1 /* $NetBSD: fss.c,v 1.83 2012/07/28 16:14:17 hannken Exp $ */ 2 3 /*- 4 * Copyright (c) 2003 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Juergen Hannken-Illjes. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 /* 33 * File system snapshot disk driver. 34 * 35 * Block/character interface to the snapshot of a mounted file system. 36 */ 37 38 #include <sys/cdefs.h> 39 __KERNEL_RCSID(0, "$NetBSD: fss.c,v 1.83 2012/07/28 16:14:17 hannken Exp $"); 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/namei.h> 44 #include <sys/proc.h> 45 #include <sys/errno.h> 46 #include <sys/malloc.h> 47 #include <sys/buf.h> 48 #include <sys/ioctl.h> 49 #include <sys/disklabel.h> 50 #include <sys/device.h> 51 #include <sys/disk.h> 52 #include <sys/stat.h> 53 #include <sys/mount.h> 54 #include <sys/vnode.h> 55 #include <sys/file.h> 56 #include <sys/uio.h> 57 #include <sys/conf.h> 58 #include <sys/kthread.h> 59 #include <sys/fstrans.h> 60 #include <sys/simplelock.h> 61 #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 69 void fssattach(int); 70 71 dev_type_open(fss_open); 72 dev_type_close(fss_close); 73 dev_type_read(fss_read); 74 dev_type_write(fss_write); 75 dev_type_ioctl(fss_ioctl); 76 dev_type_strategy(fss_strategy); 77 dev_type_dump(fss_dump); 78 dev_type_size(fss_size); 79 80 static void fss_unmount_hook(struct mount *); 81 static int fss_copy_on_write(void *, struct buf *, bool); 82 static inline void fss_error(struct fss_softc *, const char *); 83 static int fss_create_files(struct fss_softc *, struct fss_set *, 84 off_t *, struct lwp *); 85 static int fss_create_snapshot(struct fss_softc *, struct fss_set *, 86 struct lwp *); 87 static int fss_delete_snapshot(struct fss_softc *, struct lwp *); 88 static int fss_softc_alloc(struct fss_softc *); 89 static void fss_softc_free(struct fss_softc *); 90 static int fss_read_cluster(struct fss_softc *, u_int32_t); 91 static void fss_bs_thread(void *); 92 static int fss_bs_io(struct fss_softc *, fss_io_type, 93 u_int32_t, off_t, int, void *); 94 static u_int32_t *fss_bs_indir(struct fss_softc *, u_int32_t); 95 96 static kmutex_t fss_device_lock; /* Protect all units. */ 97 static int fss_num_attached = 0; /* Number of attached devices. */ 98 static struct vfs_hooks fss_vfs_hooks = { 99 .vh_unmount = fss_unmount_hook 100 }; 101 102 const struct bdevsw fss_bdevsw = { 103 fss_open, fss_close, fss_strategy, fss_ioctl, 104 fss_dump, fss_size, D_DISK | D_MPSAFE 105 }; 106 107 const 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 }; 111 112 static int fss_match(device_t, cfdata_t, void *); 113 static void fss_attach(device_t, device_t, void *); 114 static int fss_detach(device_t, int); 115 116 CFATTACH_DECL_NEW(fss, sizeof(struct fss_softc), 117 fss_match, fss_attach, fss_detach, NULL); 118 extern struct cfdriver fss_cd; 119 120 void 121 fssattach(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 } 128 129 static int 130 fss_match(device_t self, cfdata_t cfdata, void *aux) 131 { 132 return 1; 133 } 134 135 static void 136 fss_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 } 156 157 static int 158 fss_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 } 180 181 int 182 fss_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 } 215 216 int 217 fss_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 restart: 227 mutex_enter(&sc->sc_slock); 228 if ((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) != mflag) { 229 sc->sc_flags &= ~mflag; 230 mutex_exit(&sc->sc_slock); 231 return 0; 232 } 233 if ((sc->sc_flags & FSS_ACTIVE) != 0 && 234 (sc->sc_uflags & FSS_UNCONFIG_ON_CLOSE) != 0) { 235 sc->sc_uflags &= ~FSS_UNCONFIG_ON_CLOSE; 236 mutex_exit(&sc->sc_slock); 237 error = fss_ioctl(dev, FSSIOCCLR, NULL, FWRITE, l); 238 goto restart; 239 } 240 if ((sc->sc_flags & FSS_ACTIVE) != 0) { 241 mutex_exit(&sc->sc_slock); 242 return error; 243 } 244 if (! mutex_tryenter(&fss_device_lock)) { 245 mutex_exit(&sc->sc_slock); 246 goto restart; 247 } 248 249 KASSERT((sc->sc_flags & FSS_ACTIVE) == 0); 250 KASSERT((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) == mflag); 251 mutex_exit(&sc->sc_slock); 252 cf = device_cfdata(sc->sc_dev); 253 error = config_detach(sc->sc_dev, DETACH_QUIET); 254 if (! error) 255 free(cf, M_DEVBUF); 256 mutex_exit(&fss_device_lock); 257 258 return error; 259 } 260 261 void 262 fss_strategy(struct buf *bp) 263 { 264 const bool write = ((bp->b_flags & B_READ) != B_READ); 265 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(bp->b_dev)); 266 267 mutex_enter(&sc->sc_slock); 268 269 if (write || !FSS_ISVALID(sc)) { 270 271 mutex_exit(&sc->sc_slock); 272 273 bp->b_error = (write ? EROFS : ENXIO); 274 bp->b_resid = bp->b_bcount; 275 biodone(bp); 276 return; 277 } 278 279 bp->b_rawblkno = bp->b_blkno; 280 bufq_put(sc->sc_bufq, bp); 281 cv_signal(&sc->sc_work_cv); 282 283 mutex_exit(&sc->sc_slock); 284 } 285 286 int 287 fss_read(dev_t dev, struct uio *uio, int flags) 288 { 289 return physio(fss_strategy, NULL, dev, B_READ, minphys, uio); 290 } 291 292 int 293 fss_write(dev_t dev, struct uio *uio, int flags) 294 { 295 return physio(fss_strategy, NULL, dev, B_WRITE, minphys, uio); 296 } 297 298 int 299 fss_ioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 300 { 301 int error; 302 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(dev)); 303 struct fss_set _fss; 304 struct fss_set *fss = (struct fss_set *)data; 305 struct fss_set50 *fss50 = (struct fss_set50 *)data; 306 struct fss_get *fsg = (struct fss_get *)data; 307 #ifndef _LP64 308 struct fss_get50 *fsg50 = (struct fss_get50 *)data; 309 #endif 310 311 switch (cmd) { 312 case FSSIOCSET50: 313 fss = &_fss; 314 fss->fss_mount = fss50->fss_mount; 315 fss->fss_bstore = fss50->fss_bstore; 316 fss->fss_csize = fss50->fss_csize; 317 fss->fss_flags = 0; 318 /* Fall through */ 319 case FSSIOCSET: 320 mutex_enter(&sc->sc_lock); 321 if ((flag & FWRITE) == 0) 322 error = EPERM; 323 else if ((sc->sc_flags & FSS_ACTIVE) != 0) 324 error = EBUSY; 325 else 326 error = fss_create_snapshot(sc, fss, l); 327 if (error == 0) 328 sc->sc_uflags = fss->fss_flags; 329 mutex_exit(&sc->sc_lock); 330 break; 331 332 case FSSIOCCLR: 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 = ENXIO; 338 else 339 error = fss_delete_snapshot(sc, l); 340 mutex_exit(&sc->sc_lock); 341 break; 342 343 #ifndef _LP64 344 case FSSIOCGET50: 345 mutex_enter(&sc->sc_lock); 346 switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) { 347 case FSS_ACTIVE: 348 memcpy(fsg50->fsg_mount, sc->sc_mntname, MNAMELEN); 349 fsg50->fsg_csize = FSS_CLSIZE(sc); 350 timeval_to_timeval50(&sc->sc_time, &fsg50->fsg_time); 351 fsg50->fsg_mount_size = sc->sc_clcount; 352 fsg50->fsg_bs_size = sc->sc_clnext; 353 error = 0; 354 break; 355 case FSS_PERSISTENT | FSS_ACTIVE: 356 memcpy(fsg50->fsg_mount, sc->sc_mntname, MNAMELEN); 357 fsg50->fsg_csize = 0; 358 timeval_to_timeval50(&sc->sc_time, &fsg50->fsg_time); 359 fsg50->fsg_mount_size = 0; 360 fsg50->fsg_bs_size = 0; 361 error = 0; 362 break; 363 default: 364 error = ENXIO; 365 break; 366 } 367 mutex_exit(&sc->sc_lock); 368 break; 369 #endif /* _LP64 */ 370 371 case FSSIOCGET: 372 mutex_enter(&sc->sc_lock); 373 switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) { 374 case FSS_ACTIVE: 375 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 376 fsg->fsg_csize = FSS_CLSIZE(sc); 377 fsg->fsg_time = sc->sc_time; 378 fsg->fsg_mount_size = sc->sc_clcount; 379 fsg->fsg_bs_size = sc->sc_clnext; 380 error = 0; 381 break; 382 case FSS_PERSISTENT | FSS_ACTIVE: 383 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 384 fsg->fsg_csize = 0; 385 fsg->fsg_time = sc->sc_time; 386 fsg->fsg_mount_size = 0; 387 fsg->fsg_bs_size = 0; 388 error = 0; 389 break; 390 default: 391 error = ENXIO; 392 break; 393 } 394 mutex_exit(&sc->sc_lock); 395 break; 396 397 case FSSIOFSET: 398 mutex_enter(&sc->sc_slock); 399 sc->sc_uflags = *(int *)data; 400 mutex_exit(&sc->sc_slock); 401 error = 0; 402 break; 403 404 case FSSIOFGET: 405 mutex_enter(&sc->sc_slock); 406 *(int *)data = sc->sc_uflags; 407 mutex_exit(&sc->sc_slock); 408 error = 0; 409 break; 410 411 default: 412 error = EINVAL; 413 break; 414 } 415 416 return error; 417 } 418 419 int 420 fss_size(dev_t dev) 421 { 422 return -1; 423 } 424 425 int 426 fss_dump(dev_t dev, daddr_t blkno, void *va, 427 size_t size) 428 { 429 return EROFS; 430 } 431 432 /* 433 * An error occurred reading or writing the snapshot or backing store. 434 * If it is the first error log to console. 435 * The caller holds the mutex. 436 */ 437 static inline void 438 fss_error(struct fss_softc *sc, const char *msg) 439 { 440 441 if ((sc->sc_flags & (FSS_ACTIVE|FSS_ERROR)) == FSS_ACTIVE) 442 aprint_error_dev(sc->sc_dev, "snapshot invalid: %s\n", msg); 443 if ((sc->sc_flags & FSS_ACTIVE) == FSS_ACTIVE) 444 sc->sc_flags |= FSS_ERROR; 445 } 446 447 /* 448 * Allocate the variable sized parts of the softc and 449 * fork the kernel thread. 450 * 451 * The fields sc_clcount, sc_clshift, sc_cache_size and sc_indir_size 452 * must be initialized. 453 */ 454 static int 455 fss_softc_alloc(struct fss_softc *sc) 456 { 457 int i, error; 458 459 if ((sc->sc_flags & FSS_PERSISTENT) == 0) { 460 sc->sc_copied = 461 kmem_zalloc(howmany(sc->sc_clcount, NBBY), KM_SLEEP); 462 if (sc->sc_copied == NULL) 463 return(ENOMEM); 464 465 sc->sc_cache = kmem_alloc(sc->sc_cache_size * 466 sizeof(struct fss_cache), KM_SLEEP); 467 if (sc->sc_cache == NULL) 468 return(ENOMEM); 469 470 for (i = 0; i < sc->sc_cache_size; i++) { 471 sc->sc_cache[i].fc_type = FSS_CACHE_FREE; 472 sc->sc_cache[i].fc_data = 473 kmem_alloc(FSS_CLSIZE(sc), KM_SLEEP); 474 if (sc->sc_cache[i].fc_data == NULL) 475 return(ENOMEM); 476 cv_init(&sc->sc_cache[i].fc_state_cv, "cowwait1"); 477 } 478 479 sc->sc_indir_valid = 480 kmem_zalloc(howmany(sc->sc_indir_size, NBBY), KM_SLEEP); 481 if (sc->sc_indir_valid == NULL) 482 return(ENOMEM); 483 484 sc->sc_indir_data = kmem_zalloc(FSS_CLSIZE(sc), KM_SLEEP); 485 if (sc->sc_indir_data == NULL) 486 return(ENOMEM); 487 } else { 488 sc->sc_copied = NULL; 489 sc->sc_cache = NULL; 490 sc->sc_indir_valid = NULL; 491 sc->sc_indir_data = NULL; 492 } 493 494 sc->sc_flags |= FSS_BS_THREAD; 495 if ((error = kthread_create(PRI_BIO, KTHREAD_MUSTJOIN, NULL, 496 fss_bs_thread, sc, &sc->sc_bs_lwp, 497 "%s", device_xname(sc->sc_dev))) != 0) { 498 sc->sc_flags &= ~FSS_BS_THREAD; 499 return error; 500 } 501 502 disk_attach(sc->sc_dkdev); 503 504 return 0; 505 } 506 507 /* 508 * Free the variable sized parts of the softc. 509 */ 510 static void 511 fss_softc_free(struct fss_softc *sc) 512 { 513 int i; 514 515 if ((sc->sc_flags & FSS_BS_THREAD) != 0) { 516 mutex_enter(&sc->sc_slock); 517 sc->sc_flags &= ~FSS_BS_THREAD; 518 cv_signal(&sc->sc_work_cv); 519 mutex_exit(&sc->sc_slock); 520 kthread_join(sc->sc_bs_lwp); 521 522 disk_detach(sc->sc_dkdev); 523 } 524 525 if (sc->sc_copied != NULL) 526 kmem_free(sc->sc_copied, howmany(sc->sc_clcount, NBBY)); 527 sc->sc_copied = NULL; 528 529 if (sc->sc_cache != NULL) { 530 for (i = 0; i < sc->sc_cache_size; i++) 531 if (sc->sc_cache[i].fc_data != NULL) { 532 cv_destroy(&sc->sc_cache[i].fc_state_cv); 533 kmem_free(sc->sc_cache[i].fc_data, 534 FSS_CLSIZE(sc)); 535 } 536 kmem_free(sc->sc_cache, 537 sc->sc_cache_size*sizeof(struct fss_cache)); 538 } 539 sc->sc_cache = NULL; 540 541 if (sc->sc_indir_valid != NULL) 542 kmem_free(sc->sc_indir_valid, howmany(sc->sc_indir_size, NBBY)); 543 sc->sc_indir_valid = NULL; 544 545 if (sc->sc_indir_data != NULL) 546 kmem_free(sc->sc_indir_data, FSS_CLSIZE(sc)); 547 sc->sc_indir_data = NULL; 548 } 549 550 /* 551 * Set all active snapshots on this file system into ERROR state. 552 */ 553 static void 554 fss_unmount_hook(struct mount *mp) 555 { 556 int i; 557 struct fss_softc *sc; 558 559 mutex_enter(&fss_device_lock); 560 for (i = 0; i < fss_cd.cd_ndevs; i++) { 561 if ((sc = device_lookup_private(&fss_cd, i)) == NULL) 562 continue; 563 mutex_enter(&sc->sc_slock); 564 if ((sc->sc_flags & FSS_ACTIVE) != 0 && 565 sc->sc_mount == mp) 566 fss_error(sc, "forced unmount"); 567 mutex_exit(&sc->sc_slock); 568 } 569 mutex_exit(&fss_device_lock); 570 } 571 572 /* 573 * A buffer is written to the snapshotted block device. Copy to 574 * backing store if needed. 575 */ 576 static int 577 fss_copy_on_write(void *v, struct buf *bp, bool data_valid) 578 { 579 int error; 580 u_int32_t cl, ch, c; 581 struct fss_softc *sc = v; 582 583 mutex_enter(&sc->sc_slock); 584 if (!FSS_ISVALID(sc)) { 585 mutex_exit(&sc->sc_slock); 586 return 0; 587 } 588 589 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 590 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 591 error = 0; 592 if (curlwp == uvm.pagedaemon_lwp) { 593 for (c = cl; c <= ch; c++) 594 if (isclr(sc->sc_copied, c)) { 595 error = ENOMEM; 596 break; 597 } 598 } 599 mutex_exit(&sc->sc_slock); 600 601 if (error == 0) 602 for (c = cl; c <= ch; c++) { 603 error = fss_read_cluster(sc, c); 604 if (error) 605 break; 606 } 607 608 return error; 609 } 610 611 /* 612 * Lookup and open needed files. 613 * 614 * For file system internal snapshot initializes sc_mntname, sc_mount, 615 * sc_bs_vp and sc_time. 616 * 617 * Otherwise returns dev and size of the underlying block device. 618 * Initializes sc_mntname, sc_mount, sc_bdev, sc_bs_vp and sc_mount 619 */ 620 static int 621 fss_create_files(struct fss_softc *sc, struct fss_set *fss, 622 off_t *bsize, struct lwp *l) 623 { 624 int error, bits, fsbsize; 625 uint64_t numsec; 626 unsigned int secsize; 627 struct timespec ts; 628 /* nd -> nd2 to reduce mistakes while updating only some namei calls */ 629 struct pathbuf *pb2; 630 struct nameidata nd2; 631 struct vnode *vp; 632 633 /* 634 * Get the mounted file system. 635 */ 636 637 error = namei_simple_user(fss->fss_mount, 638 NSM_FOLLOW_NOEMULROOT, &vp); 639 if (error != 0) 640 return error; 641 642 if ((vp->v_vflag & VV_ROOT) != VV_ROOT) { 643 vrele(vp); 644 return EINVAL; 645 } 646 647 sc->sc_mount = vp->v_mount; 648 memcpy(sc->sc_mntname, sc->sc_mount->mnt_stat.f_mntonname, MNAMELEN); 649 650 vrele(vp); 651 652 /* 653 * Check for file system internal snapshot. 654 */ 655 656 error = namei_simple_user(fss->fss_bstore, 657 NSM_FOLLOW_NOEMULROOT, &vp); 658 if (error != 0) 659 return error; 660 661 if (vp->v_type == VREG && vp->v_mount == sc->sc_mount) { 662 sc->sc_flags |= FSS_PERSISTENT; 663 sc->sc_bs_vp = vp; 664 665 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 666 bits = sizeof(sc->sc_bs_bshift)*NBBY; 667 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < bits; 668 sc->sc_bs_bshift++) 669 if (FSS_FSBSIZE(sc) == fsbsize) 670 break; 671 if (sc->sc_bs_bshift >= bits) 672 return EINVAL; 673 674 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 675 sc->sc_clshift = 0; 676 677 if ((fss->fss_flags & FSS_UNLINK_ON_CREATE) != 0) { 678 error = do_sys_unlink(fss->fss_bstore, UIO_USERSPACE); 679 if (error) 680 return error; 681 } 682 error = vn_lock(vp, LK_EXCLUSIVE); 683 if (error != 0) 684 return error; 685 error = VFS_SNAPSHOT(sc->sc_mount, sc->sc_bs_vp, &ts); 686 TIMESPEC_TO_TIMEVAL(&sc->sc_time, &ts); 687 688 VOP_UNLOCK(sc->sc_bs_vp); 689 690 return error; 691 } 692 vrele(vp); 693 694 /* 695 * Get the block device it is mounted on. 696 */ 697 698 error = namei_simple_kernel(sc->sc_mount->mnt_stat.f_mntfromname, 699 NSM_FOLLOW_NOEMULROOT, &vp); 700 if (error != 0) 701 return error; 702 703 if (vp->v_type != VBLK) { 704 vrele(vp); 705 return EINVAL; 706 } 707 708 sc->sc_bdev = vp->v_rdev; 709 710 /* 711 * Get the block device size. 712 */ 713 714 error = getdisksize(vp, &numsec, &secsize); 715 vrele(vp); 716 if (error) 717 return error; 718 719 *bsize = (off_t)numsec*secsize; 720 721 /* 722 * Get the backing store 723 */ 724 725 error = pathbuf_copyin(fss->fss_bstore, &pb2); 726 if (error) { 727 return error; 728 } 729 NDINIT(&nd2, LOOKUP, FOLLOW, pb2); 730 if ((error = vn_open(&nd2, FREAD|FWRITE, 0)) != 0) { 731 pathbuf_destroy(pb2); 732 return error; 733 } 734 VOP_UNLOCK(nd2.ni_vp); 735 736 sc->sc_bs_vp = nd2.ni_vp; 737 738 if (nd2.ni_vp->v_type != VREG && nd2.ni_vp->v_type != VCHR) { 739 pathbuf_destroy(pb2); 740 return EINVAL; 741 } 742 pathbuf_destroy(pb2); 743 744 if ((fss->fss_flags & FSS_UNLINK_ON_CREATE) != 0) { 745 error = do_sys_unlink(fss->fss_bstore, UIO_USERSPACE); 746 if (error) 747 return error; 748 } 749 if (sc->sc_bs_vp->v_type == VREG) { 750 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 751 if (fsbsize & (fsbsize-1)) /* No power of two */ 752 return EINVAL; 753 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < 32; 754 sc->sc_bs_bshift++) 755 if (FSS_FSBSIZE(sc) == fsbsize) 756 break; 757 if (sc->sc_bs_bshift >= 32) 758 return EINVAL; 759 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 760 } else { 761 sc->sc_bs_bshift = DEV_BSHIFT; 762 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 763 } 764 765 return 0; 766 } 767 768 /* 769 * Create a snapshot. 770 */ 771 static int 772 fss_create_snapshot(struct fss_softc *sc, struct fss_set *fss, struct lwp *l) 773 { 774 int len, error; 775 u_int32_t csize; 776 off_t bsize; 777 778 bsize = 0; /* XXX gcc */ 779 780 /* 781 * Open needed files. 782 */ 783 if ((error = fss_create_files(sc, fss, &bsize, l)) != 0) 784 goto bad; 785 786 if (sc->sc_flags & FSS_PERSISTENT) { 787 fss_softc_alloc(sc); 788 sc->sc_flags |= FSS_ACTIVE; 789 return 0; 790 } 791 792 /* 793 * Set cluster size. Must be a power of two and 794 * a multiple of backing store block size. 795 */ 796 if (fss->fss_csize <= 0) 797 csize = MAXPHYS; 798 else 799 csize = fss->fss_csize; 800 if (bsize/csize > FSS_CLUSTER_MAX) 801 csize = bsize/FSS_CLUSTER_MAX+1; 802 803 for (sc->sc_clshift = sc->sc_bs_bshift; sc->sc_clshift < 32; 804 sc->sc_clshift++) 805 if (FSS_CLSIZE(sc) >= csize) 806 break; 807 if (sc->sc_clshift >= 32) { 808 error = EINVAL; 809 goto bad; 810 } 811 sc->sc_clmask = FSS_CLSIZE(sc)-1; 812 813 /* 814 * Set number of cache slots. 815 */ 816 if (FSS_CLSIZE(sc) <= 8192) 817 sc->sc_cache_size = 32; 818 else if (FSS_CLSIZE(sc) <= 65536) 819 sc->sc_cache_size = 8; 820 else 821 sc->sc_cache_size = 4; 822 823 /* 824 * Set number of clusters and size of last cluster. 825 */ 826 sc->sc_clcount = FSS_BTOCL(sc, bsize-1)+1; 827 sc->sc_clresid = FSS_CLOFF(sc, bsize-1)+1; 828 829 /* 830 * Set size of indirect table. 831 */ 832 len = sc->sc_clcount*sizeof(u_int32_t); 833 sc->sc_indir_size = FSS_BTOCL(sc, len)+1; 834 sc->sc_clnext = sc->sc_indir_size; 835 sc->sc_indir_cur = 0; 836 837 if ((error = fss_softc_alloc(sc)) != 0) 838 goto bad; 839 840 /* 841 * Activate the snapshot. 842 */ 843 844 if ((error = vfs_suspend(sc->sc_mount, 0)) != 0) 845 goto bad; 846 847 microtime(&sc->sc_time); 848 849 error = fscow_establish(sc->sc_mount, fss_copy_on_write, sc); 850 if (error == 0) 851 sc->sc_flags |= FSS_ACTIVE; 852 853 vfs_resume(sc->sc_mount); 854 855 if (error != 0) 856 goto bad; 857 858 aprint_debug_dev(sc->sc_dev, "%s snapshot active\n", sc->sc_mntname); 859 aprint_debug_dev(sc->sc_dev, 860 "%u clusters of %u, %u cache slots, %u indir clusters\n", 861 sc->sc_clcount, FSS_CLSIZE(sc), 862 sc->sc_cache_size, sc->sc_indir_size); 863 864 return 0; 865 866 bad: 867 fss_softc_free(sc); 868 if (sc->sc_bs_vp != NULL) { 869 if (sc->sc_flags & FSS_PERSISTENT) 870 vrele(sc->sc_bs_vp); 871 else 872 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 873 } 874 sc->sc_bs_vp = NULL; 875 876 return error; 877 } 878 879 /* 880 * Delete a snapshot. 881 */ 882 static int 883 fss_delete_snapshot(struct fss_softc *sc, struct lwp *l) 884 { 885 886 if ((sc->sc_flags & FSS_PERSISTENT) == 0) 887 fscow_disestablish(sc->sc_mount, fss_copy_on_write, sc); 888 889 mutex_enter(&sc->sc_slock); 890 sc->sc_flags &= ~(FSS_ACTIVE|FSS_ERROR); 891 sc->sc_mount = NULL; 892 sc->sc_bdev = NODEV; 893 mutex_exit(&sc->sc_slock); 894 895 fss_softc_free(sc); 896 if (sc->sc_flags & FSS_PERSISTENT) 897 vrele(sc->sc_bs_vp); 898 else 899 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 900 sc->sc_bs_vp = NULL; 901 sc->sc_flags &= ~FSS_PERSISTENT; 902 903 return 0; 904 } 905 906 /* 907 * Read a cluster from the snapshotted block device to the cache. 908 */ 909 static int 910 fss_read_cluster(struct fss_softc *sc, u_int32_t cl) 911 { 912 int error, todo, offset, len; 913 daddr_t dblk; 914 struct buf *bp, *mbp; 915 struct fss_cache *scp, *scl; 916 917 /* 918 * Get a free cache slot. 919 */ 920 scl = sc->sc_cache+sc->sc_cache_size; 921 922 mutex_enter(&sc->sc_slock); 923 924 restart: 925 if (isset(sc->sc_copied, cl) || !FSS_ISVALID(sc)) { 926 mutex_exit(&sc->sc_slock); 927 return 0; 928 } 929 930 for (scp = sc->sc_cache; scp < scl; scp++) 931 if (scp->fc_cluster == cl) { 932 if (scp->fc_type == FSS_CACHE_VALID) { 933 mutex_exit(&sc->sc_slock); 934 return 0; 935 } else if (scp->fc_type == FSS_CACHE_BUSY) { 936 cv_wait(&scp->fc_state_cv, &sc->sc_slock); 937 goto restart; 938 } 939 } 940 941 for (scp = sc->sc_cache; scp < scl; scp++) 942 if (scp->fc_type == FSS_CACHE_FREE) { 943 scp->fc_type = FSS_CACHE_BUSY; 944 scp->fc_cluster = cl; 945 break; 946 } 947 if (scp >= scl) { 948 cv_wait(&sc->sc_cache_cv, &sc->sc_slock); 949 goto restart; 950 } 951 952 mutex_exit(&sc->sc_slock); 953 954 /* 955 * Start the read. 956 */ 957 dblk = btodb(FSS_CLTOB(sc, cl)); 958 if (cl == sc->sc_clcount-1) { 959 todo = sc->sc_clresid; 960 memset((char *)scp->fc_data + todo, 0, FSS_CLSIZE(sc) - todo); 961 } else 962 todo = FSS_CLSIZE(sc); 963 offset = 0; 964 mbp = getiobuf(NULL, true); 965 mbp->b_bufsize = todo; 966 mbp->b_data = scp->fc_data; 967 mbp->b_resid = mbp->b_bcount = todo; 968 mbp->b_flags = B_READ; 969 mbp->b_cflags = BC_BUSY; 970 mbp->b_dev = sc->sc_bdev; 971 while (todo > 0) { 972 len = todo; 973 if (len > MAXPHYS) 974 len = MAXPHYS; 975 if (btodb(FSS_CLTOB(sc, cl)) == dblk && len == todo) 976 bp = mbp; 977 else { 978 bp = getiobuf(NULL, true); 979 nestiobuf_setup(mbp, bp, offset, len); 980 } 981 bp->b_lblkno = 0; 982 bp->b_blkno = dblk; 983 bdev_strategy(bp); 984 dblk += btodb(len); 985 offset += len; 986 todo -= len; 987 } 988 error = biowait(mbp); 989 putiobuf(mbp); 990 991 mutex_enter(&sc->sc_slock); 992 scp->fc_type = (error ? FSS_CACHE_FREE : FSS_CACHE_VALID); 993 cv_broadcast(&scp->fc_state_cv); 994 if (error == 0) { 995 setbit(sc->sc_copied, scp->fc_cluster); 996 cv_signal(&sc->sc_work_cv); 997 } 998 mutex_exit(&sc->sc_slock); 999 1000 return error; 1001 } 1002 1003 /* 1004 * Read/write clusters from/to backing store. 1005 * For persistent snapshots must be called with cl == 0. off is the 1006 * offset into the snapshot. 1007 */ 1008 static int 1009 fss_bs_io(struct fss_softc *sc, fss_io_type rw, 1010 u_int32_t cl, off_t off, int len, void *data) 1011 { 1012 int error; 1013 1014 off += FSS_CLTOB(sc, cl); 1015 1016 vn_lock(sc->sc_bs_vp, LK_EXCLUSIVE|LK_RETRY); 1017 1018 error = vn_rdwr((rw == FSS_READ ? UIO_READ : UIO_WRITE), sc->sc_bs_vp, 1019 data, len, off, UIO_SYSSPACE, 1020 IO_ADV_ENCODE(POSIX_FADV_NOREUSE) | IO_NODELOCKED, 1021 sc->sc_bs_lwp->l_cred, NULL, NULL); 1022 if (error == 0) { 1023 mutex_enter(sc->sc_bs_vp->v_interlock); 1024 error = VOP_PUTPAGES(sc->sc_bs_vp, trunc_page(off), 1025 round_page(off+len), PGO_CLEANIT | PGO_FREE | PGO_SYNCIO); 1026 } 1027 1028 VOP_UNLOCK(sc->sc_bs_vp); 1029 1030 return error; 1031 } 1032 1033 /* 1034 * Get a pointer to the indirect slot for this cluster. 1035 */ 1036 static u_int32_t * 1037 fss_bs_indir(struct fss_softc *sc, u_int32_t cl) 1038 { 1039 u_int32_t icl; 1040 int ioff; 1041 1042 icl = cl/(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 1043 ioff = cl%(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 1044 1045 if (sc->sc_indir_cur == icl) 1046 return &sc->sc_indir_data[ioff]; 1047 1048 if (sc->sc_indir_dirty) { 1049 if (fss_bs_io(sc, FSS_WRITE, sc->sc_indir_cur, 0, 1050 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 1051 return NULL; 1052 setbit(sc->sc_indir_valid, sc->sc_indir_cur); 1053 } 1054 1055 sc->sc_indir_dirty = 0; 1056 sc->sc_indir_cur = icl; 1057 1058 if (isset(sc->sc_indir_valid, sc->sc_indir_cur)) { 1059 if (fss_bs_io(sc, FSS_READ, sc->sc_indir_cur, 0, 1060 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 1061 return NULL; 1062 } else 1063 memset(sc->sc_indir_data, 0, FSS_CLSIZE(sc)); 1064 1065 return &sc->sc_indir_data[ioff]; 1066 } 1067 1068 /* 1069 * The kernel thread (one for every active snapshot). 1070 * 1071 * After wakeup it cleans the cache and runs the I/O requests. 1072 */ 1073 static void 1074 fss_bs_thread(void *arg) 1075 { 1076 bool thread_idle, is_valid; 1077 int error, i, todo, len, crotor, is_read; 1078 long off; 1079 char *addr; 1080 u_int32_t c, cl, ch, *indirp; 1081 struct buf *bp, *nbp; 1082 struct fss_softc *sc; 1083 struct fss_cache *scp, *scl; 1084 1085 sc = arg; 1086 scl = sc->sc_cache+sc->sc_cache_size; 1087 crotor = 0; 1088 thread_idle = false; 1089 1090 mutex_enter(&sc->sc_slock); 1091 1092 for (;;) { 1093 if (thread_idle) 1094 cv_wait(&sc->sc_work_cv, &sc->sc_slock); 1095 thread_idle = true; 1096 if ((sc->sc_flags & FSS_BS_THREAD) == 0) { 1097 mutex_exit(&sc->sc_slock); 1098 kthread_exit(0); 1099 } 1100 1101 /* 1102 * Process I/O requests (persistent) 1103 */ 1104 1105 if (sc->sc_flags & FSS_PERSISTENT) { 1106 if ((bp = bufq_get(sc->sc_bufq)) == NULL) 1107 continue; 1108 is_valid = FSS_ISVALID(sc); 1109 is_read = (bp->b_flags & B_READ); 1110 thread_idle = false; 1111 mutex_exit(&sc->sc_slock); 1112 1113 if (is_valid) { 1114 disk_busy(sc->sc_dkdev); 1115 error = fss_bs_io(sc, FSS_READ, 0, 1116 dbtob(bp->b_blkno), bp->b_bcount, 1117 bp->b_data); 1118 disk_unbusy(sc->sc_dkdev, 1119 (error ? 0 : bp->b_bcount), is_read); 1120 } else 1121 error = ENXIO; 1122 1123 bp->b_error = error; 1124 bp->b_resid = (error ? bp->b_bcount : 0); 1125 biodone(bp); 1126 1127 mutex_enter(&sc->sc_slock); 1128 continue; 1129 } 1130 1131 /* 1132 * Clean the cache 1133 */ 1134 for (i = 0; i < sc->sc_cache_size; i++) { 1135 crotor = (crotor + 1) % sc->sc_cache_size; 1136 scp = sc->sc_cache + crotor; 1137 if (scp->fc_type != FSS_CACHE_VALID) 1138 continue; 1139 mutex_exit(&sc->sc_slock); 1140 1141 thread_idle = false; 1142 indirp = fss_bs_indir(sc, scp->fc_cluster); 1143 if (indirp != NULL) { 1144 error = fss_bs_io(sc, FSS_WRITE, sc->sc_clnext, 1145 0, FSS_CLSIZE(sc), scp->fc_data); 1146 } else 1147 error = EIO; 1148 1149 mutex_enter(&sc->sc_slock); 1150 if (error == 0) { 1151 *indirp = sc->sc_clnext++; 1152 sc->sc_indir_dirty = 1; 1153 } else 1154 fss_error(sc, "write error on backing store"); 1155 1156 scp->fc_type = FSS_CACHE_FREE; 1157 cv_signal(&sc->sc_cache_cv); 1158 break; 1159 } 1160 1161 /* 1162 * Process I/O requests 1163 */ 1164 if ((bp = bufq_get(sc->sc_bufq)) == NULL) 1165 continue; 1166 is_valid = FSS_ISVALID(sc); 1167 is_read = (bp->b_flags & B_READ); 1168 thread_idle = false; 1169 1170 if (!is_valid) { 1171 mutex_exit(&sc->sc_slock); 1172 1173 bp->b_error = ENXIO; 1174 bp->b_resid = bp->b_bcount; 1175 biodone(bp); 1176 1177 mutex_enter(&sc->sc_slock); 1178 continue; 1179 } 1180 1181 disk_busy(sc->sc_dkdev); 1182 1183 /* 1184 * First read from the snapshotted block device unless 1185 * this request is completely covered by backing store. 1186 */ 1187 1188 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 1189 off = FSS_CLOFF(sc, dbtob(bp->b_blkno)); 1190 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 1191 error = 0; 1192 bp->b_resid = 0; 1193 bp->b_error = 0; 1194 for (c = cl; c <= ch; c++) { 1195 if (isset(sc->sc_copied, c)) 1196 continue; 1197 mutex_exit(&sc->sc_slock); 1198 1199 /* Not on backing store, read from device. */ 1200 nbp = getiobuf(NULL, true); 1201 nbp->b_flags = B_READ; 1202 nbp->b_resid = nbp->b_bcount = bp->b_bcount; 1203 nbp->b_bufsize = bp->b_bcount; 1204 nbp->b_data = bp->b_data; 1205 nbp->b_blkno = bp->b_blkno; 1206 nbp->b_lblkno = 0; 1207 nbp->b_dev = sc->sc_bdev; 1208 SET(nbp->b_cflags, BC_BUSY); /* mark buffer busy */ 1209 1210 bdev_strategy(nbp); 1211 1212 error = biowait(nbp); 1213 if (error != 0) { 1214 bp->b_resid = bp->b_bcount; 1215 bp->b_error = nbp->b_error; 1216 disk_unbusy(sc->sc_dkdev, 0, is_read); 1217 biodone(bp); 1218 } 1219 putiobuf(nbp); 1220 1221 mutex_enter(&sc->sc_slock); 1222 break; 1223 } 1224 if (error) 1225 continue; 1226 1227 /* 1228 * Replace those parts that have been saved to backing store. 1229 */ 1230 1231 addr = bp->b_data; 1232 todo = bp->b_bcount; 1233 for (c = cl; c <= ch; c++, off = 0, todo -= len, addr += len) { 1234 len = FSS_CLSIZE(sc)-off; 1235 if (len > todo) 1236 len = todo; 1237 if (isclr(sc->sc_copied, c)) 1238 continue; 1239 mutex_exit(&sc->sc_slock); 1240 1241 indirp = fss_bs_indir(sc, c); 1242 if (indirp == NULL || *indirp == 0) { 1243 /* 1244 * Not on backing store. Either in cache 1245 * or hole in the snapshotted block device. 1246 */ 1247 1248 mutex_enter(&sc->sc_slock); 1249 for (scp = sc->sc_cache; scp < scl; scp++) 1250 if (scp->fc_type == FSS_CACHE_VALID && 1251 scp->fc_cluster == c) 1252 break; 1253 if (scp < scl) 1254 memcpy(addr, (char *)scp->fc_data+off, 1255 len); 1256 else 1257 memset(addr, 0, len); 1258 continue; 1259 } 1260 1261 /* 1262 * Read from backing store. 1263 */ 1264 error = 1265 fss_bs_io(sc, FSS_READ, *indirp, off, len, addr); 1266 1267 mutex_enter(&sc->sc_slock); 1268 if (error) { 1269 bp->b_resid = bp->b_bcount; 1270 bp->b_error = error; 1271 break; 1272 } 1273 } 1274 mutex_exit(&sc->sc_slock); 1275 1276 disk_unbusy(sc->sc_dkdev, (error ? 0 : bp->b_bcount), is_read); 1277 biodone(bp); 1278 1279 mutex_enter(&sc->sc_slock); 1280 } 1281 } 1282 1283 #ifdef _MODULE 1284 1285 #include <sys/module.h> 1286 1287 MODULE(MODULE_CLASS_DRIVER, fss, NULL); 1288 CFDRIVER_DECL(fss, DV_DISK, NULL); 1289 1290 static int 1291 fss_modcmd(modcmd_t cmd, void *arg) 1292 { 1293 int bmajor = -1, cmajor = -1, error = 0; 1294 1295 switch (cmd) { 1296 case MODULE_CMD_INIT: 1297 mutex_init(&fss_device_lock, MUTEX_DEFAULT, IPL_NONE); 1298 error = config_cfdriver_attach(&fss_cd); 1299 if (error) { 1300 mutex_destroy(&fss_device_lock); 1301 break; 1302 } 1303 error = config_cfattach_attach(fss_cd.cd_name, &fss_ca); 1304 if (error) { 1305 config_cfdriver_detach(&fss_cd); 1306 mutex_destroy(&fss_device_lock); 1307 break; 1308 } 1309 error = devsw_attach(fss_cd.cd_name, 1310 &fss_bdevsw, &bmajor, &fss_cdevsw, &cmajor); 1311 if (error == EEXIST) 1312 error = 0; 1313 if (error) { 1314 config_cfattach_detach(fss_cd.cd_name, &fss_ca); 1315 config_cfdriver_detach(&fss_cd); 1316 mutex_destroy(&fss_device_lock); 1317 break; 1318 } 1319 break; 1320 1321 case MODULE_CMD_FINI: 1322 error = config_cfattach_detach(fss_cd.cd_name, &fss_ca); 1323 if (error) 1324 break; 1325 config_cfdriver_detach(&fss_cd); 1326 devsw_detach(&fss_bdevsw, &fss_cdevsw); 1327 mutex_destroy(&fss_device_lock); 1328 break; 1329 1330 default: 1331 error = ENOTTY; 1332 break; 1333 } 1334 1335 return error; 1336 } 1337 1338 #endif /* _MODULE */ 1339