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_recover_faults; 52 int hammer_debug_critical; /* non-zero enter debugger on error */ 53 int hammer_cluster_enable = 1; /* enable read clustering by default */ 54 int hammer_live_dedup = 0; 55 int hammer_tdmux_ticks; 56 int hammer_count_fsyncs; 57 int hammer_count_inodes; 58 int hammer_count_iqueued; 59 int hammer_count_reclaims; 60 int hammer_count_records; 61 int hammer_count_record_datas; 62 int hammer_count_volumes; 63 int hammer_count_buffers; 64 int hammer_count_nodes; 65 int64_t hammer_count_extra_space_used; 66 int64_t hammer_stats_btree_lookups; 67 int64_t hammer_stats_btree_searches; 68 int64_t hammer_stats_btree_inserts; 69 int64_t hammer_stats_btree_deletes; 70 int64_t hammer_stats_btree_elements; 71 int64_t hammer_stats_btree_splits; 72 int64_t hammer_stats_btree_iterations; 73 int64_t hammer_stats_btree_root_iterations; 74 int64_t hammer_stats_record_iterations; 75 76 int64_t hammer_stats_file_read; 77 int64_t hammer_stats_file_write; 78 int64_t hammer_stats_file_iopsr; 79 int64_t hammer_stats_file_iopsw; 80 int64_t hammer_stats_disk_read; 81 int64_t hammer_stats_disk_write; 82 int64_t hammer_stats_inode_flushes; 83 int64_t hammer_stats_commits; 84 int64_t hammer_stats_undo; 85 int64_t hammer_stats_redo; 86 87 long hammer_count_dirtybufspace; /* global */ 88 int hammer_count_refedbufs; /* global */ 89 int hammer_count_reservations; 90 long hammer_count_io_running_read; 91 long hammer_count_io_running_write; 92 int hammer_count_io_locked; 93 long hammer_limit_dirtybufspace; /* per-mount */ 94 int hammer_limit_recs; /* as a whole XXX */ 95 int hammer_limit_inode_recs = 2048; /* per inode */ 96 int hammer_limit_reclaims; 97 int hammer_live_dedup_cache_size = DEDUP_CACHE_SIZE; 98 int hammer_limit_redo = 4096 * 1024; /* per inode */ 99 int hammer_autoflush = 500; /* auto flush (typ on reclaim) */ 100 int hammer_bio_count; 101 int hammer_verify_zone; 102 int hammer_verify_data = 1; 103 int hammer_write_mode; 104 int hammer_double_buffer; 105 int hammer_btree_full_undo = 1; 106 int hammer_yield_check = 16; 107 int hammer_fsync_mode = 3; 108 int64_t hammer_contention_count; 109 int64_t hammer_zone_limit; 110 111 /* 112 * Live dedup debug counters (sysctls are writable so that counters 113 * can be reset from userspace). 114 */ 115 int64_t hammer_live_dedup_vnode_bcmps = 0; 116 int64_t hammer_live_dedup_device_bcmps = 0; 117 int64_t hammer_live_dedup_findblk_failures = 0; 118 int64_t hammer_live_dedup_bmap_saves = 0; 119 120 121 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem"); 122 123 SYSCTL_INT(_vfs_hammer, OID_AUTO, supported_version, CTLFLAG_RD, 124 &hammer_supported_version, 0, ""); 125 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW, 126 &hammer_debug_general, 0, ""); 127 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW, 128 &hammer_debug_io, 0, ""); 129 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW, 130 &hammer_debug_inode, 0, ""); 131 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW, 132 &hammer_debug_locks, 0, ""); 133 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW, 134 &hammer_debug_btree, 0, ""); 135 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW, 136 &hammer_debug_tid, 0, ""); 137 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW, 138 &hammer_debug_recover, 0, ""); 139 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW, 140 &hammer_debug_recover_faults, 0, ""); 141 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_critical, CTLFLAG_RW, 142 &hammer_debug_critical, 0, ""); 143 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW, 144 &hammer_cluster_enable, 0, ""); 145 /* 146 * 0 - live dedup is disabled 147 * 1 - dedup cache is populated on reads only 148 * 2 - dedup cache is populated on both reads and writes 149 * 150 * LIVE_DEDUP IS DISABLED PERMANENTLY! This feature appears to cause 151 * blockmap corruption over time so we've turned it off permanently. 152 */ 153 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup, CTLFLAG_RD, 154 &hammer_live_dedup, 0, "Enable live dedup (experimental)"); 155 SYSCTL_INT(_vfs_hammer, OID_AUTO, tdmux_ticks, CTLFLAG_RW, 156 &hammer_tdmux_ticks, 0, "Hammer tdmux ticks"); 157 158 SYSCTL_LONG(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW, 159 &hammer_limit_dirtybufspace, 0, ""); 160 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW, 161 &hammer_limit_recs, 0, ""); 162 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_inode_recs, CTLFLAG_RW, 163 &hammer_limit_inode_recs, 0, ""); 164 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_reclaims, CTLFLAG_RW, 165 &hammer_limit_reclaims, 0, ""); 166 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup_cache_size, CTLFLAG_RW, 167 &hammer_live_dedup_cache_size, 0, 168 "Number of cache entries"); 169 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_redo, CTLFLAG_RW, 170 &hammer_limit_redo, 0, ""); 171 172 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD, 173 &hammer_count_fsyncs, 0, ""); 174 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD, 175 &hammer_count_inodes, 0, ""); 176 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD, 177 &hammer_count_iqueued, 0, ""); 178 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaims, CTLFLAG_RD, 179 &hammer_count_reclaims, 0, ""); 180 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD, 181 &hammer_count_records, 0, ""); 182 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD, 183 &hammer_count_record_datas, 0, ""); 184 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD, 185 &hammer_count_volumes, 0, ""); 186 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD, 187 &hammer_count_buffers, 0, ""); 188 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD, 189 &hammer_count_nodes, 0, ""); 190 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD, 191 &hammer_count_extra_space_used, 0, ""); 192 193 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD, 194 &hammer_stats_btree_searches, 0, ""); 195 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD, 196 &hammer_stats_btree_lookups, 0, ""); 197 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD, 198 &hammer_stats_btree_inserts, 0, ""); 199 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD, 200 &hammer_stats_btree_deletes, 0, ""); 201 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD, 202 &hammer_stats_btree_elements, 0, ""); 203 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD, 204 &hammer_stats_btree_splits, 0, ""); 205 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD, 206 &hammer_stats_btree_iterations, 0, ""); 207 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_root_iterations, CTLFLAG_RD, 208 &hammer_stats_btree_root_iterations, 0, ""); 209 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD, 210 &hammer_stats_record_iterations, 0, ""); 211 212 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD, 213 &hammer_stats_file_read, 0, ""); 214 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD, 215 &hammer_stats_file_write, 0, ""); 216 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD, 217 &hammer_stats_file_iopsr, 0, ""); 218 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD, 219 &hammer_stats_file_iopsw, 0, ""); 220 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD, 221 &hammer_stats_disk_read, 0, ""); 222 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD, 223 &hammer_stats_disk_write, 0, ""); 224 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD, 225 &hammer_stats_inode_flushes, 0, ""); 226 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD, 227 &hammer_stats_commits, 0, ""); 228 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_undo, CTLFLAG_RD, 229 &hammer_stats_undo, 0, ""); 230 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_redo, CTLFLAG_RD, 231 &hammer_stats_redo, 0, ""); 232 233 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_vnode_bcmps, CTLFLAG_RW, 234 &hammer_live_dedup_vnode_bcmps, 0, 235 "successful vnode buffer comparisons"); 236 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_device_bcmps, CTLFLAG_RW, 237 &hammer_live_dedup_device_bcmps, 0, 238 "successful device buffer comparisons"); 239 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_findblk_failures, CTLFLAG_RW, 240 &hammer_live_dedup_findblk_failures, 0, 241 "block lookup failures for comparison"); 242 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_bmap_saves, CTLFLAG_RW, 243 &hammer_live_dedup_bmap_saves, 0, 244 "useful physical block lookups"); 245 246 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD, 247 &hammer_count_dirtybufspace, 0, ""); 248 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD, 249 &hammer_count_refedbufs, 0, ""); 250 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD, 251 &hammer_count_reservations, 0, ""); 252 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD, 253 &hammer_count_io_running_read, 0, ""); 254 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD, 255 &hammer_count_io_locked, 0, ""); 256 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD, 257 &hammer_count_io_running_write, 0, ""); 258 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW, 259 &hammer_zone_limit, 0, ""); 260 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW, 261 &hammer_contention_count, 0, ""); 262 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW, 263 &hammer_autoflush, 0, ""); 264 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW, 265 &hammer_verify_zone, 0, ""); 266 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW, 267 &hammer_verify_data, 0, ""); 268 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW, 269 &hammer_write_mode, 0, ""); 270 SYSCTL_INT(_vfs_hammer, OID_AUTO, double_buffer, CTLFLAG_RW, 271 &hammer_double_buffer, 0, ""); 272 SYSCTL_INT(_vfs_hammer, OID_AUTO, btree_full_undo, CTLFLAG_RW, 273 &hammer_btree_full_undo, 0, ""); 274 SYSCTL_INT(_vfs_hammer, OID_AUTO, yield_check, CTLFLAG_RW, 275 &hammer_yield_check, 0, ""); 276 SYSCTL_INT(_vfs_hammer, OID_AUTO, fsync_mode, CTLFLAG_RW, 277 &hammer_fsync_mode, 0, ""); 278 279 /* KTR_INFO_MASTER(hammer); */ 280 281 /* 282 * VFS ABI 283 */ 284 static void hammer_free_hmp(struct mount *mp); 285 286 static int hammer_vfs_mount(struct mount *mp, char *path, caddr_t data, 287 struct ucred *cred); 288 static int hammer_vfs_unmount(struct mount *mp, int mntflags); 289 static int hammer_vfs_root(struct mount *mp, struct vnode **vpp); 290 static int hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, 291 struct ucred *cred); 292 static int hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, 293 struct ucred *cred); 294 static int hammer_vfs_sync(struct mount *mp, int waitfor); 295 static int hammer_vfs_vget(struct mount *mp, struct vnode *dvp, 296 ino_t ino, struct vnode **vpp); 297 static int hammer_vfs_init(struct vfsconf *conf); 298 static int hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp, 299 struct fid *fhp, struct vnode **vpp); 300 static int hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp); 301 static int hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam, 302 int *exflagsp, struct ucred **credanonp); 303 304 305 static struct vfsops hammer_vfsops = { 306 .vfs_mount = hammer_vfs_mount, 307 .vfs_unmount = hammer_vfs_unmount, 308 .vfs_root = hammer_vfs_root, 309 .vfs_statfs = hammer_vfs_statfs, 310 .vfs_statvfs = hammer_vfs_statvfs, 311 .vfs_sync = hammer_vfs_sync, 312 .vfs_vget = hammer_vfs_vget, 313 .vfs_init = hammer_vfs_init, 314 .vfs_vptofh = hammer_vfs_vptofh, 315 .vfs_fhtovp = hammer_vfs_fhtovp, 316 .vfs_checkexp = hammer_vfs_checkexp 317 }; 318 319 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", ""); 320 321 VFS_SET(hammer_vfsops, hammer, 0); 322 MODULE_VERSION(hammer, 1); 323 324 static int 325 hammer_vfs_init(struct vfsconf *conf) 326 { 327 long n; 328 329 /* 330 * Wait up to this long for an exclusive deadlock to clear 331 * before acquiring a new shared lock on the ip. The deadlock 332 * may have occured on a b-tree node related to the ip. 333 */ 334 if (hammer_tdmux_ticks == 0) 335 hammer_tdmux_ticks = hz / 5; 336 337 /* 338 * Autosize, but be careful because a hammer filesystem's 339 * reserve is partially calculated based on dirtybufspace, 340 * so we simply cannot allow it to get too large. 341 */ 342 if (hammer_limit_recs == 0) { 343 n = nbuf * 25; 344 if (n > kmalloc_limit(M_HAMMER) / 512) 345 n = kmalloc_limit(M_HAMMER) / 512; 346 if (n > 2 * 1024 * 1024) 347 n = 2 * 1024 * 1024; 348 hammer_limit_recs = (int)n; 349 } 350 if (hammer_limit_dirtybufspace == 0) { 351 hammer_limit_dirtybufspace = hidirtybufspace / 2; 352 if (hammer_limit_dirtybufspace < 1L * 1024 * 1024) 353 hammer_limit_dirtybufspace = 1024L * 1024; 354 if (hammer_limit_dirtybufspace > 1024L * 1024 * 1024) 355 hammer_limit_dirtybufspace = 1024L * 1024 * 1024; 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 int nvolumes; 384 char *next_volume_ptr = NULL; 385 386 /* 387 * Accept hammer_mount_info. mntpt is NULL for root mounts at boot. 388 */ 389 if (mntpt == NULL) { 390 bzero(&info, sizeof(info)); 391 info.asof = 0; 392 info.hflags = 0; 393 info.nvolumes = 1; 394 395 next_volume_ptr = mp->mnt_stat.f_mntfromname; 396 397 /* Count number of volumes separated by ':' */ 398 for (char *p = next_volume_ptr; *p != '\0'; ++p) { 399 if (*p == ':') { 400 ++info.nvolumes; 401 } 402 } 403 404 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */ 405 } else { 406 if ((error = copyin(data, &info, sizeof(info))) != 0) 407 return (error); 408 } 409 410 /* 411 * updating or new mount 412 */ 413 if (mp->mnt_flag & MNT_UPDATE) { 414 hmp = (void *)mp->mnt_data; 415 KKASSERT(hmp != NULL); 416 } else { 417 if (info.nvolumes <= 0 || info.nvolumes > HAMMER_MAX_VOLUMES) 418 return (EINVAL); 419 hmp = NULL; 420 } 421 422 /* 423 * master-id validation. The master id may not be changed by a 424 * mount update. 425 */ 426 if (info.hflags & HMNT_MASTERID || info.hflags & HMNT_NOMIRROR) { 427 if (hmp && hmp->master_id != info.master_id) { 428 hkprintf("cannot change master id 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; 448 449 /* 450 * Make sure kmalloc type limits are set appropriately. 451 * 452 * Our inode kmalloc group is sized based on maxvnodes 453 * (controlled by the system, not us). 454 */ 455 kmalloc_create(&hmp->m_misc, "HAMMER-others"); 456 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes"); 457 458 kmalloc_raise_limit(hmp->m_inodes, 0); /* unlimited */ 459 460 hmp->root_btree_beg.localization = 0x00000000U; 461 hmp->root_btree_beg.obj_id = -0x8000000000000000LL; 462 hmp->root_btree_beg.key = -0x8000000000000000LL; 463 hmp->root_btree_beg.create_tid = 1; 464 hmp->root_btree_beg.delete_tid = 1; 465 hmp->root_btree_beg.rec_type = 0; 466 hmp->root_btree_beg.obj_type = 0; 467 hmp->root_btree_beg.btype = HAMMER_BTREE_TYPE_NONE; 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 hmp->root_btree_end.btype = HAMMER_BTREE_TYPE_NONE; 477 478 hmp->krate.freq = 1; /* maximum reporting rate (hz) */ 479 hmp->krate.count = -16; /* initial burst */ 480 hmp->kdiag.freq = 1; /* maximum reporting rate (hz) */ 481 hmp->kdiag.count = -16; /* initial burst */ 482 483 hmp->sync_lock.refs = 1; 484 hmp->free_lock.refs = 1; 485 hmp->undo_lock.refs = 1; 486 hmp->blkmap_lock.refs = 1; 487 hmp->snapshot_lock.refs = 1; 488 hmp->volume_lock.refs = 1; 489 490 TAILQ_INIT(&hmp->delay_list); 491 TAILQ_INIT(&hmp->flush_group_list); 492 TAILQ_INIT(&hmp->objid_cache_list); 493 TAILQ_INIT(&hmp->undo_lru_list); 494 TAILQ_INIT(&hmp->reclaim_list); 495 496 RB_INIT(&hmp->rb_dedup_crc_root); 497 RB_INIT(&hmp->rb_dedup_off_root); 498 TAILQ_INIT(&hmp->dedup_lru_list); 499 } 500 hmp->hflags &= ~HMNT_USERFLAGS; 501 hmp->hflags |= info.hflags & HMNT_USERFLAGS; 502 503 hmp->master_id = master_id; 504 505 if (info.asof) { 506 mp->mnt_flag |= MNT_RDONLY; 507 hmp->asof = info.asof; 508 } else { 509 hmp->asof = HAMMER_MAX_TID; 510 } 511 512 hmp->volume_to_remove = -1; 513 514 /* 515 * Re-open read-write if originally read-only, or vise-versa. 516 * 517 * When going from read-only to read-write execute the stage2 518 * recovery if it has not already been run. 519 */ 520 if (mp->mnt_flag & MNT_UPDATE) { 521 lwkt_gettoken(&hmp->fs_token); 522 error = 0; 523 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) { 524 hkprintf("read-only -> read-write\n"); 525 hmp->ronly = 0; 526 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL, 527 hammer_adjust_volume_mode, NULL); 528 rootvol = hammer_get_root_volume(hmp, &error); 529 if (rootvol) { 530 hammer_recover_flush_buffers(hmp, rootvol, 1); 531 error = hammer_recover_stage2(hmp, rootvol); 532 bcopy(rootvol->ondisk->vol0_blockmap, 533 hmp->blockmap, 534 sizeof(hmp->blockmap)); 535 hammer_rel_volume(rootvol, 0); 536 } 537 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL, 538 hammer_reload_inode, NULL); 539 /* kernel clears MNT_RDONLY */ 540 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) { 541 hkprintf("read-write -> read-only\n"); 542 hmp->ronly = 1; /* messy */ 543 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL, 544 hammer_reload_inode, NULL); 545 hmp->ronly = 0; 546 hammer_flusher_sync(hmp); 547 hammer_flusher_sync(hmp); 548 hammer_flusher_sync(hmp); 549 hmp->ronly = 1; 550 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL, 551 hammer_adjust_volume_mode, NULL); 552 } 553 lwkt_reltoken(&hmp->fs_token); 554 return(error); 555 } 556 557 RB_INIT(&hmp->rb_vols_root); 558 RB_INIT(&hmp->rb_inos_root); 559 RB_INIT(&hmp->rb_redo_root); 560 RB_INIT(&hmp->rb_nods_root); 561 RB_INIT(&hmp->rb_undo_root); 562 RB_INIT(&hmp->rb_resv_root); 563 RB_INIT(&hmp->rb_bufs_root); 564 RB_INIT(&hmp->rb_pfsm_root); 565 566 hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0); 567 568 RB_INIT(&hmp->volu_root); 569 RB_INIT(&hmp->undo_root); 570 RB_INIT(&hmp->data_root); 571 RB_INIT(&hmp->meta_root); 572 RB_INIT(&hmp->lose_root); 573 TAILQ_INIT(&hmp->iorun_list); 574 575 lwkt_token_init(&hmp->fs_token, "hammerfs"); 576 lwkt_token_init(&hmp->io_token, "hammerio"); 577 578 lwkt_gettoken(&hmp->fs_token); 579 580 /* 581 * Load volumes 582 */ 583 path = objcache_get(namei_oc, M_WAITOK); 584 hmp->nvolumes = -1; 585 for (i = 0; i < info.nvolumes; ++i) { 586 if (mntpt == NULL) { 587 /* 588 * Root mount. 589 */ 590 KKASSERT(next_volume_ptr != NULL); 591 strcpy(path, ""); 592 if (*next_volume_ptr != '/') { 593 /* relative path */ 594 strcpy(path, "/dev/"); 595 } 596 int k; 597 for (k = strlen(path); k < MAXPATHLEN-1; ++k) { 598 if (*next_volume_ptr == '\0') { 599 break; 600 } else if (*next_volume_ptr == ':') { 601 ++next_volume_ptr; 602 break; 603 } else { 604 path[k] = *next_volume_ptr; 605 ++next_volume_ptr; 606 } 607 } 608 path[k] = '\0'; 609 610 error = 0; 611 cdev_t dev = kgetdiskbyname(path); 612 error = bdevvp(dev, &devvp); 613 if (error) { 614 hdkprintf("can't find devvp\n"); 615 } 616 } else { 617 error = copyin(&info.volumes[i], &upath, 618 sizeof(char *)); 619 if (error == 0) 620 error = copyinstr(upath, path, 621 MAXPATHLEN, NULL); 622 } 623 if (error == 0) 624 error = hammer_install_volume(hmp, path, devvp, NULL); 625 if (error) 626 break; 627 } 628 objcache_put(namei_oc, path); 629 630 /* 631 * Make sure we found a root volume 632 */ 633 if (hmp->rootvol == NULL) { 634 if (error == EBUSY) { 635 hdkprintf("The volumes are probably mounted\n"); 636 } else { 637 hdkprintf("No root volume found!\n"); 638 error = EINVAL; 639 } 640 goto failed; 641 } 642 643 /* 644 * Check that all required volumes are available 645 */ 646 if (error == 0 && hammer_mountcheck_volumes(hmp)) { 647 hdkprintf("Missing volumes, cannot mount!\n"); 648 error = EINVAL; 649 goto failed; 650 } 651 652 /* 653 * Other errors 654 */ 655 if (error) { 656 hdkprintf("Failed to load volumes!\n"); 657 goto failed; 658 } 659 660 nvolumes = hammer_get_installed_volumes(hmp); 661 if (hmp->nvolumes != nvolumes) { 662 hdkprintf("volume header says %d volumes, but %d installed\n", 663 hmp->nvolumes, nvolumes); 664 error = EINVAL; 665 goto failed; 666 } 667 668 /* 669 * No errors, setup enough of the mount point so we can lookup the 670 * root vnode. 671 */ 672 mp->mnt_iosize_max = MAXPHYS; 673 mp->mnt_kern_flag |= MNTK_FSMID; 674 mp->mnt_kern_flag |= MNTK_THR_SYNC; /* new vsyncscan semantics */ 675 676 /* 677 * MPSAFE code. Note that VOPs and VFSops which are not MPSAFE 678 * will acquire a per-mount token prior to entry and release it 679 * on return. 680 */ 681 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE; 682 683 /* 684 * note: f_iosize is used by vnode_pager_haspage() when constructing 685 * its VOP_BMAP call. 686 */ 687 mp->mnt_stat.f_iosize = HAMMER_BUFSIZE; 688 mp->mnt_stat.f_bsize = HAMMER_BUFSIZE; 689 690 mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE; 691 mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE; 692 693 mp->mnt_maxsymlinklen = 255; 694 mp->mnt_flag |= MNT_LOCAL; 695 696 vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops); 697 vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops); 698 vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops); 699 700 /* 701 * The root volume's ondisk pointer is only valid if we hold a 702 * reference to it. 703 */ 704 rootvol = hammer_get_root_volume(hmp, &error); 705 if (error) 706 goto failed; 707 708 /* 709 * Perform any necessary UNDO operations. The recovery code does 710 * call hammer_undo_lookup() so we have to pre-cache the blockmap, 711 * and then re-copy it again after recovery is complete. 712 * 713 * If this is a read-only mount the UNDO information is retained 714 * in memory in the form of dirty buffer cache buffers, and not 715 * written back to the media. 716 */ 717 bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap, 718 sizeof(hmp->blockmap)); 719 720 /* 721 * Check filesystem version 722 */ 723 hmp->version = rootvol->ondisk->vol_version; 724 if (hmp->version < HAMMER_VOL_VERSION_MIN || 725 hmp->version > HAMMER_VOL_VERSION_MAX) { 726 hkprintf("mount unsupported fs version %d\n", hmp->version); 727 error = ERANGE; 728 goto done; 729 } 730 731 /* 732 * The undo_rec_limit limits the size of flush groups to avoid 733 * blowing out the UNDO FIFO. This calculation is typically in 734 * the tens of thousands and is designed primarily when small 735 * HAMMER filesystems are created. 736 */ 737 hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100; 738 if (hammer_debug_general & 0x0001) 739 hkprintf("undo_rec_limit %d\n", hmp->undo_rec_limit); 740 741 /* 742 * NOTE: Recover stage1 not only handles meta-data recovery, it 743 * also sets hmp->undo_seqno for HAMMER VERSION 4+ filesystems. 744 */ 745 error = hammer_recover_stage1(hmp, rootvol); 746 if (error) { 747 kprintf("Failed to recover HAMMER filesystem on mount\n"); 748 goto done; 749 } 750 751 /* 752 * Finish setup now that we have a good root volume. 753 * vol_name is a filesystem label string. 754 */ 755 ksnprintf(mp->mnt_stat.f_mntfromname, 756 sizeof(mp->mnt_stat.f_mntfromname), "%s", 757 rootvol->ondisk->vol_name); 758 mp->mnt_stat.f_fsid.val[0] = 759 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8); 760 mp->mnt_stat.f_fsid.val[1] = 761 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8); 762 mp->mnt_stat.f_fsid.val[1] &= HAMMER_LOCALIZE_MASK; 763 764 mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid; 765 mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid, 766 sizeof(mp->mnt_vstat.f_fsid_uuid)); 767 768 /* 769 * Certain often-modified fields in the root volume are cached in 770 * the hammer_mount structure so we do not have to generate lots 771 * of little UNDO structures for them. 772 * 773 * Recopy after recovery. This also has the side effect of 774 * setting our cached undo FIFO's first_offset, which serves to 775 * placemark the FIFO start for the NEXT flush cycle while the 776 * on-disk first_offset represents the LAST flush cycle. 777 */ 778 hmp->next_tid = rootvol->ondisk->vol0_next_tid; 779 hmp->flush_tid1 = hmp->next_tid; 780 hmp->flush_tid2 = hmp->next_tid; 781 bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap, 782 sizeof(hmp->blockmap)); 783 hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks; 784 785 hammer_flusher_create(hmp); 786 787 /* 788 * Locate the root directory with an obj_id of 1. 789 */ 790 error = hammer_vfs_vget(mp, NULL, HAMMER_OBJID_ROOT, &rootvp); 791 if (error) 792 goto done; 793 vput(rootvp); 794 if (hmp->ronly == 0) 795 error = hammer_recover_stage2(hmp, rootvol); 796 797 /* 798 * If the stage2 recovery fails be sure to clean out all cached 799 * vnodes before throwing away the mount structure or bad things 800 * will happen. 801 */ 802 if (error) 803 vflush(mp, 0, 0); 804 805 done: 806 if ((mp->mnt_flag & MNT_UPDATE) == 0) { 807 /* New mount */ 808 809 /* Populate info for mount point (NULL pad)*/ 810 bzero(mp->mnt_stat.f_mntonname, MNAMELEN); 811 size_t size; 812 if (mntpt) { 813 copyinstr(mntpt, mp->mnt_stat.f_mntonname, 814 MNAMELEN -1, &size); 815 } else { /* Root mount */ 816 mp->mnt_stat.f_mntonname[0] = '/'; 817 } 818 } 819 (void)VFS_STATFS(mp, &mp->mnt_stat, cred); 820 hammer_rel_volume(rootvol, 0); 821 failed: 822 /* 823 * Cleanup and return. 824 */ 825 if (error) { 826 /* called with fs_token held */ 827 hammer_free_hmp(mp); 828 } else { 829 lwkt_reltoken(&hmp->fs_token); 830 } 831 return (error); 832 } 833 834 static int 835 hammer_vfs_unmount(struct mount *mp, int mntflags) 836 { 837 hammer_mount_t hmp = (void *)mp->mnt_data; 838 int flags; 839 int error; 840 841 /* 842 * Clean out the vnodes 843 */ 844 lwkt_gettoken(&hmp->fs_token); 845 flags = 0; 846 if (mntflags & MNT_FORCE) 847 flags |= FORCECLOSE; 848 error = vflush(mp, 0, flags); 849 850 /* 851 * Clean up the internal mount structure and related entities. This 852 * may issue I/O. 853 */ 854 if (error == 0) { 855 /* called with fs_token held */ 856 hammer_free_hmp(mp); 857 } else { 858 lwkt_reltoken(&hmp->fs_token); 859 } 860 return(error); 861 } 862 863 /* 864 * Clean up the internal mount structure and disassociate it from the mount. 865 * This may issue I/O. 866 * 867 * Called with fs_token held. 868 */ 869 static void 870 hammer_free_hmp(struct mount *mp) 871 { 872 hammer_mount_t hmp = (void *)mp->mnt_data; 873 hammer_flush_group_t flg; 874 875 /* 876 * Flush anything dirty. This won't even run if the 877 * filesystem errored-out. 878 */ 879 hammer_flush_dirty(hmp, 30); 880 881 /* 882 * If the mount had a critical error we have to destroy any 883 * remaining inodes before we can finish cleaning up the flusher. 884 */ 885 if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) { 886 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL, 887 hammer_destroy_inode_callback, NULL); 888 } 889 890 /* 891 * There shouldn't be any inodes left now and any left over 892 * flush groups should now be empty. 893 */ 894 KKASSERT(RB_EMPTY(&hmp->rb_inos_root)); 895 while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) { 896 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry); 897 KKASSERT(RB_EMPTY(&flg->flush_tree)); 898 if (flg->refs) { 899 hkprintf("Warning, flush_group %p was " 900 "not empty on umount!\n", flg); 901 } 902 kfree(flg, hmp->m_misc); 903 } 904 905 /* 906 * We can finally destroy the flusher 907 */ 908 hammer_flusher_destroy(hmp); 909 910 /* 911 * We may have held recovered buffers due to a read-only mount. 912 * These must be discarded. 913 */ 914 if (hmp->ronly) 915 hammer_recover_flush_buffers(hmp, NULL, -1); 916 917 /* 918 * Unload buffers and then volumes 919 */ 920 RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL, 921 hammer_unload_buffer, NULL); 922 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL, 923 hammer_unload_volume, NULL); 924 925 mp->mnt_data = NULL; 926 mp->mnt_flag &= ~MNT_LOCAL; 927 hmp->mp = NULL; 928 hammer_destroy_objid_cache(hmp); 929 hammer_destroy_dedup_cache(hmp); 930 if (hmp->dedup_free_cache != NULL) { 931 kfree(hmp->dedup_free_cache, hmp->m_misc); 932 hmp->dedup_free_cache = NULL; 933 } 934 kmalloc_destroy(&hmp->m_misc); 935 kmalloc_destroy(&hmp->m_inodes); 936 lwkt_reltoken(&hmp->fs_token); 937 kfree(hmp, M_HAMMER); 938 } 939 940 /* 941 * Report critical errors. ip may be NULL. 942 */ 943 void 944 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip, 945 int error, const char *msg) 946 { 947 hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR; 948 949 hmkrateprintf(&hmp->krate, hmp, 950 "Critical error inode=%jd error=%d %s\n", 951 (intmax_t)(ip ? ip->obj_id : -1), 952 error, msg); 953 954 if (hmp->ronly == 0) { 955 hmp->ronly = 2; /* special errored read-only mode */ 956 hmp->mp->mnt_flag |= MNT_RDONLY; 957 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL, 958 hammer_adjust_volume_mode, NULL); 959 hmkprintf(hmp, "Forcing read-only mode\n"); 960 } 961 hmp->error = error; 962 if (hammer_debug_critical) 963 Debugger("Entering debugger"); 964 } 965 966 967 /* 968 * Obtain a vnode for the specified inode number. An exclusively locked 969 * vnode is returned. 970 */ 971 int 972 hammer_vfs_vget(struct mount *mp, struct vnode *dvp, 973 ino_t ino, struct vnode **vpp) 974 { 975 struct hammer_transaction trans; 976 struct hammer_mount *hmp = (void *)mp->mnt_data; 977 struct hammer_inode *ip; 978 int error; 979 u_int32_t localization; 980 981 lwkt_gettoken(&hmp->fs_token); 982 hammer_simple_transaction(&trans, hmp); 983 984 /* 985 * If a directory vnode is supplied (mainly NFS) then we can acquire 986 * the PFS domain from it. Otherwise we would only be able to vget 987 * inodes in the root PFS. 988 */ 989 if (dvp) { 990 localization = HAMMER_DEF_LOCALIZATION + 991 VTOI(dvp)->obj_localization; 992 } else { 993 localization = HAMMER_DEF_LOCALIZATION; 994 } 995 996 /* 997 * Lookup the requested HAMMER inode. The structure must be 998 * left unlocked while we manipulate the related vnode to avoid 999 * a deadlock. 1000 */ 1001 ip = hammer_get_inode(&trans, NULL, ino, 1002 hmp->asof, localization, 1003 0, &error); 1004 if (ip == NULL) { 1005 *vpp = NULL; 1006 } else { 1007 error = hammer_get_vnode(ip, vpp); 1008 hammer_rel_inode(ip, 0); 1009 } 1010 hammer_done_transaction(&trans); 1011 lwkt_reltoken(&hmp->fs_token); 1012 return (error); 1013 } 1014 1015 /* 1016 * Return the root vnode for the filesystem. 1017 * 1018 * HAMMER stores the root vnode in the hammer_mount structure so 1019 * getting it is easy. 1020 */ 1021 static int 1022 hammer_vfs_root(struct mount *mp, struct vnode **vpp) 1023 { 1024 int error; 1025 1026 error = hammer_vfs_vget(mp, NULL, HAMMER_OBJID_ROOT, vpp); 1027 return (error); 1028 } 1029 1030 static int 1031 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred) 1032 { 1033 struct hammer_mount *hmp = (void *)mp->mnt_data; 1034 hammer_volume_t volume; 1035 hammer_volume_ondisk_t ondisk; 1036 int error; 1037 int64_t bfree; 1038 int64_t breserved; 1039 1040 lwkt_gettoken(&hmp->fs_token); 1041 volume = hammer_get_root_volume(hmp, &error); 1042 if (error) { 1043 lwkt_reltoken(&hmp->fs_token); 1044 return(error); 1045 } 1046 ondisk = volume->ondisk; 1047 1048 /* 1049 * Basic stats 1050 */ 1051 _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved); 1052 mp->mnt_stat.f_files = ondisk->vol0_stat_inodes; 1053 bfree = ondisk->vol0_stat_freebigblocks * HAMMER_BIGBLOCK_SIZE; 1054 hammer_rel_volume(volume, 0); 1055 1056 mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE; 1057 mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree; 1058 if (mp->mnt_stat.f_files < 0) 1059 mp->mnt_stat.f_files = 0; 1060 1061 *sbp = mp->mnt_stat; 1062 lwkt_reltoken(&hmp->fs_token); 1063 return(0); 1064 } 1065 1066 static int 1067 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred) 1068 { 1069 struct hammer_mount *hmp = (void *)mp->mnt_data; 1070 hammer_volume_t volume; 1071 hammer_volume_ondisk_t ondisk; 1072 int error; 1073 int64_t bfree; 1074 int64_t breserved; 1075 1076 lwkt_gettoken(&hmp->fs_token); 1077 volume = hammer_get_root_volume(hmp, &error); 1078 if (error) { 1079 lwkt_reltoken(&hmp->fs_token); 1080 return(error); 1081 } 1082 ondisk = volume->ondisk; 1083 1084 /* 1085 * Basic stats 1086 */ 1087 _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved); 1088 mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes; 1089 bfree = ondisk->vol0_stat_freebigblocks * HAMMER_BIGBLOCK_SIZE; 1090 hammer_rel_volume(volume, 0); 1091 1092 mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE; 1093 mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree; 1094 if (mp->mnt_vstat.f_files < 0) 1095 mp->mnt_vstat.f_files = 0; 1096 *sbp = mp->mnt_vstat; 1097 lwkt_reltoken(&hmp->fs_token); 1098 return(0); 1099 } 1100 1101 /* 1102 * Sync the filesystem. Currently we have to run it twice, the second 1103 * one will advance the undo start index to the end index, so if a crash 1104 * occurs no undos will be run on mount. 1105 * 1106 * We do not sync the filesystem if we are called from a panic. If we did 1107 * we might end up blowing up a sync that was already in progress. 1108 */ 1109 static int 1110 hammer_vfs_sync(struct mount *mp, int waitfor) 1111 { 1112 struct hammer_mount *hmp = (void *)mp->mnt_data; 1113 int error; 1114 1115 lwkt_gettoken(&hmp->fs_token); 1116 if (panicstr == NULL) { 1117 error = hammer_sync_hmp(hmp, waitfor); 1118 } else { 1119 error = EIO; 1120 } 1121 lwkt_reltoken(&hmp->fs_token); 1122 return (error); 1123 } 1124 1125 /* 1126 * Convert a vnode to a file handle. 1127 * 1128 * Accesses read-only fields on already-referenced structures so 1129 * no token is needed. 1130 */ 1131 static int 1132 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp) 1133 { 1134 hammer_inode_t ip; 1135 1136 KKASSERT(MAXFIDSZ >= 16); 1137 ip = VTOI(vp); 1138 fhp->fid_len = offsetof(struct fid, fid_data[16]); 1139 fhp->fid_ext = ip->obj_localization >> 16; 1140 bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id)); 1141 bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof)); 1142 return(0); 1143 } 1144 1145 1146 /* 1147 * Convert a file handle back to a vnode. 1148 * 1149 * Use rootvp to enforce PFS isolation when a PFS is exported via a 1150 * null mount. 1151 */ 1152 static int 1153 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp, 1154 struct fid *fhp, struct vnode **vpp) 1155 { 1156 hammer_mount_t hmp = (void *)mp->mnt_data; 1157 struct hammer_transaction trans; 1158 struct hammer_inode *ip; 1159 struct hammer_inode_info info; 1160 int error; 1161 u_int32_t localization; 1162 1163 bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id)); 1164 bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof)); 1165 if (rootvp) 1166 localization = VTOI(rootvp)->obj_localization; 1167 else 1168 localization = (u_int32_t)fhp->fid_ext << 16; 1169 1170 lwkt_gettoken(&hmp->fs_token); 1171 hammer_simple_transaction(&trans, hmp); 1172 1173 /* 1174 * Get/allocate the hammer_inode structure. The structure must be 1175 * unlocked while we manipulate the related vnode to avoid a 1176 * deadlock. 1177 */ 1178 ip = hammer_get_inode(&trans, NULL, info.obj_id, 1179 info.obj_asof, localization, 0, &error); 1180 if (ip) { 1181 error = hammer_get_vnode(ip, vpp); 1182 hammer_rel_inode(ip, 0); 1183 } else { 1184 *vpp = NULL; 1185 } 1186 hammer_done_transaction(&trans); 1187 lwkt_reltoken(&hmp->fs_token); 1188 return (error); 1189 } 1190 1191 static int 1192 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam, 1193 int *exflagsp, struct ucred **credanonp) 1194 { 1195 hammer_mount_t hmp = (void *)mp->mnt_data; 1196 struct netcred *np; 1197 int error; 1198 1199 lwkt_gettoken(&hmp->fs_token); 1200 np = vfs_export_lookup(mp, &hmp->export, nam); 1201 if (np) { 1202 *exflagsp = np->netc_exflags; 1203 *credanonp = &np->netc_anon; 1204 error = 0; 1205 } else { 1206 error = EACCES; 1207 } 1208 lwkt_reltoken(&hmp->fs_token); 1209 return (error); 1210 1211 } 1212 1213 int 1214 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export) 1215 { 1216 hammer_mount_t hmp = (void *)mp->mnt_data; 1217 int error; 1218 1219 lwkt_gettoken(&hmp->fs_token); 1220 1221 switch(op) { 1222 case MOUNTCTL_SET_EXPORT: 1223 error = vfs_export(mp, &hmp->export, export); 1224 break; 1225 default: 1226 error = EOPNOTSUPP; 1227 break; 1228 } 1229 lwkt_reltoken(&hmp->fs_token); 1230 1231 return(error); 1232 } 1233 1234