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 35 #include <sys/mountctl.h> 36 37 #include "hammer.h" 38 39 /* 40 * NOTE! Global statistics may not be MPSAFE so HAMMER never uses them 41 * in conditionals. 42 */ 43 int hammer_supported_version = HAMMER_VOL_VERSION_DEFAULT; 44 int hammer_debug_io; 45 int hammer_debug_general; 46 int hammer_debug_inode; 47 int hammer_debug_locks; 48 int hammer_debug_btree; 49 int hammer_debug_tid; 50 int hammer_debug_recover; /* -1 will disable, +1 will force */ 51 int hammer_debug_critical; /* non-zero enter debugger on error */ 52 int hammer_cluster_enable = 1; /* enable read clustering by default */ 53 int hammer_live_dedup = 0; 54 int hammer_tdmux_ticks; 55 int hammer_count_fsyncs; 56 int hammer_count_inodes; 57 int hammer_count_iqueued; 58 int hammer_count_reclaims; 59 int hammer_count_records; 60 int hammer_count_record_datas; 61 int hammer_count_volumes; 62 int hammer_count_buffers; 63 int hammer_count_nodes; 64 int64_t hammer_count_extra_space_used; 65 int64_t hammer_stats_btree_lookups; 66 int64_t hammer_stats_btree_searches; 67 int64_t hammer_stats_btree_inserts; 68 int64_t hammer_stats_btree_deletes; 69 int64_t hammer_stats_btree_elements; 70 int64_t hammer_stats_btree_splits; 71 int64_t hammer_stats_btree_iterations; 72 int64_t hammer_stats_btree_root_iterations; 73 int64_t hammer_stats_record_iterations; 74 75 int64_t hammer_stats_file_read; 76 int64_t hammer_stats_file_write; 77 int64_t hammer_stats_file_iopsr; 78 int64_t hammer_stats_file_iopsw; 79 int64_t hammer_stats_disk_read; 80 int64_t hammer_stats_disk_write; 81 int64_t hammer_stats_inode_flushes; 82 int64_t hammer_stats_commits; 83 int64_t hammer_stats_undo; 84 int64_t hammer_stats_redo; 85 86 long hammer_count_dirtybufspace; /* global */ 87 int hammer_count_refedbufs; /* global */ 88 int hammer_count_reservations; 89 long hammer_count_io_running_read; 90 long hammer_count_io_running_write; 91 int hammer_count_io_locked; 92 long hammer_limit_dirtybufspace; /* per-mount */ 93 int hammer_limit_recs; /* as a whole XXX */ 94 int hammer_limit_inode_recs = 2048; /* per inode */ 95 int hammer_limit_reclaims; 96 int hammer_live_dedup_cache_size = DEDUP_CACHE_SIZE; 97 int hammer_limit_redo = 4096 * 1024; /* per inode */ 98 int hammer_autoflush = 500; /* auto flush (typ on reclaim) */ 99 int hammer_verify_zone; 100 int hammer_verify_data = 1; 101 int hammer_double_buffer; 102 int hammer_btree_full_undo = 1; 103 int hammer_yield_check = 16; 104 int hammer_fsync_mode = 3; 105 int64_t hammer_contention_count; 106 107 int hammer_noatime = 1; 108 TUNABLE_INT("vfs.hammer.noatime", &hammer_noatime); 109 110 /* 111 * Live dedup debug counters (sysctls are writable so that counters 112 * can be reset from userspace). 113 */ 114 int64_t hammer_live_dedup_vnode_bcmps = 0; 115 int64_t hammer_live_dedup_device_bcmps = 0; 116 int64_t hammer_live_dedup_findblk_failures = 0; 117 int64_t hammer_live_dedup_bmap_saves = 0; 118 119 120 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem"); 121 122 SYSCTL_INT(_vfs_hammer, OID_AUTO, supported_version, CTLFLAG_RD, 123 &hammer_supported_version, 0, ""); 124 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW, 125 &hammer_debug_general, 0, ""); 126 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW, 127 &hammer_debug_io, 0, ""); 128 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW, 129 &hammer_debug_inode, 0, ""); 130 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW, 131 &hammer_debug_locks, 0, ""); 132 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW, 133 &hammer_debug_btree, 0, ""); 134 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW, 135 &hammer_debug_tid, 0, ""); 136 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW, 137 &hammer_debug_recover, 0, ""); 138 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_critical, CTLFLAG_RW, 139 &hammer_debug_critical, 0, ""); 140 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW, 141 &hammer_cluster_enable, 0, ""); 142 /* 143 * 0 - live dedup is disabled 144 * 1 - dedup cache is populated on reads only 145 * 2 - dedup cache is populated on both reads and writes 146 * 147 * LIVE_DEDUP IS DISABLED PERMANENTLY! This feature appears to cause 148 * blockmap corruption over time so we've turned it off permanently. 149 */ 150 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup, CTLFLAG_RD, 151 &hammer_live_dedup, 0, "Enable live dedup (experimental)"); 152 SYSCTL_INT(_vfs_hammer, OID_AUTO, tdmux_ticks, CTLFLAG_RW, 153 &hammer_tdmux_ticks, 0, "Hammer tdmux ticks"); 154 155 SYSCTL_LONG(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW, 156 &hammer_limit_dirtybufspace, 0, ""); 157 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW, 158 &hammer_limit_recs, 0, ""); 159 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_inode_recs, CTLFLAG_RW, 160 &hammer_limit_inode_recs, 0, ""); 161 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_reclaims, CTLFLAG_RW, 162 &hammer_limit_reclaims, 0, ""); 163 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup_cache_size, CTLFLAG_RW, 164 &hammer_live_dedup_cache_size, 0, 165 "Number of cache entries"); 166 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_redo, CTLFLAG_RW, 167 &hammer_limit_redo, 0, ""); 168 169 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD, 170 &hammer_count_fsyncs, 0, ""); 171 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD, 172 &hammer_count_inodes, 0, ""); 173 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD, 174 &hammer_count_iqueued, 0, ""); 175 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaims, CTLFLAG_RD, 176 &hammer_count_reclaims, 0, ""); 177 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD, 178 &hammer_count_records, 0, ""); 179 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD, 180 &hammer_count_record_datas, 0, ""); 181 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD, 182 &hammer_count_volumes, 0, ""); 183 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD, 184 &hammer_count_buffers, 0, ""); 185 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD, 186 &hammer_count_nodes, 0, ""); 187 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD, 188 &hammer_count_extra_space_used, 0, ""); 189 190 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD, 191 &hammer_stats_btree_searches, 0, ""); 192 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD, 193 &hammer_stats_btree_lookups, 0, ""); 194 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD, 195 &hammer_stats_btree_inserts, 0, ""); 196 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD, 197 &hammer_stats_btree_deletes, 0, ""); 198 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD, 199 &hammer_stats_btree_elements, 0, ""); 200 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD, 201 &hammer_stats_btree_splits, 0, ""); 202 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD, 203 &hammer_stats_btree_iterations, 0, ""); 204 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_root_iterations, CTLFLAG_RD, 205 &hammer_stats_btree_root_iterations, 0, ""); 206 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD, 207 &hammer_stats_record_iterations, 0, ""); 208 209 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD, 210 &hammer_stats_file_read, 0, ""); 211 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD, 212 &hammer_stats_file_write, 0, ""); 213 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD, 214 &hammer_stats_file_iopsr, 0, ""); 215 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD, 216 &hammer_stats_file_iopsw, 0, ""); 217 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD, 218 &hammer_stats_disk_read, 0, ""); 219 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD, 220 &hammer_stats_disk_write, 0, ""); 221 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD, 222 &hammer_stats_inode_flushes, 0, ""); 223 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD, 224 &hammer_stats_commits, 0, ""); 225 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_undo, CTLFLAG_RD, 226 &hammer_stats_undo, 0, ""); 227 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_redo, CTLFLAG_RD, 228 &hammer_stats_redo, 0, ""); 229 230 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_vnode_bcmps, CTLFLAG_RW, 231 &hammer_live_dedup_vnode_bcmps, 0, 232 "successful vnode buffer comparisons"); 233 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_device_bcmps, CTLFLAG_RW, 234 &hammer_live_dedup_device_bcmps, 0, 235 "successful device buffer comparisons"); 236 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_findblk_failures, CTLFLAG_RW, 237 &hammer_live_dedup_findblk_failures, 0, 238 "block lookup failures for comparison"); 239 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_bmap_saves, CTLFLAG_RW, 240 &hammer_live_dedup_bmap_saves, 0, 241 "useful physical block lookups"); 242 243 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD, 244 &hammer_count_dirtybufspace, 0, ""); 245 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD, 246 &hammer_count_refedbufs, 0, ""); 247 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD, 248 &hammer_count_reservations, 0, ""); 249 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD, 250 &hammer_count_io_running_read, 0, ""); 251 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD, 252 &hammer_count_io_locked, 0, ""); 253 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD, 254 &hammer_count_io_running_write, 0, ""); 255 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW, 256 &hammer_contention_count, 0, ""); 257 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW, 258 &hammer_autoflush, 0, ""); 259 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW, 260 &hammer_verify_zone, 0, ""); 261 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW, 262 &hammer_verify_data, 0, ""); 263 SYSCTL_INT(_vfs_hammer, OID_AUTO, double_buffer, CTLFLAG_RW, 264 &hammer_double_buffer, 0, ""); 265 SYSCTL_INT(_vfs_hammer, OID_AUTO, btree_full_undo, CTLFLAG_RW, 266 &hammer_btree_full_undo, 0, ""); 267 SYSCTL_INT(_vfs_hammer, OID_AUTO, yield_check, CTLFLAG_RW, 268 &hammer_yield_check, 0, ""); 269 SYSCTL_INT(_vfs_hammer, OID_AUTO, fsync_mode, CTLFLAG_RW, 270 &hammer_fsync_mode, 0, ""); 271 272 /* KTR_INFO_MASTER(hammer); */ 273 274 /* 275 * VFS ABI 276 */ 277 static void hammer_free_hmp(struct mount *mp); 278 279 static int hammer_vfs_mount(struct mount *mp, char *path, caddr_t data, 280 struct ucred *cred); 281 static int hammer_vfs_unmount(struct mount *mp, int mntflags); 282 static int hammer_vfs_root(struct mount *mp, struct vnode **vpp); 283 static int hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, 284 struct ucred *cred); 285 static int hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, 286 struct ucred *cred); 287 static int hammer_vfs_sync(struct mount *mp, int waitfor); 288 static int hammer_vfs_vget(struct mount *mp, struct vnode *dvp, 289 ino_t ino, struct vnode **vpp); 290 static int hammer_vfs_init(struct vfsconf *conf); 291 static int hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp, 292 struct fid *fhp, struct vnode **vpp); 293 static int hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp); 294 static int hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam, 295 int *exflagsp, struct ucred **credanonp); 296 297 298 static struct vfsops hammer_vfsops = { 299 .vfs_mount = hammer_vfs_mount, 300 .vfs_unmount = hammer_vfs_unmount, 301 .vfs_root = hammer_vfs_root, 302 .vfs_statfs = hammer_vfs_statfs, 303 .vfs_statvfs = hammer_vfs_statvfs, 304 .vfs_sync = hammer_vfs_sync, 305 .vfs_vget = hammer_vfs_vget, 306 .vfs_init = hammer_vfs_init, 307 .vfs_vptofh = hammer_vfs_vptofh, 308 .vfs_fhtovp = hammer_vfs_fhtovp, 309 .vfs_checkexp = hammer_vfs_checkexp 310 }; 311 312 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", ""); 313 314 VFS_SET(hammer_vfsops, hammer, 0); 315 MODULE_VERSION(hammer, 1); 316 317 static int 318 hammer_vfs_init(struct vfsconf *conf) 319 { 320 long n; 321 322 /* 323 * Wait up to this long for an exclusive deadlock to clear 324 * before acquiring a new shared lock on the ip. The deadlock 325 * may have occured on a b-tree node related to the ip. 326 */ 327 if (hammer_tdmux_ticks == 0) 328 hammer_tdmux_ticks = hz / 5; 329 330 /* 331 * Autosize, but be careful because a hammer filesystem's 332 * reserve is partially calculated based on dirtybufspace, 333 * so we simply cannot allow it to get too large. 334 */ 335 if (hammer_limit_recs == 0) { 336 n = nbuf * 25; 337 if (n > kmalloc_limit(M_HAMMER) / 512) 338 n = kmalloc_limit(M_HAMMER) / 512; 339 if (n > 2 * 1024 * 1024) 340 n = 2 * 1024 * 1024; 341 hammer_limit_recs = (int)n; 342 } 343 if (hammer_limit_dirtybufspace == 0) { 344 hammer_limit_dirtybufspace = hidirtybufspace / 2; 345 if (hammer_limit_dirtybufspace < 1L * 1024 * 1024) 346 hammer_limit_dirtybufspace = 1024L * 1024; 347 if (hammer_limit_dirtybufspace > 1024L * 1024 * 1024) 348 hammer_limit_dirtybufspace = 1024L * 1024 * 1024; 349 } 350 351 /* 352 * The hammer_inode structure detaches from the vnode on reclaim. 353 * This limits the number of inodes in this state to prevent a 354 * memory pool blowout. 355 */ 356 if (hammer_limit_reclaims == 0) 357 hammer_limit_reclaims = desiredvnodes / 10; 358 359 return(0); 360 } 361 362 static int 363 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data, 364 struct ucred *cred) 365 { 366 struct hammer_mount_info info; 367 hammer_mount_t hmp; 368 hammer_volume_t rootvol; 369 struct vnode *rootvp; 370 struct vnode *devvp = NULL; 371 const char *upath; /* volume name in userspace */ 372 char *path; /* volume name in system space */ 373 int error; 374 int i; 375 int master_id; 376 int nvolumes; 377 char *next_volume_ptr = NULL; 378 379 if (hammer_noatime) { 380 /* Force noatime */ 381 mp->mnt_flag |= MNT_NOATIME; 382 } 383 384 /* 385 * Accept hammer_mount_info. mntpt is NULL for root mounts at boot. 386 */ 387 if (mntpt == NULL) { 388 bzero(&info, sizeof(info)); 389 info.asof = 0; 390 info.hflags = 0; 391 info.nvolumes = 1; 392 393 next_volume_ptr = mp->mnt_stat.f_mntfromname; 394 395 /* Count number of volumes separated by ':' */ 396 for (char *p = next_volume_ptr; *p != '\0'; ++p) { 397 if (*p == ':') { 398 ++info.nvolumes; 399 } 400 } 401 402 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */ 403 } else { 404 if ((error = copyin(data, &info, sizeof(info))) != 0) 405 return (error); 406 } 407 408 /* 409 * updating or new mount 410 */ 411 if (mp->mnt_flag & MNT_UPDATE) { 412 hmp = (void *)mp->mnt_data; 413 KKASSERT(hmp != NULL); 414 } else { 415 if (info.nvolumes <= 0 || info.nvolumes > HAMMER_MAX_VOLUMES) 416 return (EINVAL); 417 hmp = NULL; 418 } 419 420 /* 421 * master-id validation. The master id may not be changed by a 422 * mount update. 423 */ 424 if (info.hflags & HMNT_MASTERID || info.hflags & HMNT_NOMIRROR) { 425 if (hmp && hmp->master_id != info.master_id) { 426 hkprintf("cannot change master id with mount update\n"); 427 return(EINVAL); 428 } 429 master_id = info.master_id; 430 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS) 431 return (EINVAL); 432 } else { 433 if (hmp) 434 master_id = hmp->master_id; 435 else 436 master_id = 0; 437 } 438 439 /* 440 * Internal mount data structure 441 */ 442 if (hmp == NULL) { 443 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO); 444 mp->mnt_data = (qaddr_t)hmp; 445 hmp->mp = mp; 446 447 /* 448 * Make sure kmalloc type limits are set appropriately. 449 * 450 * Our inode kmalloc group is sized based on maxvnodes 451 * (controlled by the system, not us). 452 */ 453 kmalloc_create(&hmp->m_misc, "HAMMER-others"); 454 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes"); 455 456 kmalloc_raise_limit(hmp->m_inodes, 0); /* unlimited */ 457 458 hmp->root_btree_beg.localization = 0x00000000U; 459 hmp->root_btree_beg.obj_id = -0x8000000000000000LL; 460 hmp->root_btree_beg.key = -0x8000000000000000LL; 461 hmp->root_btree_beg.create_tid = 1; 462 hmp->root_btree_beg.delete_tid = 1; 463 hmp->root_btree_beg.rec_type = 0; 464 hmp->root_btree_beg.obj_type = 0; 465 hmp->root_btree_beg.btype = HAMMER_BTREE_TYPE_NONE; 466 467 hmp->root_btree_end.localization = 0xFFFFFFFFU; 468 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL; 469 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL; 470 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL; 471 hmp->root_btree_end.delete_tid = 0; /* special case */ 472 hmp->root_btree_end.rec_type = 0xFFFFU; 473 hmp->root_btree_end.obj_type = 0; 474 hmp->root_btree_end.btype = HAMMER_BTREE_TYPE_NONE; 475 476 hmp->krate.freq = 1; /* maximum reporting rate (hz) */ 477 hmp->krate.count = -16; /* initial burst */ 478 hmp->kdiag.freq = 1; /* maximum reporting rate (hz) */ 479 hmp->kdiag.count = -16; /* initial burst */ 480 481 hmp->sync_lock.refs = 1; 482 hmp->undo_lock.refs = 1; 483 hmp->blkmap_lock.refs = 1; 484 hmp->snapshot_lock.refs = 1; 485 hmp->volume_lock.refs = 1; 486 487 TAILQ_INIT(&hmp->delay_list); 488 TAILQ_INIT(&hmp->flush_group_list); 489 TAILQ_INIT(&hmp->objid_cache_list); 490 TAILQ_INIT(&hmp->undo_lru_list); 491 TAILQ_INIT(&hmp->reclaim_list); 492 493 RB_INIT(&hmp->rb_dedup_crc_root); 494 RB_INIT(&hmp->rb_dedup_off_root); 495 TAILQ_INIT(&hmp->dedup_lru_list); 496 } 497 hmp->hflags &= ~HMNT_USERFLAGS; 498 hmp->hflags |= info.hflags & HMNT_USERFLAGS; 499 500 hmp->master_id = master_id; 501 502 if (info.asof) { 503 mp->mnt_flag |= MNT_RDONLY; 504 hmp->asof = info.asof; 505 } else { 506 hmp->asof = HAMMER_MAX_TID; 507 } 508 509 hmp->volume_to_remove = -1; 510 511 /* 512 * Re-open read-write if originally read-only, or vise-versa. 513 * 514 * When going from read-only to read-write execute the stage2 515 * recovery if it has not already been run. 516 */ 517 if (mp->mnt_flag & MNT_UPDATE) { 518 lwkt_gettoken(&hmp->fs_token); 519 error = 0; 520 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) { 521 hkprintf("read-only -> read-write\n"); 522 hmp->ronly = 0; 523 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL, 524 hammer_adjust_volume_mode, NULL); 525 rootvol = hammer_get_root_volume(hmp, &error); 526 if (rootvol) { 527 hammer_recover_flush_buffers(hmp, rootvol, 1); 528 error = hammer_recover_stage2(hmp, rootvol); 529 bcopy(rootvol->ondisk->vol0_blockmap, 530 hmp->blockmap, 531 sizeof(hmp->blockmap)); 532 hammer_rel_volume(rootvol, 0); 533 } 534 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL, 535 hammer_reload_inode, NULL); 536 /* kernel clears MNT_RDONLY */ 537 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) { 538 hkprintf("read-write -> read-only\n"); 539 hmp->ronly = 1; /* messy */ 540 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL, 541 hammer_reload_inode, NULL); 542 hmp->ronly = 0; 543 hammer_flusher_sync(hmp); 544 hammer_flusher_sync(hmp); 545 hammer_flusher_sync(hmp); 546 hmp->ronly = 1; 547 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL, 548 hammer_adjust_volume_mode, NULL); 549 } 550 lwkt_reltoken(&hmp->fs_token); 551 return(error); 552 } 553 554 RB_INIT(&hmp->rb_vols_root); 555 RB_INIT(&hmp->rb_inos_root); 556 RB_INIT(&hmp->rb_redo_root); 557 RB_INIT(&hmp->rb_nods_root); 558 RB_INIT(&hmp->rb_undo_root); 559 RB_INIT(&hmp->rb_resv_root); 560 RB_INIT(&hmp->rb_bufs_root); 561 RB_INIT(&hmp->rb_pfsm_root); 562 563 hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0); 564 565 RB_INIT(&hmp->volu_root); 566 RB_INIT(&hmp->undo_root); 567 RB_INIT(&hmp->data_root); 568 RB_INIT(&hmp->meta_root); 569 RB_INIT(&hmp->lose_root); 570 TAILQ_INIT(&hmp->iorun_list); 571 572 lwkt_token_init(&hmp->fs_token, "hammerfs"); 573 lwkt_token_init(&hmp->io_token, "hammerio"); 574 575 lwkt_gettoken(&hmp->fs_token); 576 577 /* 578 * Load volumes 579 */ 580 path = objcache_get(namei_oc, M_WAITOK); 581 hmp->nvolumes = -1; 582 for (i = 0; i < info.nvolumes; ++i) { 583 if (mntpt == NULL) { 584 /* 585 * Root mount. 586 */ 587 KKASSERT(next_volume_ptr != NULL); 588 strcpy(path, ""); 589 if (*next_volume_ptr != '/') { 590 /* relative path */ 591 strcpy(path, "/dev/"); 592 } 593 int k; 594 for (k = strlen(path); k < MAXPATHLEN-1; ++k) { 595 if (*next_volume_ptr == '\0') { 596 break; 597 } else if (*next_volume_ptr == ':') { 598 ++next_volume_ptr; 599 break; 600 } else { 601 path[k] = *next_volume_ptr; 602 ++next_volume_ptr; 603 } 604 } 605 path[k] = '\0'; 606 607 error = 0; 608 cdev_t dev = kgetdiskbyname(path); 609 error = bdevvp(dev, &devvp); 610 if (error) { 611 hdkprintf("can't find devvp\n"); 612 } 613 } else { 614 error = copyin(&info.volumes[i], &upath, 615 sizeof(char *)); 616 if (error == 0) 617 error = copyinstr(upath, path, 618 MAXPATHLEN, NULL); 619 } 620 if (error == 0) 621 error = hammer_install_volume(hmp, path, devvp, NULL); 622 if (error) 623 break; 624 } 625 objcache_put(namei_oc, path); 626 627 /* 628 * Make sure we found a root volume 629 */ 630 if (hmp->rootvol == NULL) { 631 if (error == EBUSY) { 632 hdkprintf("The volumes are probably mounted\n"); 633 } else { 634 hdkprintf("No root volume found!\n"); 635 error = EINVAL; 636 } 637 goto failed; 638 } 639 640 /* 641 * Check that all required volumes are available 642 */ 643 if (error == 0 && hammer_mountcheck_volumes(hmp)) { 644 hdkprintf("Missing volumes, cannot mount!\n"); 645 error = EINVAL; 646 goto failed; 647 } 648 649 /* 650 * Other errors 651 */ 652 if (error) { 653 hdkprintf("Failed to load volumes!\n"); 654 goto failed; 655 } 656 657 nvolumes = hammer_get_installed_volumes(hmp); 658 if (hmp->nvolumes != nvolumes) { 659 hdkprintf("volume header says %d volumes, but %d installed\n", 660 hmp->nvolumes, nvolumes); 661 error = EINVAL; 662 goto failed; 663 } 664 665 /* 666 * No errors, setup enough of the mount point so we can lookup the 667 * root vnode. 668 */ 669 mp->mnt_iosize_max = MAXPHYS; 670 mp->mnt_kern_flag |= MNTK_FSMID; 671 mp->mnt_kern_flag |= MNTK_THR_SYNC; /* new vsyncscan semantics */ 672 673 /* 674 * MPSAFE code. Note that VOPs and VFSops which are not MPSAFE 675 * will acquire a per-mount token prior to entry and release it 676 * on return. 677 */ 678 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE; 679 680 /* 681 * note: f_iosize is used by vnode_pager_haspage() when constructing 682 * its VOP_BMAP call. 683 */ 684 mp->mnt_stat.f_iosize = HAMMER_BUFSIZE; 685 mp->mnt_stat.f_bsize = HAMMER_BUFSIZE; 686 687 mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE; 688 mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE; 689 690 mp->mnt_maxsymlinklen = 255; 691 mp->mnt_flag |= MNT_LOCAL; 692 693 vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops); 694 vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops); 695 vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops); 696 697 /* 698 * The root volume's ondisk pointer is only valid if we hold a 699 * reference to it. 700 */ 701 rootvol = hammer_get_root_volume(hmp, &error); 702 if (error) 703 goto failed; 704 705 /* 706 * Perform any necessary UNDO operations. The recovery code does 707 * call hammer_undo_lookup() so we have to pre-cache the blockmap, 708 * and then re-copy it again after recovery is complete. 709 * 710 * If this is a read-only mount the UNDO information is retained 711 * in memory in the form of dirty buffer cache buffers, and not 712 * written back to the media. 713 */ 714 bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap, 715 sizeof(hmp->blockmap)); 716 717 /* 718 * Check filesystem version 719 */ 720 hmp->version = rootvol->ondisk->vol_version; 721 if (hmp->version < HAMMER_VOL_VERSION_MIN || 722 hmp->version > HAMMER_VOL_VERSION_MAX) { 723 hkprintf("mount unsupported fs version %d\n", hmp->version); 724 error = ERANGE; 725 goto done; 726 } 727 728 /* 729 * The undo_rec_limit limits the size of flush groups to avoid 730 * blowing out the UNDO FIFO. This calculation is typically in 731 * the tens of thousands and is designed primarily when small 732 * HAMMER filesystems are created. 733 */ 734 hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100; 735 if (hammer_debug_general & 0x0001) 736 hkprintf("undo_rec_limit %d\n", hmp->undo_rec_limit); 737 738 /* 739 * NOTE: Recover stage1 not only handles meta-data recovery, it 740 * also sets hmp->undo_seqno for HAMMER VERSION 4+ filesystems. 741 */ 742 error = hammer_recover_stage1(hmp, rootvol); 743 if (error) { 744 kprintf("Failed to recover HAMMER filesystem on mount\n"); 745 goto done; 746 } 747 748 /* 749 * Finish setup now that we have a good root volume. 750 * vol_name is a filesystem label string. 751 */ 752 ksnprintf(mp->mnt_stat.f_mntfromname, 753 sizeof(mp->mnt_stat.f_mntfromname), "%s", 754 rootvol->ondisk->vol_name); 755 mp->mnt_stat.f_fsid.val[0] = 756 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8); 757 mp->mnt_stat.f_fsid.val[1] = 758 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8); 759 mp->mnt_stat.f_fsid.val[1] &= HAMMER_LOCALIZE_MASK; 760 761 mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid; 762 mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid, 763 sizeof(mp->mnt_vstat.f_fsid_uuid)); 764 765 /* 766 * Certain often-modified fields in the root volume are cached in 767 * the hammer_mount structure so we do not have to generate lots 768 * of little UNDO structures for them. 769 * 770 * Recopy after recovery. This also has the side effect of 771 * setting our cached undo FIFO's first_offset, which serves to 772 * placemark the FIFO start for the NEXT flush cycle while the 773 * on-disk first_offset represents the LAST flush cycle. 774 */ 775 hmp->next_tid = rootvol->ondisk->vol0_next_tid; 776 hmp->flush_tid1 = hmp->next_tid; 777 hmp->flush_tid2 = hmp->next_tid; 778 bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap, 779 sizeof(hmp->blockmap)); 780 hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks; 781 782 hammer_flusher_create(hmp); 783 784 /* 785 * Locate the root directory with an obj_id of 1. 786 */ 787 error = hammer_vfs_root(mp, &rootvp); 788 if (error) 789 goto done; 790 vput(rootvp); 791 if (hmp->ronly == 0) 792 error = hammer_recover_stage2(hmp, rootvol); 793 794 /* 795 * If the stage2 recovery fails be sure to clean out all cached 796 * vnodes before throwing away the mount structure or bad things 797 * will happen. 798 */ 799 if (error) 800 vflush(mp, 0, 0); 801 802 done: 803 if ((mp->mnt_flag & MNT_UPDATE) == 0) { 804 /* New mount */ 805 806 /* Populate info for mount point (NULL pad)*/ 807 bzero(mp->mnt_stat.f_mntonname, MNAMELEN); 808 size_t size; 809 if (mntpt) { 810 copyinstr(mntpt, mp->mnt_stat.f_mntonname, 811 MNAMELEN -1, &size); 812 } else { /* Root mount */ 813 mp->mnt_stat.f_mntonname[0] = '/'; 814 } 815 } 816 (void)VFS_STATFS(mp, &mp->mnt_stat, cred); 817 hammer_rel_volume(rootvol, 0); 818 failed: 819 /* 820 * Cleanup and return. 821 */ 822 if (error) { 823 /* called with fs_token held */ 824 hammer_free_hmp(mp); 825 } else { 826 lwkt_reltoken(&hmp->fs_token); 827 } 828 return (error); 829 } 830 831 static int 832 hammer_vfs_unmount(struct mount *mp, int mntflags) 833 { 834 hammer_mount_t hmp = (void *)mp->mnt_data; 835 int flags; 836 int error; 837 838 /* 839 * Clean out the vnodes 840 */ 841 lwkt_gettoken(&hmp->fs_token); 842 flags = 0; 843 if (mntflags & MNT_FORCE) 844 flags |= FORCECLOSE; 845 error = vflush(mp, 0, flags); 846 847 /* 848 * Clean up the internal mount structure and related entities. This 849 * may issue I/O. 850 */ 851 if (error == 0) { 852 /* called with fs_token held */ 853 hammer_free_hmp(mp); 854 } else { 855 lwkt_reltoken(&hmp->fs_token); 856 } 857 return(error); 858 } 859 860 /* 861 * Clean up the internal mount structure and disassociate it from the mount. 862 * This may issue I/O. 863 * 864 * Called with fs_token held. 865 */ 866 static void 867 hammer_free_hmp(struct mount *mp) 868 { 869 hammer_mount_t hmp = (void *)mp->mnt_data; 870 hammer_flush_group_t flg; 871 872 /* 873 * Flush anything dirty. This won't even run if the 874 * filesystem errored-out. 875 */ 876 hammer_flush_dirty(hmp, 30); 877 878 /* 879 * If the mount had a critical error we have to destroy any 880 * remaining inodes before we can finish cleaning up the flusher. 881 */ 882 if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) { 883 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL, 884 hammer_destroy_inode_callback, NULL); 885 } 886 887 /* 888 * There shouldn't be any inodes left now and any left over 889 * flush groups should now be empty. 890 */ 891 KKASSERT(RB_EMPTY(&hmp->rb_inos_root)); 892 while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) { 893 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry); 894 KKASSERT(RB_EMPTY(&flg->flush_tree)); 895 if (flg->refs) { 896 hkprintf("Warning, flush_group %p was " 897 "not empty on umount!\n", flg); 898 } 899 kfree(flg, hmp->m_misc); 900 } 901 902 /* 903 * We can finally destroy the flusher 904 */ 905 hammer_flusher_destroy(hmp); 906 907 /* 908 * We may have held recovered buffers due to a read-only mount. 909 * These must be discarded. 910 */ 911 if (hmp->ronly) 912 hammer_recover_flush_buffers(hmp, NULL, -1); 913 914 /* 915 * Unload buffers and then volumes 916 */ 917 RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL, 918 hammer_unload_buffer, NULL); 919 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL, 920 hammer_unload_volume, NULL); 921 922 mp->mnt_data = NULL; 923 mp->mnt_flag &= ~MNT_LOCAL; 924 hmp->mp = NULL; 925 hammer_destroy_objid_cache(hmp); 926 hammer_destroy_dedup_cache(hmp); 927 if (hmp->dedup_free_cache != NULL) { 928 kfree(hmp->dedup_free_cache, hmp->m_misc); 929 hmp->dedup_free_cache = NULL; 930 } 931 kmalloc_destroy(&hmp->m_misc); 932 kmalloc_destroy(&hmp->m_inodes); 933 lwkt_reltoken(&hmp->fs_token); 934 kfree(hmp, M_HAMMER); 935 } 936 937 /* 938 * Report critical errors. ip may be NULL. 939 */ 940 void 941 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip, 942 int error, const char *msg) 943 { 944 hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR; 945 946 hmkrateprintf(&hmp->krate, hmp, 947 "Critical error inode=%jd error=%d %s\n", 948 (intmax_t)(ip ? ip->obj_id : -1), 949 error, msg); 950 951 if (hmp->ronly == 0) { 952 hmp->ronly = 2; /* special errored read-only mode */ 953 hmp->mp->mnt_flag |= MNT_RDONLY; 954 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL, 955 hammer_adjust_volume_mode, NULL); 956 hmkprintf(hmp, "Forcing read-only mode\n"); 957 } 958 hmp->error = error; 959 if (hammer_debug_critical) 960 Debugger("Entering debugger"); 961 } 962 963 964 /* 965 * Obtain a vnode for the specified inode number. An exclusively locked 966 * vnode is returned. 967 */ 968 int 969 hammer_vfs_vget(struct mount *mp, struct vnode *dvp, 970 ino_t ino, struct vnode **vpp) 971 { 972 struct hammer_transaction trans; 973 struct hammer_mount *hmp = (void *)mp->mnt_data; 974 struct hammer_inode *ip; 975 int error; 976 uint32_t localization; 977 978 lwkt_gettoken(&hmp->fs_token); 979 hammer_simple_transaction(&trans, hmp); 980 981 /* 982 * If a directory vnode is supplied (mainly NFS) then we can acquire 983 * the PFS domain from it. Otherwise we would only be able to vget 984 * inodes in the root PFS. 985 */ 986 if (dvp) { 987 localization = HAMMER_DEF_LOCALIZATION | 988 VTOI(dvp)->obj_localization; 989 } else { 990 localization = HAMMER_DEF_LOCALIZATION; 991 } 992 993 /* 994 * Lookup the requested HAMMER inode. The structure must be 995 * left unlocked while we manipulate the related vnode to avoid 996 * a deadlock. 997 */ 998 ip = hammer_get_inode(&trans, NULL, ino, 999 hmp->asof, localization, 1000 0, &error); 1001 if (ip == NULL) { 1002 *vpp = NULL; 1003 } else { 1004 error = hammer_get_vnode(ip, vpp); 1005 hammer_rel_inode(ip, 0); 1006 } 1007 hammer_done_transaction(&trans); 1008 lwkt_reltoken(&hmp->fs_token); 1009 return (error); 1010 } 1011 1012 /* 1013 * Return the root vnode for the filesystem. 1014 * 1015 * HAMMER stores the root vnode in the hammer_mount structure so 1016 * getting it is easy. 1017 */ 1018 static int 1019 hammer_vfs_root(struct mount *mp, struct vnode **vpp) 1020 { 1021 int error; 1022 1023 error = hammer_vfs_vget(mp, NULL, HAMMER_OBJID_ROOT, vpp); 1024 return (error); 1025 } 1026 1027 static int 1028 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred) 1029 { 1030 struct hammer_mount *hmp = (void *)mp->mnt_data; 1031 hammer_volume_t volume; 1032 hammer_volume_ondisk_t ondisk; 1033 int error; 1034 int64_t bfree; 1035 int64_t breserved; 1036 1037 lwkt_gettoken(&hmp->fs_token); 1038 volume = hammer_get_root_volume(hmp, &error); 1039 if (error) { 1040 lwkt_reltoken(&hmp->fs_token); 1041 return(error); 1042 } 1043 ondisk = volume->ondisk; 1044 1045 /* 1046 * Basic stats 1047 */ 1048 _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved); 1049 mp->mnt_stat.f_files = ondisk->vol0_stat_inodes; 1050 bfree = ondisk->vol0_stat_freebigblocks * HAMMER_BIGBLOCK_SIZE; 1051 hammer_rel_volume(volume, 0); 1052 1053 mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE; 1054 mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree; 1055 if (mp->mnt_stat.f_files < 0) 1056 mp->mnt_stat.f_files = 0; 1057 1058 *sbp = mp->mnt_stat; 1059 lwkt_reltoken(&hmp->fs_token); 1060 return(0); 1061 } 1062 1063 static int 1064 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred) 1065 { 1066 struct hammer_mount *hmp = (void *)mp->mnt_data; 1067 hammer_volume_t volume; 1068 hammer_volume_ondisk_t ondisk; 1069 int error; 1070 int64_t bfree; 1071 int64_t breserved; 1072 1073 lwkt_gettoken(&hmp->fs_token); 1074 volume = hammer_get_root_volume(hmp, &error); 1075 if (error) { 1076 lwkt_reltoken(&hmp->fs_token); 1077 return(error); 1078 } 1079 ondisk = volume->ondisk; 1080 1081 /* 1082 * Basic stats 1083 */ 1084 _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved); 1085 mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes; 1086 bfree = ondisk->vol0_stat_freebigblocks * HAMMER_BIGBLOCK_SIZE; 1087 hammer_rel_volume(volume, 0); 1088 1089 mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE; 1090 mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree; 1091 if (mp->mnt_vstat.f_files < 0) 1092 mp->mnt_vstat.f_files = 0; 1093 *sbp = mp->mnt_vstat; 1094 lwkt_reltoken(&hmp->fs_token); 1095 return(0); 1096 } 1097 1098 /* 1099 * Sync the filesystem. Currently we have to run it twice, the second 1100 * one will advance the undo start index to the end index, so if a crash 1101 * occurs no undos will be run on mount. 1102 * 1103 * We do not sync the filesystem if we are called from a panic. If we did 1104 * we might end up blowing up a sync that was already in progress. 1105 */ 1106 static int 1107 hammer_vfs_sync(struct mount *mp, int waitfor) 1108 { 1109 struct hammer_mount *hmp = (void *)mp->mnt_data; 1110 int error; 1111 1112 lwkt_gettoken(&hmp->fs_token); 1113 if (panicstr == NULL) { 1114 error = hammer_sync_hmp(hmp, waitfor); 1115 } else { 1116 error = EIO; 1117 } 1118 lwkt_reltoken(&hmp->fs_token); 1119 return (error); 1120 } 1121 1122 /* 1123 * Convert a vnode to a file handle. 1124 * 1125 * Accesses read-only fields on already-referenced structures so 1126 * no token is needed. 1127 */ 1128 static int 1129 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp) 1130 { 1131 hammer_inode_t ip; 1132 1133 KKASSERT(MAXFIDSZ >= 16); 1134 ip = VTOI(vp); 1135 fhp->fid_len = offsetof(struct fid, fid_data[16]); 1136 fhp->fid_ext = lo_to_pfs(ip->obj_localization); 1137 bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id)); 1138 bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof)); 1139 return(0); 1140 } 1141 1142 1143 /* 1144 * Convert a file handle back to a vnode. 1145 * 1146 * Use rootvp to enforce PFS isolation when a PFS is exported via a 1147 * null mount. 1148 */ 1149 static int 1150 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp, 1151 struct fid *fhp, struct vnode **vpp) 1152 { 1153 hammer_mount_t hmp = (void *)mp->mnt_data; 1154 struct hammer_transaction trans; 1155 struct hammer_inode *ip; 1156 struct hammer_inode_info info; 1157 int error; 1158 uint32_t localization; 1159 1160 bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id)); 1161 bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof)); 1162 if (rootvp) 1163 localization = VTOI(rootvp)->obj_localization; 1164 else 1165 localization = pfs_to_lo(fhp->fid_ext); 1166 1167 lwkt_gettoken(&hmp->fs_token); 1168 hammer_simple_transaction(&trans, hmp); 1169 1170 /* 1171 * Get/allocate the hammer_inode structure. The structure must be 1172 * unlocked while we manipulate the related vnode to avoid a 1173 * deadlock. 1174 */ 1175 ip = hammer_get_inode(&trans, NULL, info.obj_id, 1176 info.obj_asof, localization, 0, &error); 1177 if (ip) { 1178 error = hammer_get_vnode(ip, vpp); 1179 hammer_rel_inode(ip, 0); 1180 } else { 1181 *vpp = NULL; 1182 } 1183 hammer_done_transaction(&trans); 1184 lwkt_reltoken(&hmp->fs_token); 1185 return (error); 1186 } 1187 1188 static int 1189 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam, 1190 int *exflagsp, struct ucred **credanonp) 1191 { 1192 hammer_mount_t hmp = (void *)mp->mnt_data; 1193 struct netcred *np; 1194 int error; 1195 1196 lwkt_gettoken(&hmp->fs_token); 1197 np = vfs_export_lookup(mp, &hmp->export, nam); 1198 if (np) { 1199 *exflagsp = np->netc_exflags; 1200 *credanonp = &np->netc_anon; 1201 error = 0; 1202 } else { 1203 error = EACCES; 1204 } 1205 lwkt_reltoken(&hmp->fs_token); 1206 return (error); 1207 1208 } 1209 1210 int 1211 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export) 1212 { 1213 hammer_mount_t hmp = (void *)mp->mnt_data; 1214 int error; 1215 1216 lwkt_gettoken(&hmp->fs_token); 1217 1218 switch(op) { 1219 case MOUNTCTL_SET_EXPORT: 1220 error = vfs_export(mp, &hmp->export, export); 1221 break; 1222 default: 1223 error = EOPNOTSUPP; 1224 break; 1225 } 1226 lwkt_reltoken(&hmp->fs_token); 1227 1228 return(error); 1229 } 1230 1231