1 /* $NetBSD: fss.c,v 1.72 2010/12/27 18:41:07 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 1.72 2010/12/27 18:41:07 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 mutex_exit(&fss_device_lock); 201 return ENOMEM; 202 } 203 } 204 205 mutex_enter(&sc->sc_slock); 206 207 sc->sc_flags |= mflag; 208 209 mutex_exit(&sc->sc_slock); 210 mutex_exit(&fss_device_lock); 211 212 return 0; 213 } 214 215 int 216 fss_close(dev_t dev, int flags, int mode, struct lwp *l) 217 { 218 int mflag, error; 219 cfdata_t cf; 220 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(dev)); 221 222 mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN); 223 error = 0; 224 225 restart: 226 mutex_enter(&sc->sc_slock); 227 if ((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) != mflag) { 228 sc->sc_flags &= ~mflag; 229 mutex_exit(&sc->sc_slock); 230 return 0; 231 } 232 if ((sc->sc_flags & FSS_ACTIVE) != 0 && 233 (sc->sc_uflags & FSS_UNCONFIG_ON_CLOSE) != 0) { 234 sc->sc_uflags &= ~FSS_UNCONFIG_ON_CLOSE; 235 mutex_exit(&sc->sc_slock); 236 error = fss_ioctl(dev, FSSIOCCLR, NULL, FWRITE, l); 237 goto restart; 238 } 239 if ((sc->sc_flags & FSS_ACTIVE) != 0) { 240 mutex_exit(&sc->sc_slock); 241 return error; 242 } 243 if (! mutex_tryenter(&fss_device_lock)) { 244 mutex_exit(&sc->sc_slock); 245 goto restart; 246 } 247 248 KASSERT((sc->sc_flags & FSS_ACTIVE) == 0); 249 KASSERT((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) == mflag); 250 mutex_exit(&sc->sc_slock); 251 cf = device_cfdata(sc->sc_dev); 252 error = config_detach(sc->sc_dev, DETACH_QUIET); 253 if (! error) 254 free(cf, M_DEVBUF); 255 mutex_exit(&fss_device_lock); 256 257 return error; 258 } 259 260 void 261 fss_strategy(struct buf *bp) 262 { 263 const bool write = ((bp->b_flags & B_READ) != B_READ); 264 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(bp->b_dev)); 265 266 mutex_enter(&sc->sc_slock); 267 268 if (write || !FSS_ISVALID(sc)) { 269 270 mutex_exit(&sc->sc_slock); 271 272 bp->b_error = (write ? EROFS : ENXIO); 273 bp->b_resid = bp->b_bcount; 274 biodone(bp); 275 return; 276 } 277 278 bp->b_rawblkno = bp->b_blkno; 279 bufq_put(sc->sc_bufq, bp); 280 cv_signal(&sc->sc_work_cv); 281 282 mutex_exit(&sc->sc_slock); 283 } 284 285 int 286 fss_read(dev_t dev, struct uio *uio, int flags) 287 { 288 return physio(fss_strategy, NULL, dev, B_READ, minphys, uio); 289 } 290 291 int 292 fss_write(dev_t dev, struct uio *uio, int flags) 293 { 294 return physio(fss_strategy, NULL, dev, B_WRITE, minphys, uio); 295 } 296 297 int 298 fss_ioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 299 { 300 int error; 301 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(dev)); 302 struct fss_set *fss = (struct fss_set *)data; 303 struct fss_get *fsg = (struct fss_get *)data; 304 305 switch (cmd) { 306 case FSSIOCSET: 307 mutex_enter(&sc->sc_lock); 308 if ((flag & FWRITE) == 0) 309 error = EPERM; 310 else if ((sc->sc_flags & FSS_ACTIVE) != 0) 311 error = EBUSY; 312 else 313 error = fss_create_snapshot(sc, fss, l); 314 mutex_exit(&sc->sc_lock); 315 break; 316 317 case FSSIOCCLR: 318 mutex_enter(&sc->sc_lock); 319 if ((flag & FWRITE) == 0) 320 error = EPERM; 321 else if ((sc->sc_flags & FSS_ACTIVE) == 0) 322 error = ENXIO; 323 else 324 error = fss_delete_snapshot(sc, l); 325 mutex_exit(&sc->sc_lock); 326 break; 327 328 case FSSIOCGET: 329 mutex_enter(&sc->sc_lock); 330 switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) { 331 case FSS_ACTIVE: 332 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 333 fsg->fsg_csize = FSS_CLSIZE(sc); 334 fsg->fsg_time = sc->sc_time; 335 fsg->fsg_mount_size = sc->sc_clcount; 336 fsg->fsg_bs_size = sc->sc_clnext; 337 error = 0; 338 break; 339 case FSS_PERSISTENT | FSS_ACTIVE: 340 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 341 fsg->fsg_csize = 0; 342 fsg->fsg_time = sc->sc_time; 343 fsg->fsg_mount_size = 0; 344 fsg->fsg_bs_size = 0; 345 error = 0; 346 break; 347 default: 348 error = ENXIO; 349 break; 350 } 351 mutex_exit(&sc->sc_lock); 352 break; 353 354 case FSSIOFSET: 355 mutex_enter(&sc->sc_slock); 356 sc->sc_uflags = *(int *)data; 357 mutex_exit(&sc->sc_slock); 358 error = 0; 359 break; 360 361 case FSSIOFGET: 362 mutex_enter(&sc->sc_slock); 363 *(int *)data = sc->sc_uflags; 364 mutex_exit(&sc->sc_slock); 365 error = 0; 366 break; 367 368 default: 369 error = EINVAL; 370 break; 371 } 372 373 return error; 374 } 375 376 int 377 fss_size(dev_t dev) 378 { 379 return -1; 380 } 381 382 int 383 fss_dump(dev_t dev, daddr_t blkno, void *va, 384 size_t size) 385 { 386 return EROFS; 387 } 388 389 /* 390 * An error occurred reading or writing the snapshot or backing store. 391 * If it is the first error log to console. 392 * The caller holds the mutex. 393 */ 394 static inline void 395 fss_error(struct fss_softc *sc, const char *msg) 396 { 397 398 if ((sc->sc_flags & (FSS_ACTIVE|FSS_ERROR)) == FSS_ACTIVE) 399 aprint_error_dev(sc->sc_dev, "snapshot invalid: %s\n", msg); 400 if ((sc->sc_flags & FSS_ACTIVE) == FSS_ACTIVE) 401 sc->sc_flags |= FSS_ERROR; 402 } 403 404 /* 405 * Allocate the variable sized parts of the softc and 406 * fork the kernel thread. 407 * 408 * The fields sc_clcount, sc_clshift, sc_cache_size and sc_indir_size 409 * must be initialized. 410 */ 411 static int 412 fss_softc_alloc(struct fss_softc *sc) 413 { 414 int i, error; 415 416 if ((sc->sc_flags & FSS_PERSISTENT) == 0) { 417 sc->sc_copied = 418 kmem_zalloc(howmany(sc->sc_clcount, NBBY), KM_SLEEP); 419 if (sc->sc_copied == NULL) 420 return(ENOMEM); 421 422 sc->sc_cache = kmem_alloc(sc->sc_cache_size * 423 sizeof(struct fss_cache), KM_SLEEP); 424 if (sc->sc_cache == NULL) 425 return(ENOMEM); 426 427 for (i = 0; i < sc->sc_cache_size; i++) { 428 sc->sc_cache[i].fc_type = FSS_CACHE_FREE; 429 sc->sc_cache[i].fc_data = 430 kmem_alloc(FSS_CLSIZE(sc), KM_SLEEP); 431 if (sc->sc_cache[i].fc_data == NULL) 432 return(ENOMEM); 433 cv_init(&sc->sc_cache[i].fc_state_cv, "cowwait1"); 434 } 435 436 sc->sc_indir_valid = 437 kmem_zalloc(howmany(sc->sc_indir_size, NBBY), KM_SLEEP); 438 if (sc->sc_indir_valid == NULL) 439 return(ENOMEM); 440 441 sc->sc_indir_data = kmem_zalloc(FSS_CLSIZE(sc), KM_SLEEP); 442 if (sc->sc_indir_data == NULL) 443 return(ENOMEM); 444 } else { 445 sc->sc_copied = NULL; 446 sc->sc_cache = NULL; 447 sc->sc_indir_valid = NULL; 448 sc->sc_indir_data = NULL; 449 } 450 451 sc->sc_flags |= FSS_BS_THREAD; 452 if ((error = kthread_create(PRI_BIO, 0, NULL, fss_bs_thread, sc, 453 &sc->sc_bs_lwp, device_xname(sc->sc_dev))) != 0) { 454 sc->sc_flags &= ~FSS_BS_THREAD; 455 return error; 456 } 457 458 disk_attach(sc->sc_dkdev); 459 460 return 0; 461 } 462 463 /* 464 * Free the variable sized parts of the softc. 465 */ 466 static void 467 fss_softc_free(struct fss_softc *sc) 468 { 469 int i; 470 471 if ((sc->sc_flags & FSS_BS_THREAD) != 0) { 472 mutex_enter(&sc->sc_slock); 473 sc->sc_flags &= ~FSS_BS_THREAD; 474 cv_signal(&sc->sc_work_cv); 475 while (sc->sc_bs_lwp != NULL) 476 kpause("fssdetach", false, 1, &sc->sc_slock); 477 mutex_exit(&sc->sc_slock); 478 479 disk_detach(sc->sc_dkdev); 480 } 481 482 if (sc->sc_copied != NULL) 483 kmem_free(sc->sc_copied, howmany(sc->sc_clcount, NBBY)); 484 sc->sc_copied = NULL; 485 486 if (sc->sc_cache != NULL) { 487 for (i = 0; i < sc->sc_cache_size; i++) 488 if (sc->sc_cache[i].fc_data != NULL) { 489 cv_destroy(&sc->sc_cache[i].fc_state_cv); 490 kmem_free(sc->sc_cache[i].fc_data, 491 FSS_CLSIZE(sc)); 492 } 493 kmem_free(sc->sc_cache, 494 sc->sc_cache_size*sizeof(struct fss_cache)); 495 } 496 sc->sc_cache = NULL; 497 498 if (sc->sc_indir_valid != NULL) 499 kmem_free(sc->sc_indir_valid, howmany(sc->sc_indir_size, NBBY)); 500 sc->sc_indir_valid = NULL; 501 502 if (sc->sc_indir_data != NULL) 503 kmem_free(sc->sc_indir_data, FSS_CLSIZE(sc)); 504 sc->sc_indir_data = NULL; 505 } 506 507 /* 508 * Set all active snapshots on this file system into ERROR state. 509 */ 510 static void 511 fss_unmount_hook(struct mount *mp) 512 { 513 int i; 514 struct fss_softc *sc; 515 516 mutex_enter(&fss_device_lock); 517 for (i = 0; i < fss_cd.cd_ndevs; i++) { 518 if ((sc = device_lookup_private(&fss_cd, i)) == NULL) 519 continue; 520 mutex_enter(&sc->sc_slock); 521 if ((sc->sc_flags & FSS_ACTIVE) != 0 && 522 sc->sc_mount == mp) 523 fss_error(sc, "forced unmount"); 524 mutex_exit(&sc->sc_slock); 525 } 526 mutex_exit(&fss_device_lock); 527 } 528 529 /* 530 * A buffer is written to the snapshotted block device. Copy to 531 * backing store if needed. 532 */ 533 static int 534 fss_copy_on_write(void *v, struct buf *bp, bool data_valid) 535 { 536 int error; 537 u_int32_t cl, ch, c; 538 struct fss_softc *sc = v; 539 540 mutex_enter(&sc->sc_slock); 541 if (!FSS_ISVALID(sc)) { 542 mutex_exit(&sc->sc_slock); 543 return 0; 544 } 545 546 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 547 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 548 error = 0; 549 if (curlwp == uvm.pagedaemon_lwp) { 550 for (c = cl; c <= ch; c++) 551 if (isclr(sc->sc_copied, c)) { 552 error = ENOMEM; 553 break; 554 } 555 } 556 mutex_exit(&sc->sc_slock); 557 558 if (error == 0) 559 for (c = cl; c <= ch; c++) { 560 error = fss_read_cluster(sc, c); 561 if (error) 562 break; 563 } 564 565 return error; 566 } 567 568 /* 569 * Lookup and open needed files. 570 * 571 * For file system internal snapshot initializes sc_mntname, sc_mount, 572 * sc_bs_vp and sc_time. 573 * 574 * Otherwise returns dev and size of the underlying block device. 575 * Initializes sc_mntname, sc_mount, sc_bdev, sc_bs_vp and sc_mount 576 */ 577 static int 578 fss_create_files(struct fss_softc *sc, struct fss_set *fss, 579 off_t *bsize, struct lwp *l) 580 { 581 int error, bits, fsbsize; 582 struct timespec ts; 583 struct partinfo dpart; 584 /* nd -> nd2 to reduce mistakes while updating only some namei calls */ 585 struct pathbuf *pb2; 586 struct nameidata nd2; 587 struct vnode *vp; 588 589 /* 590 * Get the mounted file system. 591 */ 592 593 error = namei_simple_user(fss->fss_mount, 594 NSM_FOLLOW_NOEMULROOT, &vp); 595 if (error != 0) 596 return error; 597 598 if ((vp->v_vflag & VV_ROOT) != VV_ROOT) { 599 vrele(vp); 600 return EINVAL; 601 } 602 603 sc->sc_mount = vp->v_mount; 604 memcpy(sc->sc_mntname, sc->sc_mount->mnt_stat.f_mntonname, MNAMELEN); 605 606 vrele(vp); 607 608 /* 609 * Check for file system internal snapshot. 610 */ 611 612 error = namei_simple_user(fss->fss_bstore, 613 NSM_FOLLOW_NOEMULROOT, &vp); 614 if (error != 0) 615 return error; 616 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 617 if (error != 0) { 618 vrele(vp); 619 return error; 620 } 621 622 if (vp->v_type == VREG && vp->v_mount == sc->sc_mount) { 623 sc->sc_flags |= FSS_PERSISTENT; 624 sc->sc_bs_vp = vp; 625 626 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 627 bits = sizeof(sc->sc_bs_bshift)*NBBY; 628 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < bits; 629 sc->sc_bs_bshift++) 630 if (FSS_FSBSIZE(sc) == fsbsize) 631 break; 632 if (sc->sc_bs_bshift >= bits) { 633 VOP_UNLOCK(sc->sc_bs_vp); 634 return EINVAL; 635 } 636 637 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 638 sc->sc_clshift = 0; 639 640 error = VFS_SNAPSHOT(sc->sc_mount, sc->sc_bs_vp, &ts); 641 TIMESPEC_TO_TIMEVAL(&sc->sc_time, &ts); 642 643 VOP_UNLOCK(sc->sc_bs_vp); 644 645 return error; 646 } 647 vput(vp); 648 649 /* 650 * Get the block device it is mounted on. 651 */ 652 653 error = namei_simple_kernel(sc->sc_mount->mnt_stat.f_mntfromname, 654 NSM_FOLLOW_NOEMULROOT, &vp); 655 if (error != 0) 656 return error; 657 658 if (vp->v_type != VBLK) { 659 vrele(vp); 660 return EINVAL; 661 } 662 663 sc->sc_bdev = vp->v_rdev; 664 vrele(vp); 665 666 /* 667 * Get the block device size. 668 */ 669 670 error = bdev_ioctl(sc->sc_bdev, DIOCGPART, &dpart, FREAD, l); 671 if (error) 672 return error; 673 674 *bsize = (off_t)dpart.disklab->d_secsize*dpart.part->p_size; 675 676 /* 677 * Get the backing store 678 */ 679 680 error = pathbuf_copyin(fss->fss_bstore, &pb2); 681 if (error) { 682 return error; 683 } 684 NDINIT(&nd2, LOOKUP, FOLLOW, pb2); 685 if ((error = vn_open(&nd2, FREAD|FWRITE, 0)) != 0) { 686 pathbuf_destroy(pb2); 687 return error; 688 } 689 VOP_UNLOCK(nd2.ni_vp); 690 691 sc->sc_bs_vp = nd2.ni_vp; 692 693 if (nd2.ni_vp->v_type != VREG && nd2.ni_vp->v_type != VCHR) { 694 pathbuf_destroy(pb2); 695 return EINVAL; 696 } 697 pathbuf_destroy(pb2); 698 699 if (sc->sc_bs_vp->v_type == VREG) { 700 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 701 if (fsbsize & (fsbsize-1)) /* No power of two */ 702 return EINVAL; 703 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < 32; 704 sc->sc_bs_bshift++) 705 if (FSS_FSBSIZE(sc) == fsbsize) 706 break; 707 if (sc->sc_bs_bshift >= 32) 708 return EINVAL; 709 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 710 } else { 711 sc->sc_bs_bshift = DEV_BSHIFT; 712 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 713 } 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 error = fscow_establish(sc->sc_mount, fss_copy_on_write, sc); 800 if (error == 0) 801 sc->sc_flags |= FSS_ACTIVE; 802 803 vfs_resume(sc->sc_mount); 804 805 if (error != 0) 806 goto bad; 807 808 aprint_debug_dev(sc->sc_dev, "%s snapshot active\n", sc->sc_mntname); 809 aprint_debug_dev(sc->sc_dev, 810 "%u clusters of %u, %u cache slots, %u indir clusters\n", 811 sc->sc_clcount, FSS_CLSIZE(sc), 812 sc->sc_cache_size, sc->sc_indir_size); 813 814 return 0; 815 816 bad: 817 fss_softc_free(sc); 818 if (sc->sc_bs_vp != NULL) { 819 if (sc->sc_flags & FSS_PERSISTENT) 820 vn_close(sc->sc_bs_vp, FREAD, l->l_cred); 821 else 822 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 823 } 824 sc->sc_bs_vp = NULL; 825 826 return error; 827 } 828 829 /* 830 * Delete a snapshot. 831 */ 832 static int 833 fss_delete_snapshot(struct fss_softc *sc, struct lwp *l) 834 { 835 836 if ((sc->sc_flags & FSS_PERSISTENT) == 0) 837 fscow_disestablish(sc->sc_mount, fss_copy_on_write, sc); 838 839 mutex_enter(&sc->sc_slock); 840 sc->sc_flags &= ~(FSS_ACTIVE|FSS_ERROR); 841 sc->sc_mount = NULL; 842 sc->sc_bdev = NODEV; 843 mutex_exit(&sc->sc_slock); 844 845 fss_softc_free(sc); 846 if (sc->sc_flags & FSS_PERSISTENT) 847 vn_close(sc->sc_bs_vp, FREAD, l->l_cred); 848 else 849 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 850 sc->sc_bs_vp = NULL; 851 sc->sc_flags &= ~FSS_PERSISTENT; 852 853 return 0; 854 } 855 856 /* 857 * Read a cluster from the snapshotted block device to the cache. 858 */ 859 static int 860 fss_read_cluster(struct fss_softc *sc, u_int32_t cl) 861 { 862 int error, todo, offset, len; 863 daddr_t dblk; 864 struct buf *bp, *mbp; 865 struct fss_cache *scp, *scl; 866 867 /* 868 * Get a free cache slot. 869 */ 870 scl = sc->sc_cache+sc->sc_cache_size; 871 872 mutex_enter(&sc->sc_slock); 873 874 restart: 875 if (isset(sc->sc_copied, cl) || !FSS_ISVALID(sc)) { 876 mutex_exit(&sc->sc_slock); 877 return 0; 878 } 879 880 for (scp = sc->sc_cache; scp < scl; scp++) 881 if (scp->fc_cluster == cl) { 882 if (scp->fc_type == FSS_CACHE_VALID) { 883 mutex_exit(&sc->sc_slock); 884 return 0; 885 } else if (scp->fc_type == FSS_CACHE_BUSY) { 886 cv_wait(&scp->fc_state_cv, &sc->sc_slock); 887 goto restart; 888 } 889 } 890 891 for (scp = sc->sc_cache; scp < scl; scp++) 892 if (scp->fc_type == FSS_CACHE_FREE) { 893 scp->fc_type = FSS_CACHE_BUSY; 894 scp->fc_cluster = cl; 895 break; 896 } 897 if (scp >= scl) { 898 cv_wait(&sc->sc_cache_cv, &sc->sc_slock); 899 goto restart; 900 } 901 902 mutex_exit(&sc->sc_slock); 903 904 /* 905 * Start the read. 906 */ 907 dblk = btodb(FSS_CLTOB(sc, cl)); 908 if (cl == sc->sc_clcount-1) { 909 todo = sc->sc_clresid; 910 memset((char *)scp->fc_data + todo, 0, FSS_CLSIZE(sc) - todo); 911 } else 912 todo = FSS_CLSIZE(sc); 913 offset = 0; 914 mbp = getiobuf(NULL, true); 915 mbp->b_bufsize = todo; 916 mbp->b_data = scp->fc_data; 917 mbp->b_resid = mbp->b_bcount = todo; 918 mbp->b_flags = B_READ; 919 mbp->b_cflags = BC_BUSY; 920 mbp->b_dev = sc->sc_bdev; 921 while (todo > 0) { 922 len = todo; 923 if (len > MAXPHYS) 924 len = MAXPHYS; 925 if (btodb(FSS_CLTOB(sc, cl)) == dblk && len == todo) 926 bp = mbp; 927 else { 928 bp = getiobuf(NULL, true); 929 nestiobuf_setup(mbp, bp, offset, len); 930 } 931 bp->b_lblkno = 0; 932 bp->b_blkno = dblk; 933 bdev_strategy(bp); 934 dblk += btodb(len); 935 offset += len; 936 todo -= len; 937 } 938 error = biowait(mbp); 939 putiobuf(mbp); 940 941 mutex_enter(&sc->sc_slock); 942 scp->fc_type = (error ? FSS_CACHE_FREE : FSS_CACHE_VALID); 943 cv_broadcast(&scp->fc_state_cv); 944 if (error == 0) { 945 setbit(sc->sc_copied, scp->fc_cluster); 946 cv_signal(&sc->sc_work_cv); 947 } 948 mutex_exit(&sc->sc_slock); 949 950 return error; 951 } 952 953 /* 954 * Read/write clusters from/to backing store. 955 * For persistent snapshots must be called with cl == 0. off is the 956 * offset into the snapshot. 957 */ 958 static int 959 fss_bs_io(struct fss_softc *sc, fss_io_type rw, 960 u_int32_t cl, off_t off, int len, void *data) 961 { 962 int error; 963 964 off += FSS_CLTOB(sc, cl); 965 966 vn_lock(sc->sc_bs_vp, LK_EXCLUSIVE|LK_RETRY); 967 968 error = vn_rdwr((rw == FSS_READ ? UIO_READ : UIO_WRITE), sc->sc_bs_vp, 969 data, len, off, UIO_SYSSPACE, IO_UNIT|IO_NODELOCKED, 970 sc->sc_bs_lwp->l_cred, NULL, NULL); 971 if (error == 0) { 972 mutex_enter(&sc->sc_bs_vp->v_interlock); 973 error = VOP_PUTPAGES(sc->sc_bs_vp, trunc_page(off), 974 round_page(off+len), PGO_CLEANIT|PGO_SYNCIO|PGO_FREE); 975 } 976 977 VOP_UNLOCK(sc->sc_bs_vp); 978 979 return error; 980 } 981 982 /* 983 * Get a pointer to the indirect slot for this cluster. 984 */ 985 static u_int32_t * 986 fss_bs_indir(struct fss_softc *sc, u_int32_t cl) 987 { 988 u_int32_t icl; 989 int ioff; 990 991 icl = cl/(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 992 ioff = cl%(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 993 994 if (sc->sc_indir_cur == icl) 995 return &sc->sc_indir_data[ioff]; 996 997 if (sc->sc_indir_dirty) { 998 if (fss_bs_io(sc, FSS_WRITE, sc->sc_indir_cur, 0, 999 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 1000 return NULL; 1001 setbit(sc->sc_indir_valid, sc->sc_indir_cur); 1002 } 1003 1004 sc->sc_indir_dirty = 0; 1005 sc->sc_indir_cur = icl; 1006 1007 if (isset(sc->sc_indir_valid, sc->sc_indir_cur)) { 1008 if (fss_bs_io(sc, FSS_READ, sc->sc_indir_cur, 0, 1009 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 1010 return NULL; 1011 } else 1012 memset(sc->sc_indir_data, 0, FSS_CLSIZE(sc)); 1013 1014 return &sc->sc_indir_data[ioff]; 1015 } 1016 1017 /* 1018 * The kernel thread (one for every active snapshot). 1019 * 1020 * After wakeup it cleans the cache and runs the I/O requests. 1021 */ 1022 static void 1023 fss_bs_thread(void *arg) 1024 { 1025 bool thread_idle, is_valid; 1026 int error, i, todo, len, crotor, is_read; 1027 long off; 1028 char *addr; 1029 u_int32_t c, cl, ch, *indirp; 1030 struct buf *bp, *nbp; 1031 struct fss_softc *sc; 1032 struct fss_cache *scp, *scl; 1033 1034 sc = arg; 1035 scl = sc->sc_cache+sc->sc_cache_size; 1036 crotor = 0; 1037 thread_idle = false; 1038 1039 mutex_enter(&sc->sc_slock); 1040 1041 for (;;) { 1042 if (thread_idle) 1043 cv_wait(&sc->sc_work_cv, &sc->sc_slock); 1044 thread_idle = true; 1045 if ((sc->sc_flags & FSS_BS_THREAD) == 0) { 1046 sc->sc_bs_lwp = NULL; 1047 mutex_exit(&sc->sc_slock); 1048 kthread_exit(0); 1049 } 1050 1051 /* 1052 * Process I/O requests (persistent) 1053 */ 1054 1055 if (sc->sc_flags & FSS_PERSISTENT) { 1056 if ((bp = bufq_get(sc->sc_bufq)) == NULL) 1057 continue; 1058 is_valid = FSS_ISVALID(sc); 1059 is_read = (bp->b_flags & B_READ); 1060 thread_idle = false; 1061 mutex_exit(&sc->sc_slock); 1062 1063 if (is_valid) { 1064 disk_busy(sc->sc_dkdev); 1065 error = fss_bs_io(sc, FSS_READ, 0, 1066 dbtob(bp->b_blkno), bp->b_bcount, 1067 bp->b_data); 1068 disk_unbusy(sc->sc_dkdev, 1069 (error ? 0 : bp->b_bcount), is_read); 1070 } else 1071 error = ENXIO; 1072 1073 bp->b_error = error; 1074 bp->b_resid = (error ? bp->b_bcount : 0); 1075 biodone(bp); 1076 1077 mutex_enter(&sc->sc_slock); 1078 continue; 1079 } 1080 1081 /* 1082 * Clean the cache 1083 */ 1084 for (i = 0; i < sc->sc_cache_size; i++) { 1085 crotor = (crotor + 1) % sc->sc_cache_size; 1086 scp = sc->sc_cache + crotor; 1087 if (scp->fc_type != FSS_CACHE_VALID) 1088 continue; 1089 mutex_exit(&sc->sc_slock); 1090 1091 thread_idle = false; 1092 indirp = fss_bs_indir(sc, scp->fc_cluster); 1093 if (indirp != NULL) { 1094 error = fss_bs_io(sc, FSS_WRITE, sc->sc_clnext, 1095 0, FSS_CLSIZE(sc), scp->fc_data); 1096 } else 1097 error = EIO; 1098 1099 mutex_enter(&sc->sc_slock); 1100 if (error == 0) { 1101 *indirp = sc->sc_clnext++; 1102 sc->sc_indir_dirty = 1; 1103 } else 1104 fss_error(sc, "write error on backing store"); 1105 1106 scp->fc_type = FSS_CACHE_FREE; 1107 cv_signal(&sc->sc_cache_cv); 1108 break; 1109 } 1110 1111 /* 1112 * Process I/O requests 1113 */ 1114 if ((bp = bufq_get(sc->sc_bufq)) == NULL) 1115 continue; 1116 is_valid = FSS_ISVALID(sc); 1117 is_read = (bp->b_flags & B_READ); 1118 thread_idle = false; 1119 1120 if (!is_valid) { 1121 mutex_exit(&sc->sc_slock); 1122 1123 bp->b_error = ENXIO; 1124 bp->b_resid = bp->b_bcount; 1125 biodone(bp); 1126 1127 mutex_enter(&sc->sc_slock); 1128 continue; 1129 } 1130 1131 disk_busy(sc->sc_dkdev); 1132 1133 /* 1134 * First read from the snapshotted block device unless 1135 * this request is completely covered by backing store. 1136 */ 1137 1138 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 1139 off = FSS_CLOFF(sc, dbtob(bp->b_blkno)); 1140 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 1141 error = 0; 1142 bp->b_resid = 0; 1143 bp->b_error = 0; 1144 for (c = cl; c <= ch; c++) { 1145 if (isset(sc->sc_copied, c)) 1146 continue; 1147 mutex_exit(&sc->sc_slock); 1148 1149 /* Not on backing store, read from device. */ 1150 nbp = getiobuf(NULL, true); 1151 nbp->b_flags = B_READ; 1152 nbp->b_resid = nbp->b_bcount = bp->b_bcount; 1153 nbp->b_bufsize = bp->b_bcount; 1154 nbp->b_data = bp->b_data; 1155 nbp->b_blkno = bp->b_blkno; 1156 nbp->b_lblkno = 0; 1157 nbp->b_dev = sc->sc_bdev; 1158 SET(nbp->b_cflags, BC_BUSY); /* mark buffer busy */ 1159 1160 bdev_strategy(nbp); 1161 1162 error = biowait(nbp); 1163 if (error != 0) { 1164 bp->b_resid = bp->b_bcount; 1165 bp->b_error = nbp->b_error; 1166 disk_unbusy(sc->sc_dkdev, 0, is_read); 1167 biodone(bp); 1168 } 1169 putiobuf(nbp); 1170 1171 mutex_enter(&sc->sc_slock); 1172 break; 1173 } 1174 if (error) 1175 continue; 1176 1177 /* 1178 * Replace those parts that have been saved to backing store. 1179 */ 1180 1181 addr = bp->b_data; 1182 todo = bp->b_bcount; 1183 for (c = cl; c <= ch; c++, off = 0, todo -= len, addr += len) { 1184 len = FSS_CLSIZE(sc)-off; 1185 if (len > todo) 1186 len = todo; 1187 if (isclr(sc->sc_copied, c)) 1188 continue; 1189 mutex_exit(&sc->sc_slock); 1190 1191 indirp = fss_bs_indir(sc, c); 1192 if (indirp == NULL || *indirp == 0) { 1193 /* 1194 * Not on backing store. Either in cache 1195 * or hole in the snapshotted block device. 1196 */ 1197 1198 mutex_enter(&sc->sc_slock); 1199 for (scp = sc->sc_cache; scp < scl; scp++) 1200 if (scp->fc_type == FSS_CACHE_VALID && 1201 scp->fc_cluster == c) 1202 break; 1203 if (scp < scl) 1204 memcpy(addr, (char *)scp->fc_data+off, 1205 len); 1206 else 1207 memset(addr, 0, len); 1208 continue; 1209 } 1210 1211 /* 1212 * Read from backing store. 1213 */ 1214 error = 1215 fss_bs_io(sc, FSS_READ, *indirp, off, len, addr); 1216 1217 mutex_enter(&sc->sc_slock); 1218 if (error) { 1219 bp->b_resid = bp->b_bcount; 1220 bp->b_error = error; 1221 break; 1222 } 1223 } 1224 mutex_exit(&sc->sc_slock); 1225 1226 disk_unbusy(sc->sc_dkdev, (error ? 0 : bp->b_bcount), is_read); 1227 biodone(bp); 1228 1229 mutex_enter(&sc->sc_slock); 1230 } 1231 } 1232 1233 #ifdef _MODULE 1234 1235 #include <sys/module.h> 1236 1237 MODULE(MODULE_CLASS_DRIVER, fss, NULL); 1238 CFDRIVER_DECL(fss, DV_DISK, NULL); 1239 1240 static int 1241 fss_modcmd(modcmd_t cmd, void *arg) 1242 { 1243 int bmajor = -1, cmajor = -1, error = 0; 1244 1245 switch (cmd) { 1246 case MODULE_CMD_INIT: 1247 mutex_init(&fss_device_lock, MUTEX_DEFAULT, IPL_NONE); 1248 error = config_cfdriver_attach(&fss_cd); 1249 if (error) { 1250 mutex_destroy(&fss_device_lock); 1251 break; 1252 } 1253 error = config_cfattach_attach(fss_cd.cd_name, &fss_ca); 1254 if (error) { 1255 config_cfdriver_detach(&fss_cd); 1256 mutex_destroy(&fss_device_lock); 1257 break; 1258 } 1259 error = devsw_attach(fss_cd.cd_name, 1260 &fss_bdevsw, &bmajor, &fss_cdevsw, &cmajor); 1261 if (error == EEXIST) 1262 error = 0; 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