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