vfs_wapbl.c revision 1.7
1 1.7 joerg /* $NetBSD: vfs_wapbl.c,v 1.7 2008/11/17 19:31:47 joerg Exp $ */ 2 1.2 simonb 3 1.2 simonb /*- 4 1.2 simonb * Copyright (c) 2003,2008 The NetBSD Foundation, Inc. 5 1.2 simonb * All rights reserved. 6 1.2 simonb * 7 1.2 simonb * This code is derived from software contributed to The NetBSD Foundation 8 1.2 simonb * by Wasabi Systems, Inc. 9 1.2 simonb * 10 1.2 simonb * Redistribution and use in source and binary forms, with or without 11 1.2 simonb * modification, are permitted provided that the following conditions 12 1.2 simonb * are met: 13 1.2 simonb * 1. Redistributions of source code must retain the above copyright 14 1.2 simonb * notice, this list of conditions and the following disclaimer. 15 1.2 simonb * 2. Redistributions in binary form must reproduce the above copyright 16 1.2 simonb * notice, this list of conditions and the following disclaimer in the 17 1.2 simonb * documentation and/or other materials provided with the distribution. 18 1.2 simonb * 19 1.2 simonb * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 1.2 simonb * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 1.2 simonb * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 1.2 simonb * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 1.2 simonb * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 1.2 simonb * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 1.2 simonb * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 1.2 simonb * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 1.2 simonb * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 1.2 simonb * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 1.2 simonb * POSSIBILITY OF SUCH DAMAGE. 30 1.2 simonb */ 31 1.2 simonb 32 1.2 simonb /* 33 1.2 simonb * This implements file system independent write ahead filesystem logging. 34 1.2 simonb */ 35 1.4 joerg 36 1.4 joerg #define WAPBL_INTERNAL 37 1.4 joerg 38 1.2 simonb #include <sys/cdefs.h> 39 1.7 joerg __KERNEL_RCSID(0, "$NetBSD: vfs_wapbl.c,v 1.7 2008/11/17 19:31:47 joerg Exp $"); 40 1.2 simonb 41 1.2 simonb #include <sys/param.h> 42 1.2 simonb 43 1.2 simonb #ifdef _KERNEL 44 1.2 simonb #include <sys/param.h> 45 1.2 simonb #include <sys/namei.h> 46 1.2 simonb #include <sys/proc.h> 47 1.2 simonb #include <sys/uio.h> 48 1.2 simonb #include <sys/vnode.h> 49 1.2 simonb #include <sys/file.h> 50 1.2 simonb #include <sys/malloc.h> 51 1.2 simonb #include <sys/resourcevar.h> 52 1.2 simonb #include <sys/conf.h> 53 1.2 simonb #include <sys/mount.h> 54 1.2 simonb #include <sys/kernel.h> 55 1.2 simonb #include <sys/kauth.h> 56 1.2 simonb #include <sys/mutex.h> 57 1.2 simonb #include <sys/atomic.h> 58 1.2 simonb #include <sys/wapbl.h> 59 1.2 simonb 60 1.2 simonb #if WAPBL_UVM_ALLOC 61 1.2 simonb #include <uvm/uvm.h> 62 1.2 simonb #endif 63 1.2 simonb 64 1.2 simonb #include <miscfs/specfs/specdev.h> 65 1.2 simonb 66 1.2 simonb MALLOC_JUSTDEFINE(M_WAPBL, "wapbl", "write-ahead physical block logging"); 67 1.2 simonb #define wapbl_malloc(s) malloc((s), M_WAPBL, M_WAITOK) 68 1.2 simonb #define wapbl_free(a) free((a), M_WAPBL) 69 1.2 simonb #define wapbl_calloc(n, s) malloc((n)*(s), M_WAPBL, M_WAITOK | M_ZERO) 70 1.2 simonb 71 1.2 simonb #else /* !_KERNEL */ 72 1.2 simonb #include <assert.h> 73 1.2 simonb #include <errno.h> 74 1.2 simonb #include <stdio.h> 75 1.2 simonb #include <stdbool.h> 76 1.2 simonb #include <stdlib.h> 77 1.2 simonb #include <string.h> 78 1.2 simonb 79 1.2 simonb #include <sys/time.h> 80 1.2 simonb #include <sys/wapbl.h> 81 1.2 simonb 82 1.2 simonb #define KDASSERT(x) assert(x) 83 1.2 simonb #define KASSERT(x) assert(x) 84 1.2 simonb #define wapbl_malloc(s) malloc(s) 85 1.2 simonb #define wapbl_free(a) free(a) 86 1.2 simonb #define wapbl_calloc(n, s) calloc((n), (s)) 87 1.2 simonb 88 1.2 simonb #endif /* !_KERNEL */ 89 1.2 simonb 90 1.2 simonb /* 91 1.2 simonb * INTERNAL DATA STRUCTURES 92 1.2 simonb */ 93 1.2 simonb 94 1.2 simonb /* 95 1.2 simonb * This structure holds per-mount log information. 96 1.2 simonb * 97 1.2 simonb * Legend: a = atomic access only 98 1.2 simonb * r = read-only after init 99 1.2 simonb * l = rwlock held 100 1.2 simonb * m = mutex held 101 1.2 simonb * u = unlocked access ok 102 1.2 simonb * b = bufcache_lock held 103 1.2 simonb */ 104 1.2 simonb struct wapbl { 105 1.2 simonb struct vnode *wl_logvp; /* r: log here */ 106 1.2 simonb struct vnode *wl_devvp; /* r: log on this device */ 107 1.2 simonb struct mount *wl_mount; /* r: mountpoint wl is associated with */ 108 1.2 simonb daddr_t wl_logpbn; /* r: Physical block number of start of log */ 109 1.2 simonb int wl_log_dev_bshift; /* r: logarithm of device block size of log 110 1.2 simonb device */ 111 1.2 simonb int wl_fs_dev_bshift; /* r: logarithm of device block size of 112 1.2 simonb filesystem device */ 113 1.2 simonb 114 1.3 yamt unsigned wl_lock_count; /* m: Count of transactions in progress */ 115 1.2 simonb 116 1.2 simonb size_t wl_circ_size; /* r: Number of bytes in buffer of log */ 117 1.2 simonb size_t wl_circ_off; /* r: Number of bytes reserved at start */ 118 1.2 simonb 119 1.2 simonb size_t wl_bufcount_max; /* r: Number of buffers reserved for log */ 120 1.2 simonb size_t wl_bufbytes_max; /* r: Number of buf bytes reserved for log */ 121 1.2 simonb 122 1.2 simonb off_t wl_head; /* l: Byte offset of log head */ 123 1.2 simonb off_t wl_tail; /* l: Byte offset of log tail */ 124 1.2 simonb /* 125 1.2 simonb * head == tail == 0 means log is empty 126 1.2 simonb * head == tail != 0 means log is full 127 1.2 simonb * see assertions in wapbl_advance() for other boundary conditions. 128 1.2 simonb * only truncate moves the tail, except when flush sets it to 129 1.2 simonb * wl_header_size only flush moves the head, except when truncate 130 1.2 simonb * sets it to 0. 131 1.2 simonb */ 132 1.2 simonb 133 1.2 simonb struct wapbl_wc_header *wl_wc_header; /* l */ 134 1.2 simonb void *wl_wc_scratch; /* l: scratch space (XXX: por que?!?) */ 135 1.2 simonb 136 1.2 simonb kmutex_t wl_mtx; /* u: short-term lock */ 137 1.2 simonb krwlock_t wl_rwlock; /* u: File system transaction lock */ 138 1.2 simonb 139 1.2 simonb /* 140 1.2 simonb * Must be held while accessing 141 1.2 simonb * wl_count or wl_bufs or head or tail 142 1.2 simonb */ 143 1.2 simonb 144 1.2 simonb /* 145 1.2 simonb * Callback called from within the flush routine to flush any extra 146 1.2 simonb * bits. Note that flush may be skipped without calling this if 147 1.2 simonb * there are no outstanding buffers in the transaction. 148 1.2 simonb */ 149 1.5 joerg #if _KERNEL 150 1.2 simonb wapbl_flush_fn_t wl_flush; /* r */ 151 1.2 simonb wapbl_flush_fn_t wl_flush_abort;/* r */ 152 1.5 joerg #endif 153 1.2 simonb 154 1.2 simonb size_t wl_bufbytes; /* m: Byte count of pages in wl_bufs */ 155 1.2 simonb size_t wl_bufcount; /* m: Count of buffers in wl_bufs */ 156 1.2 simonb size_t wl_bcount; /* m: Total bcount of wl_bufs */ 157 1.2 simonb 158 1.2 simonb LIST_HEAD(, buf) wl_bufs; /* m: Buffers in current transaction */ 159 1.2 simonb 160 1.2 simonb kcondvar_t wl_reclaimable_cv; /* m (obviously) */ 161 1.2 simonb size_t wl_reclaimable_bytes; /* m: Amount of space available for 162 1.2 simonb reclamation by truncate */ 163 1.2 simonb int wl_error_count; /* m: # of wl_entries with errors */ 164 1.2 simonb size_t wl_reserved_bytes; /* never truncate log smaller than this */ 165 1.2 simonb 166 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 167 1.2 simonb size_t wl_unsynced_bufbytes; /* Byte count of unsynced buffers */ 168 1.2 simonb #endif 169 1.2 simonb 170 1.2 simonb daddr_t *wl_deallocblks;/* l: address of block */ 171 1.2 simonb int *wl_dealloclens; /* l: size of block (fragments, kom ihg) */ 172 1.2 simonb int wl_dealloccnt; /* l: total count */ 173 1.2 simonb int wl_dealloclim; /* l: max count */ 174 1.2 simonb 175 1.2 simonb /* hashtable of inode numbers for allocated but unlinked inodes */ 176 1.2 simonb /* synch ??? */ 177 1.2 simonb LIST_HEAD(wapbl_ino_head, wapbl_ino) *wl_inohash; 178 1.2 simonb u_long wl_inohashmask; 179 1.2 simonb int wl_inohashcnt; 180 1.2 simonb 181 1.2 simonb SIMPLEQ_HEAD(, wapbl_entry) wl_entries; /* On disk transaction 182 1.2 simonb accounting */ 183 1.2 simonb }; 184 1.2 simonb 185 1.2 simonb #ifdef WAPBL_DEBUG_PRINT 186 1.2 simonb int wapbl_debug_print = WAPBL_DEBUG_PRINT; 187 1.2 simonb #endif 188 1.2 simonb 189 1.2 simonb /****************************************************************/ 190 1.2 simonb #ifdef _KERNEL 191 1.2 simonb 192 1.2 simonb #ifdef WAPBL_DEBUG 193 1.2 simonb struct wapbl *wapbl_debug_wl; 194 1.2 simonb #endif 195 1.2 simonb 196 1.2 simonb static int wapbl_write_commit(struct wapbl *wl, off_t head, off_t tail); 197 1.2 simonb static int wapbl_write_blocks(struct wapbl *wl, off_t *offp); 198 1.2 simonb static int wapbl_write_revocations(struct wapbl *wl, off_t *offp); 199 1.2 simonb static int wapbl_write_inodes(struct wapbl *wl, off_t *offp); 200 1.2 simonb #endif /* _KERNEL */ 201 1.2 simonb 202 1.2 simonb static int wapbl_replay_prescan(struct wapbl_replay *wr); 203 1.2 simonb static int wapbl_replay_get_inodes(struct wapbl_replay *wr); 204 1.2 simonb 205 1.2 simonb static __inline size_t wapbl_space_free(size_t avail, off_t head, 206 1.2 simonb off_t tail); 207 1.2 simonb static __inline size_t wapbl_space_used(size_t avail, off_t head, 208 1.2 simonb off_t tail); 209 1.2 simonb 210 1.2 simonb #ifdef _KERNEL 211 1.2 simonb 212 1.2 simonb #define WAPBL_INODETRK_SIZE 83 213 1.2 simonb static int wapbl_ino_pool_refcount; 214 1.2 simonb static struct pool wapbl_ino_pool; 215 1.2 simonb struct wapbl_ino { 216 1.2 simonb LIST_ENTRY(wapbl_ino) wi_hash; 217 1.2 simonb ino_t wi_ino; 218 1.2 simonb mode_t wi_mode; 219 1.2 simonb }; 220 1.2 simonb 221 1.2 simonb static void wapbl_inodetrk_init(struct wapbl *wl, u_int size); 222 1.2 simonb static void wapbl_inodetrk_free(struct wapbl *wl); 223 1.2 simonb static struct wapbl_ino *wapbl_inodetrk_get(struct wapbl *wl, ino_t ino); 224 1.2 simonb 225 1.2 simonb static size_t wapbl_transaction_len(struct wapbl *wl); 226 1.2 simonb static __inline size_t wapbl_transaction_inodes_len(struct wapbl *wl); 227 1.2 simonb 228 1.4 joerg #ifdef DEBUG 229 1.4 joerg int wapbl_replay_verify(struct wapbl_replay *, struct vnode *); 230 1.4 joerg #endif 231 1.4 joerg 232 1.4 joerg static int wapbl_replay_isopen1(struct wapbl_replay *); 233 1.4 joerg 234 1.2 simonb /* 235 1.2 simonb * This is useful for debugging. If set, the log will 236 1.2 simonb * only be truncated when necessary. 237 1.2 simonb */ 238 1.2 simonb int wapbl_lazy_truncate = 0; 239 1.2 simonb 240 1.2 simonb struct wapbl_ops wapbl_ops = { 241 1.2 simonb .wo_wapbl_discard = wapbl_discard, 242 1.2 simonb .wo_wapbl_replay_isopen = wapbl_replay_isopen1, 243 1.6 joerg .wo_wapbl_replay_can_read = wapbl_replay_can_read, 244 1.2 simonb .wo_wapbl_replay_read = wapbl_replay_read, 245 1.2 simonb .wo_wapbl_add_buf = wapbl_add_buf, 246 1.2 simonb .wo_wapbl_remove_buf = wapbl_remove_buf, 247 1.2 simonb .wo_wapbl_resize_buf = wapbl_resize_buf, 248 1.2 simonb .wo_wapbl_begin = wapbl_begin, 249 1.2 simonb .wo_wapbl_end = wapbl_end, 250 1.2 simonb .wo_wapbl_junlock_assert= wapbl_junlock_assert, 251 1.2 simonb 252 1.2 simonb /* XXX: the following is only used to say "this is a wapbl buf" */ 253 1.2 simonb .wo_wapbl_biodone = wapbl_biodone, 254 1.2 simonb }; 255 1.2 simonb 256 1.2 simonb void 257 1.2 simonb wapbl_init() 258 1.2 simonb { 259 1.2 simonb 260 1.2 simonb malloc_type_attach(M_WAPBL); 261 1.2 simonb } 262 1.2 simonb 263 1.2 simonb int 264 1.2 simonb wapbl_start(struct wapbl ** wlp, struct mount *mp, struct vnode *vp, 265 1.2 simonb daddr_t off, size_t count, size_t blksize, struct wapbl_replay *wr, 266 1.2 simonb wapbl_flush_fn_t flushfn, wapbl_flush_fn_t flushabortfn) 267 1.2 simonb { 268 1.2 simonb struct wapbl *wl; 269 1.2 simonb struct vnode *devvp; 270 1.2 simonb daddr_t logpbn; 271 1.2 simonb int error; 272 1.2 simonb int log_dev_bshift = DEV_BSHIFT; 273 1.2 simonb int fs_dev_bshift = DEV_BSHIFT; 274 1.2 simonb int run; 275 1.2 simonb 276 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_OPEN, ("wapbl_start: vp=%p off=%" PRId64 277 1.2 simonb " count=%zu blksize=%zu\n", vp, off, count, blksize)); 278 1.2 simonb 279 1.2 simonb if (log_dev_bshift > fs_dev_bshift) { 280 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_OPEN, 281 1.2 simonb ("wapbl: log device's block size cannot be larger " 282 1.2 simonb "than filesystem's\n")); 283 1.2 simonb /* 284 1.2 simonb * Not currently implemented, although it could be if 285 1.2 simonb * needed someday. 286 1.2 simonb */ 287 1.2 simonb return ENOSYS; 288 1.2 simonb } 289 1.2 simonb 290 1.2 simonb if (off < 0) 291 1.2 simonb return EINVAL; 292 1.2 simonb 293 1.2 simonb if (blksize < DEV_BSIZE) 294 1.2 simonb return EINVAL; 295 1.2 simonb if (blksize % DEV_BSIZE) 296 1.2 simonb return EINVAL; 297 1.2 simonb 298 1.2 simonb /* XXXTODO: verify that the full load is writable */ 299 1.2 simonb 300 1.2 simonb /* 301 1.2 simonb * XXX check for minimum log size 302 1.2 simonb * minimum is governed by minimum amount of space 303 1.2 simonb * to complete a transaction. (probably truncate) 304 1.2 simonb */ 305 1.2 simonb /* XXX for now pick something minimal */ 306 1.2 simonb if ((count * blksize) < MAXPHYS) { 307 1.2 simonb return ENOSPC; 308 1.2 simonb } 309 1.2 simonb 310 1.2 simonb if ((error = VOP_BMAP(vp, off, &devvp, &logpbn, &run)) != 0) { 311 1.2 simonb return error; 312 1.2 simonb } 313 1.2 simonb 314 1.2 simonb wl = wapbl_calloc(1, sizeof(*wl)); 315 1.2 simonb rw_init(&wl->wl_rwlock); 316 1.2 simonb mutex_init(&wl->wl_mtx, MUTEX_DEFAULT, IPL_NONE); 317 1.2 simonb cv_init(&wl->wl_reclaimable_cv, "wapblrec"); 318 1.2 simonb LIST_INIT(&wl->wl_bufs); 319 1.2 simonb SIMPLEQ_INIT(&wl->wl_entries); 320 1.2 simonb 321 1.2 simonb wl->wl_logvp = vp; 322 1.2 simonb wl->wl_devvp = devvp; 323 1.2 simonb wl->wl_mount = mp; 324 1.2 simonb wl->wl_logpbn = logpbn; 325 1.2 simonb wl->wl_log_dev_bshift = log_dev_bshift; 326 1.2 simonb wl->wl_fs_dev_bshift = fs_dev_bshift; 327 1.2 simonb 328 1.2 simonb wl->wl_flush = flushfn; 329 1.2 simonb wl->wl_flush_abort = flushabortfn; 330 1.2 simonb 331 1.2 simonb /* Reserve two log device blocks for the commit headers */ 332 1.2 simonb wl->wl_circ_off = 2<<wl->wl_log_dev_bshift; 333 1.2 simonb wl->wl_circ_size = ((count * blksize) - wl->wl_circ_off); 334 1.2 simonb /* truncate the log usage to a multiple of log_dev_bshift */ 335 1.2 simonb wl->wl_circ_size >>= wl->wl_log_dev_bshift; 336 1.2 simonb wl->wl_circ_size <<= wl->wl_log_dev_bshift; 337 1.2 simonb 338 1.2 simonb /* 339 1.2 simonb * wl_bufbytes_max limits the size of the in memory transaction space. 340 1.2 simonb * - Since buffers are allocated and accounted for in units of 341 1.2 simonb * PAGE_SIZE it is required to be a multiple of PAGE_SIZE 342 1.2 simonb * (i.e. 1<<PAGE_SHIFT) 343 1.2 simonb * - Since the log device has to be written in units of 344 1.2 simonb * 1<<wl_log_dev_bshift it is required to be a mulitple of 345 1.2 simonb * 1<<wl_log_dev_bshift. 346 1.2 simonb * - Since filesystem will provide data in units of 1<<wl_fs_dev_bshift, 347 1.2 simonb * it is convenient to be a multiple of 1<<wl_fs_dev_bshift. 348 1.2 simonb * Therefore it must be multiple of the least common multiple of those 349 1.2 simonb * three quantities. Fortunately, all of those quantities are 350 1.2 simonb * guaranteed to be a power of two, and the least common multiple of 351 1.2 simonb * a set of numbers which are all powers of two is simply the maximum 352 1.2 simonb * of those numbers. Finally, the maximum logarithm of a power of two 353 1.2 simonb * is the same as the log of the maximum power of two. So we can do 354 1.2 simonb * the following operations to size wl_bufbytes_max: 355 1.2 simonb */ 356 1.2 simonb 357 1.2 simonb /* XXX fix actual number of pages reserved per filesystem. */ 358 1.2 simonb wl->wl_bufbytes_max = MIN(wl->wl_circ_size, buf_memcalc() / 2); 359 1.2 simonb 360 1.2 simonb /* Round wl_bufbytes_max to the largest power of two constraint */ 361 1.2 simonb wl->wl_bufbytes_max >>= PAGE_SHIFT; 362 1.2 simonb wl->wl_bufbytes_max <<= PAGE_SHIFT; 363 1.2 simonb wl->wl_bufbytes_max >>= wl->wl_log_dev_bshift; 364 1.2 simonb wl->wl_bufbytes_max <<= wl->wl_log_dev_bshift; 365 1.2 simonb wl->wl_bufbytes_max >>= wl->wl_fs_dev_bshift; 366 1.2 simonb wl->wl_bufbytes_max <<= wl->wl_fs_dev_bshift; 367 1.2 simonb 368 1.2 simonb /* XXX maybe use filesystem fragment size instead of 1024 */ 369 1.2 simonb /* XXX fix actual number of buffers reserved per filesystem. */ 370 1.2 simonb wl->wl_bufcount_max = (nbuf / 2) * 1024; 371 1.2 simonb 372 1.2 simonb /* XXX tie this into resource estimation */ 373 1.2 simonb wl->wl_dealloclim = 2 * btodb(wl->wl_bufbytes_max); 374 1.2 simonb 375 1.2 simonb #if WAPBL_UVM_ALLOC 376 1.2 simonb wl->wl_deallocblks = (void *) uvm_km_zalloc(kernel_map, 377 1.2 simonb round_page(sizeof(*wl->wl_deallocblks) * wl->wl_dealloclim)); 378 1.2 simonb KASSERT(wl->wl_deallocblks != NULL); 379 1.2 simonb wl->wl_dealloclens = (void *) uvm_km_zalloc(kernel_map, 380 1.2 simonb round_page(sizeof(*wl->wl_dealloclens) * wl->wl_dealloclim)); 381 1.2 simonb KASSERT(wl->wl_dealloclens != NULL); 382 1.2 simonb #else 383 1.2 simonb wl->wl_deallocblks = wapbl_malloc(sizeof(*wl->wl_deallocblks) * 384 1.2 simonb wl->wl_dealloclim); 385 1.2 simonb wl->wl_dealloclens = wapbl_malloc(sizeof(*wl->wl_dealloclens) * 386 1.2 simonb wl->wl_dealloclim); 387 1.2 simonb #endif 388 1.2 simonb 389 1.2 simonb wapbl_inodetrk_init(wl, WAPBL_INODETRK_SIZE); 390 1.2 simonb 391 1.2 simonb /* Initialize the commit header */ 392 1.2 simonb { 393 1.2 simonb struct wapbl_wc_header *wc; 394 1.2 simonb size_t len = 1<<wl->wl_log_dev_bshift; 395 1.2 simonb wc = wapbl_calloc(1, len); 396 1.2 simonb wc->wc_type = WAPBL_WC_HEADER; 397 1.2 simonb wc->wc_len = len; 398 1.2 simonb wc->wc_circ_off = wl->wl_circ_off; 399 1.2 simonb wc->wc_circ_size = wl->wl_circ_size; 400 1.2 simonb /* XXX wc->wc_fsid */ 401 1.2 simonb wc->wc_log_dev_bshift = wl->wl_log_dev_bshift; 402 1.2 simonb wc->wc_fs_dev_bshift = wl->wl_fs_dev_bshift; 403 1.2 simonb wl->wl_wc_header = wc; 404 1.2 simonb wl->wl_wc_scratch = wapbl_malloc(len); 405 1.2 simonb } 406 1.2 simonb 407 1.2 simonb /* 408 1.2 simonb * if there was an existing set of unlinked but 409 1.2 simonb * allocated inodes, preserve it in the new 410 1.2 simonb * log. 411 1.2 simonb */ 412 1.2 simonb if (wr && wr->wr_inodescnt) { 413 1.2 simonb int i; 414 1.2 simonb 415 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_REPLAY, 416 1.2 simonb ("wapbl_start: reusing log with %d inodes\n", 417 1.2 simonb wr->wr_inodescnt)); 418 1.2 simonb 419 1.2 simonb /* 420 1.2 simonb * Its only valid to reuse the replay log if its 421 1.2 simonb * the same as the new log we just opened. 422 1.2 simonb */ 423 1.2 simonb KDASSERT(!wapbl_replay_isopen(wr)); 424 1.2 simonb KASSERT(devvp->v_rdev == wr->wr_devvp->v_rdev); 425 1.2 simonb KASSERT(logpbn == wr->wr_logpbn); 426 1.2 simonb KASSERT(wl->wl_circ_size == wr->wr_wc_header.wc_circ_size); 427 1.2 simonb KASSERT(wl->wl_circ_off == wr->wr_wc_header.wc_circ_off); 428 1.2 simonb KASSERT(wl->wl_log_dev_bshift == 429 1.2 simonb wr->wr_wc_header.wc_log_dev_bshift); 430 1.2 simonb KASSERT(wl->wl_fs_dev_bshift == 431 1.2 simonb wr->wr_wc_header.wc_fs_dev_bshift); 432 1.2 simonb 433 1.2 simonb wl->wl_wc_header->wc_generation = 434 1.2 simonb wr->wr_wc_header.wc_generation + 1; 435 1.2 simonb 436 1.2 simonb for (i = 0; i < wr->wr_inodescnt; i++) 437 1.2 simonb wapbl_register_inode(wl, wr->wr_inodes[i].wr_inumber, 438 1.2 simonb wr->wr_inodes[i].wr_imode); 439 1.2 simonb 440 1.2 simonb /* Make sure new transaction won't overwrite old inodes list */ 441 1.2 simonb KDASSERT(wapbl_transaction_len(wl) <= 442 1.2 simonb wapbl_space_free(wl->wl_circ_size, wr->wr_inodeshead, 443 1.2 simonb wr->wr_inodestail)); 444 1.2 simonb 445 1.2 simonb wl->wl_head = wl->wl_tail = wr->wr_inodeshead; 446 1.2 simonb wl->wl_reclaimable_bytes = wl->wl_reserved_bytes = 447 1.2 simonb wapbl_transaction_len(wl); 448 1.2 simonb 449 1.2 simonb error = wapbl_write_inodes(wl, &wl->wl_head); 450 1.2 simonb if (error) 451 1.2 simonb goto errout; 452 1.2 simonb 453 1.2 simonb KASSERT(wl->wl_head != wl->wl_tail); 454 1.2 simonb KASSERT(wl->wl_head != 0); 455 1.2 simonb } 456 1.2 simonb 457 1.2 simonb error = wapbl_write_commit(wl, wl->wl_head, wl->wl_tail); 458 1.2 simonb if (error) { 459 1.2 simonb goto errout; 460 1.2 simonb } 461 1.2 simonb 462 1.2 simonb *wlp = wl; 463 1.2 simonb #if defined(WAPBL_DEBUG) 464 1.2 simonb wapbl_debug_wl = wl; 465 1.2 simonb #endif 466 1.2 simonb 467 1.2 simonb return 0; 468 1.2 simonb errout: 469 1.2 simonb wapbl_discard(wl); 470 1.2 simonb wapbl_free(wl->wl_wc_scratch); 471 1.2 simonb wapbl_free(wl->wl_wc_header); 472 1.2 simonb #if WAPBL_UVM_ALLOC 473 1.2 simonb uvm_km_free_wakeup(kernel_map, (vaddr_t) wl->wl_deallocblks, 474 1.2 simonb round_page(sizeof(*wl->wl_deallocblks * 475 1.2 simonb wl->wl_dealloclim))); 476 1.2 simonb uvm_km_free_wakeup(kernel_map, (vaddr_t) wl->wl_dealloclens, 477 1.2 simonb round_page(sizeof(*wl->wl_dealloclens * 478 1.2 simonb wl->wl_dealloclim))); 479 1.2 simonb #else 480 1.2 simonb wapbl_free(wl->wl_deallocblks); 481 1.2 simonb wapbl_free(wl->wl_dealloclens); 482 1.2 simonb #endif 483 1.2 simonb wapbl_inodetrk_free(wl); 484 1.2 simonb wapbl_free(wl); 485 1.2 simonb 486 1.2 simonb return error; 487 1.2 simonb } 488 1.2 simonb 489 1.2 simonb /* 490 1.2 simonb * Like wapbl_flush, only discards the transaction 491 1.2 simonb * completely 492 1.2 simonb */ 493 1.2 simonb 494 1.2 simonb void 495 1.2 simonb wapbl_discard(struct wapbl *wl) 496 1.2 simonb { 497 1.2 simonb struct wapbl_entry *we; 498 1.2 simonb struct buf *bp; 499 1.2 simonb int i; 500 1.2 simonb 501 1.2 simonb /* 502 1.2 simonb * XXX we may consider using upgrade here 503 1.2 simonb * if we want to call flush from inside a transaction 504 1.2 simonb */ 505 1.2 simonb rw_enter(&wl->wl_rwlock, RW_WRITER); 506 1.2 simonb wl->wl_flush(wl->wl_mount, wl->wl_deallocblks, wl->wl_dealloclens, 507 1.2 simonb wl->wl_dealloccnt); 508 1.2 simonb 509 1.2 simonb #ifdef WAPBL_DEBUG_PRINT 510 1.2 simonb { 511 1.2 simonb struct wapbl_entry *we; 512 1.2 simonb pid_t pid = -1; 513 1.2 simonb lwpid_t lid = -1; 514 1.2 simonb if (curproc) 515 1.2 simonb pid = curproc->p_pid; 516 1.2 simonb if (curlwp) 517 1.2 simonb lid = curlwp->l_lid; 518 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 519 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_DISCARD, 520 1.2 simonb ("wapbl_discard: thread %d.%d discarding " 521 1.2 simonb "transaction\n" 522 1.2 simonb "\tbufcount=%zu bufbytes=%zu bcount=%zu " 523 1.2 simonb "deallocs=%d inodes=%d\n" 524 1.2 simonb "\terrcnt = %u, reclaimable=%zu reserved=%zu " 525 1.2 simonb "unsynced=%zu\n", 526 1.2 simonb pid, lid, wl->wl_bufcount, wl->wl_bufbytes, 527 1.2 simonb wl->wl_bcount, wl->wl_dealloccnt, 528 1.2 simonb wl->wl_inohashcnt, wl->wl_error_count, 529 1.2 simonb wl->wl_reclaimable_bytes, wl->wl_reserved_bytes, 530 1.2 simonb wl->wl_unsynced_bufbytes)); 531 1.2 simonb SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) { 532 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_DISCARD, 533 1.2 simonb ("\tentry: bufcount = %zu, reclaimable = %zu, " 534 1.2 simonb "error = %d, unsynced = %zu\n", 535 1.2 simonb we->we_bufcount, we->we_reclaimable_bytes, 536 1.2 simonb we->we_error, we->we_unsynced_bufbytes)); 537 1.2 simonb } 538 1.2 simonb #else /* !WAPBL_DEBUG_BUFBYTES */ 539 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_DISCARD, 540 1.2 simonb ("wapbl_discard: thread %d.%d discarding transaction\n" 541 1.2 simonb "\tbufcount=%zu bufbytes=%zu bcount=%zu " 542 1.2 simonb "deallocs=%d inodes=%d\n" 543 1.2 simonb "\terrcnt = %u, reclaimable=%zu reserved=%zu\n", 544 1.2 simonb pid, lid, wl->wl_bufcount, wl->wl_bufbytes, 545 1.2 simonb wl->wl_bcount, wl->wl_dealloccnt, 546 1.2 simonb wl->wl_inohashcnt, wl->wl_error_count, 547 1.2 simonb wl->wl_reclaimable_bytes, wl->wl_reserved_bytes)); 548 1.2 simonb SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) { 549 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_DISCARD, 550 1.2 simonb ("\tentry: bufcount = %zu, reclaimable = %zu, " 551 1.2 simonb "error = %d\n", 552 1.2 simonb we->we_bufcount, we->we_reclaimable_bytes, 553 1.2 simonb we->we_error)); 554 1.2 simonb } 555 1.2 simonb #endif /* !WAPBL_DEBUG_BUFBYTES */ 556 1.2 simonb } 557 1.2 simonb #endif /* WAPBL_DEBUG_PRINT */ 558 1.2 simonb 559 1.2 simonb for (i = 0; i <= wl->wl_inohashmask; i++) { 560 1.2 simonb struct wapbl_ino_head *wih; 561 1.2 simonb struct wapbl_ino *wi; 562 1.2 simonb 563 1.2 simonb wih = &wl->wl_inohash[i]; 564 1.2 simonb while ((wi = LIST_FIRST(wih)) != NULL) { 565 1.2 simonb LIST_REMOVE(wi, wi_hash); 566 1.2 simonb pool_put(&wapbl_ino_pool, wi); 567 1.2 simonb KASSERT(wl->wl_inohashcnt > 0); 568 1.2 simonb wl->wl_inohashcnt--; 569 1.2 simonb } 570 1.2 simonb } 571 1.2 simonb 572 1.2 simonb /* 573 1.2 simonb * clean buffer list 574 1.2 simonb */ 575 1.2 simonb mutex_enter(&bufcache_lock); 576 1.2 simonb mutex_enter(&wl->wl_mtx); 577 1.2 simonb while ((bp = LIST_FIRST(&wl->wl_bufs)) != NULL) { 578 1.2 simonb if (bbusy(bp, 0, 0, &wl->wl_mtx) == 0) { 579 1.2 simonb /* 580 1.2 simonb * The buffer will be unlocked and 581 1.2 simonb * removed from the transaction in brelse 582 1.2 simonb */ 583 1.2 simonb mutex_exit(&wl->wl_mtx); 584 1.2 simonb brelsel(bp, 0); 585 1.2 simonb mutex_enter(&wl->wl_mtx); 586 1.2 simonb } 587 1.2 simonb } 588 1.2 simonb mutex_exit(&wl->wl_mtx); 589 1.2 simonb mutex_exit(&bufcache_lock); 590 1.2 simonb 591 1.2 simonb /* 592 1.2 simonb * Remove references to this wl from wl_entries, free any which 593 1.2 simonb * no longer have buffers, others will be freed in wapbl_biodone 594 1.2 simonb * when they no longer have any buffers. 595 1.2 simonb */ 596 1.2 simonb while ((we = SIMPLEQ_FIRST(&wl->wl_entries)) != NULL) { 597 1.2 simonb SIMPLEQ_REMOVE_HEAD(&wl->wl_entries, we_entries); 598 1.2 simonb /* XXX should we be accumulating wl_error_count 599 1.2 simonb * and increasing reclaimable bytes ? */ 600 1.2 simonb we->we_wapbl = NULL; 601 1.2 simonb if (we->we_bufcount == 0) { 602 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 603 1.2 simonb KASSERT(we->we_unsynced_bufbytes == 0); 604 1.2 simonb #endif 605 1.2 simonb wapbl_free(we); 606 1.2 simonb } 607 1.2 simonb } 608 1.2 simonb 609 1.2 simonb /* Discard list of deallocs */ 610 1.2 simonb wl->wl_dealloccnt = 0; 611 1.2 simonb /* XXX should we clear wl_reserved_bytes? */ 612 1.2 simonb 613 1.2 simonb KASSERT(wl->wl_bufbytes == 0); 614 1.2 simonb KASSERT(wl->wl_bcount == 0); 615 1.2 simonb KASSERT(wl->wl_bufcount == 0); 616 1.2 simonb KASSERT(LIST_EMPTY(&wl->wl_bufs)); 617 1.2 simonb KASSERT(SIMPLEQ_EMPTY(&wl->wl_entries)); 618 1.2 simonb KASSERT(wl->wl_inohashcnt == 0); 619 1.2 simonb 620 1.2 simonb rw_exit(&wl->wl_rwlock); 621 1.2 simonb } 622 1.2 simonb 623 1.2 simonb int 624 1.2 simonb wapbl_stop(struct wapbl *wl, int force) 625 1.2 simonb { 626 1.2 simonb struct vnode *vp; 627 1.2 simonb int error; 628 1.2 simonb 629 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_OPEN, ("wapbl_stop called\n")); 630 1.2 simonb error = wapbl_flush(wl, 1); 631 1.2 simonb if (error) { 632 1.2 simonb if (force) 633 1.2 simonb wapbl_discard(wl); 634 1.2 simonb else 635 1.2 simonb return error; 636 1.2 simonb } 637 1.2 simonb 638 1.2 simonb /* Unlinked inodes persist after a flush */ 639 1.2 simonb if (wl->wl_inohashcnt) { 640 1.2 simonb if (force) { 641 1.2 simonb wapbl_discard(wl); 642 1.2 simonb } else { 643 1.2 simonb return EBUSY; 644 1.2 simonb } 645 1.2 simonb } 646 1.2 simonb 647 1.2 simonb KASSERT(wl->wl_bufbytes == 0); 648 1.2 simonb KASSERT(wl->wl_bcount == 0); 649 1.2 simonb KASSERT(wl->wl_bufcount == 0); 650 1.2 simonb KASSERT(LIST_EMPTY(&wl->wl_bufs)); 651 1.2 simonb KASSERT(wl->wl_dealloccnt == 0); 652 1.2 simonb KASSERT(SIMPLEQ_EMPTY(&wl->wl_entries)); 653 1.2 simonb KASSERT(wl->wl_inohashcnt == 0); 654 1.2 simonb 655 1.2 simonb vp = wl->wl_logvp; 656 1.2 simonb 657 1.2 simonb wapbl_free(wl->wl_wc_scratch); 658 1.2 simonb wapbl_free(wl->wl_wc_header); 659 1.2 simonb #if WAPBL_UVM_ALLOC 660 1.2 simonb uvm_km_free_wakeup(kernel_map, (vaddr_t) wl->wl_deallocblks, 661 1.2 simonb round_page(sizeof(*wl->wl_deallocblks * 662 1.2 simonb wl->wl_dealloclim))); 663 1.2 simonb uvm_km_free_wakeup(kernel_map, (vaddr_t) wl->wl_dealloclens, 664 1.2 simonb round_page(sizeof(*wl->wl_dealloclens * 665 1.2 simonb wl->wl_dealloclim))); 666 1.2 simonb #else 667 1.2 simonb wapbl_free(wl->wl_deallocblks); 668 1.2 simonb wapbl_free(wl->wl_dealloclens); 669 1.2 simonb #endif 670 1.2 simonb wapbl_inodetrk_free(wl); 671 1.2 simonb 672 1.2 simonb cv_destroy(&wl->wl_reclaimable_cv); 673 1.2 simonb mutex_destroy(&wl->wl_mtx); 674 1.2 simonb rw_destroy(&wl->wl_rwlock); 675 1.2 simonb wapbl_free(wl); 676 1.2 simonb 677 1.2 simonb return 0; 678 1.2 simonb } 679 1.2 simonb 680 1.2 simonb static int 681 1.2 simonb wapbl_doio(void *data, size_t len, struct vnode *devvp, daddr_t pbn, int flags) 682 1.2 simonb { 683 1.2 simonb struct pstats *pstats = curlwp->l_proc->p_stats; 684 1.2 simonb struct buf *bp; 685 1.2 simonb int error; 686 1.2 simonb 687 1.2 simonb KASSERT((flags & ~(B_WRITE | B_READ)) == 0); 688 1.2 simonb KASSERT(devvp->v_type == VBLK); 689 1.2 simonb 690 1.2 simonb if ((flags & (B_WRITE | B_READ)) == B_WRITE) { 691 1.2 simonb mutex_enter(&devvp->v_interlock); 692 1.2 simonb devvp->v_numoutput++; 693 1.2 simonb mutex_exit(&devvp->v_interlock); 694 1.2 simonb pstats->p_ru.ru_oublock++; 695 1.2 simonb } else { 696 1.2 simonb pstats->p_ru.ru_inblock++; 697 1.2 simonb } 698 1.2 simonb 699 1.2 simonb bp = getiobuf(devvp, true); 700 1.2 simonb bp->b_flags = flags; 701 1.2 simonb bp->b_cflags = BC_BUSY; /* silly & dubious */ 702 1.2 simonb bp->b_dev = devvp->v_rdev; 703 1.2 simonb bp->b_data = data; 704 1.2 simonb bp->b_bufsize = bp->b_resid = bp->b_bcount = len; 705 1.2 simonb bp->b_blkno = pbn; 706 1.2 simonb 707 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_IO, 708 1.2 simonb ("wapbl_doio: %s %d bytes at block %"PRId64" on dev 0x%x\n", 709 1.2 simonb BUF_ISWRITE(bp) ? "write" : "read", bp->b_bcount, 710 1.2 simonb bp->b_blkno, bp->b_dev)); 711 1.2 simonb 712 1.2 simonb VOP_STRATEGY(devvp, bp); 713 1.2 simonb 714 1.2 simonb error = biowait(bp); 715 1.2 simonb putiobuf(bp); 716 1.2 simonb 717 1.2 simonb if (error) { 718 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_ERROR, 719 1.2 simonb ("wapbl_doio: %s %zu bytes at block %" PRId64 720 1.2 simonb " on dev 0x%x failed with error %d\n", 721 1.2 simonb (((flags & (B_WRITE | B_READ)) == B_WRITE) ? 722 1.2 simonb "write" : "read"), 723 1.2 simonb len, pbn, devvp->v_rdev, error)); 724 1.2 simonb } 725 1.2 simonb 726 1.2 simonb return error; 727 1.2 simonb } 728 1.2 simonb 729 1.2 simonb int 730 1.2 simonb wapbl_write(void *data, size_t len, struct vnode *devvp, daddr_t pbn) 731 1.2 simonb { 732 1.2 simonb 733 1.2 simonb return wapbl_doio(data, len, devvp, pbn, B_WRITE); 734 1.2 simonb } 735 1.2 simonb 736 1.2 simonb int 737 1.2 simonb wapbl_read(void *data, size_t len, struct vnode *devvp, daddr_t pbn) 738 1.2 simonb { 739 1.2 simonb 740 1.2 simonb return wapbl_doio(data, len, devvp, pbn, B_READ); 741 1.2 simonb } 742 1.2 simonb 743 1.2 simonb /* 744 1.2 simonb * Off is byte offset returns new offset for next write 745 1.2 simonb * handles log wraparound 746 1.2 simonb */ 747 1.2 simonb static int 748 1.2 simonb wapbl_circ_write(struct wapbl *wl, void *data, size_t len, off_t *offp) 749 1.2 simonb { 750 1.2 simonb size_t slen; 751 1.2 simonb off_t off = *offp; 752 1.2 simonb int error; 753 1.2 simonb 754 1.2 simonb KDASSERT(((len >> wl->wl_log_dev_bshift) << 755 1.2 simonb wl->wl_log_dev_bshift) == len); 756 1.2 simonb 757 1.2 simonb if (off < wl->wl_circ_off) 758 1.2 simonb off = wl->wl_circ_off; 759 1.2 simonb slen = wl->wl_circ_off + wl->wl_circ_size - off; 760 1.2 simonb if (slen < len) { 761 1.2 simonb error = wapbl_write(data, slen, wl->wl_devvp, 762 1.2 simonb wl->wl_logpbn + (off >> wl->wl_log_dev_bshift)); 763 1.2 simonb if (error) 764 1.2 simonb return error; 765 1.2 simonb data = (uint8_t *)data + slen; 766 1.2 simonb len -= slen; 767 1.2 simonb off = wl->wl_circ_off; 768 1.2 simonb } 769 1.2 simonb error = wapbl_write(data, len, wl->wl_devvp, 770 1.2 simonb wl->wl_logpbn + (off >> wl->wl_log_dev_bshift)); 771 1.2 simonb if (error) 772 1.2 simonb return error; 773 1.2 simonb off += len; 774 1.2 simonb if (off >= wl->wl_circ_off + wl->wl_circ_size) 775 1.2 simonb off = wl->wl_circ_off; 776 1.2 simonb *offp = off; 777 1.2 simonb return 0; 778 1.2 simonb } 779 1.2 simonb 780 1.2 simonb /****************************************************************/ 781 1.2 simonb 782 1.2 simonb int 783 1.2 simonb wapbl_begin(struct wapbl *wl, const char *file, int line) 784 1.2 simonb { 785 1.2 simonb int doflush; 786 1.2 simonb unsigned lockcount; 787 1.2 simonb krw_t op; 788 1.2 simonb 789 1.2 simonb KDASSERT(wl); 790 1.2 simonb 791 1.2 simonb /* 792 1.2 simonb * XXX: The original code calls for the use of a RW_READER lock 793 1.2 simonb * here, but it turns out there are performance issues with high 794 1.2 simonb * metadata-rate workloads (e.g. multiple simultaneous tar 795 1.2 simonb * extractions). For now, we force the lock to be RW_WRITER, 796 1.2 simonb * since that currently has the best performance characteristics 797 1.2 simonb * (even for a single tar-file extraction). 798 1.2 simonb * 799 1.2 simonb */ 800 1.2 simonb #define WAPBL_DEBUG_SERIALIZE 1 801 1.2 simonb 802 1.2 simonb #ifdef WAPBL_DEBUG_SERIALIZE 803 1.2 simonb op = RW_WRITER; 804 1.2 simonb #else 805 1.2 simonb op = RW_READER; 806 1.2 simonb #endif 807 1.2 simonb 808 1.2 simonb /* 809 1.2 simonb * XXX this needs to be made much more sophisticated. 810 1.2 simonb * perhaps each wapbl_begin could reserve a specified 811 1.2 simonb * number of buffers and bytes. 812 1.2 simonb */ 813 1.2 simonb mutex_enter(&wl->wl_mtx); 814 1.2 simonb lockcount = wl->wl_lock_count; 815 1.2 simonb doflush = ((wl->wl_bufbytes + (lockcount * MAXPHYS)) > 816 1.2 simonb wl->wl_bufbytes_max / 2) || 817 1.2 simonb ((wl->wl_bufcount + (lockcount * 10)) > 818 1.2 simonb wl->wl_bufcount_max / 2) || 819 1.2 simonb (wapbl_transaction_len(wl) > wl->wl_circ_size / 2); 820 1.2 simonb mutex_exit(&wl->wl_mtx); 821 1.2 simonb 822 1.2 simonb if (doflush) { 823 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_FLUSH, 824 1.2 simonb ("force flush lockcnt=%d bufbytes=%zu " 825 1.2 simonb "(max=%zu) bufcount=%zu (max=%zu)\n", 826 1.2 simonb lockcount, wl->wl_bufbytes, 827 1.2 simonb wl->wl_bufbytes_max, wl->wl_bufcount, 828 1.2 simonb wl->wl_bufcount_max)); 829 1.2 simonb } 830 1.2 simonb 831 1.2 simonb if (doflush) { 832 1.2 simonb int error = wapbl_flush(wl, 0); 833 1.2 simonb if (error) 834 1.2 simonb return error; 835 1.2 simonb } 836 1.2 simonb 837 1.2 simonb rw_enter(&wl->wl_rwlock, op); 838 1.2 simonb mutex_enter(&wl->wl_mtx); 839 1.2 simonb wl->wl_lock_count++; 840 1.2 simonb mutex_exit(&wl->wl_mtx); 841 1.2 simonb 842 1.2 simonb #if defined(WAPBL_DEBUG_PRINT) && defined(WAPBL_DEBUG_SERIALIZE) 843 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_TRANSACTION, 844 1.2 simonb ("wapbl_begin thread %d.%d with bufcount=%zu " 845 1.2 simonb "bufbytes=%zu bcount=%zu at %s:%d\n", 846 1.2 simonb curproc->p_pid, curlwp->l_lid, wl->wl_bufcount, 847 1.2 simonb wl->wl_bufbytes, wl->wl_bcount, file, line)); 848 1.2 simonb #endif 849 1.2 simonb 850 1.2 simonb return 0; 851 1.2 simonb } 852 1.2 simonb 853 1.2 simonb void 854 1.2 simonb wapbl_end(struct wapbl *wl) 855 1.2 simonb { 856 1.2 simonb 857 1.2 simonb #if defined(WAPBL_DEBUG_PRINT) && defined(WAPBL_DEBUG_SERIALIZE) 858 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_TRANSACTION, 859 1.2 simonb ("wapbl_end thread %d.%d with bufcount=%zu " 860 1.2 simonb "bufbytes=%zu bcount=%zu\n", 861 1.2 simonb curproc->p_pid, curlwp->l_lid, wl->wl_bufcount, 862 1.2 simonb wl->wl_bufbytes, wl->wl_bcount)); 863 1.2 simonb #endif 864 1.2 simonb 865 1.2 simonb mutex_enter(&wl->wl_mtx); 866 1.2 simonb KASSERT(wl->wl_lock_count > 0); 867 1.2 simonb wl->wl_lock_count--; 868 1.2 simonb mutex_exit(&wl->wl_mtx); 869 1.2 simonb 870 1.2 simonb rw_exit(&wl->wl_rwlock); 871 1.2 simonb } 872 1.2 simonb 873 1.2 simonb void 874 1.2 simonb wapbl_add_buf(struct wapbl *wl, struct buf * bp) 875 1.2 simonb { 876 1.2 simonb 877 1.2 simonb KASSERT(bp->b_cflags & BC_BUSY); 878 1.2 simonb KASSERT(bp->b_vp); 879 1.2 simonb 880 1.2 simonb wapbl_jlock_assert(wl); 881 1.2 simonb 882 1.2 simonb #if 0 883 1.2 simonb /* 884 1.2 simonb * XXX this might be an issue for swapfiles. 885 1.2 simonb * see uvm_swap.c:1702 886 1.2 simonb * 887 1.2 simonb * XXX2 why require it then? leap of semantics? 888 1.2 simonb */ 889 1.2 simonb KASSERT((bp->b_cflags & BC_NOCACHE) == 0); 890 1.2 simonb #endif 891 1.2 simonb 892 1.2 simonb mutex_enter(&wl->wl_mtx); 893 1.2 simonb if (bp->b_flags & B_LOCKED) { 894 1.2 simonb LIST_REMOVE(bp, b_wapbllist); 895 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_BUFFER2, 896 1.2 simonb ("wapbl_add_buf thread %d.%d re-adding buf %p " 897 1.2 simonb "with %d bytes %d bcount\n", 898 1.2 simonb curproc->p_pid, curlwp->l_lid, bp, bp->b_bufsize, 899 1.2 simonb bp->b_bcount)); 900 1.2 simonb } else { 901 1.2 simonb /* unlocked by dirty buffers shouldn't exist */ 902 1.2 simonb KASSERT(!(bp->b_oflags & BO_DELWRI)); 903 1.2 simonb wl->wl_bufbytes += bp->b_bufsize; 904 1.2 simonb wl->wl_bcount += bp->b_bcount; 905 1.2 simonb wl->wl_bufcount++; 906 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_BUFFER, 907 1.2 simonb ("wapbl_add_buf thread %d.%d adding buf %p " 908 1.2 simonb "with %d bytes %d bcount\n", 909 1.2 simonb curproc->p_pid, curlwp->l_lid, bp, bp->b_bufsize, 910 1.2 simonb bp->b_bcount)); 911 1.2 simonb } 912 1.2 simonb LIST_INSERT_HEAD(&wl->wl_bufs, bp, b_wapbllist); 913 1.2 simonb mutex_exit(&wl->wl_mtx); 914 1.2 simonb 915 1.2 simonb bp->b_flags |= B_LOCKED; 916 1.2 simonb } 917 1.2 simonb 918 1.2 simonb static void 919 1.2 simonb wapbl_remove_buf_locked(struct wapbl * wl, struct buf *bp) 920 1.2 simonb { 921 1.2 simonb 922 1.2 simonb KASSERT(mutex_owned(&wl->wl_mtx)); 923 1.2 simonb KASSERT(bp->b_cflags & BC_BUSY); 924 1.2 simonb wapbl_jlock_assert(wl); 925 1.2 simonb 926 1.2 simonb #if 0 927 1.2 simonb /* 928 1.2 simonb * XXX this might be an issue for swapfiles. 929 1.2 simonb * see uvm_swap.c:1725 930 1.2 simonb * 931 1.2 simonb * XXXdeux: see above 932 1.2 simonb */ 933 1.2 simonb KASSERT((bp->b_flags & BC_NOCACHE) == 0); 934 1.2 simonb #endif 935 1.2 simonb KASSERT(bp->b_flags & B_LOCKED); 936 1.2 simonb 937 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_BUFFER, 938 1.2 simonb ("wapbl_remove_buf thread %d.%d removing buf %p with " 939 1.2 simonb "%d bytes %d bcount\n", 940 1.2 simonb curproc->p_pid, curlwp->l_lid, bp, bp->b_bufsize, bp->b_bcount)); 941 1.2 simonb 942 1.2 simonb KASSERT(wl->wl_bufbytes >= bp->b_bufsize); 943 1.2 simonb wl->wl_bufbytes -= bp->b_bufsize; 944 1.2 simonb KASSERT(wl->wl_bcount >= bp->b_bcount); 945 1.2 simonb wl->wl_bcount -= bp->b_bcount; 946 1.2 simonb KASSERT(wl->wl_bufcount > 0); 947 1.2 simonb wl->wl_bufcount--; 948 1.2 simonb KASSERT((wl->wl_bufcount == 0) == (wl->wl_bufbytes == 0)); 949 1.2 simonb KASSERT((wl->wl_bufcount == 0) == (wl->wl_bcount == 0)); 950 1.2 simonb LIST_REMOVE(bp, b_wapbllist); 951 1.2 simonb 952 1.2 simonb bp->b_flags &= ~B_LOCKED; 953 1.2 simonb } 954 1.2 simonb 955 1.2 simonb /* called from brelsel() in vfs_bio among other places */ 956 1.2 simonb void 957 1.2 simonb wapbl_remove_buf(struct wapbl * wl, struct buf *bp) 958 1.2 simonb { 959 1.2 simonb 960 1.2 simonb mutex_enter(&wl->wl_mtx); 961 1.2 simonb wapbl_remove_buf_locked(wl, bp); 962 1.2 simonb mutex_exit(&wl->wl_mtx); 963 1.2 simonb } 964 1.2 simonb 965 1.2 simonb void 966 1.2 simonb wapbl_resize_buf(struct wapbl *wl, struct buf *bp, long oldsz, long oldcnt) 967 1.2 simonb { 968 1.2 simonb 969 1.2 simonb KASSERT(bp->b_cflags & BC_BUSY); 970 1.2 simonb 971 1.2 simonb /* 972 1.2 simonb * XXX: why does this depend on B_LOCKED? otherwise the buf 973 1.2 simonb * is not for a transaction? if so, why is this called in the 974 1.2 simonb * first place? 975 1.2 simonb */ 976 1.2 simonb if (bp->b_flags & B_LOCKED) { 977 1.2 simonb mutex_enter(&wl->wl_mtx); 978 1.2 simonb wl->wl_bufbytes += bp->b_bufsize - oldsz; 979 1.2 simonb wl->wl_bcount += bp->b_bcount - oldcnt; 980 1.2 simonb mutex_exit(&wl->wl_mtx); 981 1.2 simonb } 982 1.2 simonb } 983 1.2 simonb 984 1.2 simonb #endif /* _KERNEL */ 985 1.2 simonb 986 1.2 simonb /****************************************************************/ 987 1.2 simonb /* Some utility inlines */ 988 1.2 simonb 989 1.2 simonb /* This is used to advance the pointer at old to new value at old+delta */ 990 1.2 simonb static __inline off_t 991 1.2 simonb wapbl_advance(size_t size, size_t off, off_t old, size_t delta) 992 1.2 simonb { 993 1.2 simonb off_t new; 994 1.2 simonb 995 1.2 simonb /* Define acceptable ranges for inputs. */ 996 1.2 simonb KASSERT(delta <= size); 997 1.2 simonb KASSERT((old == 0) || (old >= off)); 998 1.2 simonb KASSERT(old < (size + off)); 999 1.2 simonb 1000 1.2 simonb if ((old == 0) && (delta != 0)) 1001 1.2 simonb new = off + delta; 1002 1.2 simonb else if ((old + delta) < (size + off)) 1003 1.2 simonb new = old + delta; 1004 1.2 simonb else 1005 1.2 simonb new = (old + delta) - size; 1006 1.2 simonb 1007 1.2 simonb /* Note some interesting axioms */ 1008 1.2 simonb KASSERT((delta != 0) || (new == old)); 1009 1.2 simonb KASSERT((delta == 0) || (new != 0)); 1010 1.2 simonb KASSERT((delta != (size)) || (new == old)); 1011 1.2 simonb 1012 1.2 simonb /* Define acceptable ranges for output. */ 1013 1.2 simonb KASSERT((new == 0) || (new >= off)); 1014 1.2 simonb KASSERT(new < (size + off)); 1015 1.2 simonb return new; 1016 1.2 simonb } 1017 1.2 simonb 1018 1.2 simonb static __inline size_t 1019 1.2 simonb wapbl_space_used(size_t avail, off_t head, off_t tail) 1020 1.2 simonb { 1021 1.2 simonb 1022 1.2 simonb if (tail == 0) { 1023 1.2 simonb KASSERT(head == 0); 1024 1.2 simonb return 0; 1025 1.2 simonb } 1026 1.2 simonb return ((head + (avail - 1) - tail) % avail) + 1; 1027 1.2 simonb } 1028 1.2 simonb 1029 1.2 simonb static __inline size_t 1030 1.2 simonb wapbl_space_free(size_t avail, off_t head, off_t tail) 1031 1.2 simonb { 1032 1.2 simonb 1033 1.2 simonb return avail - wapbl_space_used(avail, head, tail); 1034 1.2 simonb } 1035 1.2 simonb 1036 1.2 simonb static __inline void 1037 1.2 simonb wapbl_advance_head(size_t size, size_t off, size_t delta, off_t *headp, 1038 1.2 simonb off_t *tailp) 1039 1.2 simonb { 1040 1.2 simonb off_t head = *headp; 1041 1.2 simonb off_t tail = *tailp; 1042 1.2 simonb 1043 1.2 simonb KASSERT(delta <= wapbl_space_free(size, head, tail)); 1044 1.2 simonb head = wapbl_advance(size, off, head, delta); 1045 1.2 simonb if ((tail == 0) && (head != 0)) 1046 1.2 simonb tail = off; 1047 1.2 simonb *headp = head; 1048 1.2 simonb *tailp = tail; 1049 1.2 simonb } 1050 1.2 simonb 1051 1.2 simonb static __inline void 1052 1.2 simonb wapbl_advance_tail(size_t size, size_t off, size_t delta, off_t *headp, 1053 1.2 simonb off_t *tailp) 1054 1.2 simonb { 1055 1.2 simonb off_t head = *headp; 1056 1.2 simonb off_t tail = *tailp; 1057 1.2 simonb 1058 1.2 simonb KASSERT(delta <= wapbl_space_used(size, head, tail)); 1059 1.2 simonb tail = wapbl_advance(size, off, tail, delta); 1060 1.2 simonb if (head == tail) { 1061 1.2 simonb head = tail = 0; 1062 1.2 simonb } 1063 1.2 simonb *headp = head; 1064 1.2 simonb *tailp = tail; 1065 1.2 simonb } 1066 1.2 simonb 1067 1.2 simonb #ifdef _KERNEL 1068 1.2 simonb 1069 1.2 simonb /****************************************************************/ 1070 1.2 simonb 1071 1.2 simonb /* 1072 1.2 simonb * Remove transactions whose buffers are completely flushed to disk. 1073 1.2 simonb * Will block until at least minfree space is available. 1074 1.2 simonb * only intended to be called from inside wapbl_flush and therefore 1075 1.2 simonb * does not protect against commit races with itself or with flush. 1076 1.2 simonb */ 1077 1.2 simonb static int 1078 1.2 simonb wapbl_truncate(struct wapbl *wl, size_t minfree, int waitonly) 1079 1.2 simonb { 1080 1.2 simonb size_t delta; 1081 1.2 simonb size_t avail; 1082 1.2 simonb off_t head; 1083 1.2 simonb off_t tail; 1084 1.2 simonb int error = 0; 1085 1.2 simonb 1086 1.2 simonb KASSERT(minfree <= (wl->wl_circ_size - wl->wl_reserved_bytes)); 1087 1.2 simonb KASSERT(rw_write_held(&wl->wl_rwlock)); 1088 1.2 simonb 1089 1.2 simonb mutex_enter(&wl->wl_mtx); 1090 1.2 simonb 1091 1.2 simonb /* 1092 1.2 simonb * First check to see if we have to do a commit 1093 1.2 simonb * at all. 1094 1.2 simonb */ 1095 1.2 simonb avail = wapbl_space_free(wl->wl_circ_size, wl->wl_head, wl->wl_tail); 1096 1.2 simonb if (minfree < avail) { 1097 1.2 simonb mutex_exit(&wl->wl_mtx); 1098 1.2 simonb return 0; 1099 1.2 simonb } 1100 1.2 simonb minfree -= avail; 1101 1.2 simonb while ((wl->wl_error_count == 0) && 1102 1.2 simonb (wl->wl_reclaimable_bytes < minfree)) { 1103 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_TRUNCATE, 1104 1.2 simonb ("wapbl_truncate: sleeping on %p wl=%p bytes=%zd " 1105 1.2 simonb "minfree=%zd\n", 1106 1.2 simonb &wl->wl_reclaimable_bytes, wl, wl->wl_reclaimable_bytes, 1107 1.2 simonb minfree)); 1108 1.2 simonb 1109 1.2 simonb cv_wait(&wl->wl_reclaimable_cv, &wl->wl_mtx); 1110 1.2 simonb } 1111 1.2 simonb if (wl->wl_reclaimable_bytes < minfree) { 1112 1.2 simonb KASSERT(wl->wl_error_count); 1113 1.2 simonb /* XXX maybe get actual error from buffer instead someday? */ 1114 1.2 simonb error = EIO; 1115 1.2 simonb } 1116 1.2 simonb head = wl->wl_head; 1117 1.2 simonb tail = wl->wl_tail; 1118 1.2 simonb delta = wl->wl_reclaimable_bytes; 1119 1.2 simonb 1120 1.2 simonb /* If all of of the entries are flushed, then be sure to keep 1121 1.2 simonb * the reserved bytes reserved. Watch out for discarded transactions, 1122 1.2 simonb * which could leave more bytes reserved than are reclaimable. 1123 1.2 simonb */ 1124 1.2 simonb if (SIMPLEQ_EMPTY(&wl->wl_entries) && 1125 1.2 simonb (delta >= wl->wl_reserved_bytes)) { 1126 1.2 simonb delta -= wl->wl_reserved_bytes; 1127 1.2 simonb } 1128 1.2 simonb wapbl_advance_tail(wl->wl_circ_size, wl->wl_circ_off, delta, &head, 1129 1.2 simonb &tail); 1130 1.2 simonb KDASSERT(wl->wl_reserved_bytes <= 1131 1.2 simonb wapbl_space_used(wl->wl_circ_size, head, tail)); 1132 1.2 simonb mutex_exit(&wl->wl_mtx); 1133 1.2 simonb 1134 1.2 simonb if (error) 1135 1.2 simonb return error; 1136 1.2 simonb 1137 1.2 simonb if (waitonly) 1138 1.2 simonb return 0; 1139 1.2 simonb 1140 1.2 simonb /* 1141 1.2 simonb * This is where head, tail and delta are unprotected 1142 1.2 simonb * from races against itself or flush. This is ok since 1143 1.2 simonb * we only call this routine from inside flush itself. 1144 1.2 simonb * 1145 1.2 simonb * XXX: how can it race against itself when accessed only 1146 1.2 simonb * from behind the write-locked rwlock? 1147 1.2 simonb */ 1148 1.2 simonb error = wapbl_write_commit(wl, head, tail); 1149 1.2 simonb if (error) 1150 1.2 simonb return error; 1151 1.2 simonb 1152 1.2 simonb wl->wl_head = head; 1153 1.2 simonb wl->wl_tail = tail; 1154 1.2 simonb 1155 1.2 simonb mutex_enter(&wl->wl_mtx); 1156 1.2 simonb KASSERT(wl->wl_reclaimable_bytes >= delta); 1157 1.2 simonb wl->wl_reclaimable_bytes -= delta; 1158 1.2 simonb mutex_exit(&wl->wl_mtx); 1159 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_TRUNCATE, 1160 1.2 simonb ("wapbl_truncate thread %d.%d truncating %zu bytes\n", 1161 1.2 simonb curproc->p_pid, curlwp->l_lid, delta)); 1162 1.2 simonb 1163 1.2 simonb return 0; 1164 1.2 simonb } 1165 1.2 simonb 1166 1.2 simonb /****************************************************************/ 1167 1.2 simonb 1168 1.2 simonb void 1169 1.2 simonb wapbl_biodone(struct buf *bp) 1170 1.2 simonb { 1171 1.2 simonb struct wapbl_entry *we = bp->b_private; 1172 1.2 simonb struct wapbl *wl = we->we_wapbl; 1173 1.2 simonb 1174 1.2 simonb /* 1175 1.2 simonb * Handle possible flushing of buffers after log has been 1176 1.2 simonb * decomissioned. 1177 1.2 simonb */ 1178 1.2 simonb if (!wl) { 1179 1.2 simonb KASSERT(we->we_bufcount > 0); 1180 1.2 simonb we->we_bufcount--; 1181 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1182 1.2 simonb KASSERT(we->we_unsynced_bufbytes >= bp->b_bufsize); 1183 1.2 simonb we->we_unsynced_bufbytes -= bp->b_bufsize; 1184 1.2 simonb #endif 1185 1.2 simonb 1186 1.2 simonb if (we->we_bufcount == 0) { 1187 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1188 1.2 simonb KASSERT(we->we_unsynced_bufbytes == 0); 1189 1.2 simonb #endif 1190 1.2 simonb wapbl_free(we); 1191 1.2 simonb } 1192 1.2 simonb 1193 1.2 simonb brelse(bp, 0); 1194 1.2 simonb return; 1195 1.2 simonb } 1196 1.2 simonb 1197 1.2 simonb #ifdef ohbother 1198 1.2 simonb KDASSERT(bp->b_flags & B_DONE); 1199 1.2 simonb KDASSERT(!(bp->b_flags & B_DELWRI)); 1200 1.2 simonb KDASSERT(bp->b_flags & B_ASYNC); 1201 1.2 simonb KDASSERT(bp->b_flags & B_BUSY); 1202 1.2 simonb KDASSERT(!(bp->b_flags & B_LOCKED)); 1203 1.2 simonb KDASSERT(!(bp->b_flags & B_READ)); 1204 1.2 simonb KDASSERT(!(bp->b_flags & B_INVAL)); 1205 1.2 simonb KDASSERT(!(bp->b_flags & B_NOCACHE)); 1206 1.2 simonb #endif 1207 1.2 simonb 1208 1.2 simonb if (bp->b_error) { 1209 1.2 simonb #ifdef notyet /* Can't currently handle possible dirty buffer reuse */ 1210 1.2 simonb XXXpooka: interfaces not fully updated 1211 1.2 simonb Note: this was not enabled in the original patch 1212 1.2 simonb against netbsd4 either. I don't know if comment 1213 1.2 simonb above is true or not. 1214 1.2 simonb 1215 1.2 simonb /* 1216 1.2 simonb * If an error occurs, report the error and leave the 1217 1.2 simonb * buffer as a delayed write on the LRU queue. 1218 1.2 simonb * restarting the write would likely result in 1219 1.2 simonb * an error spinloop, so let it be done harmlessly 1220 1.2 simonb * by the syncer. 1221 1.2 simonb */ 1222 1.2 simonb bp->b_flags &= ~(B_DONE); 1223 1.2 simonb simple_unlock(&bp->b_interlock); 1224 1.2 simonb 1225 1.2 simonb if (we->we_error == 0) { 1226 1.2 simonb mutex_enter(&wl->wl_mtx); 1227 1.2 simonb wl->wl_error_count++; 1228 1.2 simonb mutex_exit(&wl->wl_mtx); 1229 1.2 simonb cv_broadcast(&wl->wl_reclaimable_cv); 1230 1.2 simonb } 1231 1.2 simonb we->we_error = bp->b_error; 1232 1.2 simonb bp->b_error = 0; 1233 1.2 simonb brelse(bp); 1234 1.2 simonb return; 1235 1.2 simonb #else 1236 1.2 simonb /* For now, just mark the log permanently errored out */ 1237 1.2 simonb 1238 1.2 simonb mutex_enter(&wl->wl_mtx); 1239 1.2 simonb if (wl->wl_error_count == 0) { 1240 1.2 simonb wl->wl_error_count++; 1241 1.2 simonb cv_broadcast(&wl->wl_reclaimable_cv); 1242 1.2 simonb } 1243 1.2 simonb mutex_exit(&wl->wl_mtx); 1244 1.2 simonb #endif 1245 1.2 simonb } 1246 1.2 simonb 1247 1.2 simonb mutex_enter(&wl->wl_mtx); 1248 1.2 simonb 1249 1.2 simonb KASSERT(we->we_bufcount > 0); 1250 1.2 simonb we->we_bufcount--; 1251 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1252 1.2 simonb KASSERT(we->we_unsynced_bufbytes >= bp->b_bufsize); 1253 1.2 simonb we->we_unsynced_bufbytes -= bp->b_bufsize; 1254 1.2 simonb KASSERT(wl->wl_unsynced_bufbytes >= bp->b_bufsize); 1255 1.2 simonb wl->wl_unsynced_bufbytes -= bp->b_bufsize; 1256 1.2 simonb #endif 1257 1.2 simonb 1258 1.2 simonb /* 1259 1.2 simonb * If the current transaction can be reclaimed, start 1260 1.2 simonb * at the beginning and reclaim any consecutive reclaimable 1261 1.2 simonb * transactions. If we successfully reclaim anything, 1262 1.2 simonb * then wakeup anyone waiting for the reclaim. 1263 1.2 simonb */ 1264 1.2 simonb if (we->we_bufcount == 0) { 1265 1.2 simonb size_t delta = 0; 1266 1.2 simonb int errcnt = 0; 1267 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1268 1.2 simonb KDASSERT(we->we_unsynced_bufbytes == 0); 1269 1.2 simonb #endif 1270 1.2 simonb /* 1271 1.2 simonb * clear any posted error, since the buffer it came from 1272 1.2 simonb * has successfully flushed by now 1273 1.2 simonb */ 1274 1.2 simonb while ((we = SIMPLEQ_FIRST(&wl->wl_entries)) && 1275 1.2 simonb (we->we_bufcount == 0)) { 1276 1.2 simonb delta += we->we_reclaimable_bytes; 1277 1.2 simonb if (we->we_error) 1278 1.2 simonb errcnt++; 1279 1.2 simonb SIMPLEQ_REMOVE_HEAD(&wl->wl_entries, we_entries); 1280 1.2 simonb wapbl_free(we); 1281 1.2 simonb } 1282 1.2 simonb 1283 1.2 simonb if (delta) { 1284 1.2 simonb wl->wl_reclaimable_bytes += delta; 1285 1.2 simonb KASSERT(wl->wl_error_count >= errcnt); 1286 1.2 simonb wl->wl_error_count -= errcnt; 1287 1.2 simonb cv_broadcast(&wl->wl_reclaimable_cv); 1288 1.2 simonb } 1289 1.2 simonb } 1290 1.2 simonb 1291 1.2 simonb mutex_exit(&wl->wl_mtx); 1292 1.2 simonb brelse(bp, 0); 1293 1.2 simonb } 1294 1.2 simonb 1295 1.2 simonb /* 1296 1.2 simonb * Write transactions to disk + start I/O for contents 1297 1.2 simonb */ 1298 1.2 simonb int 1299 1.2 simonb wapbl_flush(struct wapbl *wl, int waitfor) 1300 1.2 simonb { 1301 1.2 simonb struct buf *bp; 1302 1.2 simonb struct wapbl_entry *we; 1303 1.2 simonb off_t off; 1304 1.2 simonb off_t head; 1305 1.2 simonb off_t tail; 1306 1.2 simonb size_t delta = 0; 1307 1.2 simonb size_t flushsize; 1308 1.2 simonb size_t reserved; 1309 1.2 simonb int error = 0; 1310 1.2 simonb 1311 1.2 simonb /* 1312 1.2 simonb * Do a quick check to see if a full flush can be skipped 1313 1.2 simonb * This assumes that the flush callback does not need to be called 1314 1.2 simonb * unless there are other outstanding bufs. 1315 1.2 simonb */ 1316 1.2 simonb if (!waitfor) { 1317 1.2 simonb size_t nbufs; 1318 1.2 simonb mutex_enter(&wl->wl_mtx); /* XXX need mutex here to 1319 1.2 simonb protect the KASSERTS */ 1320 1.2 simonb nbufs = wl->wl_bufcount; 1321 1.2 simonb KASSERT((wl->wl_bufcount == 0) == (wl->wl_bufbytes == 0)); 1322 1.2 simonb KASSERT((wl->wl_bufcount == 0) == (wl->wl_bcount == 0)); 1323 1.2 simonb mutex_exit(&wl->wl_mtx); 1324 1.2 simonb if (nbufs == 0) 1325 1.2 simonb return 0; 1326 1.2 simonb } 1327 1.2 simonb 1328 1.2 simonb /* 1329 1.2 simonb * XXX we may consider using LK_UPGRADE here 1330 1.2 simonb * if we want to call flush from inside a transaction 1331 1.2 simonb */ 1332 1.2 simonb rw_enter(&wl->wl_rwlock, RW_WRITER); 1333 1.2 simonb wl->wl_flush(wl->wl_mount, wl->wl_deallocblks, wl->wl_dealloclens, 1334 1.2 simonb wl->wl_dealloccnt); 1335 1.2 simonb 1336 1.2 simonb /* 1337 1.2 simonb * Now that we are fully locked and flushed, 1338 1.2 simonb * do another check for nothing to do. 1339 1.2 simonb */ 1340 1.2 simonb if (wl->wl_bufcount == 0) { 1341 1.2 simonb goto out; 1342 1.2 simonb } 1343 1.2 simonb 1344 1.2 simonb #if 0 1345 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_FLUSH, 1346 1.2 simonb ("wapbl_flush thread %d.%d flushing entries with " 1347 1.2 simonb "bufcount=%zu bufbytes=%zu\n", 1348 1.2 simonb curproc->p_pid, curlwp->l_lid, wl->wl_bufcount, 1349 1.2 simonb wl->wl_bufbytes)); 1350 1.2 simonb #endif 1351 1.2 simonb 1352 1.2 simonb /* Calculate amount of space needed to flush */ 1353 1.2 simonb flushsize = wapbl_transaction_len(wl); 1354 1.2 simonb 1355 1.2 simonb if (flushsize > (wl->wl_circ_size - wl->wl_reserved_bytes)) { 1356 1.2 simonb /* 1357 1.2 simonb * XXX this could be handled more gracefully, perhaps place 1358 1.2 simonb * only a partial transaction in the log and allow the 1359 1.2 simonb * remaining to flush without the protection of the journal. 1360 1.2 simonb */ 1361 1.2 simonb panic("wapbl_flush: current transaction too big to flush\n"); 1362 1.2 simonb } 1363 1.2 simonb 1364 1.2 simonb error = wapbl_truncate(wl, flushsize, 0); 1365 1.2 simonb if (error) 1366 1.2 simonb goto out2; 1367 1.2 simonb 1368 1.2 simonb off = wl->wl_head; 1369 1.2 simonb KASSERT((off == 0) || ((off >= wl->wl_circ_off) && 1370 1.2 simonb (off < wl->wl_circ_off + wl->wl_circ_size))); 1371 1.2 simonb error = wapbl_write_blocks(wl, &off); 1372 1.2 simonb if (error) 1373 1.2 simonb goto out2; 1374 1.2 simonb error = wapbl_write_revocations(wl, &off); 1375 1.2 simonb if (error) 1376 1.2 simonb goto out2; 1377 1.2 simonb error = wapbl_write_inodes(wl, &off); 1378 1.2 simonb if (error) 1379 1.2 simonb goto out2; 1380 1.2 simonb 1381 1.2 simonb reserved = 0; 1382 1.2 simonb if (wl->wl_inohashcnt) 1383 1.2 simonb reserved = wapbl_transaction_inodes_len(wl); 1384 1.2 simonb 1385 1.2 simonb head = wl->wl_head; 1386 1.2 simonb tail = wl->wl_tail; 1387 1.2 simonb 1388 1.2 simonb wapbl_advance_head(wl->wl_circ_size, wl->wl_circ_off, flushsize, 1389 1.2 simonb &head, &tail); 1390 1.2 simonb #ifdef WAPBL_DEBUG 1391 1.2 simonb if (head != off) { 1392 1.2 simonb panic("lost head! head=%"PRIdMAX" tail=%" PRIdMAX 1393 1.2 simonb " off=%"PRIdMAX" flush=%zu\n", 1394 1.2 simonb (intmax_t)head, (intmax_t)tail, (intmax_t)off, 1395 1.2 simonb flushsize); 1396 1.2 simonb } 1397 1.2 simonb #else 1398 1.2 simonb KASSERT(head == off); 1399 1.2 simonb #endif 1400 1.2 simonb 1401 1.2 simonb /* Opportunistically move the tail forward if we can */ 1402 1.2 simonb if (!wapbl_lazy_truncate) { 1403 1.2 simonb mutex_enter(&wl->wl_mtx); 1404 1.2 simonb delta = wl->wl_reclaimable_bytes; 1405 1.2 simonb mutex_exit(&wl->wl_mtx); 1406 1.2 simonb wapbl_advance_tail(wl->wl_circ_size, wl->wl_circ_off, delta, 1407 1.2 simonb &head, &tail); 1408 1.2 simonb } 1409 1.2 simonb 1410 1.2 simonb error = wapbl_write_commit(wl, head, tail); 1411 1.2 simonb if (error) 1412 1.2 simonb goto out2; 1413 1.2 simonb 1414 1.2 simonb /* poolme? or kmemme? */ 1415 1.2 simonb we = wapbl_calloc(1, sizeof(*we)); 1416 1.2 simonb 1417 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1418 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_FLUSH, 1419 1.2 simonb ("wapbl_flush: thread %d.%d head+=%zu tail+=%zu used=%zu" 1420 1.2 simonb " unsynced=%zu" 1421 1.2 simonb "\n\tbufcount=%zu bufbytes=%zu bcount=%zu deallocs=%d " 1422 1.2 simonb "inodes=%d\n", 1423 1.2 simonb curproc->p_pid, curlwp->l_lid, flushsize, delta, 1424 1.2 simonb wapbl_space_used(wl->wl_circ_size, head, tail), 1425 1.2 simonb wl->wl_unsynced_bufbytes, wl->wl_bufcount, 1426 1.2 simonb wl->wl_bufbytes, wl->wl_bcount, wl->wl_dealloccnt, 1427 1.2 simonb wl->wl_inohashcnt)); 1428 1.2 simonb #else 1429 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_FLUSH, 1430 1.2 simonb ("wapbl_flush: thread %d.%d head+=%zu tail+=%zu used=%zu" 1431 1.2 simonb "\n\tbufcount=%zu bufbytes=%zu bcount=%zu deallocs=%d " 1432 1.2 simonb "inodes=%d\n", 1433 1.2 simonb curproc->p_pid, curlwp->l_lid, flushsize, delta, 1434 1.2 simonb wapbl_space_used(wl->wl_circ_size, head, tail), 1435 1.2 simonb wl->wl_bufcount, wl->wl_bufbytes, wl->wl_bcount, 1436 1.2 simonb wl->wl_dealloccnt, wl->wl_inohashcnt)); 1437 1.2 simonb #endif 1438 1.2 simonb 1439 1.2 simonb 1440 1.2 simonb mutex_enter(&bufcache_lock); 1441 1.2 simonb mutex_enter(&wl->wl_mtx); 1442 1.2 simonb 1443 1.2 simonb wl->wl_reserved_bytes = reserved; 1444 1.2 simonb wl->wl_head = head; 1445 1.2 simonb wl->wl_tail = tail; 1446 1.2 simonb KASSERT(wl->wl_reclaimable_bytes >= delta); 1447 1.2 simonb wl->wl_reclaimable_bytes -= delta; 1448 1.2 simonb wl->wl_dealloccnt = 0; 1449 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1450 1.2 simonb wl->wl_unsynced_bufbytes += wl->wl_bufbytes; 1451 1.2 simonb #endif 1452 1.2 simonb 1453 1.2 simonb we->we_wapbl = wl; 1454 1.2 simonb we->we_bufcount = wl->wl_bufcount; 1455 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1456 1.2 simonb we->we_unsynced_bufbytes = wl->wl_bufbytes; 1457 1.2 simonb #endif 1458 1.2 simonb we->we_reclaimable_bytes = flushsize; 1459 1.2 simonb we->we_error = 0; 1460 1.2 simonb SIMPLEQ_INSERT_TAIL(&wl->wl_entries, we, we_entries); 1461 1.2 simonb 1462 1.2 simonb /* 1463 1.2 simonb * this flushes bufs in reverse order than they were queued 1464 1.2 simonb * it shouldn't matter, but if we care we could use TAILQ instead. 1465 1.2 simonb * XXX Note they will get put on the lru queue when they flush 1466 1.2 simonb * so we might actually want to change this to preserve order. 1467 1.2 simonb */ 1468 1.2 simonb while ((bp = LIST_FIRST(&wl->wl_bufs)) != NULL) { 1469 1.2 simonb if (bbusy(bp, 0, 0, &wl->wl_mtx)) { 1470 1.2 simonb continue; 1471 1.2 simonb } 1472 1.2 simonb bp->b_iodone = wapbl_biodone; 1473 1.2 simonb bp->b_private = we; 1474 1.2 simonb bremfree(bp); 1475 1.2 simonb wapbl_remove_buf_locked(wl, bp); 1476 1.2 simonb mutex_exit(&wl->wl_mtx); 1477 1.2 simonb mutex_exit(&bufcache_lock); 1478 1.2 simonb bawrite(bp); 1479 1.2 simonb mutex_enter(&bufcache_lock); 1480 1.2 simonb mutex_enter(&wl->wl_mtx); 1481 1.2 simonb } 1482 1.2 simonb mutex_exit(&wl->wl_mtx); 1483 1.2 simonb mutex_exit(&bufcache_lock); 1484 1.2 simonb 1485 1.2 simonb #if 0 1486 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_FLUSH, 1487 1.2 simonb ("wapbl_flush thread %d.%d done flushing entries...\n", 1488 1.2 simonb curproc->p_pid, curlwp->l_lid)); 1489 1.2 simonb #endif 1490 1.2 simonb 1491 1.2 simonb out: 1492 1.2 simonb 1493 1.2 simonb /* 1494 1.2 simonb * If the waitfor flag is set, don't return until everything is 1495 1.2 simonb * fully flushed and the on disk log is empty. 1496 1.2 simonb */ 1497 1.2 simonb if (waitfor) { 1498 1.2 simonb error = wapbl_truncate(wl, wl->wl_circ_size - 1499 1.2 simonb wl->wl_reserved_bytes, wapbl_lazy_truncate); 1500 1.2 simonb } 1501 1.2 simonb 1502 1.2 simonb out2: 1503 1.2 simonb if (error) { 1504 1.2 simonb wl->wl_flush_abort(wl->wl_mount, wl->wl_deallocblks, 1505 1.2 simonb wl->wl_dealloclens, wl->wl_dealloccnt); 1506 1.2 simonb } 1507 1.2 simonb 1508 1.2 simonb #ifdef WAPBL_DEBUG_PRINT 1509 1.2 simonb if (error) { 1510 1.2 simonb pid_t pid = -1; 1511 1.2 simonb lwpid_t lid = -1; 1512 1.2 simonb if (curproc) 1513 1.2 simonb pid = curproc->p_pid; 1514 1.2 simonb if (curlwp) 1515 1.2 simonb lid = curlwp->l_lid; 1516 1.2 simonb mutex_enter(&wl->wl_mtx); 1517 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1518 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_ERROR, 1519 1.2 simonb ("wapbl_flush: thread %d.%d aborted flush: " 1520 1.2 simonb "error = %d\n" 1521 1.2 simonb "\tbufcount=%zu bufbytes=%zu bcount=%zu " 1522 1.2 simonb "deallocs=%d inodes=%d\n" 1523 1.2 simonb "\terrcnt = %d, reclaimable=%zu reserved=%zu " 1524 1.2 simonb "unsynced=%zu\n", 1525 1.2 simonb pid, lid, error, wl->wl_bufcount, 1526 1.2 simonb wl->wl_bufbytes, wl->wl_bcount, 1527 1.2 simonb wl->wl_dealloccnt, wl->wl_inohashcnt, 1528 1.2 simonb wl->wl_error_count, wl->wl_reclaimable_bytes, 1529 1.2 simonb wl->wl_reserved_bytes, wl->wl_unsynced_bufbytes)); 1530 1.2 simonb SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) { 1531 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_ERROR, 1532 1.2 simonb ("\tentry: bufcount = %zu, reclaimable = %zu, " 1533 1.2 simonb "error = %d, unsynced = %zu\n", 1534 1.2 simonb we->we_bufcount, we->we_reclaimable_bytes, 1535 1.2 simonb we->we_error, we->we_unsynced_bufbytes)); 1536 1.2 simonb } 1537 1.2 simonb #else 1538 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_ERROR, 1539 1.2 simonb ("wapbl_flush: thread %d.%d aborted flush: " 1540 1.2 simonb "error = %d\n" 1541 1.2 simonb "\tbufcount=%zu bufbytes=%zu bcount=%zu " 1542 1.2 simonb "deallocs=%d inodes=%d\n" 1543 1.2 simonb "\terrcnt = %d, reclaimable=%zu reserved=%zu\n", 1544 1.2 simonb pid, lid, error, wl->wl_bufcount, 1545 1.2 simonb wl->wl_bufbytes, wl->wl_bcount, 1546 1.2 simonb wl->wl_dealloccnt, wl->wl_inohashcnt, 1547 1.2 simonb wl->wl_error_count, wl->wl_reclaimable_bytes, 1548 1.2 simonb wl->wl_reserved_bytes)); 1549 1.2 simonb SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) { 1550 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_ERROR, 1551 1.2 simonb ("\tentry: bufcount = %zu, reclaimable = %zu, " 1552 1.2 simonb "error = %d\n", we->we_bufcount, 1553 1.2 simonb we->we_reclaimable_bytes, we->we_error)); 1554 1.2 simonb } 1555 1.2 simonb #endif 1556 1.2 simonb mutex_exit(&wl->wl_mtx); 1557 1.2 simonb } 1558 1.2 simonb #endif 1559 1.2 simonb 1560 1.2 simonb rw_exit(&wl->wl_rwlock); 1561 1.2 simonb return error; 1562 1.2 simonb } 1563 1.2 simonb 1564 1.2 simonb /****************************************************************/ 1565 1.2 simonb 1566 1.2 simonb void 1567 1.2 simonb wapbl_jlock_assert(struct wapbl *wl) 1568 1.2 simonb { 1569 1.2 simonb 1570 1.2 simonb #ifdef WAPBL_DEBUG_SERIALIZE 1571 1.2 simonb KASSERT(rw_write_held(&wl->wl_rwlock)); 1572 1.2 simonb #else 1573 1.2 simonb KASSERT(rw_read_held(&wl->wl_rwlock) || rw_write_held(&wl->wl_rwlock)); 1574 1.2 simonb #endif 1575 1.2 simonb } 1576 1.2 simonb 1577 1.2 simonb void 1578 1.2 simonb wapbl_junlock_assert(struct wapbl *wl) 1579 1.2 simonb { 1580 1.2 simonb 1581 1.2 simonb #ifdef WAPBL_DEBUG_SERIALIZE 1582 1.2 simonb KASSERT(!rw_write_held(&wl->wl_rwlock)); 1583 1.2 simonb #endif 1584 1.2 simonb } 1585 1.2 simonb 1586 1.2 simonb /****************************************************************/ 1587 1.2 simonb 1588 1.2 simonb /* locks missing */ 1589 1.2 simonb void 1590 1.2 simonb wapbl_print(struct wapbl *wl, 1591 1.2 simonb int full, 1592 1.2 simonb void (*pr)(const char *, ...)) 1593 1.2 simonb { 1594 1.2 simonb struct buf *bp; 1595 1.2 simonb struct wapbl_entry *we; 1596 1.2 simonb (*pr)("wapbl %p", wl); 1597 1.2 simonb (*pr)("\nlogvp = %p, devvp = %p, logpbn = %"PRId64"\n", 1598 1.2 simonb wl->wl_logvp, wl->wl_devvp, wl->wl_logpbn); 1599 1.2 simonb (*pr)("circ = %zu, header = %zu, head = %"PRIdMAX" tail = %"PRIdMAX"\n", 1600 1.2 simonb wl->wl_circ_size, wl->wl_circ_off, 1601 1.2 simonb (intmax_t)wl->wl_head, (intmax_t)wl->wl_tail); 1602 1.2 simonb (*pr)("fs_dev_bshift = %d, log_dev_bshift = %d\n", 1603 1.2 simonb wl->wl_log_dev_bshift, wl->wl_fs_dev_bshift); 1604 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1605 1.2 simonb (*pr)("bufcount = %zu, bufbytes = %zu bcount = %zu reclaimable = %zu " 1606 1.2 simonb "reserved = %zu errcnt = %d unsynced = %zu\n", 1607 1.2 simonb wl->wl_bufcount, wl->wl_bufbytes, wl->wl_bcount, 1608 1.2 simonb wl->wl_reclaimable_bytes, wl->wl_reserved_bytes, 1609 1.2 simonb wl->wl_error_count, wl->wl_unsynced_bufbytes); 1610 1.2 simonb #else 1611 1.2 simonb (*pr)("bufcount = %zu, bufbytes = %zu bcount = %zu reclaimable = %zu " 1612 1.2 simonb "reserved = %zu errcnt = %d\n", wl->wl_bufcount, wl->wl_bufbytes, 1613 1.2 simonb wl->wl_bcount, wl->wl_reclaimable_bytes, wl->wl_reserved_bytes, 1614 1.2 simonb wl->wl_error_count); 1615 1.2 simonb #endif 1616 1.2 simonb (*pr)("\tdealloccnt = %d, dealloclim = %d\n", 1617 1.2 simonb wl->wl_dealloccnt, wl->wl_dealloclim); 1618 1.2 simonb (*pr)("\tinohashcnt = %d, inohashmask = 0x%08x\n", 1619 1.2 simonb wl->wl_inohashcnt, wl->wl_inohashmask); 1620 1.2 simonb (*pr)("entries:\n"); 1621 1.2 simonb SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) { 1622 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1623 1.2 simonb (*pr)("\tbufcount = %zu, reclaimable = %zu, error = %d, " 1624 1.2 simonb "unsynced = %zu\n", 1625 1.2 simonb we->we_bufcount, we->we_reclaimable_bytes, 1626 1.2 simonb we->we_error, we->we_unsynced_bufbytes); 1627 1.2 simonb #else 1628 1.2 simonb (*pr)("\tbufcount = %zu, reclaimable = %zu, error = %d\n", 1629 1.2 simonb we->we_bufcount, we->we_reclaimable_bytes, we->we_error); 1630 1.2 simonb #endif 1631 1.2 simonb } 1632 1.2 simonb if (full) { 1633 1.2 simonb int cnt = 0; 1634 1.2 simonb (*pr)("bufs ="); 1635 1.2 simonb LIST_FOREACH(bp, &wl->wl_bufs, b_wapbllist) { 1636 1.2 simonb if (!LIST_NEXT(bp, b_wapbllist)) { 1637 1.2 simonb (*pr)(" %p", bp); 1638 1.2 simonb } else if ((++cnt % 6) == 0) { 1639 1.2 simonb (*pr)(" %p,\n\t", bp); 1640 1.2 simonb } else { 1641 1.2 simonb (*pr)(" %p,", bp); 1642 1.2 simonb } 1643 1.2 simonb } 1644 1.2 simonb (*pr)("\n"); 1645 1.2 simonb 1646 1.2 simonb (*pr)("dealloced blks = "); 1647 1.2 simonb { 1648 1.2 simonb int i; 1649 1.2 simonb cnt = 0; 1650 1.2 simonb for (i = 0; i < wl->wl_dealloccnt; i++) { 1651 1.2 simonb (*pr)(" %"PRId64":%d,", 1652 1.2 simonb wl->wl_deallocblks[i], 1653 1.2 simonb wl->wl_dealloclens[i]); 1654 1.2 simonb if ((++cnt % 4) == 0) { 1655 1.2 simonb (*pr)("\n\t"); 1656 1.2 simonb } 1657 1.2 simonb } 1658 1.2 simonb } 1659 1.2 simonb (*pr)("\n"); 1660 1.2 simonb 1661 1.2 simonb (*pr)("registered inodes = "); 1662 1.2 simonb { 1663 1.2 simonb int i; 1664 1.2 simonb cnt = 0; 1665 1.2 simonb for (i = 0; i <= wl->wl_inohashmask; i++) { 1666 1.2 simonb struct wapbl_ino_head *wih; 1667 1.2 simonb struct wapbl_ino *wi; 1668 1.2 simonb 1669 1.2 simonb wih = &wl->wl_inohash[i]; 1670 1.2 simonb LIST_FOREACH(wi, wih, wi_hash) { 1671 1.2 simonb if (wi->wi_ino == 0) 1672 1.2 simonb continue; 1673 1.2 simonb (*pr)(" %"PRId32"/0%06"PRIo32",", 1674 1.2 simonb wi->wi_ino, wi->wi_mode); 1675 1.2 simonb if ((++cnt % 4) == 0) { 1676 1.2 simonb (*pr)("\n\t"); 1677 1.2 simonb } 1678 1.2 simonb } 1679 1.2 simonb } 1680 1.2 simonb (*pr)("\n"); 1681 1.2 simonb } 1682 1.2 simonb } 1683 1.2 simonb } 1684 1.2 simonb 1685 1.2 simonb #if defined(WAPBL_DEBUG) || defined(DDB) 1686 1.2 simonb void 1687 1.2 simonb wapbl_dump(struct wapbl *wl) 1688 1.2 simonb { 1689 1.2 simonb #if defined(WAPBL_DEBUG) 1690 1.2 simonb if (!wl) 1691 1.2 simonb wl = wapbl_debug_wl; 1692 1.2 simonb #endif 1693 1.2 simonb if (!wl) 1694 1.2 simonb return; 1695 1.2 simonb wapbl_print(wl, 1, printf); 1696 1.2 simonb } 1697 1.2 simonb #endif 1698 1.2 simonb 1699 1.2 simonb /****************************************************************/ 1700 1.2 simonb 1701 1.2 simonb void 1702 1.2 simonb wapbl_register_deallocation(struct wapbl *wl, daddr_t blk, int len) 1703 1.2 simonb { 1704 1.2 simonb 1705 1.2 simonb wapbl_jlock_assert(wl); 1706 1.2 simonb 1707 1.2 simonb /* XXX should eventually instead tie this into resource estimation */ 1708 1.2 simonb /* XXX this KASSERT needs locking/mutex analysis */ 1709 1.2 simonb KASSERT(wl->wl_dealloccnt < wl->wl_dealloclim); 1710 1.2 simonb wl->wl_deallocblks[wl->wl_dealloccnt] = blk; 1711 1.2 simonb wl->wl_dealloclens[wl->wl_dealloccnt] = len; 1712 1.2 simonb wl->wl_dealloccnt++; 1713 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_ALLOC, 1714 1.2 simonb ("wapbl_register_deallocation: blk=%"PRId64" len=%d\n", blk, len)); 1715 1.2 simonb } 1716 1.2 simonb 1717 1.2 simonb /****************************************************************/ 1718 1.2 simonb 1719 1.2 simonb static void 1720 1.2 simonb wapbl_inodetrk_init(struct wapbl *wl, u_int size) 1721 1.2 simonb { 1722 1.2 simonb 1723 1.2 simonb wl->wl_inohash = hashinit(size, HASH_LIST, true, &wl->wl_inohashmask); 1724 1.2 simonb if (atomic_inc_uint_nv(&wapbl_ino_pool_refcount) == 1) { 1725 1.2 simonb pool_init(&wapbl_ino_pool, sizeof(struct wapbl_ino), 0, 0, 0, 1726 1.2 simonb "wapblinopl", &pool_allocator_nointr, IPL_NONE); 1727 1.2 simonb } 1728 1.2 simonb } 1729 1.2 simonb 1730 1.2 simonb static void 1731 1.2 simonb wapbl_inodetrk_free(struct wapbl *wl) 1732 1.2 simonb { 1733 1.2 simonb 1734 1.2 simonb /* XXX this KASSERT needs locking/mutex analysis */ 1735 1.2 simonb KASSERT(wl->wl_inohashcnt == 0); 1736 1.2 simonb hashdone(wl->wl_inohash, HASH_LIST, wl->wl_inohashmask); 1737 1.2 simonb if (atomic_dec_uint_nv(&wapbl_ino_pool_refcount) == 0) { 1738 1.2 simonb pool_destroy(&wapbl_ino_pool); 1739 1.2 simonb } 1740 1.2 simonb } 1741 1.2 simonb 1742 1.2 simonb static struct wapbl_ino * 1743 1.2 simonb wapbl_inodetrk_get(struct wapbl *wl, ino_t ino) 1744 1.2 simonb { 1745 1.2 simonb struct wapbl_ino_head *wih; 1746 1.2 simonb struct wapbl_ino *wi; 1747 1.2 simonb 1748 1.2 simonb KASSERT(mutex_owned(&wl->wl_mtx)); 1749 1.2 simonb 1750 1.2 simonb wih = &wl->wl_inohash[ino & wl->wl_inohashmask]; 1751 1.2 simonb LIST_FOREACH(wi, wih, wi_hash) { 1752 1.2 simonb if (ino == wi->wi_ino) 1753 1.2 simonb return wi; 1754 1.2 simonb } 1755 1.2 simonb return 0; 1756 1.2 simonb } 1757 1.2 simonb 1758 1.2 simonb void 1759 1.2 simonb wapbl_register_inode(struct wapbl *wl, ino_t ino, mode_t mode) 1760 1.2 simonb { 1761 1.2 simonb struct wapbl_ino_head *wih; 1762 1.2 simonb struct wapbl_ino *wi; 1763 1.2 simonb 1764 1.2 simonb wi = pool_get(&wapbl_ino_pool, PR_WAITOK); 1765 1.2 simonb 1766 1.2 simonb mutex_enter(&wl->wl_mtx); 1767 1.2 simonb if (wapbl_inodetrk_get(wl, ino) == NULL) { 1768 1.2 simonb wi->wi_ino = ino; 1769 1.2 simonb wi->wi_mode = mode; 1770 1.2 simonb wih = &wl->wl_inohash[ino & wl->wl_inohashmask]; 1771 1.2 simonb LIST_INSERT_HEAD(wih, wi, wi_hash); 1772 1.2 simonb wl->wl_inohashcnt++; 1773 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_INODE, 1774 1.2 simonb ("wapbl_register_inode: ino=%"PRId64"\n", ino)); 1775 1.2 simonb mutex_exit(&wl->wl_mtx); 1776 1.2 simonb } else { 1777 1.2 simonb mutex_exit(&wl->wl_mtx); 1778 1.2 simonb pool_put(&wapbl_ino_pool, wi); 1779 1.2 simonb } 1780 1.2 simonb } 1781 1.2 simonb 1782 1.2 simonb void 1783 1.2 simonb wapbl_unregister_inode(struct wapbl *wl, ino_t ino, mode_t mode) 1784 1.2 simonb { 1785 1.2 simonb struct wapbl_ino *wi; 1786 1.2 simonb 1787 1.2 simonb mutex_enter(&wl->wl_mtx); 1788 1.2 simonb wi = wapbl_inodetrk_get(wl, ino); 1789 1.2 simonb if (wi) { 1790 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_INODE, 1791 1.2 simonb ("wapbl_unregister_inode: ino=%"PRId64"\n", ino)); 1792 1.2 simonb KASSERT(wl->wl_inohashcnt > 0); 1793 1.2 simonb wl->wl_inohashcnt--; 1794 1.2 simonb LIST_REMOVE(wi, wi_hash); 1795 1.2 simonb mutex_exit(&wl->wl_mtx); 1796 1.2 simonb 1797 1.2 simonb pool_put(&wapbl_ino_pool, wi); 1798 1.2 simonb } else { 1799 1.2 simonb mutex_exit(&wl->wl_mtx); 1800 1.2 simonb } 1801 1.2 simonb } 1802 1.2 simonb 1803 1.2 simonb /****************************************************************/ 1804 1.2 simonb 1805 1.2 simonb static __inline size_t 1806 1.2 simonb wapbl_transaction_inodes_len(struct wapbl *wl) 1807 1.2 simonb { 1808 1.2 simonb int blocklen = 1<<wl->wl_log_dev_bshift; 1809 1.2 simonb int iph; 1810 1.2 simonb 1811 1.2 simonb /* Calculate number of inodes described in a inodelist header */ 1812 1.2 simonb iph = (blocklen - offsetof(struct wapbl_wc_inodelist, wc_inodes)) / 1813 1.2 simonb sizeof(((struct wapbl_wc_inodelist *)0)->wc_inodes[0]); 1814 1.2 simonb 1815 1.2 simonb KASSERT(iph > 0); 1816 1.2 simonb 1817 1.2 simonb return MAX(1, howmany(wl->wl_inohashcnt, iph))*blocklen; 1818 1.2 simonb } 1819 1.2 simonb 1820 1.2 simonb 1821 1.2 simonb /* Calculate amount of space a transaction will take on disk */ 1822 1.2 simonb static size_t 1823 1.2 simonb wapbl_transaction_len(struct wapbl *wl) 1824 1.2 simonb { 1825 1.2 simonb int blocklen = 1<<wl->wl_log_dev_bshift; 1826 1.2 simonb size_t len; 1827 1.2 simonb int bph; 1828 1.2 simonb 1829 1.2 simonb /* Calculate number of blocks described in a blocklist header */ 1830 1.2 simonb bph = (blocklen - offsetof(struct wapbl_wc_blocklist, wc_blocks)) / 1831 1.2 simonb sizeof(((struct wapbl_wc_blocklist *)0)->wc_blocks[0]); 1832 1.2 simonb 1833 1.2 simonb KASSERT(bph > 0); 1834 1.2 simonb 1835 1.2 simonb len = wl->wl_bcount; 1836 1.2 simonb len += howmany(wl->wl_bufcount, bph)*blocklen; 1837 1.2 simonb len += howmany(wl->wl_dealloccnt, bph)*blocklen; 1838 1.2 simonb len += wapbl_transaction_inodes_len(wl); 1839 1.2 simonb 1840 1.2 simonb return len; 1841 1.2 simonb } 1842 1.2 simonb 1843 1.2 simonb /* 1844 1.2 simonb * Perform commit operation 1845 1.2 simonb * 1846 1.2 simonb * Note that generation number incrementation needs to 1847 1.2 simonb * be protected against racing with other invocations 1848 1.2 simonb * of wapbl_commit. This is ok since this routine 1849 1.2 simonb * is only invoked from wapbl_flush 1850 1.2 simonb */ 1851 1.2 simonb static int 1852 1.2 simonb wapbl_write_commit(struct wapbl *wl, off_t head, off_t tail) 1853 1.2 simonb { 1854 1.2 simonb struct wapbl_wc_header *wc = wl->wl_wc_header; 1855 1.2 simonb struct timespec ts; 1856 1.2 simonb int error; 1857 1.2 simonb int force = 1; 1858 1.2 simonb 1859 1.2 simonb /* XXX Calc checksum here, instead we do this for now */ 1860 1.2 simonb error = VOP_IOCTL(wl->wl_devvp, DIOCCACHESYNC, &force, FWRITE, FSCRED); 1861 1.2 simonb if (error) { 1862 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_ERROR, 1863 1.2 simonb ("wapbl_write_commit: DIOCCACHESYNC on dev 0x%x " 1864 1.2 simonb "returned %d\n", wl->wl_devvp->v_rdev, error)); 1865 1.2 simonb } 1866 1.2 simonb 1867 1.2 simonb wc->wc_head = head; 1868 1.2 simonb wc->wc_tail = tail; 1869 1.2 simonb wc->wc_checksum = 0; 1870 1.2 simonb wc->wc_version = 1; 1871 1.2 simonb getnanotime(&ts); 1872 1.2 simonb wc->wc_time = ts.tv_sec;; 1873 1.2 simonb wc->wc_timensec = ts.tv_nsec; 1874 1.2 simonb 1875 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_WRITE, 1876 1.2 simonb ("wapbl_write_commit: head = %"PRIdMAX "tail = %"PRIdMAX"\n", 1877 1.2 simonb (intmax_t)head, (intmax_t)tail)); 1878 1.2 simonb 1879 1.2 simonb /* 1880 1.2 simonb * XXX if generation will rollover, then first zero 1881 1.2 simonb * over second commit header before trying to write both headers. 1882 1.2 simonb */ 1883 1.2 simonb 1884 1.2 simonb error = wapbl_write(wc, wc->wc_len, wl->wl_devvp, 1885 1.2 simonb wl->wl_logpbn + wc->wc_generation % 2); 1886 1.2 simonb if (error) 1887 1.2 simonb return error; 1888 1.2 simonb 1889 1.2 simonb error = VOP_IOCTL(wl->wl_devvp, DIOCCACHESYNC, &force, FWRITE, FSCRED); 1890 1.2 simonb if (error) { 1891 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_ERROR, 1892 1.2 simonb ("wapbl_write_commit: DIOCCACHESYNC on dev 0x%x " 1893 1.2 simonb "returned %d\n", wl->wl_devvp->v_rdev, error)); 1894 1.2 simonb } 1895 1.2 simonb 1896 1.2 simonb /* 1897 1.2 simonb * If the generation number was zero, write it out a second time. 1898 1.2 simonb * This handles initialization and generation number rollover 1899 1.2 simonb */ 1900 1.2 simonb if (wc->wc_generation++ == 0) { 1901 1.2 simonb error = wapbl_write_commit(wl, head, tail); 1902 1.2 simonb /* 1903 1.2 simonb * This panic should be able to be removed if we do the 1904 1.2 simonb * zero'ing mentioned above, and we are certain to roll 1905 1.2 simonb * back generation number on failure. 1906 1.2 simonb */ 1907 1.2 simonb if (error) 1908 1.2 simonb panic("wapbl_write_commit: error writing duplicate " 1909 1.2 simonb "log header: %d\n", error); 1910 1.2 simonb } 1911 1.2 simonb return 0; 1912 1.2 simonb } 1913 1.2 simonb 1914 1.2 simonb /* Returns new offset value */ 1915 1.2 simonb static int 1916 1.2 simonb wapbl_write_blocks(struct wapbl *wl, off_t *offp) 1917 1.2 simonb { 1918 1.2 simonb struct wapbl_wc_blocklist *wc = 1919 1.2 simonb (struct wapbl_wc_blocklist *)wl->wl_wc_scratch; 1920 1.2 simonb int blocklen = 1<<wl->wl_log_dev_bshift; 1921 1.2 simonb int bph; 1922 1.2 simonb struct buf *bp; 1923 1.2 simonb off_t off = *offp; 1924 1.2 simonb int error; 1925 1.7 joerg size_t padding; 1926 1.2 simonb 1927 1.2 simonb KASSERT(rw_write_held(&wl->wl_rwlock)); 1928 1.2 simonb 1929 1.2 simonb bph = (blocklen - offsetof(struct wapbl_wc_blocklist, wc_blocks)) / 1930 1.2 simonb sizeof(((struct wapbl_wc_blocklist *)0)->wc_blocks[0]); 1931 1.2 simonb 1932 1.2 simonb bp = LIST_FIRST(&wl->wl_bufs); 1933 1.2 simonb 1934 1.2 simonb while (bp) { 1935 1.2 simonb int cnt; 1936 1.2 simonb struct buf *obp = bp; 1937 1.2 simonb 1938 1.2 simonb KASSERT(bp->b_flags & B_LOCKED); 1939 1.2 simonb 1940 1.2 simonb wc->wc_type = WAPBL_WC_BLOCKS; 1941 1.2 simonb wc->wc_len = blocklen; 1942 1.2 simonb wc->wc_blkcount = 0; 1943 1.2 simonb while (bp && (wc->wc_blkcount < bph)) { 1944 1.2 simonb /* 1945 1.2 simonb * Make sure all the physical block numbers are up to 1946 1.2 simonb * date. If this is not always true on a given 1947 1.2 simonb * filesystem, then VOP_BMAP must be called. We 1948 1.2 simonb * could call VOP_BMAP here, or else in the filesystem 1949 1.2 simonb * specific flush callback, although neither of those 1950 1.2 simonb * solutions allow us to take the vnode lock. If a 1951 1.2 simonb * filesystem requires that we must take the vnode lock 1952 1.2 simonb * to call VOP_BMAP, then we can probably do it in 1953 1.2 simonb * bwrite when the vnode lock should already be held 1954 1.2 simonb * by the invoking code. 1955 1.2 simonb */ 1956 1.2 simonb KASSERT((bp->b_vp->v_type == VBLK) || 1957 1.2 simonb (bp->b_blkno != bp->b_lblkno)); 1958 1.2 simonb KASSERT(bp->b_blkno > 0); 1959 1.2 simonb 1960 1.2 simonb wc->wc_blocks[wc->wc_blkcount].wc_daddr = bp->b_blkno; 1961 1.2 simonb wc->wc_blocks[wc->wc_blkcount].wc_dlen = bp->b_bcount; 1962 1.2 simonb wc->wc_len += bp->b_bcount; 1963 1.2 simonb wc->wc_blkcount++; 1964 1.2 simonb bp = LIST_NEXT(bp, b_wapbllist); 1965 1.2 simonb } 1966 1.7 joerg if (wc->wc_len % blocklen != 0) { 1967 1.7 joerg printf("Padding WAPBL record..."); 1968 1.7 joerg padding = blocklen - wc->wc_len % blocklen; 1969 1.7 joerg wc->wc_len += padding; 1970 1.7 joerg } else { 1971 1.7 joerg padding = 0; 1972 1.7 joerg } 1973 1.7 joerg 1974 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_WRITE, 1975 1.7 joerg ("wapbl_write_blocks: len = %u (padding %zu) off = %"PRIdMAX"\n", 1976 1.7 joerg wc->wc_len, padding, (intmax_t)off)); 1977 1.2 simonb 1978 1.2 simonb error = wapbl_circ_write(wl, wc, blocklen, &off); 1979 1.2 simonb if (error) 1980 1.2 simonb return error; 1981 1.2 simonb bp = obp; 1982 1.2 simonb cnt = 0; 1983 1.2 simonb while (bp && (cnt++ < bph)) { 1984 1.2 simonb error = wapbl_circ_write(wl, bp->b_data, 1985 1.2 simonb bp->b_bcount, &off); 1986 1.2 simonb if (error) 1987 1.2 simonb return error; 1988 1.2 simonb bp = LIST_NEXT(bp, b_wapbllist); 1989 1.2 simonb } 1990 1.7 joerg if (padding) { 1991 1.7 joerg void *zero; 1992 1.7 joerg 1993 1.7 joerg zero = wapbl_malloc(padding); 1994 1.7 joerg memset(zero, 0, padding); 1995 1.7 joerg error = wapbl_circ_write(wl, zero, padding, &off); 1996 1.7 joerg wapbl_free(zero); 1997 1.7 joerg if (error) 1998 1.7 joerg return error; 1999 1.7 joerg } 2000 1.2 simonb } 2001 1.2 simonb *offp = off; 2002 1.2 simonb return 0; 2003 1.2 simonb } 2004 1.2 simonb 2005 1.2 simonb static int 2006 1.2 simonb wapbl_write_revocations(struct wapbl *wl, off_t *offp) 2007 1.2 simonb { 2008 1.2 simonb struct wapbl_wc_blocklist *wc = 2009 1.2 simonb (struct wapbl_wc_blocklist *)wl->wl_wc_scratch; 2010 1.2 simonb int i; 2011 1.2 simonb int blocklen = 1<<wl->wl_log_dev_bshift; 2012 1.2 simonb int bph; 2013 1.2 simonb off_t off = *offp; 2014 1.2 simonb int error; 2015 1.2 simonb 2016 1.2 simonb if (wl->wl_dealloccnt == 0) 2017 1.2 simonb return 0; 2018 1.2 simonb 2019 1.2 simonb bph = (blocklen - offsetof(struct wapbl_wc_blocklist, wc_blocks)) / 2020 1.2 simonb sizeof(((struct wapbl_wc_blocklist *)0)->wc_blocks[0]); 2021 1.2 simonb 2022 1.2 simonb i = 0; 2023 1.2 simonb while (i < wl->wl_dealloccnt) { 2024 1.2 simonb wc->wc_type = WAPBL_WC_REVOCATIONS; 2025 1.2 simonb wc->wc_len = blocklen; 2026 1.2 simonb wc->wc_blkcount = 0; 2027 1.2 simonb while ((i < wl->wl_dealloccnt) && (wc->wc_blkcount < bph)) { 2028 1.2 simonb wc->wc_blocks[wc->wc_blkcount].wc_daddr = 2029 1.2 simonb wl->wl_deallocblks[i]; 2030 1.2 simonb wc->wc_blocks[wc->wc_blkcount].wc_dlen = 2031 1.2 simonb wl->wl_dealloclens[i]; 2032 1.2 simonb wc->wc_blkcount++; 2033 1.2 simonb i++; 2034 1.2 simonb } 2035 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_WRITE, 2036 1.2 simonb ("wapbl_write_revocations: len = %u off = %"PRIdMAX"\n", 2037 1.2 simonb wc->wc_len, (intmax_t)off)); 2038 1.2 simonb error = wapbl_circ_write(wl, wc, blocklen, &off); 2039 1.2 simonb if (error) 2040 1.2 simonb return error; 2041 1.2 simonb } 2042 1.2 simonb *offp = off; 2043 1.2 simonb return 0; 2044 1.2 simonb } 2045 1.2 simonb 2046 1.2 simonb static int 2047 1.2 simonb wapbl_write_inodes(struct wapbl *wl, off_t *offp) 2048 1.2 simonb { 2049 1.2 simonb struct wapbl_wc_inodelist *wc = 2050 1.2 simonb (struct wapbl_wc_inodelist *)wl->wl_wc_scratch; 2051 1.2 simonb int i; 2052 1.2 simonb int blocklen = 1<<wl->wl_log_dev_bshift; 2053 1.2 simonb off_t off = *offp; 2054 1.2 simonb int error; 2055 1.2 simonb 2056 1.2 simonb struct wapbl_ino_head *wih; 2057 1.2 simonb struct wapbl_ino *wi; 2058 1.2 simonb int iph; 2059 1.2 simonb 2060 1.2 simonb iph = (blocklen - offsetof(struct wapbl_wc_inodelist, wc_inodes)) / 2061 1.2 simonb sizeof(((struct wapbl_wc_inodelist *)0)->wc_inodes[0]); 2062 1.2 simonb 2063 1.2 simonb i = 0; 2064 1.2 simonb wih = &wl->wl_inohash[0]; 2065 1.2 simonb wi = 0; 2066 1.2 simonb do { 2067 1.2 simonb wc->wc_type = WAPBL_WC_INODES; 2068 1.2 simonb wc->wc_len = blocklen; 2069 1.2 simonb wc->wc_inocnt = 0; 2070 1.2 simonb wc->wc_clear = (i == 0); 2071 1.2 simonb while ((i < wl->wl_inohashcnt) && (wc->wc_inocnt < iph)) { 2072 1.2 simonb while (!wi) { 2073 1.2 simonb KASSERT((wih - &wl->wl_inohash[0]) 2074 1.2 simonb <= wl->wl_inohashmask); 2075 1.2 simonb wi = LIST_FIRST(wih++); 2076 1.2 simonb } 2077 1.2 simonb wc->wc_inodes[wc->wc_inocnt].wc_inumber = wi->wi_ino; 2078 1.2 simonb wc->wc_inodes[wc->wc_inocnt].wc_imode = wi->wi_mode; 2079 1.2 simonb wc->wc_inocnt++; 2080 1.2 simonb i++; 2081 1.2 simonb wi = LIST_NEXT(wi, wi_hash); 2082 1.2 simonb } 2083 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_WRITE, 2084 1.2 simonb ("wapbl_write_inodes: len = %u off = %"PRIdMAX"\n", 2085 1.2 simonb wc->wc_len, (intmax_t)off)); 2086 1.2 simonb error = wapbl_circ_write(wl, wc, blocklen, &off); 2087 1.2 simonb if (error) 2088 1.2 simonb return error; 2089 1.2 simonb } while (i < wl->wl_inohashcnt); 2090 1.2 simonb 2091 1.2 simonb *offp = off; 2092 1.2 simonb return 0; 2093 1.2 simonb } 2094 1.2 simonb 2095 1.2 simonb #endif /* _KERNEL */ 2096 1.2 simonb 2097 1.2 simonb /****************************************************************/ 2098 1.2 simonb 2099 1.2 simonb #ifdef _KERNEL 2100 1.2 simonb static struct pool wapbl_blk_pool; 2101 1.2 simonb static int wapbl_blk_pool_refcount; 2102 1.2 simonb #endif 2103 1.2 simonb struct wapbl_blk { 2104 1.2 simonb LIST_ENTRY(wapbl_blk) wb_hash; 2105 1.2 simonb daddr_t wb_blk; 2106 1.2 simonb off_t wb_off; /* Offset of this block in the log */ 2107 1.2 simonb }; 2108 1.2 simonb #define WAPBL_BLKPOOL_MIN 83 2109 1.2 simonb 2110 1.2 simonb static void 2111 1.2 simonb wapbl_blkhash_init(struct wapbl_replay *wr, u_int size) 2112 1.2 simonb { 2113 1.2 simonb if (size < WAPBL_BLKPOOL_MIN) 2114 1.2 simonb size = WAPBL_BLKPOOL_MIN; 2115 1.2 simonb KASSERT(wr->wr_blkhash == 0); 2116 1.2 simonb #ifdef _KERNEL 2117 1.2 simonb wr->wr_blkhash = hashinit(size, HASH_LIST, true, &wr->wr_blkhashmask); 2118 1.2 simonb if (atomic_inc_uint_nv(&wapbl_blk_pool_refcount) == 1) { 2119 1.2 simonb pool_init(&wapbl_blk_pool, sizeof(struct wapbl_blk), 0, 0, 0, 2120 1.2 simonb "wapblblkpl", &pool_allocator_nointr, IPL_NONE); 2121 1.2 simonb } 2122 1.2 simonb #else /* ! _KERNEL */ 2123 1.2 simonb /* Manually implement hashinit */ 2124 1.2 simonb { 2125 1.2 simonb int i; 2126 1.2 simonb unsigned long hashsize; 2127 1.2 simonb for (hashsize = 1; hashsize < size; hashsize <<= 1) 2128 1.2 simonb continue; 2129 1.2 simonb wr->wr_blkhash = wapbl_malloc(hashsize * sizeof(*wr->wr_blkhash)); 2130 1.2 simonb for (i = 0; i < wr->wr_blkhashmask; i++) 2131 1.2 simonb LIST_INIT(&wr->wr_blkhash[i]); 2132 1.2 simonb wr->wr_blkhashmask = hashsize - 1; 2133 1.2 simonb } 2134 1.2 simonb #endif /* ! _KERNEL */ 2135 1.2 simonb } 2136 1.2 simonb 2137 1.2 simonb static void 2138 1.2 simonb wapbl_blkhash_free(struct wapbl_replay *wr) 2139 1.2 simonb { 2140 1.2 simonb KASSERT(wr->wr_blkhashcnt == 0); 2141 1.2 simonb #ifdef _KERNEL 2142 1.2 simonb hashdone(wr->wr_blkhash, HASH_LIST, wr->wr_blkhashmask); 2143 1.2 simonb if (atomic_dec_uint_nv(&wapbl_blk_pool_refcount) == 0) { 2144 1.2 simonb pool_destroy(&wapbl_blk_pool); 2145 1.2 simonb } 2146 1.2 simonb #else /* ! _KERNEL */ 2147 1.2 simonb wapbl_free(wr->wr_blkhash); 2148 1.2 simonb #endif /* ! _KERNEL */ 2149 1.2 simonb } 2150 1.2 simonb 2151 1.2 simonb static struct wapbl_blk * 2152 1.2 simonb wapbl_blkhash_get(struct wapbl_replay *wr, daddr_t blk) 2153 1.2 simonb { 2154 1.2 simonb struct wapbl_blk_head *wbh; 2155 1.2 simonb struct wapbl_blk *wb; 2156 1.2 simonb wbh = &wr->wr_blkhash[blk & wr->wr_blkhashmask]; 2157 1.2 simonb LIST_FOREACH(wb, wbh, wb_hash) { 2158 1.2 simonb if (blk == wb->wb_blk) 2159 1.2 simonb return wb; 2160 1.2 simonb } 2161 1.2 simonb return 0; 2162 1.2 simonb } 2163 1.2 simonb 2164 1.2 simonb static void 2165 1.2 simonb wapbl_blkhash_ins(struct wapbl_replay *wr, daddr_t blk, off_t off) 2166 1.2 simonb { 2167 1.2 simonb struct wapbl_blk_head *wbh; 2168 1.2 simonb struct wapbl_blk *wb; 2169 1.2 simonb wb = wapbl_blkhash_get(wr, blk); 2170 1.2 simonb if (wb) { 2171 1.2 simonb KASSERT(wb->wb_blk == blk); 2172 1.2 simonb wb->wb_off = off; 2173 1.2 simonb } else { 2174 1.2 simonb #ifdef _KERNEL 2175 1.2 simonb wb = pool_get(&wapbl_blk_pool, PR_WAITOK); 2176 1.2 simonb #else /* ! _KERNEL */ 2177 1.2 simonb wb = wapbl_malloc(sizeof(*wb)); 2178 1.2 simonb #endif /* ! _KERNEL */ 2179 1.2 simonb wb->wb_blk = blk; 2180 1.2 simonb wb->wb_off = off; 2181 1.2 simonb wbh = &wr->wr_blkhash[blk & wr->wr_blkhashmask]; 2182 1.2 simonb LIST_INSERT_HEAD(wbh, wb, wb_hash); 2183 1.2 simonb wr->wr_blkhashcnt++; 2184 1.2 simonb } 2185 1.2 simonb } 2186 1.2 simonb 2187 1.2 simonb static void 2188 1.2 simonb wapbl_blkhash_rem(struct wapbl_replay *wr, daddr_t blk) 2189 1.2 simonb { 2190 1.2 simonb struct wapbl_blk *wb = wapbl_blkhash_get(wr, blk); 2191 1.2 simonb if (wb) { 2192 1.2 simonb KASSERT(wr->wr_blkhashcnt > 0); 2193 1.2 simonb wr->wr_blkhashcnt--; 2194 1.2 simonb LIST_REMOVE(wb, wb_hash); 2195 1.2 simonb #ifdef _KERNEL 2196 1.2 simonb pool_put(&wapbl_blk_pool, wb); 2197 1.2 simonb #else /* ! _KERNEL */ 2198 1.2 simonb wapbl_free(wb); 2199 1.2 simonb #endif /* ! _KERNEL */ 2200 1.2 simonb } 2201 1.2 simonb } 2202 1.2 simonb 2203 1.2 simonb static void 2204 1.2 simonb wapbl_blkhash_clear(struct wapbl_replay *wr) 2205 1.2 simonb { 2206 1.2 simonb int i; 2207 1.2 simonb for (i = 0; i <= wr->wr_blkhashmask; i++) { 2208 1.2 simonb struct wapbl_blk *wb; 2209 1.2 simonb 2210 1.2 simonb while ((wb = LIST_FIRST(&wr->wr_blkhash[i]))) { 2211 1.2 simonb KASSERT(wr->wr_blkhashcnt > 0); 2212 1.2 simonb wr->wr_blkhashcnt--; 2213 1.2 simonb LIST_REMOVE(wb, wb_hash); 2214 1.2 simonb #ifdef _KERNEL 2215 1.2 simonb pool_put(&wapbl_blk_pool, wb); 2216 1.2 simonb #else /* ! _KERNEL */ 2217 1.2 simonb wapbl_free(wb); 2218 1.2 simonb #endif /* ! _KERNEL */ 2219 1.2 simonb } 2220 1.2 simonb } 2221 1.2 simonb KASSERT(wr->wr_blkhashcnt == 0); 2222 1.2 simonb } 2223 1.2 simonb 2224 1.2 simonb /****************************************************************/ 2225 1.2 simonb 2226 1.2 simonb static int 2227 1.2 simonb wapbl_circ_read(struct wapbl_replay *wr, void *data, size_t len, off_t *offp) 2228 1.2 simonb { 2229 1.2 simonb size_t slen; 2230 1.2 simonb struct wapbl_wc_header *wc = &wr->wr_wc_header; 2231 1.2 simonb off_t off = *offp; 2232 1.2 simonb int error; 2233 1.2 simonb 2234 1.2 simonb KASSERT(((len >> wc->wc_log_dev_bshift) << 2235 1.2 simonb wc->wc_log_dev_bshift) == len); 2236 1.2 simonb if (off < wc->wc_circ_off) 2237 1.2 simonb off = wc->wc_circ_off; 2238 1.2 simonb slen = wc->wc_circ_off + wc->wc_circ_size - off; 2239 1.2 simonb if (slen < len) { 2240 1.2 simonb error = wapbl_read(data, slen, wr->wr_devvp, 2241 1.2 simonb wr->wr_logpbn + (off >> wc->wc_log_dev_bshift)); 2242 1.2 simonb if (error) 2243 1.2 simonb return error; 2244 1.2 simonb data = (uint8_t *)data + slen; 2245 1.2 simonb len -= slen; 2246 1.2 simonb off = wc->wc_circ_off; 2247 1.2 simonb } 2248 1.2 simonb error = wapbl_read(data, len, wr->wr_devvp, 2249 1.2 simonb wr->wr_logpbn + (off >> wc->wc_log_dev_bshift)); 2250 1.2 simonb if (error) 2251 1.2 simonb return error; 2252 1.2 simonb off += len; 2253 1.2 simonb if (off >= wc->wc_circ_off + wc->wc_circ_size) 2254 1.2 simonb off = wc->wc_circ_off; 2255 1.2 simonb *offp = off; 2256 1.2 simonb return 0; 2257 1.2 simonb } 2258 1.2 simonb 2259 1.2 simonb static void 2260 1.2 simonb wapbl_circ_advance(struct wapbl_replay *wr, size_t len, off_t *offp) 2261 1.2 simonb { 2262 1.2 simonb size_t slen; 2263 1.2 simonb struct wapbl_wc_header *wc = &wr->wr_wc_header; 2264 1.2 simonb off_t off = *offp; 2265 1.2 simonb 2266 1.2 simonb KASSERT(((len >> wc->wc_log_dev_bshift) << 2267 1.2 simonb wc->wc_log_dev_bshift) == len); 2268 1.2 simonb 2269 1.2 simonb if (off < wc->wc_circ_off) 2270 1.2 simonb off = wc->wc_circ_off; 2271 1.2 simonb slen = wc->wc_circ_off + wc->wc_circ_size - off; 2272 1.2 simonb if (slen < len) { 2273 1.2 simonb len -= slen; 2274 1.2 simonb off = wc->wc_circ_off; 2275 1.2 simonb } 2276 1.2 simonb off += len; 2277 1.2 simonb if (off >= wc->wc_circ_off + wc->wc_circ_size) 2278 1.2 simonb off = wc->wc_circ_off; 2279 1.2 simonb *offp = off; 2280 1.2 simonb } 2281 1.2 simonb 2282 1.2 simonb /****************************************************************/ 2283 1.2 simonb 2284 1.2 simonb int 2285 1.2 simonb wapbl_replay_start(struct wapbl_replay **wrp, struct vnode *vp, 2286 1.2 simonb daddr_t off, size_t count, size_t blksize) 2287 1.2 simonb { 2288 1.2 simonb struct wapbl_replay *wr; 2289 1.2 simonb int error; 2290 1.2 simonb struct vnode *devvp; 2291 1.2 simonb daddr_t logpbn; 2292 1.2 simonb uint8_t *scratch; 2293 1.2 simonb struct wapbl_wc_header *wch; 2294 1.2 simonb struct wapbl_wc_header *wch2; 2295 1.2 simonb /* Use this until we read the actual log header */ 2296 1.2 simonb int log_dev_bshift = DEV_BSHIFT; 2297 1.2 simonb size_t used; 2298 1.2 simonb 2299 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_REPLAY, 2300 1.2 simonb ("wapbl_replay_start: vp=%p off=%"PRId64 " count=%zu blksize=%zu\n", 2301 1.2 simonb vp, off, count, blksize)); 2302 1.2 simonb 2303 1.2 simonb if (off < 0) 2304 1.2 simonb return EINVAL; 2305 1.2 simonb 2306 1.2 simonb if (blksize < DEV_BSIZE) 2307 1.2 simonb return EINVAL; 2308 1.2 simonb if (blksize % DEV_BSIZE) 2309 1.2 simonb return EINVAL; 2310 1.2 simonb 2311 1.2 simonb #ifdef _KERNEL 2312 1.2 simonb #if 0 2313 1.2 simonb /* XXX vp->v_size isn't reliably set for VBLK devices, 2314 1.2 simonb * especially root. However, we might still want to verify 2315 1.2 simonb * that the full load is readable */ 2316 1.2 simonb if ((off + count) * blksize > vp->v_size) 2317 1.2 simonb return EINVAL; 2318 1.2 simonb #endif 2319 1.2 simonb 2320 1.2 simonb if ((error = VOP_BMAP(vp, off, &devvp, &logpbn, 0)) != 0) { 2321 1.2 simonb return error; 2322 1.2 simonb } 2323 1.2 simonb #else /* ! _KERNEL */ 2324 1.2 simonb devvp = vp; 2325 1.2 simonb logpbn = off; 2326 1.2 simonb #endif /* ! _KERNEL */ 2327 1.2 simonb 2328 1.2 simonb scratch = wapbl_malloc(MAXBSIZE); 2329 1.2 simonb 2330 1.2 simonb error = wapbl_read(scratch, 2<<log_dev_bshift, devvp, logpbn); 2331 1.2 simonb if (error) 2332 1.2 simonb goto errout; 2333 1.2 simonb 2334 1.2 simonb wch = (struct wapbl_wc_header *)scratch; 2335 1.2 simonb wch2 = 2336 1.2 simonb (struct wapbl_wc_header *)(scratch + (1<<log_dev_bshift)); 2337 1.2 simonb /* XXX verify checksums and magic numbers */ 2338 1.2 simonb if (wch->wc_type != WAPBL_WC_HEADER) { 2339 1.2 simonb printf("Unrecognized wapbl magic: 0x%08x\n", wch->wc_type); 2340 1.2 simonb error = EFTYPE; 2341 1.2 simonb goto errout; 2342 1.2 simonb } 2343 1.2 simonb 2344 1.2 simonb if (wch2->wc_generation > wch->wc_generation) 2345 1.2 simonb wch = wch2; 2346 1.2 simonb 2347 1.2 simonb wr = wapbl_calloc(1, sizeof(*wr)); 2348 1.2 simonb 2349 1.2 simonb wr->wr_logvp = vp; 2350 1.2 simonb wr->wr_devvp = devvp; 2351 1.2 simonb wr->wr_logpbn = logpbn; 2352 1.2 simonb 2353 1.2 simonb wr->wr_scratch = scratch; 2354 1.2 simonb 2355 1.2 simonb memcpy(&wr->wr_wc_header, wch, sizeof(wr->wr_wc_header)); 2356 1.2 simonb 2357 1.2 simonb used = wapbl_space_used(wch->wc_circ_size, wch->wc_head, wch->wc_tail); 2358 1.2 simonb 2359 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_REPLAY, 2360 1.2 simonb ("wapbl_replay: head=%"PRId64" tail=%"PRId64" off=%"PRId64 2361 1.2 simonb " len=%"PRId64" used=%zu\n", 2362 1.2 simonb wch->wc_head, wch->wc_tail, wch->wc_circ_off, 2363 1.2 simonb wch->wc_circ_size, used)); 2364 1.2 simonb 2365 1.2 simonb wapbl_blkhash_init(wr, (used >> wch->wc_fs_dev_bshift)); 2366 1.2 simonb error = wapbl_replay_prescan(wr); 2367 1.2 simonb if (error) { 2368 1.2 simonb wapbl_replay_stop(wr); 2369 1.2 simonb wapbl_replay_free(wr); 2370 1.2 simonb return error; 2371 1.2 simonb } 2372 1.2 simonb 2373 1.2 simonb error = wapbl_replay_get_inodes(wr); 2374 1.2 simonb if (error) { 2375 1.2 simonb wapbl_replay_stop(wr); 2376 1.2 simonb wapbl_replay_free(wr); 2377 1.2 simonb return error; 2378 1.2 simonb } 2379 1.2 simonb 2380 1.2 simonb *wrp = wr; 2381 1.2 simonb return 0; 2382 1.2 simonb 2383 1.2 simonb errout: 2384 1.2 simonb wapbl_free(scratch); 2385 1.2 simonb return error; 2386 1.2 simonb } 2387 1.2 simonb 2388 1.2 simonb void 2389 1.2 simonb wapbl_replay_stop(struct wapbl_replay *wr) 2390 1.2 simonb { 2391 1.2 simonb 2392 1.4 joerg if (!wapbl_replay_isopen(wr)) 2393 1.4 joerg return; 2394 1.4 joerg 2395 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_REPLAY, ("wapbl_replay_stop called\n")); 2396 1.2 simonb 2397 1.2 simonb wapbl_free(wr->wr_scratch); 2398 1.2 simonb wr->wr_scratch = 0; 2399 1.2 simonb 2400 1.2 simonb wr->wr_logvp = 0; 2401 1.2 simonb 2402 1.2 simonb wapbl_blkhash_clear(wr); 2403 1.2 simonb wapbl_blkhash_free(wr); 2404 1.2 simonb } 2405 1.2 simonb 2406 1.2 simonb void 2407 1.2 simonb wapbl_replay_free(struct wapbl_replay *wr) 2408 1.2 simonb { 2409 1.2 simonb 2410 1.2 simonb KDASSERT(!wapbl_replay_isopen(wr)); 2411 1.2 simonb 2412 1.2 simonb if (wr->wr_inodes) 2413 1.2 simonb wapbl_free(wr->wr_inodes); 2414 1.2 simonb wapbl_free(wr); 2415 1.2 simonb } 2416 1.2 simonb 2417 1.4 joerg #ifdef _KERNEL 2418 1.2 simonb int 2419 1.2 simonb wapbl_replay_isopen1(struct wapbl_replay *wr) 2420 1.2 simonb { 2421 1.2 simonb 2422 1.2 simonb return wapbl_replay_isopen(wr); 2423 1.2 simonb } 2424 1.4 joerg #endif 2425 1.2 simonb 2426 1.2 simonb static int 2427 1.2 simonb wapbl_replay_prescan(struct wapbl_replay *wr) 2428 1.2 simonb { 2429 1.2 simonb off_t off; 2430 1.2 simonb struct wapbl_wc_header *wch = &wr->wr_wc_header; 2431 1.2 simonb int error; 2432 1.2 simonb 2433 1.2 simonb int logblklen = 1<<wch->wc_log_dev_bshift; 2434 1.2 simonb int fsblklen = 1<<wch->wc_fs_dev_bshift; 2435 1.2 simonb 2436 1.2 simonb wapbl_blkhash_clear(wr); 2437 1.2 simonb 2438 1.2 simonb off = wch->wc_tail; 2439 1.2 simonb while (off != wch->wc_head) { 2440 1.2 simonb struct wapbl_wc_null *wcn; 2441 1.2 simonb off_t saveoff = off; 2442 1.2 simonb error = wapbl_circ_read(wr, wr->wr_scratch, logblklen, &off); 2443 1.2 simonb if (error) 2444 1.2 simonb goto errout; 2445 1.2 simonb wcn = (struct wapbl_wc_null *)wr->wr_scratch; 2446 1.2 simonb switch (wcn->wc_type) { 2447 1.2 simonb case WAPBL_WC_BLOCKS: 2448 1.2 simonb { 2449 1.2 simonb struct wapbl_wc_blocklist *wc = 2450 1.2 simonb (struct wapbl_wc_blocklist *)wr->wr_scratch; 2451 1.2 simonb int i; 2452 1.2 simonb for (i = 0; i < wc->wc_blkcount; i++) { 2453 1.2 simonb int j, n; 2454 1.2 simonb /* 2455 1.2 simonb * Enter each physical block into the 2456 1.2 simonb * hashtable independently 2457 1.2 simonb */ 2458 1.2 simonb n = wc->wc_blocks[i].wc_dlen >> 2459 1.2 simonb wch->wc_fs_dev_bshift; 2460 1.2 simonb for (j = 0; j < n; j++) { 2461 1.2 simonb wapbl_blkhash_ins(wr, 2462 1.2 simonb wc->wc_blocks[i].wc_daddr + j, 2463 1.2 simonb off); 2464 1.2 simonb wapbl_circ_advance(wr, 2465 1.2 simonb fsblklen, &off); 2466 1.2 simonb } 2467 1.2 simonb } 2468 1.2 simonb } 2469 1.2 simonb break; 2470 1.2 simonb 2471 1.2 simonb case WAPBL_WC_REVOCATIONS: 2472 1.2 simonb { 2473 1.2 simonb struct wapbl_wc_blocklist *wc = 2474 1.2 simonb (struct wapbl_wc_blocklist *)wr->wr_scratch; 2475 1.2 simonb int i; 2476 1.2 simonb for (i = 0; i < wc->wc_blkcount; i++) { 2477 1.2 simonb int j, n; 2478 1.2 simonb /* 2479 1.2 simonb * Remove any blocks found from the 2480 1.2 simonb * hashtable 2481 1.2 simonb */ 2482 1.2 simonb n = wc->wc_blocks[i].wc_dlen >> 2483 1.2 simonb wch->wc_fs_dev_bshift; 2484 1.2 simonb for (j = 0; j < n; j++) { 2485 1.2 simonb wapbl_blkhash_rem(wr, 2486 1.2 simonb wc->wc_blocks[i].wc_daddr + j); 2487 1.2 simonb } 2488 1.2 simonb } 2489 1.2 simonb } 2490 1.2 simonb break; 2491 1.2 simonb 2492 1.2 simonb case WAPBL_WC_INODES: 2493 1.2 simonb { 2494 1.2 simonb struct wapbl_wc_inodelist *wc = 2495 1.2 simonb (struct wapbl_wc_inodelist *)wr->wr_scratch; 2496 1.2 simonb /* 2497 1.2 simonb * Keep track of where we found this so we 2498 1.2 simonb * can use it later 2499 1.2 simonb */ 2500 1.2 simonb if (wc->wc_clear) { 2501 1.2 simonb wr->wr_inodestail = saveoff; 2502 1.2 simonb wr->wr_inodescnt = 0; 2503 1.2 simonb } 2504 1.2 simonb if (wr->wr_inodestail) 2505 1.2 simonb wr->wr_inodeshead = off; 2506 1.2 simonb wr->wr_inodescnt += wc->wc_inocnt; 2507 1.2 simonb } 2508 1.2 simonb break; 2509 1.2 simonb default: 2510 1.2 simonb printf("Unrecognized wapbl type: 0x%08x\n", 2511 1.2 simonb wcn->wc_type); 2512 1.2 simonb error = EFTYPE; 2513 1.2 simonb goto errout; 2514 1.2 simonb } 2515 1.2 simonb wapbl_circ_advance(wr, wcn->wc_len, &saveoff); 2516 1.2 simonb if (off != saveoff) { 2517 1.2 simonb printf("wapbl_replay: corrupted records\n"); 2518 1.2 simonb error = EFTYPE; 2519 1.2 simonb goto errout; 2520 1.2 simonb } 2521 1.2 simonb } 2522 1.2 simonb return 0; 2523 1.2 simonb 2524 1.2 simonb errout: 2525 1.2 simonb wapbl_blkhash_clear(wr); 2526 1.2 simonb return error; 2527 1.2 simonb } 2528 1.2 simonb 2529 1.2 simonb static int 2530 1.2 simonb wapbl_replay_get_inodes(struct wapbl_replay *wr) 2531 1.2 simonb { 2532 1.2 simonb off_t off; 2533 1.2 simonb struct wapbl_wc_header *wch = &wr->wr_wc_header; 2534 1.2 simonb int logblklen = 1<<wch->wc_log_dev_bshift; 2535 1.2 simonb int cnt= 0; 2536 1.2 simonb 2537 1.2 simonb KDASSERT(wapbl_replay_isopen(wr)); 2538 1.2 simonb 2539 1.2 simonb if (wr->wr_inodescnt == 0) 2540 1.2 simonb return 0; 2541 1.2 simonb 2542 1.2 simonb KASSERT(!wr->wr_inodes); 2543 1.2 simonb 2544 1.2 simonb wr->wr_inodes = wapbl_malloc(wr->wr_inodescnt*sizeof(wr->wr_inodes[0])); 2545 1.2 simonb 2546 1.2 simonb off = wr->wr_inodestail; 2547 1.2 simonb 2548 1.2 simonb while (off != wr->wr_inodeshead) { 2549 1.2 simonb struct wapbl_wc_null *wcn; 2550 1.2 simonb int error; 2551 1.2 simonb off_t saveoff = off; 2552 1.2 simonb error = wapbl_circ_read(wr, wr->wr_scratch, logblklen, &off); 2553 1.2 simonb if (error) { 2554 1.2 simonb wapbl_free(wr->wr_inodes); 2555 1.2 simonb wr->wr_inodes = 0; 2556 1.2 simonb return error; 2557 1.2 simonb } 2558 1.2 simonb wcn = (struct wapbl_wc_null *)wr->wr_scratch; 2559 1.2 simonb switch (wcn->wc_type) { 2560 1.2 simonb case WAPBL_WC_BLOCKS: 2561 1.2 simonb case WAPBL_WC_REVOCATIONS: 2562 1.2 simonb break; 2563 1.2 simonb case WAPBL_WC_INODES: 2564 1.2 simonb { 2565 1.2 simonb struct wapbl_wc_inodelist *wc = 2566 1.2 simonb (struct wapbl_wc_inodelist *)wr->wr_scratch; 2567 1.2 simonb /* 2568 1.2 simonb * Keep track of where we found this so we 2569 1.2 simonb * can use it later 2570 1.2 simonb */ 2571 1.2 simonb if (wc->wc_clear) { 2572 1.2 simonb cnt = 0; 2573 1.2 simonb } 2574 1.2 simonb /* This memcpy assumes that wr_inodes is 2575 1.2 simonb * laid out the same as wc_inodes. */ 2576 1.2 simonb memcpy(&wr->wr_inodes[cnt], wc->wc_inodes, 2577 1.2 simonb wc->wc_inocnt*sizeof(wc->wc_inodes[0])); 2578 1.2 simonb cnt += wc->wc_inocnt; 2579 1.2 simonb } 2580 1.2 simonb break; 2581 1.2 simonb default: 2582 1.2 simonb KASSERT(0); 2583 1.2 simonb } 2584 1.2 simonb off = saveoff; 2585 1.2 simonb wapbl_circ_advance(wr, wcn->wc_len, &off); 2586 1.2 simonb } 2587 1.2 simonb KASSERT(cnt == wr->wr_inodescnt); 2588 1.2 simonb return 0; 2589 1.2 simonb } 2590 1.2 simonb 2591 1.2 simonb #ifdef DEBUG 2592 1.2 simonb int 2593 1.2 simonb wapbl_replay_verify(struct wapbl_replay *wr, struct vnode *fsdevvp) 2594 1.2 simonb { 2595 1.2 simonb off_t off; 2596 1.2 simonb struct wapbl_wc_header *wch = &wr->wr_wc_header; 2597 1.2 simonb int mismatchcnt = 0; 2598 1.2 simonb int logblklen = 1<<wch->wc_log_dev_bshift; 2599 1.2 simonb int fsblklen = 1<<wch->wc_fs_dev_bshift; 2600 1.2 simonb void *scratch1 = wapbl_malloc(MAXBSIZE); 2601 1.2 simonb void *scratch2 = wapbl_malloc(MAXBSIZE); 2602 1.2 simonb int error = 0; 2603 1.2 simonb 2604 1.2 simonb KDASSERT(wapbl_replay_isopen(wr)); 2605 1.2 simonb 2606 1.2 simonb off = wch->wc_tail; 2607 1.2 simonb while (off != wch->wc_head) { 2608 1.2 simonb struct wapbl_wc_null *wcn; 2609 1.2 simonb #ifdef DEBUG 2610 1.2 simonb off_t saveoff = off; 2611 1.2 simonb #endif 2612 1.2 simonb error = wapbl_circ_read(wr, wr->wr_scratch, logblklen, &off); 2613 1.2 simonb if (error) 2614 1.2 simonb goto out; 2615 1.2 simonb wcn = (struct wapbl_wc_null *)wr->wr_scratch; 2616 1.2 simonb switch (wcn->wc_type) { 2617 1.2 simonb case WAPBL_WC_BLOCKS: 2618 1.2 simonb { 2619 1.2 simonb struct wapbl_wc_blocklist *wc = 2620 1.2 simonb (struct wapbl_wc_blocklist *)wr->wr_scratch; 2621 1.2 simonb int i; 2622 1.2 simonb for (i = 0; i < wc->wc_blkcount; i++) { 2623 1.2 simonb int foundcnt = 0; 2624 1.2 simonb int dirtycnt = 0; 2625 1.2 simonb int j, n; 2626 1.2 simonb /* 2627 1.2 simonb * Check each physical block into the 2628 1.2 simonb * hashtable independently 2629 1.2 simonb */ 2630 1.2 simonb n = wc->wc_blocks[i].wc_dlen >> 2631 1.2 simonb wch->wc_fs_dev_bshift; 2632 1.2 simonb for (j = 0; j < n; j++) { 2633 1.2 simonb struct wapbl_blk *wb = 2634 1.2 simonb wapbl_blkhash_get(wr, 2635 1.2 simonb wc->wc_blocks[i].wc_daddr + j); 2636 1.2 simonb if (wb && (wb->wb_off == off)) { 2637 1.2 simonb foundcnt++; 2638 1.2 simonb error = 2639 1.2 simonb wapbl_circ_read(wr, 2640 1.2 simonb scratch1, fsblklen, 2641 1.2 simonb &off); 2642 1.2 simonb if (error) 2643 1.2 simonb goto out; 2644 1.2 simonb error = 2645 1.2 simonb wapbl_read(scratch2, 2646 1.2 simonb fsblklen, fsdevvp, 2647 1.2 simonb wb->wb_blk); 2648 1.2 simonb if (error) 2649 1.2 simonb goto out; 2650 1.2 simonb if (memcmp(scratch1, 2651 1.2 simonb scratch2, 2652 1.2 simonb fsblklen)) { 2653 1.2 simonb printf( 2654 1.2 simonb "wapbl_verify: mismatch block %"PRId64" at off %"PRIdMAX"\n", 2655 1.2 simonb wb->wb_blk, (intmax_t)off); 2656 1.2 simonb dirtycnt++; 2657 1.2 simonb mismatchcnt++; 2658 1.2 simonb } 2659 1.2 simonb } else { 2660 1.2 simonb wapbl_circ_advance(wr, 2661 1.2 simonb fsblklen, &off); 2662 1.2 simonb } 2663 1.2 simonb } 2664 1.2 simonb #if 0 2665 1.2 simonb /* 2666 1.2 simonb * If all of the blocks in an entry 2667 1.2 simonb * are clean, then remove all of its 2668 1.2 simonb * blocks from the hashtable since they 2669 1.2 simonb * never will need replay. 2670 1.2 simonb */ 2671 1.2 simonb if ((foundcnt != 0) && 2672 1.2 simonb (dirtycnt == 0)) { 2673 1.2 simonb off = saveoff; 2674 1.2 simonb wapbl_circ_advance(wr, 2675 1.2 simonb logblklen, &off); 2676 1.2 simonb for (j = 0; j < n; j++) { 2677 1.2 simonb struct wapbl_blk *wb = 2678 1.2 simonb wapbl_blkhash_get(wr, 2679 1.2 simonb wc->wc_blocks[i].wc_daddr + j); 2680 1.2 simonb if (wb && 2681 1.2 simonb (wb->wb_off == off)) { 2682 1.2 simonb wapbl_blkhash_rem(wr, wb->wb_blk); 2683 1.2 simonb } 2684 1.2 simonb wapbl_circ_advance(wr, 2685 1.2 simonb fsblklen, &off); 2686 1.2 simonb } 2687 1.2 simonb } 2688 1.2 simonb #endif 2689 1.2 simonb } 2690 1.2 simonb } 2691 1.2 simonb break; 2692 1.2 simonb case WAPBL_WC_REVOCATIONS: 2693 1.2 simonb case WAPBL_WC_INODES: 2694 1.2 simonb break; 2695 1.2 simonb default: 2696 1.2 simonb KASSERT(0); 2697 1.2 simonb } 2698 1.2 simonb #ifdef DEBUG 2699 1.2 simonb wapbl_circ_advance(wr, wcn->wc_len, &saveoff); 2700 1.2 simonb KASSERT(off == saveoff); 2701 1.2 simonb #endif 2702 1.2 simonb } 2703 1.2 simonb out: 2704 1.2 simonb wapbl_free(scratch1); 2705 1.2 simonb wapbl_free(scratch2); 2706 1.2 simonb if (!error && mismatchcnt) 2707 1.2 simonb error = EFTYPE; 2708 1.2 simonb return error; 2709 1.2 simonb } 2710 1.2 simonb #endif 2711 1.2 simonb 2712 1.2 simonb int 2713 1.2 simonb wapbl_replay_write(struct wapbl_replay *wr, struct vnode *fsdevvp) 2714 1.2 simonb { 2715 1.2 simonb off_t off; 2716 1.2 simonb struct wapbl_wc_header *wch = &wr->wr_wc_header; 2717 1.2 simonb int logblklen = 1<<wch->wc_log_dev_bshift; 2718 1.2 simonb int fsblklen = 1<<wch->wc_fs_dev_bshift; 2719 1.2 simonb void *scratch1 = wapbl_malloc(MAXBSIZE); 2720 1.2 simonb int error = 0; 2721 1.2 simonb 2722 1.2 simonb KDASSERT(wapbl_replay_isopen(wr)); 2723 1.2 simonb 2724 1.2 simonb /* 2725 1.2 simonb * This parses the journal for replay, although it could 2726 1.2 simonb * just as easily walk the hashtable instead. 2727 1.2 simonb */ 2728 1.2 simonb 2729 1.2 simonb off = wch->wc_tail; 2730 1.2 simonb while (off != wch->wc_head) { 2731 1.2 simonb struct wapbl_wc_null *wcn; 2732 1.2 simonb #ifdef DEBUG 2733 1.2 simonb off_t saveoff = off; 2734 1.2 simonb #endif 2735 1.2 simonb error = wapbl_circ_read(wr, wr->wr_scratch, logblklen, &off); 2736 1.2 simonb if (error) 2737 1.2 simonb goto out; 2738 1.2 simonb wcn = (struct wapbl_wc_null *)wr->wr_scratch; 2739 1.2 simonb switch (wcn->wc_type) { 2740 1.2 simonb case WAPBL_WC_BLOCKS: 2741 1.2 simonb { 2742 1.2 simonb struct wapbl_wc_blocklist *wc = 2743 1.2 simonb (struct wapbl_wc_blocklist *)wr->wr_scratch; 2744 1.2 simonb int i; 2745 1.2 simonb for (i = 0; i < wc->wc_blkcount; i++) { 2746 1.2 simonb int j, n; 2747 1.2 simonb /* 2748 1.2 simonb * Check each physical block against 2749 1.2 simonb * the hashtable independently 2750 1.2 simonb */ 2751 1.2 simonb n = wc->wc_blocks[i].wc_dlen >> 2752 1.2 simonb wch->wc_fs_dev_bshift; 2753 1.2 simonb for (j = 0; j < n; j++) { 2754 1.2 simonb struct wapbl_blk *wb = 2755 1.2 simonb wapbl_blkhash_get(wr, 2756 1.2 simonb wc->wc_blocks[i].wc_daddr + j); 2757 1.2 simonb if (wb && (wb->wb_off == off)) { 2758 1.2 simonb error = wapbl_circ_read( 2759 1.2 simonb wr, scratch1, 2760 1.2 simonb fsblklen, &off); 2761 1.2 simonb if (error) 2762 1.2 simonb goto out; 2763 1.2 simonb error = 2764 1.2 simonb wapbl_write(scratch1, 2765 1.2 simonb fsblklen, fsdevvp, 2766 1.2 simonb wb->wb_blk); 2767 1.2 simonb if (error) 2768 1.2 simonb goto out; 2769 1.2 simonb } else { 2770 1.2 simonb wapbl_circ_advance(wr, 2771 1.2 simonb fsblklen, &off); 2772 1.2 simonb } 2773 1.2 simonb } 2774 1.2 simonb } 2775 1.2 simonb } 2776 1.2 simonb break; 2777 1.2 simonb case WAPBL_WC_REVOCATIONS: 2778 1.2 simonb case WAPBL_WC_INODES: 2779 1.2 simonb break; 2780 1.2 simonb default: 2781 1.2 simonb KASSERT(0); 2782 1.2 simonb } 2783 1.2 simonb #ifdef DEBUG 2784 1.2 simonb wapbl_circ_advance(wr, wcn->wc_len, &saveoff); 2785 1.2 simonb KASSERT(off == saveoff); 2786 1.2 simonb #endif 2787 1.2 simonb } 2788 1.2 simonb out: 2789 1.2 simonb wapbl_free(scratch1); 2790 1.2 simonb return error; 2791 1.2 simonb } 2792 1.2 simonb 2793 1.2 simonb int 2794 1.6 joerg wapbl_replay_can_read(struct wapbl_replay *wr, daddr_t blk, long len) 2795 1.6 joerg { 2796 1.6 joerg struct wapbl_wc_header *wch = &wr->wr_wc_header; 2797 1.6 joerg int fsblklen = 1<<wch->wc_fs_dev_bshift; 2798 1.6 joerg 2799 1.6 joerg KDASSERT(wapbl_replay_isopen(wr)); 2800 1.6 joerg KASSERT((len % fsblklen) == 0); 2801 1.6 joerg 2802 1.6 joerg while (len != 0) { 2803 1.6 joerg struct wapbl_blk *wb = wapbl_blkhash_get(wr, blk); 2804 1.6 joerg if (wb) 2805 1.6 joerg return 1; 2806 1.6 joerg len -= fsblklen; 2807 1.6 joerg } 2808 1.6 joerg return 0; 2809 1.6 joerg } 2810 1.6 joerg 2811 1.6 joerg int 2812 1.2 simonb wapbl_replay_read(struct wapbl_replay *wr, void *data, daddr_t blk, long len) 2813 1.2 simonb { 2814 1.2 simonb struct wapbl_wc_header *wch = &wr->wr_wc_header; 2815 1.2 simonb int fsblklen = 1<<wch->wc_fs_dev_bshift; 2816 1.2 simonb 2817 1.2 simonb KDASSERT(wapbl_replay_isopen(wr)); 2818 1.2 simonb 2819 1.2 simonb KASSERT((len % fsblklen) == 0); 2820 1.2 simonb 2821 1.2 simonb while (len != 0) { 2822 1.2 simonb struct wapbl_blk *wb = wapbl_blkhash_get(wr, blk); 2823 1.2 simonb if (wb) { 2824 1.2 simonb off_t off = wb->wb_off; 2825 1.2 simonb int error; 2826 1.2 simonb error = wapbl_circ_read(wr, data, fsblklen, &off); 2827 1.2 simonb if (error) 2828 1.2 simonb return error; 2829 1.2 simonb } 2830 1.2 simonb data = (uint8_t *)data + fsblklen; 2831 1.2 simonb len -= fsblklen; 2832 1.2 simonb blk++; 2833 1.2 simonb } 2834 1.2 simonb return 0; 2835 1.2 simonb } 2836
Indexes created Thu Sep 25 16:09:42 GMT 2025