vfs_wapbl.c revision 1.3.8.1
1 1.3.8.1 snj /* $NetBSD: vfs_wapbl.c,v 1.3.8.1 2009/02/24 04:13:35 snj Exp $ */ 2 1.2 simonb 3 1.2 simonb /*- 4 1.3.8.1 snj * Copyright (c) 2003, 2008, 2009 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 1.2 simonb #include <sys/cdefs.h> 36 1.3.8.1 snj __KERNEL_RCSID(0, "$NetBSD: vfs_wapbl.c,v 1.3.8.1 2009/02/24 04:13:35 snj Exp $"); 37 1.2 simonb 38 1.2 simonb #include <sys/param.h> 39 1.2 simonb 40 1.2 simonb #ifdef _KERNEL 41 1.2 simonb #include <sys/param.h> 42 1.2 simonb #include <sys/namei.h> 43 1.2 simonb #include <sys/proc.h> 44 1.2 simonb #include <sys/uio.h> 45 1.2 simonb #include <sys/vnode.h> 46 1.2 simonb #include <sys/file.h> 47 1.2 simonb #include <sys/malloc.h> 48 1.2 simonb #include <sys/resourcevar.h> 49 1.2 simonb #include <sys/conf.h> 50 1.2 simonb #include <sys/mount.h> 51 1.2 simonb #include <sys/kernel.h> 52 1.2 simonb #include <sys/kauth.h> 53 1.2 simonb #include <sys/mutex.h> 54 1.2 simonb #include <sys/atomic.h> 55 1.2 simonb #include <sys/wapbl.h> 56 1.2 simonb 57 1.2 simonb #if WAPBL_UVM_ALLOC 58 1.2 simonb #include <uvm/uvm.h> 59 1.2 simonb #endif 60 1.2 simonb 61 1.2 simonb #include <miscfs/specfs/specdev.h> 62 1.2 simonb 63 1.2 simonb MALLOC_JUSTDEFINE(M_WAPBL, "wapbl", "write-ahead physical block logging"); 64 1.2 simonb #define wapbl_malloc(s) malloc((s), M_WAPBL, M_WAITOK) 65 1.2 simonb #define wapbl_free(a) free((a), M_WAPBL) 66 1.2 simonb #define wapbl_calloc(n, s) malloc((n)*(s), M_WAPBL, M_WAITOK | M_ZERO) 67 1.2 simonb 68 1.2 simonb #else /* !_KERNEL */ 69 1.2 simonb #include <assert.h> 70 1.2 simonb #include <errno.h> 71 1.2 simonb #include <stdio.h> 72 1.2 simonb #include <stdbool.h> 73 1.2 simonb #include <stdlib.h> 74 1.2 simonb #include <string.h> 75 1.2 simonb 76 1.2 simonb #include <sys/time.h> 77 1.2 simonb #include <sys/wapbl.h> 78 1.2 simonb 79 1.2 simonb #define KDASSERT(x) assert(x) 80 1.2 simonb #define KASSERT(x) assert(x) 81 1.2 simonb #define wapbl_malloc(s) malloc(s) 82 1.2 simonb #define wapbl_free(a) free(a) 83 1.2 simonb #define wapbl_calloc(n, s) calloc((n), (s)) 84 1.2 simonb 85 1.2 simonb #endif /* !_KERNEL */ 86 1.2 simonb 87 1.2 simonb /* 88 1.2 simonb * INTERNAL DATA STRUCTURES 89 1.2 simonb */ 90 1.2 simonb 91 1.2 simonb /* 92 1.2 simonb * This structure holds per-mount log information. 93 1.2 simonb * 94 1.2 simonb * Legend: a = atomic access only 95 1.2 simonb * r = read-only after init 96 1.2 simonb * l = rwlock held 97 1.2 simonb * m = mutex held 98 1.2 simonb * u = unlocked access ok 99 1.2 simonb * b = bufcache_lock held 100 1.2 simonb */ 101 1.2 simonb struct wapbl { 102 1.2 simonb struct vnode *wl_logvp; /* r: log here */ 103 1.2 simonb struct vnode *wl_devvp; /* r: log on this device */ 104 1.2 simonb struct mount *wl_mount; /* r: mountpoint wl is associated with */ 105 1.2 simonb daddr_t wl_logpbn; /* r: Physical block number of start of log */ 106 1.2 simonb int wl_log_dev_bshift; /* r: logarithm of device block size of log 107 1.2 simonb device */ 108 1.2 simonb int wl_fs_dev_bshift; /* r: logarithm of device block size of 109 1.2 simonb filesystem device */ 110 1.2 simonb 111 1.3 yamt unsigned wl_lock_count; /* m: Count of transactions in progress */ 112 1.2 simonb 113 1.2 simonb size_t wl_circ_size; /* r: Number of bytes in buffer of log */ 114 1.2 simonb size_t wl_circ_off; /* r: Number of bytes reserved at start */ 115 1.2 simonb 116 1.2 simonb size_t wl_bufcount_max; /* r: Number of buffers reserved for log */ 117 1.2 simonb size_t wl_bufbytes_max; /* r: Number of buf bytes reserved for log */ 118 1.2 simonb 119 1.2 simonb off_t wl_head; /* l: Byte offset of log head */ 120 1.2 simonb off_t wl_tail; /* l: Byte offset of log tail */ 121 1.2 simonb /* 122 1.2 simonb * head == tail == 0 means log is empty 123 1.2 simonb * head == tail != 0 means log is full 124 1.2 simonb * see assertions in wapbl_advance() for other boundary conditions. 125 1.2 simonb * only truncate moves the tail, except when flush sets it to 126 1.2 simonb * wl_header_size only flush moves the head, except when truncate 127 1.2 simonb * sets it to 0. 128 1.2 simonb */ 129 1.2 simonb 130 1.2 simonb struct wapbl_wc_header *wl_wc_header; /* l */ 131 1.2 simonb void *wl_wc_scratch; /* l: scratch space (XXX: por que?!?) */ 132 1.2 simonb 133 1.2 simonb kmutex_t wl_mtx; /* u: short-term lock */ 134 1.2 simonb krwlock_t wl_rwlock; /* u: File system transaction lock */ 135 1.2 simonb 136 1.2 simonb /* 137 1.2 simonb * Must be held while accessing 138 1.2 simonb * wl_count or wl_bufs or head or tail 139 1.2 simonb */ 140 1.2 simonb 141 1.2 simonb /* 142 1.2 simonb * Callback called from within the flush routine to flush any extra 143 1.2 simonb * bits. Note that flush may be skipped without calling this if 144 1.2 simonb * there are no outstanding buffers in the transaction. 145 1.2 simonb */ 146 1.2 simonb wapbl_flush_fn_t wl_flush; /* r */ 147 1.2 simonb wapbl_flush_fn_t wl_flush_abort;/* r */ 148 1.2 simonb 149 1.2 simonb size_t wl_bufbytes; /* m: Byte count of pages in wl_bufs */ 150 1.2 simonb size_t wl_bufcount; /* m: Count of buffers in wl_bufs */ 151 1.2 simonb size_t wl_bcount; /* m: Total bcount of wl_bufs */ 152 1.2 simonb 153 1.2 simonb LIST_HEAD(, buf) wl_bufs; /* m: Buffers in current transaction */ 154 1.2 simonb 155 1.2 simonb kcondvar_t wl_reclaimable_cv; /* m (obviously) */ 156 1.2 simonb size_t wl_reclaimable_bytes; /* m: Amount of space available for 157 1.2 simonb reclamation by truncate */ 158 1.2 simonb int wl_error_count; /* m: # of wl_entries with errors */ 159 1.2 simonb size_t wl_reserved_bytes; /* never truncate log smaller than this */ 160 1.2 simonb 161 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 162 1.2 simonb size_t wl_unsynced_bufbytes; /* Byte count of unsynced buffers */ 163 1.2 simonb #endif 164 1.2 simonb 165 1.2 simonb daddr_t *wl_deallocblks;/* l: address of block */ 166 1.2 simonb int *wl_dealloclens; /* l: size of block (fragments, kom ihg) */ 167 1.2 simonb int wl_dealloccnt; /* l: total count */ 168 1.2 simonb int wl_dealloclim; /* l: max count */ 169 1.2 simonb 170 1.2 simonb /* hashtable of inode numbers for allocated but unlinked inodes */ 171 1.2 simonb /* synch ??? */ 172 1.2 simonb LIST_HEAD(wapbl_ino_head, wapbl_ino) *wl_inohash; 173 1.2 simonb u_long wl_inohashmask; 174 1.2 simonb int wl_inohashcnt; 175 1.2 simonb 176 1.2 simonb SIMPLEQ_HEAD(, wapbl_entry) wl_entries; /* On disk transaction 177 1.2 simonb accounting */ 178 1.2 simonb }; 179 1.2 simonb 180 1.2 simonb #ifdef WAPBL_DEBUG_PRINT 181 1.2 simonb int wapbl_debug_print = WAPBL_DEBUG_PRINT; 182 1.2 simonb #endif 183 1.2 simonb 184 1.2 simonb /****************************************************************/ 185 1.2 simonb #ifdef _KERNEL 186 1.2 simonb 187 1.2 simonb #ifdef WAPBL_DEBUG 188 1.2 simonb struct wapbl *wapbl_debug_wl; 189 1.2 simonb #endif 190 1.2 simonb 191 1.2 simonb static int wapbl_write_commit(struct wapbl *wl, off_t head, off_t tail); 192 1.2 simonb static int wapbl_write_blocks(struct wapbl *wl, off_t *offp); 193 1.2 simonb static int wapbl_write_revocations(struct wapbl *wl, off_t *offp); 194 1.2 simonb static int wapbl_write_inodes(struct wapbl *wl, off_t *offp); 195 1.2 simonb #endif /* _KERNEL */ 196 1.2 simonb 197 1.2 simonb static int wapbl_replay_prescan(struct wapbl_replay *wr); 198 1.2 simonb static int wapbl_replay_get_inodes(struct wapbl_replay *wr); 199 1.2 simonb 200 1.2 simonb static __inline size_t wapbl_space_free(size_t avail, off_t head, 201 1.2 simonb off_t tail); 202 1.2 simonb static __inline size_t wapbl_space_used(size_t avail, off_t head, 203 1.2 simonb off_t tail); 204 1.2 simonb 205 1.2 simonb #ifdef _KERNEL 206 1.2 simonb 207 1.2 simonb #define WAPBL_INODETRK_SIZE 83 208 1.2 simonb static int wapbl_ino_pool_refcount; 209 1.2 simonb static struct pool wapbl_ino_pool; 210 1.2 simonb struct wapbl_ino { 211 1.2 simonb LIST_ENTRY(wapbl_ino) wi_hash; 212 1.2 simonb ino_t wi_ino; 213 1.2 simonb mode_t wi_mode; 214 1.2 simonb }; 215 1.2 simonb 216 1.2 simonb static void wapbl_inodetrk_init(struct wapbl *wl, u_int size); 217 1.2 simonb static void wapbl_inodetrk_free(struct wapbl *wl); 218 1.2 simonb static struct wapbl_ino *wapbl_inodetrk_get(struct wapbl *wl, ino_t ino); 219 1.2 simonb 220 1.2 simonb static size_t wapbl_transaction_len(struct wapbl *wl); 221 1.2 simonb static __inline size_t wapbl_transaction_inodes_len(struct wapbl *wl); 222 1.2 simonb 223 1.2 simonb /* 224 1.2 simonb * This is useful for debugging. If set, the log will 225 1.2 simonb * only be truncated when necessary. 226 1.2 simonb */ 227 1.2 simonb int wapbl_lazy_truncate = 0; 228 1.2 simonb 229 1.2 simonb struct wapbl_ops wapbl_ops = { 230 1.2 simonb .wo_wapbl_discard = wapbl_discard, 231 1.2 simonb .wo_wapbl_replay_isopen = wapbl_replay_isopen1, 232 1.2 simonb .wo_wapbl_replay_read = wapbl_replay_read, 233 1.2 simonb .wo_wapbl_add_buf = wapbl_add_buf, 234 1.2 simonb .wo_wapbl_remove_buf = wapbl_remove_buf, 235 1.2 simonb .wo_wapbl_resize_buf = wapbl_resize_buf, 236 1.2 simonb .wo_wapbl_begin = wapbl_begin, 237 1.2 simonb .wo_wapbl_end = wapbl_end, 238 1.2 simonb .wo_wapbl_junlock_assert= wapbl_junlock_assert, 239 1.2 simonb 240 1.2 simonb /* XXX: the following is only used to say "this is a wapbl buf" */ 241 1.2 simonb .wo_wapbl_biodone = wapbl_biodone, 242 1.2 simonb }; 243 1.2 simonb 244 1.2 simonb void 245 1.2 simonb wapbl_init() 246 1.2 simonb { 247 1.2 simonb 248 1.2 simonb malloc_type_attach(M_WAPBL); 249 1.2 simonb } 250 1.2 simonb 251 1.2 simonb int 252 1.2 simonb wapbl_start(struct wapbl ** wlp, struct mount *mp, struct vnode *vp, 253 1.2 simonb daddr_t off, size_t count, size_t blksize, struct wapbl_replay *wr, 254 1.2 simonb wapbl_flush_fn_t flushfn, wapbl_flush_fn_t flushabortfn) 255 1.2 simonb { 256 1.2 simonb struct wapbl *wl; 257 1.2 simonb struct vnode *devvp; 258 1.2 simonb daddr_t logpbn; 259 1.2 simonb int error; 260 1.2 simonb int log_dev_bshift = DEV_BSHIFT; 261 1.2 simonb int fs_dev_bshift = DEV_BSHIFT; 262 1.2 simonb int run; 263 1.2 simonb 264 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_OPEN, ("wapbl_start: vp=%p off=%" PRId64 265 1.2 simonb " count=%zu blksize=%zu\n", vp, off, count, blksize)); 266 1.2 simonb 267 1.2 simonb if (log_dev_bshift > fs_dev_bshift) { 268 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_OPEN, 269 1.2 simonb ("wapbl: log device's block size cannot be larger " 270 1.2 simonb "than filesystem's\n")); 271 1.2 simonb /* 272 1.2 simonb * Not currently implemented, although it could be if 273 1.2 simonb * needed someday. 274 1.2 simonb */ 275 1.2 simonb return ENOSYS; 276 1.2 simonb } 277 1.2 simonb 278 1.2 simonb if (off < 0) 279 1.2 simonb return EINVAL; 280 1.2 simonb 281 1.2 simonb if (blksize < DEV_BSIZE) 282 1.2 simonb return EINVAL; 283 1.2 simonb if (blksize % DEV_BSIZE) 284 1.2 simonb return EINVAL; 285 1.2 simonb 286 1.2 simonb /* XXXTODO: verify that the full load is writable */ 287 1.2 simonb 288 1.2 simonb /* 289 1.2 simonb * XXX check for minimum log size 290 1.2 simonb * minimum is governed by minimum amount of space 291 1.2 simonb * to complete a transaction. (probably truncate) 292 1.2 simonb */ 293 1.2 simonb /* XXX for now pick something minimal */ 294 1.2 simonb if ((count * blksize) < MAXPHYS) { 295 1.2 simonb return ENOSPC; 296 1.2 simonb } 297 1.2 simonb 298 1.2 simonb if ((error = VOP_BMAP(vp, off, &devvp, &logpbn, &run)) != 0) { 299 1.2 simonb return error; 300 1.2 simonb } 301 1.2 simonb 302 1.2 simonb wl = wapbl_calloc(1, sizeof(*wl)); 303 1.2 simonb rw_init(&wl->wl_rwlock); 304 1.2 simonb mutex_init(&wl->wl_mtx, MUTEX_DEFAULT, IPL_NONE); 305 1.2 simonb cv_init(&wl->wl_reclaimable_cv, "wapblrec"); 306 1.2 simonb LIST_INIT(&wl->wl_bufs); 307 1.2 simonb SIMPLEQ_INIT(&wl->wl_entries); 308 1.2 simonb 309 1.2 simonb wl->wl_logvp = vp; 310 1.2 simonb wl->wl_devvp = devvp; 311 1.2 simonb wl->wl_mount = mp; 312 1.2 simonb wl->wl_logpbn = logpbn; 313 1.2 simonb wl->wl_log_dev_bshift = log_dev_bshift; 314 1.2 simonb wl->wl_fs_dev_bshift = fs_dev_bshift; 315 1.2 simonb 316 1.2 simonb wl->wl_flush = flushfn; 317 1.2 simonb wl->wl_flush_abort = flushabortfn; 318 1.2 simonb 319 1.2 simonb /* Reserve two log device blocks for the commit headers */ 320 1.2 simonb wl->wl_circ_off = 2<<wl->wl_log_dev_bshift; 321 1.2 simonb wl->wl_circ_size = ((count * blksize) - wl->wl_circ_off); 322 1.2 simonb /* truncate the log usage to a multiple of log_dev_bshift */ 323 1.2 simonb wl->wl_circ_size >>= wl->wl_log_dev_bshift; 324 1.2 simonb wl->wl_circ_size <<= wl->wl_log_dev_bshift; 325 1.2 simonb 326 1.2 simonb /* 327 1.2 simonb * wl_bufbytes_max limits the size of the in memory transaction space. 328 1.2 simonb * - Since buffers are allocated and accounted for in units of 329 1.2 simonb * PAGE_SIZE it is required to be a multiple of PAGE_SIZE 330 1.2 simonb * (i.e. 1<<PAGE_SHIFT) 331 1.2 simonb * - Since the log device has to be written in units of 332 1.2 simonb * 1<<wl_log_dev_bshift it is required to be a mulitple of 333 1.2 simonb * 1<<wl_log_dev_bshift. 334 1.2 simonb * - Since filesystem will provide data in units of 1<<wl_fs_dev_bshift, 335 1.2 simonb * it is convenient to be a multiple of 1<<wl_fs_dev_bshift. 336 1.2 simonb * Therefore it must be multiple of the least common multiple of those 337 1.2 simonb * three quantities. Fortunately, all of those quantities are 338 1.2 simonb * guaranteed to be a power of two, and the least common multiple of 339 1.2 simonb * a set of numbers which are all powers of two is simply the maximum 340 1.2 simonb * of those numbers. Finally, the maximum logarithm of a power of two 341 1.2 simonb * is the same as the log of the maximum power of two. So we can do 342 1.2 simonb * the following operations to size wl_bufbytes_max: 343 1.2 simonb */ 344 1.2 simonb 345 1.2 simonb /* XXX fix actual number of pages reserved per filesystem. */ 346 1.2 simonb wl->wl_bufbytes_max = MIN(wl->wl_circ_size, buf_memcalc() / 2); 347 1.2 simonb 348 1.2 simonb /* Round wl_bufbytes_max to the largest power of two constraint */ 349 1.2 simonb wl->wl_bufbytes_max >>= PAGE_SHIFT; 350 1.2 simonb wl->wl_bufbytes_max <<= PAGE_SHIFT; 351 1.2 simonb wl->wl_bufbytes_max >>= wl->wl_log_dev_bshift; 352 1.2 simonb wl->wl_bufbytes_max <<= wl->wl_log_dev_bshift; 353 1.2 simonb wl->wl_bufbytes_max >>= wl->wl_fs_dev_bshift; 354 1.2 simonb wl->wl_bufbytes_max <<= wl->wl_fs_dev_bshift; 355 1.2 simonb 356 1.2 simonb /* XXX maybe use filesystem fragment size instead of 1024 */ 357 1.2 simonb /* XXX fix actual number of buffers reserved per filesystem. */ 358 1.2 simonb wl->wl_bufcount_max = (nbuf / 2) * 1024; 359 1.2 simonb 360 1.2 simonb /* XXX tie this into resource estimation */ 361 1.2 simonb wl->wl_dealloclim = 2 * btodb(wl->wl_bufbytes_max); 362 1.2 simonb 363 1.2 simonb #if WAPBL_UVM_ALLOC 364 1.2 simonb wl->wl_deallocblks = (void *) uvm_km_zalloc(kernel_map, 365 1.2 simonb round_page(sizeof(*wl->wl_deallocblks) * wl->wl_dealloclim)); 366 1.2 simonb KASSERT(wl->wl_deallocblks != NULL); 367 1.2 simonb wl->wl_dealloclens = (void *) uvm_km_zalloc(kernel_map, 368 1.2 simonb round_page(sizeof(*wl->wl_dealloclens) * wl->wl_dealloclim)); 369 1.2 simonb KASSERT(wl->wl_dealloclens != NULL); 370 1.2 simonb #else 371 1.2 simonb wl->wl_deallocblks = wapbl_malloc(sizeof(*wl->wl_deallocblks) * 372 1.2 simonb wl->wl_dealloclim); 373 1.2 simonb wl->wl_dealloclens = wapbl_malloc(sizeof(*wl->wl_dealloclens) * 374 1.2 simonb wl->wl_dealloclim); 375 1.2 simonb #endif 376 1.2 simonb 377 1.2 simonb wapbl_inodetrk_init(wl, WAPBL_INODETRK_SIZE); 378 1.2 simonb 379 1.2 simonb /* Initialize the commit header */ 380 1.2 simonb { 381 1.2 simonb struct wapbl_wc_header *wc; 382 1.2 simonb size_t len = 1<<wl->wl_log_dev_bshift; 383 1.2 simonb wc = wapbl_calloc(1, len); 384 1.2 simonb wc->wc_type = WAPBL_WC_HEADER; 385 1.2 simonb wc->wc_len = len; 386 1.2 simonb wc->wc_circ_off = wl->wl_circ_off; 387 1.2 simonb wc->wc_circ_size = wl->wl_circ_size; 388 1.2 simonb /* XXX wc->wc_fsid */ 389 1.2 simonb wc->wc_log_dev_bshift = wl->wl_log_dev_bshift; 390 1.2 simonb wc->wc_fs_dev_bshift = wl->wl_fs_dev_bshift; 391 1.2 simonb wl->wl_wc_header = wc; 392 1.2 simonb wl->wl_wc_scratch = wapbl_malloc(len); 393 1.2 simonb } 394 1.2 simonb 395 1.2 simonb /* 396 1.2 simonb * if there was an existing set of unlinked but 397 1.2 simonb * allocated inodes, preserve it in the new 398 1.2 simonb * log. 399 1.2 simonb */ 400 1.2 simonb if (wr && wr->wr_inodescnt) { 401 1.2 simonb int i; 402 1.2 simonb 403 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_REPLAY, 404 1.2 simonb ("wapbl_start: reusing log with %d inodes\n", 405 1.2 simonb wr->wr_inodescnt)); 406 1.2 simonb 407 1.2 simonb /* 408 1.2 simonb * Its only valid to reuse the replay log if its 409 1.2 simonb * the same as the new log we just opened. 410 1.2 simonb */ 411 1.2 simonb KDASSERT(!wapbl_replay_isopen(wr)); 412 1.2 simonb KASSERT(devvp->v_rdev == wr->wr_devvp->v_rdev); 413 1.2 simonb KASSERT(logpbn == wr->wr_logpbn); 414 1.2 simonb KASSERT(wl->wl_circ_size == wr->wr_wc_header.wc_circ_size); 415 1.2 simonb KASSERT(wl->wl_circ_off == wr->wr_wc_header.wc_circ_off); 416 1.2 simonb KASSERT(wl->wl_log_dev_bshift == 417 1.2 simonb wr->wr_wc_header.wc_log_dev_bshift); 418 1.2 simonb KASSERT(wl->wl_fs_dev_bshift == 419 1.2 simonb wr->wr_wc_header.wc_fs_dev_bshift); 420 1.2 simonb 421 1.2 simonb wl->wl_wc_header->wc_generation = 422 1.2 simonb wr->wr_wc_header.wc_generation + 1; 423 1.2 simonb 424 1.2 simonb for (i = 0; i < wr->wr_inodescnt; i++) 425 1.2 simonb wapbl_register_inode(wl, wr->wr_inodes[i].wr_inumber, 426 1.2 simonb wr->wr_inodes[i].wr_imode); 427 1.2 simonb 428 1.2 simonb /* Make sure new transaction won't overwrite old inodes list */ 429 1.2 simonb KDASSERT(wapbl_transaction_len(wl) <= 430 1.2 simonb wapbl_space_free(wl->wl_circ_size, wr->wr_inodeshead, 431 1.2 simonb wr->wr_inodestail)); 432 1.2 simonb 433 1.2 simonb wl->wl_head = wl->wl_tail = wr->wr_inodeshead; 434 1.2 simonb wl->wl_reclaimable_bytes = wl->wl_reserved_bytes = 435 1.2 simonb wapbl_transaction_len(wl); 436 1.2 simonb 437 1.2 simonb error = wapbl_write_inodes(wl, &wl->wl_head); 438 1.2 simonb if (error) 439 1.2 simonb goto errout; 440 1.2 simonb 441 1.2 simonb KASSERT(wl->wl_head != wl->wl_tail); 442 1.2 simonb KASSERT(wl->wl_head != 0); 443 1.2 simonb } 444 1.2 simonb 445 1.2 simonb error = wapbl_write_commit(wl, wl->wl_head, wl->wl_tail); 446 1.2 simonb if (error) { 447 1.2 simonb goto errout; 448 1.2 simonb } 449 1.2 simonb 450 1.2 simonb *wlp = wl; 451 1.2 simonb #if defined(WAPBL_DEBUG) 452 1.2 simonb wapbl_debug_wl = wl; 453 1.2 simonb #endif 454 1.2 simonb 455 1.2 simonb return 0; 456 1.2 simonb errout: 457 1.2 simonb wapbl_discard(wl); 458 1.2 simonb wapbl_free(wl->wl_wc_scratch); 459 1.2 simonb wapbl_free(wl->wl_wc_header); 460 1.2 simonb #if WAPBL_UVM_ALLOC 461 1.2 simonb uvm_km_free_wakeup(kernel_map, (vaddr_t) wl->wl_deallocblks, 462 1.2 simonb round_page(sizeof(*wl->wl_deallocblks * 463 1.2 simonb wl->wl_dealloclim))); 464 1.2 simonb uvm_km_free_wakeup(kernel_map, (vaddr_t) wl->wl_dealloclens, 465 1.2 simonb round_page(sizeof(*wl->wl_dealloclens * 466 1.2 simonb wl->wl_dealloclim))); 467 1.2 simonb #else 468 1.2 simonb wapbl_free(wl->wl_deallocblks); 469 1.2 simonb wapbl_free(wl->wl_dealloclens); 470 1.2 simonb #endif 471 1.2 simonb wapbl_inodetrk_free(wl); 472 1.2 simonb wapbl_free(wl); 473 1.2 simonb 474 1.2 simonb return error; 475 1.2 simonb } 476 1.2 simonb 477 1.2 simonb /* 478 1.2 simonb * Like wapbl_flush, only discards the transaction 479 1.2 simonb * completely 480 1.2 simonb */ 481 1.2 simonb 482 1.2 simonb void 483 1.2 simonb wapbl_discard(struct wapbl *wl) 484 1.2 simonb { 485 1.2 simonb struct wapbl_entry *we; 486 1.2 simonb struct buf *bp; 487 1.2 simonb int i; 488 1.2 simonb 489 1.2 simonb /* 490 1.2 simonb * XXX we may consider using upgrade here 491 1.2 simonb * if we want to call flush from inside a transaction 492 1.2 simonb */ 493 1.2 simonb rw_enter(&wl->wl_rwlock, RW_WRITER); 494 1.2 simonb wl->wl_flush(wl->wl_mount, wl->wl_deallocblks, wl->wl_dealloclens, 495 1.2 simonb wl->wl_dealloccnt); 496 1.2 simonb 497 1.2 simonb #ifdef WAPBL_DEBUG_PRINT 498 1.2 simonb { 499 1.2 simonb struct wapbl_entry *we; 500 1.2 simonb pid_t pid = -1; 501 1.2 simonb lwpid_t lid = -1; 502 1.2 simonb if (curproc) 503 1.2 simonb pid = curproc->p_pid; 504 1.2 simonb if (curlwp) 505 1.2 simonb lid = curlwp->l_lid; 506 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 507 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_DISCARD, 508 1.2 simonb ("wapbl_discard: thread %d.%d discarding " 509 1.2 simonb "transaction\n" 510 1.2 simonb "\tbufcount=%zu bufbytes=%zu bcount=%zu " 511 1.2 simonb "deallocs=%d inodes=%d\n" 512 1.2 simonb "\terrcnt = %u, reclaimable=%zu reserved=%zu " 513 1.2 simonb "unsynced=%zu\n", 514 1.2 simonb pid, lid, wl->wl_bufcount, wl->wl_bufbytes, 515 1.2 simonb wl->wl_bcount, wl->wl_dealloccnt, 516 1.2 simonb wl->wl_inohashcnt, wl->wl_error_count, 517 1.2 simonb wl->wl_reclaimable_bytes, wl->wl_reserved_bytes, 518 1.2 simonb wl->wl_unsynced_bufbytes)); 519 1.2 simonb SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) { 520 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_DISCARD, 521 1.2 simonb ("\tentry: bufcount = %zu, reclaimable = %zu, " 522 1.2 simonb "error = %d, unsynced = %zu\n", 523 1.2 simonb we->we_bufcount, we->we_reclaimable_bytes, 524 1.2 simonb we->we_error, we->we_unsynced_bufbytes)); 525 1.2 simonb } 526 1.2 simonb #else /* !WAPBL_DEBUG_BUFBYTES */ 527 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_DISCARD, 528 1.2 simonb ("wapbl_discard: thread %d.%d discarding transaction\n" 529 1.2 simonb "\tbufcount=%zu bufbytes=%zu bcount=%zu " 530 1.2 simonb "deallocs=%d inodes=%d\n" 531 1.2 simonb "\terrcnt = %u, reclaimable=%zu reserved=%zu\n", 532 1.2 simonb pid, lid, wl->wl_bufcount, wl->wl_bufbytes, 533 1.2 simonb wl->wl_bcount, wl->wl_dealloccnt, 534 1.2 simonb wl->wl_inohashcnt, wl->wl_error_count, 535 1.2 simonb wl->wl_reclaimable_bytes, wl->wl_reserved_bytes)); 536 1.2 simonb SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) { 537 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_DISCARD, 538 1.2 simonb ("\tentry: bufcount = %zu, reclaimable = %zu, " 539 1.2 simonb "error = %d\n", 540 1.2 simonb we->we_bufcount, we->we_reclaimable_bytes, 541 1.2 simonb we->we_error)); 542 1.2 simonb } 543 1.2 simonb #endif /* !WAPBL_DEBUG_BUFBYTES */ 544 1.2 simonb } 545 1.2 simonb #endif /* WAPBL_DEBUG_PRINT */ 546 1.2 simonb 547 1.2 simonb for (i = 0; i <= wl->wl_inohashmask; i++) { 548 1.2 simonb struct wapbl_ino_head *wih; 549 1.2 simonb struct wapbl_ino *wi; 550 1.2 simonb 551 1.2 simonb wih = &wl->wl_inohash[i]; 552 1.2 simonb while ((wi = LIST_FIRST(wih)) != NULL) { 553 1.2 simonb LIST_REMOVE(wi, wi_hash); 554 1.2 simonb pool_put(&wapbl_ino_pool, wi); 555 1.2 simonb KASSERT(wl->wl_inohashcnt > 0); 556 1.2 simonb wl->wl_inohashcnt--; 557 1.2 simonb } 558 1.2 simonb } 559 1.2 simonb 560 1.2 simonb /* 561 1.2 simonb * clean buffer list 562 1.2 simonb */ 563 1.2 simonb mutex_enter(&bufcache_lock); 564 1.2 simonb mutex_enter(&wl->wl_mtx); 565 1.2 simonb while ((bp = LIST_FIRST(&wl->wl_bufs)) != NULL) { 566 1.2 simonb if (bbusy(bp, 0, 0, &wl->wl_mtx) == 0) { 567 1.2 simonb /* 568 1.2 simonb * The buffer will be unlocked and 569 1.2 simonb * removed from the transaction in brelse 570 1.2 simonb */ 571 1.2 simonb mutex_exit(&wl->wl_mtx); 572 1.2 simonb brelsel(bp, 0); 573 1.2 simonb mutex_enter(&wl->wl_mtx); 574 1.2 simonb } 575 1.2 simonb } 576 1.2 simonb mutex_exit(&wl->wl_mtx); 577 1.2 simonb mutex_exit(&bufcache_lock); 578 1.2 simonb 579 1.2 simonb /* 580 1.2 simonb * Remove references to this wl from wl_entries, free any which 581 1.2 simonb * no longer have buffers, others will be freed in wapbl_biodone 582 1.2 simonb * when they no longer have any buffers. 583 1.2 simonb */ 584 1.2 simonb while ((we = SIMPLEQ_FIRST(&wl->wl_entries)) != NULL) { 585 1.2 simonb SIMPLEQ_REMOVE_HEAD(&wl->wl_entries, we_entries); 586 1.2 simonb /* XXX should we be accumulating wl_error_count 587 1.2 simonb * and increasing reclaimable bytes ? */ 588 1.2 simonb we->we_wapbl = NULL; 589 1.2 simonb if (we->we_bufcount == 0) { 590 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 591 1.2 simonb KASSERT(we->we_unsynced_bufbytes == 0); 592 1.2 simonb #endif 593 1.2 simonb wapbl_free(we); 594 1.2 simonb } 595 1.2 simonb } 596 1.2 simonb 597 1.2 simonb /* Discard list of deallocs */ 598 1.2 simonb wl->wl_dealloccnt = 0; 599 1.2 simonb /* XXX should we clear wl_reserved_bytes? */ 600 1.2 simonb 601 1.2 simonb KASSERT(wl->wl_bufbytes == 0); 602 1.2 simonb KASSERT(wl->wl_bcount == 0); 603 1.2 simonb KASSERT(wl->wl_bufcount == 0); 604 1.2 simonb KASSERT(LIST_EMPTY(&wl->wl_bufs)); 605 1.2 simonb KASSERT(SIMPLEQ_EMPTY(&wl->wl_entries)); 606 1.2 simonb KASSERT(wl->wl_inohashcnt == 0); 607 1.2 simonb 608 1.2 simonb rw_exit(&wl->wl_rwlock); 609 1.2 simonb } 610 1.2 simonb 611 1.2 simonb int 612 1.2 simonb wapbl_stop(struct wapbl *wl, int force) 613 1.2 simonb { 614 1.2 simonb struct vnode *vp; 615 1.2 simonb int error; 616 1.2 simonb 617 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_OPEN, ("wapbl_stop called\n")); 618 1.2 simonb error = wapbl_flush(wl, 1); 619 1.2 simonb if (error) { 620 1.2 simonb if (force) 621 1.2 simonb wapbl_discard(wl); 622 1.2 simonb else 623 1.2 simonb return error; 624 1.2 simonb } 625 1.2 simonb 626 1.2 simonb /* Unlinked inodes persist after a flush */ 627 1.2 simonb if (wl->wl_inohashcnt) { 628 1.2 simonb if (force) { 629 1.2 simonb wapbl_discard(wl); 630 1.2 simonb } else { 631 1.2 simonb return EBUSY; 632 1.2 simonb } 633 1.2 simonb } 634 1.2 simonb 635 1.2 simonb KASSERT(wl->wl_bufbytes == 0); 636 1.2 simonb KASSERT(wl->wl_bcount == 0); 637 1.2 simonb KASSERT(wl->wl_bufcount == 0); 638 1.2 simonb KASSERT(LIST_EMPTY(&wl->wl_bufs)); 639 1.2 simonb KASSERT(wl->wl_dealloccnt == 0); 640 1.2 simonb KASSERT(SIMPLEQ_EMPTY(&wl->wl_entries)); 641 1.2 simonb KASSERT(wl->wl_inohashcnt == 0); 642 1.2 simonb 643 1.2 simonb vp = wl->wl_logvp; 644 1.2 simonb 645 1.2 simonb wapbl_free(wl->wl_wc_scratch); 646 1.2 simonb wapbl_free(wl->wl_wc_header); 647 1.2 simonb #if WAPBL_UVM_ALLOC 648 1.2 simonb uvm_km_free_wakeup(kernel_map, (vaddr_t) wl->wl_deallocblks, 649 1.2 simonb round_page(sizeof(*wl->wl_deallocblks * 650 1.2 simonb wl->wl_dealloclim))); 651 1.2 simonb uvm_km_free_wakeup(kernel_map, (vaddr_t) wl->wl_dealloclens, 652 1.2 simonb round_page(sizeof(*wl->wl_dealloclens * 653 1.2 simonb wl->wl_dealloclim))); 654 1.2 simonb #else 655 1.2 simonb wapbl_free(wl->wl_deallocblks); 656 1.2 simonb wapbl_free(wl->wl_dealloclens); 657 1.2 simonb #endif 658 1.2 simonb wapbl_inodetrk_free(wl); 659 1.2 simonb 660 1.2 simonb cv_destroy(&wl->wl_reclaimable_cv); 661 1.2 simonb mutex_destroy(&wl->wl_mtx); 662 1.2 simonb rw_destroy(&wl->wl_rwlock); 663 1.2 simonb wapbl_free(wl); 664 1.2 simonb 665 1.2 simonb return 0; 666 1.2 simonb } 667 1.2 simonb 668 1.2 simonb static int 669 1.2 simonb wapbl_doio(void *data, size_t len, struct vnode *devvp, daddr_t pbn, int flags) 670 1.2 simonb { 671 1.2 simonb struct pstats *pstats = curlwp->l_proc->p_stats; 672 1.2 simonb struct buf *bp; 673 1.2 simonb int error; 674 1.2 simonb 675 1.2 simonb KASSERT((flags & ~(B_WRITE | B_READ)) == 0); 676 1.2 simonb KASSERT(devvp->v_type == VBLK); 677 1.2 simonb 678 1.2 simonb if ((flags & (B_WRITE | B_READ)) == B_WRITE) { 679 1.2 simonb mutex_enter(&devvp->v_interlock); 680 1.2 simonb devvp->v_numoutput++; 681 1.2 simonb mutex_exit(&devvp->v_interlock); 682 1.2 simonb pstats->p_ru.ru_oublock++; 683 1.2 simonb } else { 684 1.2 simonb pstats->p_ru.ru_inblock++; 685 1.2 simonb } 686 1.2 simonb 687 1.2 simonb bp = getiobuf(devvp, true); 688 1.2 simonb bp->b_flags = flags; 689 1.2 simonb bp->b_cflags = BC_BUSY; /* silly & dubious */ 690 1.2 simonb bp->b_dev = devvp->v_rdev; 691 1.2 simonb bp->b_data = data; 692 1.2 simonb bp->b_bufsize = bp->b_resid = bp->b_bcount = len; 693 1.2 simonb bp->b_blkno = pbn; 694 1.2 simonb 695 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_IO, 696 1.2 simonb ("wapbl_doio: %s %d bytes at block %"PRId64" on dev 0x%x\n", 697 1.2 simonb BUF_ISWRITE(bp) ? "write" : "read", bp->b_bcount, 698 1.2 simonb bp->b_blkno, bp->b_dev)); 699 1.2 simonb 700 1.2 simonb VOP_STRATEGY(devvp, bp); 701 1.2 simonb 702 1.2 simonb error = biowait(bp); 703 1.2 simonb putiobuf(bp); 704 1.2 simonb 705 1.2 simonb if (error) { 706 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_ERROR, 707 1.2 simonb ("wapbl_doio: %s %zu bytes at block %" PRId64 708 1.2 simonb " on dev 0x%x failed with error %d\n", 709 1.2 simonb (((flags & (B_WRITE | B_READ)) == B_WRITE) ? 710 1.2 simonb "write" : "read"), 711 1.2 simonb len, pbn, devvp->v_rdev, error)); 712 1.2 simonb } 713 1.2 simonb 714 1.2 simonb return error; 715 1.2 simonb } 716 1.2 simonb 717 1.2 simonb int 718 1.2 simonb wapbl_write(void *data, size_t len, struct vnode *devvp, daddr_t pbn) 719 1.2 simonb { 720 1.2 simonb 721 1.2 simonb return wapbl_doio(data, len, devvp, pbn, B_WRITE); 722 1.2 simonb } 723 1.2 simonb 724 1.2 simonb int 725 1.2 simonb wapbl_read(void *data, size_t len, struct vnode *devvp, daddr_t pbn) 726 1.2 simonb { 727 1.2 simonb 728 1.2 simonb return wapbl_doio(data, len, devvp, pbn, B_READ); 729 1.2 simonb } 730 1.2 simonb 731 1.2 simonb /* 732 1.2 simonb * Off is byte offset returns new offset for next write 733 1.2 simonb * handles log wraparound 734 1.2 simonb */ 735 1.2 simonb static int 736 1.2 simonb wapbl_circ_write(struct wapbl *wl, void *data, size_t len, off_t *offp) 737 1.2 simonb { 738 1.2 simonb size_t slen; 739 1.2 simonb off_t off = *offp; 740 1.2 simonb int error; 741 1.2 simonb 742 1.2 simonb KDASSERT(((len >> wl->wl_log_dev_bshift) << 743 1.2 simonb wl->wl_log_dev_bshift) == len); 744 1.2 simonb 745 1.2 simonb if (off < wl->wl_circ_off) 746 1.2 simonb off = wl->wl_circ_off; 747 1.2 simonb slen = wl->wl_circ_off + wl->wl_circ_size - off; 748 1.2 simonb if (slen < len) { 749 1.2 simonb error = wapbl_write(data, slen, wl->wl_devvp, 750 1.2 simonb wl->wl_logpbn + (off >> wl->wl_log_dev_bshift)); 751 1.2 simonb if (error) 752 1.2 simonb return error; 753 1.2 simonb data = (uint8_t *)data + slen; 754 1.2 simonb len -= slen; 755 1.2 simonb off = wl->wl_circ_off; 756 1.2 simonb } 757 1.2 simonb error = wapbl_write(data, len, wl->wl_devvp, 758 1.2 simonb wl->wl_logpbn + (off >> wl->wl_log_dev_bshift)); 759 1.2 simonb if (error) 760 1.2 simonb return error; 761 1.2 simonb off += len; 762 1.2 simonb if (off >= wl->wl_circ_off + wl->wl_circ_size) 763 1.2 simonb off = wl->wl_circ_off; 764 1.2 simonb *offp = off; 765 1.2 simonb return 0; 766 1.2 simonb } 767 1.2 simonb 768 1.2 simonb /****************************************************************/ 769 1.2 simonb 770 1.2 simonb int 771 1.2 simonb wapbl_begin(struct wapbl *wl, const char *file, int line) 772 1.2 simonb { 773 1.2 simonb int doflush; 774 1.2 simonb unsigned lockcount; 775 1.2 simonb 776 1.2 simonb KDASSERT(wl); 777 1.2 simonb 778 1.2 simonb /* 779 1.2 simonb * XXX this needs to be made much more sophisticated. 780 1.2 simonb * perhaps each wapbl_begin could reserve a specified 781 1.2 simonb * number of buffers and bytes. 782 1.2 simonb */ 783 1.2 simonb mutex_enter(&wl->wl_mtx); 784 1.2 simonb lockcount = wl->wl_lock_count; 785 1.2 simonb doflush = ((wl->wl_bufbytes + (lockcount * MAXPHYS)) > 786 1.2 simonb wl->wl_bufbytes_max / 2) || 787 1.2 simonb ((wl->wl_bufcount + (lockcount * 10)) > 788 1.2 simonb wl->wl_bufcount_max / 2) || 789 1.2 simonb (wapbl_transaction_len(wl) > wl->wl_circ_size / 2); 790 1.2 simonb mutex_exit(&wl->wl_mtx); 791 1.2 simonb 792 1.2 simonb if (doflush) { 793 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_FLUSH, 794 1.2 simonb ("force flush lockcnt=%d bufbytes=%zu " 795 1.2 simonb "(max=%zu) bufcount=%zu (max=%zu)\n", 796 1.2 simonb lockcount, wl->wl_bufbytes, 797 1.2 simonb wl->wl_bufbytes_max, wl->wl_bufcount, 798 1.2 simonb wl->wl_bufcount_max)); 799 1.2 simonb } 800 1.2 simonb 801 1.2 simonb if (doflush) { 802 1.2 simonb int error = wapbl_flush(wl, 0); 803 1.2 simonb if (error) 804 1.2 simonb return error; 805 1.2 simonb } 806 1.2 simonb 807 1.3.8.1 snj rw_enter(&wl->wl_rwlock, RW_READER); 808 1.2 simonb mutex_enter(&wl->wl_mtx); 809 1.2 simonb wl->wl_lock_count++; 810 1.2 simonb mutex_exit(&wl->wl_mtx); 811 1.2 simonb 812 1.3.8.1 snj #if defined(WAPBL_DEBUG_PRINT) 813 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_TRANSACTION, 814 1.2 simonb ("wapbl_begin thread %d.%d with bufcount=%zu " 815 1.2 simonb "bufbytes=%zu bcount=%zu at %s:%d\n", 816 1.2 simonb curproc->p_pid, curlwp->l_lid, wl->wl_bufcount, 817 1.2 simonb wl->wl_bufbytes, wl->wl_bcount, file, line)); 818 1.2 simonb #endif 819 1.2 simonb 820 1.2 simonb return 0; 821 1.2 simonb } 822 1.2 simonb 823 1.2 simonb void 824 1.2 simonb wapbl_end(struct wapbl *wl) 825 1.2 simonb { 826 1.2 simonb 827 1.3.8.1 snj #if defined(WAPBL_DEBUG_PRINT) 828 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_TRANSACTION, 829 1.2 simonb ("wapbl_end thread %d.%d with bufcount=%zu " 830 1.2 simonb "bufbytes=%zu bcount=%zu\n", 831 1.2 simonb curproc->p_pid, curlwp->l_lid, wl->wl_bufcount, 832 1.2 simonb wl->wl_bufbytes, wl->wl_bcount)); 833 1.2 simonb #endif 834 1.2 simonb 835 1.2 simonb mutex_enter(&wl->wl_mtx); 836 1.2 simonb KASSERT(wl->wl_lock_count > 0); 837 1.2 simonb wl->wl_lock_count--; 838 1.2 simonb mutex_exit(&wl->wl_mtx); 839 1.2 simonb 840 1.2 simonb rw_exit(&wl->wl_rwlock); 841 1.2 simonb } 842 1.2 simonb 843 1.2 simonb void 844 1.2 simonb wapbl_add_buf(struct wapbl *wl, struct buf * bp) 845 1.2 simonb { 846 1.2 simonb 847 1.2 simonb KASSERT(bp->b_cflags & BC_BUSY); 848 1.2 simonb KASSERT(bp->b_vp); 849 1.2 simonb 850 1.2 simonb wapbl_jlock_assert(wl); 851 1.2 simonb 852 1.2 simonb #if 0 853 1.2 simonb /* 854 1.2 simonb * XXX this might be an issue for swapfiles. 855 1.2 simonb * see uvm_swap.c:1702 856 1.2 simonb * 857 1.2 simonb * XXX2 why require it then? leap of semantics? 858 1.2 simonb */ 859 1.2 simonb KASSERT((bp->b_cflags & BC_NOCACHE) == 0); 860 1.2 simonb #endif 861 1.2 simonb 862 1.2 simonb mutex_enter(&wl->wl_mtx); 863 1.2 simonb if (bp->b_flags & B_LOCKED) { 864 1.2 simonb LIST_REMOVE(bp, b_wapbllist); 865 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_BUFFER2, 866 1.2 simonb ("wapbl_add_buf thread %d.%d re-adding buf %p " 867 1.2 simonb "with %d bytes %d bcount\n", 868 1.2 simonb curproc->p_pid, curlwp->l_lid, bp, bp->b_bufsize, 869 1.2 simonb bp->b_bcount)); 870 1.2 simonb } else { 871 1.2 simonb /* unlocked by dirty buffers shouldn't exist */ 872 1.2 simonb KASSERT(!(bp->b_oflags & BO_DELWRI)); 873 1.2 simonb wl->wl_bufbytes += bp->b_bufsize; 874 1.2 simonb wl->wl_bcount += bp->b_bcount; 875 1.2 simonb wl->wl_bufcount++; 876 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_BUFFER, 877 1.2 simonb ("wapbl_add_buf thread %d.%d adding buf %p " 878 1.2 simonb "with %d bytes %d bcount\n", 879 1.2 simonb curproc->p_pid, curlwp->l_lid, bp, bp->b_bufsize, 880 1.2 simonb bp->b_bcount)); 881 1.2 simonb } 882 1.2 simonb LIST_INSERT_HEAD(&wl->wl_bufs, bp, b_wapbllist); 883 1.2 simonb mutex_exit(&wl->wl_mtx); 884 1.2 simonb 885 1.2 simonb bp->b_flags |= B_LOCKED; 886 1.2 simonb } 887 1.2 simonb 888 1.2 simonb static void 889 1.2 simonb wapbl_remove_buf_locked(struct wapbl * wl, struct buf *bp) 890 1.2 simonb { 891 1.2 simonb 892 1.2 simonb KASSERT(mutex_owned(&wl->wl_mtx)); 893 1.2 simonb KASSERT(bp->b_cflags & BC_BUSY); 894 1.2 simonb wapbl_jlock_assert(wl); 895 1.2 simonb 896 1.2 simonb #if 0 897 1.2 simonb /* 898 1.2 simonb * XXX this might be an issue for swapfiles. 899 1.2 simonb * see uvm_swap.c:1725 900 1.2 simonb * 901 1.2 simonb * XXXdeux: see above 902 1.2 simonb */ 903 1.2 simonb KASSERT((bp->b_flags & BC_NOCACHE) == 0); 904 1.2 simonb #endif 905 1.2 simonb KASSERT(bp->b_flags & B_LOCKED); 906 1.2 simonb 907 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_BUFFER, 908 1.2 simonb ("wapbl_remove_buf thread %d.%d removing buf %p with " 909 1.2 simonb "%d bytes %d bcount\n", 910 1.2 simonb curproc->p_pid, curlwp->l_lid, bp, bp->b_bufsize, bp->b_bcount)); 911 1.2 simonb 912 1.2 simonb KASSERT(wl->wl_bufbytes >= bp->b_bufsize); 913 1.2 simonb wl->wl_bufbytes -= bp->b_bufsize; 914 1.2 simonb KASSERT(wl->wl_bcount >= bp->b_bcount); 915 1.2 simonb wl->wl_bcount -= bp->b_bcount; 916 1.2 simonb KASSERT(wl->wl_bufcount > 0); 917 1.2 simonb wl->wl_bufcount--; 918 1.2 simonb KASSERT((wl->wl_bufcount == 0) == (wl->wl_bufbytes == 0)); 919 1.2 simonb KASSERT((wl->wl_bufcount == 0) == (wl->wl_bcount == 0)); 920 1.2 simonb LIST_REMOVE(bp, b_wapbllist); 921 1.2 simonb 922 1.2 simonb bp->b_flags &= ~B_LOCKED; 923 1.2 simonb } 924 1.2 simonb 925 1.2 simonb /* called from brelsel() in vfs_bio among other places */ 926 1.2 simonb void 927 1.2 simonb wapbl_remove_buf(struct wapbl * wl, struct buf *bp) 928 1.2 simonb { 929 1.2 simonb 930 1.2 simonb mutex_enter(&wl->wl_mtx); 931 1.2 simonb wapbl_remove_buf_locked(wl, bp); 932 1.2 simonb mutex_exit(&wl->wl_mtx); 933 1.2 simonb } 934 1.2 simonb 935 1.2 simonb void 936 1.2 simonb wapbl_resize_buf(struct wapbl *wl, struct buf *bp, long oldsz, long oldcnt) 937 1.2 simonb { 938 1.2 simonb 939 1.2 simonb KASSERT(bp->b_cflags & BC_BUSY); 940 1.2 simonb 941 1.2 simonb /* 942 1.2 simonb * XXX: why does this depend on B_LOCKED? otherwise the buf 943 1.2 simonb * is not for a transaction? if so, why is this called in the 944 1.2 simonb * first place? 945 1.2 simonb */ 946 1.2 simonb if (bp->b_flags & B_LOCKED) { 947 1.2 simonb mutex_enter(&wl->wl_mtx); 948 1.2 simonb wl->wl_bufbytes += bp->b_bufsize - oldsz; 949 1.2 simonb wl->wl_bcount += bp->b_bcount - oldcnt; 950 1.2 simonb mutex_exit(&wl->wl_mtx); 951 1.2 simonb } 952 1.2 simonb } 953 1.2 simonb 954 1.2 simonb #endif /* _KERNEL */ 955 1.2 simonb 956 1.2 simonb /****************************************************************/ 957 1.2 simonb /* Some utility inlines */ 958 1.2 simonb 959 1.2 simonb /* This is used to advance the pointer at old to new value at old+delta */ 960 1.2 simonb static __inline off_t 961 1.2 simonb wapbl_advance(size_t size, size_t off, off_t old, size_t delta) 962 1.2 simonb { 963 1.2 simonb off_t new; 964 1.2 simonb 965 1.2 simonb /* Define acceptable ranges for inputs. */ 966 1.2 simonb KASSERT(delta <= size); 967 1.2 simonb KASSERT((old == 0) || (old >= off)); 968 1.2 simonb KASSERT(old < (size + off)); 969 1.2 simonb 970 1.2 simonb if ((old == 0) && (delta != 0)) 971 1.2 simonb new = off + delta; 972 1.2 simonb else if ((old + delta) < (size + off)) 973 1.2 simonb new = old + delta; 974 1.2 simonb else 975 1.2 simonb new = (old + delta) - size; 976 1.2 simonb 977 1.2 simonb /* Note some interesting axioms */ 978 1.2 simonb KASSERT((delta != 0) || (new == old)); 979 1.2 simonb KASSERT((delta == 0) || (new != 0)); 980 1.2 simonb KASSERT((delta != (size)) || (new == old)); 981 1.2 simonb 982 1.2 simonb /* Define acceptable ranges for output. */ 983 1.2 simonb KASSERT((new == 0) || (new >= off)); 984 1.2 simonb KASSERT(new < (size + off)); 985 1.2 simonb return new; 986 1.2 simonb } 987 1.2 simonb 988 1.2 simonb static __inline size_t 989 1.2 simonb wapbl_space_used(size_t avail, off_t head, off_t tail) 990 1.2 simonb { 991 1.2 simonb 992 1.2 simonb if (tail == 0) { 993 1.2 simonb KASSERT(head == 0); 994 1.2 simonb return 0; 995 1.2 simonb } 996 1.2 simonb return ((head + (avail - 1) - tail) % avail) + 1; 997 1.2 simonb } 998 1.2 simonb 999 1.2 simonb static __inline size_t 1000 1.2 simonb wapbl_space_free(size_t avail, off_t head, off_t tail) 1001 1.2 simonb { 1002 1.2 simonb 1003 1.2 simonb return avail - wapbl_space_used(avail, head, tail); 1004 1.2 simonb } 1005 1.2 simonb 1006 1.2 simonb static __inline void 1007 1.2 simonb wapbl_advance_head(size_t size, size_t off, size_t delta, off_t *headp, 1008 1.2 simonb off_t *tailp) 1009 1.2 simonb { 1010 1.2 simonb off_t head = *headp; 1011 1.2 simonb off_t tail = *tailp; 1012 1.2 simonb 1013 1.2 simonb KASSERT(delta <= wapbl_space_free(size, head, tail)); 1014 1.2 simonb head = wapbl_advance(size, off, head, delta); 1015 1.2 simonb if ((tail == 0) && (head != 0)) 1016 1.2 simonb tail = off; 1017 1.2 simonb *headp = head; 1018 1.2 simonb *tailp = tail; 1019 1.2 simonb } 1020 1.2 simonb 1021 1.2 simonb static __inline void 1022 1.2 simonb wapbl_advance_tail(size_t size, size_t off, size_t delta, off_t *headp, 1023 1.2 simonb off_t *tailp) 1024 1.2 simonb { 1025 1.2 simonb off_t head = *headp; 1026 1.2 simonb off_t tail = *tailp; 1027 1.2 simonb 1028 1.2 simonb KASSERT(delta <= wapbl_space_used(size, head, tail)); 1029 1.2 simonb tail = wapbl_advance(size, off, tail, delta); 1030 1.2 simonb if (head == tail) { 1031 1.2 simonb head = tail = 0; 1032 1.2 simonb } 1033 1.2 simonb *headp = head; 1034 1.2 simonb *tailp = tail; 1035 1.2 simonb } 1036 1.2 simonb 1037 1.2 simonb #ifdef _KERNEL 1038 1.2 simonb 1039 1.2 simonb /****************************************************************/ 1040 1.2 simonb 1041 1.2 simonb /* 1042 1.2 simonb * Remove transactions whose buffers are completely flushed to disk. 1043 1.2 simonb * Will block until at least minfree space is available. 1044 1.2 simonb * only intended to be called from inside wapbl_flush and therefore 1045 1.2 simonb * does not protect against commit races with itself or with flush. 1046 1.2 simonb */ 1047 1.2 simonb static int 1048 1.2 simonb wapbl_truncate(struct wapbl *wl, size_t minfree, int waitonly) 1049 1.2 simonb { 1050 1.2 simonb size_t delta; 1051 1.2 simonb size_t avail; 1052 1.2 simonb off_t head; 1053 1.2 simonb off_t tail; 1054 1.2 simonb int error = 0; 1055 1.2 simonb 1056 1.2 simonb KASSERT(minfree <= (wl->wl_circ_size - wl->wl_reserved_bytes)); 1057 1.2 simonb KASSERT(rw_write_held(&wl->wl_rwlock)); 1058 1.2 simonb 1059 1.2 simonb mutex_enter(&wl->wl_mtx); 1060 1.2 simonb 1061 1.2 simonb /* 1062 1.2 simonb * First check to see if we have to do a commit 1063 1.2 simonb * at all. 1064 1.2 simonb */ 1065 1.2 simonb avail = wapbl_space_free(wl->wl_circ_size, wl->wl_head, wl->wl_tail); 1066 1.2 simonb if (minfree < avail) { 1067 1.2 simonb mutex_exit(&wl->wl_mtx); 1068 1.2 simonb return 0; 1069 1.2 simonb } 1070 1.2 simonb minfree -= avail; 1071 1.2 simonb while ((wl->wl_error_count == 0) && 1072 1.2 simonb (wl->wl_reclaimable_bytes < minfree)) { 1073 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_TRUNCATE, 1074 1.2 simonb ("wapbl_truncate: sleeping on %p wl=%p bytes=%zd " 1075 1.2 simonb "minfree=%zd\n", 1076 1.2 simonb &wl->wl_reclaimable_bytes, wl, wl->wl_reclaimable_bytes, 1077 1.2 simonb minfree)); 1078 1.2 simonb 1079 1.2 simonb cv_wait(&wl->wl_reclaimable_cv, &wl->wl_mtx); 1080 1.2 simonb } 1081 1.2 simonb if (wl->wl_reclaimable_bytes < minfree) { 1082 1.2 simonb KASSERT(wl->wl_error_count); 1083 1.2 simonb /* XXX maybe get actual error from buffer instead someday? */ 1084 1.2 simonb error = EIO; 1085 1.2 simonb } 1086 1.2 simonb head = wl->wl_head; 1087 1.2 simonb tail = wl->wl_tail; 1088 1.2 simonb delta = wl->wl_reclaimable_bytes; 1089 1.2 simonb 1090 1.2 simonb /* If all of of the entries are flushed, then be sure to keep 1091 1.2 simonb * the reserved bytes reserved. Watch out for discarded transactions, 1092 1.2 simonb * which could leave more bytes reserved than are reclaimable. 1093 1.2 simonb */ 1094 1.2 simonb if (SIMPLEQ_EMPTY(&wl->wl_entries) && 1095 1.2 simonb (delta >= wl->wl_reserved_bytes)) { 1096 1.2 simonb delta -= wl->wl_reserved_bytes; 1097 1.2 simonb } 1098 1.2 simonb wapbl_advance_tail(wl->wl_circ_size, wl->wl_circ_off, delta, &head, 1099 1.2 simonb &tail); 1100 1.2 simonb KDASSERT(wl->wl_reserved_bytes <= 1101 1.2 simonb wapbl_space_used(wl->wl_circ_size, head, tail)); 1102 1.2 simonb mutex_exit(&wl->wl_mtx); 1103 1.2 simonb 1104 1.2 simonb if (error) 1105 1.2 simonb return error; 1106 1.2 simonb 1107 1.2 simonb if (waitonly) 1108 1.2 simonb return 0; 1109 1.2 simonb 1110 1.2 simonb /* 1111 1.2 simonb * This is where head, tail and delta are unprotected 1112 1.2 simonb * from races against itself or flush. This is ok since 1113 1.2 simonb * we only call this routine from inside flush itself. 1114 1.2 simonb * 1115 1.2 simonb * XXX: how can it race against itself when accessed only 1116 1.2 simonb * from behind the write-locked rwlock? 1117 1.2 simonb */ 1118 1.2 simonb error = wapbl_write_commit(wl, head, tail); 1119 1.2 simonb if (error) 1120 1.2 simonb return error; 1121 1.2 simonb 1122 1.2 simonb wl->wl_head = head; 1123 1.2 simonb wl->wl_tail = tail; 1124 1.2 simonb 1125 1.2 simonb mutex_enter(&wl->wl_mtx); 1126 1.2 simonb KASSERT(wl->wl_reclaimable_bytes >= delta); 1127 1.2 simonb wl->wl_reclaimable_bytes -= delta; 1128 1.2 simonb mutex_exit(&wl->wl_mtx); 1129 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_TRUNCATE, 1130 1.2 simonb ("wapbl_truncate thread %d.%d truncating %zu bytes\n", 1131 1.2 simonb curproc->p_pid, curlwp->l_lid, delta)); 1132 1.2 simonb 1133 1.2 simonb return 0; 1134 1.2 simonb } 1135 1.2 simonb 1136 1.2 simonb /****************************************************************/ 1137 1.2 simonb 1138 1.2 simonb void 1139 1.2 simonb wapbl_biodone(struct buf *bp) 1140 1.2 simonb { 1141 1.2 simonb struct wapbl_entry *we = bp->b_private; 1142 1.2 simonb struct wapbl *wl = we->we_wapbl; 1143 1.2 simonb 1144 1.2 simonb /* 1145 1.2 simonb * Handle possible flushing of buffers after log has been 1146 1.2 simonb * decomissioned. 1147 1.2 simonb */ 1148 1.2 simonb if (!wl) { 1149 1.2 simonb KASSERT(we->we_bufcount > 0); 1150 1.2 simonb we->we_bufcount--; 1151 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1152 1.2 simonb KASSERT(we->we_unsynced_bufbytes >= bp->b_bufsize); 1153 1.2 simonb we->we_unsynced_bufbytes -= bp->b_bufsize; 1154 1.2 simonb #endif 1155 1.2 simonb 1156 1.2 simonb if (we->we_bufcount == 0) { 1157 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1158 1.2 simonb KASSERT(we->we_unsynced_bufbytes == 0); 1159 1.2 simonb #endif 1160 1.2 simonb wapbl_free(we); 1161 1.2 simonb } 1162 1.2 simonb 1163 1.2 simonb brelse(bp, 0); 1164 1.2 simonb return; 1165 1.2 simonb } 1166 1.2 simonb 1167 1.2 simonb #ifdef ohbother 1168 1.2 simonb KDASSERT(bp->b_flags & B_DONE); 1169 1.2 simonb KDASSERT(!(bp->b_flags & B_DELWRI)); 1170 1.2 simonb KDASSERT(bp->b_flags & B_ASYNC); 1171 1.2 simonb KDASSERT(bp->b_flags & B_BUSY); 1172 1.2 simonb KDASSERT(!(bp->b_flags & B_LOCKED)); 1173 1.2 simonb KDASSERT(!(bp->b_flags & B_READ)); 1174 1.2 simonb KDASSERT(!(bp->b_flags & B_INVAL)); 1175 1.2 simonb KDASSERT(!(bp->b_flags & B_NOCACHE)); 1176 1.2 simonb #endif 1177 1.2 simonb 1178 1.2 simonb if (bp->b_error) { 1179 1.2 simonb #ifdef notyet /* Can't currently handle possible dirty buffer reuse */ 1180 1.2 simonb XXXpooka: interfaces not fully updated 1181 1.2 simonb Note: this was not enabled in the original patch 1182 1.2 simonb against netbsd4 either. I don't know if comment 1183 1.2 simonb above is true or not. 1184 1.2 simonb 1185 1.2 simonb /* 1186 1.2 simonb * If an error occurs, report the error and leave the 1187 1.2 simonb * buffer as a delayed write on the LRU queue. 1188 1.2 simonb * restarting the write would likely result in 1189 1.2 simonb * an error spinloop, so let it be done harmlessly 1190 1.2 simonb * by the syncer. 1191 1.2 simonb */ 1192 1.2 simonb bp->b_flags &= ~(B_DONE); 1193 1.2 simonb simple_unlock(&bp->b_interlock); 1194 1.2 simonb 1195 1.2 simonb if (we->we_error == 0) { 1196 1.2 simonb mutex_enter(&wl->wl_mtx); 1197 1.2 simonb wl->wl_error_count++; 1198 1.2 simonb mutex_exit(&wl->wl_mtx); 1199 1.2 simonb cv_broadcast(&wl->wl_reclaimable_cv); 1200 1.2 simonb } 1201 1.2 simonb we->we_error = bp->b_error; 1202 1.2 simonb bp->b_error = 0; 1203 1.2 simonb brelse(bp); 1204 1.2 simonb return; 1205 1.2 simonb #else 1206 1.2 simonb /* For now, just mark the log permanently errored out */ 1207 1.2 simonb 1208 1.2 simonb mutex_enter(&wl->wl_mtx); 1209 1.2 simonb if (wl->wl_error_count == 0) { 1210 1.2 simonb wl->wl_error_count++; 1211 1.2 simonb cv_broadcast(&wl->wl_reclaimable_cv); 1212 1.2 simonb } 1213 1.2 simonb mutex_exit(&wl->wl_mtx); 1214 1.2 simonb #endif 1215 1.2 simonb } 1216 1.2 simonb 1217 1.2 simonb mutex_enter(&wl->wl_mtx); 1218 1.2 simonb 1219 1.2 simonb KASSERT(we->we_bufcount > 0); 1220 1.2 simonb we->we_bufcount--; 1221 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1222 1.2 simonb KASSERT(we->we_unsynced_bufbytes >= bp->b_bufsize); 1223 1.2 simonb we->we_unsynced_bufbytes -= bp->b_bufsize; 1224 1.2 simonb KASSERT(wl->wl_unsynced_bufbytes >= bp->b_bufsize); 1225 1.2 simonb wl->wl_unsynced_bufbytes -= bp->b_bufsize; 1226 1.2 simonb #endif 1227 1.2 simonb 1228 1.2 simonb /* 1229 1.2 simonb * If the current transaction can be reclaimed, start 1230 1.2 simonb * at the beginning and reclaim any consecutive reclaimable 1231 1.2 simonb * transactions. If we successfully reclaim anything, 1232 1.2 simonb * then wakeup anyone waiting for the reclaim. 1233 1.2 simonb */ 1234 1.2 simonb if (we->we_bufcount == 0) { 1235 1.2 simonb size_t delta = 0; 1236 1.2 simonb int errcnt = 0; 1237 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1238 1.2 simonb KDASSERT(we->we_unsynced_bufbytes == 0); 1239 1.2 simonb #endif 1240 1.2 simonb /* 1241 1.2 simonb * clear any posted error, since the buffer it came from 1242 1.2 simonb * has successfully flushed by now 1243 1.2 simonb */ 1244 1.2 simonb while ((we = SIMPLEQ_FIRST(&wl->wl_entries)) && 1245 1.2 simonb (we->we_bufcount == 0)) { 1246 1.2 simonb delta += we->we_reclaimable_bytes; 1247 1.2 simonb if (we->we_error) 1248 1.2 simonb errcnt++; 1249 1.2 simonb SIMPLEQ_REMOVE_HEAD(&wl->wl_entries, we_entries); 1250 1.2 simonb wapbl_free(we); 1251 1.2 simonb } 1252 1.2 simonb 1253 1.2 simonb if (delta) { 1254 1.2 simonb wl->wl_reclaimable_bytes += delta; 1255 1.2 simonb KASSERT(wl->wl_error_count >= errcnt); 1256 1.2 simonb wl->wl_error_count -= errcnt; 1257 1.2 simonb cv_broadcast(&wl->wl_reclaimable_cv); 1258 1.2 simonb } 1259 1.2 simonb } 1260 1.2 simonb 1261 1.2 simonb mutex_exit(&wl->wl_mtx); 1262 1.2 simonb brelse(bp, 0); 1263 1.2 simonb } 1264 1.2 simonb 1265 1.2 simonb /* 1266 1.2 simonb * Write transactions to disk + start I/O for contents 1267 1.2 simonb */ 1268 1.2 simonb int 1269 1.2 simonb wapbl_flush(struct wapbl *wl, int waitfor) 1270 1.2 simonb { 1271 1.2 simonb struct buf *bp; 1272 1.2 simonb struct wapbl_entry *we; 1273 1.2 simonb off_t off; 1274 1.2 simonb off_t head; 1275 1.2 simonb off_t tail; 1276 1.2 simonb size_t delta = 0; 1277 1.2 simonb size_t flushsize; 1278 1.2 simonb size_t reserved; 1279 1.2 simonb int error = 0; 1280 1.2 simonb 1281 1.2 simonb /* 1282 1.2 simonb * Do a quick check to see if a full flush can be skipped 1283 1.2 simonb * This assumes that the flush callback does not need to be called 1284 1.2 simonb * unless there are other outstanding bufs. 1285 1.2 simonb */ 1286 1.2 simonb if (!waitfor) { 1287 1.2 simonb size_t nbufs; 1288 1.2 simonb mutex_enter(&wl->wl_mtx); /* XXX need mutex here to 1289 1.2 simonb protect the KASSERTS */ 1290 1.2 simonb nbufs = wl->wl_bufcount; 1291 1.2 simonb KASSERT((wl->wl_bufcount == 0) == (wl->wl_bufbytes == 0)); 1292 1.2 simonb KASSERT((wl->wl_bufcount == 0) == (wl->wl_bcount == 0)); 1293 1.2 simonb mutex_exit(&wl->wl_mtx); 1294 1.2 simonb if (nbufs == 0) 1295 1.2 simonb return 0; 1296 1.2 simonb } 1297 1.2 simonb 1298 1.2 simonb /* 1299 1.2 simonb * XXX we may consider using LK_UPGRADE here 1300 1.2 simonb * if we want to call flush from inside a transaction 1301 1.2 simonb */ 1302 1.2 simonb rw_enter(&wl->wl_rwlock, RW_WRITER); 1303 1.2 simonb wl->wl_flush(wl->wl_mount, wl->wl_deallocblks, wl->wl_dealloclens, 1304 1.2 simonb wl->wl_dealloccnt); 1305 1.2 simonb 1306 1.2 simonb /* 1307 1.2 simonb * Now that we are fully locked and flushed, 1308 1.2 simonb * do another check for nothing to do. 1309 1.2 simonb */ 1310 1.2 simonb if (wl->wl_bufcount == 0) { 1311 1.2 simonb goto out; 1312 1.2 simonb } 1313 1.2 simonb 1314 1.2 simonb #if 0 1315 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_FLUSH, 1316 1.2 simonb ("wapbl_flush thread %d.%d flushing entries with " 1317 1.2 simonb "bufcount=%zu bufbytes=%zu\n", 1318 1.2 simonb curproc->p_pid, curlwp->l_lid, wl->wl_bufcount, 1319 1.2 simonb wl->wl_bufbytes)); 1320 1.2 simonb #endif 1321 1.2 simonb 1322 1.2 simonb /* Calculate amount of space needed to flush */ 1323 1.2 simonb flushsize = wapbl_transaction_len(wl); 1324 1.2 simonb 1325 1.2 simonb if (flushsize > (wl->wl_circ_size - wl->wl_reserved_bytes)) { 1326 1.2 simonb /* 1327 1.2 simonb * XXX this could be handled more gracefully, perhaps place 1328 1.2 simonb * only a partial transaction in the log and allow the 1329 1.2 simonb * remaining to flush without the protection of the journal. 1330 1.2 simonb */ 1331 1.2 simonb panic("wapbl_flush: current transaction too big to flush\n"); 1332 1.2 simonb } 1333 1.2 simonb 1334 1.2 simonb error = wapbl_truncate(wl, flushsize, 0); 1335 1.2 simonb if (error) 1336 1.2 simonb goto out2; 1337 1.2 simonb 1338 1.2 simonb off = wl->wl_head; 1339 1.2 simonb KASSERT((off == 0) || ((off >= wl->wl_circ_off) && 1340 1.2 simonb (off < wl->wl_circ_off + wl->wl_circ_size))); 1341 1.2 simonb error = wapbl_write_blocks(wl, &off); 1342 1.2 simonb if (error) 1343 1.2 simonb goto out2; 1344 1.2 simonb error = wapbl_write_revocations(wl, &off); 1345 1.2 simonb if (error) 1346 1.2 simonb goto out2; 1347 1.2 simonb error = wapbl_write_inodes(wl, &off); 1348 1.2 simonb if (error) 1349 1.2 simonb goto out2; 1350 1.2 simonb 1351 1.2 simonb reserved = 0; 1352 1.2 simonb if (wl->wl_inohashcnt) 1353 1.2 simonb reserved = wapbl_transaction_inodes_len(wl); 1354 1.2 simonb 1355 1.2 simonb head = wl->wl_head; 1356 1.2 simonb tail = wl->wl_tail; 1357 1.2 simonb 1358 1.2 simonb wapbl_advance_head(wl->wl_circ_size, wl->wl_circ_off, flushsize, 1359 1.2 simonb &head, &tail); 1360 1.2 simonb #ifdef WAPBL_DEBUG 1361 1.2 simonb if (head != off) { 1362 1.2 simonb panic("lost head! head=%"PRIdMAX" tail=%" PRIdMAX 1363 1.2 simonb " off=%"PRIdMAX" flush=%zu\n", 1364 1.2 simonb (intmax_t)head, (intmax_t)tail, (intmax_t)off, 1365 1.2 simonb flushsize); 1366 1.2 simonb } 1367 1.2 simonb #else 1368 1.2 simonb KASSERT(head == off); 1369 1.2 simonb #endif 1370 1.2 simonb 1371 1.2 simonb /* Opportunistically move the tail forward if we can */ 1372 1.2 simonb if (!wapbl_lazy_truncate) { 1373 1.2 simonb mutex_enter(&wl->wl_mtx); 1374 1.2 simonb delta = wl->wl_reclaimable_bytes; 1375 1.2 simonb mutex_exit(&wl->wl_mtx); 1376 1.2 simonb wapbl_advance_tail(wl->wl_circ_size, wl->wl_circ_off, delta, 1377 1.2 simonb &head, &tail); 1378 1.2 simonb } 1379 1.2 simonb 1380 1.2 simonb error = wapbl_write_commit(wl, head, tail); 1381 1.2 simonb if (error) 1382 1.2 simonb goto out2; 1383 1.2 simonb 1384 1.2 simonb /* poolme? or kmemme? */ 1385 1.2 simonb we = wapbl_calloc(1, sizeof(*we)); 1386 1.2 simonb 1387 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1388 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_FLUSH, 1389 1.2 simonb ("wapbl_flush: thread %d.%d head+=%zu tail+=%zu used=%zu" 1390 1.2 simonb " unsynced=%zu" 1391 1.2 simonb "\n\tbufcount=%zu bufbytes=%zu bcount=%zu deallocs=%d " 1392 1.2 simonb "inodes=%d\n", 1393 1.2 simonb curproc->p_pid, curlwp->l_lid, flushsize, delta, 1394 1.2 simonb wapbl_space_used(wl->wl_circ_size, head, tail), 1395 1.2 simonb wl->wl_unsynced_bufbytes, wl->wl_bufcount, 1396 1.2 simonb wl->wl_bufbytes, wl->wl_bcount, wl->wl_dealloccnt, 1397 1.2 simonb wl->wl_inohashcnt)); 1398 1.2 simonb #else 1399 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_FLUSH, 1400 1.2 simonb ("wapbl_flush: thread %d.%d head+=%zu tail+=%zu used=%zu" 1401 1.2 simonb "\n\tbufcount=%zu bufbytes=%zu bcount=%zu deallocs=%d " 1402 1.2 simonb "inodes=%d\n", 1403 1.2 simonb curproc->p_pid, curlwp->l_lid, flushsize, delta, 1404 1.2 simonb wapbl_space_used(wl->wl_circ_size, head, tail), 1405 1.2 simonb wl->wl_bufcount, wl->wl_bufbytes, wl->wl_bcount, 1406 1.2 simonb wl->wl_dealloccnt, wl->wl_inohashcnt)); 1407 1.2 simonb #endif 1408 1.2 simonb 1409 1.2 simonb 1410 1.2 simonb mutex_enter(&bufcache_lock); 1411 1.2 simonb mutex_enter(&wl->wl_mtx); 1412 1.2 simonb 1413 1.2 simonb wl->wl_reserved_bytes = reserved; 1414 1.2 simonb wl->wl_head = head; 1415 1.2 simonb wl->wl_tail = tail; 1416 1.2 simonb KASSERT(wl->wl_reclaimable_bytes >= delta); 1417 1.2 simonb wl->wl_reclaimable_bytes -= delta; 1418 1.2 simonb wl->wl_dealloccnt = 0; 1419 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1420 1.2 simonb wl->wl_unsynced_bufbytes += wl->wl_bufbytes; 1421 1.2 simonb #endif 1422 1.2 simonb 1423 1.2 simonb we->we_wapbl = wl; 1424 1.2 simonb we->we_bufcount = wl->wl_bufcount; 1425 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1426 1.2 simonb we->we_unsynced_bufbytes = wl->wl_bufbytes; 1427 1.2 simonb #endif 1428 1.2 simonb we->we_reclaimable_bytes = flushsize; 1429 1.2 simonb we->we_error = 0; 1430 1.2 simonb SIMPLEQ_INSERT_TAIL(&wl->wl_entries, we, we_entries); 1431 1.2 simonb 1432 1.2 simonb /* 1433 1.2 simonb * this flushes bufs in reverse order than they were queued 1434 1.2 simonb * it shouldn't matter, but if we care we could use TAILQ instead. 1435 1.2 simonb * XXX Note they will get put on the lru queue when they flush 1436 1.2 simonb * so we might actually want to change this to preserve order. 1437 1.2 simonb */ 1438 1.2 simonb while ((bp = LIST_FIRST(&wl->wl_bufs)) != NULL) { 1439 1.2 simonb if (bbusy(bp, 0, 0, &wl->wl_mtx)) { 1440 1.2 simonb continue; 1441 1.2 simonb } 1442 1.2 simonb bp->b_iodone = wapbl_biodone; 1443 1.2 simonb bp->b_private = we; 1444 1.2 simonb bremfree(bp); 1445 1.2 simonb wapbl_remove_buf_locked(wl, bp); 1446 1.2 simonb mutex_exit(&wl->wl_mtx); 1447 1.2 simonb mutex_exit(&bufcache_lock); 1448 1.2 simonb bawrite(bp); 1449 1.2 simonb mutex_enter(&bufcache_lock); 1450 1.2 simonb mutex_enter(&wl->wl_mtx); 1451 1.2 simonb } 1452 1.2 simonb mutex_exit(&wl->wl_mtx); 1453 1.2 simonb mutex_exit(&bufcache_lock); 1454 1.2 simonb 1455 1.2 simonb #if 0 1456 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_FLUSH, 1457 1.2 simonb ("wapbl_flush thread %d.%d done flushing entries...\n", 1458 1.2 simonb curproc->p_pid, curlwp->l_lid)); 1459 1.2 simonb #endif 1460 1.2 simonb 1461 1.2 simonb out: 1462 1.2 simonb 1463 1.2 simonb /* 1464 1.2 simonb * If the waitfor flag is set, don't return until everything is 1465 1.2 simonb * fully flushed and the on disk log is empty. 1466 1.2 simonb */ 1467 1.2 simonb if (waitfor) { 1468 1.2 simonb error = wapbl_truncate(wl, wl->wl_circ_size - 1469 1.2 simonb wl->wl_reserved_bytes, wapbl_lazy_truncate); 1470 1.2 simonb } 1471 1.2 simonb 1472 1.2 simonb out2: 1473 1.2 simonb if (error) { 1474 1.2 simonb wl->wl_flush_abort(wl->wl_mount, wl->wl_deallocblks, 1475 1.2 simonb wl->wl_dealloclens, wl->wl_dealloccnt); 1476 1.2 simonb } 1477 1.2 simonb 1478 1.2 simonb #ifdef WAPBL_DEBUG_PRINT 1479 1.2 simonb if (error) { 1480 1.2 simonb pid_t pid = -1; 1481 1.2 simonb lwpid_t lid = -1; 1482 1.2 simonb if (curproc) 1483 1.2 simonb pid = curproc->p_pid; 1484 1.2 simonb if (curlwp) 1485 1.2 simonb lid = curlwp->l_lid; 1486 1.2 simonb mutex_enter(&wl->wl_mtx); 1487 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1488 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_ERROR, 1489 1.2 simonb ("wapbl_flush: thread %d.%d aborted flush: " 1490 1.2 simonb "error = %d\n" 1491 1.2 simonb "\tbufcount=%zu bufbytes=%zu bcount=%zu " 1492 1.2 simonb "deallocs=%d inodes=%d\n" 1493 1.2 simonb "\terrcnt = %d, reclaimable=%zu reserved=%zu " 1494 1.2 simonb "unsynced=%zu\n", 1495 1.2 simonb pid, lid, error, wl->wl_bufcount, 1496 1.2 simonb wl->wl_bufbytes, wl->wl_bcount, 1497 1.2 simonb wl->wl_dealloccnt, wl->wl_inohashcnt, 1498 1.2 simonb wl->wl_error_count, wl->wl_reclaimable_bytes, 1499 1.2 simonb wl->wl_reserved_bytes, wl->wl_unsynced_bufbytes)); 1500 1.2 simonb SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) { 1501 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_ERROR, 1502 1.2 simonb ("\tentry: bufcount = %zu, reclaimable = %zu, " 1503 1.2 simonb "error = %d, unsynced = %zu\n", 1504 1.2 simonb we->we_bufcount, we->we_reclaimable_bytes, 1505 1.2 simonb we->we_error, we->we_unsynced_bufbytes)); 1506 1.2 simonb } 1507 1.2 simonb #else 1508 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_ERROR, 1509 1.2 simonb ("wapbl_flush: thread %d.%d aborted flush: " 1510 1.2 simonb "error = %d\n" 1511 1.2 simonb "\tbufcount=%zu bufbytes=%zu bcount=%zu " 1512 1.2 simonb "deallocs=%d inodes=%d\n" 1513 1.2 simonb "\terrcnt = %d, reclaimable=%zu reserved=%zu\n", 1514 1.2 simonb pid, lid, error, wl->wl_bufcount, 1515 1.2 simonb wl->wl_bufbytes, wl->wl_bcount, 1516 1.2 simonb wl->wl_dealloccnt, wl->wl_inohashcnt, 1517 1.2 simonb wl->wl_error_count, wl->wl_reclaimable_bytes, 1518 1.2 simonb wl->wl_reserved_bytes)); 1519 1.2 simonb SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) { 1520 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_ERROR, 1521 1.2 simonb ("\tentry: bufcount = %zu, reclaimable = %zu, " 1522 1.2 simonb "error = %d\n", we->we_bufcount, 1523 1.2 simonb we->we_reclaimable_bytes, we->we_error)); 1524 1.2 simonb } 1525 1.2 simonb #endif 1526 1.2 simonb mutex_exit(&wl->wl_mtx); 1527 1.2 simonb } 1528 1.2 simonb #endif 1529 1.2 simonb 1530 1.2 simonb rw_exit(&wl->wl_rwlock); 1531 1.2 simonb return error; 1532 1.2 simonb } 1533 1.2 simonb 1534 1.2 simonb /****************************************************************/ 1535 1.2 simonb 1536 1.2 simonb void 1537 1.2 simonb wapbl_jlock_assert(struct wapbl *wl) 1538 1.2 simonb { 1539 1.2 simonb 1540 1.3.8.1 snj KASSERT(rw_lock_held(&wl->wl_rwlock)); 1541 1.2 simonb } 1542 1.2 simonb 1543 1.2 simonb void 1544 1.2 simonb wapbl_junlock_assert(struct wapbl *wl) 1545 1.2 simonb { 1546 1.2 simonb 1547 1.2 simonb KASSERT(!rw_write_held(&wl->wl_rwlock)); 1548 1.2 simonb } 1549 1.2 simonb 1550 1.2 simonb /****************************************************************/ 1551 1.2 simonb 1552 1.2 simonb /* locks missing */ 1553 1.2 simonb void 1554 1.2 simonb wapbl_print(struct wapbl *wl, 1555 1.2 simonb int full, 1556 1.2 simonb void (*pr)(const char *, ...)) 1557 1.2 simonb { 1558 1.2 simonb struct buf *bp; 1559 1.2 simonb struct wapbl_entry *we; 1560 1.2 simonb (*pr)("wapbl %p", wl); 1561 1.2 simonb (*pr)("\nlogvp = %p, devvp = %p, logpbn = %"PRId64"\n", 1562 1.2 simonb wl->wl_logvp, wl->wl_devvp, wl->wl_logpbn); 1563 1.2 simonb (*pr)("circ = %zu, header = %zu, head = %"PRIdMAX" tail = %"PRIdMAX"\n", 1564 1.2 simonb wl->wl_circ_size, wl->wl_circ_off, 1565 1.2 simonb (intmax_t)wl->wl_head, (intmax_t)wl->wl_tail); 1566 1.2 simonb (*pr)("fs_dev_bshift = %d, log_dev_bshift = %d\n", 1567 1.2 simonb wl->wl_log_dev_bshift, wl->wl_fs_dev_bshift); 1568 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1569 1.2 simonb (*pr)("bufcount = %zu, bufbytes = %zu bcount = %zu reclaimable = %zu " 1570 1.2 simonb "reserved = %zu errcnt = %d unsynced = %zu\n", 1571 1.2 simonb wl->wl_bufcount, wl->wl_bufbytes, wl->wl_bcount, 1572 1.2 simonb wl->wl_reclaimable_bytes, wl->wl_reserved_bytes, 1573 1.2 simonb wl->wl_error_count, wl->wl_unsynced_bufbytes); 1574 1.2 simonb #else 1575 1.2 simonb (*pr)("bufcount = %zu, bufbytes = %zu bcount = %zu reclaimable = %zu " 1576 1.2 simonb "reserved = %zu errcnt = %d\n", wl->wl_bufcount, wl->wl_bufbytes, 1577 1.2 simonb wl->wl_bcount, wl->wl_reclaimable_bytes, wl->wl_reserved_bytes, 1578 1.2 simonb wl->wl_error_count); 1579 1.2 simonb #endif 1580 1.2 simonb (*pr)("\tdealloccnt = %d, dealloclim = %d\n", 1581 1.2 simonb wl->wl_dealloccnt, wl->wl_dealloclim); 1582 1.2 simonb (*pr)("\tinohashcnt = %d, inohashmask = 0x%08x\n", 1583 1.2 simonb wl->wl_inohashcnt, wl->wl_inohashmask); 1584 1.2 simonb (*pr)("entries:\n"); 1585 1.2 simonb SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) { 1586 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1587 1.2 simonb (*pr)("\tbufcount = %zu, reclaimable = %zu, error = %d, " 1588 1.2 simonb "unsynced = %zu\n", 1589 1.2 simonb we->we_bufcount, we->we_reclaimable_bytes, 1590 1.2 simonb we->we_error, we->we_unsynced_bufbytes); 1591 1.2 simonb #else 1592 1.2 simonb (*pr)("\tbufcount = %zu, reclaimable = %zu, error = %d\n", 1593 1.2 simonb we->we_bufcount, we->we_reclaimable_bytes, we->we_error); 1594 1.2 simonb #endif 1595 1.2 simonb } 1596 1.2 simonb if (full) { 1597 1.2 simonb int cnt = 0; 1598 1.2 simonb (*pr)("bufs ="); 1599 1.2 simonb LIST_FOREACH(bp, &wl->wl_bufs, b_wapbllist) { 1600 1.2 simonb if (!LIST_NEXT(bp, b_wapbllist)) { 1601 1.2 simonb (*pr)(" %p", bp); 1602 1.2 simonb } else if ((++cnt % 6) == 0) { 1603 1.2 simonb (*pr)(" %p,\n\t", bp); 1604 1.2 simonb } else { 1605 1.2 simonb (*pr)(" %p,", bp); 1606 1.2 simonb } 1607 1.2 simonb } 1608 1.2 simonb (*pr)("\n"); 1609 1.2 simonb 1610 1.2 simonb (*pr)("dealloced blks = "); 1611 1.2 simonb { 1612 1.2 simonb int i; 1613 1.2 simonb cnt = 0; 1614 1.2 simonb for (i = 0; i < wl->wl_dealloccnt; i++) { 1615 1.2 simonb (*pr)(" %"PRId64":%d,", 1616 1.2 simonb wl->wl_deallocblks[i], 1617 1.2 simonb wl->wl_dealloclens[i]); 1618 1.2 simonb if ((++cnt % 4) == 0) { 1619 1.2 simonb (*pr)("\n\t"); 1620 1.2 simonb } 1621 1.2 simonb } 1622 1.2 simonb } 1623 1.2 simonb (*pr)("\n"); 1624 1.2 simonb 1625 1.2 simonb (*pr)("registered inodes = "); 1626 1.2 simonb { 1627 1.2 simonb int i; 1628 1.2 simonb cnt = 0; 1629 1.2 simonb for (i = 0; i <= wl->wl_inohashmask; i++) { 1630 1.2 simonb struct wapbl_ino_head *wih; 1631 1.2 simonb struct wapbl_ino *wi; 1632 1.2 simonb 1633 1.2 simonb wih = &wl->wl_inohash[i]; 1634 1.2 simonb LIST_FOREACH(wi, wih, wi_hash) { 1635 1.2 simonb if (wi->wi_ino == 0) 1636 1.2 simonb continue; 1637 1.2 simonb (*pr)(" %"PRId32"/0%06"PRIo32",", 1638 1.2 simonb wi->wi_ino, wi->wi_mode); 1639 1.2 simonb if ((++cnt % 4) == 0) { 1640 1.2 simonb (*pr)("\n\t"); 1641 1.2 simonb } 1642 1.2 simonb } 1643 1.2 simonb } 1644 1.2 simonb (*pr)("\n"); 1645 1.2 simonb } 1646 1.2 simonb } 1647 1.2 simonb } 1648 1.2 simonb 1649 1.2 simonb #if defined(WAPBL_DEBUG) || defined(DDB) 1650 1.2 simonb void 1651 1.2 simonb wapbl_dump(struct wapbl *wl) 1652 1.2 simonb { 1653 1.2 simonb #if defined(WAPBL_DEBUG) 1654 1.2 simonb if (!wl) 1655 1.2 simonb wl = wapbl_debug_wl; 1656 1.2 simonb #endif 1657 1.2 simonb if (!wl) 1658 1.2 simonb return; 1659 1.2 simonb wapbl_print(wl, 1, printf); 1660 1.2 simonb } 1661 1.2 simonb #endif 1662 1.2 simonb 1663 1.2 simonb /****************************************************************/ 1664 1.2 simonb 1665 1.2 simonb void 1666 1.2 simonb wapbl_register_deallocation(struct wapbl *wl, daddr_t blk, int len) 1667 1.2 simonb { 1668 1.2 simonb 1669 1.2 simonb wapbl_jlock_assert(wl); 1670 1.2 simonb 1671 1.2 simonb /* XXX should eventually instead tie this into resource estimation */ 1672 1.2 simonb /* XXX this KASSERT needs locking/mutex analysis */ 1673 1.2 simonb KASSERT(wl->wl_dealloccnt < wl->wl_dealloclim); 1674 1.2 simonb wl->wl_deallocblks[wl->wl_dealloccnt] = blk; 1675 1.2 simonb wl->wl_dealloclens[wl->wl_dealloccnt] = len; 1676 1.2 simonb wl->wl_dealloccnt++; 1677 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_ALLOC, 1678 1.2 simonb ("wapbl_register_deallocation: blk=%"PRId64" len=%d\n", blk, len)); 1679 1.2 simonb } 1680 1.2 simonb 1681 1.2 simonb /****************************************************************/ 1682 1.2 simonb 1683 1.2 simonb static void 1684 1.2 simonb wapbl_inodetrk_init(struct wapbl *wl, u_int size) 1685 1.2 simonb { 1686 1.2 simonb 1687 1.2 simonb wl->wl_inohash = hashinit(size, HASH_LIST, true, &wl->wl_inohashmask); 1688 1.2 simonb if (atomic_inc_uint_nv(&wapbl_ino_pool_refcount) == 1) { 1689 1.2 simonb pool_init(&wapbl_ino_pool, sizeof(struct wapbl_ino), 0, 0, 0, 1690 1.2 simonb "wapblinopl", &pool_allocator_nointr, IPL_NONE); 1691 1.2 simonb } 1692 1.2 simonb } 1693 1.2 simonb 1694 1.2 simonb static void 1695 1.2 simonb wapbl_inodetrk_free(struct wapbl *wl) 1696 1.2 simonb { 1697 1.2 simonb 1698 1.2 simonb /* XXX this KASSERT needs locking/mutex analysis */ 1699 1.2 simonb KASSERT(wl->wl_inohashcnt == 0); 1700 1.2 simonb hashdone(wl->wl_inohash, HASH_LIST, wl->wl_inohashmask); 1701 1.2 simonb if (atomic_dec_uint_nv(&wapbl_ino_pool_refcount) == 0) { 1702 1.2 simonb pool_destroy(&wapbl_ino_pool); 1703 1.2 simonb } 1704 1.2 simonb } 1705 1.2 simonb 1706 1.2 simonb static struct wapbl_ino * 1707 1.2 simonb wapbl_inodetrk_get(struct wapbl *wl, ino_t ino) 1708 1.2 simonb { 1709 1.2 simonb struct wapbl_ino_head *wih; 1710 1.2 simonb struct wapbl_ino *wi; 1711 1.2 simonb 1712 1.2 simonb KASSERT(mutex_owned(&wl->wl_mtx)); 1713 1.2 simonb 1714 1.2 simonb wih = &wl->wl_inohash[ino & wl->wl_inohashmask]; 1715 1.2 simonb LIST_FOREACH(wi, wih, wi_hash) { 1716 1.2 simonb if (ino == wi->wi_ino) 1717 1.2 simonb return wi; 1718 1.2 simonb } 1719 1.2 simonb return 0; 1720 1.2 simonb } 1721 1.2 simonb 1722 1.2 simonb void 1723 1.2 simonb wapbl_register_inode(struct wapbl *wl, ino_t ino, mode_t mode) 1724 1.2 simonb { 1725 1.2 simonb struct wapbl_ino_head *wih; 1726 1.2 simonb struct wapbl_ino *wi; 1727 1.2 simonb 1728 1.2 simonb wi = pool_get(&wapbl_ino_pool, PR_WAITOK); 1729 1.2 simonb 1730 1.2 simonb mutex_enter(&wl->wl_mtx); 1731 1.2 simonb if (wapbl_inodetrk_get(wl, ino) == NULL) { 1732 1.2 simonb wi->wi_ino = ino; 1733 1.2 simonb wi->wi_mode = mode; 1734 1.2 simonb wih = &wl->wl_inohash[ino & wl->wl_inohashmask]; 1735 1.2 simonb LIST_INSERT_HEAD(wih, wi, wi_hash); 1736 1.2 simonb wl->wl_inohashcnt++; 1737 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_INODE, 1738 1.2 simonb ("wapbl_register_inode: ino=%"PRId64"\n", ino)); 1739 1.2 simonb mutex_exit(&wl->wl_mtx); 1740 1.2 simonb } else { 1741 1.2 simonb mutex_exit(&wl->wl_mtx); 1742 1.2 simonb pool_put(&wapbl_ino_pool, wi); 1743 1.2 simonb } 1744 1.2 simonb } 1745 1.2 simonb 1746 1.2 simonb void 1747 1.2 simonb wapbl_unregister_inode(struct wapbl *wl, ino_t ino, mode_t mode) 1748 1.2 simonb { 1749 1.2 simonb struct wapbl_ino *wi; 1750 1.2 simonb 1751 1.2 simonb mutex_enter(&wl->wl_mtx); 1752 1.2 simonb wi = wapbl_inodetrk_get(wl, ino); 1753 1.2 simonb if (wi) { 1754 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_INODE, 1755 1.2 simonb ("wapbl_unregister_inode: ino=%"PRId64"\n", ino)); 1756 1.2 simonb KASSERT(wl->wl_inohashcnt > 0); 1757 1.2 simonb wl->wl_inohashcnt--; 1758 1.2 simonb LIST_REMOVE(wi, wi_hash); 1759 1.2 simonb mutex_exit(&wl->wl_mtx); 1760 1.2 simonb 1761 1.2 simonb pool_put(&wapbl_ino_pool, wi); 1762 1.2 simonb } else { 1763 1.2 simonb mutex_exit(&wl->wl_mtx); 1764 1.2 simonb } 1765 1.2 simonb } 1766 1.2 simonb 1767 1.2 simonb /****************************************************************/ 1768 1.2 simonb 1769 1.2 simonb static __inline size_t 1770 1.2 simonb wapbl_transaction_inodes_len(struct wapbl *wl) 1771 1.2 simonb { 1772 1.2 simonb int blocklen = 1<<wl->wl_log_dev_bshift; 1773 1.2 simonb int iph; 1774 1.2 simonb 1775 1.2 simonb /* Calculate number of inodes described in a inodelist header */ 1776 1.2 simonb iph = (blocklen - offsetof(struct wapbl_wc_inodelist, wc_inodes)) / 1777 1.2 simonb sizeof(((struct wapbl_wc_inodelist *)0)->wc_inodes[0]); 1778 1.2 simonb 1779 1.2 simonb KASSERT(iph > 0); 1780 1.2 simonb 1781 1.2 simonb return MAX(1, howmany(wl->wl_inohashcnt, iph))*blocklen; 1782 1.2 simonb } 1783 1.2 simonb 1784 1.2 simonb 1785 1.2 simonb /* Calculate amount of space a transaction will take on disk */ 1786 1.2 simonb static size_t 1787 1.2 simonb wapbl_transaction_len(struct wapbl *wl) 1788 1.2 simonb { 1789 1.2 simonb int blocklen = 1<<wl->wl_log_dev_bshift; 1790 1.2 simonb size_t len; 1791 1.2 simonb int bph; 1792 1.2 simonb 1793 1.2 simonb /* Calculate number of blocks described in a blocklist header */ 1794 1.2 simonb bph = (blocklen - offsetof(struct wapbl_wc_blocklist, wc_blocks)) / 1795 1.2 simonb sizeof(((struct wapbl_wc_blocklist *)0)->wc_blocks[0]); 1796 1.2 simonb 1797 1.2 simonb KASSERT(bph > 0); 1798 1.2 simonb 1799 1.2 simonb len = wl->wl_bcount; 1800 1.2 simonb len += howmany(wl->wl_bufcount, bph)*blocklen; 1801 1.2 simonb len += howmany(wl->wl_dealloccnt, bph)*blocklen; 1802 1.2 simonb len += wapbl_transaction_inodes_len(wl); 1803 1.2 simonb 1804 1.2 simonb return len; 1805 1.2 simonb } 1806 1.2 simonb 1807 1.2 simonb /* 1808 1.2 simonb * Perform commit operation 1809 1.2 simonb * 1810 1.2 simonb * Note that generation number incrementation needs to 1811 1.2 simonb * be protected against racing with other invocations 1812 1.2 simonb * of wapbl_commit. This is ok since this routine 1813 1.2 simonb * is only invoked from wapbl_flush 1814 1.2 simonb */ 1815 1.2 simonb static int 1816 1.2 simonb wapbl_write_commit(struct wapbl *wl, off_t head, off_t tail) 1817 1.2 simonb { 1818 1.2 simonb struct wapbl_wc_header *wc = wl->wl_wc_header; 1819 1.2 simonb struct timespec ts; 1820 1.2 simonb int error; 1821 1.2 simonb int force = 1; 1822 1.2 simonb 1823 1.2 simonb /* XXX Calc checksum here, instead we do this for now */ 1824 1.2 simonb error = VOP_IOCTL(wl->wl_devvp, DIOCCACHESYNC, &force, FWRITE, FSCRED); 1825 1.2 simonb if (error) { 1826 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_ERROR, 1827 1.2 simonb ("wapbl_write_commit: DIOCCACHESYNC on dev 0x%x " 1828 1.2 simonb "returned %d\n", wl->wl_devvp->v_rdev, error)); 1829 1.2 simonb } 1830 1.2 simonb 1831 1.2 simonb wc->wc_head = head; 1832 1.2 simonb wc->wc_tail = tail; 1833 1.2 simonb wc->wc_checksum = 0; 1834 1.2 simonb wc->wc_version = 1; 1835 1.2 simonb getnanotime(&ts); 1836 1.2 simonb wc->wc_time = ts.tv_sec;; 1837 1.2 simonb wc->wc_timensec = ts.tv_nsec; 1838 1.2 simonb 1839 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_WRITE, 1840 1.2 simonb ("wapbl_write_commit: head = %"PRIdMAX "tail = %"PRIdMAX"\n", 1841 1.2 simonb (intmax_t)head, (intmax_t)tail)); 1842 1.2 simonb 1843 1.2 simonb /* 1844 1.2 simonb * XXX if generation will rollover, then first zero 1845 1.2 simonb * over second commit header before trying to write both headers. 1846 1.2 simonb */ 1847 1.2 simonb 1848 1.2 simonb error = wapbl_write(wc, wc->wc_len, wl->wl_devvp, 1849 1.2 simonb wl->wl_logpbn + wc->wc_generation % 2); 1850 1.2 simonb if (error) 1851 1.2 simonb return error; 1852 1.2 simonb 1853 1.2 simonb error = VOP_IOCTL(wl->wl_devvp, DIOCCACHESYNC, &force, FWRITE, FSCRED); 1854 1.2 simonb if (error) { 1855 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_ERROR, 1856 1.2 simonb ("wapbl_write_commit: DIOCCACHESYNC on dev 0x%x " 1857 1.2 simonb "returned %d\n", wl->wl_devvp->v_rdev, error)); 1858 1.2 simonb } 1859 1.2 simonb 1860 1.2 simonb /* 1861 1.2 simonb * If the generation number was zero, write it out a second time. 1862 1.2 simonb * This handles initialization and generation number rollover 1863 1.2 simonb */ 1864 1.2 simonb if (wc->wc_generation++ == 0) { 1865 1.2 simonb error = wapbl_write_commit(wl, head, tail); 1866 1.2 simonb /* 1867 1.2 simonb * This panic should be able to be removed if we do the 1868 1.2 simonb * zero'ing mentioned above, and we are certain to roll 1869 1.2 simonb * back generation number on failure. 1870 1.2 simonb */ 1871 1.2 simonb if (error) 1872 1.2 simonb panic("wapbl_write_commit: error writing duplicate " 1873 1.2 simonb "log header: %d\n", error); 1874 1.2 simonb } 1875 1.2 simonb return 0; 1876 1.2 simonb } 1877 1.2 simonb 1878 1.2 simonb /* Returns new offset value */ 1879 1.2 simonb static int 1880 1.2 simonb wapbl_write_blocks(struct wapbl *wl, off_t *offp) 1881 1.2 simonb { 1882 1.2 simonb struct wapbl_wc_blocklist *wc = 1883 1.2 simonb (struct wapbl_wc_blocklist *)wl->wl_wc_scratch; 1884 1.2 simonb int blocklen = 1<<wl->wl_log_dev_bshift; 1885 1.2 simonb int bph; 1886 1.2 simonb struct buf *bp; 1887 1.2 simonb off_t off = *offp; 1888 1.2 simonb int error; 1889 1.2 simonb 1890 1.2 simonb KASSERT(rw_write_held(&wl->wl_rwlock)); 1891 1.2 simonb 1892 1.2 simonb bph = (blocklen - offsetof(struct wapbl_wc_blocklist, wc_blocks)) / 1893 1.2 simonb sizeof(((struct wapbl_wc_blocklist *)0)->wc_blocks[0]); 1894 1.2 simonb 1895 1.2 simonb bp = LIST_FIRST(&wl->wl_bufs); 1896 1.2 simonb 1897 1.2 simonb while (bp) { 1898 1.2 simonb int cnt; 1899 1.2 simonb struct buf *obp = bp; 1900 1.2 simonb 1901 1.2 simonb KASSERT(bp->b_flags & B_LOCKED); 1902 1.2 simonb 1903 1.2 simonb wc->wc_type = WAPBL_WC_BLOCKS; 1904 1.2 simonb wc->wc_len = blocklen; 1905 1.2 simonb wc->wc_blkcount = 0; 1906 1.2 simonb while (bp && (wc->wc_blkcount < bph)) { 1907 1.2 simonb /* 1908 1.2 simonb * Make sure all the physical block numbers are up to 1909 1.2 simonb * date. If this is not always true on a given 1910 1.2 simonb * filesystem, then VOP_BMAP must be called. We 1911 1.2 simonb * could call VOP_BMAP here, or else in the filesystem 1912 1.2 simonb * specific flush callback, although neither of those 1913 1.2 simonb * solutions allow us to take the vnode lock. If a 1914 1.2 simonb * filesystem requires that we must take the vnode lock 1915 1.2 simonb * to call VOP_BMAP, then we can probably do it in 1916 1.2 simonb * bwrite when the vnode lock should already be held 1917 1.2 simonb * by the invoking code. 1918 1.2 simonb */ 1919 1.2 simonb KASSERT((bp->b_vp->v_type == VBLK) || 1920 1.2 simonb (bp->b_blkno != bp->b_lblkno)); 1921 1.2 simonb KASSERT(bp->b_blkno > 0); 1922 1.2 simonb 1923 1.2 simonb wc->wc_blocks[wc->wc_blkcount].wc_daddr = bp->b_blkno; 1924 1.2 simonb wc->wc_blocks[wc->wc_blkcount].wc_dlen = bp->b_bcount; 1925 1.2 simonb wc->wc_len += bp->b_bcount; 1926 1.2 simonb wc->wc_blkcount++; 1927 1.2 simonb bp = LIST_NEXT(bp, b_wapbllist); 1928 1.2 simonb } 1929 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_WRITE, 1930 1.2 simonb ("wapbl_write_blocks: len = %u off = %"PRIdMAX"\n", 1931 1.2 simonb wc->wc_len, (intmax_t)off)); 1932 1.2 simonb 1933 1.2 simonb error = wapbl_circ_write(wl, wc, blocklen, &off); 1934 1.2 simonb if (error) 1935 1.2 simonb return error; 1936 1.2 simonb bp = obp; 1937 1.2 simonb cnt = 0; 1938 1.2 simonb while (bp && (cnt++ < bph)) { 1939 1.2 simonb error = wapbl_circ_write(wl, bp->b_data, 1940 1.2 simonb bp->b_bcount, &off); 1941 1.2 simonb if (error) 1942 1.2 simonb return error; 1943 1.2 simonb bp = LIST_NEXT(bp, b_wapbllist); 1944 1.2 simonb } 1945 1.2 simonb } 1946 1.2 simonb *offp = off; 1947 1.2 simonb return 0; 1948 1.2 simonb } 1949 1.2 simonb 1950 1.2 simonb static int 1951 1.2 simonb wapbl_write_revocations(struct wapbl *wl, off_t *offp) 1952 1.2 simonb { 1953 1.2 simonb struct wapbl_wc_blocklist *wc = 1954 1.2 simonb (struct wapbl_wc_blocklist *)wl->wl_wc_scratch; 1955 1.2 simonb int i; 1956 1.2 simonb int blocklen = 1<<wl->wl_log_dev_bshift; 1957 1.2 simonb int bph; 1958 1.2 simonb off_t off = *offp; 1959 1.2 simonb int error; 1960 1.2 simonb 1961 1.2 simonb if (wl->wl_dealloccnt == 0) 1962 1.2 simonb return 0; 1963 1.2 simonb 1964 1.2 simonb bph = (blocklen - offsetof(struct wapbl_wc_blocklist, wc_blocks)) / 1965 1.2 simonb sizeof(((struct wapbl_wc_blocklist *)0)->wc_blocks[0]); 1966 1.2 simonb 1967 1.2 simonb i = 0; 1968 1.2 simonb while (i < wl->wl_dealloccnt) { 1969 1.2 simonb wc->wc_type = WAPBL_WC_REVOCATIONS; 1970 1.2 simonb wc->wc_len = blocklen; 1971 1.2 simonb wc->wc_blkcount = 0; 1972 1.2 simonb while ((i < wl->wl_dealloccnt) && (wc->wc_blkcount < bph)) { 1973 1.2 simonb wc->wc_blocks[wc->wc_blkcount].wc_daddr = 1974 1.2 simonb wl->wl_deallocblks[i]; 1975 1.2 simonb wc->wc_blocks[wc->wc_blkcount].wc_dlen = 1976 1.2 simonb wl->wl_dealloclens[i]; 1977 1.2 simonb wc->wc_blkcount++; 1978 1.2 simonb i++; 1979 1.2 simonb } 1980 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_WRITE, 1981 1.2 simonb ("wapbl_write_revocations: len = %u off = %"PRIdMAX"\n", 1982 1.2 simonb wc->wc_len, (intmax_t)off)); 1983 1.2 simonb error = wapbl_circ_write(wl, wc, blocklen, &off); 1984 1.2 simonb if (error) 1985 1.2 simonb return error; 1986 1.2 simonb } 1987 1.2 simonb *offp = off; 1988 1.2 simonb return 0; 1989 1.2 simonb } 1990 1.2 simonb 1991 1.2 simonb static int 1992 1.2 simonb wapbl_write_inodes(struct wapbl *wl, off_t *offp) 1993 1.2 simonb { 1994 1.2 simonb struct wapbl_wc_inodelist *wc = 1995 1.2 simonb (struct wapbl_wc_inodelist *)wl->wl_wc_scratch; 1996 1.2 simonb int i; 1997 1.2 simonb int blocklen = 1<<wl->wl_log_dev_bshift; 1998 1.2 simonb off_t off = *offp; 1999 1.2 simonb int error; 2000 1.2 simonb 2001 1.2 simonb struct wapbl_ino_head *wih; 2002 1.2 simonb struct wapbl_ino *wi; 2003 1.2 simonb int iph; 2004 1.2 simonb 2005 1.2 simonb iph = (blocklen - offsetof(struct wapbl_wc_inodelist, wc_inodes)) / 2006 1.2 simonb sizeof(((struct wapbl_wc_inodelist *)0)->wc_inodes[0]); 2007 1.2 simonb 2008 1.2 simonb i = 0; 2009 1.2 simonb wih = &wl->wl_inohash[0]; 2010 1.2 simonb wi = 0; 2011 1.2 simonb do { 2012 1.2 simonb wc->wc_type = WAPBL_WC_INODES; 2013 1.2 simonb wc->wc_len = blocklen; 2014 1.2 simonb wc->wc_inocnt = 0; 2015 1.2 simonb wc->wc_clear = (i == 0); 2016 1.2 simonb while ((i < wl->wl_inohashcnt) && (wc->wc_inocnt < iph)) { 2017 1.2 simonb while (!wi) { 2018 1.2 simonb KASSERT((wih - &wl->wl_inohash[0]) 2019 1.2 simonb <= wl->wl_inohashmask); 2020 1.2 simonb wi = LIST_FIRST(wih++); 2021 1.2 simonb } 2022 1.2 simonb wc->wc_inodes[wc->wc_inocnt].wc_inumber = wi->wi_ino; 2023 1.2 simonb wc->wc_inodes[wc->wc_inocnt].wc_imode = wi->wi_mode; 2024 1.2 simonb wc->wc_inocnt++; 2025 1.2 simonb i++; 2026 1.2 simonb wi = LIST_NEXT(wi, wi_hash); 2027 1.2 simonb } 2028 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_WRITE, 2029 1.2 simonb ("wapbl_write_inodes: len = %u off = %"PRIdMAX"\n", 2030 1.2 simonb wc->wc_len, (intmax_t)off)); 2031 1.2 simonb error = wapbl_circ_write(wl, wc, blocklen, &off); 2032 1.2 simonb if (error) 2033 1.2 simonb return error; 2034 1.2 simonb } while (i < wl->wl_inohashcnt); 2035 1.2 simonb 2036 1.2 simonb *offp = off; 2037 1.2 simonb return 0; 2038 1.2 simonb } 2039 1.2 simonb 2040 1.2 simonb #endif /* _KERNEL */ 2041 1.2 simonb 2042 1.2 simonb /****************************************************************/ 2043 1.2 simonb 2044 1.2 simonb #ifdef _KERNEL 2045 1.2 simonb static struct pool wapbl_blk_pool; 2046 1.2 simonb static int wapbl_blk_pool_refcount; 2047 1.2 simonb #endif 2048 1.2 simonb struct wapbl_blk { 2049 1.2 simonb LIST_ENTRY(wapbl_blk) wb_hash; 2050 1.2 simonb daddr_t wb_blk; 2051 1.2 simonb off_t wb_off; /* Offset of this block in the log */ 2052 1.2 simonb }; 2053 1.2 simonb #define WAPBL_BLKPOOL_MIN 83 2054 1.2 simonb 2055 1.2 simonb static void 2056 1.2 simonb wapbl_blkhash_init(struct wapbl_replay *wr, u_int size) 2057 1.2 simonb { 2058 1.2 simonb if (size < WAPBL_BLKPOOL_MIN) 2059 1.2 simonb size = WAPBL_BLKPOOL_MIN; 2060 1.2 simonb KASSERT(wr->wr_blkhash == 0); 2061 1.2 simonb #ifdef _KERNEL 2062 1.2 simonb wr->wr_blkhash = hashinit(size, HASH_LIST, true, &wr->wr_blkhashmask); 2063 1.2 simonb if (atomic_inc_uint_nv(&wapbl_blk_pool_refcount) == 1) { 2064 1.2 simonb pool_init(&wapbl_blk_pool, sizeof(struct wapbl_blk), 0, 0, 0, 2065 1.2 simonb "wapblblkpl", &pool_allocator_nointr, IPL_NONE); 2066 1.2 simonb } 2067 1.2 simonb #else /* ! _KERNEL */ 2068 1.2 simonb /* Manually implement hashinit */ 2069 1.2 simonb { 2070 1.2 simonb int i; 2071 1.2 simonb unsigned long hashsize; 2072 1.2 simonb for (hashsize = 1; hashsize < size; hashsize <<= 1) 2073 1.2 simonb continue; 2074 1.2 simonb wr->wr_blkhash = wapbl_malloc(hashsize * sizeof(*wr->wr_blkhash)); 2075 1.2 simonb for (i = 0; i < wr->wr_blkhashmask; i++) 2076 1.2 simonb LIST_INIT(&wr->wr_blkhash[i]); 2077 1.2 simonb wr->wr_blkhashmask = hashsize - 1; 2078 1.2 simonb } 2079 1.2 simonb #endif /* ! _KERNEL */ 2080 1.2 simonb } 2081 1.2 simonb 2082 1.2 simonb static void 2083 1.2 simonb wapbl_blkhash_free(struct wapbl_replay *wr) 2084 1.2 simonb { 2085 1.2 simonb KASSERT(wr->wr_blkhashcnt == 0); 2086 1.2 simonb #ifdef _KERNEL 2087 1.2 simonb hashdone(wr->wr_blkhash, HASH_LIST, wr->wr_blkhashmask); 2088 1.2 simonb if (atomic_dec_uint_nv(&wapbl_blk_pool_refcount) == 0) { 2089 1.2 simonb pool_destroy(&wapbl_blk_pool); 2090 1.2 simonb } 2091 1.2 simonb #else /* ! _KERNEL */ 2092 1.2 simonb wapbl_free(wr->wr_blkhash); 2093 1.2 simonb #endif /* ! _KERNEL */ 2094 1.2 simonb } 2095 1.2 simonb 2096 1.2 simonb static struct wapbl_blk * 2097 1.2 simonb wapbl_blkhash_get(struct wapbl_replay *wr, daddr_t blk) 2098 1.2 simonb { 2099 1.2 simonb struct wapbl_blk_head *wbh; 2100 1.2 simonb struct wapbl_blk *wb; 2101 1.2 simonb wbh = &wr->wr_blkhash[blk & wr->wr_blkhashmask]; 2102 1.2 simonb LIST_FOREACH(wb, wbh, wb_hash) { 2103 1.2 simonb if (blk == wb->wb_blk) 2104 1.2 simonb return wb; 2105 1.2 simonb } 2106 1.2 simonb return 0; 2107 1.2 simonb } 2108 1.2 simonb 2109 1.2 simonb static void 2110 1.2 simonb wapbl_blkhash_ins(struct wapbl_replay *wr, daddr_t blk, off_t off) 2111 1.2 simonb { 2112 1.2 simonb struct wapbl_blk_head *wbh; 2113 1.2 simonb struct wapbl_blk *wb; 2114 1.2 simonb wb = wapbl_blkhash_get(wr, blk); 2115 1.2 simonb if (wb) { 2116 1.2 simonb KASSERT(wb->wb_blk == blk); 2117 1.2 simonb wb->wb_off = off; 2118 1.2 simonb } else { 2119 1.2 simonb #ifdef _KERNEL 2120 1.2 simonb wb = pool_get(&wapbl_blk_pool, PR_WAITOK); 2121 1.2 simonb #else /* ! _KERNEL */ 2122 1.2 simonb wb = wapbl_malloc(sizeof(*wb)); 2123 1.2 simonb #endif /* ! _KERNEL */ 2124 1.2 simonb wb->wb_blk = blk; 2125 1.2 simonb wb->wb_off = off; 2126 1.2 simonb wbh = &wr->wr_blkhash[blk & wr->wr_blkhashmask]; 2127 1.2 simonb LIST_INSERT_HEAD(wbh, wb, wb_hash); 2128 1.2 simonb wr->wr_blkhashcnt++; 2129 1.2 simonb } 2130 1.2 simonb } 2131 1.2 simonb 2132 1.2 simonb static void 2133 1.2 simonb wapbl_blkhash_rem(struct wapbl_replay *wr, daddr_t blk) 2134 1.2 simonb { 2135 1.2 simonb struct wapbl_blk *wb = wapbl_blkhash_get(wr, blk); 2136 1.2 simonb if (wb) { 2137 1.2 simonb KASSERT(wr->wr_blkhashcnt > 0); 2138 1.2 simonb wr->wr_blkhashcnt--; 2139 1.2 simonb LIST_REMOVE(wb, wb_hash); 2140 1.2 simonb #ifdef _KERNEL 2141 1.2 simonb pool_put(&wapbl_blk_pool, wb); 2142 1.2 simonb #else /* ! _KERNEL */ 2143 1.2 simonb wapbl_free(wb); 2144 1.2 simonb #endif /* ! _KERNEL */ 2145 1.2 simonb } 2146 1.2 simonb } 2147 1.2 simonb 2148 1.2 simonb static void 2149 1.2 simonb wapbl_blkhash_clear(struct wapbl_replay *wr) 2150 1.2 simonb { 2151 1.2 simonb int i; 2152 1.2 simonb for (i = 0; i <= wr->wr_blkhashmask; i++) { 2153 1.2 simonb struct wapbl_blk *wb; 2154 1.2 simonb 2155 1.2 simonb while ((wb = LIST_FIRST(&wr->wr_blkhash[i]))) { 2156 1.2 simonb KASSERT(wr->wr_blkhashcnt > 0); 2157 1.2 simonb wr->wr_blkhashcnt--; 2158 1.2 simonb LIST_REMOVE(wb, wb_hash); 2159 1.2 simonb #ifdef _KERNEL 2160 1.2 simonb pool_put(&wapbl_blk_pool, wb); 2161 1.2 simonb #else /* ! _KERNEL */ 2162 1.2 simonb wapbl_free(wb); 2163 1.2 simonb #endif /* ! _KERNEL */ 2164 1.2 simonb } 2165 1.2 simonb } 2166 1.2 simonb KASSERT(wr->wr_blkhashcnt == 0); 2167 1.2 simonb } 2168 1.2 simonb 2169 1.2 simonb /****************************************************************/ 2170 1.2 simonb 2171 1.2 simonb static int 2172 1.2 simonb wapbl_circ_read(struct wapbl_replay *wr, void *data, size_t len, off_t *offp) 2173 1.2 simonb { 2174 1.2 simonb size_t slen; 2175 1.2 simonb struct wapbl_wc_header *wc = &wr->wr_wc_header; 2176 1.2 simonb off_t off = *offp; 2177 1.2 simonb int error; 2178 1.2 simonb 2179 1.2 simonb KASSERT(((len >> wc->wc_log_dev_bshift) << 2180 1.2 simonb wc->wc_log_dev_bshift) == len); 2181 1.2 simonb if (off < wc->wc_circ_off) 2182 1.2 simonb off = wc->wc_circ_off; 2183 1.2 simonb slen = wc->wc_circ_off + wc->wc_circ_size - off; 2184 1.2 simonb if (slen < len) { 2185 1.2 simonb error = wapbl_read(data, slen, wr->wr_devvp, 2186 1.2 simonb wr->wr_logpbn + (off >> wc->wc_log_dev_bshift)); 2187 1.2 simonb if (error) 2188 1.2 simonb return error; 2189 1.2 simonb data = (uint8_t *)data + slen; 2190 1.2 simonb len -= slen; 2191 1.2 simonb off = wc->wc_circ_off; 2192 1.2 simonb } 2193 1.2 simonb error = wapbl_read(data, len, wr->wr_devvp, 2194 1.2 simonb wr->wr_logpbn + (off >> wc->wc_log_dev_bshift)); 2195 1.2 simonb if (error) 2196 1.2 simonb return error; 2197 1.2 simonb off += len; 2198 1.2 simonb if (off >= wc->wc_circ_off + wc->wc_circ_size) 2199 1.2 simonb off = wc->wc_circ_off; 2200 1.2 simonb *offp = off; 2201 1.2 simonb return 0; 2202 1.2 simonb } 2203 1.2 simonb 2204 1.2 simonb static void 2205 1.2 simonb wapbl_circ_advance(struct wapbl_replay *wr, size_t len, off_t *offp) 2206 1.2 simonb { 2207 1.2 simonb size_t slen; 2208 1.2 simonb struct wapbl_wc_header *wc = &wr->wr_wc_header; 2209 1.2 simonb off_t off = *offp; 2210 1.2 simonb 2211 1.2 simonb KASSERT(((len >> wc->wc_log_dev_bshift) << 2212 1.2 simonb wc->wc_log_dev_bshift) == len); 2213 1.2 simonb 2214 1.2 simonb if (off < wc->wc_circ_off) 2215 1.2 simonb off = wc->wc_circ_off; 2216 1.2 simonb slen = wc->wc_circ_off + wc->wc_circ_size - off; 2217 1.2 simonb if (slen < len) { 2218 1.2 simonb len -= slen; 2219 1.2 simonb off = wc->wc_circ_off; 2220 1.2 simonb } 2221 1.2 simonb off += len; 2222 1.2 simonb if (off >= wc->wc_circ_off + wc->wc_circ_size) 2223 1.2 simonb off = wc->wc_circ_off; 2224 1.2 simonb *offp = off; 2225 1.2 simonb } 2226 1.2 simonb 2227 1.2 simonb /****************************************************************/ 2228 1.2 simonb 2229 1.2 simonb int 2230 1.2 simonb wapbl_replay_start(struct wapbl_replay **wrp, struct vnode *vp, 2231 1.2 simonb daddr_t off, size_t count, size_t blksize) 2232 1.2 simonb { 2233 1.2 simonb struct wapbl_replay *wr; 2234 1.2 simonb int error; 2235 1.2 simonb struct vnode *devvp; 2236 1.2 simonb daddr_t logpbn; 2237 1.2 simonb uint8_t *scratch; 2238 1.2 simonb struct wapbl_wc_header *wch; 2239 1.2 simonb struct wapbl_wc_header *wch2; 2240 1.2 simonb /* Use this until we read the actual log header */ 2241 1.2 simonb int log_dev_bshift = DEV_BSHIFT; 2242 1.2 simonb size_t used; 2243 1.2 simonb 2244 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_REPLAY, 2245 1.2 simonb ("wapbl_replay_start: vp=%p off=%"PRId64 " count=%zu blksize=%zu\n", 2246 1.2 simonb vp, off, count, blksize)); 2247 1.2 simonb 2248 1.2 simonb if (off < 0) 2249 1.2 simonb return EINVAL; 2250 1.2 simonb 2251 1.2 simonb if (blksize < DEV_BSIZE) 2252 1.2 simonb return EINVAL; 2253 1.2 simonb if (blksize % DEV_BSIZE) 2254 1.2 simonb return EINVAL; 2255 1.2 simonb 2256 1.2 simonb #ifdef _KERNEL 2257 1.2 simonb #if 0 2258 1.2 simonb /* XXX vp->v_size isn't reliably set for VBLK devices, 2259 1.2 simonb * especially root. However, we might still want to verify 2260 1.2 simonb * that the full load is readable */ 2261 1.2 simonb if ((off + count) * blksize > vp->v_size) 2262 1.2 simonb return EINVAL; 2263 1.2 simonb #endif 2264 1.2 simonb 2265 1.2 simonb if ((error = VOP_BMAP(vp, off, &devvp, &logpbn, 0)) != 0) { 2266 1.2 simonb return error; 2267 1.2 simonb } 2268 1.2 simonb #else /* ! _KERNEL */ 2269 1.2 simonb devvp = vp; 2270 1.2 simonb logpbn = off; 2271 1.2 simonb #endif /* ! _KERNEL */ 2272 1.2 simonb 2273 1.2 simonb scratch = wapbl_malloc(MAXBSIZE); 2274 1.2 simonb 2275 1.2 simonb error = wapbl_read(scratch, 2<<log_dev_bshift, devvp, logpbn); 2276 1.2 simonb if (error) 2277 1.2 simonb goto errout; 2278 1.2 simonb 2279 1.2 simonb wch = (struct wapbl_wc_header *)scratch; 2280 1.2 simonb wch2 = 2281 1.2 simonb (struct wapbl_wc_header *)(scratch + (1<<log_dev_bshift)); 2282 1.2 simonb /* XXX verify checksums and magic numbers */ 2283 1.2 simonb if (wch->wc_type != WAPBL_WC_HEADER) { 2284 1.2 simonb printf("Unrecognized wapbl magic: 0x%08x\n", wch->wc_type); 2285 1.2 simonb error = EFTYPE; 2286 1.2 simonb goto errout; 2287 1.2 simonb } 2288 1.2 simonb 2289 1.2 simonb if (wch2->wc_generation > wch->wc_generation) 2290 1.2 simonb wch = wch2; 2291 1.2 simonb 2292 1.2 simonb wr = wapbl_calloc(1, sizeof(*wr)); 2293 1.2 simonb 2294 1.2 simonb wr->wr_logvp = vp; 2295 1.2 simonb wr->wr_devvp = devvp; 2296 1.2 simonb wr->wr_logpbn = logpbn; 2297 1.2 simonb 2298 1.2 simonb wr->wr_scratch = scratch; 2299 1.2 simonb 2300 1.2 simonb memcpy(&wr->wr_wc_header, wch, sizeof(wr->wr_wc_header)); 2301 1.2 simonb 2302 1.2 simonb used = wapbl_space_used(wch->wc_circ_size, wch->wc_head, wch->wc_tail); 2303 1.2 simonb 2304 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_REPLAY, 2305 1.2 simonb ("wapbl_replay: head=%"PRId64" tail=%"PRId64" off=%"PRId64 2306 1.2 simonb " len=%"PRId64" used=%zu\n", 2307 1.2 simonb wch->wc_head, wch->wc_tail, wch->wc_circ_off, 2308 1.2 simonb wch->wc_circ_size, used)); 2309 1.2 simonb 2310 1.2 simonb wapbl_blkhash_init(wr, (used >> wch->wc_fs_dev_bshift)); 2311 1.2 simonb error = wapbl_replay_prescan(wr); 2312 1.2 simonb if (error) { 2313 1.2 simonb wapbl_replay_stop(wr); 2314 1.2 simonb wapbl_replay_free(wr); 2315 1.2 simonb return error; 2316 1.2 simonb } 2317 1.2 simonb 2318 1.2 simonb error = wapbl_replay_get_inodes(wr); 2319 1.2 simonb if (error) { 2320 1.2 simonb wapbl_replay_stop(wr); 2321 1.2 simonb wapbl_replay_free(wr); 2322 1.2 simonb return error; 2323 1.2 simonb } 2324 1.2 simonb 2325 1.2 simonb *wrp = wr; 2326 1.2 simonb return 0; 2327 1.2 simonb 2328 1.2 simonb errout: 2329 1.2 simonb wapbl_free(scratch); 2330 1.2 simonb return error; 2331 1.2 simonb } 2332 1.2 simonb 2333 1.2 simonb void 2334 1.2 simonb wapbl_replay_stop(struct wapbl_replay *wr) 2335 1.2 simonb { 2336 1.2 simonb 2337 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_REPLAY, ("wapbl_replay_stop called\n")); 2338 1.2 simonb 2339 1.2 simonb KDASSERT(wapbl_replay_isopen(wr)); 2340 1.2 simonb 2341 1.2 simonb wapbl_free(wr->wr_scratch); 2342 1.2 simonb wr->wr_scratch = 0; 2343 1.2 simonb 2344 1.2 simonb wr->wr_logvp = 0; 2345 1.2 simonb 2346 1.2 simonb wapbl_blkhash_clear(wr); 2347 1.2 simonb wapbl_blkhash_free(wr); 2348 1.2 simonb } 2349 1.2 simonb 2350 1.2 simonb void 2351 1.2 simonb wapbl_replay_free(struct wapbl_replay *wr) 2352 1.2 simonb { 2353 1.2 simonb 2354 1.2 simonb KDASSERT(!wapbl_replay_isopen(wr)); 2355 1.2 simonb 2356 1.2 simonb if (wr->wr_inodes) 2357 1.2 simonb wapbl_free(wr->wr_inodes); 2358 1.2 simonb wapbl_free(wr); 2359 1.2 simonb } 2360 1.2 simonb 2361 1.2 simonb int 2362 1.2 simonb wapbl_replay_isopen1(struct wapbl_replay *wr) 2363 1.2 simonb { 2364 1.2 simonb 2365 1.2 simonb return wapbl_replay_isopen(wr); 2366 1.2 simonb } 2367 1.2 simonb 2368 1.2 simonb static int 2369 1.2 simonb wapbl_replay_prescan(struct wapbl_replay *wr) 2370 1.2 simonb { 2371 1.2 simonb off_t off; 2372 1.2 simonb struct wapbl_wc_header *wch = &wr->wr_wc_header; 2373 1.2 simonb int error; 2374 1.2 simonb 2375 1.2 simonb int logblklen = 1<<wch->wc_log_dev_bshift; 2376 1.2 simonb int fsblklen = 1<<wch->wc_fs_dev_bshift; 2377 1.2 simonb 2378 1.2 simonb wapbl_blkhash_clear(wr); 2379 1.2 simonb 2380 1.2 simonb off = wch->wc_tail; 2381 1.2 simonb while (off != wch->wc_head) { 2382 1.2 simonb struct wapbl_wc_null *wcn; 2383 1.2 simonb off_t saveoff = off; 2384 1.2 simonb error = wapbl_circ_read(wr, wr->wr_scratch, logblklen, &off); 2385 1.2 simonb if (error) 2386 1.2 simonb goto errout; 2387 1.2 simonb wcn = (struct wapbl_wc_null *)wr->wr_scratch; 2388 1.2 simonb switch (wcn->wc_type) { 2389 1.2 simonb case WAPBL_WC_BLOCKS: 2390 1.2 simonb { 2391 1.2 simonb struct wapbl_wc_blocklist *wc = 2392 1.2 simonb (struct wapbl_wc_blocklist *)wr->wr_scratch; 2393 1.2 simonb int i; 2394 1.2 simonb for (i = 0; i < wc->wc_blkcount; i++) { 2395 1.2 simonb int j, n; 2396 1.2 simonb /* 2397 1.2 simonb * Enter each physical block into the 2398 1.2 simonb * hashtable independently 2399 1.2 simonb */ 2400 1.2 simonb n = wc->wc_blocks[i].wc_dlen >> 2401 1.2 simonb wch->wc_fs_dev_bshift; 2402 1.2 simonb for (j = 0; j < n; j++) { 2403 1.2 simonb wapbl_blkhash_ins(wr, 2404 1.2 simonb wc->wc_blocks[i].wc_daddr + j, 2405 1.2 simonb off); 2406 1.2 simonb wapbl_circ_advance(wr, 2407 1.2 simonb fsblklen, &off); 2408 1.2 simonb } 2409 1.2 simonb } 2410 1.2 simonb } 2411 1.2 simonb break; 2412 1.2 simonb 2413 1.2 simonb case WAPBL_WC_REVOCATIONS: 2414 1.2 simonb { 2415 1.2 simonb struct wapbl_wc_blocklist *wc = 2416 1.2 simonb (struct wapbl_wc_blocklist *)wr->wr_scratch; 2417 1.2 simonb int i; 2418 1.2 simonb for (i = 0; i < wc->wc_blkcount; i++) { 2419 1.2 simonb int j, n; 2420 1.2 simonb /* 2421 1.2 simonb * Remove any blocks found from the 2422 1.2 simonb * hashtable 2423 1.2 simonb */ 2424 1.2 simonb n = wc->wc_blocks[i].wc_dlen >> 2425 1.2 simonb wch->wc_fs_dev_bshift; 2426 1.2 simonb for (j = 0; j < n; j++) { 2427 1.2 simonb wapbl_blkhash_rem(wr, 2428 1.2 simonb wc->wc_blocks[i].wc_daddr + j); 2429 1.2 simonb } 2430 1.2 simonb } 2431 1.2 simonb } 2432 1.2 simonb break; 2433 1.2 simonb 2434 1.2 simonb case WAPBL_WC_INODES: 2435 1.2 simonb { 2436 1.2 simonb struct wapbl_wc_inodelist *wc = 2437 1.2 simonb (struct wapbl_wc_inodelist *)wr->wr_scratch; 2438 1.2 simonb /* 2439 1.2 simonb * Keep track of where we found this so we 2440 1.2 simonb * can use it later 2441 1.2 simonb */ 2442 1.2 simonb if (wc->wc_clear) { 2443 1.2 simonb wr->wr_inodestail = saveoff; 2444 1.2 simonb wr->wr_inodescnt = 0; 2445 1.2 simonb } 2446 1.2 simonb if (wr->wr_inodestail) 2447 1.2 simonb wr->wr_inodeshead = off; 2448 1.2 simonb wr->wr_inodescnt += wc->wc_inocnt; 2449 1.2 simonb } 2450 1.2 simonb break; 2451 1.2 simonb default: 2452 1.2 simonb printf("Unrecognized wapbl type: 0x%08x\n", 2453 1.2 simonb wcn->wc_type); 2454 1.2 simonb error = EFTYPE; 2455 1.2 simonb goto errout; 2456 1.2 simonb } 2457 1.2 simonb wapbl_circ_advance(wr, wcn->wc_len, &saveoff); 2458 1.2 simonb if (off != saveoff) { 2459 1.2 simonb printf("wapbl_replay: corrupted records\n"); 2460 1.2 simonb error = EFTYPE; 2461 1.2 simonb goto errout; 2462 1.2 simonb } 2463 1.2 simonb } 2464 1.2 simonb return 0; 2465 1.2 simonb 2466 1.2 simonb errout: 2467 1.2 simonb wapbl_blkhash_clear(wr); 2468 1.2 simonb return error; 2469 1.2 simonb } 2470 1.2 simonb 2471 1.2 simonb static int 2472 1.2 simonb wapbl_replay_get_inodes(struct wapbl_replay *wr) 2473 1.2 simonb { 2474 1.2 simonb off_t off; 2475 1.2 simonb struct wapbl_wc_header *wch = &wr->wr_wc_header; 2476 1.2 simonb int logblklen = 1<<wch->wc_log_dev_bshift; 2477 1.2 simonb int cnt= 0; 2478 1.2 simonb 2479 1.2 simonb KDASSERT(wapbl_replay_isopen(wr)); 2480 1.2 simonb 2481 1.2 simonb if (wr->wr_inodescnt == 0) 2482 1.2 simonb return 0; 2483 1.2 simonb 2484 1.2 simonb KASSERT(!wr->wr_inodes); 2485 1.2 simonb 2486 1.2 simonb wr->wr_inodes = wapbl_malloc(wr->wr_inodescnt*sizeof(wr->wr_inodes[0])); 2487 1.2 simonb 2488 1.2 simonb off = wr->wr_inodestail; 2489 1.2 simonb 2490 1.2 simonb while (off != wr->wr_inodeshead) { 2491 1.2 simonb struct wapbl_wc_null *wcn; 2492 1.2 simonb int error; 2493 1.2 simonb off_t saveoff = off; 2494 1.2 simonb error = wapbl_circ_read(wr, wr->wr_scratch, logblklen, &off); 2495 1.2 simonb if (error) { 2496 1.2 simonb wapbl_free(wr->wr_inodes); 2497 1.2 simonb wr->wr_inodes = 0; 2498 1.2 simonb return error; 2499 1.2 simonb } 2500 1.2 simonb wcn = (struct wapbl_wc_null *)wr->wr_scratch; 2501 1.2 simonb switch (wcn->wc_type) { 2502 1.2 simonb case WAPBL_WC_BLOCKS: 2503 1.2 simonb case WAPBL_WC_REVOCATIONS: 2504 1.2 simonb break; 2505 1.2 simonb case WAPBL_WC_INODES: 2506 1.2 simonb { 2507 1.2 simonb struct wapbl_wc_inodelist *wc = 2508 1.2 simonb (struct wapbl_wc_inodelist *)wr->wr_scratch; 2509 1.2 simonb /* 2510 1.2 simonb * Keep track of where we found this so we 2511 1.2 simonb * can use it later 2512 1.2 simonb */ 2513 1.2 simonb if (wc->wc_clear) { 2514 1.2 simonb cnt = 0; 2515 1.2 simonb } 2516 1.2 simonb /* This memcpy assumes that wr_inodes is 2517 1.2 simonb * laid out the same as wc_inodes. */ 2518 1.2 simonb memcpy(&wr->wr_inodes[cnt], wc->wc_inodes, 2519 1.2 simonb wc->wc_inocnt*sizeof(wc->wc_inodes[0])); 2520 1.2 simonb cnt += wc->wc_inocnt; 2521 1.2 simonb } 2522 1.2 simonb break; 2523 1.2 simonb default: 2524 1.2 simonb KASSERT(0); 2525 1.2 simonb } 2526 1.2 simonb off = saveoff; 2527 1.2 simonb wapbl_circ_advance(wr, wcn->wc_len, &off); 2528 1.2 simonb } 2529 1.2 simonb KASSERT(cnt == wr->wr_inodescnt); 2530 1.2 simonb return 0; 2531 1.2 simonb } 2532 1.2 simonb 2533 1.2 simonb #ifdef DEBUG 2534 1.2 simonb int 2535 1.2 simonb wapbl_replay_verify(struct wapbl_replay *wr, struct vnode *fsdevvp) 2536 1.2 simonb { 2537 1.2 simonb off_t off; 2538 1.2 simonb struct wapbl_wc_header *wch = &wr->wr_wc_header; 2539 1.2 simonb int mismatchcnt = 0; 2540 1.2 simonb int logblklen = 1<<wch->wc_log_dev_bshift; 2541 1.2 simonb int fsblklen = 1<<wch->wc_fs_dev_bshift; 2542 1.2 simonb void *scratch1 = wapbl_malloc(MAXBSIZE); 2543 1.2 simonb void *scratch2 = wapbl_malloc(MAXBSIZE); 2544 1.2 simonb int error = 0; 2545 1.2 simonb 2546 1.2 simonb KDASSERT(wapbl_replay_isopen(wr)); 2547 1.2 simonb 2548 1.2 simonb off = wch->wc_tail; 2549 1.2 simonb while (off != wch->wc_head) { 2550 1.2 simonb struct wapbl_wc_null *wcn; 2551 1.2 simonb #ifdef DEBUG 2552 1.2 simonb off_t saveoff = off; 2553 1.2 simonb #endif 2554 1.2 simonb error = wapbl_circ_read(wr, wr->wr_scratch, logblklen, &off); 2555 1.2 simonb if (error) 2556 1.2 simonb goto out; 2557 1.2 simonb wcn = (struct wapbl_wc_null *)wr->wr_scratch; 2558 1.2 simonb switch (wcn->wc_type) { 2559 1.2 simonb case WAPBL_WC_BLOCKS: 2560 1.2 simonb { 2561 1.2 simonb struct wapbl_wc_blocklist *wc = 2562 1.2 simonb (struct wapbl_wc_blocklist *)wr->wr_scratch; 2563 1.2 simonb int i; 2564 1.2 simonb for (i = 0; i < wc->wc_blkcount; i++) { 2565 1.2 simonb int foundcnt = 0; 2566 1.2 simonb int dirtycnt = 0; 2567 1.2 simonb int j, n; 2568 1.2 simonb /* 2569 1.2 simonb * Check each physical block into the 2570 1.2 simonb * hashtable independently 2571 1.2 simonb */ 2572 1.2 simonb n = wc->wc_blocks[i].wc_dlen >> 2573 1.2 simonb wch->wc_fs_dev_bshift; 2574 1.2 simonb for (j = 0; j < n; j++) { 2575 1.2 simonb struct wapbl_blk *wb = 2576 1.2 simonb wapbl_blkhash_get(wr, 2577 1.2 simonb wc->wc_blocks[i].wc_daddr + j); 2578 1.2 simonb if (wb && (wb->wb_off == off)) { 2579 1.2 simonb foundcnt++; 2580 1.2 simonb error = 2581 1.2 simonb wapbl_circ_read(wr, 2582 1.2 simonb scratch1, fsblklen, 2583 1.2 simonb &off); 2584 1.2 simonb if (error) 2585 1.2 simonb goto out; 2586 1.2 simonb error = 2587 1.2 simonb wapbl_read(scratch2, 2588 1.2 simonb fsblklen, fsdevvp, 2589 1.2 simonb wb->wb_blk); 2590 1.2 simonb if (error) 2591 1.2 simonb goto out; 2592 1.2 simonb if (memcmp(scratch1, 2593 1.2 simonb scratch2, 2594 1.2 simonb fsblklen)) { 2595 1.2 simonb printf( 2596 1.2 simonb "wapbl_verify: mismatch block %"PRId64" at off %"PRIdMAX"\n", 2597 1.2 simonb wb->wb_blk, (intmax_t)off); 2598 1.2 simonb dirtycnt++; 2599 1.2 simonb mismatchcnt++; 2600 1.2 simonb } 2601 1.2 simonb } else { 2602 1.2 simonb wapbl_circ_advance(wr, 2603 1.2 simonb fsblklen, &off); 2604 1.2 simonb } 2605 1.2 simonb } 2606 1.2 simonb #if 0 2607 1.2 simonb /* 2608 1.2 simonb * If all of the blocks in an entry 2609 1.2 simonb * are clean, then remove all of its 2610 1.2 simonb * blocks from the hashtable since they 2611 1.2 simonb * never will need replay. 2612 1.2 simonb */ 2613 1.2 simonb if ((foundcnt != 0) && 2614 1.2 simonb (dirtycnt == 0)) { 2615 1.2 simonb off = saveoff; 2616 1.2 simonb wapbl_circ_advance(wr, 2617 1.2 simonb logblklen, &off); 2618 1.2 simonb for (j = 0; j < n; j++) { 2619 1.2 simonb struct wapbl_blk *wb = 2620 1.2 simonb wapbl_blkhash_get(wr, 2621 1.2 simonb wc->wc_blocks[i].wc_daddr + j); 2622 1.2 simonb if (wb && 2623 1.2 simonb (wb->wb_off == off)) { 2624 1.2 simonb wapbl_blkhash_rem(wr, wb->wb_blk); 2625 1.2 simonb } 2626 1.2 simonb wapbl_circ_advance(wr, 2627 1.2 simonb fsblklen, &off); 2628 1.2 simonb } 2629 1.2 simonb } 2630 1.2 simonb #endif 2631 1.2 simonb } 2632 1.2 simonb } 2633 1.2 simonb break; 2634 1.2 simonb case WAPBL_WC_REVOCATIONS: 2635 1.2 simonb case WAPBL_WC_INODES: 2636 1.2 simonb break; 2637 1.2 simonb default: 2638 1.2 simonb KASSERT(0); 2639 1.2 simonb } 2640 1.2 simonb #ifdef DEBUG 2641 1.2 simonb wapbl_circ_advance(wr, wcn->wc_len, &saveoff); 2642 1.2 simonb KASSERT(off == saveoff); 2643 1.2 simonb #endif 2644 1.2 simonb } 2645 1.2 simonb out: 2646 1.2 simonb wapbl_free(scratch1); 2647 1.2 simonb wapbl_free(scratch2); 2648 1.2 simonb if (!error && mismatchcnt) 2649 1.2 simonb error = EFTYPE; 2650 1.2 simonb return error; 2651 1.2 simonb } 2652 1.2 simonb #endif 2653 1.2 simonb 2654 1.2 simonb int 2655 1.2 simonb wapbl_replay_write(struct wapbl_replay *wr, struct vnode *fsdevvp) 2656 1.2 simonb { 2657 1.2 simonb off_t off; 2658 1.2 simonb struct wapbl_wc_header *wch = &wr->wr_wc_header; 2659 1.2 simonb int logblklen = 1<<wch->wc_log_dev_bshift; 2660 1.2 simonb int fsblklen = 1<<wch->wc_fs_dev_bshift; 2661 1.2 simonb void *scratch1 = wapbl_malloc(MAXBSIZE); 2662 1.2 simonb int error = 0; 2663 1.2 simonb 2664 1.2 simonb KDASSERT(wapbl_replay_isopen(wr)); 2665 1.2 simonb 2666 1.2 simonb /* 2667 1.2 simonb * This parses the journal for replay, although it could 2668 1.2 simonb * just as easily walk the hashtable instead. 2669 1.2 simonb */ 2670 1.2 simonb 2671 1.2 simonb off = wch->wc_tail; 2672 1.2 simonb while (off != wch->wc_head) { 2673 1.2 simonb struct wapbl_wc_null *wcn; 2674 1.2 simonb #ifdef DEBUG 2675 1.2 simonb off_t saveoff = off; 2676 1.2 simonb #endif 2677 1.2 simonb error = wapbl_circ_read(wr, wr->wr_scratch, logblklen, &off); 2678 1.2 simonb if (error) 2679 1.2 simonb goto out; 2680 1.2 simonb wcn = (struct wapbl_wc_null *)wr->wr_scratch; 2681 1.2 simonb switch (wcn->wc_type) { 2682 1.2 simonb case WAPBL_WC_BLOCKS: 2683 1.2 simonb { 2684 1.2 simonb struct wapbl_wc_blocklist *wc = 2685 1.2 simonb (struct wapbl_wc_blocklist *)wr->wr_scratch; 2686 1.2 simonb int i; 2687 1.2 simonb for (i = 0; i < wc->wc_blkcount; i++) { 2688 1.2 simonb int j, n; 2689 1.2 simonb /* 2690 1.2 simonb * Check each physical block against 2691 1.2 simonb * the hashtable independently 2692 1.2 simonb */ 2693 1.2 simonb n = wc->wc_blocks[i].wc_dlen >> 2694 1.2 simonb wch->wc_fs_dev_bshift; 2695 1.2 simonb for (j = 0; j < n; j++) { 2696 1.2 simonb struct wapbl_blk *wb = 2697 1.2 simonb wapbl_blkhash_get(wr, 2698 1.2 simonb wc->wc_blocks[i].wc_daddr + j); 2699 1.2 simonb if (wb && (wb->wb_off == off)) { 2700 1.2 simonb error = wapbl_circ_read( 2701 1.2 simonb wr, scratch1, 2702 1.2 simonb fsblklen, &off); 2703 1.2 simonb if (error) 2704 1.2 simonb goto out; 2705 1.2 simonb error = 2706 1.2 simonb wapbl_write(scratch1, 2707 1.2 simonb fsblklen, fsdevvp, 2708 1.2 simonb wb->wb_blk); 2709 1.2 simonb if (error) 2710 1.2 simonb goto out; 2711 1.2 simonb } else { 2712 1.2 simonb wapbl_circ_advance(wr, 2713 1.2 simonb fsblklen, &off); 2714 1.2 simonb } 2715 1.2 simonb } 2716 1.2 simonb } 2717 1.2 simonb } 2718 1.2 simonb break; 2719 1.2 simonb case WAPBL_WC_REVOCATIONS: 2720 1.2 simonb case WAPBL_WC_INODES: 2721 1.2 simonb break; 2722 1.2 simonb default: 2723 1.2 simonb KASSERT(0); 2724 1.2 simonb } 2725 1.2 simonb #ifdef DEBUG 2726 1.2 simonb wapbl_circ_advance(wr, wcn->wc_len, &saveoff); 2727 1.2 simonb KASSERT(off == saveoff); 2728 1.2 simonb #endif 2729 1.2 simonb } 2730 1.2 simonb out: 2731 1.2 simonb wapbl_free(scratch1); 2732 1.2 simonb return error; 2733 1.2 simonb } 2734 1.2 simonb 2735 1.2 simonb int 2736 1.2 simonb wapbl_replay_read(struct wapbl_replay *wr, void *data, daddr_t blk, long len) 2737 1.2 simonb { 2738 1.2 simonb struct wapbl_wc_header *wch = &wr->wr_wc_header; 2739 1.2 simonb int fsblklen = 1<<wch->wc_fs_dev_bshift; 2740 1.2 simonb 2741 1.2 simonb KDASSERT(wapbl_replay_isopen(wr)); 2742 1.2 simonb 2743 1.2 simonb KASSERT((len % fsblklen) == 0); 2744 1.2 simonb 2745 1.2 simonb while (len != 0) { 2746 1.2 simonb struct wapbl_blk *wb = wapbl_blkhash_get(wr, blk); 2747 1.2 simonb if (wb) { 2748 1.2 simonb off_t off = wb->wb_off; 2749 1.2 simonb int error; 2750 1.2 simonb error = wapbl_circ_read(wr, data, fsblklen, &off); 2751 1.2 simonb if (error) 2752 1.2 simonb return error; 2753 1.2 simonb } 2754 1.2 simonb data = (uint8_t *)data + fsblklen; 2755 1.2 simonb len -= fsblklen; 2756 1.2 simonb blk++; 2757 1.2 simonb } 2758 1.2 simonb return 0; 2759 1.2 simonb } 2760
Indexes created Fri Oct 03 02:09:57 GMT 2025