1 /* 2 * Copyright (c) 2007-2008 The DragonFly Project. All rights reserved. 3 * 4 * This code is derived from software contributed to The DragonFly Project 5 * by Matthew Dillon <dillon@backplane.com> 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * 3. Neither the name of The DragonFly Project nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific, prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * $DragonFly: src/sys/vfs/hammer/hammer_vfsops.c,v 1.74 2008/11/13 02:18:43 dillon Exp $ 35 */ 36 37 #include <sys/param.h> 38 #include <sys/systm.h> 39 #include <sys/kernel.h> 40 #include <sys/vnode.h> 41 #include <sys/mount.h> 42 #include <sys/malloc.h> 43 #include <sys/nlookup.h> 44 #include <sys/fcntl.h> 45 #include <sys/sysctl.h> 46 #include <sys/buf.h> 47 #include <sys/buf2.h> 48 #include "hammer.h" 49 50 int hammer_supported_version = HAMMER_VOL_VERSION_TWO; 51 int hammer_debug_io; 52 int hammer_debug_general; 53 int hammer_debug_debug = 1; /* medium-error panics */ 54 int hammer_debug_inode; 55 int hammer_debug_locks; 56 int hammer_debug_btree; 57 int hammer_debug_tid; 58 int hammer_debug_recover; /* -1 will disable, +1 will force */ 59 int hammer_debug_recover_faults; 60 int hammer_error_panic; /* panic on error levels */ 61 int hammer_cluster_enable = 1; /* enable read clustering by default */ 62 int hammer_count_fsyncs; 63 int hammer_count_inodes; 64 int hammer_count_iqueued; 65 int hammer_count_reclaiming; 66 int hammer_count_records; 67 int hammer_count_record_datas; 68 int hammer_count_volumes; 69 int hammer_count_buffers; 70 int hammer_count_nodes; 71 int64_t hammer_count_extra_space_used; 72 int64_t hammer_stats_btree_lookups; 73 int64_t hammer_stats_btree_searches; 74 int64_t hammer_stats_btree_inserts; 75 int64_t hammer_stats_btree_deletes; 76 int64_t hammer_stats_btree_elements; 77 int64_t hammer_stats_btree_splits; 78 int64_t hammer_stats_btree_iterations; 79 int64_t hammer_stats_btree_root_iterations; 80 int64_t hammer_stats_record_iterations; 81 82 int64_t hammer_stats_file_read; 83 int64_t hammer_stats_file_write; 84 int64_t hammer_stats_file_iopsr; 85 int64_t hammer_stats_file_iopsw; 86 int64_t hammer_stats_disk_read; 87 int64_t hammer_stats_disk_write; 88 int64_t hammer_stats_inode_flushes; 89 int64_t hammer_stats_commits; 90 int64_t hammer_stats_undo; 91 92 int hammer_count_dirtybufspace; /* global */ 93 int hammer_count_refedbufs; /* global */ 94 int hammer_count_reservations; 95 int hammer_count_io_running_read; 96 int hammer_count_io_running_write; 97 int hammer_count_io_locked; 98 int hammer_limit_dirtybufspace; /* per-mount */ 99 int hammer_limit_recs; /* as a whole XXX */ 100 int hammer_limit_inode_recs = 1024; /* per inode */ 101 int hammer_autoflush = 2000; /* auto flush */ 102 int hammer_bio_count; 103 int hammer_verify_zone; 104 int hammer_verify_data = 1; 105 int hammer_write_mode; 106 int hammer_yield_check = 16; 107 int hammer_fsync_mode; 108 int64_t hammer_contention_count; 109 int64_t hammer_zone_limit; 110 111 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem"); 112 SYSCTL_INT(_vfs_hammer, OID_AUTO, supported_version, CTLFLAG_RD, 113 &hammer_supported_version, 0, ""); 114 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW, 115 &hammer_debug_general, 0, ""); 116 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW, 117 &hammer_debug_io, 0, ""); 118 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW, 119 &hammer_debug_debug, 0, ""); 120 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW, 121 &hammer_debug_inode, 0, ""); 122 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW, 123 &hammer_debug_locks, 0, ""); 124 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW, 125 &hammer_debug_btree, 0, ""); 126 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW, 127 &hammer_debug_tid, 0, ""); 128 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW, 129 &hammer_debug_recover, 0, ""); 130 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW, 131 &hammer_debug_recover_faults, 0, ""); 132 SYSCTL_INT(_vfs_hammer, OID_AUTO, error_panic, CTLFLAG_RW, 133 &hammer_error_panic, 0, ""); 134 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW, 135 &hammer_cluster_enable, 0, ""); 136 137 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW, 138 &hammer_limit_dirtybufspace, 0, ""); 139 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW, 140 &hammer_limit_recs, 0, ""); 141 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_inode_recs, CTLFLAG_RW, 142 &hammer_limit_inode_recs, 0, ""); 143 144 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD, 145 &hammer_count_fsyncs, 0, ""); 146 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD, 147 &hammer_count_inodes, 0, ""); 148 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD, 149 &hammer_count_iqueued, 0, ""); 150 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaiming, CTLFLAG_RD, 151 &hammer_count_reclaiming, 0, ""); 152 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD, 153 &hammer_count_records, 0, ""); 154 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD, 155 &hammer_count_record_datas, 0, ""); 156 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD, 157 &hammer_count_volumes, 0, ""); 158 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD, 159 &hammer_count_buffers, 0, ""); 160 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD, 161 &hammer_count_nodes, 0, ""); 162 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD, 163 &hammer_count_extra_space_used, 0, ""); 164 165 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD, 166 &hammer_stats_btree_searches, 0, ""); 167 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD, 168 &hammer_stats_btree_lookups, 0, ""); 169 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD, 170 &hammer_stats_btree_inserts, 0, ""); 171 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD, 172 &hammer_stats_btree_deletes, 0, ""); 173 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD, 174 &hammer_stats_btree_elements, 0, ""); 175 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD, 176 &hammer_stats_btree_splits, 0, ""); 177 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD, 178 &hammer_stats_btree_iterations, 0, ""); 179 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_root_iterations, CTLFLAG_RD, 180 &hammer_stats_btree_root_iterations, 0, ""); 181 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD, 182 &hammer_stats_record_iterations, 0, ""); 183 184 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD, 185 &hammer_stats_file_read, 0, ""); 186 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD, 187 &hammer_stats_file_write, 0, ""); 188 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD, 189 &hammer_stats_file_iopsr, 0, ""); 190 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD, 191 &hammer_stats_file_iopsw, 0, ""); 192 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD, 193 &hammer_stats_disk_read, 0, ""); 194 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD, 195 &hammer_stats_disk_write, 0, ""); 196 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD, 197 &hammer_stats_inode_flushes, 0, ""); 198 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD, 199 &hammer_stats_commits, 0, ""); 200 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_undo, CTLFLAG_RD, 201 &hammer_stats_undo, 0, ""); 202 203 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD, 204 &hammer_count_dirtybufspace, 0, ""); 205 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD, 206 &hammer_count_refedbufs, 0, ""); 207 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD, 208 &hammer_count_reservations, 0, ""); 209 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD, 210 &hammer_count_io_running_read, 0, ""); 211 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD, 212 &hammer_count_io_locked, 0, ""); 213 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD, 214 &hammer_count_io_running_write, 0, ""); 215 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW, 216 &hammer_zone_limit, 0, ""); 217 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW, 218 &hammer_contention_count, 0, ""); 219 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW, 220 &hammer_autoflush, 0, ""); 221 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW, 222 &hammer_verify_zone, 0, ""); 223 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW, 224 &hammer_verify_data, 0, ""); 225 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW, 226 &hammer_write_mode, 0, ""); 227 SYSCTL_INT(_vfs_hammer, OID_AUTO, yield_check, CTLFLAG_RW, 228 &hammer_yield_check, 0, ""); 229 SYSCTL_INT(_vfs_hammer, OID_AUTO, fsync_mode, CTLFLAG_RW, 230 &hammer_fsync_mode, 0, ""); 231 232 KTR_INFO_MASTER(hammer); 233 234 /* 235 * VFS ABI 236 */ 237 static void hammer_free_hmp(struct mount *mp); 238 239 static int hammer_vfs_mount(struct mount *mp, char *path, caddr_t data, 240 struct ucred *cred); 241 static int hammer_vfs_unmount(struct mount *mp, int mntflags); 242 static int hammer_vfs_root(struct mount *mp, struct vnode **vpp); 243 static int hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, 244 struct ucred *cred); 245 static int hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, 246 struct ucred *cred); 247 static int hammer_vfs_sync(struct mount *mp, int waitfor); 248 static int hammer_vfs_vget(struct mount *mp, struct vnode *dvp, 249 ino_t ino, struct vnode **vpp); 250 static int hammer_vfs_init(struct vfsconf *conf); 251 static int hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp, 252 struct fid *fhp, struct vnode **vpp); 253 static int hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp); 254 static int hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam, 255 int *exflagsp, struct ucred **credanonp); 256 257 258 static struct vfsops hammer_vfsops = { 259 .vfs_mount = hammer_vfs_mount, 260 .vfs_unmount = hammer_vfs_unmount, 261 .vfs_root = hammer_vfs_root, 262 .vfs_statfs = hammer_vfs_statfs, 263 .vfs_statvfs = hammer_vfs_statvfs, 264 .vfs_sync = hammer_vfs_sync, 265 .vfs_vget = hammer_vfs_vget, 266 .vfs_init = hammer_vfs_init, 267 .vfs_vptofh = hammer_vfs_vptofh, 268 .vfs_fhtovp = hammer_vfs_fhtovp, 269 .vfs_checkexp = hammer_vfs_checkexp 270 }; 271 272 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", ""); 273 274 VFS_SET(hammer_vfsops, hammer, 0); 275 MODULE_VERSION(hammer, 1); 276 277 static int 278 hammer_vfs_init(struct vfsconf *conf) 279 { 280 int n; 281 282 if (hammer_limit_recs == 0) { 283 hammer_limit_recs = nbuf * 25; 284 n = kmalloc_limit(M_HAMMER) / 512; 285 if (hammer_limit_recs > n) 286 hammer_limit_recs = n; 287 } 288 if (hammer_limit_dirtybufspace == 0) { 289 hammer_limit_dirtybufspace = hidirtybufspace / 2; 290 if (hammer_limit_dirtybufspace < 100) 291 hammer_limit_dirtybufspace = 100; 292 } 293 return(0); 294 } 295 296 static int 297 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data, 298 struct ucred *cred) 299 { 300 struct hammer_mount_info info; 301 hammer_mount_t hmp; 302 hammer_volume_t rootvol; 303 struct vnode *rootvp; 304 struct vnode *devvp = NULL; 305 const char *upath; /* volume name in userspace */ 306 char *path; /* volume name in system space */ 307 int error; 308 int i; 309 int master_id; 310 int maxinodes; 311 312 /* 313 * Accept hammer_mount_info. mntpt is NULL for root mounts at boot. 314 */ 315 if (mntpt == NULL) { 316 if ((error = bdevvp(rootdev, &devvp))) { 317 kprintf("hammer_mountroot: can't find devvp\n"); 318 return (error); 319 } 320 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */ 321 bzero(&info, sizeof(info)); 322 info.asof = 0; 323 info.hflags = 0; 324 info.nvolumes = 1; 325 } else { 326 if ((error = copyin(data, &info, sizeof(info))) != 0) 327 return (error); 328 } 329 330 /* 331 * updating or new mount 332 */ 333 if (mp->mnt_flag & MNT_UPDATE) { 334 hmp = (void *)mp->mnt_data; 335 KKASSERT(hmp != NULL); 336 } else { 337 if (info.nvolumes <= 0 || info.nvolumes >= 32768) 338 return (EINVAL); 339 hmp = NULL; 340 } 341 342 /* 343 * master-id validation. The master id may not be changed by a 344 * mount update. 345 */ 346 if (info.hflags & HMNT_MASTERID) { 347 if (hmp && hmp->master_id != info.master_id) { 348 kprintf("hammer: cannot change master id " 349 "with mount update\n"); 350 return(EINVAL); 351 } 352 master_id = info.master_id; 353 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS) 354 return (EINVAL); 355 } else { 356 if (hmp) 357 master_id = hmp->master_id; 358 else 359 master_id = 0; 360 } 361 362 /* 363 * Interal mount data structure 364 */ 365 if (hmp == NULL) { 366 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO); 367 mp->mnt_data = (qaddr_t)hmp; 368 hmp->mp = mp; 369 /*TAILQ_INIT(&hmp->recycle_list);*/ 370 371 /* 372 * Make sure kmalloc type limits are set appropriately. If root 373 * increases the vnode limit you may have to do a dummy remount 374 * to adjust the HAMMER inode limit. 375 */ 376 kmalloc_create(&hmp->m_misc, "HAMMER-others"); 377 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes"); 378 379 maxinodes = desiredvnodes + desiredvnodes / 5 + 380 HAMMER_RECLAIM_WAIT; 381 kmalloc_raise_limit(hmp->m_inodes, 382 maxinodes * sizeof(struct hammer_inode)); 383 384 hmp->root_btree_beg.localization = 0x00000000U; 385 hmp->root_btree_beg.obj_id = -0x8000000000000000LL; 386 hmp->root_btree_beg.key = -0x8000000000000000LL; 387 hmp->root_btree_beg.create_tid = 1; 388 hmp->root_btree_beg.delete_tid = 1; 389 hmp->root_btree_beg.rec_type = 0; 390 hmp->root_btree_beg.obj_type = 0; 391 392 hmp->root_btree_end.localization = 0xFFFFFFFFU; 393 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL; 394 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL; 395 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL; 396 hmp->root_btree_end.delete_tid = 0; /* special case */ 397 hmp->root_btree_end.rec_type = 0xFFFFU; 398 hmp->root_btree_end.obj_type = 0; 399 400 hmp->krate.freq = 1; /* maximum reporting rate (hz) */ 401 hmp->krate.count = -16; /* initial burst */ 402 403 hmp->sync_lock.refs = 1; 404 hmp->free_lock.refs = 1; 405 hmp->undo_lock.refs = 1; 406 hmp->blkmap_lock.refs = 1; 407 408 TAILQ_INIT(&hmp->delay_list); 409 TAILQ_INIT(&hmp->flush_group_list); 410 TAILQ_INIT(&hmp->objid_cache_list); 411 TAILQ_INIT(&hmp->undo_lru_list); 412 TAILQ_INIT(&hmp->reclaim_list); 413 } 414 hmp->hflags &= ~HMNT_USERFLAGS; 415 hmp->hflags |= info.hflags & HMNT_USERFLAGS; 416 417 hmp->master_id = master_id; 418 419 if (info.asof) { 420 mp->mnt_flag |= MNT_RDONLY; 421 hmp->asof = info.asof; 422 } else { 423 hmp->asof = HAMMER_MAX_TID; 424 } 425 426 /* 427 * Re-open read-write if originally read-only, or vise-versa. 428 */ 429 if (mp->mnt_flag & MNT_UPDATE) { 430 error = 0; 431 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) { 432 kprintf("HAMMER read-only -> read-write\n"); 433 hmp->ronly = 0; 434 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL, 435 hammer_adjust_volume_mode, NULL); 436 rootvol = hammer_get_root_volume(hmp, &error); 437 if (rootvol) { 438 hammer_recover_flush_buffers(hmp, rootvol, 1); 439 bcopy(rootvol->ondisk->vol0_blockmap, 440 hmp->blockmap, 441 sizeof(hmp->blockmap)); 442 hammer_rel_volume(rootvol, 0); 443 } 444 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL, 445 hammer_reload_inode, NULL); 446 /* kernel clears MNT_RDONLY */ 447 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) { 448 kprintf("HAMMER read-write -> read-only\n"); 449 hmp->ronly = 1; /* messy */ 450 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL, 451 hammer_reload_inode, NULL); 452 hmp->ronly = 0; 453 hammer_flusher_sync(hmp); 454 hammer_flusher_sync(hmp); 455 hammer_flusher_sync(hmp); 456 hmp->ronly = 1; 457 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL, 458 hammer_adjust_volume_mode, NULL); 459 } 460 return(error); 461 } 462 463 RB_INIT(&hmp->rb_vols_root); 464 RB_INIT(&hmp->rb_inos_root); 465 RB_INIT(&hmp->rb_nods_root); 466 RB_INIT(&hmp->rb_undo_root); 467 RB_INIT(&hmp->rb_resv_root); 468 RB_INIT(&hmp->rb_bufs_root); 469 RB_INIT(&hmp->rb_pfsm_root); 470 471 hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0); 472 473 TAILQ_INIT(&hmp->volu_list); 474 TAILQ_INIT(&hmp->undo_list); 475 TAILQ_INIT(&hmp->data_list); 476 TAILQ_INIT(&hmp->meta_list); 477 TAILQ_INIT(&hmp->lose_list); 478 479 /* 480 * Load volumes 481 */ 482 path = objcache_get(namei_oc, M_WAITOK); 483 hmp->nvolumes = -1; 484 for (i = 0; i < info.nvolumes; ++i) { 485 if (mntpt == NULL) { 486 /* 487 * Root mount. 488 * Only one volume; and no need for copyin. 489 */ 490 KKASSERT(info.nvolumes == 1); 491 ksnprintf(path, MAXPATHLEN, "/dev/%s", 492 mp->mnt_stat.f_mntfromname); 493 error = 0; 494 } else { 495 error = copyin(&info.volumes[i], &upath, 496 sizeof(char *)); 497 if (error == 0) 498 error = copyinstr(upath, path, 499 MAXPATHLEN, NULL); 500 } 501 if (error == 0) 502 error = hammer_install_volume(hmp, path, devvp); 503 if (error) 504 break; 505 } 506 objcache_put(namei_oc, path); 507 508 /* 509 * Make sure we found a root volume 510 */ 511 if (error == 0 && hmp->rootvol == NULL) { 512 kprintf("hammer_mount: No root volume found!\n"); 513 error = EINVAL; 514 } 515 516 /* 517 * Check that all required volumes are available 518 */ 519 if (error == 0 && hammer_mountcheck_volumes(hmp)) { 520 kprintf("hammer_mount: Missing volumes, cannot mount!\n"); 521 error = EINVAL; 522 } 523 524 if (error) { 525 hammer_free_hmp(mp); 526 return (error); 527 } 528 529 /* 530 * No errors, setup enough of the mount point so we can lookup the 531 * root vnode. 532 */ 533 mp->mnt_iosize_max = MAXPHYS; 534 mp->mnt_kern_flag |= MNTK_FSMID; 535 536 /* 537 * note: f_iosize is used by vnode_pager_haspage() when constructing 538 * its VOP_BMAP call. 539 */ 540 mp->mnt_stat.f_iosize = HAMMER_BUFSIZE; 541 mp->mnt_stat.f_bsize = HAMMER_BUFSIZE; 542 543 mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE; 544 mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE; 545 546 mp->mnt_maxsymlinklen = 255; 547 mp->mnt_flag |= MNT_LOCAL; 548 549 vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops); 550 vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops); 551 vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops); 552 553 /* 554 * The root volume's ondisk pointer is only valid if we hold a 555 * reference to it. 556 */ 557 rootvol = hammer_get_root_volume(hmp, &error); 558 if (error) 559 goto failed; 560 561 /* 562 * Perform any necessary UNDO operations. The recovery code does 563 * call hammer_undo_lookup() so we have to pre-cache the blockmap, 564 * and then re-copy it again after recovery is complete. 565 * 566 * If this is a read-only mount the UNDO information is retained 567 * in memory in the form of dirty buffer cache buffers, and not 568 * written back to the media. 569 */ 570 bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap, 571 sizeof(hmp->blockmap)); 572 573 /* 574 * Check filesystem version 575 */ 576 hmp->version = rootvol->ondisk->vol_version; 577 if (hmp->version < HAMMER_VOL_VERSION_MIN || 578 hmp->version > HAMMER_VOL_VERSION_MAX) { 579 kprintf("HAMMER: mount unsupported fs version %d\n", 580 hmp->version); 581 error = ERANGE; 582 goto done; 583 } 584 585 /* 586 * The undo_rec_limit limits the size of flush groups to avoid 587 * blowing out the UNDO FIFO. This calculation is typically in 588 * the tens of thousands and is designed primarily when small 589 * HAMMER filesystems are created. 590 */ 591 hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100; 592 if (hammer_debug_general & 0x0001) 593 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit); 594 595 error = hammer_recover(hmp, rootvol); 596 if (error) { 597 kprintf("Failed to recover HAMMER filesystem on mount\n"); 598 goto done; 599 } 600 601 /* 602 * Finish setup now that we have a good root volume. 603 * 604 * The top 16 bits of fsid.val[1] is a pfs id. 605 */ 606 ksnprintf(mp->mnt_stat.f_mntfromname, 607 sizeof(mp->mnt_stat.f_mntfromname), "%s", 608 rootvol->ondisk->vol_name); 609 mp->mnt_stat.f_fsid.val[0] = 610 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8); 611 mp->mnt_stat.f_fsid.val[1] = 612 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8); 613 mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF; 614 615 mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid; 616 mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid, 617 sizeof(mp->mnt_vstat.f_fsid_uuid)); 618 619 /* 620 * Certain often-modified fields in the root volume are cached in 621 * the hammer_mount structure so we do not have to generate lots 622 * of little UNDO structures for them. 623 * 624 * Recopy after recovery. This also has the side effect of 625 * setting our cached undo FIFO's first_offset, which serves to 626 * placemark the FIFO start for the NEXT flush cycle while the 627 * on-disk first_offset represents the LAST flush cycle. 628 */ 629 hmp->next_tid = rootvol->ondisk->vol0_next_tid; 630 hmp->flush_tid1 = hmp->next_tid; 631 hmp->flush_tid2 = hmp->next_tid; 632 bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap, 633 sizeof(hmp->blockmap)); 634 hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks; 635 636 hammer_flusher_create(hmp); 637 638 /* 639 * Locate the root directory using the root cluster's B-Tree as a 640 * starting point. The root directory uses an obj_id of 1. 641 * 642 * FUTURE: Leave the root directory cached referenced but unlocked 643 * in hmp->rootvp (need to flush it on unmount). 644 */ 645 error = hammer_vfs_vget(mp, NULL, 1, &rootvp); 646 if (error) 647 goto done; 648 vput(rootvp); 649 /*vn_unlock(hmp->rootvp);*/ 650 651 done: 652 hammer_rel_volume(rootvol, 0); 653 failed: 654 /* 655 * Cleanup and return. 656 */ 657 if (error) 658 hammer_free_hmp(mp); 659 return (error); 660 } 661 662 static int 663 hammer_vfs_unmount(struct mount *mp, int mntflags) 664 { 665 #if 0 666 struct hammer_mount *hmp = (void *)mp->mnt_data; 667 #endif 668 int flags; 669 int error; 670 671 /* 672 * Clean out the vnodes 673 */ 674 flags = 0; 675 if (mntflags & MNT_FORCE) 676 flags |= FORCECLOSE; 677 if ((error = vflush(mp, 0, flags)) != 0) 678 return (error); 679 680 /* 681 * Clean up the internal mount structure and related entities. This 682 * may issue I/O. 683 */ 684 hammer_free_hmp(mp); 685 return(0); 686 } 687 688 /* 689 * Clean up the internal mount structure and disassociate it from the mount. 690 * This may issue I/O. 691 */ 692 static void 693 hammer_free_hmp(struct mount *mp) 694 { 695 struct hammer_mount *hmp = (void *)mp->mnt_data; 696 hammer_flush_group_t flg; 697 int count; 698 int dummy; 699 700 /* 701 * Flush anything dirty. This won't even run if the 702 * filesystem errored-out. 703 */ 704 count = 0; 705 while (hammer_flusher_haswork(hmp)) { 706 hammer_flusher_sync(hmp); 707 ++count; 708 if (count >= 5) { 709 if (count == 5) 710 kprintf("HAMMER: umount flushing."); 711 else 712 kprintf("."); 713 tsleep(&dummy, 0, "hmrufl", hz); 714 } 715 if (count == 30) { 716 kprintf("giving up\n"); 717 break; 718 } 719 } 720 if (count >= 5 && count < 30) 721 kprintf("\n"); 722 723 /* 724 * If the mount had a critical error we have to destroy any 725 * remaining inodes before we can finish cleaning up the flusher. 726 */ 727 if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) { 728 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL, 729 hammer_destroy_inode_callback, NULL); 730 } 731 732 /* 733 * There shouldn't be any inodes left now and any left over 734 * flush groups should now be empty. 735 */ 736 KKASSERT(RB_EMPTY(&hmp->rb_inos_root)); 737 while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) { 738 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry); 739 KKASSERT(TAILQ_EMPTY(&flg->flush_list)); 740 if (flg->refs) { 741 kprintf("HAMMER: Warning, flush_group %p was " 742 "not empty on umount!\n", flg); 743 } 744 kfree(flg, hmp->m_misc); 745 } 746 747 /* 748 * We can finally destroy the flusher 749 */ 750 hammer_flusher_destroy(hmp); 751 752 /* 753 * We may have held recovered buffers due to a read-only mount. 754 * These must be discarded. 755 */ 756 if (hmp->ronly) 757 hammer_recover_flush_buffers(hmp, NULL, -1); 758 759 /* 760 * Unload buffers and then volumes 761 */ 762 RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL, 763 hammer_unload_buffer, NULL); 764 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL, 765 hammer_unload_volume, NULL); 766 767 mp->mnt_data = NULL; 768 mp->mnt_flag &= ~MNT_LOCAL; 769 hmp->mp = NULL; 770 hammer_destroy_objid_cache(hmp); 771 kmalloc_destroy(&hmp->m_misc); 772 kmalloc_destroy(&hmp->m_inodes); 773 kfree(hmp, M_HAMMER); 774 } 775 776 /* 777 * Report critical errors. ip may be NULL. 778 */ 779 void 780 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip, 781 int error, const char *msg) 782 { 783 hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR; 784 785 krateprintf(&hmp->krate, 786 "HAMMER(%s): Critical error inode=%jd error=%d %s\n", 787 hmp->mp->mnt_stat.f_mntfromname, 788 (intmax_t)(ip ? ip->obj_id : -1), 789 error, msg); 790 791 if (hmp->ronly == 0) { 792 hmp->ronly = 2; /* special errored read-only mode */ 793 hmp->mp->mnt_flag |= MNT_RDONLY; 794 kprintf("HAMMER(%s): Forcing read-only mode\n", 795 hmp->mp->mnt_stat.f_mntfromname); 796 } 797 hmp->error = error; 798 if (hammer_error_panic > 2) 799 Debugger("Entering debugger"); 800 } 801 802 803 /* 804 * Obtain a vnode for the specified inode number. An exclusively locked 805 * vnode is returned. 806 */ 807 int 808 hammer_vfs_vget(struct mount *mp, struct vnode *dvp, 809 ino_t ino, struct vnode **vpp) 810 { 811 struct hammer_transaction trans; 812 struct hammer_mount *hmp = (void *)mp->mnt_data; 813 struct hammer_inode *ip; 814 int error; 815 u_int32_t localization; 816 817 hammer_simple_transaction(&trans, hmp); 818 819 /* 820 * If a directory vnode is supplied (mainly NFS) then we can acquire 821 * the PFS domain from it. Otherwise we would only be able to vget 822 * inodes in the root PFS. 823 */ 824 if (dvp) { 825 localization = HAMMER_DEF_LOCALIZATION + 826 VTOI(dvp)->obj_localization; 827 } else { 828 localization = HAMMER_DEF_LOCALIZATION; 829 } 830 831 /* 832 * Lookup the requested HAMMER inode. The structure must be 833 * left unlocked while we manipulate the related vnode to avoid 834 * a deadlock. 835 */ 836 ip = hammer_get_inode(&trans, NULL, ino, 837 hmp->asof, localization, 838 0, &error); 839 if (ip == NULL) { 840 *vpp = NULL; 841 hammer_done_transaction(&trans); 842 return(error); 843 } 844 error = hammer_get_vnode(ip, vpp); 845 hammer_rel_inode(ip, 0); 846 hammer_done_transaction(&trans); 847 return (error); 848 } 849 850 /* 851 * Return the root vnode for the filesystem. 852 * 853 * HAMMER stores the root vnode in the hammer_mount structure so 854 * getting it is easy. 855 */ 856 static int 857 hammer_vfs_root(struct mount *mp, struct vnode **vpp) 858 { 859 #if 0 860 struct hammer_mount *hmp = (void *)mp->mnt_data; 861 #endif 862 int error; 863 864 error = hammer_vfs_vget(mp, NULL, 1, vpp); 865 return (error); 866 } 867 868 static int 869 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred) 870 { 871 struct hammer_mount *hmp = (void *)mp->mnt_data; 872 hammer_volume_t volume; 873 hammer_volume_ondisk_t ondisk; 874 int error; 875 int64_t bfree; 876 int64_t breserved; 877 878 volume = hammer_get_root_volume(hmp, &error); 879 if (error) 880 return(error); 881 ondisk = volume->ondisk; 882 883 /* 884 * Basic stats 885 */ 886 _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved); 887 mp->mnt_stat.f_files = ondisk->vol0_stat_inodes; 888 bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE; 889 hammer_rel_volume(volume, 0); 890 891 mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE; 892 mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree; 893 if (mp->mnt_stat.f_files < 0) 894 mp->mnt_stat.f_files = 0; 895 896 *sbp = mp->mnt_stat; 897 return(0); 898 } 899 900 static int 901 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred) 902 { 903 struct hammer_mount *hmp = (void *)mp->mnt_data; 904 hammer_volume_t volume; 905 hammer_volume_ondisk_t ondisk; 906 int error; 907 int64_t bfree; 908 int64_t breserved; 909 910 volume = hammer_get_root_volume(hmp, &error); 911 if (error) 912 return(error); 913 ondisk = volume->ondisk; 914 915 /* 916 * Basic stats 917 */ 918 _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved); 919 mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes; 920 bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE; 921 hammer_rel_volume(volume, 0); 922 923 mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE; 924 mp->mnt_vstat.f_bavail = mp->mnt_stat.f_bfree; 925 if (mp->mnt_vstat.f_files < 0) 926 mp->mnt_vstat.f_files = 0; 927 *sbp = mp->mnt_vstat; 928 return(0); 929 } 930 931 /* 932 * Sync the filesystem. Currently we have to run it twice, the second 933 * one will advance the undo start index to the end index, so if a crash 934 * occurs no undos will be run on mount. 935 * 936 * We do not sync the filesystem if we are called from a panic. If we did 937 * we might end up blowing up a sync that was already in progress. 938 */ 939 static int 940 hammer_vfs_sync(struct mount *mp, int waitfor) 941 { 942 struct hammer_mount *hmp = (void *)mp->mnt_data; 943 int error; 944 945 if (panicstr == NULL) { 946 error = hammer_sync_hmp(hmp, waitfor); 947 } else { 948 error = EIO; 949 } 950 return (error); 951 } 952 953 /* 954 * Convert a vnode to a file handle. 955 */ 956 static int 957 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp) 958 { 959 hammer_inode_t ip; 960 961 KKASSERT(MAXFIDSZ >= 16); 962 ip = VTOI(vp); 963 fhp->fid_len = offsetof(struct fid, fid_data[16]); 964 fhp->fid_ext = ip->obj_localization >> 16; 965 bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id)); 966 bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof)); 967 return(0); 968 } 969 970 971 /* 972 * Convert a file handle back to a vnode. 973 * 974 * Use rootvp to enforce PFS isolation when a PFS is exported via a 975 * null mount. 976 */ 977 static int 978 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp, 979 struct fid *fhp, struct vnode **vpp) 980 { 981 struct hammer_transaction trans; 982 struct hammer_inode *ip; 983 struct hammer_inode_info info; 984 int error; 985 u_int32_t localization; 986 987 bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id)); 988 bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof)); 989 if (rootvp) 990 localization = VTOI(rootvp)->obj_localization; 991 else 992 localization = (u_int32_t)fhp->fid_ext << 16; 993 994 hammer_simple_transaction(&trans, (void *)mp->mnt_data); 995 996 /* 997 * Get/allocate the hammer_inode structure. The structure must be 998 * unlocked while we manipulate the related vnode to avoid a 999 * deadlock. 1000 */ 1001 ip = hammer_get_inode(&trans, NULL, info.obj_id, 1002 info.obj_asof, localization, 0, &error); 1003 if (ip == NULL) { 1004 *vpp = NULL; 1005 return(error); 1006 } 1007 error = hammer_get_vnode(ip, vpp); 1008 hammer_rel_inode(ip, 0); 1009 hammer_done_transaction(&trans); 1010 return (error); 1011 } 1012 1013 static int 1014 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam, 1015 int *exflagsp, struct ucred **credanonp) 1016 { 1017 hammer_mount_t hmp = (void *)mp->mnt_data; 1018 struct netcred *np; 1019 int error; 1020 1021 np = vfs_export_lookup(mp, &hmp->export, nam); 1022 if (np) { 1023 *exflagsp = np->netc_exflags; 1024 *credanonp = &np->netc_anon; 1025 error = 0; 1026 } else { 1027 error = EACCES; 1028 } 1029 return (error); 1030 1031 } 1032 1033 int 1034 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export) 1035 { 1036 hammer_mount_t hmp = (void *)mp->mnt_data; 1037 int error; 1038 1039 switch(op) { 1040 case MOUNTCTL_SET_EXPORT: 1041 error = vfs_export(mp, &hmp->export, export); 1042 break; 1043 default: 1044 error = EOPNOTSUPP; 1045 break; 1046 } 1047 return(error); 1048 } 1049 1050