1 /* $NetBSD: fss.c,v 1.91 2014/07/25 08:10:35 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 1.91 2014/07/25 08:10:35 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/vfs_syscalls.h> /* For do_sys_unlink(). */ 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 .d_open = fss_open, 103 .d_close = fss_close, 104 .d_strategy = fss_strategy, 105 .d_ioctl = fss_ioctl, 106 .d_dump = fss_dump, 107 .d_psize = fss_size, 108 .d_discard = nodiscard, 109 .d_flag = D_DISK | D_MPSAFE 110 }; 111 112 const struct cdevsw fss_cdevsw = { 113 .d_open = fss_open, 114 .d_close = fss_close, 115 .d_read = fss_read, 116 .d_write = fss_write, 117 .d_ioctl = fss_ioctl, 118 .d_stop = nostop, 119 .d_tty = notty, 120 .d_poll = nopoll, 121 .d_mmap = nommap, 122 .d_kqfilter = nokqfilter, 123 .d_discard = nodiscard, 124 .d_flag = D_DISK | D_MPSAFE 125 }; 126 127 static int fss_match(device_t, cfdata_t, void *); 128 static void fss_attach(device_t, device_t, void *); 129 static int fss_detach(device_t, int); 130 131 CFATTACH_DECL_NEW(fss, sizeof(struct fss_softc), 132 fss_match, fss_attach, fss_detach, NULL); 133 extern struct cfdriver fss_cd; 134 135 void 136 fssattach(int num) 137 { 138 139 mutex_init(&fss_device_lock, MUTEX_DEFAULT, IPL_NONE); 140 if (config_cfattach_attach(fss_cd.cd_name, &fss_ca)) 141 aprint_error("%s: unable to register\n", fss_cd.cd_name); 142 } 143 144 static int 145 fss_match(device_t self, cfdata_t cfdata, void *aux) 146 { 147 return 1; 148 } 149 150 static void 151 fss_attach(device_t parent, device_t self, void *aux) 152 { 153 struct fss_softc *sc = device_private(self); 154 155 sc->sc_dev = self; 156 sc->sc_bdev = NODEV; 157 mutex_init(&sc->sc_slock, MUTEX_DEFAULT, IPL_NONE); 158 mutex_init(&sc->sc_lock, MUTEX_DEFAULT, IPL_NONE); 159 cv_init(&sc->sc_work_cv, "fssbs"); 160 cv_init(&sc->sc_cache_cv, "cowwait"); 161 bufq_alloc(&sc->sc_bufq, "fcfs", 0); 162 sc->sc_dkdev = malloc(sizeof(*sc->sc_dkdev), M_DEVBUF, M_WAITOK); 163 sc->sc_dkdev->dk_info = NULL; 164 disk_init(sc->sc_dkdev, device_xname(self), NULL); 165 if (!pmf_device_register(self, NULL, NULL)) 166 aprint_error_dev(self, "couldn't establish power handler\n"); 167 168 if (fss_num_attached++ == 0) 169 vfs_hooks_attach(&fss_vfs_hooks); 170 } 171 172 static int 173 fss_detach(device_t self, int flags) 174 { 175 struct fss_softc *sc = device_private(self); 176 177 if (sc->sc_flags & FSS_ACTIVE) 178 return EBUSY; 179 180 if (--fss_num_attached == 0) 181 vfs_hooks_detach(&fss_vfs_hooks); 182 183 pmf_device_deregister(self); 184 mutex_destroy(&sc->sc_slock); 185 mutex_destroy(&sc->sc_lock); 186 cv_destroy(&sc->sc_work_cv); 187 cv_destroy(&sc->sc_cache_cv); 188 bufq_drain(sc->sc_bufq); 189 bufq_free(sc->sc_bufq); 190 disk_destroy(sc->sc_dkdev); 191 free(sc->sc_dkdev, M_DEVBUF); 192 193 return 0; 194 } 195 196 int 197 fss_open(dev_t dev, int flags, int mode, struct lwp *l) 198 { 199 int mflag; 200 cfdata_t cf; 201 struct fss_softc *sc; 202 203 mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN); 204 205 mutex_enter(&fss_device_lock); 206 207 sc = device_lookup_private(&fss_cd, minor(dev)); 208 if (sc == NULL) { 209 cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK); 210 cf->cf_name = fss_cd.cd_name; 211 cf->cf_atname = fss_cd.cd_name; 212 cf->cf_unit = minor(dev); 213 cf->cf_fstate = FSTATE_STAR; 214 sc = device_private(config_attach_pseudo(cf)); 215 if (sc == NULL) { 216 mutex_exit(&fss_device_lock); 217 return ENOMEM; 218 } 219 } 220 221 mutex_enter(&sc->sc_slock); 222 223 sc->sc_flags |= mflag; 224 225 mutex_exit(&sc->sc_slock); 226 mutex_exit(&fss_device_lock); 227 228 return 0; 229 } 230 231 int 232 fss_close(dev_t dev, int flags, int mode, struct lwp *l) 233 { 234 int mflag, error; 235 cfdata_t cf; 236 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(dev)); 237 238 mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN); 239 error = 0; 240 241 mutex_enter(&fss_device_lock); 242 restart: 243 mutex_enter(&sc->sc_slock); 244 if ((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) != mflag) { 245 sc->sc_flags &= ~mflag; 246 mutex_exit(&sc->sc_slock); 247 mutex_exit(&fss_device_lock); 248 return 0; 249 } 250 if ((sc->sc_flags & FSS_ACTIVE) != 0 && 251 (sc->sc_uflags & FSS_UNCONFIG_ON_CLOSE) != 0) { 252 sc->sc_uflags &= ~FSS_UNCONFIG_ON_CLOSE; 253 mutex_exit(&sc->sc_slock); 254 error = fss_ioctl(dev, FSSIOCCLR, NULL, FWRITE, l); 255 goto restart; 256 } 257 if ((sc->sc_flags & FSS_ACTIVE) != 0) { 258 mutex_exit(&sc->sc_slock); 259 mutex_exit(&fss_device_lock); 260 return error; 261 } 262 263 KASSERT((sc->sc_flags & FSS_ACTIVE) == 0); 264 KASSERT((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) == mflag); 265 mutex_exit(&sc->sc_slock); 266 cf = device_cfdata(sc->sc_dev); 267 error = config_detach(sc->sc_dev, DETACH_QUIET); 268 if (! error) 269 free(cf, M_DEVBUF); 270 mutex_exit(&fss_device_lock); 271 272 return error; 273 } 274 275 void 276 fss_strategy(struct buf *bp) 277 { 278 const bool write = ((bp->b_flags & B_READ) != B_READ); 279 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(bp->b_dev)); 280 281 mutex_enter(&sc->sc_slock); 282 283 if (write || !FSS_ISVALID(sc)) { 284 285 mutex_exit(&sc->sc_slock); 286 287 bp->b_error = (write ? EROFS : ENXIO); 288 bp->b_resid = bp->b_bcount; 289 biodone(bp); 290 return; 291 } 292 293 bp->b_rawblkno = bp->b_blkno; 294 bufq_put(sc->sc_bufq, bp); 295 cv_signal(&sc->sc_work_cv); 296 297 mutex_exit(&sc->sc_slock); 298 } 299 300 int 301 fss_read(dev_t dev, struct uio *uio, int flags) 302 { 303 return physio(fss_strategy, NULL, dev, B_READ, minphys, uio); 304 } 305 306 int 307 fss_write(dev_t dev, struct uio *uio, int flags) 308 { 309 return physio(fss_strategy, NULL, dev, B_WRITE, minphys, uio); 310 } 311 312 int 313 fss_ioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 314 { 315 int error; 316 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(dev)); 317 struct fss_set _fss; 318 struct fss_set *fss = (struct fss_set *)data; 319 struct fss_set50 *fss50 = (struct fss_set50 *)data; 320 struct fss_get *fsg = (struct fss_get *)data; 321 #ifndef _LP64 322 struct fss_get50 *fsg50 = (struct fss_get50 *)data; 323 #endif 324 325 switch (cmd) { 326 case FSSIOCSET50: 327 fss = &_fss; 328 fss->fss_mount = fss50->fss_mount; 329 fss->fss_bstore = fss50->fss_bstore; 330 fss->fss_csize = fss50->fss_csize; 331 fss->fss_flags = 0; 332 /* Fall through */ 333 case FSSIOCSET: 334 mutex_enter(&sc->sc_lock); 335 if ((flag & FWRITE) == 0) 336 error = EPERM; 337 else if ((sc->sc_flags & FSS_ACTIVE) != 0) 338 error = EBUSY; 339 else 340 error = fss_create_snapshot(sc, fss, l); 341 if (error == 0) 342 sc->sc_uflags = fss->fss_flags; 343 mutex_exit(&sc->sc_lock); 344 break; 345 346 case FSSIOCCLR: 347 mutex_enter(&sc->sc_lock); 348 if ((flag & FWRITE) == 0) 349 error = EPERM; 350 else if ((sc->sc_flags & FSS_ACTIVE) == 0) 351 error = ENXIO; 352 else 353 error = fss_delete_snapshot(sc, l); 354 mutex_exit(&sc->sc_lock); 355 break; 356 357 #ifndef _LP64 358 case FSSIOCGET50: 359 mutex_enter(&sc->sc_lock); 360 switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) { 361 case FSS_ACTIVE: 362 memcpy(fsg50->fsg_mount, sc->sc_mntname, MNAMELEN); 363 fsg50->fsg_csize = FSS_CLSIZE(sc); 364 timeval_to_timeval50(&sc->sc_time, &fsg50->fsg_time); 365 fsg50->fsg_mount_size = sc->sc_clcount; 366 fsg50->fsg_bs_size = sc->sc_clnext; 367 error = 0; 368 break; 369 case FSS_PERSISTENT | FSS_ACTIVE: 370 memcpy(fsg50->fsg_mount, sc->sc_mntname, MNAMELEN); 371 fsg50->fsg_csize = 0; 372 timeval_to_timeval50(&sc->sc_time, &fsg50->fsg_time); 373 fsg50->fsg_mount_size = 0; 374 fsg50->fsg_bs_size = 0; 375 error = 0; 376 break; 377 default: 378 error = ENXIO; 379 break; 380 } 381 mutex_exit(&sc->sc_lock); 382 break; 383 #endif /* _LP64 */ 384 385 case FSSIOCGET: 386 mutex_enter(&sc->sc_lock); 387 switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) { 388 case FSS_ACTIVE: 389 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 390 fsg->fsg_csize = FSS_CLSIZE(sc); 391 fsg->fsg_time = sc->sc_time; 392 fsg->fsg_mount_size = sc->sc_clcount; 393 fsg->fsg_bs_size = sc->sc_clnext; 394 error = 0; 395 break; 396 case FSS_PERSISTENT | FSS_ACTIVE: 397 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 398 fsg->fsg_csize = 0; 399 fsg->fsg_time = sc->sc_time; 400 fsg->fsg_mount_size = 0; 401 fsg->fsg_bs_size = 0; 402 error = 0; 403 break; 404 default: 405 error = ENXIO; 406 break; 407 } 408 mutex_exit(&sc->sc_lock); 409 break; 410 411 case FSSIOFSET: 412 mutex_enter(&sc->sc_slock); 413 sc->sc_uflags = *(int *)data; 414 mutex_exit(&sc->sc_slock); 415 error = 0; 416 break; 417 418 case FSSIOFGET: 419 mutex_enter(&sc->sc_slock); 420 *(int *)data = sc->sc_uflags; 421 mutex_exit(&sc->sc_slock); 422 error = 0; 423 break; 424 425 default: 426 error = EINVAL; 427 break; 428 } 429 430 return error; 431 } 432 433 int 434 fss_size(dev_t dev) 435 { 436 return -1; 437 } 438 439 int 440 fss_dump(dev_t dev, daddr_t blkno, void *va, 441 size_t size) 442 { 443 return EROFS; 444 } 445 446 /* 447 * An error occurred reading or writing the snapshot or backing store. 448 * If it is the first error log to console. 449 * The caller holds the mutex. 450 */ 451 static inline void 452 fss_error(struct fss_softc *sc, const char *msg) 453 { 454 455 if ((sc->sc_flags & (FSS_ACTIVE|FSS_ERROR)) == FSS_ACTIVE) 456 aprint_error_dev(sc->sc_dev, "snapshot invalid: %s\n", msg); 457 if ((sc->sc_flags & FSS_ACTIVE) == FSS_ACTIVE) 458 sc->sc_flags |= FSS_ERROR; 459 } 460 461 /* 462 * Allocate the variable sized parts of the softc and 463 * fork the kernel thread. 464 * 465 * The fields sc_clcount, sc_clshift, sc_cache_size and sc_indir_size 466 * must be initialized. 467 */ 468 static int 469 fss_softc_alloc(struct fss_softc *sc) 470 { 471 int i, error; 472 473 if ((sc->sc_flags & FSS_PERSISTENT) == 0) { 474 sc->sc_copied = 475 kmem_zalloc(howmany(sc->sc_clcount, NBBY), KM_SLEEP); 476 if (sc->sc_copied == NULL) 477 return(ENOMEM); 478 479 sc->sc_cache = kmem_alloc(sc->sc_cache_size * 480 sizeof(struct fss_cache), KM_SLEEP); 481 if (sc->sc_cache == NULL) 482 return(ENOMEM); 483 484 for (i = 0; i < sc->sc_cache_size; i++) { 485 sc->sc_cache[i].fc_type = FSS_CACHE_FREE; 486 sc->sc_cache[i].fc_data = 487 kmem_alloc(FSS_CLSIZE(sc), KM_SLEEP); 488 if (sc->sc_cache[i].fc_data == NULL) 489 return(ENOMEM); 490 cv_init(&sc->sc_cache[i].fc_state_cv, "cowwait1"); 491 } 492 493 sc->sc_indir_valid = 494 kmem_zalloc(howmany(sc->sc_indir_size, NBBY), KM_SLEEP); 495 if (sc->sc_indir_valid == NULL) 496 return(ENOMEM); 497 498 sc->sc_indir_data = kmem_zalloc(FSS_CLSIZE(sc), KM_SLEEP); 499 if (sc->sc_indir_data == NULL) 500 return(ENOMEM); 501 } else { 502 sc->sc_copied = NULL; 503 sc->sc_cache = NULL; 504 sc->sc_indir_valid = NULL; 505 sc->sc_indir_data = NULL; 506 } 507 508 sc->sc_flags |= FSS_BS_THREAD; 509 if ((error = kthread_create(PRI_BIO, KTHREAD_MUSTJOIN, NULL, 510 fss_bs_thread, sc, &sc->sc_bs_lwp, 511 "%s", device_xname(sc->sc_dev))) != 0) { 512 sc->sc_flags &= ~FSS_BS_THREAD; 513 return error; 514 } 515 516 disk_attach(sc->sc_dkdev); 517 518 return 0; 519 } 520 521 /* 522 * Free the variable sized parts of the softc. 523 */ 524 static void 525 fss_softc_free(struct fss_softc *sc) 526 { 527 int i; 528 529 if ((sc->sc_flags & FSS_BS_THREAD) != 0) { 530 mutex_enter(&sc->sc_slock); 531 sc->sc_flags &= ~FSS_BS_THREAD; 532 cv_signal(&sc->sc_work_cv); 533 mutex_exit(&sc->sc_slock); 534 kthread_join(sc->sc_bs_lwp); 535 536 disk_detach(sc->sc_dkdev); 537 } 538 539 if (sc->sc_copied != NULL) 540 kmem_free(sc->sc_copied, howmany(sc->sc_clcount, NBBY)); 541 sc->sc_copied = NULL; 542 543 if (sc->sc_cache != NULL) { 544 for (i = 0; i < sc->sc_cache_size; i++) 545 if (sc->sc_cache[i].fc_data != NULL) { 546 cv_destroy(&sc->sc_cache[i].fc_state_cv); 547 kmem_free(sc->sc_cache[i].fc_data, 548 FSS_CLSIZE(sc)); 549 } 550 kmem_free(sc->sc_cache, 551 sc->sc_cache_size*sizeof(struct fss_cache)); 552 } 553 sc->sc_cache = NULL; 554 555 if (sc->sc_indir_valid != NULL) 556 kmem_free(sc->sc_indir_valid, howmany(sc->sc_indir_size, NBBY)); 557 sc->sc_indir_valid = NULL; 558 559 if (sc->sc_indir_data != NULL) 560 kmem_free(sc->sc_indir_data, FSS_CLSIZE(sc)); 561 sc->sc_indir_data = NULL; 562 } 563 564 /* 565 * Set all active snapshots on this file system into ERROR state. 566 */ 567 static void 568 fss_unmount_hook(struct mount *mp) 569 { 570 int i; 571 struct fss_softc *sc; 572 573 mutex_enter(&fss_device_lock); 574 for (i = 0; i < fss_cd.cd_ndevs; i++) { 575 if ((sc = device_lookup_private(&fss_cd, i)) == NULL) 576 continue; 577 mutex_enter(&sc->sc_slock); 578 if ((sc->sc_flags & FSS_ACTIVE) != 0 && 579 sc->sc_mount == mp) 580 fss_error(sc, "forced unmount"); 581 mutex_exit(&sc->sc_slock); 582 } 583 mutex_exit(&fss_device_lock); 584 } 585 586 /* 587 * A buffer is written to the snapshotted block device. Copy to 588 * backing store if needed. 589 */ 590 static int 591 fss_copy_on_write(void *v, struct buf *bp, bool data_valid) 592 { 593 int error; 594 u_int32_t cl, ch, c; 595 struct fss_softc *sc = v; 596 597 mutex_enter(&sc->sc_slock); 598 if (!FSS_ISVALID(sc)) { 599 mutex_exit(&sc->sc_slock); 600 return 0; 601 } 602 603 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 604 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 605 error = 0; 606 if (curlwp == uvm.pagedaemon_lwp) { 607 for (c = cl; c <= ch; c++) 608 if (isclr(sc->sc_copied, c)) { 609 error = ENOMEM; 610 break; 611 } 612 } 613 mutex_exit(&sc->sc_slock); 614 615 if (error == 0) 616 for (c = cl; c <= ch; c++) { 617 error = fss_read_cluster(sc, c); 618 if (error) 619 break; 620 } 621 622 return error; 623 } 624 625 /* 626 * Lookup and open needed files. 627 * 628 * For file system internal snapshot initializes sc_mntname, sc_mount, 629 * sc_bs_vp and sc_time. 630 * 631 * Otherwise returns dev and size of the underlying block device. 632 * Initializes sc_mntname, sc_mount, sc_bdev, sc_bs_vp and sc_mount 633 */ 634 static int 635 fss_create_files(struct fss_softc *sc, struct fss_set *fss, 636 off_t *bsize, struct lwp *l) 637 { 638 int error, bits, fsbsize; 639 uint64_t numsec; 640 unsigned int secsize; 641 struct timespec ts; 642 /* nd -> nd2 to reduce mistakes while updating only some namei calls */ 643 struct pathbuf *pb2; 644 struct nameidata nd2; 645 struct vnode *vp; 646 647 /* 648 * Get the mounted file system. 649 */ 650 651 error = namei_simple_user(fss->fss_mount, 652 NSM_FOLLOW_NOEMULROOT, &vp); 653 if (error != 0) 654 return error; 655 656 if ((vp->v_vflag & VV_ROOT) != VV_ROOT) { 657 vrele(vp); 658 return EINVAL; 659 } 660 661 sc->sc_mount = vp->v_mount; 662 memcpy(sc->sc_mntname, sc->sc_mount->mnt_stat.f_mntonname, MNAMELEN); 663 664 vrele(vp); 665 666 /* 667 * Check for file system internal snapshot. 668 */ 669 670 error = namei_simple_user(fss->fss_bstore, 671 NSM_FOLLOW_NOEMULROOT, &vp); 672 if (error != 0) 673 return error; 674 675 if (vp->v_type == VREG && vp->v_mount == sc->sc_mount) { 676 sc->sc_flags |= FSS_PERSISTENT; 677 sc->sc_bs_vp = vp; 678 679 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 680 bits = sizeof(sc->sc_bs_bshift)*NBBY; 681 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < bits; 682 sc->sc_bs_bshift++) 683 if (FSS_FSBSIZE(sc) == fsbsize) 684 break; 685 if (sc->sc_bs_bshift >= bits) 686 return EINVAL; 687 688 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 689 sc->sc_clshift = 0; 690 691 if ((fss->fss_flags & FSS_UNLINK_ON_CREATE) != 0) { 692 error = do_sys_unlink(fss->fss_bstore, UIO_USERSPACE); 693 if (error) 694 return error; 695 } 696 error = vn_lock(vp, LK_EXCLUSIVE); 697 if (error != 0) 698 return error; 699 error = VFS_SNAPSHOT(sc->sc_mount, sc->sc_bs_vp, &ts); 700 TIMESPEC_TO_TIMEVAL(&sc->sc_time, &ts); 701 702 VOP_UNLOCK(sc->sc_bs_vp); 703 704 return error; 705 } 706 vrele(vp); 707 708 /* 709 * Get the block device it is mounted on and its size. 710 */ 711 712 error = spec_node_lookup_by_mount(sc->sc_mount, &vp); 713 if (error) 714 return error; 715 sc->sc_bdev = vp->v_rdev; 716 717 error = getdisksize(vp, &numsec, &secsize); 718 vrele(vp); 719 if (error) 720 return error; 721 722 *bsize = (off_t)numsec*secsize; 723 724 /* 725 * Get the backing store 726 */ 727 728 error = pathbuf_copyin(fss->fss_bstore, &pb2); 729 if (error) { 730 return error; 731 } 732 NDINIT(&nd2, LOOKUP, FOLLOW, pb2); 733 if ((error = vn_open(&nd2, FREAD|FWRITE, 0)) != 0) { 734 pathbuf_destroy(pb2); 735 return error; 736 } 737 VOP_UNLOCK(nd2.ni_vp); 738 739 sc->sc_bs_vp = nd2.ni_vp; 740 741 if (nd2.ni_vp->v_type != VREG && nd2.ni_vp->v_type != VCHR) { 742 pathbuf_destroy(pb2); 743 return EINVAL; 744 } 745 pathbuf_destroy(pb2); 746 747 if ((fss->fss_flags & FSS_UNLINK_ON_CREATE) != 0) { 748 error = do_sys_unlink(fss->fss_bstore, UIO_USERSPACE); 749 if (error) 750 return error; 751 } 752 if (sc->sc_bs_vp->v_type == VREG) { 753 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 754 if (fsbsize & (fsbsize-1)) /* No power of two */ 755 return EINVAL; 756 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < 32; 757 sc->sc_bs_bshift++) 758 if (FSS_FSBSIZE(sc) == fsbsize) 759 break; 760 if (sc->sc_bs_bshift >= 32) 761 return EINVAL; 762 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 763 } else { 764 sc->sc_bs_bshift = DEV_BSHIFT; 765 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 766 } 767 768 return 0; 769 } 770 771 /* 772 * Create a snapshot. 773 */ 774 static int 775 fss_create_snapshot(struct fss_softc *sc, struct fss_set *fss, struct lwp *l) 776 { 777 int len, error; 778 u_int32_t csize; 779 off_t bsize; 780 781 bsize = 0; /* XXX gcc */ 782 783 /* 784 * Open needed files. 785 */ 786 if ((error = fss_create_files(sc, fss, &bsize, l)) != 0) 787 goto bad; 788 789 if (sc->sc_flags & FSS_PERSISTENT) { 790 fss_softc_alloc(sc); 791 sc->sc_flags |= FSS_ACTIVE; 792 return 0; 793 } 794 795 /* 796 * Set cluster size. Must be a power of two and 797 * a multiple of backing store block size. 798 */ 799 if (fss->fss_csize <= 0) 800 csize = MAXPHYS; 801 else 802 csize = fss->fss_csize; 803 if (bsize/csize > FSS_CLUSTER_MAX) 804 csize = bsize/FSS_CLUSTER_MAX+1; 805 806 for (sc->sc_clshift = sc->sc_bs_bshift; sc->sc_clshift < 32; 807 sc->sc_clshift++) 808 if (FSS_CLSIZE(sc) >= csize) 809 break; 810 if (sc->sc_clshift >= 32) { 811 error = EINVAL; 812 goto bad; 813 } 814 sc->sc_clmask = FSS_CLSIZE(sc)-1; 815 816 /* 817 * Set number of cache slots. 818 */ 819 if (FSS_CLSIZE(sc) <= 8192) 820 sc->sc_cache_size = 32; 821 else if (FSS_CLSIZE(sc) <= 65536) 822 sc->sc_cache_size = 8; 823 else 824 sc->sc_cache_size = 4; 825 826 /* 827 * Set number of clusters and size of last cluster. 828 */ 829 sc->sc_clcount = FSS_BTOCL(sc, bsize-1)+1; 830 sc->sc_clresid = FSS_CLOFF(sc, bsize-1)+1; 831 832 /* 833 * Set size of indirect table. 834 */ 835 len = sc->sc_clcount*sizeof(u_int32_t); 836 sc->sc_indir_size = FSS_BTOCL(sc, len)+1; 837 sc->sc_clnext = sc->sc_indir_size; 838 sc->sc_indir_cur = 0; 839 840 if ((error = fss_softc_alloc(sc)) != 0) 841 goto bad; 842 843 /* 844 * Activate the snapshot. 845 */ 846 847 if ((error = vfs_suspend(sc->sc_mount, 0)) != 0) 848 goto bad; 849 850 microtime(&sc->sc_time); 851 852 error = fscow_establish(sc->sc_mount, fss_copy_on_write, sc); 853 if (error == 0) 854 sc->sc_flags |= FSS_ACTIVE; 855 856 vfs_resume(sc->sc_mount); 857 858 if (error != 0) 859 goto bad; 860 861 aprint_debug_dev(sc->sc_dev, "%s snapshot active\n", sc->sc_mntname); 862 aprint_debug_dev(sc->sc_dev, 863 "%u clusters of %u, %u cache slots, %u indir clusters\n", 864 sc->sc_clcount, FSS_CLSIZE(sc), 865 sc->sc_cache_size, sc->sc_indir_size); 866 867 return 0; 868 869 bad: 870 fss_softc_free(sc); 871 if (sc->sc_bs_vp != NULL) { 872 if (sc->sc_flags & FSS_PERSISTENT) 873 vrele(sc->sc_bs_vp); 874 else 875 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 876 } 877 sc->sc_bs_vp = NULL; 878 879 return error; 880 } 881 882 /* 883 * Delete a snapshot. 884 */ 885 static int 886 fss_delete_snapshot(struct fss_softc *sc, struct lwp *l) 887 { 888 889 if ((sc->sc_flags & FSS_PERSISTENT) == 0) 890 fscow_disestablish(sc->sc_mount, fss_copy_on_write, sc); 891 892 mutex_enter(&sc->sc_slock); 893 sc->sc_flags &= ~(FSS_ACTIVE|FSS_ERROR); 894 sc->sc_mount = NULL; 895 sc->sc_bdev = NODEV; 896 mutex_exit(&sc->sc_slock); 897 898 fss_softc_free(sc); 899 if (sc->sc_flags & FSS_PERSISTENT) 900 vrele(sc->sc_bs_vp); 901 else 902 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 903 sc->sc_bs_vp = NULL; 904 sc->sc_flags &= ~FSS_PERSISTENT; 905 906 return 0; 907 } 908 909 /* 910 * Read a cluster from the snapshotted block device to the cache. 911 */ 912 static int 913 fss_read_cluster(struct fss_softc *sc, u_int32_t cl) 914 { 915 int error, todo, offset, len; 916 daddr_t dblk; 917 struct buf *bp, *mbp; 918 struct fss_cache *scp, *scl; 919 920 /* 921 * Get a free cache slot. 922 */ 923 scl = sc->sc_cache+sc->sc_cache_size; 924 925 mutex_enter(&sc->sc_slock); 926 927 restart: 928 if (isset(sc->sc_copied, cl) || !FSS_ISVALID(sc)) { 929 mutex_exit(&sc->sc_slock); 930 return 0; 931 } 932 933 for (scp = sc->sc_cache; scp < scl; scp++) 934 if (scp->fc_cluster == cl) { 935 if (scp->fc_type == FSS_CACHE_VALID) { 936 mutex_exit(&sc->sc_slock); 937 return 0; 938 } else if (scp->fc_type == FSS_CACHE_BUSY) { 939 cv_wait(&scp->fc_state_cv, &sc->sc_slock); 940 goto restart; 941 } 942 } 943 944 for (scp = sc->sc_cache; scp < scl; scp++) 945 if (scp->fc_type == FSS_CACHE_FREE) { 946 scp->fc_type = FSS_CACHE_BUSY; 947 scp->fc_cluster = cl; 948 break; 949 } 950 if (scp >= scl) { 951 cv_wait(&sc->sc_cache_cv, &sc->sc_slock); 952 goto restart; 953 } 954 955 mutex_exit(&sc->sc_slock); 956 957 /* 958 * Start the read. 959 */ 960 dblk = btodb(FSS_CLTOB(sc, cl)); 961 if (cl == sc->sc_clcount-1) { 962 todo = sc->sc_clresid; 963 memset((char *)scp->fc_data + todo, 0, FSS_CLSIZE(sc) - todo); 964 } else 965 todo = FSS_CLSIZE(sc); 966 offset = 0; 967 mbp = getiobuf(NULL, true); 968 mbp->b_bufsize = todo; 969 mbp->b_data = scp->fc_data; 970 mbp->b_resid = mbp->b_bcount = todo; 971 mbp->b_flags = B_READ; 972 mbp->b_cflags = BC_BUSY; 973 mbp->b_dev = sc->sc_bdev; 974 while (todo > 0) { 975 len = todo; 976 if (len > MAXPHYS) 977 len = MAXPHYS; 978 if (btodb(FSS_CLTOB(sc, cl)) == dblk && len == todo) 979 bp = mbp; 980 else { 981 bp = getiobuf(NULL, true); 982 nestiobuf_setup(mbp, bp, offset, len); 983 } 984 bp->b_lblkno = 0; 985 bp->b_blkno = dblk; 986 bdev_strategy(bp); 987 dblk += btodb(len); 988 offset += len; 989 todo -= len; 990 } 991 error = biowait(mbp); 992 putiobuf(mbp); 993 994 mutex_enter(&sc->sc_slock); 995 scp->fc_type = (error ? FSS_CACHE_FREE : FSS_CACHE_VALID); 996 cv_broadcast(&scp->fc_state_cv); 997 if (error == 0) { 998 setbit(sc->sc_copied, scp->fc_cluster); 999 cv_signal(&sc->sc_work_cv); 1000 } 1001 mutex_exit(&sc->sc_slock); 1002 1003 return error; 1004 } 1005 1006 /* 1007 * Read/write clusters from/to backing store. 1008 * For persistent snapshots must be called with cl == 0. off is the 1009 * offset into the snapshot. 1010 */ 1011 static int 1012 fss_bs_io(struct fss_softc *sc, fss_io_type rw, 1013 u_int32_t cl, off_t off, int len, void *data) 1014 { 1015 int error; 1016 1017 off += FSS_CLTOB(sc, cl); 1018 1019 vn_lock(sc->sc_bs_vp, LK_EXCLUSIVE|LK_RETRY); 1020 1021 error = vn_rdwr((rw == FSS_READ ? UIO_READ : UIO_WRITE), sc->sc_bs_vp, 1022 data, len, off, UIO_SYSSPACE, 1023 IO_ADV_ENCODE(POSIX_FADV_NOREUSE) | IO_NODELOCKED, 1024 sc->sc_bs_lwp->l_cred, NULL, NULL); 1025 if (error == 0) { 1026 mutex_enter(sc->sc_bs_vp->v_interlock); 1027 error = VOP_PUTPAGES(sc->sc_bs_vp, trunc_page(off), 1028 round_page(off+len), PGO_CLEANIT | PGO_FREE | PGO_SYNCIO); 1029 } 1030 1031 VOP_UNLOCK(sc->sc_bs_vp); 1032 1033 return error; 1034 } 1035 1036 /* 1037 * Get a pointer to the indirect slot for this cluster. 1038 */ 1039 static u_int32_t * 1040 fss_bs_indir(struct fss_softc *sc, u_int32_t cl) 1041 { 1042 u_int32_t icl; 1043 int ioff; 1044 1045 icl = cl/(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 1046 ioff = cl%(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 1047 1048 if (sc->sc_indir_cur == icl) 1049 return &sc->sc_indir_data[ioff]; 1050 1051 if (sc->sc_indir_dirty) { 1052 if (fss_bs_io(sc, FSS_WRITE, sc->sc_indir_cur, 0, 1053 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 1054 return NULL; 1055 setbit(sc->sc_indir_valid, sc->sc_indir_cur); 1056 } 1057 1058 sc->sc_indir_dirty = 0; 1059 sc->sc_indir_cur = icl; 1060 1061 if (isset(sc->sc_indir_valid, sc->sc_indir_cur)) { 1062 if (fss_bs_io(sc, FSS_READ, sc->sc_indir_cur, 0, 1063 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 1064 return NULL; 1065 } else 1066 memset(sc->sc_indir_data, 0, FSS_CLSIZE(sc)); 1067 1068 return &sc->sc_indir_data[ioff]; 1069 } 1070 1071 /* 1072 * The kernel thread (one for every active snapshot). 1073 * 1074 * After wakeup it cleans the cache and runs the I/O requests. 1075 */ 1076 static void 1077 fss_bs_thread(void *arg) 1078 { 1079 bool thread_idle, is_valid; 1080 int error, i, todo, len, crotor, is_read; 1081 long off; 1082 char *addr; 1083 u_int32_t c, cl, ch, *indirp; 1084 struct buf *bp, *nbp; 1085 struct fss_softc *sc; 1086 struct fss_cache *scp, *scl; 1087 1088 sc = arg; 1089 scl = sc->sc_cache+sc->sc_cache_size; 1090 crotor = 0; 1091 thread_idle = false; 1092 1093 mutex_enter(&sc->sc_slock); 1094 1095 for (;;) { 1096 if (thread_idle) 1097 cv_wait(&sc->sc_work_cv, &sc->sc_slock); 1098 thread_idle = true; 1099 if ((sc->sc_flags & FSS_BS_THREAD) == 0) { 1100 mutex_exit(&sc->sc_slock); 1101 kthread_exit(0); 1102 } 1103 1104 /* 1105 * Process I/O requests (persistent) 1106 */ 1107 1108 if (sc->sc_flags & FSS_PERSISTENT) { 1109 if ((bp = bufq_get(sc->sc_bufq)) == NULL) 1110 continue; 1111 is_valid = FSS_ISVALID(sc); 1112 is_read = (bp->b_flags & B_READ); 1113 thread_idle = false; 1114 mutex_exit(&sc->sc_slock); 1115 1116 if (is_valid) { 1117 disk_busy(sc->sc_dkdev); 1118 error = fss_bs_io(sc, FSS_READ, 0, 1119 dbtob(bp->b_blkno), bp->b_bcount, 1120 bp->b_data); 1121 disk_unbusy(sc->sc_dkdev, 1122 (error ? 0 : bp->b_bcount), is_read); 1123 } else 1124 error = ENXIO; 1125 1126 bp->b_error = error; 1127 bp->b_resid = (error ? bp->b_bcount : 0); 1128 biodone(bp); 1129 1130 mutex_enter(&sc->sc_slock); 1131 continue; 1132 } 1133 1134 /* 1135 * Clean the cache 1136 */ 1137 for (i = 0; i < sc->sc_cache_size; i++) { 1138 crotor = (crotor + 1) % sc->sc_cache_size; 1139 scp = sc->sc_cache + crotor; 1140 if (scp->fc_type != FSS_CACHE_VALID) 1141 continue; 1142 mutex_exit(&sc->sc_slock); 1143 1144 thread_idle = false; 1145 indirp = fss_bs_indir(sc, scp->fc_cluster); 1146 if (indirp != NULL) { 1147 error = fss_bs_io(sc, FSS_WRITE, sc->sc_clnext, 1148 0, FSS_CLSIZE(sc), scp->fc_data); 1149 } else 1150 error = EIO; 1151 1152 mutex_enter(&sc->sc_slock); 1153 if (error == 0) { 1154 *indirp = sc->sc_clnext++; 1155 sc->sc_indir_dirty = 1; 1156 } else 1157 fss_error(sc, "write error on backing store"); 1158 1159 scp->fc_type = FSS_CACHE_FREE; 1160 cv_broadcast(&sc->sc_cache_cv); 1161 break; 1162 } 1163 1164 /* 1165 * Process I/O requests 1166 */ 1167 if ((bp = bufq_get(sc->sc_bufq)) == NULL) 1168 continue; 1169 is_valid = FSS_ISVALID(sc); 1170 is_read = (bp->b_flags & B_READ); 1171 thread_idle = false; 1172 1173 if (!is_valid) { 1174 mutex_exit(&sc->sc_slock); 1175 1176 bp->b_error = ENXIO; 1177 bp->b_resid = bp->b_bcount; 1178 biodone(bp); 1179 1180 mutex_enter(&sc->sc_slock); 1181 continue; 1182 } 1183 1184 disk_busy(sc->sc_dkdev); 1185 1186 /* 1187 * First read from the snapshotted block device unless 1188 * this request is completely covered by backing store. 1189 */ 1190 1191 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 1192 off = FSS_CLOFF(sc, dbtob(bp->b_blkno)); 1193 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 1194 error = 0; 1195 bp->b_resid = 0; 1196 bp->b_error = 0; 1197 for (c = cl; c <= ch; c++) { 1198 if (isset(sc->sc_copied, c)) 1199 continue; 1200 mutex_exit(&sc->sc_slock); 1201 1202 /* Not on backing store, read from device. */ 1203 nbp = getiobuf(NULL, true); 1204 nbp->b_flags = B_READ; 1205 nbp->b_resid = nbp->b_bcount = bp->b_bcount; 1206 nbp->b_bufsize = bp->b_bcount; 1207 nbp->b_data = bp->b_data; 1208 nbp->b_blkno = bp->b_blkno; 1209 nbp->b_lblkno = 0; 1210 nbp->b_dev = sc->sc_bdev; 1211 SET(nbp->b_cflags, BC_BUSY); /* mark buffer busy */ 1212 1213 bdev_strategy(nbp); 1214 1215 error = biowait(nbp); 1216 if (error != 0) { 1217 bp->b_resid = bp->b_bcount; 1218 bp->b_error = nbp->b_error; 1219 disk_unbusy(sc->sc_dkdev, 0, is_read); 1220 biodone(bp); 1221 } 1222 putiobuf(nbp); 1223 1224 mutex_enter(&sc->sc_slock); 1225 break; 1226 } 1227 if (error) 1228 continue; 1229 1230 /* 1231 * Replace those parts that have been saved to backing store. 1232 */ 1233 1234 addr = bp->b_data; 1235 todo = bp->b_bcount; 1236 for (c = cl; c <= ch; c++, off = 0, todo -= len, addr += len) { 1237 len = FSS_CLSIZE(sc)-off; 1238 if (len > todo) 1239 len = todo; 1240 if (isclr(sc->sc_copied, c)) 1241 continue; 1242 mutex_exit(&sc->sc_slock); 1243 1244 indirp = fss_bs_indir(sc, c); 1245 if (indirp == NULL || *indirp == 0) { 1246 /* 1247 * Not on backing store. Either in cache 1248 * or hole in the snapshotted block device. 1249 */ 1250 1251 mutex_enter(&sc->sc_slock); 1252 for (scp = sc->sc_cache; scp < scl; scp++) 1253 if (scp->fc_type == FSS_CACHE_VALID && 1254 scp->fc_cluster == c) 1255 break; 1256 if (scp < scl) 1257 memcpy(addr, (char *)scp->fc_data+off, 1258 len); 1259 else 1260 memset(addr, 0, len); 1261 continue; 1262 } 1263 1264 /* 1265 * Read from backing store. 1266 */ 1267 error = 1268 fss_bs_io(sc, FSS_READ, *indirp, off, len, addr); 1269 1270 mutex_enter(&sc->sc_slock); 1271 if (error) { 1272 bp->b_resid = bp->b_bcount; 1273 bp->b_error = error; 1274 break; 1275 } 1276 } 1277 mutex_exit(&sc->sc_slock); 1278 1279 disk_unbusy(sc->sc_dkdev, (error ? 0 : bp->b_bcount), is_read); 1280 biodone(bp); 1281 1282 mutex_enter(&sc->sc_slock); 1283 } 1284 } 1285 1286 #ifdef _MODULE 1287 1288 #include <sys/module.h> 1289 1290 MODULE(MODULE_CLASS_DRIVER, fss, NULL); 1291 CFDRIVER_DECL(fss, DV_DISK, NULL); 1292 1293 static int 1294 fss_modcmd(modcmd_t cmd, void *arg) 1295 { 1296 int bmajor = -1, cmajor = -1, error = 0; 1297 1298 switch (cmd) { 1299 case MODULE_CMD_INIT: 1300 mutex_init(&fss_device_lock, MUTEX_DEFAULT, IPL_NONE); 1301 error = config_cfdriver_attach(&fss_cd); 1302 if (error) { 1303 mutex_destroy(&fss_device_lock); 1304 break; 1305 } 1306 error = config_cfattach_attach(fss_cd.cd_name, &fss_ca); 1307 if (error) { 1308 config_cfdriver_detach(&fss_cd); 1309 mutex_destroy(&fss_device_lock); 1310 break; 1311 } 1312 error = devsw_attach(fss_cd.cd_name, 1313 &fss_bdevsw, &bmajor, &fss_cdevsw, &cmajor); 1314 if (error == EEXIST) 1315 error = 0; 1316 if (error) { 1317 config_cfattach_detach(fss_cd.cd_name, &fss_ca); 1318 config_cfdriver_detach(&fss_cd); 1319 mutex_destroy(&fss_device_lock); 1320 break; 1321 } 1322 break; 1323 1324 case MODULE_CMD_FINI: 1325 error = config_cfattach_detach(fss_cd.cd_name, &fss_ca); 1326 if (error) 1327 break; 1328 config_cfdriver_detach(&fss_cd); 1329 devsw_detach(&fss_bdevsw, &fss_cdevsw); 1330 mutex_destroy(&fss_device_lock); 1331 break; 1332 1333 default: 1334 error = ENOTTY; 1335 break; 1336 } 1337 1338 return error; 1339 } 1340 1341 #endif /* _MODULE */ 1342