1 /* $NetBSD: fss.c,v 1.99 2017/10/28 03:47:24 riastradh 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 1.99 2017/10/28 03:47:24 riastradh 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 #include "ioconf.h" 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); 133 134 void 135 fssattach(int num) 136 { 137 138 mutex_init(&fss_device_lock, MUTEX_DEFAULT, IPL_NONE); 139 if (config_cfattach_attach(fss_cd.cd_name, &fss_ca)) 140 aprint_error("%s: unable to register\n", fss_cd.cd_name); 141 } 142 143 static int 144 fss_match(device_t self, cfdata_t cfdata, void *aux) 145 { 146 return 1; 147 } 148 149 static void 150 fss_attach(device_t parent, device_t self, void *aux) 151 { 152 struct fss_softc *sc = device_private(self); 153 154 sc->sc_dev = self; 155 sc->sc_bdev = NODEV; 156 mutex_init(&sc->sc_slock, MUTEX_DEFAULT, IPL_NONE); 157 mutex_init(&sc->sc_lock, MUTEX_DEFAULT, IPL_NONE); 158 cv_init(&sc->sc_work_cv, "fssbs"); 159 cv_init(&sc->sc_cache_cv, "cowwait"); 160 bufq_alloc(&sc->sc_bufq, "fcfs", 0); 161 sc->sc_dkdev = malloc(sizeof(*sc->sc_dkdev), M_DEVBUF, M_WAITOK); 162 sc->sc_dkdev->dk_info = NULL; 163 disk_init(sc->sc_dkdev, device_xname(self), NULL); 164 if (!pmf_device_register(self, NULL, NULL)) 165 aprint_error_dev(self, "couldn't establish power handler\n"); 166 167 if (fss_num_attached++ == 0) 168 vfs_hooks_attach(&fss_vfs_hooks); 169 } 170 171 static int 172 fss_detach(device_t self, int flags) 173 { 174 struct fss_softc *sc = device_private(self); 175 176 if (sc->sc_flags & FSS_ACTIVE) 177 return EBUSY; 178 179 if (--fss_num_attached == 0) 180 vfs_hooks_detach(&fss_vfs_hooks); 181 182 pmf_device_deregister(self); 183 mutex_destroy(&sc->sc_slock); 184 mutex_destroy(&sc->sc_lock); 185 cv_destroy(&sc->sc_work_cv); 186 cv_destroy(&sc->sc_cache_cv); 187 bufq_drain(sc->sc_bufq); 188 bufq_free(sc->sc_bufq); 189 disk_destroy(sc->sc_dkdev); 190 free(sc->sc_dkdev, M_DEVBUF); 191 192 return 0; 193 } 194 195 int 196 fss_open(dev_t dev, int flags, int mode, struct lwp *l) 197 { 198 int mflag; 199 cfdata_t cf; 200 struct fss_softc *sc; 201 202 mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN); 203 204 mutex_enter(&fss_device_lock); 205 206 sc = device_lookup_private(&fss_cd, minor(dev)); 207 if (sc == NULL) { 208 cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK); 209 cf->cf_name = fss_cd.cd_name; 210 cf->cf_atname = fss_cd.cd_name; 211 cf->cf_unit = minor(dev); 212 cf->cf_fstate = FSTATE_STAR; 213 sc = device_private(config_attach_pseudo(cf)); 214 if (sc == NULL) { 215 mutex_exit(&fss_device_lock); 216 return ENOMEM; 217 } 218 } 219 220 mutex_enter(&sc->sc_slock); 221 222 sc->sc_flags |= mflag; 223 224 mutex_exit(&sc->sc_slock); 225 mutex_exit(&fss_device_lock); 226 227 return 0; 228 } 229 230 int 231 fss_close(dev_t dev, int flags, int mode, struct lwp *l) 232 { 233 int mflag, error; 234 cfdata_t cf; 235 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(dev)); 236 237 mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN); 238 error = 0; 239 240 mutex_enter(&fss_device_lock); 241 restart: 242 mutex_enter(&sc->sc_slock); 243 if ((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) != mflag) { 244 sc->sc_flags &= ~mflag; 245 mutex_exit(&sc->sc_slock); 246 mutex_exit(&fss_device_lock); 247 return 0; 248 } 249 if ((sc->sc_flags & FSS_ACTIVE) != 0 && 250 (sc->sc_uflags & FSS_UNCONFIG_ON_CLOSE) != 0) { 251 sc->sc_uflags &= ~FSS_UNCONFIG_ON_CLOSE; 252 mutex_exit(&sc->sc_slock); 253 error = fss_ioctl(dev, FSSIOCCLR, NULL, FWRITE, l); 254 goto restart; 255 } 256 if ((sc->sc_flags & FSS_ACTIVE) != 0) { 257 mutex_exit(&sc->sc_slock); 258 mutex_exit(&fss_device_lock); 259 return error; 260 } 261 262 KASSERT((sc->sc_flags & FSS_ACTIVE) == 0); 263 KASSERT((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) == mflag); 264 mutex_exit(&sc->sc_slock); 265 cf = device_cfdata(sc->sc_dev); 266 error = config_detach(sc->sc_dev, DETACH_QUIET); 267 if (! error) 268 free(cf, M_DEVBUF); 269 mutex_exit(&fss_device_lock); 270 271 return error; 272 } 273 274 void 275 fss_strategy(struct buf *bp) 276 { 277 const bool write = ((bp->b_flags & B_READ) != B_READ); 278 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(bp->b_dev)); 279 280 mutex_enter(&sc->sc_slock); 281 282 if (write || !FSS_ISVALID(sc)) { 283 284 mutex_exit(&sc->sc_slock); 285 286 bp->b_error = (write ? EROFS : ENXIO); 287 bp->b_resid = bp->b_bcount; 288 biodone(bp); 289 return; 290 } 291 292 bp->b_rawblkno = bp->b_blkno; 293 bufq_put(sc->sc_bufq, bp); 294 cv_signal(&sc->sc_work_cv); 295 296 mutex_exit(&sc->sc_slock); 297 } 298 299 int 300 fss_read(dev_t dev, struct uio *uio, int flags) 301 { 302 return physio(fss_strategy, NULL, dev, B_READ, minphys, uio); 303 } 304 305 int 306 fss_write(dev_t dev, struct uio *uio, int flags) 307 { 308 return physio(fss_strategy, NULL, dev, B_WRITE, minphys, uio); 309 } 310 311 int 312 fss_ioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 313 { 314 int error; 315 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(dev)); 316 struct fss_set _fss; 317 struct fss_set *fss = (struct fss_set *)data; 318 struct fss_set50 *fss50 = (struct fss_set50 *)data; 319 struct fss_get *fsg = (struct fss_get *)data; 320 #ifndef _LP64 321 struct fss_get50 *fsg50 = (struct fss_get50 *)data; 322 #endif 323 324 switch (cmd) { 325 case FSSIOCSET50: 326 fss = &_fss; 327 fss->fss_mount = fss50->fss_mount; 328 fss->fss_bstore = fss50->fss_bstore; 329 fss->fss_csize = fss50->fss_csize; 330 fss->fss_flags = 0; 331 /* Fall through */ 332 case FSSIOCSET: 333 mutex_enter(&sc->sc_lock); 334 if ((flag & FWRITE) == 0) 335 error = EPERM; 336 else if ((sc->sc_flags & FSS_ACTIVE) != 0) 337 error = EBUSY; 338 else 339 error = fss_create_snapshot(sc, fss, l); 340 if (error == 0) 341 sc->sc_uflags = fss->fss_flags; 342 mutex_exit(&sc->sc_lock); 343 break; 344 345 case FSSIOCCLR: 346 mutex_enter(&sc->sc_lock); 347 if ((flag & FWRITE) == 0) 348 error = EPERM; 349 else if ((sc->sc_flags & FSS_ACTIVE) == 0) 350 error = ENXIO; 351 else 352 error = fss_delete_snapshot(sc, l); 353 mutex_exit(&sc->sc_lock); 354 break; 355 356 #ifndef _LP64 357 case FSSIOCGET50: 358 mutex_enter(&sc->sc_lock); 359 switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) { 360 case FSS_ACTIVE: 361 memcpy(fsg50->fsg_mount, sc->sc_mntname, MNAMELEN); 362 fsg50->fsg_csize = FSS_CLSIZE(sc); 363 timeval_to_timeval50(&sc->sc_time, &fsg50->fsg_time); 364 fsg50->fsg_mount_size = sc->sc_clcount; 365 fsg50->fsg_bs_size = sc->sc_clnext; 366 error = 0; 367 break; 368 case FSS_PERSISTENT | FSS_ACTIVE: 369 memcpy(fsg50->fsg_mount, sc->sc_mntname, MNAMELEN); 370 fsg50->fsg_csize = 0; 371 timeval_to_timeval50(&sc->sc_time, &fsg50->fsg_time); 372 fsg50->fsg_mount_size = 0; 373 fsg50->fsg_bs_size = 0; 374 error = 0; 375 break; 376 default: 377 error = ENXIO; 378 break; 379 } 380 mutex_exit(&sc->sc_lock); 381 break; 382 #endif /* _LP64 */ 383 384 case FSSIOCGET: 385 mutex_enter(&sc->sc_lock); 386 switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) { 387 case FSS_ACTIVE: 388 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 389 fsg->fsg_csize = FSS_CLSIZE(sc); 390 fsg->fsg_time = sc->sc_time; 391 fsg->fsg_mount_size = sc->sc_clcount; 392 fsg->fsg_bs_size = sc->sc_clnext; 393 error = 0; 394 break; 395 case FSS_PERSISTENT | FSS_ACTIVE: 396 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 397 fsg->fsg_csize = 0; 398 fsg->fsg_time = sc->sc_time; 399 fsg->fsg_mount_size = 0; 400 fsg->fsg_bs_size = 0; 401 error = 0; 402 break; 403 default: 404 error = ENXIO; 405 break; 406 } 407 mutex_exit(&sc->sc_lock); 408 break; 409 410 case FSSIOFSET: 411 mutex_enter(&sc->sc_slock); 412 sc->sc_uflags = *(int *)data; 413 mutex_exit(&sc->sc_slock); 414 error = 0; 415 break; 416 417 case FSSIOFGET: 418 mutex_enter(&sc->sc_slock); 419 *(int *)data = sc->sc_uflags; 420 mutex_exit(&sc->sc_slock); 421 error = 0; 422 break; 423 424 default: 425 error = EINVAL; 426 break; 427 } 428 429 return error; 430 } 431 432 int 433 fss_size(dev_t dev) 434 { 435 return -1; 436 } 437 438 int 439 fss_dump(dev_t dev, daddr_t blkno, void *va, 440 size_t size) 441 { 442 return EROFS; 443 } 444 445 /* 446 * An error occurred reading or writing the snapshot or backing store. 447 * If it is the first error log to console and disestablish cow handler. 448 * The caller holds the mutex. 449 */ 450 static inline void 451 fss_error(struct fss_softc *sc, const char *msg) 452 { 453 454 if ((sc->sc_flags & (FSS_ACTIVE | FSS_ERROR)) != FSS_ACTIVE) 455 return; 456 457 aprint_error_dev(sc->sc_dev, "snapshot invalid: %s\n", msg); 458 if ((sc->sc_flags & FSS_PERSISTENT) == 0) 459 fscow_disestablish(sc->sc_mount, fss_copy_on_write, sc); 460 sc->sc_flags |= FSS_ERROR; 461 } 462 463 /* 464 * Allocate the variable sized parts of the softc and 465 * fork the kernel thread. 466 * 467 * The fields sc_clcount, sc_clshift, sc_cache_size and sc_indir_size 468 * must be initialized. 469 */ 470 static int 471 fss_softc_alloc(struct fss_softc *sc) 472 { 473 int i, error; 474 475 if ((sc->sc_flags & FSS_PERSISTENT) == 0) { 476 sc->sc_copied = 477 kmem_zalloc(howmany(sc->sc_clcount, NBBY), KM_SLEEP); 478 sc->sc_cache = kmem_alloc(sc->sc_cache_size * 479 sizeof(struct fss_cache), KM_SLEEP); 480 for (i = 0; i < sc->sc_cache_size; i++) { 481 sc->sc_cache[i].fc_type = FSS_CACHE_FREE; 482 sc->sc_cache[i].fc_data = 483 kmem_alloc(FSS_CLSIZE(sc), KM_SLEEP); 484 cv_init(&sc->sc_cache[i].fc_state_cv, "cowwait1"); 485 } 486 487 sc->sc_indir_valid = 488 kmem_zalloc(howmany(sc->sc_indir_size, NBBY), KM_SLEEP); 489 sc->sc_indir_data = kmem_zalloc(FSS_CLSIZE(sc), KM_SLEEP); 490 } else { 491 sc->sc_copied = NULL; 492 sc->sc_cache = NULL; 493 sc->sc_indir_valid = NULL; 494 sc->sc_indir_data = NULL; 495 } 496 497 sc->sc_flags |= FSS_BS_THREAD; 498 if ((error = kthread_create(PRI_BIO, KTHREAD_MUSTJOIN, NULL, 499 fss_bs_thread, sc, &sc->sc_bs_lwp, 500 "%s", device_xname(sc->sc_dev))) != 0) { 501 sc->sc_flags &= ~FSS_BS_THREAD; 502 return error; 503 } 504 505 disk_attach(sc->sc_dkdev); 506 507 return 0; 508 } 509 510 /* 511 * Free the variable sized parts of the softc. 512 */ 513 static void 514 fss_softc_free(struct fss_softc *sc) 515 { 516 int i; 517 518 if ((sc->sc_flags & FSS_BS_THREAD) != 0) { 519 mutex_enter(&sc->sc_slock); 520 sc->sc_flags &= ~FSS_BS_THREAD; 521 cv_signal(&sc->sc_work_cv); 522 mutex_exit(&sc->sc_slock); 523 kthread_join(sc->sc_bs_lwp); 524 525 disk_detach(sc->sc_dkdev); 526 } 527 528 if (sc->sc_copied != NULL) 529 kmem_free(sc->sc_copied, howmany(sc->sc_clcount, NBBY)); 530 sc->sc_copied = NULL; 531 532 if (sc->sc_cache != NULL) { 533 for (i = 0; i < sc->sc_cache_size; i++) 534 if (sc->sc_cache[i].fc_data != NULL) { 535 cv_destroy(&sc->sc_cache[i].fc_state_cv); 536 kmem_free(sc->sc_cache[i].fc_data, 537 FSS_CLSIZE(sc)); 538 } 539 kmem_free(sc->sc_cache, 540 sc->sc_cache_size*sizeof(struct fss_cache)); 541 } 542 sc->sc_cache = NULL; 543 544 if (sc->sc_indir_valid != NULL) 545 kmem_free(sc->sc_indir_valid, howmany(sc->sc_indir_size, NBBY)); 546 sc->sc_indir_valid = NULL; 547 548 if (sc->sc_indir_data != NULL) 549 kmem_free(sc->sc_indir_data, FSS_CLSIZE(sc)); 550 sc->sc_indir_data = NULL; 551 } 552 553 /* 554 * Set all active snapshots on this file system into ERROR state. 555 */ 556 static void 557 fss_unmount_hook(struct mount *mp) 558 { 559 int i; 560 struct fss_softc *sc; 561 562 mutex_enter(&fss_device_lock); 563 for (i = 0; i < fss_cd.cd_ndevs; i++) { 564 if ((sc = device_lookup_private(&fss_cd, i)) == NULL) 565 continue; 566 mutex_enter(&sc->sc_slock); 567 if ((sc->sc_flags & FSS_ACTIVE) != 0 && sc->sc_mount == mp) 568 fss_error(sc, "forced by unmount"); 569 mutex_exit(&sc->sc_slock); 570 } 571 mutex_exit(&fss_device_lock); 572 } 573 574 /* 575 * A buffer is written to the snapshotted block device. Copy to 576 * backing store if needed. 577 */ 578 static int 579 fss_copy_on_write(void *v, struct buf *bp, bool data_valid) 580 { 581 int error; 582 u_int32_t cl, ch, c; 583 struct fss_softc *sc = v; 584 585 mutex_enter(&sc->sc_slock); 586 if (!FSS_ISVALID(sc)) { 587 mutex_exit(&sc->sc_slock); 588 return 0; 589 } 590 591 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 592 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 593 error = 0; 594 if (curlwp == uvm.pagedaemon_lwp) { 595 for (c = cl; c <= ch; c++) 596 if (isclr(sc->sc_copied, c)) { 597 error = ENOMEM; 598 break; 599 } 600 } 601 mutex_exit(&sc->sc_slock); 602 603 if (error == 0) 604 for (c = cl; c <= ch; c++) { 605 error = fss_read_cluster(sc, c); 606 if (error) 607 break; 608 } 609 610 return error; 611 } 612 613 /* 614 * Lookup and open needed files. 615 * 616 * For file system internal snapshot initializes sc_mntname, sc_mount, 617 * sc_bs_vp and sc_time. 618 * 619 * Otherwise returns dev and size of the underlying block device. 620 * Initializes sc_mntname, sc_mount, sc_bdev, sc_bs_vp and sc_mount 621 */ 622 static int 623 fss_create_files(struct fss_softc *sc, struct fss_set *fss, 624 off_t *bsize, struct lwp *l) 625 { 626 int error, bits, fsbsize; 627 uint64_t numsec; 628 unsigned int secsize; 629 struct timespec ts; 630 /* nd -> nd2 to reduce mistakes while updating only some namei calls */ 631 struct pathbuf *pb2; 632 struct nameidata nd2; 633 struct vnode *vp; 634 635 /* 636 * Get the mounted file system. 637 */ 638 639 error = namei_simple_user(fss->fss_mount, 640 NSM_FOLLOW_NOEMULROOT, &vp); 641 if (error != 0) 642 return error; 643 644 if ((vp->v_vflag & VV_ROOT) != VV_ROOT) { 645 vrele(vp); 646 return EINVAL; 647 } 648 649 sc->sc_mount = vp->v_mount; 650 memcpy(sc->sc_mntname, sc->sc_mount->mnt_stat.f_mntonname, MNAMELEN); 651 652 vrele(vp); 653 654 /* 655 * Check for file system internal snapshot. 656 */ 657 658 error = namei_simple_user(fss->fss_bstore, 659 NSM_FOLLOW_NOEMULROOT, &vp); 660 if (error != 0) 661 return error; 662 663 if (vp->v_type == VREG && vp->v_mount == sc->sc_mount) { 664 sc->sc_flags |= FSS_PERSISTENT; 665 sc->sc_bs_vp = vp; 666 667 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 668 bits = sizeof(sc->sc_bs_bshift)*NBBY; 669 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < bits; 670 sc->sc_bs_bshift++) 671 if (FSS_FSBSIZE(sc) == fsbsize) 672 break; 673 if (sc->sc_bs_bshift >= bits) 674 return EINVAL; 675 676 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 677 sc->sc_clshift = 0; 678 679 if ((fss->fss_flags & FSS_UNLINK_ON_CREATE) != 0) { 680 error = do_sys_unlink(fss->fss_bstore, UIO_USERSPACE); 681 if (error) 682 return error; 683 } 684 error = vn_lock(vp, LK_EXCLUSIVE); 685 if (error != 0) 686 return error; 687 error = VFS_SNAPSHOT(sc->sc_mount, sc->sc_bs_vp, &ts); 688 TIMESPEC_TO_TIMEVAL(&sc->sc_time, &ts); 689 690 VOP_UNLOCK(sc->sc_bs_vp); 691 692 return error; 693 } 694 vrele(vp); 695 696 /* 697 * Get the block device it is mounted on and its size. 698 */ 699 700 error = spec_node_lookup_by_mount(sc->sc_mount, &vp); 701 if (error) 702 return error; 703 sc->sc_bdev = vp->v_rdev; 704 705 error = getdisksize(vp, &numsec, &secsize); 706 vrele(vp); 707 if (error) 708 return error; 709 710 *bsize = (off_t)numsec*secsize; 711 712 /* 713 * Get the backing store 714 */ 715 716 error = pathbuf_copyin(fss->fss_bstore, &pb2); 717 if (error) { 718 return error; 719 } 720 NDINIT(&nd2, LOOKUP, FOLLOW, pb2); 721 if ((error = vn_open(&nd2, FREAD|FWRITE, 0)) != 0) { 722 pathbuf_destroy(pb2); 723 return error; 724 } 725 VOP_UNLOCK(nd2.ni_vp); 726 727 sc->sc_bs_vp = nd2.ni_vp; 728 729 if (nd2.ni_vp->v_type != VREG && nd2.ni_vp->v_type != VCHR) { 730 pathbuf_destroy(pb2); 731 return EINVAL; 732 } 733 pathbuf_destroy(pb2); 734 735 if ((fss->fss_flags & FSS_UNLINK_ON_CREATE) != 0) { 736 error = do_sys_unlink(fss->fss_bstore, UIO_USERSPACE); 737 if (error) 738 return error; 739 } 740 if (sc->sc_bs_vp->v_type == VREG) { 741 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 742 if (fsbsize & (fsbsize-1)) /* No power of two */ 743 return EINVAL; 744 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < 32; 745 sc->sc_bs_bshift++) 746 if (FSS_FSBSIZE(sc) == fsbsize) 747 break; 748 if (sc->sc_bs_bshift >= 32) 749 return EINVAL; 750 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 751 } else { 752 sc->sc_bs_bshift = DEV_BSHIFT; 753 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 754 } 755 756 return 0; 757 } 758 759 /* 760 * Create a snapshot. 761 */ 762 static int 763 fss_create_snapshot(struct fss_softc *sc, struct fss_set *fss, struct lwp *l) 764 { 765 int len, error; 766 u_int32_t csize; 767 off_t bsize; 768 769 bsize = 0; /* XXX gcc */ 770 771 /* 772 * Open needed files. 773 */ 774 if ((error = fss_create_files(sc, fss, &bsize, l)) != 0) 775 goto bad; 776 777 if (sc->sc_flags & FSS_PERSISTENT) { 778 fss_softc_alloc(sc); 779 sc->sc_flags |= FSS_ACTIVE; 780 return 0; 781 } 782 783 /* 784 * Set cluster size. Must be a power of two and 785 * a multiple of backing store block size. 786 */ 787 if (fss->fss_csize <= 0) 788 csize = MAXPHYS; 789 else 790 csize = fss->fss_csize; 791 if (bsize/csize > FSS_CLUSTER_MAX) 792 csize = bsize/FSS_CLUSTER_MAX+1; 793 794 for (sc->sc_clshift = sc->sc_bs_bshift; sc->sc_clshift < 32; 795 sc->sc_clshift++) 796 if (FSS_CLSIZE(sc) >= csize) 797 break; 798 if (sc->sc_clshift >= 32) { 799 error = EINVAL; 800 goto bad; 801 } 802 sc->sc_clmask = FSS_CLSIZE(sc)-1; 803 804 /* 805 * Set number of cache slots. 806 */ 807 if (FSS_CLSIZE(sc) <= 8192) 808 sc->sc_cache_size = 32; 809 else if (FSS_CLSIZE(sc) <= 65536) 810 sc->sc_cache_size = 8; 811 else 812 sc->sc_cache_size = 4; 813 814 /* 815 * Set number of clusters and size of last cluster. 816 */ 817 sc->sc_clcount = FSS_BTOCL(sc, bsize-1)+1; 818 sc->sc_clresid = FSS_CLOFF(sc, bsize-1)+1; 819 820 /* 821 * Set size of indirect table. 822 */ 823 len = sc->sc_clcount*sizeof(u_int32_t); 824 sc->sc_indir_size = FSS_BTOCL(sc, len)+1; 825 sc->sc_clnext = sc->sc_indir_size; 826 sc->sc_indir_cur = 0; 827 828 if ((error = fss_softc_alloc(sc)) != 0) 829 goto bad; 830 831 /* 832 * Activate the snapshot. 833 */ 834 835 if ((error = vfs_suspend(sc->sc_mount, 0)) != 0) 836 goto bad; 837 838 microtime(&sc->sc_time); 839 840 vrele_flush(sc->sc_mount); 841 error = VFS_SYNC(sc->sc_mount, MNT_WAIT, curlwp->l_cred); 842 if (error == 0) 843 error = fscow_establish(sc->sc_mount, fss_copy_on_write, sc); 844 if (error == 0) 845 sc->sc_flags |= FSS_ACTIVE; 846 847 vfs_resume(sc->sc_mount); 848 849 if (error != 0) 850 goto bad; 851 852 aprint_debug_dev(sc->sc_dev, "%s snapshot active\n", sc->sc_mntname); 853 aprint_debug_dev(sc->sc_dev, 854 "%u clusters of %u, %u cache slots, %u indir clusters\n", 855 sc->sc_clcount, FSS_CLSIZE(sc), 856 sc->sc_cache_size, sc->sc_indir_size); 857 858 return 0; 859 860 bad: 861 fss_softc_free(sc); 862 if (sc->sc_bs_vp != NULL) { 863 if (sc->sc_flags & FSS_PERSISTENT) 864 vrele(sc->sc_bs_vp); 865 else 866 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 867 } 868 sc->sc_bs_vp = NULL; 869 870 return error; 871 } 872 873 /* 874 * Delete a snapshot. 875 */ 876 static int 877 fss_delete_snapshot(struct fss_softc *sc, struct lwp *l) 878 { 879 880 if ((sc->sc_flags & (FSS_PERSISTENT | FSS_ERROR)) == 0) 881 fscow_disestablish(sc->sc_mount, fss_copy_on_write, sc); 882 883 mutex_enter(&sc->sc_slock); 884 sc->sc_flags &= ~(FSS_ACTIVE|FSS_ERROR); 885 sc->sc_mount = NULL; 886 sc->sc_bdev = NODEV; 887 mutex_exit(&sc->sc_slock); 888 889 fss_softc_free(sc); 890 if (sc->sc_flags & FSS_PERSISTENT) 891 vrele(sc->sc_bs_vp); 892 else 893 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 894 sc->sc_bs_vp = NULL; 895 sc->sc_flags &= ~FSS_PERSISTENT; 896 897 return 0; 898 } 899 900 /* 901 * Read a cluster from the snapshotted block device to the cache. 902 */ 903 static int 904 fss_read_cluster(struct fss_softc *sc, u_int32_t cl) 905 { 906 int error, todo, offset, len; 907 daddr_t dblk; 908 struct buf *bp, *mbp; 909 struct fss_cache *scp, *scl; 910 911 /* 912 * Get a free cache slot. 913 */ 914 scl = sc->sc_cache+sc->sc_cache_size; 915 916 mutex_enter(&sc->sc_slock); 917 918 restart: 919 if (isset(sc->sc_copied, cl) || !FSS_ISVALID(sc)) { 920 mutex_exit(&sc->sc_slock); 921 return 0; 922 } 923 924 for (scp = sc->sc_cache; scp < scl; scp++) 925 if (scp->fc_cluster == cl) { 926 if (scp->fc_type == FSS_CACHE_VALID) { 927 mutex_exit(&sc->sc_slock); 928 return 0; 929 } else if (scp->fc_type == FSS_CACHE_BUSY) { 930 cv_wait(&scp->fc_state_cv, &sc->sc_slock); 931 goto restart; 932 } 933 } 934 935 for (scp = sc->sc_cache; scp < scl; scp++) 936 if (scp->fc_type == FSS_CACHE_FREE) { 937 scp->fc_type = FSS_CACHE_BUSY; 938 scp->fc_cluster = cl; 939 break; 940 } 941 if (scp >= scl) { 942 cv_wait(&sc->sc_cache_cv, &sc->sc_slock); 943 goto restart; 944 } 945 946 mutex_exit(&sc->sc_slock); 947 948 /* 949 * Start the read. 950 */ 951 dblk = btodb(FSS_CLTOB(sc, cl)); 952 if (cl == sc->sc_clcount-1) { 953 todo = sc->sc_clresid; 954 memset((char *)scp->fc_data + todo, 0, FSS_CLSIZE(sc) - todo); 955 } else 956 todo = FSS_CLSIZE(sc); 957 offset = 0; 958 mbp = getiobuf(NULL, true); 959 mbp->b_bufsize = todo; 960 mbp->b_data = scp->fc_data; 961 mbp->b_resid = mbp->b_bcount = todo; 962 mbp->b_flags = B_READ; 963 mbp->b_cflags = BC_BUSY; 964 mbp->b_dev = sc->sc_bdev; 965 while (todo > 0) { 966 len = todo; 967 if (len > MAXPHYS) 968 len = MAXPHYS; 969 if (btodb(FSS_CLTOB(sc, cl)) == dblk && len == todo) 970 bp = mbp; 971 else { 972 bp = getiobuf(NULL, true); 973 nestiobuf_setup(mbp, bp, offset, len); 974 } 975 bp->b_lblkno = 0; 976 bp->b_blkno = dblk; 977 bdev_strategy(bp); 978 dblk += btodb(len); 979 offset += len; 980 todo -= len; 981 } 982 error = biowait(mbp); 983 putiobuf(mbp); 984 985 mutex_enter(&sc->sc_slock); 986 scp->fc_type = (error ? FSS_CACHE_FREE : FSS_CACHE_VALID); 987 cv_broadcast(&scp->fc_state_cv); 988 if (error == 0) { 989 setbit(sc->sc_copied, scp->fc_cluster); 990 cv_signal(&sc->sc_work_cv); 991 } 992 mutex_exit(&sc->sc_slock); 993 994 return error; 995 } 996 997 /* 998 * Read/write clusters from/to backing store. 999 * For persistent snapshots must be called with cl == 0. off is the 1000 * offset into the snapshot. 1001 */ 1002 static int 1003 fss_bs_io(struct fss_softc *sc, fss_io_type rw, 1004 u_int32_t cl, off_t off, int len, void *data) 1005 { 1006 int error; 1007 1008 off += FSS_CLTOB(sc, cl); 1009 1010 vn_lock(sc->sc_bs_vp, LK_EXCLUSIVE|LK_RETRY); 1011 1012 error = vn_rdwr((rw == FSS_READ ? UIO_READ : UIO_WRITE), sc->sc_bs_vp, 1013 data, len, off, UIO_SYSSPACE, 1014 IO_ADV_ENCODE(POSIX_FADV_NOREUSE) | IO_NODELOCKED, 1015 sc->sc_bs_lwp->l_cred, NULL, NULL); 1016 if (error == 0) { 1017 mutex_enter(sc->sc_bs_vp->v_interlock); 1018 error = VOP_PUTPAGES(sc->sc_bs_vp, trunc_page(off), 1019 round_page(off+len), PGO_CLEANIT | PGO_FREE | PGO_SYNCIO); 1020 } 1021 1022 VOP_UNLOCK(sc->sc_bs_vp); 1023 1024 return error; 1025 } 1026 1027 /* 1028 * Get a pointer to the indirect slot for this cluster. 1029 */ 1030 static u_int32_t * 1031 fss_bs_indir(struct fss_softc *sc, u_int32_t cl) 1032 { 1033 u_int32_t icl; 1034 int ioff; 1035 1036 icl = cl/(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 1037 ioff = cl%(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 1038 1039 if (sc->sc_indir_cur == icl) 1040 return &sc->sc_indir_data[ioff]; 1041 1042 if (sc->sc_indir_dirty) { 1043 if (fss_bs_io(sc, FSS_WRITE, sc->sc_indir_cur, 0, 1044 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 1045 return NULL; 1046 setbit(sc->sc_indir_valid, sc->sc_indir_cur); 1047 } 1048 1049 sc->sc_indir_dirty = 0; 1050 sc->sc_indir_cur = icl; 1051 1052 if (isset(sc->sc_indir_valid, sc->sc_indir_cur)) { 1053 if (fss_bs_io(sc, FSS_READ, sc->sc_indir_cur, 0, 1054 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 1055 return NULL; 1056 } else 1057 memset(sc->sc_indir_data, 0, FSS_CLSIZE(sc)); 1058 1059 return &sc->sc_indir_data[ioff]; 1060 } 1061 1062 /* 1063 * The kernel thread (one for every active snapshot). 1064 * 1065 * After wakeup it cleans the cache and runs the I/O requests. 1066 */ 1067 static void 1068 fss_bs_thread(void *arg) 1069 { 1070 bool thread_idle, is_valid; 1071 int error, i, todo, len, crotor, is_read; 1072 long off; 1073 char *addr; 1074 u_int32_t c, cl, ch, *indirp; 1075 struct buf *bp, *nbp; 1076 struct fss_softc *sc; 1077 struct fss_cache *scp, *scl; 1078 1079 sc = arg; 1080 scl = sc->sc_cache+sc->sc_cache_size; 1081 crotor = 0; 1082 thread_idle = false; 1083 1084 mutex_enter(&sc->sc_slock); 1085 1086 for (;;) { 1087 if (thread_idle) 1088 cv_wait(&sc->sc_work_cv, &sc->sc_slock); 1089 thread_idle = true; 1090 if ((sc->sc_flags & FSS_BS_THREAD) == 0) { 1091 mutex_exit(&sc->sc_slock); 1092 kthread_exit(0); 1093 } 1094 1095 /* 1096 * Process I/O requests (persistent) 1097 */ 1098 1099 if (sc->sc_flags & FSS_PERSISTENT) { 1100 if ((bp = bufq_get(sc->sc_bufq)) == NULL) 1101 continue; 1102 is_valid = FSS_ISVALID(sc); 1103 is_read = (bp->b_flags & B_READ); 1104 thread_idle = false; 1105 mutex_exit(&sc->sc_slock); 1106 1107 if (is_valid) { 1108 disk_busy(sc->sc_dkdev); 1109 error = fss_bs_io(sc, FSS_READ, 0, 1110 dbtob(bp->b_blkno), bp->b_bcount, 1111 bp->b_data); 1112 disk_unbusy(sc->sc_dkdev, 1113 (error ? 0 : bp->b_bcount), is_read); 1114 } else 1115 error = ENXIO; 1116 1117 bp->b_error = error; 1118 bp->b_resid = (error ? bp->b_bcount : 0); 1119 biodone(bp); 1120 1121 mutex_enter(&sc->sc_slock); 1122 continue; 1123 } 1124 1125 /* 1126 * Clean the cache 1127 */ 1128 for (i = 0; i < sc->sc_cache_size; i++) { 1129 crotor = (crotor + 1) % sc->sc_cache_size; 1130 scp = sc->sc_cache + crotor; 1131 if (scp->fc_type != FSS_CACHE_VALID) 1132 continue; 1133 mutex_exit(&sc->sc_slock); 1134 1135 thread_idle = false; 1136 indirp = fss_bs_indir(sc, scp->fc_cluster); 1137 if (indirp != NULL) { 1138 error = fss_bs_io(sc, FSS_WRITE, sc->sc_clnext, 1139 0, FSS_CLSIZE(sc), scp->fc_data); 1140 } else 1141 error = EIO; 1142 1143 mutex_enter(&sc->sc_slock); 1144 if (error == 0) { 1145 *indirp = sc->sc_clnext++; 1146 sc->sc_indir_dirty = 1; 1147 } else 1148 fss_error(sc, "write error on backing store"); 1149 1150 scp->fc_type = FSS_CACHE_FREE; 1151 cv_broadcast(&sc->sc_cache_cv); 1152 break; 1153 } 1154 1155 /* 1156 * Process I/O requests 1157 */ 1158 if ((bp = bufq_get(sc->sc_bufq)) == NULL) 1159 continue; 1160 is_valid = FSS_ISVALID(sc); 1161 is_read = (bp->b_flags & B_READ); 1162 thread_idle = false; 1163 1164 if (!is_valid) { 1165 mutex_exit(&sc->sc_slock); 1166 1167 bp->b_error = ENXIO; 1168 bp->b_resid = bp->b_bcount; 1169 biodone(bp); 1170 1171 mutex_enter(&sc->sc_slock); 1172 continue; 1173 } 1174 1175 disk_busy(sc->sc_dkdev); 1176 1177 /* 1178 * First read from the snapshotted block device unless 1179 * this request is completely covered by backing store. 1180 */ 1181 1182 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 1183 off = FSS_CLOFF(sc, dbtob(bp->b_blkno)); 1184 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 1185 error = 0; 1186 bp->b_resid = 0; 1187 bp->b_error = 0; 1188 for (c = cl; c <= ch; c++) { 1189 if (isset(sc->sc_copied, c)) 1190 continue; 1191 mutex_exit(&sc->sc_slock); 1192 1193 /* Not on backing store, read from device. */ 1194 nbp = getiobuf(NULL, true); 1195 nbp->b_flags = B_READ; 1196 nbp->b_resid = nbp->b_bcount = bp->b_bcount; 1197 nbp->b_bufsize = bp->b_bcount; 1198 nbp->b_data = bp->b_data; 1199 nbp->b_blkno = bp->b_blkno; 1200 nbp->b_lblkno = 0; 1201 nbp->b_dev = sc->sc_bdev; 1202 SET(nbp->b_cflags, BC_BUSY); /* mark buffer busy */ 1203 1204 bdev_strategy(nbp); 1205 1206 error = biowait(nbp); 1207 if (error != 0) { 1208 bp->b_resid = bp->b_bcount; 1209 bp->b_error = nbp->b_error; 1210 disk_unbusy(sc->sc_dkdev, 0, is_read); 1211 biodone(bp); 1212 } 1213 putiobuf(nbp); 1214 1215 mutex_enter(&sc->sc_slock); 1216 break; 1217 } 1218 if (error) 1219 continue; 1220 1221 /* 1222 * Replace those parts that have been saved to backing store. 1223 */ 1224 1225 addr = bp->b_data; 1226 todo = bp->b_bcount; 1227 for (c = cl; c <= ch; c++, off = 0, todo -= len, addr += len) { 1228 len = FSS_CLSIZE(sc)-off; 1229 if (len > todo) 1230 len = todo; 1231 if (isclr(sc->sc_copied, c)) 1232 continue; 1233 mutex_exit(&sc->sc_slock); 1234 1235 indirp = fss_bs_indir(sc, c); 1236 if (indirp == NULL || *indirp == 0) { 1237 /* 1238 * Not on backing store. Either in cache 1239 * or hole in the snapshotted block device. 1240 */ 1241 1242 mutex_enter(&sc->sc_slock); 1243 for (scp = sc->sc_cache; scp < scl; scp++) 1244 if (scp->fc_type == FSS_CACHE_VALID && 1245 scp->fc_cluster == c) 1246 break; 1247 if (scp < scl) 1248 memcpy(addr, (char *)scp->fc_data+off, 1249 len); 1250 else 1251 memset(addr, 0, len); 1252 continue; 1253 } 1254 1255 /* 1256 * Read from backing store. 1257 */ 1258 error = 1259 fss_bs_io(sc, FSS_READ, *indirp, off, len, addr); 1260 1261 mutex_enter(&sc->sc_slock); 1262 if (error) { 1263 bp->b_resid = bp->b_bcount; 1264 bp->b_error = error; 1265 break; 1266 } 1267 } 1268 mutex_exit(&sc->sc_slock); 1269 1270 disk_unbusy(sc->sc_dkdev, (error ? 0 : bp->b_bcount), is_read); 1271 biodone(bp); 1272 1273 mutex_enter(&sc->sc_slock); 1274 } 1275 } 1276 1277 #ifdef _MODULE 1278 1279 #include <sys/module.h> 1280 1281 MODULE(MODULE_CLASS_DRIVER, fss, NULL); 1282 CFDRIVER_DECL(fss, DV_DISK, NULL); 1283 1284 devmajor_t fss_bmajor = -1, fss_cmajor = -1; 1285 1286 static int 1287 fss_modcmd(modcmd_t cmd, void *arg) 1288 { 1289 int error = 0; 1290 1291 switch (cmd) { 1292 case MODULE_CMD_INIT: 1293 mutex_init(&fss_device_lock, MUTEX_DEFAULT, IPL_NONE); 1294 error = config_cfdriver_attach(&fss_cd); 1295 if (error) { 1296 mutex_destroy(&fss_device_lock); 1297 break; 1298 } 1299 error = config_cfattach_attach(fss_cd.cd_name, &fss_ca); 1300 if (error) { 1301 config_cfdriver_detach(&fss_cd); 1302 mutex_destroy(&fss_device_lock); 1303 break; 1304 } 1305 error = devsw_attach(fss_cd.cd_name, 1306 &fss_bdevsw, &fss_bmajor, &fss_cdevsw, &fss_cmajor); 1307 1308 if (error) { 1309 config_cfattach_detach(fss_cd.cd_name, &fss_ca); 1310 config_cfdriver_detach(&fss_cd); 1311 mutex_destroy(&fss_device_lock); 1312 break; 1313 } 1314 break; 1315 1316 case MODULE_CMD_FINI: 1317 devsw_detach(&fss_bdevsw, &fss_cdevsw); 1318 error = config_cfattach_detach(fss_cd.cd_name, &fss_ca); 1319 if (error) { 1320 devsw_attach(fss_cd.cd_name, &fss_bdevsw, &fss_bmajor, 1321 &fss_cdevsw, &fss_cmajor); 1322 break; 1323 } 1324 config_cfdriver_detach(&fss_cd); 1325 mutex_destroy(&fss_device_lock); 1326 break; 1327 1328 default: 1329 error = ENOTTY; 1330 break; 1331 } 1332 1333 return error; 1334 } 1335 1336 #endif /* _MODULE */ 1337