1 /* $NetBSD: fss.c,v 1.86 2013/02/13 14:03:48 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 1.86 2013/02/13 14:03:48 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 mutex_enter(&fss_device_lock); 227 restart: 228 mutex_enter(&sc->sc_slock); 229 if ((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) != mflag) { 230 sc->sc_flags &= ~mflag; 231 mutex_exit(&sc->sc_slock); 232 mutex_exit(&fss_device_lock); 233 return 0; 234 } 235 if ((sc->sc_flags & FSS_ACTIVE) != 0 && 236 (sc->sc_uflags & FSS_UNCONFIG_ON_CLOSE) != 0) { 237 sc->sc_uflags &= ~FSS_UNCONFIG_ON_CLOSE; 238 mutex_exit(&sc->sc_slock); 239 error = fss_ioctl(dev, FSSIOCCLR, NULL, FWRITE, l); 240 goto restart; 241 } 242 if ((sc->sc_flags & FSS_ACTIVE) != 0) { 243 mutex_exit(&sc->sc_slock); 244 mutex_exit(&fss_device_lock); 245 return error; 246 } 247 248 KASSERT((sc->sc_flags & FSS_ACTIVE) == 0); 249 KASSERT((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) == mflag); 250 mutex_exit(&sc->sc_slock); 251 cf = device_cfdata(sc->sc_dev); 252 error = config_detach(sc->sc_dev, DETACH_QUIET); 253 if (! error) 254 free(cf, M_DEVBUF); 255 mutex_exit(&fss_device_lock); 256 257 return error; 258 } 259 260 void 261 fss_strategy(struct buf *bp) 262 { 263 const bool write = ((bp->b_flags & B_READ) != B_READ); 264 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(bp->b_dev)); 265 266 mutex_enter(&sc->sc_slock); 267 268 if (write || !FSS_ISVALID(sc)) { 269 270 mutex_exit(&sc->sc_slock); 271 272 bp->b_error = (write ? EROFS : ENXIO); 273 bp->b_resid = bp->b_bcount; 274 biodone(bp); 275 return; 276 } 277 278 bp->b_rawblkno = bp->b_blkno; 279 bufq_put(sc->sc_bufq, bp); 280 cv_signal(&sc->sc_work_cv); 281 282 mutex_exit(&sc->sc_slock); 283 } 284 285 int 286 fss_read(dev_t dev, struct uio *uio, int flags) 287 { 288 return physio(fss_strategy, NULL, dev, B_READ, minphys, uio); 289 } 290 291 int 292 fss_write(dev_t dev, struct uio *uio, int flags) 293 { 294 return physio(fss_strategy, NULL, dev, B_WRITE, minphys, uio); 295 } 296 297 int 298 fss_ioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 299 { 300 int error; 301 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(dev)); 302 struct fss_set _fss; 303 struct fss_set *fss = (struct fss_set *)data; 304 struct fss_set50 *fss50 = (struct fss_set50 *)data; 305 struct fss_get *fsg = (struct fss_get *)data; 306 #ifndef _LP64 307 struct fss_get50 *fsg50 = (struct fss_get50 *)data; 308 #endif 309 310 switch (cmd) { 311 case FSSIOCSET50: 312 fss = &_fss; 313 fss->fss_mount = fss50->fss_mount; 314 fss->fss_bstore = fss50->fss_bstore; 315 fss->fss_csize = fss50->fss_csize; 316 fss->fss_flags = 0; 317 /* Fall through */ 318 case FSSIOCSET: 319 mutex_enter(&sc->sc_lock); 320 if ((flag & FWRITE) == 0) 321 error = EPERM; 322 else if ((sc->sc_flags & FSS_ACTIVE) != 0) 323 error = EBUSY; 324 else 325 error = fss_create_snapshot(sc, fss, l); 326 if (error == 0) 327 sc->sc_uflags = fss->fss_flags; 328 mutex_exit(&sc->sc_lock); 329 break; 330 331 case FSSIOCCLR: 332 mutex_enter(&sc->sc_lock); 333 if ((flag & FWRITE) == 0) 334 error = EPERM; 335 else if ((sc->sc_flags & FSS_ACTIVE) == 0) 336 error = ENXIO; 337 else 338 error = fss_delete_snapshot(sc, l); 339 mutex_exit(&sc->sc_lock); 340 break; 341 342 #ifndef _LP64 343 case FSSIOCGET50: 344 mutex_enter(&sc->sc_lock); 345 switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) { 346 case FSS_ACTIVE: 347 memcpy(fsg50->fsg_mount, sc->sc_mntname, MNAMELEN); 348 fsg50->fsg_csize = FSS_CLSIZE(sc); 349 timeval_to_timeval50(&sc->sc_time, &fsg50->fsg_time); 350 fsg50->fsg_mount_size = sc->sc_clcount; 351 fsg50->fsg_bs_size = sc->sc_clnext; 352 error = 0; 353 break; 354 case FSS_PERSISTENT | FSS_ACTIVE: 355 memcpy(fsg50->fsg_mount, sc->sc_mntname, MNAMELEN); 356 fsg50->fsg_csize = 0; 357 timeval_to_timeval50(&sc->sc_time, &fsg50->fsg_time); 358 fsg50->fsg_mount_size = 0; 359 fsg50->fsg_bs_size = 0; 360 error = 0; 361 break; 362 default: 363 error = ENXIO; 364 break; 365 } 366 mutex_exit(&sc->sc_lock); 367 break; 368 #endif /* _LP64 */ 369 370 case FSSIOCGET: 371 mutex_enter(&sc->sc_lock); 372 switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) { 373 case FSS_ACTIVE: 374 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 375 fsg->fsg_csize = FSS_CLSIZE(sc); 376 fsg->fsg_time = sc->sc_time; 377 fsg->fsg_mount_size = sc->sc_clcount; 378 fsg->fsg_bs_size = sc->sc_clnext; 379 error = 0; 380 break; 381 case FSS_PERSISTENT | FSS_ACTIVE: 382 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 383 fsg->fsg_csize = 0; 384 fsg->fsg_time = sc->sc_time; 385 fsg->fsg_mount_size = 0; 386 fsg->fsg_bs_size = 0; 387 error = 0; 388 break; 389 default: 390 error = ENXIO; 391 break; 392 } 393 mutex_exit(&sc->sc_lock); 394 break; 395 396 case FSSIOFSET: 397 mutex_enter(&sc->sc_slock); 398 sc->sc_uflags = *(int *)data; 399 mutex_exit(&sc->sc_slock); 400 error = 0; 401 break; 402 403 case FSSIOFGET: 404 mutex_enter(&sc->sc_slock); 405 *(int *)data = sc->sc_uflags; 406 mutex_exit(&sc->sc_slock); 407 error = 0; 408 break; 409 410 default: 411 error = EINVAL; 412 break; 413 } 414 415 return error; 416 } 417 418 int 419 fss_size(dev_t dev) 420 { 421 return -1; 422 } 423 424 int 425 fss_dump(dev_t dev, daddr_t blkno, void *va, 426 size_t size) 427 { 428 return EROFS; 429 } 430 431 /* 432 * An error occurred reading or writing the snapshot or backing store. 433 * If it is the first error log to console. 434 * The caller holds the mutex. 435 */ 436 static inline void 437 fss_error(struct fss_softc *sc, const char *msg) 438 { 439 440 if ((sc->sc_flags & (FSS_ACTIVE|FSS_ERROR)) == FSS_ACTIVE) 441 aprint_error_dev(sc->sc_dev, "snapshot invalid: %s\n", msg); 442 if ((sc->sc_flags & FSS_ACTIVE) == FSS_ACTIVE) 443 sc->sc_flags |= FSS_ERROR; 444 } 445 446 /* 447 * Allocate the variable sized parts of the softc and 448 * fork the kernel thread. 449 * 450 * The fields sc_clcount, sc_clshift, sc_cache_size and sc_indir_size 451 * must be initialized. 452 */ 453 static int 454 fss_softc_alloc(struct fss_softc *sc) 455 { 456 int i, error; 457 458 if ((sc->sc_flags & FSS_PERSISTENT) == 0) { 459 sc->sc_copied = 460 kmem_zalloc(howmany(sc->sc_clcount, NBBY), KM_SLEEP); 461 if (sc->sc_copied == NULL) 462 return(ENOMEM); 463 464 sc->sc_cache = kmem_alloc(sc->sc_cache_size * 465 sizeof(struct fss_cache), KM_SLEEP); 466 if (sc->sc_cache == NULL) 467 return(ENOMEM); 468 469 for (i = 0; i < sc->sc_cache_size; i++) { 470 sc->sc_cache[i].fc_type = FSS_CACHE_FREE; 471 sc->sc_cache[i].fc_data = 472 kmem_alloc(FSS_CLSIZE(sc), KM_SLEEP); 473 if (sc->sc_cache[i].fc_data == NULL) 474 return(ENOMEM); 475 cv_init(&sc->sc_cache[i].fc_state_cv, "cowwait1"); 476 } 477 478 sc->sc_indir_valid = 479 kmem_zalloc(howmany(sc->sc_indir_size, NBBY), KM_SLEEP); 480 if (sc->sc_indir_valid == NULL) 481 return(ENOMEM); 482 483 sc->sc_indir_data = kmem_zalloc(FSS_CLSIZE(sc), KM_SLEEP); 484 if (sc->sc_indir_data == NULL) 485 return(ENOMEM); 486 } else { 487 sc->sc_copied = NULL; 488 sc->sc_cache = NULL; 489 sc->sc_indir_valid = NULL; 490 sc->sc_indir_data = NULL; 491 } 492 493 sc->sc_flags |= FSS_BS_THREAD; 494 if ((error = kthread_create(PRI_BIO, KTHREAD_MUSTJOIN, NULL, 495 fss_bs_thread, sc, &sc->sc_bs_lwp, 496 "%s", device_xname(sc->sc_dev))) != 0) { 497 sc->sc_flags &= ~FSS_BS_THREAD; 498 return error; 499 } 500 501 disk_attach(sc->sc_dkdev); 502 503 return 0; 504 } 505 506 /* 507 * Free the variable sized parts of the softc. 508 */ 509 static void 510 fss_softc_free(struct fss_softc *sc) 511 { 512 int i; 513 514 if ((sc->sc_flags & FSS_BS_THREAD) != 0) { 515 mutex_enter(&sc->sc_slock); 516 sc->sc_flags &= ~FSS_BS_THREAD; 517 cv_signal(&sc->sc_work_cv); 518 mutex_exit(&sc->sc_slock); 519 kthread_join(sc->sc_bs_lwp); 520 521 disk_detach(sc->sc_dkdev); 522 } 523 524 if (sc->sc_copied != NULL) 525 kmem_free(sc->sc_copied, howmany(sc->sc_clcount, NBBY)); 526 sc->sc_copied = NULL; 527 528 if (sc->sc_cache != NULL) { 529 for (i = 0; i < sc->sc_cache_size; i++) 530 if (sc->sc_cache[i].fc_data != NULL) { 531 cv_destroy(&sc->sc_cache[i].fc_state_cv); 532 kmem_free(sc->sc_cache[i].fc_data, 533 FSS_CLSIZE(sc)); 534 } 535 kmem_free(sc->sc_cache, 536 sc->sc_cache_size*sizeof(struct fss_cache)); 537 } 538 sc->sc_cache = NULL; 539 540 if (sc->sc_indir_valid != NULL) 541 kmem_free(sc->sc_indir_valid, howmany(sc->sc_indir_size, NBBY)); 542 sc->sc_indir_valid = NULL; 543 544 if (sc->sc_indir_data != NULL) 545 kmem_free(sc->sc_indir_data, FSS_CLSIZE(sc)); 546 sc->sc_indir_data = NULL; 547 } 548 549 /* 550 * Set all active snapshots on this file system into ERROR state. 551 */ 552 static void 553 fss_unmount_hook(struct mount *mp) 554 { 555 int i; 556 struct fss_softc *sc; 557 558 mutex_enter(&fss_device_lock); 559 for (i = 0; i < fss_cd.cd_ndevs; i++) { 560 if ((sc = device_lookup_private(&fss_cd, i)) == NULL) 561 continue; 562 mutex_enter(&sc->sc_slock); 563 if ((sc->sc_flags & FSS_ACTIVE) != 0 && 564 sc->sc_mount == mp) 565 fss_error(sc, "forced unmount"); 566 mutex_exit(&sc->sc_slock); 567 } 568 mutex_exit(&fss_device_lock); 569 } 570 571 /* 572 * A buffer is written to the snapshotted block device. Copy to 573 * backing store if needed. 574 */ 575 static int 576 fss_copy_on_write(void *v, struct buf *bp, bool data_valid) 577 { 578 int error; 579 u_int32_t cl, ch, c; 580 struct fss_softc *sc = v; 581 582 mutex_enter(&sc->sc_slock); 583 if (!FSS_ISVALID(sc)) { 584 mutex_exit(&sc->sc_slock); 585 return 0; 586 } 587 588 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 589 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 590 error = 0; 591 if (curlwp == uvm.pagedaemon_lwp) { 592 for (c = cl; c <= ch; c++) 593 if (isclr(sc->sc_copied, c)) { 594 error = ENOMEM; 595 break; 596 } 597 } 598 mutex_exit(&sc->sc_slock); 599 600 if (error == 0) 601 for (c = cl; c <= ch; c++) { 602 error = fss_read_cluster(sc, c); 603 if (error) 604 break; 605 } 606 607 return error; 608 } 609 610 /* 611 * Lookup and open needed files. 612 * 613 * For file system internal snapshot initializes sc_mntname, sc_mount, 614 * sc_bs_vp and sc_time. 615 * 616 * Otherwise returns dev and size of the underlying block device. 617 * Initializes sc_mntname, sc_mount, sc_bdev, sc_bs_vp and sc_mount 618 */ 619 static int 620 fss_create_files(struct fss_softc *sc, struct fss_set *fss, 621 off_t *bsize, struct lwp *l) 622 { 623 int error, bits, fsbsize; 624 uint64_t numsec; 625 unsigned int secsize; 626 struct timespec ts; 627 /* nd -> nd2 to reduce mistakes while updating only some namei calls */ 628 struct pathbuf *pb2; 629 struct nameidata nd2; 630 struct vnode *vp; 631 632 /* 633 * Get the mounted file system. 634 */ 635 636 error = namei_simple_user(fss->fss_mount, 637 NSM_FOLLOW_NOEMULROOT, &vp); 638 if (error != 0) 639 return error; 640 641 if ((vp->v_vflag & VV_ROOT) != VV_ROOT) { 642 vrele(vp); 643 return EINVAL; 644 } 645 646 sc->sc_mount = vp->v_mount; 647 memcpy(sc->sc_mntname, sc->sc_mount->mnt_stat.f_mntonname, MNAMELEN); 648 649 vrele(vp); 650 651 /* 652 * Check for file system internal snapshot. 653 */ 654 655 error = namei_simple_user(fss->fss_bstore, 656 NSM_FOLLOW_NOEMULROOT, &vp); 657 if (error != 0) 658 return error; 659 660 if (vp->v_type == VREG && vp->v_mount == sc->sc_mount) { 661 sc->sc_flags |= FSS_PERSISTENT; 662 sc->sc_bs_vp = vp; 663 664 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 665 bits = sizeof(sc->sc_bs_bshift)*NBBY; 666 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < bits; 667 sc->sc_bs_bshift++) 668 if (FSS_FSBSIZE(sc) == fsbsize) 669 break; 670 if (sc->sc_bs_bshift >= bits) 671 return EINVAL; 672 673 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 674 sc->sc_clshift = 0; 675 676 if ((fss->fss_flags & FSS_UNLINK_ON_CREATE) != 0) { 677 error = do_sys_unlink(fss->fss_bstore, UIO_USERSPACE); 678 if (error) 679 return error; 680 } 681 error = vn_lock(vp, LK_EXCLUSIVE); 682 if (error != 0) 683 return error; 684 error = VFS_SNAPSHOT(sc->sc_mount, sc->sc_bs_vp, &ts); 685 TIMESPEC_TO_TIMEVAL(&sc->sc_time, &ts); 686 687 VOP_UNLOCK(sc->sc_bs_vp); 688 689 return error; 690 } 691 vrele(vp); 692 693 /* 694 * Get the block device it is mounted on and its size. 695 */ 696 697 error = spec_node_lookup_by_mount(sc->sc_mount, &vp); 698 if (error) 699 return error; 700 sc->sc_bdev = vp->v_rdev; 701 702 error = getdisksize(vp, &numsec, &secsize); 703 vrele(vp); 704 if (error) 705 return error; 706 707 *bsize = (off_t)numsec*secsize; 708 709 /* 710 * Get the backing store 711 */ 712 713 error = pathbuf_copyin(fss->fss_bstore, &pb2); 714 if (error) { 715 return error; 716 } 717 NDINIT(&nd2, LOOKUP, FOLLOW, pb2); 718 if ((error = vn_open(&nd2, FREAD|FWRITE, 0)) != 0) { 719 pathbuf_destroy(pb2); 720 return error; 721 } 722 VOP_UNLOCK(nd2.ni_vp); 723 724 sc->sc_bs_vp = nd2.ni_vp; 725 726 if (nd2.ni_vp->v_type != VREG && nd2.ni_vp->v_type != VCHR) { 727 pathbuf_destroy(pb2); 728 return EINVAL; 729 } 730 pathbuf_destroy(pb2); 731 732 if ((fss->fss_flags & FSS_UNLINK_ON_CREATE) != 0) { 733 error = do_sys_unlink(fss->fss_bstore, UIO_USERSPACE); 734 if (error) 735 return error; 736 } 737 if (sc->sc_bs_vp->v_type == VREG) { 738 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 739 if (fsbsize & (fsbsize-1)) /* No power of two */ 740 return EINVAL; 741 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < 32; 742 sc->sc_bs_bshift++) 743 if (FSS_FSBSIZE(sc) == fsbsize) 744 break; 745 if (sc->sc_bs_bshift >= 32) 746 return EINVAL; 747 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 748 } else { 749 sc->sc_bs_bshift = DEV_BSHIFT; 750 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 751 } 752 753 return 0; 754 } 755 756 /* 757 * Create a snapshot. 758 */ 759 static int 760 fss_create_snapshot(struct fss_softc *sc, struct fss_set *fss, struct lwp *l) 761 { 762 int len, error; 763 u_int32_t csize; 764 off_t bsize; 765 766 bsize = 0; /* XXX gcc */ 767 768 /* 769 * Open needed files. 770 */ 771 if ((error = fss_create_files(sc, fss, &bsize, l)) != 0) 772 goto bad; 773 774 if (sc->sc_flags & FSS_PERSISTENT) { 775 fss_softc_alloc(sc); 776 sc->sc_flags |= FSS_ACTIVE; 777 return 0; 778 } 779 780 /* 781 * Set cluster size. Must be a power of two and 782 * a multiple of backing store block size. 783 */ 784 if (fss->fss_csize <= 0) 785 csize = MAXPHYS; 786 else 787 csize = fss->fss_csize; 788 if (bsize/csize > FSS_CLUSTER_MAX) 789 csize = bsize/FSS_CLUSTER_MAX+1; 790 791 for (sc->sc_clshift = sc->sc_bs_bshift; sc->sc_clshift < 32; 792 sc->sc_clshift++) 793 if (FSS_CLSIZE(sc) >= csize) 794 break; 795 if (sc->sc_clshift >= 32) { 796 error = EINVAL; 797 goto bad; 798 } 799 sc->sc_clmask = FSS_CLSIZE(sc)-1; 800 801 /* 802 * Set number of cache slots. 803 */ 804 if (FSS_CLSIZE(sc) <= 8192) 805 sc->sc_cache_size = 32; 806 else if (FSS_CLSIZE(sc) <= 65536) 807 sc->sc_cache_size = 8; 808 else 809 sc->sc_cache_size = 4; 810 811 /* 812 * Set number of clusters and size of last cluster. 813 */ 814 sc->sc_clcount = FSS_BTOCL(sc, bsize-1)+1; 815 sc->sc_clresid = FSS_CLOFF(sc, bsize-1)+1; 816 817 /* 818 * Set size of indirect table. 819 */ 820 len = sc->sc_clcount*sizeof(u_int32_t); 821 sc->sc_indir_size = FSS_BTOCL(sc, len)+1; 822 sc->sc_clnext = sc->sc_indir_size; 823 sc->sc_indir_cur = 0; 824 825 if ((error = fss_softc_alloc(sc)) != 0) 826 goto bad; 827 828 /* 829 * Activate the snapshot. 830 */ 831 832 if ((error = vfs_suspend(sc->sc_mount, 0)) != 0) 833 goto bad; 834 835 microtime(&sc->sc_time); 836 837 error = fscow_establish(sc->sc_mount, fss_copy_on_write, sc); 838 if (error == 0) 839 sc->sc_flags |= FSS_ACTIVE; 840 841 vfs_resume(sc->sc_mount); 842 843 if (error != 0) 844 goto bad; 845 846 aprint_debug_dev(sc->sc_dev, "%s snapshot active\n", sc->sc_mntname); 847 aprint_debug_dev(sc->sc_dev, 848 "%u clusters of %u, %u cache slots, %u indir clusters\n", 849 sc->sc_clcount, FSS_CLSIZE(sc), 850 sc->sc_cache_size, sc->sc_indir_size); 851 852 return 0; 853 854 bad: 855 fss_softc_free(sc); 856 if (sc->sc_bs_vp != NULL) { 857 if (sc->sc_flags & FSS_PERSISTENT) 858 vrele(sc->sc_bs_vp); 859 else 860 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 861 } 862 sc->sc_bs_vp = NULL; 863 864 return error; 865 } 866 867 /* 868 * Delete a snapshot. 869 */ 870 static int 871 fss_delete_snapshot(struct fss_softc *sc, struct lwp *l) 872 { 873 874 if ((sc->sc_flags & FSS_PERSISTENT) == 0) 875 fscow_disestablish(sc->sc_mount, fss_copy_on_write, sc); 876 877 mutex_enter(&sc->sc_slock); 878 sc->sc_flags &= ~(FSS_ACTIVE|FSS_ERROR); 879 sc->sc_mount = NULL; 880 sc->sc_bdev = NODEV; 881 mutex_exit(&sc->sc_slock); 882 883 fss_softc_free(sc); 884 if (sc->sc_flags & FSS_PERSISTENT) 885 vrele(sc->sc_bs_vp); 886 else 887 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 888 sc->sc_bs_vp = NULL; 889 sc->sc_flags &= ~FSS_PERSISTENT; 890 891 return 0; 892 } 893 894 /* 895 * Read a cluster from the snapshotted block device to the cache. 896 */ 897 static int 898 fss_read_cluster(struct fss_softc *sc, u_int32_t cl) 899 { 900 int error, todo, offset, len; 901 daddr_t dblk; 902 struct buf *bp, *mbp; 903 struct fss_cache *scp, *scl; 904 905 /* 906 * Get a free cache slot. 907 */ 908 scl = sc->sc_cache+sc->sc_cache_size; 909 910 mutex_enter(&sc->sc_slock); 911 912 restart: 913 if (isset(sc->sc_copied, cl) || !FSS_ISVALID(sc)) { 914 mutex_exit(&sc->sc_slock); 915 return 0; 916 } 917 918 for (scp = sc->sc_cache; scp < scl; scp++) 919 if (scp->fc_cluster == cl) { 920 if (scp->fc_type == FSS_CACHE_VALID) { 921 mutex_exit(&sc->sc_slock); 922 return 0; 923 } else if (scp->fc_type == FSS_CACHE_BUSY) { 924 cv_wait(&scp->fc_state_cv, &sc->sc_slock); 925 goto restart; 926 } 927 } 928 929 for (scp = sc->sc_cache; scp < scl; scp++) 930 if (scp->fc_type == FSS_CACHE_FREE) { 931 scp->fc_type = FSS_CACHE_BUSY; 932 scp->fc_cluster = cl; 933 break; 934 } 935 if (scp >= scl) { 936 cv_wait(&sc->sc_cache_cv, &sc->sc_slock); 937 goto restart; 938 } 939 940 mutex_exit(&sc->sc_slock); 941 942 /* 943 * Start the read. 944 */ 945 dblk = btodb(FSS_CLTOB(sc, cl)); 946 if (cl == sc->sc_clcount-1) { 947 todo = sc->sc_clresid; 948 memset((char *)scp->fc_data + todo, 0, FSS_CLSIZE(sc) - todo); 949 } else 950 todo = FSS_CLSIZE(sc); 951 offset = 0; 952 mbp = getiobuf(NULL, true); 953 mbp->b_bufsize = todo; 954 mbp->b_data = scp->fc_data; 955 mbp->b_resid = mbp->b_bcount = todo; 956 mbp->b_flags = B_READ; 957 mbp->b_cflags = BC_BUSY; 958 mbp->b_dev = sc->sc_bdev; 959 while (todo > 0) { 960 len = todo; 961 if (len > MAXPHYS) 962 len = MAXPHYS; 963 if (btodb(FSS_CLTOB(sc, cl)) == dblk && len == todo) 964 bp = mbp; 965 else { 966 bp = getiobuf(NULL, true); 967 nestiobuf_setup(mbp, bp, offset, len); 968 } 969 bp->b_lblkno = 0; 970 bp->b_blkno = dblk; 971 bdev_strategy(bp); 972 dblk += btodb(len); 973 offset += len; 974 todo -= len; 975 } 976 error = biowait(mbp); 977 putiobuf(mbp); 978 979 mutex_enter(&sc->sc_slock); 980 scp->fc_type = (error ? FSS_CACHE_FREE : FSS_CACHE_VALID); 981 cv_broadcast(&scp->fc_state_cv); 982 if (error == 0) { 983 setbit(sc->sc_copied, scp->fc_cluster); 984 cv_signal(&sc->sc_work_cv); 985 } 986 mutex_exit(&sc->sc_slock); 987 988 return error; 989 } 990 991 /* 992 * Read/write clusters from/to backing store. 993 * For persistent snapshots must be called with cl == 0. off is the 994 * offset into the snapshot. 995 */ 996 static int 997 fss_bs_io(struct fss_softc *sc, fss_io_type rw, 998 u_int32_t cl, off_t off, int len, void *data) 999 { 1000 int error; 1001 1002 off += FSS_CLTOB(sc, cl); 1003 1004 vn_lock(sc->sc_bs_vp, LK_EXCLUSIVE|LK_RETRY); 1005 1006 error = vn_rdwr((rw == FSS_READ ? UIO_READ : UIO_WRITE), sc->sc_bs_vp, 1007 data, len, off, UIO_SYSSPACE, 1008 IO_ADV_ENCODE(POSIX_FADV_NOREUSE) | IO_NODELOCKED, 1009 sc->sc_bs_lwp->l_cred, NULL, NULL); 1010 if (error == 0) { 1011 mutex_enter(sc->sc_bs_vp->v_interlock); 1012 error = VOP_PUTPAGES(sc->sc_bs_vp, trunc_page(off), 1013 round_page(off+len), PGO_CLEANIT | PGO_FREE | PGO_SYNCIO); 1014 } 1015 1016 VOP_UNLOCK(sc->sc_bs_vp); 1017 1018 return error; 1019 } 1020 1021 /* 1022 * Get a pointer to the indirect slot for this cluster. 1023 */ 1024 static u_int32_t * 1025 fss_bs_indir(struct fss_softc *sc, u_int32_t cl) 1026 { 1027 u_int32_t icl; 1028 int ioff; 1029 1030 icl = cl/(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 1031 ioff = cl%(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 1032 1033 if (sc->sc_indir_cur == icl) 1034 return &sc->sc_indir_data[ioff]; 1035 1036 if (sc->sc_indir_dirty) { 1037 if (fss_bs_io(sc, FSS_WRITE, sc->sc_indir_cur, 0, 1038 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 1039 return NULL; 1040 setbit(sc->sc_indir_valid, sc->sc_indir_cur); 1041 } 1042 1043 sc->sc_indir_dirty = 0; 1044 sc->sc_indir_cur = icl; 1045 1046 if (isset(sc->sc_indir_valid, sc->sc_indir_cur)) { 1047 if (fss_bs_io(sc, FSS_READ, sc->sc_indir_cur, 0, 1048 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 1049 return NULL; 1050 } else 1051 memset(sc->sc_indir_data, 0, FSS_CLSIZE(sc)); 1052 1053 return &sc->sc_indir_data[ioff]; 1054 } 1055 1056 /* 1057 * The kernel thread (one for every active snapshot). 1058 * 1059 * After wakeup it cleans the cache and runs the I/O requests. 1060 */ 1061 static void 1062 fss_bs_thread(void *arg) 1063 { 1064 bool thread_idle, is_valid; 1065 int error, i, todo, len, crotor, is_read; 1066 long off; 1067 char *addr; 1068 u_int32_t c, cl, ch, *indirp; 1069 struct buf *bp, *nbp; 1070 struct fss_softc *sc; 1071 struct fss_cache *scp, *scl; 1072 1073 sc = arg; 1074 scl = sc->sc_cache+sc->sc_cache_size; 1075 crotor = 0; 1076 thread_idle = false; 1077 1078 mutex_enter(&sc->sc_slock); 1079 1080 for (;;) { 1081 if (thread_idle) 1082 cv_wait(&sc->sc_work_cv, &sc->sc_slock); 1083 thread_idle = true; 1084 if ((sc->sc_flags & FSS_BS_THREAD) == 0) { 1085 mutex_exit(&sc->sc_slock); 1086 kthread_exit(0); 1087 } 1088 1089 /* 1090 * Process I/O requests (persistent) 1091 */ 1092 1093 if (sc->sc_flags & FSS_PERSISTENT) { 1094 if ((bp = bufq_get(sc->sc_bufq)) == NULL) 1095 continue; 1096 is_valid = FSS_ISVALID(sc); 1097 is_read = (bp->b_flags & B_READ); 1098 thread_idle = false; 1099 mutex_exit(&sc->sc_slock); 1100 1101 if (is_valid) { 1102 disk_busy(sc->sc_dkdev); 1103 error = fss_bs_io(sc, FSS_READ, 0, 1104 dbtob(bp->b_blkno), bp->b_bcount, 1105 bp->b_data); 1106 disk_unbusy(sc->sc_dkdev, 1107 (error ? 0 : bp->b_bcount), is_read); 1108 } else 1109 error = ENXIO; 1110 1111 bp->b_error = error; 1112 bp->b_resid = (error ? bp->b_bcount : 0); 1113 biodone(bp); 1114 1115 mutex_enter(&sc->sc_slock); 1116 continue; 1117 } 1118 1119 /* 1120 * Clean the cache 1121 */ 1122 for (i = 0; i < sc->sc_cache_size; i++) { 1123 crotor = (crotor + 1) % sc->sc_cache_size; 1124 scp = sc->sc_cache + crotor; 1125 if (scp->fc_type != FSS_CACHE_VALID) 1126 continue; 1127 mutex_exit(&sc->sc_slock); 1128 1129 thread_idle = false; 1130 indirp = fss_bs_indir(sc, scp->fc_cluster); 1131 if (indirp != NULL) { 1132 error = fss_bs_io(sc, FSS_WRITE, sc->sc_clnext, 1133 0, FSS_CLSIZE(sc), scp->fc_data); 1134 } else 1135 error = EIO; 1136 1137 mutex_enter(&sc->sc_slock); 1138 if (error == 0) { 1139 *indirp = sc->sc_clnext++; 1140 sc->sc_indir_dirty = 1; 1141 } else 1142 fss_error(sc, "write error on backing store"); 1143 1144 scp->fc_type = FSS_CACHE_FREE; 1145 cv_signal(&sc->sc_cache_cv); 1146 break; 1147 } 1148 1149 /* 1150 * Process I/O requests 1151 */ 1152 if ((bp = bufq_get(sc->sc_bufq)) == NULL) 1153 continue; 1154 is_valid = FSS_ISVALID(sc); 1155 is_read = (bp->b_flags & B_READ); 1156 thread_idle = false; 1157 1158 if (!is_valid) { 1159 mutex_exit(&sc->sc_slock); 1160 1161 bp->b_error = ENXIO; 1162 bp->b_resid = bp->b_bcount; 1163 biodone(bp); 1164 1165 mutex_enter(&sc->sc_slock); 1166 continue; 1167 } 1168 1169 disk_busy(sc->sc_dkdev); 1170 1171 /* 1172 * First read from the snapshotted block device unless 1173 * this request is completely covered by backing store. 1174 */ 1175 1176 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 1177 off = FSS_CLOFF(sc, dbtob(bp->b_blkno)); 1178 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 1179 error = 0; 1180 bp->b_resid = 0; 1181 bp->b_error = 0; 1182 for (c = cl; c <= ch; c++) { 1183 if (isset(sc->sc_copied, c)) 1184 continue; 1185 mutex_exit(&sc->sc_slock); 1186 1187 /* Not on backing store, read from device. */ 1188 nbp = getiobuf(NULL, true); 1189 nbp->b_flags = B_READ; 1190 nbp->b_resid = nbp->b_bcount = bp->b_bcount; 1191 nbp->b_bufsize = bp->b_bcount; 1192 nbp->b_data = bp->b_data; 1193 nbp->b_blkno = bp->b_blkno; 1194 nbp->b_lblkno = 0; 1195 nbp->b_dev = sc->sc_bdev; 1196 SET(nbp->b_cflags, BC_BUSY); /* mark buffer busy */ 1197 1198 bdev_strategy(nbp); 1199 1200 error = biowait(nbp); 1201 if (error != 0) { 1202 bp->b_resid = bp->b_bcount; 1203 bp->b_error = nbp->b_error; 1204 disk_unbusy(sc->sc_dkdev, 0, is_read); 1205 biodone(bp); 1206 } 1207 putiobuf(nbp); 1208 1209 mutex_enter(&sc->sc_slock); 1210 break; 1211 } 1212 if (error) 1213 continue; 1214 1215 /* 1216 * Replace those parts that have been saved to backing store. 1217 */ 1218 1219 addr = bp->b_data; 1220 todo = bp->b_bcount; 1221 for (c = cl; c <= ch; c++, off = 0, todo -= len, addr += len) { 1222 len = FSS_CLSIZE(sc)-off; 1223 if (len > todo) 1224 len = todo; 1225 if (isclr(sc->sc_copied, c)) 1226 continue; 1227 mutex_exit(&sc->sc_slock); 1228 1229 indirp = fss_bs_indir(sc, c); 1230 if (indirp == NULL || *indirp == 0) { 1231 /* 1232 * Not on backing store. Either in cache 1233 * or hole in the snapshotted block device. 1234 */ 1235 1236 mutex_enter(&sc->sc_slock); 1237 for (scp = sc->sc_cache; scp < scl; scp++) 1238 if (scp->fc_type == FSS_CACHE_VALID && 1239 scp->fc_cluster == c) 1240 break; 1241 if (scp < scl) 1242 memcpy(addr, (char *)scp->fc_data+off, 1243 len); 1244 else 1245 memset(addr, 0, len); 1246 continue; 1247 } 1248 1249 /* 1250 * Read from backing store. 1251 */ 1252 error = 1253 fss_bs_io(sc, FSS_READ, *indirp, off, len, addr); 1254 1255 mutex_enter(&sc->sc_slock); 1256 if (error) { 1257 bp->b_resid = bp->b_bcount; 1258 bp->b_error = error; 1259 break; 1260 } 1261 } 1262 mutex_exit(&sc->sc_slock); 1263 1264 disk_unbusy(sc->sc_dkdev, (error ? 0 : bp->b_bcount), is_read); 1265 biodone(bp); 1266 1267 mutex_enter(&sc->sc_slock); 1268 } 1269 } 1270 1271 #ifdef _MODULE 1272 1273 #include <sys/module.h> 1274 1275 MODULE(MODULE_CLASS_DRIVER, fss, NULL); 1276 CFDRIVER_DECL(fss, DV_DISK, NULL); 1277 1278 static int 1279 fss_modcmd(modcmd_t cmd, void *arg) 1280 { 1281 int bmajor = -1, cmajor = -1, error = 0; 1282 1283 switch (cmd) { 1284 case MODULE_CMD_INIT: 1285 mutex_init(&fss_device_lock, MUTEX_DEFAULT, IPL_NONE); 1286 error = config_cfdriver_attach(&fss_cd); 1287 if (error) { 1288 mutex_destroy(&fss_device_lock); 1289 break; 1290 } 1291 error = config_cfattach_attach(fss_cd.cd_name, &fss_ca); 1292 if (error) { 1293 config_cfdriver_detach(&fss_cd); 1294 mutex_destroy(&fss_device_lock); 1295 break; 1296 } 1297 error = devsw_attach(fss_cd.cd_name, 1298 &fss_bdevsw, &bmajor, &fss_cdevsw, &cmajor); 1299 if (error == EEXIST) 1300 error = 0; 1301 if (error) { 1302 config_cfattach_detach(fss_cd.cd_name, &fss_ca); 1303 config_cfdriver_detach(&fss_cd); 1304 mutex_destroy(&fss_device_lock); 1305 break; 1306 } 1307 break; 1308 1309 case MODULE_CMD_FINI: 1310 error = config_cfattach_detach(fss_cd.cd_name, &fss_ca); 1311 if (error) 1312 break; 1313 config_cfdriver_detach(&fss_cd); 1314 devsw_detach(&fss_bdevsw, &fss_cdevsw); 1315 mutex_destroy(&fss_device_lock); 1316 break; 1317 1318 default: 1319 error = ENOTTY; 1320 break; 1321 } 1322 1323 return error; 1324 } 1325 1326 #endif /* _MODULE */ 1327