1/* $NetBSD: fss.c,v 1.81.4.2 2013/02/10 23:57:25 riz 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 1.81.4.2 2013/02/10 23:57:25 riz 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#include <sys/vfs_syscalls.h> /* For do_sys_unlink(). */ 62 63#include <miscfs/specfs/specdev.h> 64 65#include <dev/fssvar.h> 66 67#include <uvm/uvm.h> 68 69void fssattach(int); 70 71dev_type_open(fss_open); 72dev_type_close(fss_close); 73dev_type_read(fss_read); 74dev_type_write(fss_write); 75dev_type_ioctl(fss_ioctl); 76dev_type_strategy(fss_strategy); 77dev_type_dump(fss_dump); 78dev_type_size(fss_size); 79 80static void fss_unmount_hook(struct mount *); 81static int fss_copy_on_write(void *, struct buf *, bool); 82static inline void fss_error(struct fss_softc *, const char *); 83static int fss_create_files(struct fss_softc *, struct fss_set *, 84 off_t *, struct lwp *); 85static int fss_create_snapshot(struct fss_softc *, struct fss_set *, 86 struct lwp *); 87static int fss_delete_snapshot(struct fss_softc *, struct lwp *); 88static int fss_softc_alloc(struct fss_softc *); 89static void fss_softc_free(struct fss_softc *); 90static int fss_read_cluster(struct fss_softc *, u_int32_t); 91static void fss_bs_thread(void *); 92static int fss_bs_io(struct fss_softc *, fss_io_type, 93 u_int32_t, off_t, int, void *); 94static u_int32_t *fss_bs_indir(struct fss_softc *, u_int32_t); 95 96static kmutex_t fss_device_lock; /* Protect all units. */ 97static int fss_num_attached = 0; /* Number of attached devices. */ 98static struct vfs_hooks fss_vfs_hooks = { 99 .vh_unmount = fss_unmount_hook 100}; 101 102const struct bdevsw fss_bdevsw = { 103 fss_open, fss_close, fss_strategy, fss_ioctl, 104 fss_dump, fss_size, D_DISK | D_MPSAFE 105}; 106 107const struct cdevsw fss_cdevsw = { 108 fss_open, fss_close, fss_read, fss_write, fss_ioctl, 109 nostop, notty, nopoll, nommap, nokqfilter, D_DISK | D_MPSAFE 110}; 111 112static int fss_match(device_t, cfdata_t, void *); 113static void fss_attach(device_t, device_t, void *); 114static int fss_detach(device_t, int); 115 116CFATTACH_DECL_NEW(fss, sizeof(struct fss_softc), 117 fss_match, fss_attach, fss_detach, NULL); 118extern struct cfdriver fss_cd; 119 120void 121fssattach(int num) 122{ 123 124 mutex_init(&fss_device_lock, MUTEX_DEFAULT, IPL_NONE); 125 if (config_cfattach_attach(fss_cd.cd_name, &fss_ca)) 126 aprint_error("%s: unable to register\n", fss_cd.cd_name); 127} 128 129static int 130fss_match(device_t self, cfdata_t cfdata, void *aux) 131{ 132 return 1; 133} 134 135static void 136fss_attach(device_t parent, device_t self, void *aux) 137{ 138 struct fss_softc *sc = device_private(self); 139 140 sc->sc_dev = self; 141 sc->sc_bdev = NODEV; 142 mutex_init(&sc->sc_slock, MUTEX_DEFAULT, IPL_NONE); 143 mutex_init(&sc->sc_lock, MUTEX_DEFAULT, IPL_NONE); 144 cv_init(&sc->sc_work_cv, "fssbs"); 145 cv_init(&sc->sc_cache_cv, "cowwait"); 146 bufq_alloc(&sc->sc_bufq, "fcfs", 0); 147 sc->sc_dkdev = malloc(sizeof(*sc->sc_dkdev), M_DEVBUF, M_WAITOK); 148 sc->sc_dkdev->dk_info = NULL; 149 disk_init(sc->sc_dkdev, device_xname(self), NULL); 150 if (!pmf_device_register(self, NULL, NULL)) 151 aprint_error_dev(self, "couldn't establish power handler\n"); 152 153 if (fss_num_attached++ == 0) 154 vfs_hooks_attach(&fss_vfs_hooks); 155} 156 157static int 158fss_detach(device_t self, int flags) 159{ 160 struct fss_softc *sc = device_private(self); 161 162 if (sc->sc_flags & FSS_ACTIVE) 163 return EBUSY; 164 165 if (--fss_num_attached == 0) 166 vfs_hooks_detach(&fss_vfs_hooks); 167 168 pmf_device_deregister(self); 169 mutex_destroy(&sc->sc_slock); 170 mutex_destroy(&sc->sc_lock); 171 cv_destroy(&sc->sc_work_cv); 172 cv_destroy(&sc->sc_cache_cv); 173 bufq_drain(sc->sc_bufq); 174 bufq_free(sc->sc_bufq); 175 disk_destroy(sc->sc_dkdev); 176 free(sc->sc_dkdev, M_DEVBUF); 177 178 return 0; 179} 180 181int 182fss_open(dev_t dev, int flags, int mode, struct lwp *l) 183{ 184 int mflag; 185 cfdata_t cf; 186 struct fss_softc *sc; 187 188 mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN); 189 190 mutex_enter(&fss_device_lock); 191 192 sc = device_lookup_private(&fss_cd, minor(dev)); 193 if (sc == NULL) { 194 cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK); 195 cf->cf_name = fss_cd.cd_name; 196 cf->cf_atname = fss_cd.cd_name; 197 cf->cf_unit = minor(dev); 198 cf->cf_fstate = FSTATE_STAR; 199 sc = device_private(config_attach_pseudo(cf)); 200 if (sc == NULL) { 201 mutex_exit(&fss_device_lock); 202 return ENOMEM; 203 } 204 } 205 206 mutex_enter(&sc->sc_slock); 207 208 sc->sc_flags |= mflag; 209 210 mutex_exit(&sc->sc_slock); 211 mutex_exit(&fss_device_lock); 212 213 return 0; 214} 215 216int 217fss_close(dev_t dev, int flags, int mode, struct lwp *l) 218{ 219 int mflag, error; 220 cfdata_t cf; 221 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(dev)); 222 223 mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN); 224 error = 0; 225 226 mutex_enter(&fss_device_lock); 227restart: 228 mutex_enter(&sc->sc_slock); 229 if ((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) != mflag) { 230 sc->sc_flags &= ~mflag; 231 mutex_exit(&sc->sc_slock); 232 mutex_exit(&fss_device_lock); 233 return 0; 234 } 235 if ((sc->sc_flags & FSS_ACTIVE) != 0 && 236 (sc->sc_uflags & FSS_UNCONFIG_ON_CLOSE) != 0) { 237 sc->sc_uflags &= ~FSS_UNCONFIG_ON_CLOSE; 238 mutex_exit(&sc->sc_slock); 239 error = fss_ioctl(dev, FSSIOCCLR, NULL, FWRITE, l); 240 goto restart; 241 } 242 if ((sc->sc_flags & FSS_ACTIVE) != 0) { 243 mutex_exit(&sc->sc_slock); 244 mutex_exit(&fss_device_lock); 245 return error; 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 260void 261fss_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 285int 286fss_read(dev_t dev, struct uio *uio, int flags) 287{ 288 return physio(fss_strategy, NULL, dev, B_READ, minphys, uio); 289} 290 291int 292fss_write(dev_t dev, struct uio *uio, int flags) 293{ 294 return physio(fss_strategy, NULL, dev, B_WRITE, minphys, uio); 295} 296 297int 298fss_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; 303 struct fss_set *fss = (struct fss_set *)data; 304 struct fss_set50 *fss50 = (struct fss_set50 *)data; 305 struct fss_get *fsg = (struct fss_get *)data; 306#ifndef _LP64 307 struct fss_get50 *fsg50 = (struct fss_get50 *)data; 308#endif 309 310 switch (cmd) { 311 case FSSIOCSET50: 312 fss = &_fss; 313 fss->fss_mount = fss50->fss_mount; 314 fss->fss_bstore = fss50->fss_bstore; 315 fss->fss_csize = fss50->fss_csize; 316 fss->fss_flags = 0; 317 /* Fall through */ 318 case FSSIOCSET: 319 mutex_enter(&sc->sc_lock); 320 if ((flag & FWRITE) == 0) 321 error = EPERM; 322 else if ((sc->sc_flags & FSS_ACTIVE) != 0) 323 error = EBUSY; 324 else 325 error = fss_create_snapshot(sc, fss, l); 326 if (error == 0) 327 sc->sc_uflags = fss->fss_flags; 328 mutex_exit(&sc->sc_lock); 329 break; 330 331 case FSSIOCCLR: 332 mutex_enter(&sc->sc_lock); 333 if ((flag & FWRITE) == 0) 334 error = EPERM; 335 else if ((sc->sc_flags & FSS_ACTIVE) == 0) 336 error = ENXIO; 337 else 338 error = fss_delete_snapshot(sc, l); 339 mutex_exit(&sc->sc_lock); 340 break; 341 342#ifndef _LP64 343 case FSSIOCGET50: 344 mutex_enter(&sc->sc_lock); 345 switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) { 346 case FSS_ACTIVE: 347 memcpy(fsg50->fsg_mount, sc->sc_mntname, MNAMELEN); 348 fsg50->fsg_csize = FSS_CLSIZE(sc); 349 timeval_to_timeval50(&sc->sc_time, &fsg50->fsg_time); 350 fsg50->fsg_mount_size = sc->sc_clcount; 351 fsg50->fsg_bs_size = sc->sc_clnext; 352 error = 0; 353 break; 354 case FSS_PERSISTENT | FSS_ACTIVE: 355 memcpy(fsg50->fsg_mount, sc->sc_mntname, MNAMELEN); 356 fsg50->fsg_csize = 0; 357 timeval_to_timeval50(&sc->sc_time, &fsg50->fsg_time); 358 fsg50->fsg_mount_size = 0; 359 fsg50->fsg_bs_size = 0; 360 error = 0; 361 break; 362 default: 363 error = ENXIO; 364 break; 365 } 366 mutex_exit(&sc->sc_lock); 367 break; 368#endif /* _LP64 */ 369 370 case FSSIOCGET: 371 mutex_enter(&sc->sc_lock); 372 switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) { 373 case FSS_ACTIVE: 374 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 375 fsg->fsg_csize = FSS_CLSIZE(sc); 376 fsg->fsg_time = sc->sc_time; 377 fsg->fsg_mount_size = sc->sc_clcount; 378 fsg->fsg_bs_size = sc->sc_clnext; 379 error = 0; 380 break; 381 case FSS_PERSISTENT | FSS_ACTIVE: 382 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 383 fsg->fsg_csize = 0; 384 fsg->fsg_time = sc->sc_time; 385 fsg->fsg_mount_size = 0; 386 fsg->fsg_bs_size = 0; 387 error = 0; 388 break; 389 default: 390 error = ENXIO; 391 break; 392 } 393 mutex_exit(&sc->sc_lock); 394 break; 395 396 case FSSIOFSET: 397 mutex_enter(&sc->sc_slock); 398 sc->sc_uflags = *(int *)data; 399 mutex_exit(&sc->sc_slock); 400 error = 0; 401 break; 402 403 case FSSIOFGET: 404 mutex_enter(&sc->sc_slock); 405 *(int *)data = sc->sc_uflags; 406 mutex_exit(&sc->sc_slock); 407 error = 0; 408 break; 409 410 default: 411 error = EINVAL; 412 break; 413 } 414 415 return error; 416} 417 418int 419fss_size(dev_t dev) 420{ 421 return -1; 422} 423 424int 425fss_dump(dev_t dev, daddr_t blkno, void *va, 426 size_t size) 427{ 428 return EROFS; 429} 430 431/* 432 * An error occurred reading or writing the snapshot or backing store. 433 * If it is the first error log to console. 434 * The caller holds the mutex. 435 */ 436static inline void 437fss_error(struct fss_softc *sc, const char *msg) 438{ 439 440 if ((sc->sc_flags & (FSS_ACTIVE|FSS_ERROR)) == FSS_ACTIVE) 441 aprint_error_dev(sc->sc_dev, "snapshot invalid: %s\n", msg); 442 if ((sc->sc_flags & FSS_ACTIVE) == FSS_ACTIVE) 443 sc->sc_flags |= FSS_ERROR; 444} 445 446/* 447 * Allocate the variable sized parts of the softc and 448 * fork the kernel thread. 449 * 450 * The fields sc_clcount, sc_clshift, sc_cache_size and sc_indir_size 451 * must be initialized. 452 */ 453static int 454fss_softc_alloc(struct fss_softc *sc) 455{ 456 int i, error; 457 458 if ((sc->sc_flags & FSS_PERSISTENT) == 0) { 459 sc->sc_copied = 460 kmem_zalloc(howmany(sc->sc_clcount, NBBY), KM_SLEEP); 461 if (sc->sc_copied == NULL) 462 return(ENOMEM); 463 464 sc->sc_cache = kmem_alloc(sc->sc_cache_size * 465 sizeof(struct fss_cache), KM_SLEEP); 466 if (sc->sc_cache == NULL) 467 return(ENOMEM); 468 469 for (i = 0; i < sc->sc_cache_size; i++) { 470 sc->sc_cache[i].fc_type = FSS_CACHE_FREE; 471 sc->sc_cache[i].fc_data = 472 kmem_alloc(FSS_CLSIZE(sc), KM_SLEEP); 473 if (sc->sc_cache[i].fc_data == NULL) 474 return(ENOMEM); 475 cv_init(&sc->sc_cache[i].fc_state_cv, "cowwait1"); 476 } 477 478 sc->sc_indir_valid = 479 kmem_zalloc(howmany(sc->sc_indir_size, NBBY), KM_SLEEP); 480 if (sc->sc_indir_valid == NULL) 481 return(ENOMEM); 482 483 sc->sc_indir_data = kmem_zalloc(FSS_CLSIZE(sc), KM_SLEEP); 484 if (sc->sc_indir_data == NULL) 485 return(ENOMEM); 486 } else { 487 sc->sc_copied = NULL; 488 sc->sc_cache = NULL; 489 sc->sc_indir_valid = NULL; 490 sc->sc_indir_data = NULL; 491 } 492 493 sc->sc_flags |= FSS_BS_THREAD; 494 if ((error = kthread_create(PRI_BIO, KTHREAD_MUSTJOIN, NULL, 495 fss_bs_thread, sc, &sc->sc_bs_lwp, 496 "%s", device_xname(sc->sc_dev))) != 0) { 497 sc->sc_flags &= ~FSS_BS_THREAD; 498 return error; 499 } 500 501 disk_attach(sc->sc_dkdev); 502 503 return 0; 504} 505 506/* 507 * Free the variable sized parts of the softc. 508 */ 509static void 510fss_softc_free(struct fss_softc *sc) 511{ 512 int i; 513 514 if ((sc->sc_flags & FSS_BS_THREAD) != 0) { 515 mutex_enter(&sc->sc_slock); 516 sc->sc_flags &= ~FSS_BS_THREAD; 517 cv_signal(&sc->sc_work_cv); 518 mutex_exit(&sc->sc_slock); 519 kthread_join(sc->sc_bs_lwp); 520 521 disk_detach(sc->sc_dkdev); 522 } 523 524 if (sc->sc_copied != NULL) 525 kmem_free(sc->sc_copied, howmany(sc->sc_clcount, NBBY)); 526 sc->sc_copied = NULL; 527 528 if (sc->sc_cache != NULL) { 529 for (i = 0; i < sc->sc_cache_size; i++) 530 if (sc->sc_cache[i].fc_data != NULL) { 531 cv_destroy(&sc->sc_cache[i].fc_state_cv); 532 kmem_free(sc->sc_cache[i].fc_data, 533 FSS_CLSIZE(sc)); 534 } 535 kmem_free(sc->sc_cache, 536 sc->sc_cache_size*sizeof(struct fss_cache)); 537 } 538 sc->sc_cache = NULL; 539 540 if (sc->sc_indir_valid != NULL) 541 kmem_free(sc->sc_indir_valid, howmany(sc->sc_indir_size, NBBY)); 542 sc->sc_indir_valid = NULL; 543 544 if (sc->sc_indir_data != NULL) 545 kmem_free(sc->sc_indir_data, FSS_CLSIZE(sc)); 546 sc->sc_indir_data = NULL; 547} 548 549/* 550 * Set all active snapshots on this file system into ERROR state. 551 */ 552static void 553fss_unmount_hook(struct mount *mp) 554{ 555 int i; 556 struct fss_softc *sc; 557 558 mutex_enter(&fss_device_lock); 559 for (i = 0; i < fss_cd.cd_ndevs; i++) { 560 if ((sc = device_lookup_private(&fss_cd, i)) == NULL) 561 continue; 562 mutex_enter(&sc->sc_slock); 563 if ((sc->sc_flags & FSS_ACTIVE) != 0 && 564 sc->sc_mount == mp) 565 fss_error(sc, "forced unmount"); 566 mutex_exit(&sc->sc_slock); 567 } 568 mutex_exit(&fss_device_lock); 569} 570 571/* 572 * A buffer is written to the snapshotted block device. Copy to 573 * backing store if needed. 574 */ 575static int 576fss_copy_on_write(void *v, struct buf *bp, bool data_valid) 577{ 578 int error; 579 u_int32_t cl, ch, c; 580 struct fss_softc *sc = v; 581 582 mutex_enter(&sc->sc_slock); 583 if (!FSS_ISVALID(sc)) { 584 mutex_exit(&sc->sc_slock); 585 return 0; 586 } 587 588 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 589 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 590 error = 0; 591 if (curlwp == uvm.pagedaemon_lwp) { 592 for (c = cl; c <= ch; c++) 593 if (isclr(sc->sc_copied, c)) { 594 error = ENOMEM; 595 break; 596 } 597 } 598 mutex_exit(&sc->sc_slock); 599 600 if (error == 0) 601 for (c = cl; c <= ch; c++) { 602 error = fss_read_cluster(sc, c); 603 if (error) 604 break; 605 } 606 607 return error; 608} 609 610/* 611 * Lookup and open needed files. 612 * 613 * For file system internal snapshot initializes sc_mntname, sc_mount, 614 * sc_bs_vp and sc_time. 615 * 616 * Otherwise returns dev and size of the underlying block device. 617 * Initializes sc_mntname, sc_mount, sc_bdev, sc_bs_vp and sc_mount 618 */ 619static int 620fss_create_files(struct fss_softc *sc, struct fss_set *fss, 621 off_t *bsize, struct lwp *l) 622{ 623 int i, error, bits, fsbsize; 624 uint64_t numsec; 625 unsigned int secsize; 626 struct timespec ts; 627 /* nd -> nd2 to reduce mistakes while updating only some namei calls */ 628 struct pathbuf *pb2; 629 struct nameidata nd2; 630 struct vnode *vp; 631 632 /* 633 * Get the mounted file system. 634 */ 635 636 error = namei_simple_user(fss->fss_mount, 637 NSM_FOLLOW_NOEMULROOT, &vp); 638 if (error != 0) 639 return error; 640 641 if ((vp->v_vflag & VV_ROOT) != VV_ROOT) { 642 vrele(vp); 643 return EINVAL; 644 } 645 646 sc->sc_mount = vp->v_mount; 647 memcpy(sc->sc_mntname, sc->sc_mount->mnt_stat.f_mntonname, MNAMELEN); 648 649 vrele(vp); 650 651 /* 652 * Check for file system internal snapshot. 653 */ 654 655 error = namei_simple_user(fss->fss_bstore, 656 NSM_FOLLOW_NOEMULROOT, &vp); 657 if (error != 0) 658 return error; 659 660 if (vp->v_type == VREG && vp->v_mount == sc->sc_mount) { 661 sc->sc_flags |= FSS_PERSISTENT; 662 sc->sc_bs_vp = vp; 663 664 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 665 bits = sizeof(sc->sc_bs_bshift)*NBBY; 666 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < bits; 667 sc->sc_bs_bshift++) 668 if (FSS_FSBSIZE(sc) == fsbsize) 669 break; 670 if (sc->sc_bs_bshift >= bits) 671 return EINVAL; 672 673 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 674 sc->sc_clshift = 0; 675 676 if ((fss->fss_flags & FSS_UNLINK_ON_CREATE) != 0) { 677 error = do_sys_unlink(fss->fss_bstore, UIO_USERSPACE); 678 if (error) 679 return error; 680 } 681 error = vn_lock(vp, LK_EXCLUSIVE); 682 if (error != 0) 683 return error; 684 error = VFS_SNAPSHOT(sc->sc_mount, sc->sc_bs_vp, &ts); 685 TIMESPEC_TO_TIMEVAL(&sc->sc_time, &ts); 686 687 VOP_UNLOCK(sc->sc_bs_vp); 688 689 return error; 690 } 691 vrele(vp); 692 693 /* 694 * Get the block device it is mounted on and its size. 695 */ 696 697 mutex_enter(&device_lock); 698 for (i = 0; i < SPECHSZ; i++) { 699 for (vp = specfs_hash[i]; vp; vp = vp->v_specnext) { 700 if (vp->v_type == VBLK && 701 vp == vp->v_specnode->sn_dev->sd_bdevvp && 702 vp->v_specmountpoint == sc->sc_mount) 703 break; 704 } 705 if (vp != NULL) 706 break; 707 } 708 if (vp == NULL) { 709 mutex_exit(&device_lock); 710 return EINVAL; 711 } 712 mutex_enter(vp->v_interlock); 713 mutex_exit(&device_lock); 714 error = vget(vp, 0); 715 if (error) 716 return error; 717 sc->sc_bdev = vp->v_rdev; 718 719 error = getdisksize(vp, &numsec, &secsize); 720 vrele(vp); 721 if (error) 722 return error; 723 724 *bsize = (off_t)numsec*secsize; 725 726 /* 727 * Get the backing store 728 */ 729 730 error = pathbuf_copyin(fss->fss_bstore, &pb2); 731 if (error) { 732 return error; 733 } 734 NDINIT(&nd2, LOOKUP, FOLLOW, pb2); 735 if ((error = vn_open(&nd2, FREAD|FWRITE, 0)) != 0) { 736 pathbuf_destroy(pb2); 737 return error; 738 } 739 VOP_UNLOCK(nd2.ni_vp); 740 741 sc->sc_bs_vp = nd2.ni_vp; 742 743 if (nd2.ni_vp->v_type != VREG && nd2.ni_vp->v_type != VCHR) { 744 pathbuf_destroy(pb2); 745 return EINVAL; 746 } 747 pathbuf_destroy(pb2); 748 749 if ((fss->fss_flags & FSS_UNLINK_ON_CREATE) != 0) { 750 error = do_sys_unlink(fss->fss_bstore, UIO_USERSPACE); 751 if (error) 752 return error; 753 } 754 if (sc->sc_bs_vp->v_type == VREG) { 755 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 756 if (fsbsize & (fsbsize-1)) /* No power of two */ 757 return EINVAL; 758 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < 32; 759 sc->sc_bs_bshift++) 760 if (FSS_FSBSIZE(sc) == fsbsize) 761 break; 762 if (sc->sc_bs_bshift >= 32) 763 return EINVAL; 764 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 765 } else { 766 sc->sc_bs_bshift = DEV_BSHIFT; 767 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 768 } 769 770 return 0; 771} 772 773/* 774 * Create a snapshot. 775 */ 776static int 777fss_create_snapshot(struct fss_softc *sc, struct fss_set *fss, struct lwp *l) 778{ 779 int len, error; 780 u_int32_t csize; 781 off_t bsize; 782 783 bsize = 0; /* XXX gcc */ 784 785 /* 786 * Open needed files. 787 */ 788 if ((error = fss_create_files(sc, fss, &bsize, l)) != 0) 789 goto bad; 790 791 if (sc->sc_flags & FSS_PERSISTENT) { 792 fss_softc_alloc(sc); 793 sc->sc_flags |= FSS_ACTIVE; 794 return 0; 795 } 796 797 /* 798 * Set cluster size. Must be a power of two and 799 * a multiple of backing store block size. 800 */ 801 if (fss->fss_csize <= 0) 802 csize = MAXPHYS; 803 else 804 csize = fss->fss_csize; 805 if (bsize/csize > FSS_CLUSTER_MAX) 806 csize = bsize/FSS_CLUSTER_MAX+1; 807 808 for (sc->sc_clshift = sc->sc_bs_bshift; sc->sc_clshift < 32; 809 sc->sc_clshift++) 810 if (FSS_CLSIZE(sc) >= csize) 811 break; 812 if (sc->sc_clshift >= 32) { 813 error = EINVAL; 814 goto bad; 815 } 816 sc->sc_clmask = FSS_CLSIZE(sc)-1; 817 818 /* 819 * Set number of cache slots. 820 */ 821 if (FSS_CLSIZE(sc) <= 8192) 822 sc->sc_cache_size = 32; 823 else if (FSS_CLSIZE(sc) <= 65536) 824 sc->sc_cache_size = 8; 825 else 826 sc->sc_cache_size = 4; 827 828 /* 829 * Set number of clusters and size of last cluster. 830 */ 831 sc->sc_clcount = FSS_BTOCL(sc, bsize-1)+1; 832 sc->sc_clresid = FSS_CLOFF(sc, bsize-1)+1; 833 834 /* 835 * Set size of indirect table. 836 */ 837 len = sc->sc_clcount*sizeof(u_int32_t); 838 sc->sc_indir_size = FSS_BTOCL(sc, len)+1; 839 sc->sc_clnext = sc->sc_indir_size; 840 sc->sc_indir_cur = 0; 841 842 if ((error = fss_softc_alloc(sc)) != 0) 843 goto bad; 844 845 /* 846 * Activate the snapshot. 847 */ 848 849 if ((error = vfs_suspend(sc->sc_mount, 0)) != 0) 850 goto bad; 851 852 microtime(&sc->sc_time); 853 854 error = fscow_establish(sc->sc_mount, fss_copy_on_write, sc); 855 if (error == 0) 856 sc->sc_flags |= FSS_ACTIVE; 857 858 vfs_resume(sc->sc_mount); 859 860 if (error != 0) 861 goto bad; 862 863 aprint_debug_dev(sc->sc_dev, "%s snapshot active\n", sc->sc_mntname); 864 aprint_debug_dev(sc->sc_dev, 865 "%u clusters of %u, %u cache slots, %u indir clusters\n", 866 sc->sc_clcount, FSS_CLSIZE(sc), 867 sc->sc_cache_size, sc->sc_indir_size); 868 869 return 0; 870 871bad: 872 fss_softc_free(sc); 873 if (sc->sc_bs_vp != NULL) { 874 if (sc->sc_flags & FSS_PERSISTENT) 875 vrele(sc->sc_bs_vp); 876 else 877 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 878 } 879 sc->sc_bs_vp = NULL; 880 881 return error; 882} 883 884/* 885 * Delete a snapshot. 886 */ 887static int 888fss_delete_snapshot(struct fss_softc *sc, struct lwp *l) 889{ 890 891 if ((sc->sc_flags & FSS_PERSISTENT) == 0) 892 fscow_disestablish(sc->sc_mount, fss_copy_on_write, sc); 893 894 mutex_enter(&sc->sc_slock); 895 sc->sc_flags &= ~(FSS_ACTIVE|FSS_ERROR); 896 sc->sc_mount = NULL; 897 sc->sc_bdev = NODEV; 898 mutex_exit(&sc->sc_slock); 899 900 fss_softc_free(sc); 901 if (sc->sc_flags & FSS_PERSISTENT) 902 vrele(sc->sc_bs_vp); 903 else 904 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 905 sc->sc_bs_vp = NULL; 906 sc->sc_flags &= ~FSS_PERSISTENT; 907 908 return 0; 909} 910 911/* 912 * Read a cluster from the snapshotted block device to the cache. 913 */ 914static int 915fss_read_cluster(struct fss_softc *sc, u_int32_t cl) 916{ 917 int error, todo, offset, len; 918 daddr_t dblk; 919 struct buf *bp, *mbp; 920 struct fss_cache *scp, *scl; 921 922 /* 923 * Get a free cache slot. 924 */ 925 scl = sc->sc_cache+sc->sc_cache_size; 926 927 mutex_enter(&sc->sc_slock); 928 929restart: 930 if (isset(sc->sc_copied, cl) || !FSS_ISVALID(sc)) { 931 mutex_exit(&sc->sc_slock); 932 return 0; 933 } 934 935 for (scp = sc->sc_cache; scp < scl; scp++) 936 if (scp->fc_cluster == cl) { 937 if (scp->fc_type == FSS_CACHE_VALID) { 938 mutex_exit(&sc->sc_slock); 939 return 0; 940 } else if (scp->fc_type == FSS_CACHE_BUSY) { 941 cv_wait(&scp->fc_state_cv, &sc->sc_slock); 942 goto restart; 943 } 944 } 945 946 for (scp = sc->sc_cache; scp < scl; scp++) 947 if (scp->fc_type == FSS_CACHE_FREE) { 948 scp->fc_type = FSS_CACHE_BUSY; 949 scp->fc_cluster = cl; 950 break; 951 } 952 if (scp >= scl) { 953 cv_wait(&sc->sc_cache_cv, &sc->sc_slock); 954 goto restart; 955 } 956 957 mutex_exit(&sc->sc_slock); 958 959 /* 960 * Start the read. 961 */ 962 dblk = btodb(FSS_CLTOB(sc, cl)); 963 if (cl == sc->sc_clcount-1) { 964 todo = sc->sc_clresid; 965 memset((char *)scp->fc_data + todo, 0, FSS_CLSIZE(sc) - todo); 966 } else 967 todo = FSS_CLSIZE(sc); 968 offset = 0; 969 mbp = getiobuf(NULL, true); 970 mbp->b_bufsize = todo; 971 mbp->b_data = scp->fc_data; 972 mbp->b_resid = mbp->b_bcount = todo; 973 mbp->b_flags = B_READ; 974 mbp->b_cflags = BC_BUSY; 975 mbp->b_dev = sc->sc_bdev; 976 while (todo > 0) { 977 len = todo; 978 if (len > MAXPHYS) 979 len = MAXPHYS; 980 if (btodb(FSS_CLTOB(sc, cl)) == dblk && len == todo) 981 bp = mbp; 982 else { 983 bp = getiobuf(NULL, true); 984 nestiobuf_setup(mbp, bp, offset, len); 985 } 986 bp->b_lblkno = 0; 987 bp->b_blkno = dblk; 988 bdev_strategy(bp); 989 dblk += btodb(len); 990 offset += len; 991 todo -= len; 992 } 993 error = biowait(mbp); 994 putiobuf(mbp); 995 996 mutex_enter(&sc->sc_slock); 997 scp->fc_type = (error ? FSS_CACHE_FREE : FSS_CACHE_VALID); 998 cv_broadcast(&scp->fc_state_cv); 999 if (error == 0) { 1000 setbit(sc->sc_copied, scp->fc_cluster); 1001 cv_signal(&sc->sc_work_cv); 1002 } 1003 mutex_exit(&sc->sc_slock); 1004 1005 return error; 1006} 1007 1008/* 1009 * Read/write clusters from/to backing store. 1010 * For persistent snapshots must be called with cl == 0. off is the 1011 * offset into the snapshot. 1012 */ 1013static int 1014fss_bs_io(struct fss_softc *sc, fss_io_type rw, 1015 u_int32_t cl, off_t off, int len, void *data) 1016{ 1017 int error; 1018 1019 off += FSS_CLTOB(sc, cl); 1020 1021 vn_lock(sc->sc_bs_vp, LK_EXCLUSIVE|LK_RETRY); 1022 1023 error = vn_rdwr((rw == FSS_READ ? UIO_READ : UIO_WRITE), sc->sc_bs_vp, 1024 data, len, off, UIO_SYSSPACE, 1025 IO_ADV_ENCODE(POSIX_FADV_NOREUSE) | IO_NODELOCKED, 1026 sc->sc_bs_lwp->l_cred, NULL, NULL); 1027 if (error == 0) { 1028 mutex_enter(sc->sc_bs_vp->v_interlock); 1029 error = VOP_PUTPAGES(sc->sc_bs_vp, trunc_page(off), 1030 round_page(off+len), PGO_CLEANIT | PGO_FREE | PGO_SYNCIO); 1031 } 1032 1033 VOP_UNLOCK(sc->sc_bs_vp); 1034 1035 return error; 1036} 1037 1038/* 1039 * Get a pointer to the indirect slot for this cluster. 1040 */ 1041static u_int32_t * 1042fss_bs_indir(struct fss_softc *sc, u_int32_t cl) 1043{ 1044 u_int32_t icl; 1045 int ioff; 1046 1047 icl = cl/(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 1048 ioff = cl%(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 1049 1050 if (sc->sc_indir_cur == icl) 1051 return &sc->sc_indir_data[ioff]; 1052 1053 if (sc->sc_indir_dirty) { 1054 if (fss_bs_io(sc, FSS_WRITE, sc->sc_indir_cur, 0, 1055 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 1056 return NULL; 1057 setbit(sc->sc_indir_valid, sc->sc_indir_cur); 1058 } 1059 1060 sc->sc_indir_dirty = 0; 1061 sc->sc_indir_cur = icl; 1062 1063 if (isset(sc->sc_indir_valid, sc->sc_indir_cur)) { 1064 if (fss_bs_io(sc, FSS_READ, sc->sc_indir_cur, 0, 1065 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 1066 return NULL; 1067 } else 1068 memset(sc->sc_indir_data, 0, FSS_CLSIZE(sc)); 1069 1070 return &sc->sc_indir_data[ioff]; 1071} 1072 1073/* 1074 * The kernel thread (one for every active snapshot). 1075 * 1076 * After wakeup it cleans the cache and runs the I/O requests. 1077 */ 1078static void 1079fss_bs_thread(void *arg) 1080{ 1081 bool thread_idle, is_valid; 1082 int error, i, todo, len, crotor, is_read; 1083 long off; 1084 char *addr; 1085 u_int32_t c, cl, ch, *indirp; 1086 struct buf *bp, *nbp; 1087 struct fss_softc *sc; 1088 struct fss_cache *scp, *scl; 1089 1090 sc = arg; 1091 scl = sc->sc_cache+sc->sc_cache_size; 1092 crotor = 0; 1093 thread_idle = false; 1094 1095 mutex_enter(&sc->sc_slock); 1096 1097 for (;;) { 1098 if (thread_idle) 1099 cv_wait(&sc->sc_work_cv, &sc->sc_slock); 1100 thread_idle = true; 1101 if ((sc->sc_flags & FSS_BS_THREAD) == 0) { 1102 mutex_exit(&sc->sc_slock); 1103 kthread_exit(0); 1104 } 1105 1106 /* 1107 * Process I/O requests (persistent) 1108 */ 1109 1110 if (sc->sc_flags & FSS_PERSISTENT) { 1111 if ((bp = bufq_get(sc->sc_bufq)) == NULL) 1112 continue; 1113 is_valid = FSS_ISVALID(sc); 1114 is_read = (bp->b_flags & B_READ); 1115 thread_idle = false; 1116 mutex_exit(&sc->sc_slock); 1117 1118 if (is_valid) { 1119 disk_busy(sc->sc_dkdev); 1120 error = fss_bs_io(sc, FSS_READ, 0, 1121 dbtob(bp->b_blkno), bp->b_bcount, 1122 bp->b_data); 1123 disk_unbusy(sc->sc_dkdev, 1124 (error ? 0 : bp->b_bcount), is_read); 1125 } else 1126 error = ENXIO; 1127 1128 bp->b_error = error; 1129 bp->b_resid = (error ? bp->b_bcount : 0); 1130 biodone(bp); 1131 1132 mutex_enter(&sc->sc_slock); 1133 continue; 1134 } 1135 1136 /* 1137 * Clean the cache 1138 */ 1139 for (i = 0; i < sc->sc_cache_size; i++) { 1140 crotor = (crotor + 1) % sc->sc_cache_size; 1141 scp = sc->sc_cache + crotor; 1142 if (scp->fc_type != FSS_CACHE_VALID) 1143 continue; 1144 mutex_exit(&sc->sc_slock); 1145 1146 thread_idle = false; 1147 indirp = fss_bs_indir(sc, scp->fc_cluster); 1148 if (indirp != NULL) { 1149 error = fss_bs_io(sc, FSS_WRITE, sc->sc_clnext, 1150 0, FSS_CLSIZE(sc), scp->fc_data); 1151 } else 1152 error = EIO; 1153 1154 mutex_enter(&sc->sc_slock); 1155 if (error == 0) { 1156 *indirp = sc->sc_clnext++; 1157 sc->sc_indir_dirty = 1; 1158 } else 1159 fss_error(sc, "write error on backing store"); 1160 1161 scp->fc_type = FSS_CACHE_FREE; 1162 cv_signal(&sc->sc_cache_cv); 1163 break; 1164 } 1165 1166 /* 1167 * Process I/O requests 1168 */ 1169 if ((bp = bufq_get(sc->sc_bufq)) == NULL) 1170 continue; 1171 is_valid = FSS_ISVALID(sc); 1172 is_read = (bp->b_flags & B_READ); 1173 thread_idle = false; 1174 1175 if (!is_valid) { 1176 mutex_exit(&sc->sc_slock); 1177 1178 bp->b_error = ENXIO; 1179 bp->b_resid = bp->b_bcount; 1180 biodone(bp); 1181 1182 mutex_enter(&sc->sc_slock); 1183 continue; 1184 } 1185 1186 disk_busy(sc->sc_dkdev); 1187 1188 /* 1189 * First read from the snapshotted block device unless 1190 * this request is completely covered by backing store. 1191 */ 1192 1193 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 1194 off = FSS_CLOFF(sc, dbtob(bp->b_blkno)); 1195 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 1196 error = 0; 1197 bp->b_resid = 0; 1198 bp->b_error = 0; 1199 for (c = cl; c <= ch; c++) { 1200 if (isset(sc->sc_copied, c)) 1201 continue; 1202 mutex_exit(&sc->sc_slock); 1203 1204 /* Not on backing store, read from device. */ 1205 nbp = getiobuf(NULL, true); 1206 nbp->b_flags = B_READ; 1207 nbp->b_resid = nbp->b_bcount = bp->b_bcount; 1208 nbp->b_bufsize = bp->b_bcount; 1209 nbp->b_data = bp->b_data; 1210 nbp->b_blkno = bp->b_blkno; 1211 nbp->b_lblkno = 0; 1212 nbp->b_dev = sc->sc_bdev; 1213 SET(nbp->b_cflags, BC_BUSY); /* mark buffer busy */ 1214 1215 bdev_strategy(nbp); 1216 1217 error = biowait(nbp); 1218 if (error != 0) { 1219 bp->b_resid = bp->b_bcount; 1220 bp->b_error = nbp->b_error; 1221 disk_unbusy(sc->sc_dkdev, 0, is_read); 1222 biodone(bp); 1223 } 1224 putiobuf(nbp); 1225 1226 mutex_enter(&sc->sc_slock); 1227 break; 1228 } 1229 if (error) 1230 continue; 1231 1232 /* 1233 * Replace those parts that have been saved to backing store. 1234 */ 1235 1236 addr = bp->b_data; 1237 todo = bp->b_bcount; 1238 for (c = cl; c <= ch; c++, off = 0, todo -= len, addr += len) { 1239 len = FSS_CLSIZE(sc)-off; 1240 if (len > todo) 1241 len = todo; 1242 if (isclr(sc->sc_copied, c)) 1243 continue; 1244 mutex_exit(&sc->sc_slock); 1245 1246 indirp = fss_bs_indir(sc, c); 1247 if (indirp == NULL || *indirp == 0) { 1248 /* 1249 * Not on backing store. Either in cache 1250 * or hole in the snapshotted block device. 1251 */ 1252 1253 mutex_enter(&sc->sc_slock); 1254 for (scp = sc->sc_cache; scp < scl; scp++) 1255 if (scp->fc_type == FSS_CACHE_VALID && 1256 scp->fc_cluster == c) 1257 break; 1258 if (scp < scl) 1259 memcpy(addr, (char *)scp->fc_data+off, 1260 len); 1261 else 1262 memset(addr, 0, len); 1263 continue; 1264 } 1265 1266 /* 1267 * Read from backing store. 1268 */ 1269 error = 1270 fss_bs_io(sc, FSS_READ, *indirp, off, len, addr); 1271 1272 mutex_enter(&sc->sc_slock); 1273 if (error) { 1274 bp->b_resid = bp->b_bcount; 1275 bp->b_error = error; 1276 break; 1277 } 1278 } 1279 mutex_exit(&sc->sc_slock); 1280 1281 disk_unbusy(sc->sc_dkdev, (error ? 0 : bp->b_bcount), is_read); 1282 biodone(bp); 1283 1284 mutex_enter(&sc->sc_slock); 1285 } 1286} 1287 1288#ifdef _MODULE 1289 1290#include <sys/module.h> 1291 1292MODULE(MODULE_CLASS_DRIVER, fss, NULL); 1293CFDRIVER_DECL(fss, DV_DISK, NULL); 1294 1295static int 1296fss_modcmd(modcmd_t cmd, void *arg) 1297{ 1298 int bmajor = -1, cmajor = -1, error = 0; 1299 1300 switch (cmd) { 1301 case MODULE_CMD_INIT: 1302 mutex_init(&fss_device_lock, MUTEX_DEFAULT, IPL_NONE); 1303 error = config_cfdriver_attach(&fss_cd); 1304 if (error) { 1305 mutex_destroy(&fss_device_lock); 1306 break; 1307 } 1308 error = config_cfattach_attach(fss_cd.cd_name, &fss_ca); 1309 if (error) { 1310 config_cfdriver_detach(&fss_cd); 1311 mutex_destroy(&fss_device_lock); 1312 break; 1313 } 1314 error = devsw_attach(fss_cd.cd_name, 1315 &fss_bdevsw, &bmajor, &fss_cdevsw, &cmajor); 1316 if (error == EEXIST) 1317 error = 0; 1318 if (error) { 1319 config_cfattach_detach(fss_cd.cd_name, &fss_ca); 1320 config_cfdriver_detach(&fss_cd); 1321 mutex_destroy(&fss_device_lock); 1322 break; 1323 } 1324 break; 1325 1326 case MODULE_CMD_FINI: 1327 error = config_cfattach_detach(fss_cd.cd_name, &fss_ca); 1328 if (error) 1329 break; 1330 config_cfdriver_detach(&fss_cd); 1331 devsw_detach(&fss_bdevsw, &fss_cdevsw); 1332 mutex_destroy(&fss_device_lock); 1333 break; 1334 1335 default: 1336 error = ENOTTY; 1337 break; 1338 } 1339 1340 return error; 1341} 1342 1343#endif /* _MODULE */ 1344