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 * $DragonFly: src/sys/vfs/hammer/hammer_vfsops.c,v 1.65 2008/07/18 00:19:53 dillon Exp $ 35 */ 36 37 #include <sys/param.h> 38 #include <sys/systm.h> 39 #include <sys/kernel.h> 40 #include <sys/vnode.h> 41 #include <sys/mount.h> 42 #include <sys/malloc.h> 43 #include <sys/nlookup.h> 44 #include <sys/fcntl.h> 45 #include <sys/sysctl.h> 46 #include <sys/buf.h> 47 #include <sys/buf2.h> 48 #include "hammer.h" 49 50 int hammer_debug_io; 51 int hammer_debug_general; 52 int hammer_debug_debug = 1; /* medium-error panics */ 53 int hammer_debug_inode; 54 int hammer_debug_locks; 55 int hammer_debug_btree; 56 int hammer_debug_tid; 57 int hammer_debug_recover; /* -1 will disable, +1 will force */ 58 int hammer_debug_recover_faults; 59 int hammer_cluster_enable = 1; /* enable read clustering by default */ 60 int hammer_count_fsyncs; 61 int hammer_count_inodes; 62 int hammer_count_iqueued; 63 int hammer_count_reclaiming; 64 int hammer_count_records; 65 int hammer_count_record_datas; 66 int hammer_count_volumes; 67 int hammer_count_buffers; 68 int hammer_count_nodes; 69 int64_t hammer_count_extra_space_used; 70 int64_t hammer_stats_btree_lookups; 71 int64_t hammer_stats_btree_searches; 72 int64_t hammer_stats_btree_inserts; 73 int64_t hammer_stats_btree_deletes; 74 int64_t hammer_stats_btree_elements; 75 int64_t hammer_stats_btree_splits; 76 int64_t hammer_stats_btree_iterations; 77 int64_t hammer_stats_record_iterations; 78 79 int64_t hammer_stats_file_read; 80 int64_t hammer_stats_file_write; 81 int64_t hammer_stats_file_iopsr; 82 int64_t hammer_stats_file_iopsw; 83 int64_t hammer_stats_disk_read; 84 int64_t hammer_stats_disk_write; 85 int64_t hammer_stats_inode_flushes; 86 int64_t hammer_stats_commits; 87 88 int hammer_count_dirtybufspace; /* global */ 89 int hammer_count_refedbufs; /* global */ 90 int hammer_count_reservations; 91 int hammer_count_io_running_read; 92 int hammer_count_io_running_write; 93 int hammer_count_io_locked; 94 int hammer_limit_dirtybufspace; /* per-mount */ 95 int hammer_limit_recs; /* as a whole XXX */ 96 int hammer_limit_iqueued; /* per-mount */ 97 int hammer_bio_count; 98 int hammer_verify_zone; 99 int hammer_verify_data = 1; 100 int hammer_write_mode; 101 int64_t hammer_contention_count; 102 int64_t hammer_zone_limit; 103 104 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem"); 105 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW, 106 &hammer_debug_general, 0, ""); 107 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW, 108 &hammer_debug_io, 0, ""); 109 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW, 110 &hammer_debug_debug, 0, ""); 111 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW, 112 &hammer_debug_inode, 0, ""); 113 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW, 114 &hammer_debug_locks, 0, ""); 115 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW, 116 &hammer_debug_btree, 0, ""); 117 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW, 118 &hammer_debug_tid, 0, ""); 119 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW, 120 &hammer_debug_recover, 0, ""); 121 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW, 122 &hammer_debug_recover_faults, 0, ""); 123 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW, 124 &hammer_cluster_enable, 0, ""); 125 126 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW, 127 &hammer_limit_dirtybufspace, 0, ""); 128 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW, 129 &hammer_limit_recs, 0, ""); 130 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_iqueued, CTLFLAG_RW, 131 &hammer_limit_iqueued, 0, ""); 132 133 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD, 134 &hammer_count_fsyncs, 0, ""); 135 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD, 136 &hammer_count_inodes, 0, ""); 137 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD, 138 &hammer_count_iqueued, 0, ""); 139 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaiming, CTLFLAG_RD, 140 &hammer_count_reclaiming, 0, ""); 141 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD, 142 &hammer_count_records, 0, ""); 143 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD, 144 &hammer_count_record_datas, 0, ""); 145 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD, 146 &hammer_count_volumes, 0, ""); 147 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD, 148 &hammer_count_buffers, 0, ""); 149 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD, 150 &hammer_count_nodes, 0, ""); 151 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD, 152 &hammer_count_extra_space_used, 0, ""); 153 154 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD, 155 &hammer_stats_btree_searches, 0, ""); 156 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD, 157 &hammer_stats_btree_lookups, 0, ""); 158 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD, 159 &hammer_stats_btree_inserts, 0, ""); 160 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD, 161 &hammer_stats_btree_deletes, 0, ""); 162 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD, 163 &hammer_stats_btree_elements, 0, ""); 164 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD, 165 &hammer_stats_btree_splits, 0, ""); 166 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD, 167 &hammer_stats_btree_iterations, 0, ""); 168 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD, 169 &hammer_stats_record_iterations, 0, ""); 170 171 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD, 172 &hammer_stats_file_read, 0, ""); 173 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD, 174 &hammer_stats_file_write, 0, ""); 175 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD, 176 &hammer_stats_file_iopsr, 0, ""); 177 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD, 178 &hammer_stats_file_iopsw, 0, ""); 179 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD, 180 &hammer_stats_disk_read, 0, ""); 181 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD, 182 &hammer_stats_disk_write, 0, ""); 183 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD, 184 &hammer_stats_inode_flushes, 0, ""); 185 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD, 186 &hammer_stats_commits, 0, ""); 187 188 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD, 189 &hammer_count_dirtybufspace, 0, ""); 190 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD, 191 &hammer_count_refedbufs, 0, ""); 192 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD, 193 &hammer_count_reservations, 0, ""); 194 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD, 195 &hammer_count_io_running_read, 0, ""); 196 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD, 197 &hammer_count_io_locked, 0, ""); 198 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD, 199 &hammer_count_io_running_write, 0, ""); 200 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW, 201 &hammer_zone_limit, 0, ""); 202 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW, 203 &hammer_contention_count, 0, ""); 204 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW, 205 &hammer_verify_zone, 0, ""); 206 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW, 207 &hammer_verify_data, 0, ""); 208 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW, 209 &hammer_write_mode, 0, ""); 210 211 KTR_INFO_MASTER(hammer); 212 213 /* 214 * VFS ABI 215 */ 216 static void hammer_free_hmp(struct mount *mp); 217 218 static int hammer_vfs_mount(struct mount *mp, char *path, caddr_t data, 219 struct ucred *cred); 220 static int hammer_vfs_unmount(struct mount *mp, int mntflags); 221 static int hammer_vfs_root(struct mount *mp, struct vnode **vpp); 222 static int hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, 223 struct ucred *cred); 224 static int hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, 225 struct ucred *cred); 226 static int hammer_vfs_sync(struct mount *mp, int waitfor); 227 static int hammer_vfs_vget(struct mount *mp, ino_t ino, 228 struct vnode **vpp); 229 static int hammer_vfs_init(struct vfsconf *conf); 230 static int hammer_vfs_fhtovp(struct mount *mp, struct fid *fhp, 231 struct vnode **vpp); 232 static int hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp); 233 static int hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam, 234 int *exflagsp, struct ucred **credanonp); 235 236 237 static struct vfsops hammer_vfsops = { 238 .vfs_mount = hammer_vfs_mount, 239 .vfs_unmount = hammer_vfs_unmount, 240 .vfs_root = hammer_vfs_root, 241 .vfs_statfs = hammer_vfs_statfs, 242 .vfs_statvfs = hammer_vfs_statvfs, 243 .vfs_sync = hammer_vfs_sync, 244 .vfs_vget = hammer_vfs_vget, 245 .vfs_init = hammer_vfs_init, 246 .vfs_vptofh = hammer_vfs_vptofh, 247 .vfs_fhtovp = hammer_vfs_fhtovp, 248 .vfs_checkexp = hammer_vfs_checkexp 249 }; 250 251 MALLOC_DEFINE(M_HAMMER, "hammer-mount", "hammer mount"); 252 253 VFS_SET(hammer_vfsops, hammer, 0); 254 MODULE_VERSION(hammer, 1); 255 256 static int 257 hammer_vfs_init(struct vfsconf *conf) 258 { 259 int n; 260 261 if (hammer_limit_recs == 0) { 262 hammer_limit_recs = nbuf * 25; 263 n = kmalloc_limit(M_HAMMER) / 512; 264 if (hammer_limit_recs > n) 265 hammer_limit_recs = n; 266 } 267 if (hammer_limit_dirtybufspace == 0) { 268 hammer_limit_dirtybufspace = hidirtybufspace / 2; 269 if (hammer_limit_dirtybufspace < 100) 270 hammer_limit_dirtybufspace = 100; 271 } 272 if (hammer_limit_iqueued == 0) 273 hammer_limit_iqueued = desiredvnodes / 5; 274 return(0); 275 } 276 277 static int 278 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data, 279 struct ucred *cred) 280 { 281 struct hammer_mount_info info; 282 hammer_mount_t hmp; 283 hammer_volume_t rootvol; 284 struct vnode *rootvp; 285 const char *upath; /* volume name in userspace */ 286 char *path; /* volume name in system space */ 287 int error; 288 int i; 289 290 if ((error = copyin(data, &info, sizeof(info))) != 0) 291 return (error); 292 if ((mp->mnt_flag & MNT_UPDATE) == 0) { 293 if (info.nvolumes <= 0 || info.nvolumes >= 32768) 294 return (EINVAL); 295 } 296 if ((info.hflags & HMNT_MASTERID) && 297 (info.masterid < -1 || info.masterid >= HAMMER_MAX_MASTERS)) { 298 return (EINVAL); 299 } 300 301 /* 302 * Interal mount data structure 303 */ 304 if (mp->mnt_flag & MNT_UPDATE) { 305 hmp = (void *)mp->mnt_data; 306 KKASSERT(hmp != NULL); 307 } else { 308 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO); 309 mp->mnt_data = (qaddr_t)hmp; 310 hmp->mp = mp; 311 hmp->namekey_iterator = mycpu->gd_time_seconds; 312 /*TAILQ_INIT(&hmp->recycle_list);*/ 313 314 hmp->root_btree_beg.localization = 0x00000000U; 315 hmp->root_btree_beg.obj_id = -0x8000000000000000LL; 316 hmp->root_btree_beg.key = -0x8000000000000000LL; 317 hmp->root_btree_beg.create_tid = 1; 318 hmp->root_btree_beg.delete_tid = 1; 319 hmp->root_btree_beg.rec_type = 0; 320 hmp->root_btree_beg.obj_type = 0; 321 322 hmp->root_btree_end.localization = 0xFFFFFFFFU; 323 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL; 324 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL; 325 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL; 326 hmp->root_btree_end.delete_tid = 0; /* special case */ 327 hmp->root_btree_end.rec_type = 0xFFFFU; 328 hmp->root_btree_end.obj_type = 0; 329 330 hmp->krate.freq = 1; /* maximum reporting rate (hz) */ 331 hmp->krate.count = -16; /* initial burst */ 332 333 hmp->sync_lock.refs = 1; 334 hmp->free_lock.refs = 1; 335 hmp->undo_lock.refs = 1; 336 hmp->blkmap_lock.refs = 1; 337 338 TAILQ_INIT(&hmp->delay_list); 339 TAILQ_INIT(&hmp->flush_group_list); 340 TAILQ_INIT(&hmp->objid_cache_list); 341 TAILQ_INIT(&hmp->undo_lru_list); 342 TAILQ_INIT(&hmp->reclaim_list); 343 } 344 hmp->hflags &= ~HMNT_USERFLAGS; 345 hmp->hflags |= info.hflags & HMNT_USERFLAGS; 346 if (info.hflags & HMNT_MASTERID) 347 hmp->masterid = info.masterid; 348 else 349 hmp->masterid = -1; 350 if (info.asof) { 351 kprintf("ASOF\n"); 352 mp->mnt_flag |= MNT_RDONLY; 353 hmp->asof = info.asof; 354 } else { 355 hmp->asof = HAMMER_MAX_TID; 356 } 357 358 /* 359 * Re-open read-write if originally read-only, or vise-versa. 360 */ 361 if (mp->mnt_flag & MNT_UPDATE) { 362 error = 0; 363 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) { 364 kprintf("HAMMER read-only -> read-write\n"); 365 hmp->ronly = 0; 366 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL, 367 hammer_adjust_volume_mode, NULL); 368 rootvol = hammer_get_root_volume(hmp, &error); 369 if (rootvol) { 370 hammer_recover_flush_buffers(hmp, rootvol, 1); 371 bcopy(rootvol->ondisk->vol0_blockmap, 372 hmp->blockmap, 373 sizeof(hmp->blockmap)); 374 hammer_rel_volume(rootvol, 0); 375 } 376 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL, 377 hammer_reload_inode, NULL); 378 /* kernel clears MNT_RDONLY */ 379 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) { 380 kprintf("HAMMER read-write -> read-only\n"); 381 hmp->ronly = 1; /* messy */ 382 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL, 383 hammer_reload_inode, NULL); 384 hmp->ronly = 0; 385 hammer_flusher_sync(hmp); 386 hammer_flusher_sync(hmp); 387 hammer_flusher_sync(hmp); 388 hmp->ronly = 1; 389 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL, 390 hammer_adjust_volume_mode, NULL); 391 } 392 return(error); 393 } 394 395 RB_INIT(&hmp->rb_vols_root); 396 RB_INIT(&hmp->rb_inos_root); 397 RB_INIT(&hmp->rb_nods_root); 398 RB_INIT(&hmp->rb_undo_root); 399 RB_INIT(&hmp->rb_resv_root); 400 RB_INIT(&hmp->rb_bufs_root); 401 RB_INIT(&hmp->rb_pfsm_root); 402 403 hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0); 404 405 TAILQ_INIT(&hmp->volu_list); 406 TAILQ_INIT(&hmp->undo_list); 407 TAILQ_INIT(&hmp->data_list); 408 TAILQ_INIT(&hmp->meta_list); 409 TAILQ_INIT(&hmp->lose_list); 410 411 /* 412 * Load volumes 413 */ 414 path = objcache_get(namei_oc, M_WAITOK); 415 hmp->nvolumes = -1; 416 for (i = 0; i < info.nvolumes; ++i) { 417 error = copyin(&info.volumes[i], &upath, sizeof(char *)); 418 if (error == 0) 419 error = copyinstr(upath, path, MAXPATHLEN, NULL); 420 if (error == 0) 421 error = hammer_install_volume(hmp, path); 422 if (error) 423 break; 424 } 425 objcache_put(namei_oc, path); 426 427 /* 428 * Make sure we found a root volume 429 */ 430 if (error == 0 && hmp->rootvol == NULL) { 431 kprintf("hammer_mount: No root volume found!\n"); 432 error = EINVAL; 433 } 434 435 /* 436 * Check that all required volumes are available 437 */ 438 if (error == 0 && hammer_mountcheck_volumes(hmp)) { 439 kprintf("hammer_mount: Missing volumes, cannot mount!\n"); 440 error = EINVAL; 441 } 442 443 if (error) { 444 hammer_free_hmp(mp); 445 return (error); 446 } 447 448 /* 449 * No errors, setup enough of the mount point so we can lookup the 450 * root vnode. 451 */ 452 mp->mnt_iosize_max = MAXPHYS; 453 mp->mnt_kern_flag |= MNTK_FSMID; 454 455 /* 456 * note: f_iosize is used by vnode_pager_haspage() when constructing 457 * its VOP_BMAP call. 458 */ 459 mp->mnt_stat.f_iosize = HAMMER_BUFSIZE; 460 mp->mnt_stat.f_bsize = HAMMER_BUFSIZE; 461 462 mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE; 463 mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE; 464 465 mp->mnt_maxsymlinklen = 255; 466 mp->mnt_flag |= MNT_LOCAL; 467 468 vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops); 469 vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops); 470 vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops); 471 472 /* 473 * The root volume's ondisk pointer is only valid if we hold a 474 * reference to it. 475 */ 476 rootvol = hammer_get_root_volume(hmp, &error); 477 if (error) 478 goto failed; 479 480 /* 481 * Perform any necessary UNDO operations. The recovery code does 482 * call hammer_undo_lookup() so we have to pre-cache the blockmap, 483 * and then re-copy it again after recovery is complete. 484 * 485 * If this is a read-only mount the UNDO information is retained 486 * in memory in the form of dirty buffer cache buffers, and not 487 * written back to the media. 488 */ 489 bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap, 490 sizeof(hmp->blockmap)); 491 492 /* 493 * The undo_rec_limit limits the size of flush groups to avoid 494 * blowing out the UNDO FIFO. This calculation is typically in 495 * the tens of thousands and is designed primarily when small 496 * HAMMER filesystems are created. 497 */ 498 hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100; 499 if (hammer_debug_general & 0x0001) 500 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit); 501 502 error = hammer_recover(hmp, rootvol); 503 if (error) { 504 kprintf("Failed to recover HAMMER filesystem on mount\n"); 505 goto done; 506 } 507 508 /* 509 * Finish setup now that we have a good root volume. 510 * 511 * The top 16 bits of fsid.val[1] is a pfs id. 512 */ 513 ksnprintf(mp->mnt_stat.f_mntfromname, 514 sizeof(mp->mnt_stat.f_mntfromname), "%s", 515 rootvol->ondisk->vol_name); 516 mp->mnt_stat.f_fsid.val[0] = 517 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8); 518 mp->mnt_stat.f_fsid.val[1] = 519 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8); 520 mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF; 521 522 mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid; 523 mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid, 524 sizeof(mp->mnt_vstat.f_fsid_uuid)); 525 526 /* 527 * Certain often-modified fields in the root volume are cached in 528 * the hammer_mount structure so we do not have to generate lots 529 * of little UNDO structures for them. 530 * 531 * Recopy after recovery. This also has the side effect of 532 * setting our cached undo FIFO's first_offset, which serves to 533 * placemark the FIFO start for the NEXT flush cycle while the 534 * on-disk first_offset represents the LAST flush cycle. 535 */ 536 hmp->next_tid = rootvol->ondisk->vol0_next_tid; 537 bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap, 538 sizeof(hmp->blockmap)); 539 hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks; 540 541 hammer_flusher_create(hmp); 542 543 /* 544 * Locate the root directory using the root cluster's B-Tree as a 545 * starting point. The root directory uses an obj_id of 1. 546 * 547 * FUTURE: Leave the root directory cached referenced but unlocked 548 * in hmp->rootvp (need to flush it on unmount). 549 */ 550 error = hammer_vfs_vget(mp, 1, &rootvp); 551 if (error) 552 goto done; 553 vput(rootvp); 554 /*vn_unlock(hmp->rootvp);*/ 555 556 done: 557 hammer_rel_volume(rootvol, 0); 558 failed: 559 /* 560 * Cleanup and return. 561 */ 562 if (error) 563 hammer_free_hmp(mp); 564 return (error); 565 } 566 567 static int 568 hammer_vfs_unmount(struct mount *mp, int mntflags) 569 { 570 #if 0 571 struct hammer_mount *hmp = (void *)mp->mnt_data; 572 #endif 573 int flags; 574 int error; 575 576 /* 577 * Clean out the vnodes 578 */ 579 flags = 0; 580 if (mntflags & MNT_FORCE) 581 flags |= FORCECLOSE; 582 if ((error = vflush(mp, 0, flags)) != 0) 583 return (error); 584 585 /* 586 * Clean up the internal mount structure and related entities. This 587 * may issue I/O. 588 */ 589 hammer_free_hmp(mp); 590 return(0); 591 } 592 593 /* 594 * Clean up the internal mount structure and disassociate it from the mount. 595 * This may issue I/O. 596 */ 597 static void 598 hammer_free_hmp(struct mount *mp) 599 { 600 struct hammer_mount *hmp = (void *)mp->mnt_data; 601 hammer_flush_group_t flg; 602 int count; 603 604 /* 605 * Flush anything dirty. This won't even run if the 606 * filesystem errored-out. 607 */ 608 count = 0; 609 while (hammer_flusher_haswork(hmp)) { 610 hammer_flusher_sync(hmp); 611 ++count; 612 if (count >= 5) { 613 if (count == 5) 614 kprintf("HAMMER: umount flushing."); 615 else 616 kprintf("."); 617 tsleep(hmp, 0, "hmrufl", hz); 618 } 619 if (count == 30) { 620 kprintf("giving up\n"); 621 break; 622 } 623 } 624 if (count >= 5 && count < 30) 625 kprintf("\n"); 626 627 /* 628 * If the mount had a critical error we have to destroy any 629 * remaining inodes before we can finish cleaning up the flusher. 630 */ 631 if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) { 632 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL, 633 hammer_destroy_inode_callback, NULL); 634 } 635 636 /* 637 * There shouldn't be any inodes left now and any left over 638 * flush groups should now be empty. 639 */ 640 KKASSERT(RB_EMPTY(&hmp->rb_inos_root)); 641 while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) { 642 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry); 643 KKASSERT(TAILQ_EMPTY(&flg->flush_list)); 644 if (flg->refs) { 645 kprintf("HAMMER: Warning, flush_group %p was " 646 "not empty on umount!\n", flg); 647 } 648 kfree(flg, M_HAMMER); 649 } 650 651 /* 652 * We can finally destroy the flusher 653 */ 654 hammer_flusher_destroy(hmp); 655 656 /* 657 * Unload buffers and then volumes 658 */ 659 RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL, 660 hammer_unload_buffer, NULL); 661 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL, 662 hammer_unload_volume, NULL); 663 664 mp->mnt_data = NULL; 665 mp->mnt_flag &= ~MNT_LOCAL; 666 hmp->mp = NULL; 667 hammer_destroy_objid_cache(hmp); 668 kfree(hmp, M_HAMMER); 669 } 670 671 /* 672 * Report critical errors. ip may be NULL. 673 */ 674 void 675 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip, 676 int error, const char *msg) 677 { 678 hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR; 679 krateprintf(&hmp->krate, 680 "HAMMER(%s): Critical error inode=%lld %s\n", 681 hmp->mp->mnt_stat.f_mntfromname, 682 (ip ? ip->obj_id : -1), msg); 683 if (hmp->ronly == 0) { 684 hmp->ronly = 2; /* special errored read-only mode */ 685 hmp->mp->mnt_flag |= MNT_RDONLY; 686 kprintf("HAMMER(%s): Forcing read-only mode\n", 687 hmp->mp->mnt_stat.f_mntfromname); 688 } 689 hmp->error = error; 690 } 691 692 693 /* 694 * Obtain a vnode for the specified inode number. An exclusively locked 695 * vnode is returned. 696 */ 697 int 698 hammer_vfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp) 699 { 700 struct hammer_transaction trans; 701 struct hammer_mount *hmp = (void *)mp->mnt_data; 702 struct hammer_inode *ip; 703 int error; 704 705 hammer_simple_transaction(&trans, hmp); 706 707 /* 708 * Lookup the requested HAMMER inode. The structure must be 709 * left unlocked while we manipulate the related vnode to avoid 710 * a deadlock. 711 */ 712 ip = hammer_get_inode(&trans, NULL, ino, 713 hmp->asof, HAMMER_DEF_LOCALIZATION, 714 0, &error); 715 if (ip == NULL) { 716 *vpp = NULL; 717 return(error); 718 } 719 error = hammer_get_vnode(ip, vpp); 720 hammer_rel_inode(ip, 0); 721 hammer_done_transaction(&trans); 722 return (error); 723 } 724 725 /* 726 * Return the root vnode for the filesystem. 727 * 728 * HAMMER stores the root vnode in the hammer_mount structure so 729 * getting it is easy. 730 */ 731 static int 732 hammer_vfs_root(struct mount *mp, struct vnode **vpp) 733 { 734 #if 0 735 struct hammer_mount *hmp = (void *)mp->mnt_data; 736 #endif 737 int error; 738 739 error = hammer_vfs_vget(mp, 1, vpp); 740 return (error); 741 } 742 743 static int 744 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred) 745 { 746 struct hammer_mount *hmp = (void *)mp->mnt_data; 747 hammer_volume_t volume; 748 hammer_volume_ondisk_t ondisk; 749 int error; 750 int64_t bfree; 751 752 volume = hammer_get_root_volume(hmp, &error); 753 if (error) 754 return(error); 755 ondisk = volume->ondisk; 756 757 /* 758 * Basic stats 759 */ 760 mp->mnt_stat.f_files = ondisk->vol0_stat_inodes; 761 bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE; 762 hammer_rel_volume(volume, 0); 763 764 mp->mnt_stat.f_bfree = bfree / HAMMER_BUFSIZE; 765 mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree; 766 if (mp->mnt_stat.f_files < 0) 767 mp->mnt_stat.f_files = 0; 768 769 *sbp = mp->mnt_stat; 770 return(0); 771 } 772 773 static int 774 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred) 775 { 776 struct hammer_mount *hmp = (void *)mp->mnt_data; 777 hammer_volume_t volume; 778 hammer_volume_ondisk_t ondisk; 779 int error; 780 int64_t bfree; 781 782 volume = hammer_get_root_volume(hmp, &error); 783 if (error) 784 return(error); 785 ondisk = volume->ondisk; 786 787 /* 788 * Basic stats 789 */ 790 mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes; 791 bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE; 792 hammer_rel_volume(volume, 0); 793 794 mp->mnt_vstat.f_bfree = bfree / HAMMER_BUFSIZE; 795 mp->mnt_vstat.f_bavail = mp->mnt_stat.f_bfree; 796 if (mp->mnt_vstat.f_files < 0) 797 mp->mnt_vstat.f_files = 0; 798 *sbp = mp->mnt_vstat; 799 return(0); 800 } 801 802 /* 803 * Sync the filesystem. Currently we have to run it twice, the second 804 * one will advance the undo start index to the end index, so if a crash 805 * occurs no undos will be run on mount. 806 * 807 * We do not sync the filesystem if we are called from a panic. If we did 808 * we might end up blowing up a sync that was already in progress. 809 */ 810 static int 811 hammer_vfs_sync(struct mount *mp, int waitfor) 812 { 813 struct hammer_mount *hmp = (void *)mp->mnt_data; 814 int error; 815 816 if (panicstr == NULL) { 817 error = hammer_sync_hmp(hmp, waitfor); 818 if (error == 0) 819 error = hammer_sync_hmp(hmp, waitfor); 820 } else { 821 error = EIO; 822 } 823 return (error); 824 } 825 826 /* 827 * Convert a vnode to a file handle. 828 */ 829 static int 830 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp) 831 { 832 hammer_inode_t ip; 833 834 KKASSERT(MAXFIDSZ >= 16); 835 ip = VTOI(vp); 836 fhp->fid_len = offsetof(struct fid, fid_data[16]); 837 fhp->fid_ext = ip->obj_localization >> 16; 838 bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id)); 839 bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof)); 840 return(0); 841 } 842 843 844 /* 845 * Convert a file handle back to a vnode. 846 */ 847 static int 848 hammer_vfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp) 849 { 850 struct hammer_transaction trans; 851 struct hammer_inode *ip; 852 struct hammer_inode_info info; 853 int error; 854 u_int32_t localization; 855 856 bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id)); 857 bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof)); 858 localization = (u_int32_t)fhp->fid_ext << 16; 859 860 hammer_simple_transaction(&trans, (void *)mp->mnt_data); 861 862 /* 863 * Get/allocate the hammer_inode structure. The structure must be 864 * unlocked while we manipulate the related vnode to avoid a 865 * deadlock. 866 */ 867 ip = hammer_get_inode(&trans, NULL, info.obj_id, 868 info.obj_asof, localization, 0, &error); 869 if (ip == NULL) { 870 *vpp = NULL; 871 return(error); 872 } 873 error = hammer_get_vnode(ip, vpp); 874 hammer_rel_inode(ip, 0); 875 hammer_done_transaction(&trans); 876 return (error); 877 } 878 879 static int 880 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam, 881 int *exflagsp, struct ucred **credanonp) 882 { 883 hammer_mount_t hmp = (void *)mp->mnt_data; 884 struct netcred *np; 885 int error; 886 887 np = vfs_export_lookup(mp, &hmp->export, nam); 888 if (np) { 889 *exflagsp = np->netc_exflags; 890 *credanonp = &np->netc_anon; 891 error = 0; 892 } else { 893 error = EACCES; 894 } 895 return (error); 896 897 } 898 899 int 900 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export) 901 { 902 hammer_mount_t hmp = (void *)mp->mnt_data; 903 int error; 904 905 switch(op) { 906 case MOUNTCTL_SET_EXPORT: 907 error = vfs_export(mp, &hmp->export, export); 908 break; 909 default: 910 error = EOPNOTSUPP; 911 break; 912 } 913 return(error); 914 } 915 916