1 /* $NetBSD: fss.c,v 1.63 2009/06/29 05:08:17 dholland 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 1.63 2009/06/29 05:08:17 dholland 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 /* nd -> nd2 to reduce mistakes while updating only some namei calls */ 583 struct nameidata nd2; 584 struct vnode *vp; 585 586 /* 587 * Get the mounted file system. 588 */ 589 590 error = namei_simple_user(fss->fss_mount, 591 NSM_FOLLOW_NOEMULROOT, &vp); 592 if (error != 0) 593 return error; 594 595 if ((vp->v_vflag & VV_ROOT) != VV_ROOT) { 596 vrele(vp); 597 return EINVAL; 598 } 599 600 sc->sc_mount = vp->v_mount; 601 memcpy(sc->sc_mntname, sc->sc_mount->mnt_stat.f_mntonname, MNAMELEN); 602 603 vrele(vp); 604 605 /* 606 * Check for file system internal snapshot. 607 */ 608 609 error = namei_simple_user(fss->fss_bstore, 610 NSM_FOLLOW_NOEMULROOT, &vp); 611 if (error != 0) 612 return error; 613 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 614 if (error != 0) { 615 vrele(vp); 616 return error; 617 } 618 619 if (vp->v_type == VREG && vp->v_mount == sc->sc_mount) { 620 sc->sc_flags |= FSS_PERSISTENT; 621 sc->sc_bs_vp = vp; 622 623 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 624 bits = sizeof(sc->sc_bs_bshift)*NBBY; 625 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < bits; 626 sc->sc_bs_bshift++) 627 if (FSS_FSBSIZE(sc) == fsbsize) 628 break; 629 if (sc->sc_bs_bshift >= bits) { 630 VOP_UNLOCK(sc->sc_bs_vp, 0); 631 return EINVAL; 632 } 633 634 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 635 sc->sc_clshift = 0; 636 637 error = VFS_SNAPSHOT(sc->sc_mount, sc->sc_bs_vp, &ts); 638 TIMESPEC_TO_TIMEVAL(&sc->sc_time, &ts); 639 640 VOP_UNLOCK(sc->sc_bs_vp, 0); 641 642 return error; 643 } 644 vput(vp); 645 646 /* 647 * Get the block device it is mounted on. 648 */ 649 650 error = namei_simple_kernel(sc->sc_mount->mnt_stat.f_mntfromname, 651 NSM_FOLLOW_NOEMULROOT, &vp); 652 if (error != 0) 653 return error; 654 655 if (vp->v_type != VBLK) { 656 vrele(vp); 657 return EINVAL; 658 } 659 660 sc->sc_bdev = vp->v_rdev; 661 vrele(vp); 662 663 /* 664 * Get the block device size. 665 */ 666 667 error = bdev_ioctl(sc->sc_bdev, DIOCGPART, &dpart, FREAD, l); 668 if (error) 669 return error; 670 671 *bsize = (off_t)dpart.disklab->d_secsize*dpart.part->p_size; 672 673 /* 674 * Get the backing store 675 */ 676 677 NDINIT(&nd2, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_bstore); 678 if ((error = vn_open(&nd2, FREAD|FWRITE, 0)) != 0) 679 return error; 680 VOP_UNLOCK(nd2.ni_vp, 0); 681 682 sc->sc_bs_vp = nd2.ni_vp; 683 684 if (nd2.ni_vp->v_type != VREG && nd2.ni_vp->v_type != VCHR) 685 return EINVAL; 686 687 if (sc->sc_bs_vp->v_type == VREG) { 688 error = VOP_GETATTR(sc->sc_bs_vp, &va, l->l_cred); 689 if (error != 0) 690 return error; 691 sc->sc_bs_size = va.va_size; 692 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 693 if (fsbsize & (fsbsize-1)) /* No power of two */ 694 return EINVAL; 695 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < 32; 696 sc->sc_bs_bshift++) 697 if (FSS_FSBSIZE(sc) == fsbsize) 698 break; 699 if (sc->sc_bs_bshift >= 32) 700 return EINVAL; 701 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 702 } else { 703 sc->sc_bs_bshift = DEV_BSHIFT; 704 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 705 } 706 707 /* 708 * As all IO to from/to the backing store goes through 709 * VOP_STRATEGY() clean the buffer cache to prevent 710 * cache incoherencies. 711 */ 712 if ((error = vinvalbuf(sc->sc_bs_vp, V_SAVE, l->l_cred, l, 0, 0)) != 0) 713 return error; 714 715 return 0; 716 } 717 718 /* 719 * Create a snapshot. 720 */ 721 static int 722 fss_create_snapshot(struct fss_softc *sc, struct fss_set *fss, struct lwp *l) 723 { 724 int len, error; 725 u_int32_t csize; 726 off_t bsize; 727 728 bsize = 0; /* XXX gcc */ 729 730 /* 731 * Open needed files. 732 */ 733 if ((error = fss_create_files(sc, fss, &bsize, l)) != 0) 734 goto bad; 735 736 if (sc->sc_flags & FSS_PERSISTENT) { 737 fss_softc_alloc(sc); 738 sc->sc_flags |= FSS_ACTIVE; 739 return 0; 740 } 741 742 /* 743 * Set cluster size. Must be a power of two and 744 * a multiple of backing store block size. 745 */ 746 if (fss->fss_csize <= 0) 747 csize = MAXPHYS; 748 else 749 csize = fss->fss_csize; 750 if (bsize/csize > FSS_CLUSTER_MAX) 751 csize = bsize/FSS_CLUSTER_MAX+1; 752 753 for (sc->sc_clshift = sc->sc_bs_bshift; sc->sc_clshift < 32; 754 sc->sc_clshift++) 755 if (FSS_CLSIZE(sc) >= csize) 756 break; 757 if (sc->sc_clshift >= 32) { 758 error = EINVAL; 759 goto bad; 760 } 761 sc->sc_clmask = FSS_CLSIZE(sc)-1; 762 763 /* 764 * Set number of cache slots. 765 */ 766 if (FSS_CLSIZE(sc) <= 8192) 767 sc->sc_cache_size = 32; 768 else if (FSS_CLSIZE(sc) <= 65536) 769 sc->sc_cache_size = 8; 770 else 771 sc->sc_cache_size = 4; 772 773 /* 774 * Set number of clusters and size of last cluster. 775 */ 776 sc->sc_clcount = FSS_BTOCL(sc, bsize-1)+1; 777 sc->sc_clresid = FSS_CLOFF(sc, bsize-1)+1; 778 779 /* 780 * Set size of indirect table. 781 */ 782 len = sc->sc_clcount*sizeof(u_int32_t); 783 sc->sc_indir_size = FSS_BTOCL(sc, len)+1; 784 sc->sc_clnext = sc->sc_indir_size; 785 sc->sc_indir_cur = 0; 786 787 if ((error = fss_softc_alloc(sc)) != 0) 788 goto bad; 789 790 /* 791 * Activate the snapshot. 792 */ 793 794 if ((error = vfs_suspend(sc->sc_mount, 0)) != 0) 795 goto bad; 796 797 microtime(&sc->sc_time); 798 799 if (error == 0) 800 error = fscow_establish(sc->sc_mount, 801 fss_copy_on_write, sc); 802 if (error == 0) 803 sc->sc_flags |= FSS_ACTIVE; 804 805 vfs_resume(sc->sc_mount); 806 807 if (error != 0) 808 goto bad; 809 810 aprint_debug_dev(sc->sc_dev, "%s snapshot active\n", sc->sc_mntname); 811 aprint_debug_dev(sc->sc_dev, 812 "%u clusters of %u, %u cache slots, %u indir clusters\n", 813 sc->sc_clcount, FSS_CLSIZE(sc), 814 sc->sc_cache_size, sc->sc_indir_size); 815 816 return 0; 817 818 bad: 819 fss_softc_free(sc); 820 if (sc->sc_bs_vp != NULL) { 821 if (sc->sc_flags & FSS_PERSISTENT) 822 vn_close(sc->sc_bs_vp, FREAD, l->l_cred); 823 else 824 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 825 } 826 sc->sc_bs_vp = NULL; 827 828 return error; 829 } 830 831 /* 832 * Delete a snapshot. 833 */ 834 static int 835 fss_delete_snapshot(struct fss_softc *sc, struct lwp *l) 836 { 837 838 if ((sc->sc_flags & FSS_PERSISTENT) == 0) 839 fscow_disestablish(sc->sc_mount, fss_copy_on_write, sc); 840 841 mutex_enter(&sc->sc_slock); 842 sc->sc_flags &= ~(FSS_ACTIVE|FSS_ERROR); 843 sc->sc_mount = NULL; 844 sc->sc_bdev = NODEV; 845 mutex_exit(&sc->sc_slock); 846 847 fss_softc_free(sc); 848 if (sc->sc_flags & FSS_PERSISTENT) 849 vn_close(sc->sc_bs_vp, FREAD, l->l_cred); 850 else 851 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 852 sc->sc_bs_vp = NULL; 853 sc->sc_flags &= ~FSS_PERSISTENT; 854 855 return 0; 856 } 857 858 /* 859 * Read a cluster from the snapshotted block device to the cache. 860 */ 861 static int 862 fss_read_cluster(struct fss_softc *sc, u_int32_t cl) 863 { 864 int error, todo, offset, len; 865 daddr_t dblk; 866 struct buf *bp, *mbp; 867 struct fss_cache *scp, *scl; 868 869 /* 870 * Get a free cache slot. 871 */ 872 scl = sc->sc_cache+sc->sc_cache_size; 873 874 mutex_enter(&sc->sc_slock); 875 876 restart: 877 if (isset(sc->sc_copied, cl) || !FSS_ISVALID(sc)) { 878 mutex_exit(&sc->sc_slock); 879 return 0; 880 } 881 882 for (scp = sc->sc_cache; scp < scl; scp++) 883 if (scp->fc_cluster == cl) { 884 if (scp->fc_type == FSS_CACHE_VALID) { 885 mutex_exit(&sc->sc_slock); 886 return 0; 887 } else if (scp->fc_type == FSS_CACHE_BUSY) { 888 cv_wait(&scp->fc_state_cv, &sc->sc_slock); 889 goto restart; 890 } 891 } 892 893 for (scp = sc->sc_cache; scp < scl; scp++) 894 if (scp->fc_type == FSS_CACHE_FREE) { 895 scp->fc_type = FSS_CACHE_BUSY; 896 scp->fc_cluster = cl; 897 break; 898 } 899 if (scp >= scl) { 900 cv_wait(&sc->sc_cache_cv, &sc->sc_slock); 901 goto restart; 902 } 903 904 mutex_exit(&sc->sc_slock); 905 906 /* 907 * Start the read. 908 */ 909 dblk = btodb(FSS_CLTOB(sc, cl)); 910 if (cl == sc->sc_clcount-1) { 911 todo = sc->sc_clresid; 912 memset((char *)scp->fc_data + todo, 0, FSS_CLSIZE(sc) - todo); 913 } else 914 todo = FSS_CLSIZE(sc); 915 offset = 0; 916 mbp = getiobuf(NULL, true); 917 mbp->b_bufsize = todo; 918 mbp->b_data = scp->fc_data; 919 mbp->b_resid = mbp->b_bcount = todo; 920 mbp->b_flags = B_READ; 921 mbp->b_cflags = BC_BUSY; 922 mbp->b_dev = sc->sc_bdev; 923 while (todo > 0) { 924 len = todo; 925 if (len > MAXPHYS) 926 len = MAXPHYS; 927 if (btodb(FSS_CLTOB(sc, cl)) == dblk && len == todo) 928 bp = mbp; 929 else { 930 bp = getiobuf(NULL, true); 931 nestiobuf_setup(mbp, bp, offset, len); 932 } 933 bp->b_lblkno = 0; 934 bp->b_blkno = dblk; 935 bdev_strategy(bp); 936 dblk += btodb(len); 937 offset += len; 938 todo -= len; 939 } 940 error = biowait(mbp); 941 putiobuf(mbp); 942 943 mutex_enter(&sc->sc_slock); 944 scp->fc_type = (error ? FSS_CACHE_FREE : FSS_CACHE_VALID); 945 cv_broadcast(&scp->fc_state_cv); 946 if (error == 0) { 947 setbit(sc->sc_copied, scp->fc_cluster); 948 cv_signal(&sc->sc_work_cv); 949 } 950 mutex_exit(&sc->sc_slock); 951 952 return error; 953 } 954 955 /* 956 * Read/write clusters from/to backing store. 957 * For persistent snapshots must be called with cl == 0. off is the 958 * offset into the snapshot. 959 */ 960 static int 961 fss_bs_io(struct fss_softc *sc, fss_io_type rw, 962 u_int32_t cl, off_t off, int len, void *data) 963 { 964 int error; 965 966 off += FSS_CLTOB(sc, cl); 967 968 vn_lock(sc->sc_bs_vp, LK_EXCLUSIVE|LK_RETRY); 969 970 error = vn_rdwr((rw == FSS_READ ? UIO_READ : UIO_WRITE), sc->sc_bs_vp, 971 data, len, off, UIO_SYSSPACE, IO_UNIT|IO_NODELOCKED, 972 sc->sc_bs_lwp->l_cred, NULL, NULL); 973 if (error == 0) { 974 mutex_enter(&sc->sc_bs_vp->v_interlock); 975 error = VOP_PUTPAGES(sc->sc_bs_vp, trunc_page(off), 976 round_page(off+len), PGO_CLEANIT|PGO_SYNCIO|PGO_FREE); 977 } 978 979 VOP_UNLOCK(sc->sc_bs_vp, 0); 980 981 return error; 982 } 983 984 /* 985 * Get a pointer to the indirect slot for this cluster. 986 */ 987 static u_int32_t * 988 fss_bs_indir(struct fss_softc *sc, u_int32_t cl) 989 { 990 u_int32_t icl; 991 int ioff; 992 993 icl = cl/(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 994 ioff = cl%(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 995 996 if (sc->sc_indir_cur == icl) 997 return &sc->sc_indir_data[ioff]; 998 999 if (sc->sc_indir_dirty) { 1000 if (fss_bs_io(sc, FSS_WRITE, sc->sc_indir_cur, 0, 1001 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 1002 return NULL; 1003 setbit(sc->sc_indir_valid, sc->sc_indir_cur); 1004 } 1005 1006 sc->sc_indir_dirty = 0; 1007 sc->sc_indir_cur = icl; 1008 1009 if (isset(sc->sc_indir_valid, sc->sc_indir_cur)) { 1010 if (fss_bs_io(sc, FSS_READ, sc->sc_indir_cur, 0, 1011 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 1012 return NULL; 1013 } else 1014 memset(sc->sc_indir_data, 0, FSS_CLSIZE(sc)); 1015 1016 return &sc->sc_indir_data[ioff]; 1017 } 1018 1019 /* 1020 * The kernel thread (one for every active snapshot). 1021 * 1022 * After wakeup it cleans the cache and runs the I/O requests. 1023 */ 1024 static void 1025 fss_bs_thread(void *arg) 1026 { 1027 bool thread_idle, is_valid; 1028 int error, i, todo, len, crotor, is_read; 1029 long off; 1030 char *addr; 1031 u_int32_t c, cl, ch, *indirp; 1032 struct buf *bp, *nbp; 1033 struct fss_softc *sc; 1034 struct fss_cache *scp, *scl; 1035 1036 sc = arg; 1037 scl = sc->sc_cache+sc->sc_cache_size; 1038 crotor = 0; 1039 thread_idle = false; 1040 1041 mutex_enter(&sc->sc_slock); 1042 1043 for (;;) { 1044 if (thread_idle) 1045 cv_wait(&sc->sc_work_cv, &sc->sc_slock); 1046 thread_idle = true; 1047 if ((sc->sc_flags & FSS_BS_THREAD) == 0) { 1048 sc->sc_bs_lwp = NULL; 1049 mutex_exit(&sc->sc_slock); 1050 kthread_exit(0); 1051 } 1052 1053 /* 1054 * Process I/O requests (persistent) 1055 */ 1056 1057 if (sc->sc_flags & FSS_PERSISTENT) { 1058 if ((bp = bufq_get(sc->sc_bufq)) == NULL) 1059 continue; 1060 is_valid = FSS_ISVALID(sc); 1061 is_read = (bp->b_flags & B_READ); 1062 thread_idle = false; 1063 mutex_exit(&sc->sc_slock); 1064 1065 if (is_valid) { 1066 disk_busy(sc->sc_dkdev); 1067 error = fss_bs_io(sc, FSS_READ, 0, 1068 dbtob(bp->b_blkno), bp->b_bcount, 1069 bp->b_data); 1070 disk_unbusy(sc->sc_dkdev, 1071 (error ? 0 : bp->b_bcount), is_read); 1072 } else 1073 error = ENXIO; 1074 1075 bp->b_error = error; 1076 bp->b_resid = (error ? bp->b_bcount : 0); 1077 biodone(bp); 1078 1079 mutex_enter(&sc->sc_slock); 1080 continue; 1081 } 1082 1083 /* 1084 * Clean the cache 1085 */ 1086 for (i = 0; i < sc->sc_cache_size; i++) { 1087 crotor = (crotor + 1) % sc->sc_cache_size; 1088 scp = sc->sc_cache + crotor; 1089 if (scp->fc_type != FSS_CACHE_VALID) 1090 continue; 1091 mutex_exit(&sc->sc_slock); 1092 1093 thread_idle = false; 1094 indirp = fss_bs_indir(sc, scp->fc_cluster); 1095 if (indirp != NULL) { 1096 error = fss_bs_io(sc, FSS_WRITE, sc->sc_clnext, 1097 0, FSS_CLSIZE(sc), scp->fc_data); 1098 } else 1099 error = EIO; 1100 1101 mutex_enter(&sc->sc_slock); 1102 if (error == 0) { 1103 *indirp = sc->sc_clnext++; 1104 sc->sc_indir_dirty = 1; 1105 } else 1106 fss_error(sc, "write error on backing store"); 1107 1108 scp->fc_type = FSS_CACHE_FREE; 1109 cv_signal(&sc->sc_cache_cv); 1110 break; 1111 } 1112 1113 /* 1114 * Process I/O requests 1115 */ 1116 if ((bp = bufq_get(sc->sc_bufq)) == NULL) 1117 continue; 1118 is_valid = FSS_ISVALID(sc); 1119 is_read = (bp->b_flags & B_READ); 1120 thread_idle = false; 1121 1122 if (!is_valid) { 1123 mutex_exit(&sc->sc_slock); 1124 1125 bp->b_error = ENXIO; 1126 bp->b_resid = bp->b_bcount; 1127 biodone(bp); 1128 1129 mutex_enter(&sc->sc_slock); 1130 continue; 1131 } 1132 1133 disk_busy(sc->sc_dkdev); 1134 1135 /* 1136 * First read from the snapshotted block device unless 1137 * this request is completely covered by backing store. 1138 */ 1139 1140 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 1141 off = FSS_CLOFF(sc, dbtob(bp->b_blkno)); 1142 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 1143 error = 0; 1144 bp->b_resid = 0; 1145 bp->b_error = 0; 1146 for (c = cl; c <= ch; c++) { 1147 if (isset(sc->sc_copied, c)) 1148 continue; 1149 mutex_exit(&sc->sc_slock); 1150 1151 /* Not on backing store, read from device. */ 1152 nbp = getiobuf(NULL, true); 1153 nbp->b_flags = B_READ; 1154 nbp->b_resid = nbp->b_bcount = bp->b_bcount; 1155 nbp->b_bufsize = bp->b_bcount; 1156 nbp->b_data = bp->b_data; 1157 nbp->b_blkno = bp->b_blkno; 1158 nbp->b_lblkno = 0; 1159 nbp->b_dev = sc->sc_bdev; 1160 SET(nbp->b_cflags, BC_BUSY); /* mark buffer busy */ 1161 1162 bdev_strategy(nbp); 1163 1164 error = biowait(nbp); 1165 if (error != 0) { 1166 bp->b_resid = bp->b_bcount; 1167 bp->b_error = nbp->b_error; 1168 disk_unbusy(sc->sc_dkdev, 0, is_read); 1169 biodone(bp); 1170 } 1171 putiobuf(nbp); 1172 1173 mutex_enter(&sc->sc_slock); 1174 break; 1175 } 1176 if (error) 1177 continue; 1178 1179 /* 1180 * Replace those parts that have been saved to backing store. 1181 */ 1182 1183 addr = bp->b_data; 1184 todo = bp->b_bcount; 1185 for (c = cl; c <= ch; c++, off = 0, todo -= len, addr += len) { 1186 len = FSS_CLSIZE(sc)-off; 1187 if (len > todo) 1188 len = todo; 1189 if (isclr(sc->sc_copied, c)) 1190 continue; 1191 mutex_exit(&sc->sc_slock); 1192 1193 indirp = fss_bs_indir(sc, c); 1194 if (indirp == NULL || *indirp == 0) { 1195 /* 1196 * Not on backing store. Either in cache 1197 * or hole in the snapshotted block device. 1198 */ 1199 1200 mutex_enter(&sc->sc_slock); 1201 for (scp = sc->sc_cache; scp < scl; scp++) 1202 if (scp->fc_type == FSS_CACHE_VALID && 1203 scp->fc_cluster == c) 1204 break; 1205 if (scp < scl) 1206 memcpy(addr, (char *)scp->fc_data+off, 1207 len); 1208 else 1209 memset(addr, 0, len); 1210 continue; 1211 } 1212 1213 /* 1214 * Read from backing store. 1215 */ 1216 error = 1217 fss_bs_io(sc, FSS_READ, *indirp, off, len, addr); 1218 1219 mutex_enter(&sc->sc_slock); 1220 if (error) { 1221 bp->b_resid = bp->b_bcount; 1222 bp->b_error = error; 1223 break; 1224 } 1225 } 1226 mutex_exit(&sc->sc_slock); 1227 1228 disk_unbusy(sc->sc_dkdev, (error ? 0 : bp->b_bcount), is_read); 1229 biodone(bp); 1230 1231 mutex_enter(&sc->sc_slock); 1232 } 1233 } 1234 1235 #ifdef _MODULE 1236 1237 #include <sys/module.h> 1238 1239 MODULE(MODULE_CLASS_DRIVER, fss, NULL); 1240 CFDRIVER_DECL(fss, DV_DISK, NULL); 1241 1242 static int 1243 fss_modcmd(modcmd_t cmd, void *arg) 1244 { 1245 int bmajor = -1, cmajor = -1, error = 0; 1246 1247 switch (cmd) { 1248 case MODULE_CMD_INIT: 1249 mutex_init(&fss_device_lock, MUTEX_DEFAULT, IPL_NONE); 1250 error = config_cfdriver_attach(&fss_cd); 1251 if (error) { 1252 mutex_destroy(&fss_device_lock); 1253 break; 1254 } 1255 error = config_cfattach_attach(fss_cd.cd_name, &fss_ca); 1256 if (error) { 1257 config_cfdriver_detach(&fss_cd); 1258 mutex_destroy(&fss_device_lock); 1259 break; 1260 } 1261 error = devsw_attach(fss_cd.cd_name, 1262 &fss_bdevsw, &bmajor, &fss_cdevsw, &cmajor); 1263 if (error) { 1264 config_cfattach_detach(fss_cd.cd_name, &fss_ca); 1265 config_cfdriver_detach(&fss_cd); 1266 mutex_destroy(&fss_device_lock); 1267 break; 1268 } 1269 break; 1270 1271 case MODULE_CMD_FINI: 1272 error = config_cfattach_detach(fss_cd.cd_name, &fss_ca); 1273 if (error) 1274 break; 1275 config_cfdriver_detach(&fss_cd); 1276 devsw_detach(&fss_bdevsw, &fss_cdevsw); 1277 mutex_destroy(&fss_device_lock); 1278 break; 1279 1280 default: 1281 error = ENOTTY; 1282 break; 1283 } 1284 1285 return error; 1286 } 1287 1288 #endif /* _MODULE */ 1289