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/param.h> 36 #include <sys/systm.h> 37 #include <sys/kernel.h> 38 #include <sys/vnode.h> 39 #include <sys/mount.h> 40 #include <sys/malloc.h> 41 #include <sys/nlookup.h> 42 #include <sys/fcntl.h> 43 #include <sys/sysctl.h> 44 #include <sys/buf.h> 45 #include <sys/buf2.h> 46 #include "hammer.h" 47 48 /* 49 * NOTE! Global statistics may not be MPSAFE so HAMMER never uses them 50 * in conditionals. 51 */ 52 int hammer_supported_version = HAMMER_VOL_VERSION_DEFAULT; 53 int hammer_debug_io; 54 int hammer_debug_general; 55 int hammer_debug_debug = 1; /* medium-error panics */ 56 int hammer_debug_inode; 57 int hammer_debug_locks; 58 int hammer_debug_btree; 59 int hammer_debug_tid; 60 int hammer_debug_recover; /* -1 will disable, +1 will force */ 61 int hammer_debug_recover_faults; 62 int hammer_debug_critical; /* non-zero enter debugger on error */ 63 int hammer_cluster_enable = 1; /* enable read clustering by default */ 64 int hammer_live_dedup = 0; 65 int hammer_tdmux_ticks; 66 int hammer_count_fsyncs; 67 int hammer_count_inodes; 68 int hammer_count_iqueued; 69 int hammer_count_reclaims; 70 int hammer_count_records; 71 int hammer_count_record_datas; 72 int hammer_count_volumes; 73 int hammer_count_buffers; 74 int hammer_count_nodes; 75 int64_t hammer_count_extra_space_used; 76 int64_t hammer_stats_btree_lookups; 77 int64_t hammer_stats_btree_searches; 78 int64_t hammer_stats_btree_inserts; 79 int64_t hammer_stats_btree_deletes; 80 int64_t hammer_stats_btree_elements; 81 int64_t hammer_stats_btree_splits; 82 int64_t hammer_stats_btree_iterations; 83 int64_t hammer_stats_btree_root_iterations; 84 int64_t hammer_stats_record_iterations; 85 86 int64_t hammer_stats_file_read; 87 int64_t hammer_stats_file_write; 88 int64_t hammer_stats_file_iopsr; 89 int64_t hammer_stats_file_iopsw; 90 int64_t hammer_stats_disk_read; 91 int64_t hammer_stats_disk_write; 92 int64_t hammer_stats_inode_flushes; 93 int64_t hammer_stats_commits; 94 int64_t hammer_stats_undo; 95 int64_t hammer_stats_redo; 96 97 long hammer_count_dirtybufspace; /* global */ 98 int hammer_count_refedbufs; /* global */ 99 int hammer_count_reservations; 100 long hammer_count_io_running_read; 101 long hammer_count_io_running_write; 102 int hammer_count_io_locked; 103 long hammer_limit_dirtybufspace; /* per-mount */ 104 int hammer_limit_recs; /* as a whole XXX */ 105 int hammer_limit_inode_recs = 2048; /* per inode */ 106 int hammer_limit_reclaims; 107 int hammer_live_dedup_cache_size = DEDUP_CACHE_SIZE; 108 int hammer_limit_redo = 4096 * 1024; /* per inode */ 109 int hammer_autoflush = 500; /* auto flush (typ on reclaim) */ 110 int hammer_bio_count; 111 int hammer_verify_zone; 112 int hammer_verify_data = 1; 113 int hammer_write_mode; 114 int hammer_double_buffer; 115 int hammer_yield_check = 16; 116 int hammer_fsync_mode = 3; 117 int64_t hammer_contention_count; 118 int64_t hammer_zone_limit; 119 120 /* 121 * Live dedup debug counters (sysctls are writable so that counters 122 * can be reset from userspace). 123 */ 124 int64_t hammer_live_dedup_vnode_bcmps = 0; 125 int64_t hammer_live_dedup_device_bcmps = 0; 126 int64_t hammer_live_dedup_findblk_failures = 0; 127 int64_t hammer_live_dedup_bmap_saves = 0; 128 129 130 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem"); 131 132 SYSCTL_INT(_vfs_hammer, OID_AUTO, supported_version, CTLFLAG_RD, 133 &hammer_supported_version, 0, ""); 134 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW, 135 &hammer_debug_general, 0, ""); 136 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW, 137 &hammer_debug_io, 0, ""); 138 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW, 139 &hammer_debug_debug, 0, ""); 140 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW, 141 &hammer_debug_inode, 0, ""); 142 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW, 143 &hammer_debug_locks, 0, ""); 144 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW, 145 &hammer_debug_btree, 0, ""); 146 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW, 147 &hammer_debug_tid, 0, ""); 148 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW, 149 &hammer_debug_recover, 0, ""); 150 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW, 151 &hammer_debug_recover_faults, 0, ""); 152 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_critical, CTLFLAG_RW, 153 &hammer_debug_critical, 0, ""); 154 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW, 155 &hammer_cluster_enable, 0, ""); 156 /* 157 * 0 - live dedup is disabled 158 * 1 - dedup cache is populated on reads only 159 * 2 - dedup cache is populated on both reads and writes 160 */ 161 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup, CTLFLAG_RD, 162 &hammer_live_dedup, 0, "Enable live dedup (experimental)"); 163 SYSCTL_INT(_vfs_hammer, OID_AUTO, tdmux_ticks, CTLFLAG_RW, 164 &hammer_tdmux_ticks, 0, "Hammer tdmux ticks"); 165 166 SYSCTL_LONG(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW, 167 &hammer_limit_dirtybufspace, 0, ""); 168 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW, 169 &hammer_limit_recs, 0, ""); 170 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_inode_recs, CTLFLAG_RW, 171 &hammer_limit_inode_recs, 0, ""); 172 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_reclaims, CTLFLAG_RW, 173 &hammer_limit_reclaims, 0, ""); 174 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup_cache_size, CTLFLAG_RW, 175 &hammer_live_dedup_cache_size, 0, 176 "Number of cache entries"); 177 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_redo, CTLFLAG_RW, 178 &hammer_limit_redo, 0, ""); 179 180 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD, 181 &hammer_count_fsyncs, 0, ""); 182 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD, 183 &hammer_count_inodes, 0, ""); 184 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD, 185 &hammer_count_iqueued, 0, ""); 186 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaims, CTLFLAG_RD, 187 &hammer_count_reclaims, 0, ""); 188 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD, 189 &hammer_count_records, 0, ""); 190 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD, 191 &hammer_count_record_datas, 0, ""); 192 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD, 193 &hammer_count_volumes, 0, ""); 194 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD, 195 &hammer_count_buffers, 0, ""); 196 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD, 197 &hammer_count_nodes, 0, ""); 198 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD, 199 &hammer_count_extra_space_used, 0, ""); 200 201 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD, 202 &hammer_stats_btree_searches, 0, ""); 203 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD, 204 &hammer_stats_btree_lookups, 0, ""); 205 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD, 206 &hammer_stats_btree_inserts, 0, ""); 207 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD, 208 &hammer_stats_btree_deletes, 0, ""); 209 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD, 210 &hammer_stats_btree_elements, 0, ""); 211 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD, 212 &hammer_stats_btree_splits, 0, ""); 213 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD, 214 &hammer_stats_btree_iterations, 0, ""); 215 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_root_iterations, CTLFLAG_RD, 216 &hammer_stats_btree_root_iterations, 0, ""); 217 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD, 218 &hammer_stats_record_iterations, 0, ""); 219 220 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD, 221 &hammer_stats_file_read, 0, ""); 222 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD, 223 &hammer_stats_file_write, 0, ""); 224 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD, 225 &hammer_stats_file_iopsr, 0, ""); 226 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD, 227 &hammer_stats_file_iopsw, 0, ""); 228 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD, 229 &hammer_stats_disk_read, 0, ""); 230 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD, 231 &hammer_stats_disk_write, 0, ""); 232 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD, 233 &hammer_stats_inode_flushes, 0, ""); 234 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD, 235 &hammer_stats_commits, 0, ""); 236 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_undo, CTLFLAG_RD, 237 &hammer_stats_undo, 0, ""); 238 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_redo, CTLFLAG_RD, 239 &hammer_stats_redo, 0, ""); 240 241 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_vnode_bcmps, CTLFLAG_RW, 242 &hammer_live_dedup_vnode_bcmps, 0, 243 "successful vnode buffer comparisons"); 244 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_device_bcmps, CTLFLAG_RW, 245 &hammer_live_dedup_device_bcmps, 0, 246 "successful device buffer comparisons"); 247 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_findblk_failures, CTLFLAG_RW, 248 &hammer_live_dedup_findblk_failures, 0, 249 "block lookup failures for comparison"); 250 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_bmap_saves, CTLFLAG_RW, 251 &hammer_live_dedup_bmap_saves, 0, 252 "useful physical block lookups"); 253 254 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD, 255 &hammer_count_dirtybufspace, 0, ""); 256 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD, 257 &hammer_count_refedbufs, 0, ""); 258 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD, 259 &hammer_count_reservations, 0, ""); 260 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD, 261 &hammer_count_io_running_read, 0, ""); 262 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD, 263 &hammer_count_io_locked, 0, ""); 264 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD, 265 &hammer_count_io_running_write, 0, ""); 266 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW, 267 &hammer_zone_limit, 0, ""); 268 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW, 269 &hammer_contention_count, 0, ""); 270 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW, 271 &hammer_autoflush, 0, ""); 272 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW, 273 &hammer_verify_zone, 0, ""); 274 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW, 275 &hammer_verify_data, 0, ""); 276 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW, 277 &hammer_write_mode, 0, ""); 278 SYSCTL_INT(_vfs_hammer, OID_AUTO, double_buffer, CTLFLAG_RW, 279 &hammer_double_buffer, 0, ""); 280 SYSCTL_INT(_vfs_hammer, OID_AUTO, yield_check, CTLFLAG_RW, 281 &hammer_yield_check, 0, ""); 282 SYSCTL_INT(_vfs_hammer, OID_AUTO, fsync_mode, CTLFLAG_RW, 283 &hammer_fsync_mode, 0, ""); 284 285 /* KTR_INFO_MASTER(hammer); */ 286 287 /* 288 * VFS ABI 289 */ 290 static void hammer_free_hmp(struct mount *mp); 291 292 static int hammer_vfs_mount(struct mount *mp, char *path, caddr_t data, 293 struct ucred *cred); 294 static int hammer_vfs_unmount(struct mount *mp, int mntflags); 295 static int hammer_vfs_root(struct mount *mp, struct vnode **vpp); 296 static int hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, 297 struct ucred *cred); 298 static int hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, 299 struct ucred *cred); 300 static int hammer_vfs_sync(struct mount *mp, int waitfor); 301 static int hammer_vfs_vget(struct mount *mp, struct vnode *dvp, 302 ino_t ino, struct vnode **vpp); 303 static int hammer_vfs_init(struct vfsconf *conf); 304 static int hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp, 305 struct fid *fhp, struct vnode **vpp); 306 static int hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp); 307 static int hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam, 308 int *exflagsp, struct ucred **credanonp); 309 310 311 static struct vfsops hammer_vfsops = { 312 .vfs_mount = hammer_vfs_mount, 313 .vfs_unmount = hammer_vfs_unmount, 314 .vfs_root = hammer_vfs_root, 315 .vfs_statfs = hammer_vfs_statfs, 316 .vfs_statvfs = hammer_vfs_statvfs, 317 .vfs_sync = hammer_vfs_sync, 318 .vfs_vget = hammer_vfs_vget, 319 .vfs_init = hammer_vfs_init, 320 .vfs_vptofh = hammer_vfs_vptofh, 321 .vfs_fhtovp = hammer_vfs_fhtovp, 322 .vfs_checkexp = hammer_vfs_checkexp 323 }; 324 325 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", ""); 326 327 VFS_SET(hammer_vfsops, hammer, 0); 328 MODULE_VERSION(hammer, 1); 329 330 static int 331 hammer_vfs_init(struct vfsconf *conf) 332 { 333 int n; 334 335 /* 336 * Wait up to this long for an exclusive deadlock to clear 337 * before acquiring a new shared lock on the ip. The deadlock 338 * may have occured on a b-tree node related to the ip. 339 */ 340 if (hammer_tdmux_ticks == 0) 341 hammer_tdmux_ticks = hz / 5; 342 343 /* 344 * Autosize 345 */ 346 if (hammer_limit_recs == 0) { 347 hammer_limit_recs = nbuf * 25; 348 n = kmalloc_limit(M_HAMMER) / 512; 349 if (hammer_limit_recs > n) 350 hammer_limit_recs = n; 351 } 352 if (hammer_limit_dirtybufspace == 0) { 353 hammer_limit_dirtybufspace = hidirtybufspace / 2; 354 if (hammer_limit_dirtybufspace < 100) 355 hammer_limit_dirtybufspace = 100; 356 } 357 358 /* 359 * The hammer_inode structure detaches from the vnode on reclaim. 360 * This limits the number of inodes in this state to prevent a 361 * memory pool blowout. 362 */ 363 if (hammer_limit_reclaims == 0) 364 hammer_limit_reclaims = desiredvnodes / 10; 365 366 return(0); 367 } 368 369 static int 370 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data, 371 struct ucred *cred) 372 { 373 struct hammer_mount_info info; 374 hammer_mount_t hmp; 375 hammer_volume_t rootvol; 376 struct vnode *rootvp; 377 struct vnode *devvp = NULL; 378 const char *upath; /* volume name in userspace */ 379 char *path; /* volume name in system space */ 380 int error; 381 int i; 382 int master_id; 383 char *next_volume_ptr = NULL; 384 385 /* 386 * Accept hammer_mount_info. mntpt is NULL for root mounts at boot. 387 */ 388 if (mntpt == NULL) { 389 bzero(&info, sizeof(info)); 390 info.asof = 0; 391 info.hflags = 0; 392 info.nvolumes = 1; 393 394 next_volume_ptr = mp->mnt_stat.f_mntfromname; 395 396 /* Count number of volumes separated by ':' */ 397 for (char *p = next_volume_ptr; *p != '\0'; ++p) { 398 if (*p == ':') { 399 ++info.nvolumes; 400 } 401 } 402 403 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */ 404 } else { 405 if ((error = copyin(data, &info, sizeof(info))) != 0) 406 return (error); 407 } 408 409 /* 410 * updating or new mount 411 */ 412 if (mp->mnt_flag & MNT_UPDATE) { 413 hmp = (void *)mp->mnt_data; 414 KKASSERT(hmp != NULL); 415 } else { 416 if (info.nvolumes <= 0 || info.nvolumes >= 32768) 417 return (EINVAL); 418 hmp = NULL; 419 } 420 421 /* 422 * master-id validation. The master id may not be changed by a 423 * mount update. 424 */ 425 if (info.hflags & HMNT_MASTERID) { 426 if (hmp && hmp->master_id != info.master_id) { 427 kprintf("hammer: cannot change master id " 428 "with mount update\n"); 429 return(EINVAL); 430 } 431 master_id = info.master_id; 432 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS) 433 return (EINVAL); 434 } else { 435 if (hmp) 436 master_id = hmp->master_id; 437 else 438 master_id = 0; 439 } 440 441 /* 442 * Internal mount data structure 443 */ 444 if (hmp == NULL) { 445 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO); 446 mp->mnt_data = (qaddr_t)hmp; 447 hmp->mp = mp; 448 /*TAILQ_INIT(&hmp->recycle_list);*/ 449 450 /* 451 * Make sure kmalloc type limits are set appropriately. 452 * 453 * Our inode kmalloc group is sized based on maxvnodes 454 * (controlled by the system, not us). 455 */ 456 kmalloc_create(&hmp->m_misc, "HAMMER-others"); 457 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes"); 458 459 kmalloc_raise_limit(hmp->m_inodes, 0); /* unlimited */ 460 461 hmp->root_btree_beg.localization = 0x00000000U; 462 hmp->root_btree_beg.obj_id = -0x8000000000000000LL; 463 hmp->root_btree_beg.key = -0x8000000000000000LL; 464 hmp->root_btree_beg.create_tid = 1; 465 hmp->root_btree_beg.delete_tid = 1; 466 hmp->root_btree_beg.rec_type = 0; 467 hmp->root_btree_beg.obj_type = 0; 468 469 hmp->root_btree_end.localization = 0xFFFFFFFFU; 470 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL; 471 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL; 472 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL; 473 hmp->root_btree_end.delete_tid = 0; /* special case */ 474 hmp->root_btree_end.rec_type = 0xFFFFU; 475 hmp->root_btree_end.obj_type = 0; 476 477 hmp->krate.freq = 1; /* maximum reporting rate (hz) */ 478 hmp->krate.count = -16; /* initial burst */ 479 480 hmp->sync_lock.refs = 1; 481 hmp->free_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 kprintf("HAMMER 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 kprintf("HAMMER 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 kprintf("hammer_mountroot: 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); 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 (error == 0 && hmp->rootvol == NULL) { 631 kprintf("hammer_mount: No root volume found!\n"); 632 error = EINVAL; 633 } 634 635 /* 636 * Check that all required volumes are available 637 */ 638 if (error == 0 && hammer_mountcheck_volumes(hmp)) { 639 kprintf("hammer_mount: Missing volumes, cannot mount!\n"); 640 error = EINVAL; 641 } 642 643 if (error) { 644 /* called with fs_token held */ 645 hammer_free_hmp(mp); 646 return (error); 647 } 648 649 /* 650 * No errors, setup enough of the mount point so we can lookup the 651 * root vnode. 652 */ 653 mp->mnt_iosize_max = MAXPHYS; 654 mp->mnt_kern_flag |= MNTK_FSMID; 655 656 /* 657 * MPSAFE code. Note that VOPs and VFSops which are not MPSAFE 658 * will acquire a per-mount token prior to entry and release it 659 * on return, so even if we do not specify it we no longer get 660 * the BGL regardlless of how we are flagged. 661 */ 662 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE; 663 /*MNTK_RD_MPSAFE | MNTK_GA_MPSAFE | MNTK_IN_MPSAFE;*/ 664 665 /* 666 * note: f_iosize is used by vnode_pager_haspage() when constructing 667 * its VOP_BMAP call. 668 */ 669 mp->mnt_stat.f_iosize = HAMMER_BUFSIZE; 670 mp->mnt_stat.f_bsize = HAMMER_BUFSIZE; 671 672 mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE; 673 mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE; 674 675 mp->mnt_maxsymlinklen = 255; 676 mp->mnt_flag |= MNT_LOCAL; 677 678 vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops); 679 vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops); 680 vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops); 681 682 /* 683 * The root volume's ondisk pointer is only valid if we hold a 684 * reference to it. 685 */ 686 rootvol = hammer_get_root_volume(hmp, &error); 687 if (error) 688 goto failed; 689 690 /* 691 * Perform any necessary UNDO operations. The recovery code does 692 * call hammer_undo_lookup() so we have to pre-cache the blockmap, 693 * and then re-copy it again after recovery is complete. 694 * 695 * If this is a read-only mount the UNDO information is retained 696 * in memory in the form of dirty buffer cache buffers, and not 697 * written back to the media. 698 */ 699 bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap, 700 sizeof(hmp->blockmap)); 701 702 /* 703 * Check filesystem version 704 */ 705 hmp->version = rootvol->ondisk->vol_version; 706 if (hmp->version < HAMMER_VOL_VERSION_MIN || 707 hmp->version > HAMMER_VOL_VERSION_MAX) { 708 kprintf("HAMMER: mount unsupported fs version %d\n", 709 hmp->version); 710 error = ERANGE; 711 goto done; 712 } 713 714 /* 715 * The undo_rec_limit limits the size of flush groups to avoid 716 * blowing out the UNDO FIFO. This calculation is typically in 717 * the tens of thousands and is designed primarily when small 718 * HAMMER filesystems are created. 719 */ 720 hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100; 721 if (hammer_debug_general & 0x0001) 722 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit); 723 724 /* 725 * NOTE: Recover stage1 not only handles meta-data recovery, it 726 * also sets hmp->undo_seqno for HAMMER VERSION 4+ filesystems. 727 */ 728 error = hammer_recover_stage1(hmp, rootvol); 729 if (error) { 730 kprintf("Failed to recover HAMMER filesystem on mount\n"); 731 goto done; 732 } 733 734 /* 735 * Finish setup now that we have a good root volume. 736 * 737 * The top 16 bits of fsid.val[1] is a pfs id. 738 */ 739 ksnprintf(mp->mnt_stat.f_mntfromname, 740 sizeof(mp->mnt_stat.f_mntfromname), "%s", 741 rootvol->ondisk->vol_name); 742 mp->mnt_stat.f_fsid.val[0] = 743 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8); 744 mp->mnt_stat.f_fsid.val[1] = 745 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8); 746 mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF; 747 748 mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid; 749 mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid, 750 sizeof(mp->mnt_vstat.f_fsid_uuid)); 751 752 /* 753 * Certain often-modified fields in the root volume are cached in 754 * the hammer_mount structure so we do not have to generate lots 755 * of little UNDO structures for them. 756 * 757 * Recopy after recovery. This also has the side effect of 758 * setting our cached undo FIFO's first_offset, which serves to 759 * placemark the FIFO start for the NEXT flush cycle while the 760 * on-disk first_offset represents the LAST flush cycle. 761 */ 762 hmp->next_tid = rootvol->ondisk->vol0_next_tid; 763 hmp->flush_tid1 = hmp->next_tid; 764 hmp->flush_tid2 = hmp->next_tid; 765 bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap, 766 sizeof(hmp->blockmap)); 767 hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks; 768 769 hammer_flusher_create(hmp); 770 771 /* 772 * Locate the root directory using the root cluster's B-Tree as a 773 * starting point. The root directory uses an obj_id of 1. 774 * 775 * FUTURE: Leave the root directory cached referenced but unlocked 776 * in hmp->rootvp (need to flush it on unmount). 777 */ 778 error = hammer_vfs_vget(mp, NULL, 1, &rootvp); 779 if (error) 780 goto done; 781 vput(rootvp); 782 /*vn_unlock(hmp->rootvp);*/ 783 if (hmp->ronly == 0) 784 error = hammer_recover_stage2(hmp, rootvol); 785 786 /* 787 * If the stage2 recovery fails be sure to clean out all cached 788 * vnodes before throwing away the mount structure or bad things 789 * will happen. 790 */ 791 if (error) 792 vflush(mp, 0, 0); 793 794 done: 795 if ((mp->mnt_flag & MNT_UPDATE) == 0) { 796 /* New mount */ 797 798 /* Populate info for mount point (NULL pad)*/ 799 bzero(mp->mnt_stat.f_mntonname, MNAMELEN); 800 size_t size; 801 if (mntpt) { 802 copyinstr(mntpt, mp->mnt_stat.f_mntonname, 803 MNAMELEN -1, &size); 804 } else { /* Root mount */ 805 mp->mnt_stat.f_mntonname[0] = '/'; 806 } 807 } 808 (void)VFS_STATFS(mp, &mp->mnt_stat, cred); 809 hammer_rel_volume(rootvol, 0); 810 failed: 811 /* 812 * Cleanup and return. 813 */ 814 if (error) { 815 /* called with fs_token held */ 816 hammer_free_hmp(mp); 817 } else { 818 lwkt_reltoken(&hmp->fs_token); 819 } 820 return (error); 821 } 822 823 static int 824 hammer_vfs_unmount(struct mount *mp, int mntflags) 825 { 826 hammer_mount_t hmp = (void *)mp->mnt_data; 827 int flags; 828 int error; 829 830 /* 831 * Clean out the vnodes 832 */ 833 lwkt_gettoken(&hmp->fs_token); 834 flags = 0; 835 if (mntflags & MNT_FORCE) 836 flags |= FORCECLOSE; 837 error = vflush(mp, 0, flags); 838 839 /* 840 * Clean up the internal mount structure and related entities. This 841 * may issue I/O. 842 */ 843 if (error == 0) { 844 /* called with fs_token held */ 845 hammer_free_hmp(mp); 846 } else { 847 lwkt_reltoken(&hmp->fs_token); 848 } 849 return(error); 850 } 851 852 /* 853 * Clean up the internal mount structure and disassociate it from the mount. 854 * This may issue I/O. 855 * 856 * Called with fs_token held. 857 */ 858 static void 859 hammer_free_hmp(struct mount *mp) 860 { 861 hammer_mount_t hmp = (void *)mp->mnt_data; 862 hammer_flush_group_t flg; 863 int count; 864 int dummy; 865 866 /* 867 * Flush anything dirty. This won't even run if the 868 * filesystem errored-out. 869 */ 870 count = 0; 871 while (hammer_flusher_haswork(hmp)) { 872 hammer_flusher_sync(hmp); 873 ++count; 874 if (count >= 5) { 875 if (count == 5) 876 kprintf("HAMMER: umount flushing."); 877 else 878 kprintf("."); 879 tsleep(&dummy, 0, "hmrufl", hz); 880 } 881 if (count == 30) { 882 kprintf("giving up\n"); 883 break; 884 } 885 } 886 if (count >= 5 && count < 30) 887 kprintf("\n"); 888 889 /* 890 * If the mount had a critical error we have to destroy any 891 * remaining inodes before we can finish cleaning up the flusher. 892 */ 893 if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) { 894 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL, 895 hammer_destroy_inode_callback, NULL); 896 } 897 898 /* 899 * There shouldn't be any inodes left now and any left over 900 * flush groups should now be empty. 901 */ 902 KKASSERT(RB_EMPTY(&hmp->rb_inos_root)); 903 while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) { 904 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry); 905 KKASSERT(RB_EMPTY(&flg->flush_tree)); 906 if (flg->refs) { 907 kprintf("HAMMER: Warning, flush_group %p was " 908 "not empty on umount!\n", flg); 909 } 910 kfree(flg, hmp->m_misc); 911 } 912 913 /* 914 * We can finally destroy the flusher 915 */ 916 hammer_flusher_destroy(hmp); 917 918 /* 919 * We may have held recovered buffers due to a read-only mount. 920 * These must be discarded. 921 */ 922 if (hmp->ronly) 923 hammer_recover_flush_buffers(hmp, NULL, -1); 924 925 /* 926 * Unload buffers and then volumes 927 */ 928 RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL, 929 hammer_unload_buffer, NULL); 930 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL, 931 hammer_unload_volume, NULL); 932 933 mp->mnt_data = NULL; 934 mp->mnt_flag &= ~MNT_LOCAL; 935 hmp->mp = NULL; 936 hammer_destroy_objid_cache(hmp); 937 hammer_destroy_dedup_cache(hmp); 938 if (hmp->dedup_free_cache != NULL) { 939 kfree(hmp->dedup_free_cache, hmp->m_misc); 940 hmp->dedup_free_cache = NULL; 941 } 942 kmalloc_destroy(&hmp->m_misc); 943 kmalloc_destroy(&hmp->m_inodes); 944 lwkt_reltoken(&hmp->fs_token); 945 kfree(hmp, M_HAMMER); 946 } 947 948 /* 949 * Report critical errors. ip may be NULL. 950 */ 951 void 952 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip, 953 int error, const char *msg) 954 { 955 hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR; 956 957 krateprintf(&hmp->krate, 958 "HAMMER(%s): Critical error inode=%jd error=%d %s\n", 959 hmp->mp->mnt_stat.f_mntfromname, 960 (intmax_t)(ip ? ip->obj_id : -1), 961 error, msg); 962 963 if (hmp->ronly == 0) { 964 hmp->ronly = 2; /* special errored read-only mode */ 965 hmp->mp->mnt_flag |= MNT_RDONLY; 966 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL, 967 hammer_adjust_volume_mode, NULL); 968 kprintf("HAMMER(%s): Forcing read-only mode\n", 969 hmp->mp->mnt_stat.f_mntfromname); 970 } 971 hmp->error = error; 972 if (hammer_debug_critical) 973 Debugger("Entering debugger"); 974 } 975 976 977 /* 978 * Obtain a vnode for the specified inode number. An exclusively locked 979 * vnode is returned. 980 */ 981 int 982 hammer_vfs_vget(struct mount *mp, struct vnode *dvp, 983 ino_t ino, struct vnode **vpp) 984 { 985 struct hammer_transaction trans; 986 struct hammer_mount *hmp = (void *)mp->mnt_data; 987 struct hammer_inode *ip; 988 int error; 989 u_int32_t localization; 990 991 lwkt_gettoken(&hmp->fs_token); 992 hammer_simple_transaction(&trans, hmp); 993 994 /* 995 * If a directory vnode is supplied (mainly NFS) then we can acquire 996 * the PFS domain from it. Otherwise we would only be able to vget 997 * inodes in the root PFS. 998 */ 999 if (dvp) { 1000 localization = HAMMER_DEF_LOCALIZATION + 1001 VTOI(dvp)->obj_localization; 1002 } else { 1003 localization = HAMMER_DEF_LOCALIZATION; 1004 } 1005 1006 /* 1007 * Lookup the requested HAMMER inode. The structure must be 1008 * left unlocked while we manipulate the related vnode to avoid 1009 * a deadlock. 1010 */ 1011 ip = hammer_get_inode(&trans, NULL, ino, 1012 hmp->asof, localization, 1013 0, &error); 1014 if (ip == NULL) { 1015 *vpp = NULL; 1016 } else { 1017 error = hammer_get_vnode(ip, vpp); 1018 hammer_rel_inode(ip, 0); 1019 } 1020 hammer_done_transaction(&trans); 1021 lwkt_reltoken(&hmp->fs_token); 1022 return (error); 1023 } 1024 1025 /* 1026 * Return the root vnode for the filesystem. 1027 * 1028 * HAMMER stores the root vnode in the hammer_mount structure so 1029 * getting it is easy. 1030 */ 1031 static int 1032 hammer_vfs_root(struct mount *mp, struct vnode **vpp) 1033 { 1034 int error; 1035 1036 error = hammer_vfs_vget(mp, NULL, 1, vpp); 1037 return (error); 1038 } 1039 1040 static int 1041 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred) 1042 { 1043 struct hammer_mount *hmp = (void *)mp->mnt_data; 1044 hammer_volume_t volume; 1045 hammer_volume_ondisk_t ondisk; 1046 int error; 1047 int64_t bfree; 1048 int64_t breserved; 1049 1050 lwkt_gettoken(&hmp->fs_token); 1051 volume = hammer_get_root_volume(hmp, &error); 1052 if (error) { 1053 lwkt_reltoken(&hmp->fs_token); 1054 return(error); 1055 } 1056 ondisk = volume->ondisk; 1057 1058 /* 1059 * Basic stats 1060 */ 1061 _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved); 1062 mp->mnt_stat.f_files = ondisk->vol0_stat_inodes; 1063 bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE; 1064 hammer_rel_volume(volume, 0); 1065 1066 mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE; 1067 mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree; 1068 if (mp->mnt_stat.f_files < 0) 1069 mp->mnt_stat.f_files = 0; 1070 1071 *sbp = mp->mnt_stat; 1072 lwkt_reltoken(&hmp->fs_token); 1073 return(0); 1074 } 1075 1076 static int 1077 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred) 1078 { 1079 struct hammer_mount *hmp = (void *)mp->mnt_data; 1080 hammer_volume_t volume; 1081 hammer_volume_ondisk_t ondisk; 1082 int error; 1083 int64_t bfree; 1084 int64_t breserved; 1085 1086 lwkt_gettoken(&hmp->fs_token); 1087 volume = hammer_get_root_volume(hmp, &error); 1088 if (error) { 1089 lwkt_reltoken(&hmp->fs_token); 1090 return(error); 1091 } 1092 ondisk = volume->ondisk; 1093 1094 /* 1095 * Basic stats 1096 */ 1097 _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved); 1098 mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes; 1099 bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE; 1100 hammer_rel_volume(volume, 0); 1101 1102 mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE; 1103 mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree; 1104 if (mp->mnt_vstat.f_files < 0) 1105 mp->mnt_vstat.f_files = 0; 1106 *sbp = mp->mnt_vstat; 1107 lwkt_reltoken(&hmp->fs_token); 1108 return(0); 1109 } 1110 1111 /* 1112 * Sync the filesystem. Currently we have to run it twice, the second 1113 * one will advance the undo start index to the end index, so if a crash 1114 * occurs no undos will be run on mount. 1115 * 1116 * We do not sync the filesystem if we are called from a panic. If we did 1117 * we might end up blowing up a sync that was already in progress. 1118 */ 1119 static int 1120 hammer_vfs_sync(struct mount *mp, int waitfor) 1121 { 1122 struct hammer_mount *hmp = (void *)mp->mnt_data; 1123 int error; 1124 1125 lwkt_gettoken(&hmp->fs_token); 1126 if (panicstr == NULL) { 1127 error = hammer_sync_hmp(hmp, waitfor); 1128 } else { 1129 error = EIO; 1130 } 1131 lwkt_reltoken(&hmp->fs_token); 1132 return (error); 1133 } 1134 1135 /* 1136 * Convert a vnode to a file handle. 1137 * 1138 * Accesses read-only fields on already-referenced structures so 1139 * no token is needed. 1140 */ 1141 static int 1142 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp) 1143 { 1144 hammer_inode_t ip; 1145 1146 KKASSERT(MAXFIDSZ >= 16); 1147 ip = VTOI(vp); 1148 fhp->fid_len = offsetof(struct fid, fid_data[16]); 1149 fhp->fid_ext = ip->obj_localization >> 16; 1150 bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id)); 1151 bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof)); 1152 return(0); 1153 } 1154 1155 1156 /* 1157 * Convert a file handle back to a vnode. 1158 * 1159 * Use rootvp to enforce PFS isolation when a PFS is exported via a 1160 * null mount. 1161 */ 1162 static int 1163 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp, 1164 struct fid *fhp, struct vnode **vpp) 1165 { 1166 hammer_mount_t hmp = (void *)mp->mnt_data; 1167 struct hammer_transaction trans; 1168 struct hammer_inode *ip; 1169 struct hammer_inode_info info; 1170 int error; 1171 u_int32_t localization; 1172 1173 bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id)); 1174 bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof)); 1175 if (rootvp) 1176 localization = VTOI(rootvp)->obj_localization; 1177 else 1178 localization = (u_int32_t)fhp->fid_ext << 16; 1179 1180 lwkt_gettoken(&hmp->fs_token); 1181 hammer_simple_transaction(&trans, hmp); 1182 1183 /* 1184 * Get/allocate the hammer_inode structure. The structure must be 1185 * unlocked while we manipulate the related vnode to avoid a 1186 * deadlock. 1187 */ 1188 ip = hammer_get_inode(&trans, NULL, info.obj_id, 1189 info.obj_asof, localization, 0, &error); 1190 if (ip) { 1191 error = hammer_get_vnode(ip, vpp); 1192 hammer_rel_inode(ip, 0); 1193 } else { 1194 *vpp = NULL; 1195 } 1196 hammer_done_transaction(&trans); 1197 lwkt_reltoken(&hmp->fs_token); 1198 return (error); 1199 } 1200 1201 static int 1202 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam, 1203 int *exflagsp, struct ucred **credanonp) 1204 { 1205 hammer_mount_t hmp = (void *)mp->mnt_data; 1206 struct netcred *np; 1207 int error; 1208 1209 lwkt_gettoken(&hmp->fs_token); 1210 np = vfs_export_lookup(mp, &hmp->export, nam); 1211 if (np) { 1212 *exflagsp = np->netc_exflags; 1213 *credanonp = &np->netc_anon; 1214 error = 0; 1215 } else { 1216 error = EACCES; 1217 } 1218 lwkt_reltoken(&hmp->fs_token); 1219 return (error); 1220 1221 } 1222 1223 int 1224 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export) 1225 { 1226 hammer_mount_t hmp = (void *)mp->mnt_data; 1227 int error; 1228 1229 lwkt_gettoken(&hmp->fs_token); 1230 1231 switch(op) { 1232 case MOUNTCTL_SET_EXPORT: 1233 error = vfs_export(mp, &hmp->export, export); 1234 break; 1235 default: 1236 error = EOPNOTSUPP; 1237 break; 1238 } 1239 lwkt_reltoken(&hmp->fs_token); 1240 1241 return(error); 1242 } 1243 1244