1 /* $NetBSD: fss.c,v 1.61 2008/11/05 19:30:57 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 1.61 2008/11/05 19:30:57 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 all units. */ 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 | D_MPSAFE 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 | D_MPSAFE 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 if (fss_num_attached++ == 0) 153 vfs_hooks_attach(&fss_vfs_hooks); 154 } 155 156 static int 157 fss_detach(device_t self, int flags) 158 { 159 struct fss_softc *sc = device_private(self); 160 161 if (sc->sc_flags & FSS_ACTIVE) 162 return EBUSY; 163 164 if (--fss_num_attached == 0) 165 vfs_hooks_detach(&fss_vfs_hooks); 166 167 pmf_device_deregister(self); 168 mutex_destroy(&sc->sc_slock); 169 mutex_destroy(&sc->sc_lock); 170 cv_destroy(&sc->sc_work_cv); 171 cv_destroy(&sc->sc_cache_cv); 172 bufq_drain(sc->sc_bufq); 173 bufq_free(sc->sc_bufq); 174 disk_destroy(sc->sc_dkdev); 175 free(sc->sc_dkdev, M_DEVBUF); 176 177 return 0; 178 } 179 180 int 181 fss_open(dev_t dev, int flags, int mode, struct lwp *l) 182 { 183 int mflag; 184 cfdata_t cf; 185 struct fss_softc *sc; 186 187 mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN); 188 189 mutex_enter(&fss_device_lock); 190 191 sc = device_lookup_private(&fss_cd, minor(dev)); 192 if (sc == NULL) { 193 cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK); 194 cf->cf_name = fss_cd.cd_name; 195 cf->cf_atname = fss_cd.cd_name; 196 cf->cf_unit = minor(dev); 197 cf->cf_fstate = FSTATE_STAR; 198 sc = device_private(config_attach_pseudo(cf)); 199 if (sc == NULL) 200 return ENOMEM; 201 } 202 203 mutex_enter(&sc->sc_slock); 204 205 sc->sc_flags |= mflag; 206 207 mutex_exit(&sc->sc_slock); 208 mutex_exit(&fss_device_lock); 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 error = 0; 222 223 restart: 224 mutex_enter(&sc->sc_slock); 225 if ((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) != mflag) { 226 sc->sc_flags &= ~mflag; 227 mutex_exit(&sc->sc_slock); 228 return 0; 229 } 230 if ((sc->sc_flags & FSS_ACTIVE) != 0 && 231 (sc->sc_uflags & FSS_UNCONFIG_ON_CLOSE) != 0) { 232 sc->sc_uflags &= ~FSS_UNCONFIG_ON_CLOSE; 233 mutex_exit(&sc->sc_slock); 234 error = fss_ioctl(dev, FSSIOCCLR, NULL, FWRITE, l); 235 goto restart; 236 } 237 if ((sc->sc_flags & FSS_ACTIVE) != 0) { 238 mutex_exit(&sc->sc_slock); 239 return error; 240 } 241 if (! mutex_tryenter(&fss_device_lock)) { 242 mutex_exit(&sc->sc_slock); 243 goto restart; 244 } 245 246 KASSERT((sc->sc_flags & FSS_ACTIVE) == 0); 247 KASSERT((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) == mflag); 248 mutex_exit(&sc->sc_slock); 249 cf = device_cfdata(sc->sc_dev); 250 error = config_detach(sc->sc_dev, DETACH_QUIET); 251 if (! error) 252 free(cf, M_DEVBUF); 253 mutex_exit(&fss_device_lock); 254 255 return error; 256 } 257 258 void 259 fss_strategy(struct buf *bp) 260 { 261 const bool write = ((bp->b_flags & B_READ) != B_READ); 262 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(bp->b_dev)); 263 264 mutex_enter(&sc->sc_slock); 265 266 if (write || !FSS_ISVALID(sc)) { 267 268 mutex_exit(&sc->sc_slock); 269 270 bp->b_error = (write ? EROFS : ENXIO); 271 bp->b_resid = bp->b_bcount; 272 biodone(bp); 273 return; 274 } 275 276 bp->b_rawblkno = bp->b_blkno; 277 BUFQ_PUT(sc->sc_bufq, bp); 278 cv_signal(&sc->sc_work_cv); 279 280 mutex_exit(&sc->sc_slock); 281 } 282 283 int 284 fss_read(dev_t dev, struct uio *uio, int flags) 285 { 286 return physio(fss_strategy, NULL, dev, B_READ, minphys, uio); 287 } 288 289 int 290 fss_write(dev_t dev, struct uio *uio, int flags) 291 { 292 return physio(fss_strategy, NULL, dev, B_WRITE, minphys, uio); 293 } 294 295 int 296 fss_ioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 297 { 298 int error; 299 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(dev)); 300 struct fss_set *fss = (struct fss_set *)data; 301 struct fss_get *fsg = (struct fss_get *)data; 302 303 switch (cmd) { 304 case FSSIOCSET: 305 mutex_enter(&sc->sc_lock); 306 if ((flag & FWRITE) == 0) 307 error = EPERM; 308 else if ((sc->sc_flags & FSS_ACTIVE) != 0) 309 error = EBUSY; 310 else 311 error = fss_create_snapshot(sc, fss, l); 312 mutex_exit(&sc->sc_lock); 313 break; 314 315 case FSSIOCCLR: 316 mutex_enter(&sc->sc_lock); 317 if ((flag & FWRITE) == 0) 318 error = EPERM; 319 else if ((sc->sc_flags & FSS_ACTIVE) == 0) 320 error = ENXIO; 321 else 322 error = fss_delete_snapshot(sc, l); 323 mutex_exit(&sc->sc_lock); 324 break; 325 326 case FSSIOCGET: 327 mutex_enter(&sc->sc_lock); 328 switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) { 329 case FSS_ACTIVE: 330 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 331 fsg->fsg_csize = FSS_CLSIZE(sc); 332 fsg->fsg_time = sc->sc_time; 333 fsg->fsg_mount_size = sc->sc_clcount; 334 fsg->fsg_bs_size = sc->sc_clnext; 335 error = 0; 336 break; 337 case FSS_PERSISTENT | FSS_ACTIVE: 338 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 339 fsg->fsg_csize = 0; 340 fsg->fsg_time = sc->sc_time; 341 fsg->fsg_mount_size = 0; 342 fsg->fsg_bs_size = 0; 343 error = 0; 344 break; 345 default: 346 error = ENXIO; 347 break; 348 } 349 mutex_exit(&sc->sc_lock); 350 break; 351 352 case FSSIOFSET: 353 mutex_enter(&sc->sc_slock); 354 sc->sc_uflags = *(int *)data; 355 mutex_exit(&sc->sc_slock); 356 error = 0; 357 break; 358 359 case FSSIOFGET: 360 mutex_enter(&sc->sc_slock); 361 *(int *)data = sc->sc_uflags; 362 mutex_exit(&sc->sc_slock); 363 error = 0; 364 break; 365 366 default: 367 error = EINVAL; 368 break; 369 } 370 371 return error; 372 } 373 374 int 375 fss_size(dev_t dev) 376 { 377 return -1; 378 } 379 380 int 381 fss_dump(dev_t dev, daddr_t blkno, void *va, 382 size_t size) 383 { 384 return EROFS; 385 } 386 387 /* 388 * An error occurred reading or writing the snapshot or backing store. 389 * If it is the first error log to console. 390 * The caller holds the mutex. 391 */ 392 static inline void 393 fss_error(struct fss_softc *sc, const char *msg) 394 { 395 396 if ((sc->sc_flags & (FSS_ACTIVE|FSS_ERROR)) == FSS_ACTIVE) 397 aprint_error_dev(sc->sc_dev, "snapshot invalid: %s\n", msg); 398 if ((sc->sc_flags & FSS_ACTIVE) == FSS_ACTIVE) 399 sc->sc_flags |= FSS_ERROR; 400 } 401 402 /* 403 * Allocate the variable sized parts of the softc and 404 * fork the kernel thread. 405 * 406 * The fields sc_clcount, sc_clshift, sc_cache_size and sc_indir_size 407 * must be initialized. 408 */ 409 static int 410 fss_softc_alloc(struct fss_softc *sc) 411 { 412 int i, error; 413 414 if ((sc->sc_flags & FSS_PERSISTENT) == 0) { 415 sc->sc_copied = 416 kmem_zalloc(howmany(sc->sc_clcount, NBBY), KM_SLEEP); 417 if (sc->sc_copied == NULL) 418 return(ENOMEM); 419 420 sc->sc_cache = kmem_alloc(sc->sc_cache_size * 421 sizeof(struct fss_cache), KM_SLEEP); 422 if (sc->sc_cache == NULL) 423 return(ENOMEM); 424 425 for (i = 0; i < sc->sc_cache_size; i++) { 426 sc->sc_cache[i].fc_type = FSS_CACHE_FREE; 427 sc->sc_cache[i].fc_data = 428 kmem_alloc(FSS_CLSIZE(sc), KM_SLEEP); 429 if (sc->sc_cache[i].fc_data == NULL) 430 return(ENOMEM); 431 cv_init(&sc->sc_cache[i].fc_state_cv, "cowwait1"); 432 } 433 434 sc->sc_indir_valid = 435 kmem_zalloc(howmany(sc->sc_indir_size, NBBY), KM_SLEEP); 436 if (sc->sc_indir_valid == NULL) 437 return(ENOMEM); 438 439 sc->sc_indir_data = kmem_zalloc(FSS_CLSIZE(sc), KM_SLEEP); 440 if (sc->sc_indir_data == NULL) 441 return(ENOMEM); 442 } else { 443 sc->sc_copied = NULL; 444 sc->sc_cache = NULL; 445 sc->sc_indir_valid = NULL; 446 sc->sc_indir_data = NULL; 447 } 448 449 if ((error = kthread_create(PRI_BIO, 0, NULL, fss_bs_thread, sc, 450 &sc->sc_bs_lwp, device_xname(sc->sc_dev))) != 0) 451 return error; 452 453 sc->sc_flags |= FSS_BS_THREAD; 454 455 disk_attach(sc->sc_dkdev); 456 457 return 0; 458 } 459 460 /* 461 * Free the variable sized parts of the softc. 462 */ 463 static void 464 fss_softc_free(struct fss_softc *sc) 465 { 466 int i; 467 468 if ((sc->sc_flags & FSS_BS_THREAD) != 0) { 469 mutex_enter(&sc->sc_slock); 470 sc->sc_flags &= ~FSS_BS_THREAD; 471 cv_signal(&sc->sc_work_cv); 472 while (sc->sc_bs_lwp != NULL) 473 kpause("fssdetach", false, 1, &sc->sc_slock); 474 mutex_exit(&sc->sc_slock); 475 } 476 477 disk_detach(sc->sc_dkdev); 478 479 if (sc->sc_copied != NULL) 480 kmem_free(sc->sc_copied, howmany(sc->sc_clcount, NBBY)); 481 sc->sc_copied = NULL; 482 483 if (sc->sc_cache != NULL) { 484 for (i = 0; i < sc->sc_cache_size; i++) 485 if (sc->sc_cache[i].fc_data != NULL) { 486 cv_destroy(&sc->sc_cache[i].fc_state_cv); 487 kmem_free(sc->sc_cache[i].fc_data, 488 FSS_CLSIZE(sc)); 489 } 490 kmem_free(sc->sc_cache, 491 sc->sc_cache_size*sizeof(struct fss_cache)); 492 } 493 sc->sc_cache = NULL; 494 495 if (sc->sc_indir_valid != NULL) 496 kmem_free(sc->sc_indir_valid, howmany(sc->sc_indir_size, NBBY)); 497 sc->sc_indir_valid = NULL; 498 499 if (sc->sc_indir_data != NULL) 500 kmem_free(sc->sc_indir_data, FSS_CLSIZE(sc)); 501 sc->sc_indir_data = NULL; 502 } 503 504 /* 505 * Set all active snapshots on this file system into ERROR state. 506 */ 507 static void 508 fss_unmount_hook(struct mount *mp) 509 { 510 int i; 511 struct fss_softc *sc; 512 513 mutex_enter(&fss_device_lock); 514 for (i = 0; i < fss_cd.cd_ndevs; i++) { 515 if ((sc = device_lookup_private(&fss_cd, i)) == NULL) 516 continue; 517 mutex_enter(&sc->sc_slock); 518 if ((sc->sc_flags & FSS_ACTIVE) != 0 && 519 sc->sc_mount == mp) 520 fss_error(sc, "forced unmount"); 521 mutex_exit(&sc->sc_slock); 522 } 523 mutex_exit(&fss_device_lock); 524 } 525 526 /* 527 * A buffer is written to the snapshotted block device. Copy to 528 * backing store if needed. 529 */ 530 static int 531 fss_copy_on_write(void *v, struct buf *bp, bool data_valid) 532 { 533 int error; 534 u_int32_t cl, ch, c; 535 struct fss_softc *sc = v; 536 537 mutex_enter(&sc->sc_slock); 538 if (!FSS_ISVALID(sc)) { 539 mutex_exit(&sc->sc_slock); 540 return 0; 541 } 542 543 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 544 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 545 error = 0; 546 if (curlwp == uvm.pagedaemon_lwp) { 547 for (c = cl; c <= ch; c++) 548 if (isclr(sc->sc_copied, c)) { 549 error = ENOMEM; 550 break; 551 } 552 } 553 mutex_exit(&sc->sc_slock); 554 555 if (error == 0) 556 for (c = cl; c <= ch; c++) { 557 error = fss_read_cluster(sc, c); 558 if (error) 559 break; 560 } 561 562 return error; 563 } 564 565 /* 566 * Lookup and open needed files. 567 * 568 * For file system internal snapshot initializes sc_mntname, sc_mount, 569 * sc_bs_vp and sc_time. 570 * 571 * Otherwise returns dev and size of the underlying block device. 572 * Initializes sc_mntname, sc_mount, sc_bdev, sc_bs_vp and sc_mount 573 */ 574 static int 575 fss_create_files(struct fss_softc *sc, struct fss_set *fss, 576 off_t *bsize, struct lwp *l) 577 { 578 int error, bits, fsbsize; 579 struct timespec ts; 580 struct partinfo dpart; 581 struct vattr va; 582 struct nameidata nd; 583 584 /* 585 * Get the mounted file system. 586 */ 587 588 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_mount); 589 if ((error = namei(&nd)) != 0) 590 return error; 591 592 if ((nd.ni_vp->v_vflag & VV_ROOT) != VV_ROOT) { 593 vrele(nd.ni_vp); 594 return EINVAL; 595 } 596 597 sc->sc_mount = nd.ni_vp->v_mount; 598 memcpy(sc->sc_mntname, sc->sc_mount->mnt_stat.f_mntonname, MNAMELEN); 599 600 vrele(nd.ni_vp); 601 602 /* 603 * Check for file system internal snapshot. 604 */ 605 606 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_USERSPACE, fss->fss_bstore); 607 if ((error = namei(&nd)) != 0) 608 return error; 609 610 if (nd.ni_vp->v_type == VREG && nd.ni_vp->v_mount == sc->sc_mount) { 611 sc->sc_flags |= FSS_PERSISTENT; 612 sc->sc_bs_vp = nd.ni_vp; 613 614 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 615 bits = sizeof(sc->sc_bs_bshift)*NBBY; 616 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < bits; 617 sc->sc_bs_bshift++) 618 if (FSS_FSBSIZE(sc) == fsbsize) 619 break; 620 if (sc->sc_bs_bshift >= bits) { 621 VOP_UNLOCK(sc->sc_bs_vp, 0); 622 return EINVAL; 623 } 624 625 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 626 sc->sc_clshift = 0; 627 628 error = VFS_SNAPSHOT(sc->sc_mount, sc->sc_bs_vp, &ts); 629 TIMESPEC_TO_TIMEVAL(&sc->sc_time, &ts); 630 631 VOP_UNLOCK(sc->sc_bs_vp, 0); 632 633 return error; 634 } 635 vput(nd.ni_vp); 636 637 /* 638 * Get the block device it is mounted on. 639 */ 640 641 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, 642 sc->sc_mount->mnt_stat.f_mntfromname); 643 if ((error = namei(&nd)) != 0) 644 return error; 645 646 if (nd.ni_vp->v_type != VBLK) { 647 vrele(nd.ni_vp); 648 return EINVAL; 649 } 650 651 sc->sc_bdev = nd.ni_vp->v_rdev; 652 vrele(nd.ni_vp); 653 654 /* 655 * Get the block device size. 656 */ 657 658 error = bdev_ioctl(sc->sc_bdev, DIOCGPART, &dpart, FREAD, l); 659 if (error) 660 return error; 661 662 *bsize = (off_t)dpart.disklab->d_secsize*dpart.part->p_size; 663 664 /* 665 * Get the backing store 666 */ 667 668 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_bstore); 669 if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) 670 return error; 671 VOP_UNLOCK(nd.ni_vp, 0); 672 673 sc->sc_bs_vp = nd.ni_vp; 674 675 if (nd.ni_vp->v_type != VREG && nd.ni_vp->v_type != VCHR) 676 return EINVAL; 677 678 if (sc->sc_bs_vp->v_type == VREG) { 679 error = VOP_GETATTR(sc->sc_bs_vp, &va, l->l_cred); 680 if (error != 0) 681 return error; 682 sc->sc_bs_size = va.va_size; 683 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 684 if (fsbsize & (fsbsize-1)) /* No power of two */ 685 return EINVAL; 686 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < 32; 687 sc->sc_bs_bshift++) 688 if (FSS_FSBSIZE(sc) == fsbsize) 689 break; 690 if (sc->sc_bs_bshift >= 32) 691 return EINVAL; 692 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 693 } else { 694 sc->sc_bs_bshift = DEV_BSHIFT; 695 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 696 } 697 698 /* 699 * As all IO to from/to the backing store goes through 700 * VOP_STRATEGY() clean the buffer cache to prevent 701 * cache incoherencies. 702 */ 703 if ((error = vinvalbuf(sc->sc_bs_vp, V_SAVE, l->l_cred, l, 0, 0)) != 0) 704 return error; 705 706 return 0; 707 } 708 709 /* 710 * Create a snapshot. 711 */ 712 static int 713 fss_create_snapshot(struct fss_softc *sc, struct fss_set *fss, struct lwp *l) 714 { 715 int len, error; 716 u_int32_t csize; 717 off_t bsize; 718 719 bsize = 0; /* XXX gcc */ 720 721 /* 722 * Open needed files. 723 */ 724 if ((error = fss_create_files(sc, fss, &bsize, l)) != 0) 725 goto bad; 726 727 if (sc->sc_flags & FSS_PERSISTENT) { 728 fss_softc_alloc(sc); 729 sc->sc_flags |= FSS_ACTIVE; 730 return 0; 731 } 732 733 /* 734 * Set cluster size. Must be a power of two and 735 * a multiple of backing store block size. 736 */ 737 if (fss->fss_csize <= 0) 738 csize = MAXPHYS; 739 else 740 csize = fss->fss_csize; 741 if (bsize/csize > FSS_CLUSTER_MAX) 742 csize = bsize/FSS_CLUSTER_MAX+1; 743 744 for (sc->sc_clshift = sc->sc_bs_bshift; sc->sc_clshift < 32; 745 sc->sc_clshift++) 746 if (FSS_CLSIZE(sc) >= csize) 747 break; 748 if (sc->sc_clshift >= 32) { 749 error = EINVAL; 750 goto bad; 751 } 752 sc->sc_clmask = FSS_CLSIZE(sc)-1; 753 754 /* 755 * Set number of cache slots. 756 */ 757 if (FSS_CLSIZE(sc) <= 8192) 758 sc->sc_cache_size = 32; 759 else if (FSS_CLSIZE(sc) <= 65536) 760 sc->sc_cache_size = 8; 761 else 762 sc->sc_cache_size = 4; 763 764 /* 765 * Set number of clusters and size of last cluster. 766 */ 767 sc->sc_clcount = FSS_BTOCL(sc, bsize-1)+1; 768 sc->sc_clresid = FSS_CLOFF(sc, bsize-1)+1; 769 770 /* 771 * Set size of indirect table. 772 */ 773 len = sc->sc_clcount*sizeof(u_int32_t); 774 sc->sc_indir_size = FSS_BTOCL(sc, len)+1; 775 sc->sc_clnext = sc->sc_indir_size; 776 sc->sc_indir_cur = 0; 777 778 if ((error = fss_softc_alloc(sc)) != 0) 779 goto bad; 780 781 /* 782 * Activate the snapshot. 783 */ 784 785 if ((error = vfs_suspend(sc->sc_mount, 0)) != 0) 786 goto bad; 787 788 microtime(&sc->sc_time); 789 790 if (error == 0) 791 error = fscow_establish(sc->sc_mount, 792 fss_copy_on_write, sc); 793 if (error == 0) 794 sc->sc_flags |= FSS_ACTIVE; 795 796 vfs_resume(sc->sc_mount); 797 798 if (error != 0) 799 goto bad; 800 801 aprint_debug_dev(sc->sc_dev, "%s snapshot active\n", sc->sc_mntname); 802 aprint_debug_dev(sc->sc_dev, 803 "%u clusters of %u, %u cache slots, %u indir clusters\n", 804 sc->sc_clcount, FSS_CLSIZE(sc), 805 sc->sc_cache_size, sc->sc_indir_size); 806 807 return 0; 808 809 bad: 810 fss_softc_free(sc); 811 if (sc->sc_bs_vp != NULL) { 812 if (sc->sc_flags & FSS_PERSISTENT) 813 vn_close(sc->sc_bs_vp, FREAD, l->l_cred); 814 else 815 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 816 } 817 sc->sc_bs_vp = NULL; 818 819 return error; 820 } 821 822 /* 823 * Delete a snapshot. 824 */ 825 static int 826 fss_delete_snapshot(struct fss_softc *sc, struct lwp *l) 827 { 828 829 if ((sc->sc_flags & FSS_PERSISTENT) == 0) 830 fscow_disestablish(sc->sc_mount, fss_copy_on_write, sc); 831 832 mutex_enter(&sc->sc_slock); 833 sc->sc_flags &= ~(FSS_ACTIVE|FSS_ERROR); 834 sc->sc_mount = NULL; 835 sc->sc_bdev = NODEV; 836 mutex_exit(&sc->sc_slock); 837 838 fss_softc_free(sc); 839 if (sc->sc_flags & FSS_PERSISTENT) 840 vn_close(sc->sc_bs_vp, FREAD, l->l_cred); 841 else 842 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 843 sc->sc_bs_vp = NULL; 844 sc->sc_flags &= ~FSS_PERSISTENT; 845 846 return 0; 847 } 848 849 /* 850 * Read a cluster from the snapshotted block device to the cache. 851 */ 852 static int 853 fss_read_cluster(struct fss_softc *sc, u_int32_t cl) 854 { 855 int error, todo, offset, len; 856 daddr_t dblk; 857 struct buf *bp, *mbp; 858 struct fss_cache *scp, *scl; 859 860 /* 861 * Get a free cache slot. 862 */ 863 scl = sc->sc_cache+sc->sc_cache_size; 864 865 mutex_enter(&sc->sc_slock); 866 867 restart: 868 if (isset(sc->sc_copied, cl) || !FSS_ISVALID(sc)) { 869 mutex_exit(&sc->sc_slock); 870 return 0; 871 } 872 873 for (scp = sc->sc_cache; scp < scl; scp++) 874 if (scp->fc_cluster == cl) { 875 if (scp->fc_type == FSS_CACHE_VALID) { 876 mutex_exit(&sc->sc_slock); 877 return 0; 878 } else if (scp->fc_type == FSS_CACHE_BUSY) { 879 cv_wait(&scp->fc_state_cv, &sc->sc_slock); 880 goto restart; 881 } 882 } 883 884 for (scp = sc->sc_cache; scp < scl; scp++) 885 if (scp->fc_type == FSS_CACHE_FREE) { 886 scp->fc_type = FSS_CACHE_BUSY; 887 scp->fc_cluster = cl; 888 break; 889 } 890 if (scp >= scl) { 891 cv_wait(&sc->sc_cache_cv, &sc->sc_slock); 892 goto restart; 893 } 894 895 mutex_exit(&sc->sc_slock); 896 897 /* 898 * Start the read. 899 */ 900 dblk = btodb(FSS_CLTOB(sc, cl)); 901 if (cl == sc->sc_clcount-1) { 902 todo = sc->sc_clresid; 903 memset((char *)scp->fc_data + todo, 0, FSS_CLSIZE(sc) - todo); 904 } else 905 todo = FSS_CLSIZE(sc); 906 offset = 0; 907 mbp = getiobuf(NULL, true); 908 mbp->b_bufsize = todo; 909 mbp->b_data = scp->fc_data; 910 mbp->b_resid = mbp->b_bcount = todo; 911 mbp->b_flags = B_READ; 912 mbp->b_cflags = BC_BUSY; 913 mbp->b_dev = sc->sc_bdev; 914 while (todo > 0) { 915 len = todo; 916 if (len > MAXPHYS) 917 len = MAXPHYS; 918 if (btodb(FSS_CLTOB(sc, cl)) == dblk && len == todo) 919 bp = mbp; 920 else { 921 bp = getiobuf(NULL, true); 922 nestiobuf_setup(mbp, bp, offset, len); 923 } 924 bp->b_lblkno = 0; 925 bp->b_blkno = dblk; 926 bdev_strategy(bp); 927 dblk += btodb(len); 928 offset += len; 929 todo -= len; 930 } 931 error = biowait(mbp); 932 putiobuf(mbp); 933 934 mutex_enter(&sc->sc_slock); 935 scp->fc_type = (error ? FSS_CACHE_FREE : FSS_CACHE_VALID); 936 cv_broadcast(&scp->fc_state_cv); 937 if (error == 0) { 938 setbit(sc->sc_copied, scp->fc_cluster); 939 cv_signal(&sc->sc_work_cv); 940 } 941 mutex_exit(&sc->sc_slock); 942 943 return error; 944 } 945 946 /* 947 * Read/write clusters from/to backing store. 948 * For persistent snapshots must be called with cl == 0. off is the 949 * offset into the snapshot. 950 */ 951 static int 952 fss_bs_io(struct fss_softc *sc, fss_io_type rw, 953 u_int32_t cl, off_t off, int len, void *data) 954 { 955 int error; 956 957 off += FSS_CLTOB(sc, cl); 958 959 vn_lock(sc->sc_bs_vp, LK_EXCLUSIVE|LK_RETRY); 960 961 error = vn_rdwr((rw == FSS_READ ? UIO_READ : UIO_WRITE), sc->sc_bs_vp, 962 data, len, off, UIO_SYSSPACE, IO_UNIT|IO_NODELOCKED, 963 sc->sc_bs_lwp->l_cred, NULL, NULL); 964 if (error == 0) { 965 mutex_enter(&sc->sc_bs_vp->v_interlock); 966 error = VOP_PUTPAGES(sc->sc_bs_vp, trunc_page(off), 967 round_page(off+len), PGO_CLEANIT|PGO_SYNCIO|PGO_FREE); 968 } 969 970 VOP_UNLOCK(sc->sc_bs_vp, 0); 971 972 return error; 973 } 974 975 /* 976 * Get a pointer to the indirect slot for this cluster. 977 */ 978 static u_int32_t * 979 fss_bs_indir(struct fss_softc *sc, u_int32_t cl) 980 { 981 u_int32_t icl; 982 int ioff; 983 984 icl = cl/(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 985 ioff = cl%(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 986 987 if (sc->sc_indir_cur == icl) 988 return &sc->sc_indir_data[ioff]; 989 990 if (sc->sc_indir_dirty) { 991 if (fss_bs_io(sc, FSS_WRITE, sc->sc_indir_cur, 0, 992 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 993 return NULL; 994 setbit(sc->sc_indir_valid, sc->sc_indir_cur); 995 } 996 997 sc->sc_indir_dirty = 0; 998 sc->sc_indir_cur = icl; 999 1000 if (isset(sc->sc_indir_valid, sc->sc_indir_cur)) { 1001 if (fss_bs_io(sc, FSS_READ, sc->sc_indir_cur, 0, 1002 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 1003 return NULL; 1004 } else 1005 memset(sc->sc_indir_data, 0, FSS_CLSIZE(sc)); 1006 1007 return &sc->sc_indir_data[ioff]; 1008 } 1009 1010 /* 1011 * The kernel thread (one for every active snapshot). 1012 * 1013 * After wakeup it cleans the cache and runs the I/O requests. 1014 */ 1015 static void 1016 fss_bs_thread(void *arg) 1017 { 1018 bool thread_idle, is_valid; 1019 int error, i, todo, len, crotor, is_read; 1020 long off; 1021 char *addr; 1022 u_int32_t c, cl, ch, *indirp; 1023 struct buf *bp, *nbp; 1024 struct fss_softc *sc; 1025 struct fss_cache *scp, *scl; 1026 1027 sc = arg; 1028 scl = sc->sc_cache+sc->sc_cache_size; 1029 crotor = 0; 1030 thread_idle = false; 1031 1032 mutex_enter(&sc->sc_slock); 1033 1034 for (;;) { 1035 if (thread_idle) 1036 cv_wait(&sc->sc_work_cv, &sc->sc_slock); 1037 thread_idle = true; 1038 if ((sc->sc_flags & FSS_BS_THREAD) == 0) { 1039 sc->sc_bs_lwp = NULL; 1040 mutex_exit(&sc->sc_slock); 1041 kthread_exit(0); 1042 } 1043 1044 /* 1045 * Process I/O requests (persistent) 1046 */ 1047 1048 if (sc->sc_flags & FSS_PERSISTENT) { 1049 if ((bp = BUFQ_GET(sc->sc_bufq)) == NULL) 1050 continue; 1051 is_valid = FSS_ISVALID(sc); 1052 is_read = (bp->b_flags & B_READ); 1053 thread_idle = false; 1054 mutex_exit(&sc->sc_slock); 1055 1056 if (is_valid) { 1057 disk_busy(sc->sc_dkdev); 1058 error = fss_bs_io(sc, FSS_READ, 0, 1059 dbtob(bp->b_blkno), bp->b_bcount, 1060 bp->b_data); 1061 disk_unbusy(sc->sc_dkdev, 1062 (error ? 0 : bp->b_bcount), is_read); 1063 } else 1064 error = ENXIO; 1065 1066 bp->b_error = error; 1067 bp->b_resid = (error ? bp->b_bcount : 0); 1068 biodone(bp); 1069 1070 mutex_enter(&sc->sc_slock); 1071 continue; 1072 } 1073 1074 /* 1075 * Clean the cache 1076 */ 1077 for (i = 0; i < sc->sc_cache_size; i++) { 1078 crotor = (crotor + 1) % sc->sc_cache_size; 1079 scp = sc->sc_cache + crotor; 1080 if (scp->fc_type != FSS_CACHE_VALID) 1081 continue; 1082 mutex_exit(&sc->sc_slock); 1083 1084 thread_idle = false; 1085 indirp = fss_bs_indir(sc, scp->fc_cluster); 1086 if (indirp != NULL) { 1087 error = fss_bs_io(sc, FSS_WRITE, sc->sc_clnext, 1088 0, FSS_CLSIZE(sc), scp->fc_data); 1089 } else 1090 error = EIO; 1091 1092 mutex_enter(&sc->sc_slock); 1093 if (error == 0) { 1094 *indirp = sc->sc_clnext++; 1095 sc->sc_indir_dirty = 1; 1096 } else 1097 fss_error(sc, "write error on backing store"); 1098 1099 scp->fc_type = FSS_CACHE_FREE; 1100 cv_signal(&sc->sc_cache_cv); 1101 break; 1102 } 1103 1104 /* 1105 * Process I/O requests 1106 */ 1107 if ((bp = BUFQ_GET(sc->sc_bufq)) == NULL) 1108 continue; 1109 is_valid = FSS_ISVALID(sc); 1110 is_read = (bp->b_flags & B_READ); 1111 thread_idle = false; 1112 1113 if (!is_valid) { 1114 mutex_exit(&sc->sc_slock); 1115 1116 bp->b_error = ENXIO; 1117 bp->b_resid = bp->b_bcount; 1118 biodone(bp); 1119 1120 mutex_enter(&sc->sc_slock); 1121 continue; 1122 } 1123 1124 disk_busy(sc->sc_dkdev); 1125 1126 /* 1127 * First read from the snapshotted block device unless 1128 * this request is completely covered by backing store. 1129 */ 1130 1131 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 1132 off = FSS_CLOFF(sc, dbtob(bp->b_blkno)); 1133 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 1134 error = 0; 1135 bp->b_resid = 0; 1136 bp->b_error = 0; 1137 for (c = cl; c <= ch; c++) { 1138 if (isset(sc->sc_copied, c)) 1139 continue; 1140 mutex_exit(&sc->sc_slock); 1141 1142 /* Not on backing store, read from device. */ 1143 nbp = getiobuf(NULL, true); 1144 nbp->b_flags = B_READ; 1145 nbp->b_resid = nbp->b_bcount = bp->b_bcount; 1146 nbp->b_bufsize = bp->b_bcount; 1147 nbp->b_data = bp->b_data; 1148 nbp->b_blkno = bp->b_blkno; 1149 nbp->b_lblkno = 0; 1150 nbp->b_dev = sc->sc_bdev; 1151 SET(nbp->b_cflags, BC_BUSY); /* mark buffer busy */ 1152 1153 bdev_strategy(nbp); 1154 1155 error = biowait(nbp); 1156 if (error != 0) { 1157 bp->b_resid = bp->b_bcount; 1158 bp->b_error = nbp->b_error; 1159 disk_unbusy(sc->sc_dkdev, 0, is_read); 1160 biodone(bp); 1161 } 1162 putiobuf(nbp); 1163 1164 mutex_enter(&sc->sc_slock); 1165 break; 1166 } 1167 if (error) 1168 continue; 1169 1170 /* 1171 * Replace those parts that have been saved to backing store. 1172 */ 1173 1174 addr = bp->b_data; 1175 todo = bp->b_bcount; 1176 for (c = cl; c <= ch; c++, off = 0, todo -= len, addr += len) { 1177 len = FSS_CLSIZE(sc)-off; 1178 if (len > todo) 1179 len = todo; 1180 if (isclr(sc->sc_copied, c)) 1181 continue; 1182 mutex_exit(&sc->sc_slock); 1183 1184 indirp = fss_bs_indir(sc, c); 1185 if (indirp == NULL || *indirp == 0) { 1186 /* 1187 * Not on backing store. Either in cache 1188 * or hole in the snapshotted block device. 1189 */ 1190 1191 mutex_enter(&sc->sc_slock); 1192 for (scp = sc->sc_cache; scp < scl; scp++) 1193 if (scp->fc_type == FSS_CACHE_VALID && 1194 scp->fc_cluster == c) 1195 break; 1196 if (scp < scl) 1197 memcpy(addr, (char *)scp->fc_data+off, 1198 len); 1199 else 1200 memset(addr, 0, len); 1201 continue; 1202 } 1203 1204 /* 1205 * Read from backing store. 1206 */ 1207 error = 1208 fss_bs_io(sc, FSS_READ, *indirp, off, len, addr); 1209 1210 mutex_enter(&sc->sc_slock); 1211 if (error) { 1212 bp->b_resid = bp->b_bcount; 1213 bp->b_error = error; 1214 break; 1215 } 1216 } 1217 mutex_exit(&sc->sc_slock); 1218 1219 disk_unbusy(sc->sc_dkdev, (error ? 0 : bp->b_bcount), is_read); 1220 biodone(bp); 1221 1222 mutex_enter(&sc->sc_slock); 1223 } 1224 } 1225 1226 #ifdef _MODULE 1227 1228 #include <sys/module.h> 1229 1230 MODULE(MODULE_CLASS_DRIVER, fss, NULL); 1231 CFDRIVER_DECL(fss, DV_DISK, NULL); 1232 1233 static int 1234 fss_modcmd(modcmd_t cmd, void *arg) 1235 { 1236 int bmajor = -1, cmajor = -1, error = 0; 1237 1238 switch (cmd) { 1239 case MODULE_CMD_INIT: 1240 mutex_init(&fss_device_lock, MUTEX_DEFAULT, IPL_NONE); 1241 error = config_cfdriver_attach(&fss_cd); 1242 if (error) { 1243 mutex_destroy(&fss_device_lock); 1244 break; 1245 } 1246 error = config_cfattach_attach(fss_cd.cd_name, &fss_ca); 1247 if (error) { 1248 config_cfdriver_detach(&fss_cd); 1249 mutex_destroy(&fss_device_lock); 1250 break; 1251 } 1252 error = devsw_attach(fss_cd.cd_name, 1253 &fss_bdevsw, &bmajor, &fss_cdevsw, &cmajor); 1254 if (error) { 1255 config_cfattach_detach(fss_cd.cd_name, &fss_ca); 1256 config_cfdriver_detach(&fss_cd); 1257 mutex_destroy(&fss_device_lock); 1258 break; 1259 } 1260 break; 1261 1262 case MODULE_CMD_FINI: 1263 error = config_cfattach_detach(fss_cd.cd_name, &fss_ca); 1264 if (error) 1265 break; 1266 config_cfdriver_detach(&fss_cd); 1267 devsw_detach(&fss_bdevsw, &fss_cdevsw); 1268 mutex_destroy(&fss_device_lock); 1269 break; 1270 1271 default: 1272 error = ENOTTY; 1273 break; 1274 } 1275 1276 return error; 1277 } 1278 1279 #endif /* _MODULE */ 1280