sc->sc_bufq)) == NULL) 1105 continue; 1106 1107 thread_idle = false; 1108 1109 disk_busy(sc->sc_dkdev); 1110 1111 if (!FSS_ISVALID(sc)) { 1112 bp->b_error = ENXIO; 1113 bp->b_resid = bp->b_bcount; 1114 disk_unbusy(sc->sc_dkdev, 0, (bp->b_flags & B_READ)); 1115 biodone(bp); 1116 continue; 1117 } 1118 1119 /* 1120 * First read from the snapshotted block device unless 1121 * this request is completely covered by backing store. 1122 */ 1123 1124 mutex_exit(&sc->sc_slock); 1125 1126 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 1127 off = FSS_CLOFF(sc, dbtob(bp->b_blkno)); 1128 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 1129 1130 error = 0; 1131 for (c = cl; c <= ch; c++) { 1132 if (isset(sc->sc_copied, c)) 1133 continue; 1134 /* Not on backing store, read from device. */ 1135 nbp = getiobuf(NULL, true); 1136 nbp->b_flags = B_READ; 1137 nbp->b_resid = nbp->b_bcount = bp->b_bcount; 1138 nbp->b_bufsize = bp->b_bcount; 1139 nbp->b_data = bp->b_data; 1140 nbp->b_blkno = bp->b_blkno; 1141 nbp->b_lblkno = 0; 1142 nbp->b_dev = sc->sc_bdev; 1143 SET(nbp->b_cflags, BC_BUSY); /* mark buffer busy */ 1144 1145 bdev_strategy(nbp); 1146 1147 error = biowait(nbp); 1148 if (error != 0) { 1149 bp->b_resid = bp->b_bcount; 1150 bp->b_error = nbp->b_error; 1151 disk_unbusy(sc->sc_dkdev, 0, 1152 (bp->b_flags & B_READ)); 1153 biodone(bp); 1154 } 1155 putiobuf(nbp); 1156 break; 1157 } 1158 mutex_enter(&sc->sc_slock); 1159 if (error) 1160 continue; 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 mutex_exit(&sc->sc_slock); 1178 1179 indirp = fss_bs_indir(sc, c); 1180 1181 mutex_enter(&sc->sc_slock); 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 mutex_exit(&sc->sc_slock); 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 mutex_enter(&sc->sc_slock); 1209 break; 1210 } 1211 1212 mutex_enter(&sc->sc_slock); 1213 1214 } 1215 1216 disk_unbusy(sc->sc_dkdev, bp->b_bcount - bp->b_resid, 1217 (bp->b_flags & B_READ)); 1218 1219 biodone(bp); 1220 } 1221 } 1222
Indexes created Wed Nov 11 00:26:22 UTC 2026