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