vfs_wapbl.c revision 1.3.8.2.2.2
1 1.3.8.2.2.2 riz /* $NetBSD: vfs_wapbl.c,v 1.3.8.2.2.2 2011/03/07 04:09:55 riz 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 1.3.8.2.2.2 riz __KERNEL_RCSID(0, "$NetBSD: vfs_wapbl.c,v 1.3.8.2.2.2 2011/03/07 04:09:55 riz 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 1.3.8.2.2.1 riz * lm = rwlock held writing or mutex held 99 1.2 simonb * u = unlocked access ok 100 1.2 simonb * b = bufcache_lock held 101 1.2 simonb */ 102 1.2 simonb struct wapbl { 103 1.2 simonb struct vnode *wl_logvp; /* r: log here */ 104 1.2 simonb struct vnode *wl_devvp; /* r: log on this device */ 105 1.2 simonb struct mount *wl_mount; /* r: mountpoint wl is associated with */ 106 1.2 simonb daddr_t wl_logpbn; /* r: Physical block number of start of log */ 107 1.2 simonb int wl_log_dev_bshift; /* r: logarithm of device block size of log 108 1.2 simonb device */ 109 1.2 simonb int wl_fs_dev_bshift; /* r: logarithm of device block size of 110 1.2 simonb filesystem device */ 111 1.2 simonb 112 1.3 yamt unsigned wl_lock_count; /* m: Count of transactions in progress */ 113 1.2 simonb 114 1.2 simonb size_t wl_circ_size; /* r: Number of bytes in buffer of log */ 115 1.2 simonb size_t wl_circ_off; /* r: Number of bytes reserved at start */ 116 1.2 simonb 117 1.2 simonb size_t wl_bufcount_max; /* r: Number of buffers reserved for log */ 118 1.2 simonb size_t wl_bufbytes_max; /* r: Number of buf bytes reserved for log */ 119 1.2 simonb 120 1.2 simonb off_t wl_head; /* l: Byte offset of log head */ 121 1.2 simonb off_t wl_tail; /* l: Byte offset of log tail */ 122 1.2 simonb /* 123 1.2 simonb * head == tail == 0 means log is empty 124 1.2 simonb * head == tail != 0 means log is full 125 1.2 simonb * see assertions in wapbl_advance() for other boundary conditions. 126 1.2 simonb * only truncate moves the tail, except when flush sets it to 127 1.2 simonb * wl_header_size only flush moves the head, except when truncate 128 1.2 simonb * sets it to 0. 129 1.2 simonb */ 130 1.2 simonb 131 1.2 simonb struct wapbl_wc_header *wl_wc_header; /* l */ 132 1.2 simonb void *wl_wc_scratch; /* l: scratch space (XXX: por que?!?) */ 133 1.2 simonb 134 1.2 simonb kmutex_t wl_mtx; /* u: short-term lock */ 135 1.2 simonb krwlock_t wl_rwlock; /* u: File system transaction lock */ 136 1.2 simonb 137 1.2 simonb /* 138 1.2 simonb * Must be held while accessing 139 1.2 simonb * wl_count or wl_bufs or head or tail 140 1.2 simonb */ 141 1.2 simonb 142 1.2 simonb /* 143 1.2 simonb * Callback called from within the flush routine to flush any extra 144 1.2 simonb * bits. Note that flush may be skipped without calling this if 145 1.2 simonb * there are no outstanding buffers in the transaction. 146 1.2 simonb */ 147 1.2 simonb wapbl_flush_fn_t wl_flush; /* r */ 148 1.2 simonb wapbl_flush_fn_t wl_flush_abort;/* r */ 149 1.2 simonb 150 1.2 simonb size_t wl_bufbytes; /* m: Byte count of pages in wl_bufs */ 151 1.2 simonb size_t wl_bufcount; /* m: Count of buffers in wl_bufs */ 152 1.2 simonb size_t wl_bcount; /* m: Total bcount of wl_bufs */ 153 1.2 simonb 154 1.2 simonb LIST_HEAD(, buf) wl_bufs; /* m: Buffers in current transaction */ 155 1.2 simonb 156 1.2 simonb kcondvar_t wl_reclaimable_cv; /* m (obviously) */ 157 1.2 simonb size_t wl_reclaimable_bytes; /* m: Amount of space available for 158 1.2 simonb reclamation by truncate */ 159 1.2 simonb int wl_error_count; /* m: # of wl_entries with errors */ 160 1.2 simonb size_t wl_reserved_bytes; /* never truncate log smaller than this */ 161 1.2 simonb 162 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 163 1.2 simonb size_t wl_unsynced_bufbytes; /* Byte count of unsynced buffers */ 164 1.2 simonb #endif 165 1.2 simonb 166 1.3.8.2.2.1 riz daddr_t *wl_deallocblks;/* lm: address of block */ 167 1.3.8.2.2.1 riz int *wl_dealloclens; /* lm: size of block */ 168 1.3.8.2.2.1 riz int wl_dealloccnt; /* lm: total count */ 169 1.2 simonb int wl_dealloclim; /* l: max count */ 170 1.2 simonb 171 1.2 simonb /* hashtable of inode numbers for allocated but unlinked inodes */ 172 1.2 simonb /* synch ??? */ 173 1.2 simonb LIST_HEAD(wapbl_ino_head, wapbl_ino) *wl_inohash; 174 1.2 simonb u_long wl_inohashmask; 175 1.2 simonb int wl_inohashcnt; 176 1.2 simonb 177 1.2 simonb SIMPLEQ_HEAD(, wapbl_entry) wl_entries; /* On disk transaction 178 1.2 simonb accounting */ 179 1.2 simonb }; 180 1.2 simonb 181 1.2 simonb #ifdef WAPBL_DEBUG_PRINT 182 1.2 simonb int wapbl_debug_print = WAPBL_DEBUG_PRINT; 183 1.2 simonb #endif 184 1.2 simonb 185 1.2 simonb /****************************************************************/ 186 1.2 simonb #ifdef _KERNEL 187 1.2 simonb 188 1.2 simonb #ifdef WAPBL_DEBUG 189 1.2 simonb struct wapbl *wapbl_debug_wl; 190 1.2 simonb #endif 191 1.2 simonb 192 1.2 simonb static int wapbl_write_commit(struct wapbl *wl, off_t head, off_t tail); 193 1.2 simonb static int wapbl_write_blocks(struct wapbl *wl, off_t *offp); 194 1.2 simonb static int wapbl_write_revocations(struct wapbl *wl, off_t *offp); 195 1.2 simonb static int wapbl_write_inodes(struct wapbl *wl, off_t *offp); 196 1.2 simonb #endif /* _KERNEL */ 197 1.2 simonb 198 1.2 simonb static int wapbl_replay_prescan(struct wapbl_replay *wr); 199 1.2 simonb static int wapbl_replay_get_inodes(struct wapbl_replay *wr); 200 1.2 simonb 201 1.2 simonb static __inline size_t wapbl_space_free(size_t avail, off_t head, 202 1.2 simonb off_t tail); 203 1.2 simonb static __inline size_t wapbl_space_used(size_t avail, off_t head, 204 1.2 simonb off_t tail); 205 1.2 simonb 206 1.2 simonb #ifdef _KERNEL 207 1.2 simonb 208 1.2 simonb #define WAPBL_INODETRK_SIZE 83 209 1.2 simonb static int wapbl_ino_pool_refcount; 210 1.2 simonb static struct pool wapbl_ino_pool; 211 1.2 simonb struct wapbl_ino { 212 1.2 simonb LIST_ENTRY(wapbl_ino) wi_hash; 213 1.2 simonb ino_t wi_ino; 214 1.2 simonb mode_t wi_mode; 215 1.2 simonb }; 216 1.2 simonb 217 1.2 simonb static void wapbl_inodetrk_init(struct wapbl *wl, u_int size); 218 1.2 simonb static void wapbl_inodetrk_free(struct wapbl *wl); 219 1.2 simonb static struct wapbl_ino *wapbl_inodetrk_get(struct wapbl *wl, ino_t ino); 220 1.2 simonb 221 1.2 simonb static size_t wapbl_transaction_len(struct wapbl *wl); 222 1.2 simonb static __inline size_t wapbl_transaction_inodes_len(struct wapbl *wl); 223 1.2 simonb 224 1.2 simonb /* 225 1.2 simonb * This is useful for debugging. If set, the log will 226 1.2 simonb * only be truncated when necessary. 227 1.2 simonb */ 228 1.2 simonb int wapbl_lazy_truncate = 0; 229 1.2 simonb 230 1.2 simonb struct wapbl_ops wapbl_ops = { 231 1.2 simonb .wo_wapbl_discard = wapbl_discard, 232 1.2 simonb .wo_wapbl_replay_isopen = wapbl_replay_isopen1, 233 1.2 simonb .wo_wapbl_replay_read = wapbl_replay_read, 234 1.2 simonb .wo_wapbl_add_buf = wapbl_add_buf, 235 1.2 simonb .wo_wapbl_remove_buf = wapbl_remove_buf, 236 1.2 simonb .wo_wapbl_resize_buf = wapbl_resize_buf, 237 1.2 simonb .wo_wapbl_begin = wapbl_begin, 238 1.2 simonb .wo_wapbl_end = wapbl_end, 239 1.2 simonb .wo_wapbl_junlock_assert= wapbl_junlock_assert, 240 1.2 simonb 241 1.2 simonb /* XXX: the following is only used to say "this is a wapbl buf" */ 242 1.2 simonb .wo_wapbl_biodone = wapbl_biodone, 243 1.2 simonb }; 244 1.2 simonb 245 1.2 simonb void 246 1.2 simonb wapbl_init() 247 1.2 simonb { 248 1.2 simonb 249 1.2 simonb malloc_type_attach(M_WAPBL); 250 1.2 simonb } 251 1.2 simonb 252 1.2 simonb int 253 1.2 simonb wapbl_start(struct wapbl ** wlp, struct mount *mp, struct vnode *vp, 254 1.2 simonb daddr_t off, size_t count, size_t blksize, struct wapbl_replay *wr, 255 1.2 simonb wapbl_flush_fn_t flushfn, wapbl_flush_fn_t flushabortfn) 256 1.2 simonb { 257 1.2 simonb struct wapbl *wl; 258 1.2 simonb struct vnode *devvp; 259 1.2 simonb daddr_t logpbn; 260 1.2 simonb int error; 261 1.2 simonb int log_dev_bshift = DEV_BSHIFT; 262 1.2 simonb int fs_dev_bshift = DEV_BSHIFT; 263 1.2 simonb int run; 264 1.2 simonb 265 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_OPEN, ("wapbl_start: vp=%p off=%" PRId64 266 1.2 simonb " count=%zu blksize=%zu\n", vp, off, count, blksize)); 267 1.2 simonb 268 1.2 simonb if (log_dev_bshift > fs_dev_bshift) { 269 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_OPEN, 270 1.2 simonb ("wapbl: log device's block size cannot be larger " 271 1.2 simonb "than filesystem's\n")); 272 1.2 simonb /* 273 1.2 simonb * Not currently implemented, although it could be if 274 1.2 simonb * needed someday. 275 1.2 simonb */ 276 1.2 simonb return ENOSYS; 277 1.2 simonb } 278 1.2 simonb 279 1.2 simonb if (off < 0) 280 1.2 simonb return EINVAL; 281 1.2 simonb 282 1.2 simonb if (blksize < DEV_BSIZE) 283 1.2 simonb return EINVAL; 284 1.2 simonb if (blksize % DEV_BSIZE) 285 1.2 simonb return EINVAL; 286 1.2 simonb 287 1.2 simonb /* XXXTODO: verify that the full load is writable */ 288 1.2 simonb 289 1.2 simonb /* 290 1.2 simonb * XXX check for minimum log size 291 1.2 simonb * minimum is governed by minimum amount of space 292 1.2 simonb * to complete a transaction. (probably truncate) 293 1.2 simonb */ 294 1.2 simonb /* XXX for now pick something minimal */ 295 1.2 simonb if ((count * blksize) < MAXPHYS) { 296 1.2 simonb return ENOSPC; 297 1.2 simonb } 298 1.2 simonb 299 1.2 simonb if ((error = VOP_BMAP(vp, off, &devvp, &logpbn, &run)) != 0) { 300 1.2 simonb return error; 301 1.2 simonb } 302 1.2 simonb 303 1.2 simonb wl = wapbl_calloc(1, sizeof(*wl)); 304 1.2 simonb rw_init(&wl->wl_rwlock); 305 1.2 simonb mutex_init(&wl->wl_mtx, MUTEX_DEFAULT, IPL_NONE); 306 1.2 simonb cv_init(&wl->wl_reclaimable_cv, "wapblrec"); 307 1.2 simonb LIST_INIT(&wl->wl_bufs); 308 1.2 simonb SIMPLEQ_INIT(&wl->wl_entries); 309 1.2 simonb 310 1.2 simonb wl->wl_logvp = vp; 311 1.2 simonb wl->wl_devvp = devvp; 312 1.2 simonb wl->wl_mount = mp; 313 1.2 simonb wl->wl_logpbn = logpbn; 314 1.2 simonb wl->wl_log_dev_bshift = log_dev_bshift; 315 1.2 simonb wl->wl_fs_dev_bshift = fs_dev_bshift; 316 1.2 simonb 317 1.2 simonb wl->wl_flush = flushfn; 318 1.2 simonb wl->wl_flush_abort = flushabortfn; 319 1.2 simonb 320 1.2 simonb /* Reserve two log device blocks for the commit headers */ 321 1.2 simonb wl->wl_circ_off = 2<<wl->wl_log_dev_bshift; 322 1.2 simonb wl->wl_circ_size = ((count * blksize) - wl->wl_circ_off); 323 1.2 simonb /* truncate the log usage to a multiple of log_dev_bshift */ 324 1.2 simonb wl->wl_circ_size >>= wl->wl_log_dev_bshift; 325 1.2 simonb wl->wl_circ_size <<= wl->wl_log_dev_bshift; 326 1.2 simonb 327 1.2 simonb /* 328 1.2 simonb * wl_bufbytes_max limits the size of the in memory transaction space. 329 1.2 simonb * - Since buffers are allocated and accounted for in units of 330 1.2 simonb * PAGE_SIZE it is required to be a multiple of PAGE_SIZE 331 1.2 simonb * (i.e. 1<<PAGE_SHIFT) 332 1.2 simonb * - Since the log device has to be written in units of 333 1.2 simonb * 1<<wl_log_dev_bshift it is required to be a mulitple of 334 1.2 simonb * 1<<wl_log_dev_bshift. 335 1.2 simonb * - Since filesystem will provide data in units of 1<<wl_fs_dev_bshift, 336 1.2 simonb * it is convenient to be a multiple of 1<<wl_fs_dev_bshift. 337 1.2 simonb * Therefore it must be multiple of the least common multiple of those 338 1.2 simonb * three quantities. Fortunately, all of those quantities are 339 1.2 simonb * guaranteed to be a power of two, and the least common multiple of 340 1.2 simonb * a set of numbers which are all powers of two is simply the maximum 341 1.2 simonb * of those numbers. Finally, the maximum logarithm of a power of two 342 1.2 simonb * is the same as the log of the maximum power of two. So we can do 343 1.2 simonb * the following operations to size wl_bufbytes_max: 344 1.2 simonb */ 345 1.2 simonb 346 1.2 simonb /* XXX fix actual number of pages reserved per filesystem. */ 347 1.2 simonb wl->wl_bufbytes_max = MIN(wl->wl_circ_size, buf_memcalc() / 2); 348 1.2 simonb 349 1.2 simonb /* Round wl_bufbytes_max to the largest power of two constraint */ 350 1.2 simonb wl->wl_bufbytes_max >>= PAGE_SHIFT; 351 1.2 simonb wl->wl_bufbytes_max <<= PAGE_SHIFT; 352 1.2 simonb wl->wl_bufbytes_max >>= wl->wl_log_dev_bshift; 353 1.2 simonb wl->wl_bufbytes_max <<= wl->wl_log_dev_bshift; 354 1.2 simonb wl->wl_bufbytes_max >>= wl->wl_fs_dev_bshift; 355 1.2 simonb wl->wl_bufbytes_max <<= wl->wl_fs_dev_bshift; 356 1.2 simonb 357 1.2 simonb /* XXX maybe use filesystem fragment size instead of 1024 */ 358 1.2 simonb /* XXX fix actual number of buffers reserved per filesystem. */ 359 1.2 simonb wl->wl_bufcount_max = (nbuf / 2) * 1024; 360 1.2 simonb 361 1.2 simonb /* XXX tie this into resource estimation */ 362 1.2 simonb wl->wl_dealloclim = 2 * btodb(wl->wl_bufbytes_max); 363 1.2 simonb 364 1.2 simonb #if WAPBL_UVM_ALLOC 365 1.2 simonb wl->wl_deallocblks = (void *) uvm_km_zalloc(kernel_map, 366 1.2 simonb round_page(sizeof(*wl->wl_deallocblks) * wl->wl_dealloclim)); 367 1.2 simonb KASSERT(wl->wl_deallocblks != NULL); 368 1.2 simonb wl->wl_dealloclens = (void *) uvm_km_zalloc(kernel_map, 369 1.2 simonb round_page(sizeof(*wl->wl_dealloclens) * wl->wl_dealloclim)); 370 1.2 simonb KASSERT(wl->wl_dealloclens != NULL); 371 1.2 simonb #else 372 1.2 simonb wl->wl_deallocblks = wapbl_malloc(sizeof(*wl->wl_deallocblks) * 373 1.2 simonb wl->wl_dealloclim); 374 1.2 simonb wl->wl_dealloclens = wapbl_malloc(sizeof(*wl->wl_dealloclens) * 375 1.2 simonb wl->wl_dealloclim); 376 1.2 simonb #endif 377 1.2 simonb 378 1.2 simonb wapbl_inodetrk_init(wl, WAPBL_INODETRK_SIZE); 379 1.2 simonb 380 1.2 simonb /* Initialize the commit header */ 381 1.2 simonb { 382 1.2 simonb struct wapbl_wc_header *wc; 383 1.2 simonb size_t len = 1<<wl->wl_log_dev_bshift; 384 1.2 simonb wc = wapbl_calloc(1, len); 385 1.2 simonb wc->wc_type = WAPBL_WC_HEADER; 386 1.2 simonb wc->wc_len = len; 387 1.2 simonb wc->wc_circ_off = wl->wl_circ_off; 388 1.2 simonb wc->wc_circ_size = wl->wl_circ_size; 389 1.2 simonb /* XXX wc->wc_fsid */ 390 1.2 simonb wc->wc_log_dev_bshift = wl->wl_log_dev_bshift; 391 1.2 simonb wc->wc_fs_dev_bshift = wl->wl_fs_dev_bshift; 392 1.2 simonb wl->wl_wc_header = wc; 393 1.2 simonb wl->wl_wc_scratch = wapbl_malloc(len); 394 1.2 simonb } 395 1.2 simonb 396 1.2 simonb /* 397 1.2 simonb * if there was an existing set of unlinked but 398 1.2 simonb * allocated inodes, preserve it in the new 399 1.2 simonb * log. 400 1.2 simonb */ 401 1.2 simonb if (wr && wr->wr_inodescnt) { 402 1.2 simonb int i; 403 1.2 simonb 404 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_REPLAY, 405 1.2 simonb ("wapbl_start: reusing log with %d inodes\n", 406 1.2 simonb wr->wr_inodescnt)); 407 1.2 simonb 408 1.2 simonb /* 409 1.2 simonb * Its only valid to reuse the replay log if its 410 1.2 simonb * the same as the new log we just opened. 411 1.2 simonb */ 412 1.2 simonb KDASSERT(!wapbl_replay_isopen(wr)); 413 1.2 simonb KASSERT(devvp->v_rdev == wr->wr_devvp->v_rdev); 414 1.2 simonb KASSERT(logpbn == wr->wr_logpbn); 415 1.2 simonb KASSERT(wl->wl_circ_size == wr->wr_wc_header.wc_circ_size); 416 1.2 simonb KASSERT(wl->wl_circ_off == wr->wr_wc_header.wc_circ_off); 417 1.2 simonb KASSERT(wl->wl_log_dev_bshift == 418 1.2 simonb wr->wr_wc_header.wc_log_dev_bshift); 419 1.2 simonb KASSERT(wl->wl_fs_dev_bshift == 420 1.2 simonb wr->wr_wc_header.wc_fs_dev_bshift); 421 1.2 simonb 422 1.2 simonb wl->wl_wc_header->wc_generation = 423 1.2 simonb wr->wr_wc_header.wc_generation + 1; 424 1.2 simonb 425 1.2 simonb for (i = 0; i < wr->wr_inodescnt; i++) 426 1.2 simonb wapbl_register_inode(wl, wr->wr_inodes[i].wr_inumber, 427 1.2 simonb wr->wr_inodes[i].wr_imode); 428 1.2 simonb 429 1.2 simonb /* Make sure new transaction won't overwrite old inodes list */ 430 1.2 simonb KDASSERT(wapbl_transaction_len(wl) <= 431 1.2 simonb wapbl_space_free(wl->wl_circ_size, wr->wr_inodeshead, 432 1.2 simonb wr->wr_inodestail)); 433 1.2 simonb 434 1.2 simonb wl->wl_head = wl->wl_tail = wr->wr_inodeshead; 435 1.2 simonb wl->wl_reclaimable_bytes = wl->wl_reserved_bytes = 436 1.2 simonb wapbl_transaction_len(wl); 437 1.2 simonb 438 1.2 simonb error = wapbl_write_inodes(wl, &wl->wl_head); 439 1.2 simonb if (error) 440 1.2 simonb goto errout; 441 1.2 simonb 442 1.2 simonb KASSERT(wl->wl_head != wl->wl_tail); 443 1.2 simonb KASSERT(wl->wl_head != 0); 444 1.2 simonb } 445 1.2 simonb 446 1.2 simonb error = wapbl_write_commit(wl, wl->wl_head, wl->wl_tail); 447 1.2 simonb if (error) { 448 1.2 simonb goto errout; 449 1.2 simonb } 450 1.2 simonb 451 1.2 simonb *wlp = wl; 452 1.2 simonb #if defined(WAPBL_DEBUG) 453 1.2 simonb wapbl_debug_wl = wl; 454 1.2 simonb #endif 455 1.2 simonb 456 1.2 simonb return 0; 457 1.2 simonb errout: 458 1.2 simonb wapbl_discard(wl); 459 1.2 simonb wapbl_free(wl->wl_wc_scratch); 460 1.2 simonb wapbl_free(wl->wl_wc_header); 461 1.2 simonb #if WAPBL_UVM_ALLOC 462 1.2 simonb uvm_km_free_wakeup(kernel_map, (vaddr_t) wl->wl_deallocblks, 463 1.2 simonb round_page(sizeof(*wl->wl_deallocblks * 464 1.2 simonb wl->wl_dealloclim))); 465 1.2 simonb uvm_km_free_wakeup(kernel_map, (vaddr_t) wl->wl_dealloclens, 466 1.2 simonb round_page(sizeof(*wl->wl_dealloclens * 467 1.2 simonb wl->wl_dealloclim))); 468 1.2 simonb #else 469 1.2 simonb wapbl_free(wl->wl_deallocblks); 470 1.2 simonb wapbl_free(wl->wl_dealloclens); 471 1.2 simonb #endif 472 1.2 simonb wapbl_inodetrk_free(wl); 473 1.2 simonb wapbl_free(wl); 474 1.2 simonb 475 1.2 simonb return error; 476 1.2 simonb } 477 1.2 simonb 478 1.2 simonb /* 479 1.2 simonb * Like wapbl_flush, only discards the transaction 480 1.2 simonb * completely 481 1.2 simonb */ 482 1.2 simonb 483 1.2 simonb void 484 1.2 simonb wapbl_discard(struct wapbl *wl) 485 1.2 simonb { 486 1.2 simonb struct wapbl_entry *we; 487 1.2 simonb struct buf *bp; 488 1.2 simonb int i; 489 1.2 simonb 490 1.2 simonb /* 491 1.2 simonb * XXX we may consider using upgrade here 492 1.2 simonb * if we want to call flush from inside a transaction 493 1.2 simonb */ 494 1.2 simonb rw_enter(&wl->wl_rwlock, RW_WRITER); 495 1.2 simonb wl->wl_flush(wl->wl_mount, wl->wl_deallocblks, wl->wl_dealloclens, 496 1.2 simonb wl->wl_dealloccnt); 497 1.2 simonb 498 1.2 simonb #ifdef WAPBL_DEBUG_PRINT 499 1.2 simonb { 500 1.2 simonb struct wapbl_entry *we; 501 1.2 simonb pid_t pid = -1; 502 1.2 simonb lwpid_t lid = -1; 503 1.2 simonb if (curproc) 504 1.2 simonb pid = curproc->p_pid; 505 1.2 simonb if (curlwp) 506 1.2 simonb lid = curlwp->l_lid; 507 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 508 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_DISCARD, 509 1.2 simonb ("wapbl_discard: thread %d.%d discarding " 510 1.2 simonb "transaction\n" 511 1.2 simonb "\tbufcount=%zu bufbytes=%zu bcount=%zu " 512 1.2 simonb "deallocs=%d inodes=%d\n" 513 1.2 simonb "\terrcnt = %u, reclaimable=%zu reserved=%zu " 514 1.2 simonb "unsynced=%zu\n", 515 1.2 simonb pid, lid, wl->wl_bufcount, wl->wl_bufbytes, 516 1.2 simonb wl->wl_bcount, wl->wl_dealloccnt, 517 1.2 simonb wl->wl_inohashcnt, wl->wl_error_count, 518 1.2 simonb wl->wl_reclaimable_bytes, wl->wl_reserved_bytes, 519 1.2 simonb wl->wl_unsynced_bufbytes)); 520 1.2 simonb SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) { 521 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_DISCARD, 522 1.2 simonb ("\tentry: bufcount = %zu, reclaimable = %zu, " 523 1.2 simonb "error = %d, unsynced = %zu\n", 524 1.2 simonb we->we_bufcount, we->we_reclaimable_bytes, 525 1.2 simonb we->we_error, we->we_unsynced_bufbytes)); 526 1.2 simonb } 527 1.2 simonb #else /* !WAPBL_DEBUG_BUFBYTES */ 528 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_DISCARD, 529 1.2 simonb ("wapbl_discard: thread %d.%d discarding transaction\n" 530 1.2 simonb "\tbufcount=%zu bufbytes=%zu bcount=%zu " 531 1.2 simonb "deallocs=%d inodes=%d\n" 532 1.2 simonb "\terrcnt = %u, reclaimable=%zu reserved=%zu\n", 533 1.2 simonb pid, lid, wl->wl_bufcount, wl->wl_bufbytes, 534 1.2 simonb wl->wl_bcount, wl->wl_dealloccnt, 535 1.2 simonb wl->wl_inohashcnt, wl->wl_error_count, 536 1.2 simonb wl->wl_reclaimable_bytes, wl->wl_reserved_bytes)); 537 1.2 simonb SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) { 538 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_DISCARD, 539 1.2 simonb ("\tentry: bufcount = %zu, reclaimable = %zu, " 540 1.2 simonb "error = %d\n", 541 1.2 simonb we->we_bufcount, we->we_reclaimable_bytes, 542 1.2 simonb we->we_error)); 543 1.2 simonb } 544 1.2 simonb #endif /* !WAPBL_DEBUG_BUFBYTES */ 545 1.2 simonb } 546 1.2 simonb #endif /* WAPBL_DEBUG_PRINT */ 547 1.2 simonb 548 1.2 simonb for (i = 0; i <= wl->wl_inohashmask; i++) { 549 1.2 simonb struct wapbl_ino_head *wih; 550 1.2 simonb struct wapbl_ino *wi; 551 1.2 simonb 552 1.2 simonb wih = &wl->wl_inohash[i]; 553 1.2 simonb while ((wi = LIST_FIRST(wih)) != NULL) { 554 1.2 simonb LIST_REMOVE(wi, wi_hash); 555 1.2 simonb pool_put(&wapbl_ino_pool, wi); 556 1.2 simonb KASSERT(wl->wl_inohashcnt > 0); 557 1.2 simonb wl->wl_inohashcnt--; 558 1.2 simonb } 559 1.2 simonb } 560 1.2 simonb 561 1.2 simonb /* 562 1.2 simonb * clean buffer list 563 1.2 simonb */ 564 1.2 simonb mutex_enter(&bufcache_lock); 565 1.2 simonb mutex_enter(&wl->wl_mtx); 566 1.2 simonb while ((bp = LIST_FIRST(&wl->wl_bufs)) != NULL) { 567 1.2 simonb if (bbusy(bp, 0, 0, &wl->wl_mtx) == 0) { 568 1.2 simonb /* 569 1.2 simonb * The buffer will be unlocked and 570 1.2 simonb * removed from the transaction in brelse 571 1.2 simonb */ 572 1.2 simonb mutex_exit(&wl->wl_mtx); 573 1.2 simonb brelsel(bp, 0); 574 1.2 simonb mutex_enter(&wl->wl_mtx); 575 1.2 simonb } 576 1.2 simonb } 577 1.2 simonb mutex_exit(&wl->wl_mtx); 578 1.2 simonb mutex_exit(&bufcache_lock); 579 1.2 simonb 580 1.2 simonb /* 581 1.2 simonb * Remove references to this wl from wl_entries, free any which 582 1.2 simonb * no longer have buffers, others will be freed in wapbl_biodone 583 1.2 simonb * when they no longer have any buffers. 584 1.2 simonb */ 585 1.2 simonb while ((we = SIMPLEQ_FIRST(&wl->wl_entries)) != NULL) { 586 1.2 simonb SIMPLEQ_REMOVE_HEAD(&wl->wl_entries, we_entries); 587 1.2 simonb /* XXX should we be accumulating wl_error_count 588 1.2 simonb * and increasing reclaimable bytes ? */ 589 1.2 simonb we->we_wapbl = NULL; 590 1.2 simonb if (we->we_bufcount == 0) { 591 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 592 1.2 simonb KASSERT(we->we_unsynced_bufbytes == 0); 593 1.2 simonb #endif 594 1.2 simonb wapbl_free(we); 595 1.2 simonb } 596 1.2 simonb } 597 1.2 simonb 598 1.2 simonb /* Discard list of deallocs */ 599 1.2 simonb wl->wl_dealloccnt = 0; 600 1.2 simonb /* XXX should we clear wl_reserved_bytes? */ 601 1.2 simonb 602 1.2 simonb KASSERT(wl->wl_bufbytes == 0); 603 1.2 simonb KASSERT(wl->wl_bcount == 0); 604 1.2 simonb KASSERT(wl->wl_bufcount == 0); 605 1.2 simonb KASSERT(LIST_EMPTY(&wl->wl_bufs)); 606 1.2 simonb KASSERT(SIMPLEQ_EMPTY(&wl->wl_entries)); 607 1.2 simonb KASSERT(wl->wl_inohashcnt == 0); 608 1.2 simonb 609 1.2 simonb rw_exit(&wl->wl_rwlock); 610 1.2 simonb } 611 1.2 simonb 612 1.2 simonb int 613 1.2 simonb wapbl_stop(struct wapbl *wl, int force) 614 1.2 simonb { 615 1.2 simonb struct vnode *vp; 616 1.2 simonb int error; 617 1.2 simonb 618 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_OPEN, ("wapbl_stop called\n")); 619 1.2 simonb error = wapbl_flush(wl, 1); 620 1.2 simonb if (error) { 621 1.2 simonb if (force) 622 1.2 simonb wapbl_discard(wl); 623 1.2 simonb else 624 1.2 simonb return error; 625 1.2 simonb } 626 1.2 simonb 627 1.2 simonb /* Unlinked inodes persist after a flush */ 628 1.2 simonb if (wl->wl_inohashcnt) { 629 1.2 simonb if (force) { 630 1.2 simonb wapbl_discard(wl); 631 1.2 simonb } else { 632 1.2 simonb return EBUSY; 633 1.2 simonb } 634 1.2 simonb } 635 1.2 simonb 636 1.2 simonb KASSERT(wl->wl_bufbytes == 0); 637 1.2 simonb KASSERT(wl->wl_bcount == 0); 638 1.2 simonb KASSERT(wl->wl_bufcount == 0); 639 1.2 simonb KASSERT(LIST_EMPTY(&wl->wl_bufs)); 640 1.2 simonb KASSERT(wl->wl_dealloccnt == 0); 641 1.2 simonb KASSERT(SIMPLEQ_EMPTY(&wl->wl_entries)); 642 1.2 simonb KASSERT(wl->wl_inohashcnt == 0); 643 1.2 simonb 644 1.2 simonb vp = wl->wl_logvp; 645 1.2 simonb 646 1.2 simonb wapbl_free(wl->wl_wc_scratch); 647 1.2 simonb wapbl_free(wl->wl_wc_header); 648 1.2 simonb #if WAPBL_UVM_ALLOC 649 1.2 simonb uvm_km_free_wakeup(kernel_map, (vaddr_t) wl->wl_deallocblks, 650 1.2 simonb round_page(sizeof(*wl->wl_deallocblks * 651 1.2 simonb wl->wl_dealloclim))); 652 1.2 simonb uvm_km_free_wakeup(kernel_map, (vaddr_t) wl->wl_dealloclens, 653 1.2 simonb round_page(sizeof(*wl->wl_dealloclens * 654 1.2 simonb wl->wl_dealloclim))); 655 1.2 simonb #else 656 1.2 simonb wapbl_free(wl->wl_deallocblks); 657 1.2 simonb wapbl_free(wl->wl_dealloclens); 658 1.2 simonb #endif 659 1.2 simonb wapbl_inodetrk_free(wl); 660 1.2 simonb 661 1.2 simonb cv_destroy(&wl->wl_reclaimable_cv); 662 1.2 simonb mutex_destroy(&wl->wl_mtx); 663 1.2 simonb rw_destroy(&wl->wl_rwlock); 664 1.2 simonb wapbl_free(wl); 665 1.2 simonb 666 1.2 simonb return 0; 667 1.2 simonb } 668 1.2 simonb 669 1.2 simonb static int 670 1.2 simonb wapbl_doio(void *data, size_t len, struct vnode *devvp, daddr_t pbn, int flags) 671 1.2 simonb { 672 1.2 simonb struct pstats *pstats = curlwp->l_proc->p_stats; 673 1.2 simonb struct buf *bp; 674 1.2 simonb int error; 675 1.2 simonb 676 1.2 simonb KASSERT((flags & ~(B_WRITE | B_READ)) == 0); 677 1.2 simonb KASSERT(devvp->v_type == VBLK); 678 1.2 simonb 679 1.2 simonb if ((flags & (B_WRITE | B_READ)) == B_WRITE) { 680 1.2 simonb mutex_enter(&devvp->v_interlock); 681 1.2 simonb devvp->v_numoutput++; 682 1.2 simonb mutex_exit(&devvp->v_interlock); 683 1.2 simonb pstats->p_ru.ru_oublock++; 684 1.2 simonb } else { 685 1.2 simonb pstats->p_ru.ru_inblock++; 686 1.2 simonb } 687 1.2 simonb 688 1.2 simonb bp = getiobuf(devvp, true); 689 1.2 simonb bp->b_flags = flags; 690 1.2 simonb bp->b_cflags = BC_BUSY; /* silly & dubious */ 691 1.2 simonb bp->b_dev = devvp->v_rdev; 692 1.2 simonb bp->b_data = data; 693 1.2 simonb bp->b_bufsize = bp->b_resid = bp->b_bcount = len; 694 1.2 simonb bp->b_blkno = pbn; 695 1.2 simonb 696 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_IO, 697 1.2 simonb ("wapbl_doio: %s %d bytes at block %"PRId64" on dev 0x%x\n", 698 1.2 simonb BUF_ISWRITE(bp) ? "write" : "read", bp->b_bcount, 699 1.2 simonb bp->b_blkno, bp->b_dev)); 700 1.2 simonb 701 1.2 simonb VOP_STRATEGY(devvp, bp); 702 1.2 simonb 703 1.2 simonb error = biowait(bp); 704 1.2 simonb putiobuf(bp); 705 1.2 simonb 706 1.2 simonb if (error) { 707 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_ERROR, 708 1.2 simonb ("wapbl_doio: %s %zu bytes at block %" PRId64 709 1.2 simonb " on dev 0x%x failed with error %d\n", 710 1.2 simonb (((flags & (B_WRITE | B_READ)) == B_WRITE) ? 711 1.2 simonb "write" : "read"), 712 1.2 simonb len, pbn, devvp->v_rdev, error)); 713 1.2 simonb } 714 1.2 simonb 715 1.2 simonb return error; 716 1.2 simonb } 717 1.2 simonb 718 1.2 simonb int 719 1.2 simonb wapbl_write(void *data, size_t len, struct vnode *devvp, daddr_t pbn) 720 1.2 simonb { 721 1.2 simonb 722 1.2 simonb return wapbl_doio(data, len, devvp, pbn, B_WRITE); 723 1.2 simonb } 724 1.2 simonb 725 1.2 simonb int 726 1.2 simonb wapbl_read(void *data, size_t len, struct vnode *devvp, daddr_t pbn) 727 1.2 simonb { 728 1.2 simonb 729 1.2 simonb return wapbl_doio(data, len, devvp, pbn, B_READ); 730 1.2 simonb } 731 1.2 simonb 732 1.2 simonb /* 733 1.2 simonb * Off is byte offset returns new offset for next write 734 1.2 simonb * handles log wraparound 735 1.2 simonb */ 736 1.2 simonb static int 737 1.2 simonb wapbl_circ_write(struct wapbl *wl, void *data, size_t len, off_t *offp) 738 1.2 simonb { 739 1.2 simonb size_t slen; 740 1.2 simonb off_t off = *offp; 741 1.2 simonb int error; 742 1.2 simonb 743 1.2 simonb KDASSERT(((len >> wl->wl_log_dev_bshift) << 744 1.2 simonb wl->wl_log_dev_bshift) == len); 745 1.2 simonb 746 1.2 simonb if (off < wl->wl_circ_off) 747 1.2 simonb off = wl->wl_circ_off; 748 1.2 simonb slen = wl->wl_circ_off + wl->wl_circ_size - off; 749 1.2 simonb if (slen < len) { 750 1.2 simonb error = wapbl_write(data, slen, wl->wl_devvp, 751 1.2 simonb wl->wl_logpbn + (off >> wl->wl_log_dev_bshift)); 752 1.2 simonb if (error) 753 1.2 simonb return error; 754 1.2 simonb data = (uint8_t *)data + slen; 755 1.2 simonb len -= slen; 756 1.2 simonb off = wl->wl_circ_off; 757 1.2 simonb } 758 1.2 simonb error = wapbl_write(data, len, wl->wl_devvp, 759 1.2 simonb wl->wl_logpbn + (off >> wl->wl_log_dev_bshift)); 760 1.2 simonb if (error) 761 1.2 simonb return error; 762 1.2 simonb off += len; 763 1.2 simonb if (off >= wl->wl_circ_off + wl->wl_circ_size) 764 1.2 simonb off = wl->wl_circ_off; 765 1.2 simonb *offp = off; 766 1.2 simonb return 0; 767 1.2 simonb } 768 1.2 simonb 769 1.2 simonb /****************************************************************/ 770 1.2 simonb 771 1.2 simonb int 772 1.2 simonb wapbl_begin(struct wapbl *wl, const char *file, int line) 773 1.2 simonb { 774 1.2 simonb int doflush; 775 1.2 simonb unsigned lockcount; 776 1.2 simonb 777 1.2 simonb KDASSERT(wl); 778 1.2 simonb 779 1.2 simonb /* 780 1.2 simonb * XXX this needs to be made much more sophisticated. 781 1.2 simonb * perhaps each wapbl_begin could reserve a specified 782 1.2 simonb * number of buffers and bytes. 783 1.2 simonb */ 784 1.2 simonb mutex_enter(&wl->wl_mtx); 785 1.2 simonb lockcount = wl->wl_lock_count; 786 1.2 simonb doflush = ((wl->wl_bufbytes + (lockcount * MAXPHYS)) > 787 1.2 simonb wl->wl_bufbytes_max / 2) || 788 1.2 simonb ((wl->wl_bufcount + (lockcount * 10)) > 789 1.2 simonb wl->wl_bufcount_max / 2) || 790 1.3.8.2.2.2 riz (wapbl_transaction_len(wl) > wl->wl_circ_size / 2) || 791 1.3.8.2.2.2 riz (wl->wl_dealloccnt >= 792 1.3.8.2.2.2 riz (wl->wl_dealloclim - (wl->wl_dealloclim >> 8))); 793 1.2 simonb mutex_exit(&wl->wl_mtx); 794 1.2 simonb 795 1.2 simonb if (doflush) { 796 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_FLUSH, 797 1.2 simonb ("force flush lockcnt=%d bufbytes=%zu " 798 1.3.8.2.2.2 riz "(max=%zu) bufcount=%zu (max=%zu) " 799 1.3.8.2.2.2 riz "dealloccnt %d (lim=%d)\n", 800 1.2 simonb lockcount, wl->wl_bufbytes, 801 1.2 simonb wl->wl_bufbytes_max, wl->wl_bufcount, 802 1.3.8.2.2.2 riz wl->wl_bufcount_max, 803 1.3.8.2.2.2 riz wl->wl_dealloccnt, wl->wl_dealloclim)); 804 1.2 simonb } 805 1.2 simonb 806 1.2 simonb if (doflush) { 807 1.2 simonb int error = wapbl_flush(wl, 0); 808 1.2 simonb if (error) 809 1.2 simonb return error; 810 1.2 simonb } 811 1.2 simonb 812 1.3.8.1 snj rw_enter(&wl->wl_rwlock, RW_READER); 813 1.2 simonb mutex_enter(&wl->wl_mtx); 814 1.2 simonb wl->wl_lock_count++; 815 1.2 simonb mutex_exit(&wl->wl_mtx); 816 1.2 simonb 817 1.3.8.1 snj #if defined(WAPBL_DEBUG_PRINT) 818 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_TRANSACTION, 819 1.2 simonb ("wapbl_begin thread %d.%d with bufcount=%zu " 820 1.2 simonb "bufbytes=%zu bcount=%zu at %s:%d\n", 821 1.2 simonb curproc->p_pid, curlwp->l_lid, wl->wl_bufcount, 822 1.2 simonb wl->wl_bufbytes, wl->wl_bcount, file, line)); 823 1.2 simonb #endif 824 1.2 simonb 825 1.2 simonb return 0; 826 1.2 simonb } 827 1.2 simonb 828 1.2 simonb void 829 1.2 simonb wapbl_end(struct wapbl *wl) 830 1.2 simonb { 831 1.2 simonb 832 1.3.8.1 snj #if defined(WAPBL_DEBUG_PRINT) 833 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_TRANSACTION, 834 1.2 simonb ("wapbl_end thread %d.%d with bufcount=%zu " 835 1.2 simonb "bufbytes=%zu bcount=%zu\n", 836 1.2 simonb curproc->p_pid, curlwp->l_lid, wl->wl_bufcount, 837 1.2 simonb wl->wl_bufbytes, wl->wl_bcount)); 838 1.2 simonb #endif 839 1.2 simonb 840 1.2 simonb mutex_enter(&wl->wl_mtx); 841 1.2 simonb KASSERT(wl->wl_lock_count > 0); 842 1.2 simonb wl->wl_lock_count--; 843 1.2 simonb mutex_exit(&wl->wl_mtx); 844 1.2 simonb 845 1.2 simonb rw_exit(&wl->wl_rwlock); 846 1.2 simonb } 847 1.2 simonb 848 1.2 simonb void 849 1.2 simonb wapbl_add_buf(struct wapbl *wl, struct buf * bp) 850 1.2 simonb { 851 1.2 simonb 852 1.2 simonb KASSERT(bp->b_cflags & BC_BUSY); 853 1.2 simonb KASSERT(bp->b_vp); 854 1.2 simonb 855 1.2 simonb wapbl_jlock_assert(wl); 856 1.2 simonb 857 1.2 simonb #if 0 858 1.2 simonb /* 859 1.2 simonb * XXX this might be an issue for swapfiles. 860 1.2 simonb * see uvm_swap.c:1702 861 1.2 simonb * 862 1.2 simonb * XXX2 why require it then? leap of semantics? 863 1.2 simonb */ 864 1.2 simonb KASSERT((bp->b_cflags & BC_NOCACHE) == 0); 865 1.2 simonb #endif 866 1.2 simonb 867 1.2 simonb mutex_enter(&wl->wl_mtx); 868 1.2 simonb if (bp->b_flags & B_LOCKED) { 869 1.2 simonb LIST_REMOVE(bp, b_wapbllist); 870 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_BUFFER2, 871 1.2 simonb ("wapbl_add_buf thread %d.%d re-adding buf %p " 872 1.2 simonb "with %d bytes %d bcount\n", 873 1.2 simonb curproc->p_pid, curlwp->l_lid, bp, bp->b_bufsize, 874 1.2 simonb bp->b_bcount)); 875 1.2 simonb } else { 876 1.2 simonb /* unlocked by dirty buffers shouldn't exist */ 877 1.2 simonb KASSERT(!(bp->b_oflags & BO_DELWRI)); 878 1.2 simonb wl->wl_bufbytes += bp->b_bufsize; 879 1.2 simonb wl->wl_bcount += bp->b_bcount; 880 1.2 simonb wl->wl_bufcount++; 881 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_BUFFER, 882 1.2 simonb ("wapbl_add_buf thread %d.%d adding buf %p " 883 1.2 simonb "with %d bytes %d bcount\n", 884 1.2 simonb curproc->p_pid, curlwp->l_lid, bp, bp->b_bufsize, 885 1.2 simonb bp->b_bcount)); 886 1.2 simonb } 887 1.2 simonb LIST_INSERT_HEAD(&wl->wl_bufs, bp, b_wapbllist); 888 1.2 simonb mutex_exit(&wl->wl_mtx); 889 1.2 simonb 890 1.2 simonb bp->b_flags |= B_LOCKED; 891 1.2 simonb } 892 1.2 simonb 893 1.2 simonb static void 894 1.2 simonb wapbl_remove_buf_locked(struct wapbl * wl, struct buf *bp) 895 1.2 simonb { 896 1.2 simonb 897 1.2 simonb KASSERT(mutex_owned(&wl->wl_mtx)); 898 1.2 simonb KASSERT(bp->b_cflags & BC_BUSY); 899 1.2 simonb wapbl_jlock_assert(wl); 900 1.2 simonb 901 1.2 simonb #if 0 902 1.2 simonb /* 903 1.2 simonb * XXX this might be an issue for swapfiles. 904 1.2 simonb * see uvm_swap.c:1725 905 1.2 simonb * 906 1.2 simonb * XXXdeux: see above 907 1.2 simonb */ 908 1.2 simonb KASSERT((bp->b_flags & BC_NOCACHE) == 0); 909 1.2 simonb #endif 910 1.2 simonb KASSERT(bp->b_flags & B_LOCKED); 911 1.2 simonb 912 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_BUFFER, 913 1.2 simonb ("wapbl_remove_buf thread %d.%d removing buf %p with " 914 1.2 simonb "%d bytes %d bcount\n", 915 1.2 simonb curproc->p_pid, curlwp->l_lid, bp, bp->b_bufsize, bp->b_bcount)); 916 1.2 simonb 917 1.2 simonb KASSERT(wl->wl_bufbytes >= bp->b_bufsize); 918 1.2 simonb wl->wl_bufbytes -= bp->b_bufsize; 919 1.2 simonb KASSERT(wl->wl_bcount >= bp->b_bcount); 920 1.2 simonb wl->wl_bcount -= bp->b_bcount; 921 1.2 simonb KASSERT(wl->wl_bufcount > 0); 922 1.2 simonb wl->wl_bufcount--; 923 1.2 simonb KASSERT((wl->wl_bufcount == 0) == (wl->wl_bufbytes == 0)); 924 1.2 simonb KASSERT((wl->wl_bufcount == 0) == (wl->wl_bcount == 0)); 925 1.2 simonb LIST_REMOVE(bp, b_wapbllist); 926 1.2 simonb 927 1.2 simonb bp->b_flags &= ~B_LOCKED; 928 1.2 simonb } 929 1.2 simonb 930 1.2 simonb /* called from brelsel() in vfs_bio among other places */ 931 1.2 simonb void 932 1.2 simonb wapbl_remove_buf(struct wapbl * wl, struct buf *bp) 933 1.2 simonb { 934 1.2 simonb 935 1.2 simonb mutex_enter(&wl->wl_mtx); 936 1.2 simonb wapbl_remove_buf_locked(wl, bp); 937 1.2 simonb mutex_exit(&wl->wl_mtx); 938 1.2 simonb } 939 1.2 simonb 940 1.2 simonb void 941 1.2 simonb wapbl_resize_buf(struct wapbl *wl, struct buf *bp, long oldsz, long oldcnt) 942 1.2 simonb { 943 1.2 simonb 944 1.2 simonb KASSERT(bp->b_cflags & BC_BUSY); 945 1.2 simonb 946 1.2 simonb /* 947 1.2 simonb * XXX: why does this depend on B_LOCKED? otherwise the buf 948 1.2 simonb * is not for a transaction? if so, why is this called in the 949 1.2 simonb * first place? 950 1.2 simonb */ 951 1.2 simonb if (bp->b_flags & B_LOCKED) { 952 1.2 simonb mutex_enter(&wl->wl_mtx); 953 1.2 simonb wl->wl_bufbytes += bp->b_bufsize - oldsz; 954 1.2 simonb wl->wl_bcount += bp->b_bcount - oldcnt; 955 1.2 simonb mutex_exit(&wl->wl_mtx); 956 1.2 simonb } 957 1.2 simonb } 958 1.2 simonb 959 1.2 simonb #endif /* _KERNEL */ 960 1.2 simonb 961 1.2 simonb /****************************************************************/ 962 1.2 simonb /* Some utility inlines */ 963 1.2 simonb 964 1.2 simonb /* This is used to advance the pointer at old to new value at old+delta */ 965 1.2 simonb static __inline off_t 966 1.2 simonb wapbl_advance(size_t size, size_t off, off_t old, size_t delta) 967 1.2 simonb { 968 1.2 simonb off_t new; 969 1.2 simonb 970 1.2 simonb /* Define acceptable ranges for inputs. */ 971 1.2 simonb KASSERT(delta <= size); 972 1.2 simonb KASSERT((old == 0) || (old >= off)); 973 1.2 simonb KASSERT(old < (size + off)); 974 1.2 simonb 975 1.2 simonb if ((old == 0) && (delta != 0)) 976 1.2 simonb new = off + delta; 977 1.2 simonb else if ((old + delta) < (size + off)) 978 1.2 simonb new = old + delta; 979 1.2 simonb else 980 1.2 simonb new = (old + delta) - size; 981 1.2 simonb 982 1.2 simonb /* Note some interesting axioms */ 983 1.2 simonb KASSERT((delta != 0) || (new == old)); 984 1.2 simonb KASSERT((delta == 0) || (new != 0)); 985 1.2 simonb KASSERT((delta != (size)) || (new == old)); 986 1.2 simonb 987 1.2 simonb /* Define acceptable ranges for output. */ 988 1.2 simonb KASSERT((new == 0) || (new >= off)); 989 1.2 simonb KASSERT(new < (size + off)); 990 1.2 simonb return new; 991 1.2 simonb } 992 1.2 simonb 993 1.2 simonb static __inline size_t 994 1.2 simonb wapbl_space_used(size_t avail, off_t head, off_t tail) 995 1.2 simonb { 996 1.2 simonb 997 1.2 simonb if (tail == 0) { 998 1.2 simonb KASSERT(head == 0); 999 1.2 simonb return 0; 1000 1.2 simonb } 1001 1.2 simonb return ((head + (avail - 1) - tail) % avail) + 1; 1002 1.2 simonb } 1003 1.2 simonb 1004 1.2 simonb static __inline size_t 1005 1.2 simonb wapbl_space_free(size_t avail, off_t head, off_t tail) 1006 1.2 simonb { 1007 1.2 simonb 1008 1.2 simonb return avail - wapbl_space_used(avail, head, tail); 1009 1.2 simonb } 1010 1.2 simonb 1011 1.2 simonb static __inline void 1012 1.2 simonb wapbl_advance_head(size_t size, size_t off, size_t delta, off_t *headp, 1013 1.2 simonb off_t *tailp) 1014 1.2 simonb { 1015 1.2 simonb off_t head = *headp; 1016 1.2 simonb off_t tail = *tailp; 1017 1.2 simonb 1018 1.2 simonb KASSERT(delta <= wapbl_space_free(size, head, tail)); 1019 1.2 simonb head = wapbl_advance(size, off, head, delta); 1020 1.2 simonb if ((tail == 0) && (head != 0)) 1021 1.2 simonb tail = off; 1022 1.2 simonb *headp = head; 1023 1.2 simonb *tailp = tail; 1024 1.2 simonb } 1025 1.2 simonb 1026 1.2 simonb static __inline void 1027 1.2 simonb wapbl_advance_tail(size_t size, size_t off, size_t delta, off_t *headp, 1028 1.2 simonb off_t *tailp) 1029 1.2 simonb { 1030 1.2 simonb off_t head = *headp; 1031 1.2 simonb off_t tail = *tailp; 1032 1.2 simonb 1033 1.2 simonb KASSERT(delta <= wapbl_space_used(size, head, tail)); 1034 1.2 simonb tail = wapbl_advance(size, off, tail, delta); 1035 1.2 simonb if (head == tail) { 1036 1.2 simonb head = tail = 0; 1037 1.2 simonb } 1038 1.2 simonb *headp = head; 1039 1.2 simonb *tailp = tail; 1040 1.2 simonb } 1041 1.2 simonb 1042 1.2 simonb #ifdef _KERNEL 1043 1.2 simonb 1044 1.2 simonb /****************************************************************/ 1045 1.2 simonb 1046 1.2 simonb /* 1047 1.2 simonb * Remove transactions whose buffers are completely flushed to disk. 1048 1.2 simonb * Will block until at least minfree space is available. 1049 1.2 simonb * only intended to be called from inside wapbl_flush and therefore 1050 1.2 simonb * does not protect against commit races with itself or with flush. 1051 1.2 simonb */ 1052 1.2 simonb static int 1053 1.2 simonb wapbl_truncate(struct wapbl *wl, size_t minfree, int waitonly) 1054 1.2 simonb { 1055 1.2 simonb size_t delta; 1056 1.2 simonb size_t avail; 1057 1.2 simonb off_t head; 1058 1.2 simonb off_t tail; 1059 1.2 simonb int error = 0; 1060 1.2 simonb 1061 1.2 simonb KASSERT(minfree <= (wl->wl_circ_size - wl->wl_reserved_bytes)); 1062 1.2 simonb KASSERT(rw_write_held(&wl->wl_rwlock)); 1063 1.2 simonb 1064 1.2 simonb mutex_enter(&wl->wl_mtx); 1065 1.2 simonb 1066 1.2 simonb /* 1067 1.2 simonb * First check to see if we have to do a commit 1068 1.2 simonb * at all. 1069 1.2 simonb */ 1070 1.2 simonb avail = wapbl_space_free(wl->wl_circ_size, wl->wl_head, wl->wl_tail); 1071 1.2 simonb if (minfree < avail) { 1072 1.2 simonb mutex_exit(&wl->wl_mtx); 1073 1.2 simonb return 0; 1074 1.2 simonb } 1075 1.2 simonb minfree -= avail; 1076 1.2 simonb while ((wl->wl_error_count == 0) && 1077 1.2 simonb (wl->wl_reclaimable_bytes < minfree)) { 1078 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_TRUNCATE, 1079 1.2 simonb ("wapbl_truncate: sleeping on %p wl=%p bytes=%zd " 1080 1.2 simonb "minfree=%zd\n", 1081 1.2 simonb &wl->wl_reclaimable_bytes, wl, wl->wl_reclaimable_bytes, 1082 1.2 simonb minfree)); 1083 1.2 simonb 1084 1.2 simonb cv_wait(&wl->wl_reclaimable_cv, &wl->wl_mtx); 1085 1.2 simonb } 1086 1.2 simonb if (wl->wl_reclaimable_bytes < minfree) { 1087 1.2 simonb KASSERT(wl->wl_error_count); 1088 1.2 simonb /* XXX maybe get actual error from buffer instead someday? */ 1089 1.2 simonb error = EIO; 1090 1.2 simonb } 1091 1.2 simonb head = wl->wl_head; 1092 1.2 simonb tail = wl->wl_tail; 1093 1.2 simonb delta = wl->wl_reclaimable_bytes; 1094 1.2 simonb 1095 1.2 simonb /* If all of of the entries are flushed, then be sure to keep 1096 1.2 simonb * the reserved bytes reserved. Watch out for discarded transactions, 1097 1.2 simonb * which could leave more bytes reserved than are reclaimable. 1098 1.2 simonb */ 1099 1.2 simonb if (SIMPLEQ_EMPTY(&wl->wl_entries) && 1100 1.2 simonb (delta >= wl->wl_reserved_bytes)) { 1101 1.2 simonb delta -= wl->wl_reserved_bytes; 1102 1.2 simonb } 1103 1.2 simonb wapbl_advance_tail(wl->wl_circ_size, wl->wl_circ_off, delta, &head, 1104 1.2 simonb &tail); 1105 1.2 simonb KDASSERT(wl->wl_reserved_bytes <= 1106 1.2 simonb wapbl_space_used(wl->wl_circ_size, head, tail)); 1107 1.2 simonb mutex_exit(&wl->wl_mtx); 1108 1.2 simonb 1109 1.2 simonb if (error) 1110 1.2 simonb return error; 1111 1.2 simonb 1112 1.2 simonb if (waitonly) 1113 1.2 simonb return 0; 1114 1.2 simonb 1115 1.2 simonb /* 1116 1.2 simonb * This is where head, tail and delta are unprotected 1117 1.2 simonb * from races against itself or flush. This is ok since 1118 1.2 simonb * we only call this routine from inside flush itself. 1119 1.2 simonb * 1120 1.2 simonb * XXX: how can it race against itself when accessed only 1121 1.2 simonb * from behind the write-locked rwlock? 1122 1.2 simonb */ 1123 1.2 simonb error = wapbl_write_commit(wl, head, tail); 1124 1.2 simonb if (error) 1125 1.2 simonb return error; 1126 1.2 simonb 1127 1.2 simonb wl->wl_head = head; 1128 1.2 simonb wl->wl_tail = tail; 1129 1.2 simonb 1130 1.2 simonb mutex_enter(&wl->wl_mtx); 1131 1.2 simonb KASSERT(wl->wl_reclaimable_bytes >= delta); 1132 1.2 simonb wl->wl_reclaimable_bytes -= delta; 1133 1.2 simonb mutex_exit(&wl->wl_mtx); 1134 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_TRUNCATE, 1135 1.2 simonb ("wapbl_truncate thread %d.%d truncating %zu bytes\n", 1136 1.2 simonb curproc->p_pid, curlwp->l_lid, delta)); 1137 1.2 simonb 1138 1.2 simonb return 0; 1139 1.2 simonb } 1140 1.2 simonb 1141 1.2 simonb /****************************************************************/ 1142 1.2 simonb 1143 1.2 simonb void 1144 1.2 simonb wapbl_biodone(struct buf *bp) 1145 1.2 simonb { 1146 1.2 simonb struct wapbl_entry *we = bp->b_private; 1147 1.2 simonb struct wapbl *wl = we->we_wapbl; 1148 1.2 simonb 1149 1.2 simonb /* 1150 1.2 simonb * Handle possible flushing of buffers after log has been 1151 1.2 simonb * decomissioned. 1152 1.2 simonb */ 1153 1.2 simonb if (!wl) { 1154 1.2 simonb KASSERT(we->we_bufcount > 0); 1155 1.2 simonb we->we_bufcount--; 1156 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1157 1.2 simonb KASSERT(we->we_unsynced_bufbytes >= bp->b_bufsize); 1158 1.2 simonb we->we_unsynced_bufbytes -= bp->b_bufsize; 1159 1.2 simonb #endif 1160 1.2 simonb 1161 1.2 simonb if (we->we_bufcount == 0) { 1162 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1163 1.2 simonb KASSERT(we->we_unsynced_bufbytes == 0); 1164 1.2 simonb #endif 1165 1.2 simonb wapbl_free(we); 1166 1.2 simonb } 1167 1.2 simonb 1168 1.2 simonb brelse(bp, 0); 1169 1.2 simonb return; 1170 1.2 simonb } 1171 1.2 simonb 1172 1.2 simonb #ifdef ohbother 1173 1.2 simonb KDASSERT(bp->b_flags & B_DONE); 1174 1.2 simonb KDASSERT(!(bp->b_flags & B_DELWRI)); 1175 1.2 simonb KDASSERT(bp->b_flags & B_ASYNC); 1176 1.2 simonb KDASSERT(bp->b_flags & B_BUSY); 1177 1.2 simonb KDASSERT(!(bp->b_flags & B_LOCKED)); 1178 1.2 simonb KDASSERT(!(bp->b_flags & B_READ)); 1179 1.2 simonb KDASSERT(!(bp->b_flags & B_INVAL)); 1180 1.2 simonb KDASSERT(!(bp->b_flags & B_NOCACHE)); 1181 1.2 simonb #endif 1182 1.2 simonb 1183 1.2 simonb if (bp->b_error) { 1184 1.2 simonb #ifdef notyet /* Can't currently handle possible dirty buffer reuse */ 1185 1.2 simonb XXXpooka: interfaces not fully updated 1186 1.2 simonb Note: this was not enabled in the original patch 1187 1.2 simonb against netbsd4 either. I don't know if comment 1188 1.2 simonb above is true or not. 1189 1.2 simonb 1190 1.2 simonb /* 1191 1.2 simonb * If an error occurs, report the error and leave the 1192 1.2 simonb * buffer as a delayed write on the LRU queue. 1193 1.2 simonb * restarting the write would likely result in 1194 1.2 simonb * an error spinloop, so let it be done harmlessly 1195 1.2 simonb * by the syncer. 1196 1.2 simonb */ 1197 1.2 simonb bp->b_flags &= ~(B_DONE); 1198 1.2 simonb simple_unlock(&bp->b_interlock); 1199 1.2 simonb 1200 1.2 simonb if (we->we_error == 0) { 1201 1.2 simonb mutex_enter(&wl->wl_mtx); 1202 1.2 simonb wl->wl_error_count++; 1203 1.2 simonb mutex_exit(&wl->wl_mtx); 1204 1.2 simonb cv_broadcast(&wl->wl_reclaimable_cv); 1205 1.2 simonb } 1206 1.2 simonb we->we_error = bp->b_error; 1207 1.2 simonb bp->b_error = 0; 1208 1.2 simonb brelse(bp); 1209 1.2 simonb return; 1210 1.2 simonb #else 1211 1.2 simonb /* For now, just mark the log permanently errored out */ 1212 1.2 simonb 1213 1.2 simonb mutex_enter(&wl->wl_mtx); 1214 1.2 simonb if (wl->wl_error_count == 0) { 1215 1.2 simonb wl->wl_error_count++; 1216 1.2 simonb cv_broadcast(&wl->wl_reclaimable_cv); 1217 1.2 simonb } 1218 1.2 simonb mutex_exit(&wl->wl_mtx); 1219 1.2 simonb #endif 1220 1.2 simonb } 1221 1.2 simonb 1222 1.2 simonb mutex_enter(&wl->wl_mtx); 1223 1.2 simonb 1224 1.2 simonb KASSERT(we->we_bufcount > 0); 1225 1.2 simonb we->we_bufcount--; 1226 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1227 1.2 simonb KASSERT(we->we_unsynced_bufbytes >= bp->b_bufsize); 1228 1.2 simonb we->we_unsynced_bufbytes -= bp->b_bufsize; 1229 1.2 simonb KASSERT(wl->wl_unsynced_bufbytes >= bp->b_bufsize); 1230 1.2 simonb wl->wl_unsynced_bufbytes -= bp->b_bufsize; 1231 1.2 simonb #endif 1232 1.2 simonb 1233 1.2 simonb /* 1234 1.2 simonb * If the current transaction can be reclaimed, start 1235 1.2 simonb * at the beginning and reclaim any consecutive reclaimable 1236 1.2 simonb * transactions. If we successfully reclaim anything, 1237 1.2 simonb * then wakeup anyone waiting for the reclaim. 1238 1.2 simonb */ 1239 1.2 simonb if (we->we_bufcount == 0) { 1240 1.2 simonb size_t delta = 0; 1241 1.2 simonb int errcnt = 0; 1242 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1243 1.2 simonb KDASSERT(we->we_unsynced_bufbytes == 0); 1244 1.2 simonb #endif 1245 1.2 simonb /* 1246 1.2 simonb * clear any posted error, since the buffer it came from 1247 1.2 simonb * has successfully flushed by now 1248 1.2 simonb */ 1249 1.2 simonb while ((we = SIMPLEQ_FIRST(&wl->wl_entries)) && 1250 1.2 simonb (we->we_bufcount == 0)) { 1251 1.2 simonb delta += we->we_reclaimable_bytes; 1252 1.2 simonb if (we->we_error) 1253 1.2 simonb errcnt++; 1254 1.2 simonb SIMPLEQ_REMOVE_HEAD(&wl->wl_entries, we_entries); 1255 1.2 simonb wapbl_free(we); 1256 1.2 simonb } 1257 1.2 simonb 1258 1.2 simonb if (delta) { 1259 1.2 simonb wl->wl_reclaimable_bytes += delta; 1260 1.2 simonb KASSERT(wl->wl_error_count >= errcnt); 1261 1.2 simonb wl->wl_error_count -= errcnt; 1262 1.2 simonb cv_broadcast(&wl->wl_reclaimable_cv); 1263 1.2 simonb } 1264 1.2 simonb } 1265 1.2 simonb 1266 1.2 simonb mutex_exit(&wl->wl_mtx); 1267 1.2 simonb brelse(bp, 0); 1268 1.2 simonb } 1269 1.2 simonb 1270 1.2 simonb /* 1271 1.2 simonb * Write transactions to disk + start I/O for contents 1272 1.2 simonb */ 1273 1.2 simonb int 1274 1.2 simonb wapbl_flush(struct wapbl *wl, int waitfor) 1275 1.2 simonb { 1276 1.2 simonb struct buf *bp; 1277 1.2 simonb struct wapbl_entry *we; 1278 1.2 simonb off_t off; 1279 1.2 simonb off_t head; 1280 1.2 simonb off_t tail; 1281 1.2 simonb size_t delta = 0; 1282 1.2 simonb size_t flushsize; 1283 1.2 simonb size_t reserved; 1284 1.2 simonb int error = 0; 1285 1.2 simonb 1286 1.2 simonb /* 1287 1.2 simonb * Do a quick check to see if a full flush can be skipped 1288 1.2 simonb * This assumes that the flush callback does not need to be called 1289 1.2 simonb * unless there are other outstanding bufs. 1290 1.2 simonb */ 1291 1.2 simonb if (!waitfor) { 1292 1.2 simonb size_t nbufs; 1293 1.2 simonb mutex_enter(&wl->wl_mtx); /* XXX need mutex here to 1294 1.2 simonb protect the KASSERTS */ 1295 1.2 simonb nbufs = wl->wl_bufcount; 1296 1.2 simonb KASSERT((wl->wl_bufcount == 0) == (wl->wl_bufbytes == 0)); 1297 1.2 simonb KASSERT((wl->wl_bufcount == 0) == (wl->wl_bcount == 0)); 1298 1.2 simonb mutex_exit(&wl->wl_mtx); 1299 1.2 simonb if (nbufs == 0) 1300 1.2 simonb return 0; 1301 1.2 simonb } 1302 1.2 simonb 1303 1.2 simonb /* 1304 1.2 simonb * XXX we may consider using LK_UPGRADE here 1305 1.2 simonb * if we want to call flush from inside a transaction 1306 1.2 simonb */ 1307 1.2 simonb rw_enter(&wl->wl_rwlock, RW_WRITER); 1308 1.2 simonb wl->wl_flush(wl->wl_mount, wl->wl_deallocblks, wl->wl_dealloclens, 1309 1.2 simonb wl->wl_dealloccnt); 1310 1.2 simonb 1311 1.2 simonb /* 1312 1.2 simonb * Now that we are fully locked and flushed, 1313 1.2 simonb * do another check for nothing to do. 1314 1.2 simonb */ 1315 1.2 simonb if (wl->wl_bufcount == 0) { 1316 1.2 simonb goto out; 1317 1.2 simonb } 1318 1.2 simonb 1319 1.2 simonb #if 0 1320 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_FLUSH, 1321 1.2 simonb ("wapbl_flush thread %d.%d flushing entries with " 1322 1.2 simonb "bufcount=%zu bufbytes=%zu\n", 1323 1.2 simonb curproc->p_pid, curlwp->l_lid, wl->wl_bufcount, 1324 1.2 simonb wl->wl_bufbytes)); 1325 1.2 simonb #endif 1326 1.2 simonb 1327 1.2 simonb /* Calculate amount of space needed to flush */ 1328 1.2 simonb flushsize = wapbl_transaction_len(wl); 1329 1.2 simonb 1330 1.2 simonb if (flushsize > (wl->wl_circ_size - wl->wl_reserved_bytes)) { 1331 1.2 simonb /* 1332 1.2 simonb * XXX this could be handled more gracefully, perhaps place 1333 1.2 simonb * only a partial transaction in the log and allow the 1334 1.2 simonb * remaining to flush without the protection of the journal. 1335 1.2 simonb */ 1336 1.2 simonb panic("wapbl_flush: current transaction too big to flush\n"); 1337 1.2 simonb } 1338 1.2 simonb 1339 1.2 simonb error = wapbl_truncate(wl, flushsize, 0); 1340 1.2 simonb if (error) 1341 1.2 simonb goto out2; 1342 1.2 simonb 1343 1.2 simonb off = wl->wl_head; 1344 1.2 simonb KASSERT((off == 0) || ((off >= wl->wl_circ_off) && 1345 1.2 simonb (off < wl->wl_circ_off + wl->wl_circ_size))); 1346 1.2 simonb error = wapbl_write_blocks(wl, &off); 1347 1.2 simonb if (error) 1348 1.2 simonb goto out2; 1349 1.2 simonb error = wapbl_write_revocations(wl, &off); 1350 1.2 simonb if (error) 1351 1.2 simonb goto out2; 1352 1.2 simonb error = wapbl_write_inodes(wl, &off); 1353 1.2 simonb if (error) 1354 1.2 simonb goto out2; 1355 1.2 simonb 1356 1.2 simonb reserved = 0; 1357 1.2 simonb if (wl->wl_inohashcnt) 1358 1.2 simonb reserved = wapbl_transaction_inodes_len(wl); 1359 1.2 simonb 1360 1.2 simonb head = wl->wl_head; 1361 1.2 simonb tail = wl->wl_tail; 1362 1.2 simonb 1363 1.2 simonb wapbl_advance_head(wl->wl_circ_size, wl->wl_circ_off, flushsize, 1364 1.2 simonb &head, &tail); 1365 1.2 simonb #ifdef WAPBL_DEBUG 1366 1.2 simonb if (head != off) { 1367 1.2 simonb panic("lost head! head=%"PRIdMAX" tail=%" PRIdMAX 1368 1.2 simonb " off=%"PRIdMAX" flush=%zu\n", 1369 1.2 simonb (intmax_t)head, (intmax_t)tail, (intmax_t)off, 1370 1.2 simonb flushsize); 1371 1.2 simonb } 1372 1.2 simonb #else 1373 1.2 simonb KASSERT(head == off); 1374 1.2 simonb #endif 1375 1.2 simonb 1376 1.2 simonb /* Opportunistically move the tail forward if we can */ 1377 1.2 simonb if (!wapbl_lazy_truncate) { 1378 1.2 simonb mutex_enter(&wl->wl_mtx); 1379 1.2 simonb delta = wl->wl_reclaimable_bytes; 1380 1.2 simonb mutex_exit(&wl->wl_mtx); 1381 1.2 simonb wapbl_advance_tail(wl->wl_circ_size, wl->wl_circ_off, delta, 1382 1.2 simonb &head, &tail); 1383 1.2 simonb } 1384 1.2 simonb 1385 1.2 simonb error = wapbl_write_commit(wl, head, tail); 1386 1.2 simonb if (error) 1387 1.2 simonb goto out2; 1388 1.2 simonb 1389 1.2 simonb /* poolme? or kmemme? */ 1390 1.2 simonb we = wapbl_calloc(1, sizeof(*we)); 1391 1.2 simonb 1392 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1393 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_FLUSH, 1394 1.2 simonb ("wapbl_flush: thread %d.%d head+=%zu tail+=%zu used=%zu" 1395 1.2 simonb " unsynced=%zu" 1396 1.2 simonb "\n\tbufcount=%zu bufbytes=%zu bcount=%zu deallocs=%d " 1397 1.2 simonb "inodes=%d\n", 1398 1.2 simonb curproc->p_pid, curlwp->l_lid, flushsize, delta, 1399 1.2 simonb wapbl_space_used(wl->wl_circ_size, head, tail), 1400 1.2 simonb wl->wl_unsynced_bufbytes, wl->wl_bufcount, 1401 1.2 simonb wl->wl_bufbytes, wl->wl_bcount, wl->wl_dealloccnt, 1402 1.2 simonb wl->wl_inohashcnt)); 1403 1.2 simonb #else 1404 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_FLUSH, 1405 1.2 simonb ("wapbl_flush: thread %d.%d head+=%zu tail+=%zu used=%zu" 1406 1.2 simonb "\n\tbufcount=%zu bufbytes=%zu bcount=%zu deallocs=%d " 1407 1.2 simonb "inodes=%d\n", 1408 1.2 simonb curproc->p_pid, curlwp->l_lid, flushsize, delta, 1409 1.2 simonb wapbl_space_used(wl->wl_circ_size, head, tail), 1410 1.2 simonb wl->wl_bufcount, wl->wl_bufbytes, wl->wl_bcount, 1411 1.2 simonb wl->wl_dealloccnt, wl->wl_inohashcnt)); 1412 1.2 simonb #endif 1413 1.2 simonb 1414 1.2 simonb 1415 1.2 simonb mutex_enter(&bufcache_lock); 1416 1.2 simonb mutex_enter(&wl->wl_mtx); 1417 1.2 simonb 1418 1.2 simonb wl->wl_reserved_bytes = reserved; 1419 1.2 simonb wl->wl_head = head; 1420 1.2 simonb wl->wl_tail = tail; 1421 1.2 simonb KASSERT(wl->wl_reclaimable_bytes >= delta); 1422 1.2 simonb wl->wl_reclaimable_bytes -= delta; 1423 1.2 simonb wl->wl_dealloccnt = 0; 1424 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1425 1.2 simonb wl->wl_unsynced_bufbytes += wl->wl_bufbytes; 1426 1.2 simonb #endif 1427 1.2 simonb 1428 1.2 simonb we->we_wapbl = wl; 1429 1.2 simonb we->we_bufcount = wl->wl_bufcount; 1430 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1431 1.2 simonb we->we_unsynced_bufbytes = wl->wl_bufbytes; 1432 1.2 simonb #endif 1433 1.2 simonb we->we_reclaimable_bytes = flushsize; 1434 1.2 simonb we->we_error = 0; 1435 1.2 simonb SIMPLEQ_INSERT_TAIL(&wl->wl_entries, we, we_entries); 1436 1.2 simonb 1437 1.2 simonb /* 1438 1.2 simonb * this flushes bufs in reverse order than they were queued 1439 1.2 simonb * it shouldn't matter, but if we care we could use TAILQ instead. 1440 1.2 simonb * XXX Note they will get put on the lru queue when they flush 1441 1.2 simonb * so we might actually want to change this to preserve order. 1442 1.2 simonb */ 1443 1.2 simonb while ((bp = LIST_FIRST(&wl->wl_bufs)) != NULL) { 1444 1.2 simonb if (bbusy(bp, 0, 0, &wl->wl_mtx)) { 1445 1.2 simonb continue; 1446 1.2 simonb } 1447 1.2 simonb bp->b_iodone = wapbl_biodone; 1448 1.2 simonb bp->b_private = we; 1449 1.2 simonb bremfree(bp); 1450 1.2 simonb wapbl_remove_buf_locked(wl, bp); 1451 1.2 simonb mutex_exit(&wl->wl_mtx); 1452 1.2 simonb mutex_exit(&bufcache_lock); 1453 1.2 simonb bawrite(bp); 1454 1.2 simonb mutex_enter(&bufcache_lock); 1455 1.2 simonb mutex_enter(&wl->wl_mtx); 1456 1.2 simonb } 1457 1.2 simonb mutex_exit(&wl->wl_mtx); 1458 1.2 simonb mutex_exit(&bufcache_lock); 1459 1.2 simonb 1460 1.2 simonb #if 0 1461 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_FLUSH, 1462 1.2 simonb ("wapbl_flush thread %d.%d done flushing entries...\n", 1463 1.2 simonb curproc->p_pid, curlwp->l_lid)); 1464 1.2 simonb #endif 1465 1.2 simonb 1466 1.2 simonb out: 1467 1.2 simonb 1468 1.2 simonb /* 1469 1.2 simonb * If the waitfor flag is set, don't return until everything is 1470 1.2 simonb * fully flushed and the on disk log is empty. 1471 1.2 simonb */ 1472 1.2 simonb if (waitfor) { 1473 1.2 simonb error = wapbl_truncate(wl, wl->wl_circ_size - 1474 1.2 simonb wl->wl_reserved_bytes, wapbl_lazy_truncate); 1475 1.2 simonb } 1476 1.2 simonb 1477 1.2 simonb out2: 1478 1.2 simonb if (error) { 1479 1.2 simonb wl->wl_flush_abort(wl->wl_mount, wl->wl_deallocblks, 1480 1.2 simonb wl->wl_dealloclens, wl->wl_dealloccnt); 1481 1.2 simonb } 1482 1.2 simonb 1483 1.2 simonb #ifdef WAPBL_DEBUG_PRINT 1484 1.2 simonb if (error) { 1485 1.2 simonb pid_t pid = -1; 1486 1.2 simonb lwpid_t lid = -1; 1487 1.2 simonb if (curproc) 1488 1.2 simonb pid = curproc->p_pid; 1489 1.2 simonb if (curlwp) 1490 1.2 simonb lid = curlwp->l_lid; 1491 1.2 simonb mutex_enter(&wl->wl_mtx); 1492 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1493 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_ERROR, 1494 1.2 simonb ("wapbl_flush: thread %d.%d aborted flush: " 1495 1.2 simonb "error = %d\n" 1496 1.2 simonb "\tbufcount=%zu bufbytes=%zu bcount=%zu " 1497 1.2 simonb "deallocs=%d inodes=%d\n" 1498 1.2 simonb "\terrcnt = %d, reclaimable=%zu reserved=%zu " 1499 1.2 simonb "unsynced=%zu\n", 1500 1.2 simonb pid, lid, error, wl->wl_bufcount, 1501 1.2 simonb wl->wl_bufbytes, wl->wl_bcount, 1502 1.2 simonb wl->wl_dealloccnt, wl->wl_inohashcnt, 1503 1.2 simonb wl->wl_error_count, wl->wl_reclaimable_bytes, 1504 1.2 simonb wl->wl_reserved_bytes, wl->wl_unsynced_bufbytes)); 1505 1.2 simonb SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) { 1506 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_ERROR, 1507 1.2 simonb ("\tentry: bufcount = %zu, reclaimable = %zu, " 1508 1.2 simonb "error = %d, unsynced = %zu\n", 1509 1.2 simonb we->we_bufcount, we->we_reclaimable_bytes, 1510 1.2 simonb we->we_error, we->we_unsynced_bufbytes)); 1511 1.2 simonb } 1512 1.2 simonb #else 1513 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_ERROR, 1514 1.2 simonb ("wapbl_flush: thread %d.%d aborted flush: " 1515 1.2 simonb "error = %d\n" 1516 1.2 simonb "\tbufcount=%zu bufbytes=%zu bcount=%zu " 1517 1.2 simonb "deallocs=%d inodes=%d\n" 1518 1.2 simonb "\terrcnt = %d, reclaimable=%zu reserved=%zu\n", 1519 1.2 simonb pid, lid, error, wl->wl_bufcount, 1520 1.2 simonb wl->wl_bufbytes, wl->wl_bcount, 1521 1.2 simonb wl->wl_dealloccnt, wl->wl_inohashcnt, 1522 1.2 simonb wl->wl_error_count, wl->wl_reclaimable_bytes, 1523 1.2 simonb wl->wl_reserved_bytes)); 1524 1.2 simonb SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) { 1525 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_ERROR, 1526 1.2 simonb ("\tentry: bufcount = %zu, reclaimable = %zu, " 1527 1.2 simonb "error = %d\n", we->we_bufcount, 1528 1.2 simonb we->we_reclaimable_bytes, we->we_error)); 1529 1.2 simonb } 1530 1.2 simonb #endif 1531 1.2 simonb mutex_exit(&wl->wl_mtx); 1532 1.2 simonb } 1533 1.2 simonb #endif 1534 1.2 simonb 1535 1.2 simonb rw_exit(&wl->wl_rwlock); 1536 1.2 simonb return error; 1537 1.2 simonb } 1538 1.2 simonb 1539 1.2 simonb /****************************************************************/ 1540 1.2 simonb 1541 1.2 simonb void 1542 1.2 simonb wapbl_jlock_assert(struct wapbl *wl) 1543 1.2 simonb { 1544 1.2 simonb 1545 1.3.8.1 snj KASSERT(rw_lock_held(&wl->wl_rwlock)); 1546 1.2 simonb } 1547 1.2 simonb 1548 1.2 simonb void 1549 1.2 simonb wapbl_junlock_assert(struct wapbl *wl) 1550 1.2 simonb { 1551 1.2 simonb 1552 1.2 simonb KASSERT(!rw_write_held(&wl->wl_rwlock)); 1553 1.2 simonb } 1554 1.2 simonb 1555 1.2 simonb /****************************************************************/ 1556 1.2 simonb 1557 1.2 simonb /* locks missing */ 1558 1.2 simonb void 1559 1.2 simonb wapbl_print(struct wapbl *wl, 1560 1.2 simonb int full, 1561 1.2 simonb void (*pr)(const char *, ...)) 1562 1.2 simonb { 1563 1.2 simonb struct buf *bp; 1564 1.2 simonb struct wapbl_entry *we; 1565 1.2 simonb (*pr)("wapbl %p", wl); 1566 1.2 simonb (*pr)("\nlogvp = %p, devvp = %p, logpbn = %"PRId64"\n", 1567 1.2 simonb wl->wl_logvp, wl->wl_devvp, wl->wl_logpbn); 1568 1.2 simonb (*pr)("circ = %zu, header = %zu, head = %"PRIdMAX" tail = %"PRIdMAX"\n", 1569 1.2 simonb wl->wl_circ_size, wl->wl_circ_off, 1570 1.2 simonb (intmax_t)wl->wl_head, (intmax_t)wl->wl_tail); 1571 1.2 simonb (*pr)("fs_dev_bshift = %d, log_dev_bshift = %d\n", 1572 1.2 simonb wl->wl_log_dev_bshift, wl->wl_fs_dev_bshift); 1573 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1574 1.2 simonb (*pr)("bufcount = %zu, bufbytes = %zu bcount = %zu reclaimable = %zu " 1575 1.2 simonb "reserved = %zu errcnt = %d unsynced = %zu\n", 1576 1.2 simonb wl->wl_bufcount, wl->wl_bufbytes, wl->wl_bcount, 1577 1.2 simonb wl->wl_reclaimable_bytes, wl->wl_reserved_bytes, 1578 1.2 simonb wl->wl_error_count, wl->wl_unsynced_bufbytes); 1579 1.2 simonb #else 1580 1.2 simonb (*pr)("bufcount = %zu, bufbytes = %zu bcount = %zu reclaimable = %zu " 1581 1.2 simonb "reserved = %zu errcnt = %d\n", wl->wl_bufcount, wl->wl_bufbytes, 1582 1.2 simonb wl->wl_bcount, wl->wl_reclaimable_bytes, wl->wl_reserved_bytes, 1583 1.2 simonb wl->wl_error_count); 1584 1.2 simonb #endif 1585 1.2 simonb (*pr)("\tdealloccnt = %d, dealloclim = %d\n", 1586 1.2 simonb wl->wl_dealloccnt, wl->wl_dealloclim); 1587 1.2 simonb (*pr)("\tinohashcnt = %d, inohashmask = 0x%08x\n", 1588 1.2 simonb wl->wl_inohashcnt, wl->wl_inohashmask); 1589 1.2 simonb (*pr)("entries:\n"); 1590 1.2 simonb SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) { 1591 1.2 simonb #ifdef WAPBL_DEBUG_BUFBYTES 1592 1.2 simonb (*pr)("\tbufcount = %zu, reclaimable = %zu, error = %d, " 1593 1.2 simonb "unsynced = %zu\n", 1594 1.2 simonb we->we_bufcount, we->we_reclaimable_bytes, 1595 1.2 simonb we->we_error, we->we_unsynced_bufbytes); 1596 1.2 simonb #else 1597 1.2 simonb (*pr)("\tbufcount = %zu, reclaimable = %zu, error = %d\n", 1598 1.2 simonb we->we_bufcount, we->we_reclaimable_bytes, we->we_error); 1599 1.2 simonb #endif 1600 1.2 simonb } 1601 1.2 simonb if (full) { 1602 1.2 simonb int cnt = 0; 1603 1.2 simonb (*pr)("bufs ="); 1604 1.2 simonb LIST_FOREACH(bp, &wl->wl_bufs, b_wapbllist) { 1605 1.2 simonb if (!LIST_NEXT(bp, b_wapbllist)) { 1606 1.2 simonb (*pr)(" %p", bp); 1607 1.2 simonb } else if ((++cnt % 6) == 0) { 1608 1.2 simonb (*pr)(" %p,\n\t", bp); 1609 1.2 simonb } else { 1610 1.2 simonb (*pr)(" %p,", bp); 1611 1.2 simonb } 1612 1.2 simonb } 1613 1.2 simonb (*pr)("\n"); 1614 1.2 simonb 1615 1.2 simonb (*pr)("dealloced blks = "); 1616 1.2 simonb { 1617 1.2 simonb int i; 1618 1.2 simonb cnt = 0; 1619 1.2 simonb for (i = 0; i < wl->wl_dealloccnt; i++) { 1620 1.2 simonb (*pr)(" %"PRId64":%d,", 1621 1.2 simonb wl->wl_deallocblks[i], 1622 1.2 simonb wl->wl_dealloclens[i]); 1623 1.2 simonb if ((++cnt % 4) == 0) { 1624 1.2 simonb (*pr)("\n\t"); 1625 1.2 simonb } 1626 1.2 simonb } 1627 1.2 simonb } 1628 1.2 simonb (*pr)("\n"); 1629 1.2 simonb 1630 1.2 simonb (*pr)("registered inodes = "); 1631 1.2 simonb { 1632 1.2 simonb int i; 1633 1.2 simonb cnt = 0; 1634 1.2 simonb for (i = 0; i <= wl->wl_inohashmask; i++) { 1635 1.2 simonb struct wapbl_ino_head *wih; 1636 1.2 simonb struct wapbl_ino *wi; 1637 1.2 simonb 1638 1.2 simonb wih = &wl->wl_inohash[i]; 1639 1.2 simonb LIST_FOREACH(wi, wih, wi_hash) { 1640 1.2 simonb if (wi->wi_ino == 0) 1641 1.2 simonb continue; 1642 1.2 simonb (*pr)(" %"PRId32"/0%06"PRIo32",", 1643 1.2 simonb wi->wi_ino, wi->wi_mode); 1644 1.2 simonb if ((++cnt % 4) == 0) { 1645 1.2 simonb (*pr)("\n\t"); 1646 1.2 simonb } 1647 1.2 simonb } 1648 1.2 simonb } 1649 1.2 simonb (*pr)("\n"); 1650 1.2 simonb } 1651 1.2 simonb } 1652 1.2 simonb } 1653 1.2 simonb 1654 1.2 simonb #if defined(WAPBL_DEBUG) || defined(DDB) 1655 1.2 simonb void 1656 1.2 simonb wapbl_dump(struct wapbl *wl) 1657 1.2 simonb { 1658 1.2 simonb #if defined(WAPBL_DEBUG) 1659 1.2 simonb if (!wl) 1660 1.2 simonb wl = wapbl_debug_wl; 1661 1.2 simonb #endif 1662 1.2 simonb if (!wl) 1663 1.2 simonb return; 1664 1.2 simonb wapbl_print(wl, 1, printf); 1665 1.2 simonb } 1666 1.2 simonb #endif 1667 1.2 simonb 1668 1.2 simonb /****************************************************************/ 1669 1.2 simonb 1670 1.2 simonb void 1671 1.2 simonb wapbl_register_deallocation(struct wapbl *wl, daddr_t blk, int len) 1672 1.2 simonb { 1673 1.2 simonb 1674 1.2 simonb wapbl_jlock_assert(wl); 1675 1.2 simonb 1676 1.3.8.2.2.1 riz mutex_enter(&wl->wl_mtx); 1677 1.2 simonb /* XXX should eventually instead tie this into resource estimation */ 1678 1.3.8.2.2.2 riz /* 1679 1.3.8.2.2.2 riz * XXX this panic needs locking/mutex analysis and the 1680 1.3.8.2.2.2 riz * ability to cope with the failure. 1681 1.3.8.2.2.2 riz */ 1682 1.3.8.2.2.2 riz /* XXX this XXX doesn't have enough XXX */ 1683 1.3.8.2.2.2 riz if (__predict_false(wl->wl_dealloccnt >= wl->wl_dealloclim)) 1684 1.3.8.2.2.2 riz panic("wapbl_register_deallocation: out of resources"); 1685 1.3.8.2.2.2 riz 1686 1.2 simonb wl->wl_deallocblks[wl->wl_dealloccnt] = blk; 1687 1.2 simonb wl->wl_dealloclens[wl->wl_dealloccnt] = len; 1688 1.2 simonb wl->wl_dealloccnt++; 1689 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_ALLOC, 1690 1.2 simonb ("wapbl_register_deallocation: blk=%"PRId64" len=%d\n", blk, len)); 1691 1.3.8.2.2.1 riz mutex_exit(&wl->wl_mtx); 1692 1.2 simonb } 1693 1.2 simonb 1694 1.2 simonb /****************************************************************/ 1695 1.2 simonb 1696 1.2 simonb static void 1697 1.2 simonb wapbl_inodetrk_init(struct wapbl *wl, u_int size) 1698 1.2 simonb { 1699 1.2 simonb 1700 1.2 simonb wl->wl_inohash = hashinit(size, HASH_LIST, true, &wl->wl_inohashmask); 1701 1.2 simonb if (atomic_inc_uint_nv(&wapbl_ino_pool_refcount) == 1) { 1702 1.2 simonb pool_init(&wapbl_ino_pool, sizeof(struct wapbl_ino), 0, 0, 0, 1703 1.2 simonb "wapblinopl", &pool_allocator_nointr, IPL_NONE); 1704 1.2 simonb } 1705 1.2 simonb } 1706 1.2 simonb 1707 1.2 simonb static void 1708 1.2 simonb wapbl_inodetrk_free(struct wapbl *wl) 1709 1.2 simonb { 1710 1.2 simonb 1711 1.2 simonb /* XXX this KASSERT needs locking/mutex analysis */ 1712 1.2 simonb KASSERT(wl->wl_inohashcnt == 0); 1713 1.2 simonb hashdone(wl->wl_inohash, HASH_LIST, wl->wl_inohashmask); 1714 1.2 simonb if (atomic_dec_uint_nv(&wapbl_ino_pool_refcount) == 0) { 1715 1.2 simonb pool_destroy(&wapbl_ino_pool); 1716 1.2 simonb } 1717 1.2 simonb } 1718 1.2 simonb 1719 1.2 simonb static struct wapbl_ino * 1720 1.2 simonb wapbl_inodetrk_get(struct wapbl *wl, ino_t ino) 1721 1.2 simonb { 1722 1.2 simonb struct wapbl_ino_head *wih; 1723 1.2 simonb struct wapbl_ino *wi; 1724 1.2 simonb 1725 1.2 simonb KASSERT(mutex_owned(&wl->wl_mtx)); 1726 1.2 simonb 1727 1.2 simonb wih = &wl->wl_inohash[ino & wl->wl_inohashmask]; 1728 1.2 simonb LIST_FOREACH(wi, wih, wi_hash) { 1729 1.2 simonb if (ino == wi->wi_ino) 1730 1.2 simonb return wi; 1731 1.2 simonb } 1732 1.2 simonb return 0; 1733 1.2 simonb } 1734 1.2 simonb 1735 1.2 simonb void 1736 1.2 simonb wapbl_register_inode(struct wapbl *wl, ino_t ino, mode_t mode) 1737 1.2 simonb { 1738 1.2 simonb struct wapbl_ino_head *wih; 1739 1.2 simonb struct wapbl_ino *wi; 1740 1.2 simonb 1741 1.2 simonb wi = pool_get(&wapbl_ino_pool, PR_WAITOK); 1742 1.2 simonb 1743 1.2 simonb mutex_enter(&wl->wl_mtx); 1744 1.2 simonb if (wapbl_inodetrk_get(wl, ino) == NULL) { 1745 1.2 simonb wi->wi_ino = ino; 1746 1.2 simonb wi->wi_mode = mode; 1747 1.2 simonb wih = &wl->wl_inohash[ino & wl->wl_inohashmask]; 1748 1.2 simonb LIST_INSERT_HEAD(wih, wi, wi_hash); 1749 1.2 simonb wl->wl_inohashcnt++; 1750 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_INODE, 1751 1.2 simonb ("wapbl_register_inode: ino=%"PRId64"\n", ino)); 1752 1.2 simonb mutex_exit(&wl->wl_mtx); 1753 1.2 simonb } else { 1754 1.2 simonb mutex_exit(&wl->wl_mtx); 1755 1.2 simonb pool_put(&wapbl_ino_pool, wi); 1756 1.2 simonb } 1757 1.2 simonb } 1758 1.2 simonb 1759 1.2 simonb void 1760 1.2 simonb wapbl_unregister_inode(struct wapbl *wl, ino_t ino, mode_t mode) 1761 1.2 simonb { 1762 1.2 simonb struct wapbl_ino *wi; 1763 1.2 simonb 1764 1.2 simonb mutex_enter(&wl->wl_mtx); 1765 1.2 simonb wi = wapbl_inodetrk_get(wl, ino); 1766 1.2 simonb if (wi) { 1767 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_INODE, 1768 1.2 simonb ("wapbl_unregister_inode: ino=%"PRId64"\n", ino)); 1769 1.2 simonb KASSERT(wl->wl_inohashcnt > 0); 1770 1.2 simonb wl->wl_inohashcnt--; 1771 1.2 simonb LIST_REMOVE(wi, wi_hash); 1772 1.2 simonb mutex_exit(&wl->wl_mtx); 1773 1.2 simonb 1774 1.2 simonb pool_put(&wapbl_ino_pool, wi); 1775 1.2 simonb } else { 1776 1.2 simonb mutex_exit(&wl->wl_mtx); 1777 1.2 simonb } 1778 1.2 simonb } 1779 1.2 simonb 1780 1.2 simonb /****************************************************************/ 1781 1.2 simonb 1782 1.2 simonb static __inline size_t 1783 1.2 simonb wapbl_transaction_inodes_len(struct wapbl *wl) 1784 1.2 simonb { 1785 1.2 simonb int blocklen = 1<<wl->wl_log_dev_bshift; 1786 1.2 simonb int iph; 1787 1.2 simonb 1788 1.2 simonb /* Calculate number of inodes described in a inodelist header */ 1789 1.2 simonb iph = (blocklen - offsetof(struct wapbl_wc_inodelist, wc_inodes)) / 1790 1.2 simonb sizeof(((struct wapbl_wc_inodelist *)0)->wc_inodes[0]); 1791 1.2 simonb 1792 1.2 simonb KASSERT(iph > 0); 1793 1.2 simonb 1794 1.2 simonb return MAX(1, howmany(wl->wl_inohashcnt, iph))*blocklen; 1795 1.2 simonb } 1796 1.2 simonb 1797 1.2 simonb 1798 1.2 simonb /* Calculate amount of space a transaction will take on disk */ 1799 1.2 simonb static size_t 1800 1.2 simonb wapbl_transaction_len(struct wapbl *wl) 1801 1.2 simonb { 1802 1.2 simonb int blocklen = 1<<wl->wl_log_dev_bshift; 1803 1.2 simonb size_t len; 1804 1.2 simonb int bph; 1805 1.2 simonb 1806 1.2 simonb /* Calculate number of blocks described in a blocklist header */ 1807 1.2 simonb bph = (blocklen - offsetof(struct wapbl_wc_blocklist, wc_blocks)) / 1808 1.2 simonb sizeof(((struct wapbl_wc_blocklist *)0)->wc_blocks[0]); 1809 1.2 simonb 1810 1.2 simonb KASSERT(bph > 0); 1811 1.2 simonb 1812 1.2 simonb len = wl->wl_bcount; 1813 1.2 simonb len += howmany(wl->wl_bufcount, bph)*blocklen; 1814 1.2 simonb len += howmany(wl->wl_dealloccnt, bph)*blocklen; 1815 1.2 simonb len += wapbl_transaction_inodes_len(wl); 1816 1.2 simonb 1817 1.2 simonb return len; 1818 1.2 simonb } 1819 1.2 simonb 1820 1.2 simonb /* 1821 1.2 simonb * Perform commit operation 1822 1.2 simonb * 1823 1.2 simonb * Note that generation number incrementation needs to 1824 1.2 simonb * be protected against racing with other invocations 1825 1.2 simonb * of wapbl_commit. This is ok since this routine 1826 1.2 simonb * is only invoked from wapbl_flush 1827 1.2 simonb */ 1828 1.2 simonb static int 1829 1.2 simonb wapbl_write_commit(struct wapbl *wl, off_t head, off_t tail) 1830 1.2 simonb { 1831 1.2 simonb struct wapbl_wc_header *wc = wl->wl_wc_header; 1832 1.2 simonb struct timespec ts; 1833 1.2 simonb int error; 1834 1.2 simonb int force = 1; 1835 1.2 simonb 1836 1.2 simonb /* XXX Calc checksum here, instead we do this for now */ 1837 1.2 simonb error = VOP_IOCTL(wl->wl_devvp, DIOCCACHESYNC, &force, FWRITE, FSCRED); 1838 1.2 simonb if (error) { 1839 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_ERROR, 1840 1.2 simonb ("wapbl_write_commit: DIOCCACHESYNC on dev 0x%x " 1841 1.2 simonb "returned %d\n", wl->wl_devvp->v_rdev, error)); 1842 1.2 simonb } 1843 1.2 simonb 1844 1.2 simonb wc->wc_head = head; 1845 1.2 simonb wc->wc_tail = tail; 1846 1.2 simonb wc->wc_checksum = 0; 1847 1.2 simonb wc->wc_version = 1; 1848 1.2 simonb getnanotime(&ts); 1849 1.2 simonb wc->wc_time = ts.tv_sec;; 1850 1.2 simonb wc->wc_timensec = ts.tv_nsec; 1851 1.2 simonb 1852 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_WRITE, 1853 1.2 simonb ("wapbl_write_commit: head = %"PRIdMAX "tail = %"PRIdMAX"\n", 1854 1.2 simonb (intmax_t)head, (intmax_t)tail)); 1855 1.2 simonb 1856 1.2 simonb /* 1857 1.2 simonb * XXX if generation will rollover, then first zero 1858 1.2 simonb * over second commit header before trying to write both headers. 1859 1.2 simonb */ 1860 1.2 simonb 1861 1.2 simonb error = wapbl_write(wc, wc->wc_len, wl->wl_devvp, 1862 1.2 simonb wl->wl_logpbn + wc->wc_generation % 2); 1863 1.2 simonb if (error) 1864 1.2 simonb return error; 1865 1.2 simonb 1866 1.2 simonb error = VOP_IOCTL(wl->wl_devvp, DIOCCACHESYNC, &force, FWRITE, FSCRED); 1867 1.2 simonb if (error) { 1868 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_ERROR, 1869 1.2 simonb ("wapbl_write_commit: DIOCCACHESYNC on dev 0x%x " 1870 1.2 simonb "returned %d\n", wl->wl_devvp->v_rdev, error)); 1871 1.2 simonb } 1872 1.2 simonb 1873 1.2 simonb /* 1874 1.2 simonb * If the generation number was zero, write it out a second time. 1875 1.2 simonb * This handles initialization and generation number rollover 1876 1.2 simonb */ 1877 1.2 simonb if (wc->wc_generation++ == 0) { 1878 1.2 simonb error = wapbl_write_commit(wl, head, tail); 1879 1.2 simonb /* 1880 1.2 simonb * This panic should be able to be removed if we do the 1881 1.2 simonb * zero'ing mentioned above, and we are certain to roll 1882 1.2 simonb * back generation number on failure. 1883 1.2 simonb */ 1884 1.2 simonb if (error) 1885 1.2 simonb panic("wapbl_write_commit: error writing duplicate " 1886 1.2 simonb "log header: %d\n", error); 1887 1.2 simonb } 1888 1.2 simonb return 0; 1889 1.2 simonb } 1890 1.2 simonb 1891 1.2 simonb /* Returns new offset value */ 1892 1.2 simonb static int 1893 1.2 simonb wapbl_write_blocks(struct wapbl *wl, off_t *offp) 1894 1.2 simonb { 1895 1.2 simonb struct wapbl_wc_blocklist *wc = 1896 1.2 simonb (struct wapbl_wc_blocklist *)wl->wl_wc_scratch; 1897 1.2 simonb int blocklen = 1<<wl->wl_log_dev_bshift; 1898 1.2 simonb int bph; 1899 1.2 simonb struct buf *bp; 1900 1.2 simonb off_t off = *offp; 1901 1.2 simonb int error; 1902 1.2 simonb 1903 1.2 simonb KASSERT(rw_write_held(&wl->wl_rwlock)); 1904 1.2 simonb 1905 1.2 simonb bph = (blocklen - offsetof(struct wapbl_wc_blocklist, wc_blocks)) / 1906 1.2 simonb sizeof(((struct wapbl_wc_blocklist *)0)->wc_blocks[0]); 1907 1.2 simonb 1908 1.2 simonb bp = LIST_FIRST(&wl->wl_bufs); 1909 1.2 simonb 1910 1.2 simonb while (bp) { 1911 1.2 simonb int cnt; 1912 1.2 simonb struct buf *obp = bp; 1913 1.2 simonb 1914 1.2 simonb KASSERT(bp->b_flags & B_LOCKED); 1915 1.2 simonb 1916 1.2 simonb wc->wc_type = WAPBL_WC_BLOCKS; 1917 1.2 simonb wc->wc_len = blocklen; 1918 1.2 simonb wc->wc_blkcount = 0; 1919 1.2 simonb while (bp && (wc->wc_blkcount < bph)) { 1920 1.2 simonb /* 1921 1.2 simonb * Make sure all the physical block numbers are up to 1922 1.2 simonb * date. If this is not always true on a given 1923 1.2 simonb * filesystem, then VOP_BMAP must be called. We 1924 1.2 simonb * could call VOP_BMAP here, or else in the filesystem 1925 1.2 simonb * specific flush callback, although neither of those 1926 1.2 simonb * solutions allow us to take the vnode lock. If a 1927 1.2 simonb * filesystem requires that we must take the vnode lock 1928 1.2 simonb * to call VOP_BMAP, then we can probably do it in 1929 1.2 simonb * bwrite when the vnode lock should already be held 1930 1.2 simonb * by the invoking code. 1931 1.2 simonb */ 1932 1.2 simonb KASSERT((bp->b_vp->v_type == VBLK) || 1933 1.2 simonb (bp->b_blkno != bp->b_lblkno)); 1934 1.2 simonb KASSERT(bp->b_blkno > 0); 1935 1.2 simonb 1936 1.2 simonb wc->wc_blocks[wc->wc_blkcount].wc_daddr = bp->b_blkno; 1937 1.2 simonb wc->wc_blocks[wc->wc_blkcount].wc_dlen = bp->b_bcount; 1938 1.2 simonb wc->wc_len += bp->b_bcount; 1939 1.2 simonb wc->wc_blkcount++; 1940 1.2 simonb bp = LIST_NEXT(bp, b_wapbllist); 1941 1.2 simonb } 1942 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_WRITE, 1943 1.2 simonb ("wapbl_write_blocks: len = %u off = %"PRIdMAX"\n", 1944 1.2 simonb wc->wc_len, (intmax_t)off)); 1945 1.2 simonb 1946 1.2 simonb error = wapbl_circ_write(wl, wc, blocklen, &off); 1947 1.2 simonb if (error) 1948 1.2 simonb return error; 1949 1.2 simonb bp = obp; 1950 1.2 simonb cnt = 0; 1951 1.2 simonb while (bp && (cnt++ < bph)) { 1952 1.2 simonb error = wapbl_circ_write(wl, bp->b_data, 1953 1.2 simonb bp->b_bcount, &off); 1954 1.2 simonb if (error) 1955 1.2 simonb return error; 1956 1.2 simonb bp = LIST_NEXT(bp, b_wapbllist); 1957 1.2 simonb } 1958 1.2 simonb } 1959 1.2 simonb *offp = off; 1960 1.2 simonb return 0; 1961 1.2 simonb } 1962 1.2 simonb 1963 1.2 simonb static int 1964 1.2 simonb wapbl_write_revocations(struct wapbl *wl, off_t *offp) 1965 1.2 simonb { 1966 1.2 simonb struct wapbl_wc_blocklist *wc = 1967 1.2 simonb (struct wapbl_wc_blocklist *)wl->wl_wc_scratch; 1968 1.2 simonb int i; 1969 1.2 simonb int blocklen = 1<<wl->wl_log_dev_bshift; 1970 1.2 simonb int bph; 1971 1.2 simonb off_t off = *offp; 1972 1.2 simonb int error; 1973 1.2 simonb 1974 1.2 simonb if (wl->wl_dealloccnt == 0) 1975 1.2 simonb return 0; 1976 1.2 simonb 1977 1.2 simonb bph = (blocklen - offsetof(struct wapbl_wc_blocklist, wc_blocks)) / 1978 1.2 simonb sizeof(((struct wapbl_wc_blocklist *)0)->wc_blocks[0]); 1979 1.2 simonb 1980 1.2 simonb i = 0; 1981 1.2 simonb while (i < wl->wl_dealloccnt) { 1982 1.2 simonb wc->wc_type = WAPBL_WC_REVOCATIONS; 1983 1.2 simonb wc->wc_len = blocklen; 1984 1.2 simonb wc->wc_blkcount = 0; 1985 1.2 simonb while ((i < wl->wl_dealloccnt) && (wc->wc_blkcount < bph)) { 1986 1.2 simonb wc->wc_blocks[wc->wc_blkcount].wc_daddr = 1987 1.2 simonb wl->wl_deallocblks[i]; 1988 1.2 simonb wc->wc_blocks[wc->wc_blkcount].wc_dlen = 1989 1.2 simonb wl->wl_dealloclens[i]; 1990 1.2 simonb wc->wc_blkcount++; 1991 1.2 simonb i++; 1992 1.2 simonb } 1993 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_WRITE, 1994 1.2 simonb ("wapbl_write_revocations: len = %u off = %"PRIdMAX"\n", 1995 1.2 simonb wc->wc_len, (intmax_t)off)); 1996 1.2 simonb error = wapbl_circ_write(wl, wc, blocklen, &off); 1997 1.2 simonb if (error) 1998 1.2 simonb return error; 1999 1.2 simonb } 2000 1.2 simonb *offp = off; 2001 1.2 simonb return 0; 2002 1.2 simonb } 2003 1.2 simonb 2004 1.2 simonb static int 2005 1.2 simonb wapbl_write_inodes(struct wapbl *wl, off_t *offp) 2006 1.2 simonb { 2007 1.2 simonb struct wapbl_wc_inodelist *wc = 2008 1.2 simonb (struct wapbl_wc_inodelist *)wl->wl_wc_scratch; 2009 1.2 simonb int i; 2010 1.2 simonb int blocklen = 1<<wl->wl_log_dev_bshift; 2011 1.2 simonb off_t off = *offp; 2012 1.2 simonb int error; 2013 1.2 simonb 2014 1.2 simonb struct wapbl_ino_head *wih; 2015 1.2 simonb struct wapbl_ino *wi; 2016 1.2 simonb int iph; 2017 1.2 simonb 2018 1.2 simonb iph = (blocklen - offsetof(struct wapbl_wc_inodelist, wc_inodes)) / 2019 1.2 simonb sizeof(((struct wapbl_wc_inodelist *)0)->wc_inodes[0]); 2020 1.2 simonb 2021 1.2 simonb i = 0; 2022 1.2 simonb wih = &wl->wl_inohash[0]; 2023 1.2 simonb wi = 0; 2024 1.2 simonb do { 2025 1.2 simonb wc->wc_type = WAPBL_WC_INODES; 2026 1.2 simonb wc->wc_len = blocklen; 2027 1.2 simonb wc->wc_inocnt = 0; 2028 1.2 simonb wc->wc_clear = (i == 0); 2029 1.2 simonb while ((i < wl->wl_inohashcnt) && (wc->wc_inocnt < iph)) { 2030 1.2 simonb while (!wi) { 2031 1.2 simonb KASSERT((wih - &wl->wl_inohash[0]) 2032 1.2 simonb <= wl->wl_inohashmask); 2033 1.2 simonb wi = LIST_FIRST(wih++); 2034 1.2 simonb } 2035 1.2 simonb wc->wc_inodes[wc->wc_inocnt].wc_inumber = wi->wi_ino; 2036 1.2 simonb wc->wc_inodes[wc->wc_inocnt].wc_imode = wi->wi_mode; 2037 1.2 simonb wc->wc_inocnt++; 2038 1.2 simonb i++; 2039 1.2 simonb wi = LIST_NEXT(wi, wi_hash); 2040 1.2 simonb } 2041 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_WRITE, 2042 1.2 simonb ("wapbl_write_inodes: len = %u off = %"PRIdMAX"\n", 2043 1.2 simonb wc->wc_len, (intmax_t)off)); 2044 1.2 simonb error = wapbl_circ_write(wl, wc, blocklen, &off); 2045 1.2 simonb if (error) 2046 1.2 simonb return error; 2047 1.2 simonb } while (i < wl->wl_inohashcnt); 2048 1.2 simonb 2049 1.2 simonb *offp = off; 2050 1.2 simonb return 0; 2051 1.2 simonb } 2052 1.2 simonb 2053 1.2 simonb #endif /* _KERNEL */ 2054 1.2 simonb 2055 1.2 simonb /****************************************************************/ 2056 1.2 simonb 2057 1.2 simonb #ifdef _KERNEL 2058 1.2 simonb static struct pool wapbl_blk_pool; 2059 1.2 simonb static int wapbl_blk_pool_refcount; 2060 1.2 simonb #endif 2061 1.2 simonb struct wapbl_blk { 2062 1.2 simonb LIST_ENTRY(wapbl_blk) wb_hash; 2063 1.2 simonb daddr_t wb_blk; 2064 1.2 simonb off_t wb_off; /* Offset of this block in the log */ 2065 1.2 simonb }; 2066 1.2 simonb #define WAPBL_BLKPOOL_MIN 83 2067 1.2 simonb 2068 1.2 simonb static void 2069 1.2 simonb wapbl_blkhash_init(struct wapbl_replay *wr, u_int size) 2070 1.2 simonb { 2071 1.2 simonb if (size < WAPBL_BLKPOOL_MIN) 2072 1.2 simonb size = WAPBL_BLKPOOL_MIN; 2073 1.2 simonb KASSERT(wr->wr_blkhash == 0); 2074 1.2 simonb #ifdef _KERNEL 2075 1.2 simonb wr->wr_blkhash = hashinit(size, HASH_LIST, true, &wr->wr_blkhashmask); 2076 1.2 simonb if (atomic_inc_uint_nv(&wapbl_blk_pool_refcount) == 1) { 2077 1.2 simonb pool_init(&wapbl_blk_pool, sizeof(struct wapbl_blk), 0, 0, 0, 2078 1.2 simonb "wapblblkpl", &pool_allocator_nointr, IPL_NONE); 2079 1.2 simonb } 2080 1.2 simonb #else /* ! _KERNEL */ 2081 1.2 simonb /* Manually implement hashinit */ 2082 1.2 simonb { 2083 1.2 simonb int i; 2084 1.2 simonb unsigned long hashsize; 2085 1.2 simonb for (hashsize = 1; hashsize < size; hashsize <<= 1) 2086 1.2 simonb continue; 2087 1.2 simonb wr->wr_blkhash = wapbl_malloc(hashsize * sizeof(*wr->wr_blkhash)); 2088 1.3.8.2 snj for (i = 0; i < hashsize; i++) 2089 1.2 simonb LIST_INIT(&wr->wr_blkhash[i]); 2090 1.2 simonb wr->wr_blkhashmask = hashsize - 1; 2091 1.2 simonb } 2092 1.2 simonb #endif /* ! _KERNEL */ 2093 1.2 simonb } 2094 1.2 simonb 2095 1.2 simonb static void 2096 1.2 simonb wapbl_blkhash_free(struct wapbl_replay *wr) 2097 1.2 simonb { 2098 1.2 simonb KASSERT(wr->wr_blkhashcnt == 0); 2099 1.2 simonb #ifdef _KERNEL 2100 1.2 simonb hashdone(wr->wr_blkhash, HASH_LIST, wr->wr_blkhashmask); 2101 1.2 simonb if (atomic_dec_uint_nv(&wapbl_blk_pool_refcount) == 0) { 2102 1.2 simonb pool_destroy(&wapbl_blk_pool); 2103 1.2 simonb } 2104 1.2 simonb #else /* ! _KERNEL */ 2105 1.2 simonb wapbl_free(wr->wr_blkhash); 2106 1.2 simonb #endif /* ! _KERNEL */ 2107 1.2 simonb } 2108 1.2 simonb 2109 1.2 simonb static struct wapbl_blk * 2110 1.2 simonb wapbl_blkhash_get(struct wapbl_replay *wr, daddr_t blk) 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 wbh = &wr->wr_blkhash[blk & wr->wr_blkhashmask]; 2115 1.2 simonb LIST_FOREACH(wb, wbh, wb_hash) { 2116 1.2 simonb if (blk == wb->wb_blk) 2117 1.2 simonb return wb; 2118 1.2 simonb } 2119 1.2 simonb return 0; 2120 1.2 simonb } 2121 1.2 simonb 2122 1.2 simonb static void 2123 1.2 simonb wapbl_blkhash_ins(struct wapbl_replay *wr, daddr_t blk, off_t off) 2124 1.2 simonb { 2125 1.2 simonb struct wapbl_blk_head *wbh; 2126 1.2 simonb struct wapbl_blk *wb; 2127 1.2 simonb wb = wapbl_blkhash_get(wr, blk); 2128 1.2 simonb if (wb) { 2129 1.2 simonb KASSERT(wb->wb_blk == blk); 2130 1.2 simonb wb->wb_off = off; 2131 1.2 simonb } else { 2132 1.2 simonb #ifdef _KERNEL 2133 1.2 simonb wb = pool_get(&wapbl_blk_pool, PR_WAITOK); 2134 1.2 simonb #else /* ! _KERNEL */ 2135 1.2 simonb wb = wapbl_malloc(sizeof(*wb)); 2136 1.2 simonb #endif /* ! _KERNEL */ 2137 1.2 simonb wb->wb_blk = blk; 2138 1.2 simonb wb->wb_off = off; 2139 1.2 simonb wbh = &wr->wr_blkhash[blk & wr->wr_blkhashmask]; 2140 1.2 simonb LIST_INSERT_HEAD(wbh, wb, wb_hash); 2141 1.2 simonb wr->wr_blkhashcnt++; 2142 1.2 simonb } 2143 1.2 simonb } 2144 1.2 simonb 2145 1.2 simonb static void 2146 1.2 simonb wapbl_blkhash_rem(struct wapbl_replay *wr, daddr_t blk) 2147 1.2 simonb { 2148 1.2 simonb struct wapbl_blk *wb = wapbl_blkhash_get(wr, blk); 2149 1.2 simonb if (wb) { 2150 1.2 simonb KASSERT(wr->wr_blkhashcnt > 0); 2151 1.2 simonb wr->wr_blkhashcnt--; 2152 1.2 simonb LIST_REMOVE(wb, wb_hash); 2153 1.2 simonb #ifdef _KERNEL 2154 1.2 simonb pool_put(&wapbl_blk_pool, wb); 2155 1.2 simonb #else /* ! _KERNEL */ 2156 1.2 simonb wapbl_free(wb); 2157 1.2 simonb #endif /* ! _KERNEL */ 2158 1.2 simonb } 2159 1.2 simonb } 2160 1.2 simonb 2161 1.2 simonb static void 2162 1.2 simonb wapbl_blkhash_clear(struct wapbl_replay *wr) 2163 1.2 simonb { 2164 1.2 simonb int i; 2165 1.2 simonb for (i = 0; i <= wr->wr_blkhashmask; i++) { 2166 1.2 simonb struct wapbl_blk *wb; 2167 1.2 simonb 2168 1.2 simonb while ((wb = LIST_FIRST(&wr->wr_blkhash[i]))) { 2169 1.2 simonb KASSERT(wr->wr_blkhashcnt > 0); 2170 1.2 simonb wr->wr_blkhashcnt--; 2171 1.2 simonb LIST_REMOVE(wb, wb_hash); 2172 1.2 simonb #ifdef _KERNEL 2173 1.2 simonb pool_put(&wapbl_blk_pool, wb); 2174 1.2 simonb #else /* ! _KERNEL */ 2175 1.2 simonb wapbl_free(wb); 2176 1.2 simonb #endif /* ! _KERNEL */ 2177 1.2 simonb } 2178 1.2 simonb } 2179 1.2 simonb KASSERT(wr->wr_blkhashcnt == 0); 2180 1.2 simonb } 2181 1.2 simonb 2182 1.2 simonb /****************************************************************/ 2183 1.2 simonb 2184 1.2 simonb static int 2185 1.2 simonb wapbl_circ_read(struct wapbl_replay *wr, void *data, size_t len, off_t *offp) 2186 1.2 simonb { 2187 1.2 simonb size_t slen; 2188 1.2 simonb struct wapbl_wc_header *wc = &wr->wr_wc_header; 2189 1.2 simonb off_t off = *offp; 2190 1.2 simonb int error; 2191 1.2 simonb 2192 1.2 simonb KASSERT(((len >> wc->wc_log_dev_bshift) << 2193 1.2 simonb wc->wc_log_dev_bshift) == len); 2194 1.2 simonb if (off < wc->wc_circ_off) 2195 1.2 simonb off = wc->wc_circ_off; 2196 1.2 simonb slen = wc->wc_circ_off + wc->wc_circ_size - off; 2197 1.2 simonb if (slen < len) { 2198 1.2 simonb error = wapbl_read(data, slen, wr->wr_devvp, 2199 1.2 simonb wr->wr_logpbn + (off >> wc->wc_log_dev_bshift)); 2200 1.2 simonb if (error) 2201 1.2 simonb return error; 2202 1.2 simonb data = (uint8_t *)data + slen; 2203 1.2 simonb len -= slen; 2204 1.2 simonb off = wc->wc_circ_off; 2205 1.2 simonb } 2206 1.2 simonb error = wapbl_read(data, len, wr->wr_devvp, 2207 1.2 simonb wr->wr_logpbn + (off >> wc->wc_log_dev_bshift)); 2208 1.2 simonb if (error) 2209 1.2 simonb return error; 2210 1.2 simonb off += len; 2211 1.2 simonb if (off >= wc->wc_circ_off + wc->wc_circ_size) 2212 1.2 simonb off = wc->wc_circ_off; 2213 1.2 simonb *offp = off; 2214 1.2 simonb return 0; 2215 1.2 simonb } 2216 1.2 simonb 2217 1.2 simonb static void 2218 1.2 simonb wapbl_circ_advance(struct wapbl_replay *wr, size_t len, off_t *offp) 2219 1.2 simonb { 2220 1.2 simonb size_t slen; 2221 1.2 simonb struct wapbl_wc_header *wc = &wr->wr_wc_header; 2222 1.2 simonb off_t off = *offp; 2223 1.2 simonb 2224 1.2 simonb KASSERT(((len >> wc->wc_log_dev_bshift) << 2225 1.2 simonb wc->wc_log_dev_bshift) == len); 2226 1.2 simonb 2227 1.2 simonb if (off < wc->wc_circ_off) 2228 1.2 simonb off = wc->wc_circ_off; 2229 1.2 simonb slen = wc->wc_circ_off + wc->wc_circ_size - off; 2230 1.2 simonb if (slen < len) { 2231 1.2 simonb len -= slen; 2232 1.2 simonb off = wc->wc_circ_off; 2233 1.2 simonb } 2234 1.2 simonb off += len; 2235 1.2 simonb if (off >= wc->wc_circ_off + wc->wc_circ_size) 2236 1.2 simonb off = wc->wc_circ_off; 2237 1.2 simonb *offp = off; 2238 1.2 simonb } 2239 1.2 simonb 2240 1.2 simonb /****************************************************************/ 2241 1.2 simonb 2242 1.2 simonb int 2243 1.2 simonb wapbl_replay_start(struct wapbl_replay **wrp, struct vnode *vp, 2244 1.2 simonb daddr_t off, size_t count, size_t blksize) 2245 1.2 simonb { 2246 1.2 simonb struct wapbl_replay *wr; 2247 1.2 simonb int error; 2248 1.2 simonb struct vnode *devvp; 2249 1.2 simonb daddr_t logpbn; 2250 1.2 simonb uint8_t *scratch; 2251 1.2 simonb struct wapbl_wc_header *wch; 2252 1.2 simonb struct wapbl_wc_header *wch2; 2253 1.2 simonb /* Use this until we read the actual log header */ 2254 1.2 simonb int log_dev_bshift = DEV_BSHIFT; 2255 1.2 simonb size_t used; 2256 1.2 simonb 2257 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_REPLAY, 2258 1.2 simonb ("wapbl_replay_start: vp=%p off=%"PRId64 " count=%zu blksize=%zu\n", 2259 1.2 simonb vp, off, count, blksize)); 2260 1.2 simonb 2261 1.2 simonb if (off < 0) 2262 1.2 simonb return EINVAL; 2263 1.2 simonb 2264 1.2 simonb if (blksize < DEV_BSIZE) 2265 1.2 simonb return EINVAL; 2266 1.2 simonb if (blksize % DEV_BSIZE) 2267 1.2 simonb return EINVAL; 2268 1.2 simonb 2269 1.2 simonb #ifdef _KERNEL 2270 1.2 simonb #if 0 2271 1.2 simonb /* XXX vp->v_size isn't reliably set for VBLK devices, 2272 1.2 simonb * especially root. However, we might still want to verify 2273 1.2 simonb * that the full load is readable */ 2274 1.2 simonb if ((off + count) * blksize > vp->v_size) 2275 1.2 simonb return EINVAL; 2276 1.2 simonb #endif 2277 1.2 simonb 2278 1.2 simonb if ((error = VOP_BMAP(vp, off, &devvp, &logpbn, 0)) != 0) { 2279 1.2 simonb return error; 2280 1.2 simonb } 2281 1.2 simonb #else /* ! _KERNEL */ 2282 1.2 simonb devvp = vp; 2283 1.2 simonb logpbn = off; 2284 1.2 simonb #endif /* ! _KERNEL */ 2285 1.2 simonb 2286 1.2 simonb scratch = wapbl_malloc(MAXBSIZE); 2287 1.2 simonb 2288 1.2 simonb error = wapbl_read(scratch, 2<<log_dev_bshift, devvp, logpbn); 2289 1.2 simonb if (error) 2290 1.2 simonb goto errout; 2291 1.2 simonb 2292 1.2 simonb wch = (struct wapbl_wc_header *)scratch; 2293 1.2 simonb wch2 = 2294 1.2 simonb (struct wapbl_wc_header *)(scratch + (1<<log_dev_bshift)); 2295 1.2 simonb /* XXX verify checksums and magic numbers */ 2296 1.2 simonb if (wch->wc_type != WAPBL_WC_HEADER) { 2297 1.2 simonb printf("Unrecognized wapbl magic: 0x%08x\n", wch->wc_type); 2298 1.2 simonb error = EFTYPE; 2299 1.2 simonb goto errout; 2300 1.2 simonb } 2301 1.2 simonb 2302 1.2 simonb if (wch2->wc_generation > wch->wc_generation) 2303 1.2 simonb wch = wch2; 2304 1.2 simonb 2305 1.2 simonb wr = wapbl_calloc(1, sizeof(*wr)); 2306 1.2 simonb 2307 1.2 simonb wr->wr_logvp = vp; 2308 1.2 simonb wr->wr_devvp = devvp; 2309 1.2 simonb wr->wr_logpbn = logpbn; 2310 1.2 simonb 2311 1.2 simonb wr->wr_scratch = scratch; 2312 1.2 simonb 2313 1.2 simonb memcpy(&wr->wr_wc_header, wch, sizeof(wr->wr_wc_header)); 2314 1.2 simonb 2315 1.2 simonb used = wapbl_space_used(wch->wc_circ_size, wch->wc_head, wch->wc_tail); 2316 1.2 simonb 2317 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_REPLAY, 2318 1.2 simonb ("wapbl_replay: head=%"PRId64" tail=%"PRId64" off=%"PRId64 2319 1.2 simonb " len=%"PRId64" used=%zu\n", 2320 1.2 simonb wch->wc_head, wch->wc_tail, wch->wc_circ_off, 2321 1.2 simonb wch->wc_circ_size, used)); 2322 1.2 simonb 2323 1.2 simonb wapbl_blkhash_init(wr, (used >> wch->wc_fs_dev_bshift)); 2324 1.2 simonb error = wapbl_replay_prescan(wr); 2325 1.2 simonb if (error) { 2326 1.2 simonb wapbl_replay_stop(wr); 2327 1.2 simonb wapbl_replay_free(wr); 2328 1.2 simonb return error; 2329 1.2 simonb } 2330 1.2 simonb 2331 1.2 simonb error = wapbl_replay_get_inodes(wr); 2332 1.2 simonb if (error) { 2333 1.2 simonb wapbl_replay_stop(wr); 2334 1.2 simonb wapbl_replay_free(wr); 2335 1.2 simonb return error; 2336 1.2 simonb } 2337 1.2 simonb 2338 1.2 simonb *wrp = wr; 2339 1.2 simonb return 0; 2340 1.2 simonb 2341 1.2 simonb errout: 2342 1.2 simonb wapbl_free(scratch); 2343 1.2 simonb return error; 2344 1.2 simonb } 2345 1.2 simonb 2346 1.2 simonb void 2347 1.2 simonb wapbl_replay_stop(struct wapbl_replay *wr) 2348 1.2 simonb { 2349 1.2 simonb 2350 1.2 simonb WAPBL_PRINTF(WAPBL_PRINT_REPLAY, ("wapbl_replay_stop called\n")); 2351 1.2 simonb 2352 1.2 simonb KDASSERT(wapbl_replay_isopen(wr)); 2353 1.2 simonb 2354 1.2 simonb wapbl_free(wr->wr_scratch); 2355 1.2 simonb wr->wr_scratch = 0; 2356 1.2 simonb 2357 1.2 simonb wr->wr_logvp = 0; 2358 1.2 simonb 2359 1.2 simonb wapbl_blkhash_clear(wr); 2360 1.2 simonb wapbl_blkhash_free(wr); 2361 1.2 simonb } 2362 1.2 simonb 2363 1.2 simonb void 2364 1.2 simonb wapbl_replay_free(struct wapbl_replay *wr) 2365 1.2 simonb { 2366 1.2 simonb 2367 1.2 simonb KDASSERT(!wapbl_replay_isopen(wr)); 2368 1.2 simonb 2369 1.2 simonb if (wr->wr_inodes) 2370 1.2 simonb wapbl_free(wr->wr_inodes); 2371 1.2 simonb wapbl_free(wr); 2372 1.2 simonb } 2373 1.2 simonb 2374 1.2 simonb int 2375 1.2 simonb wapbl_replay_isopen1(struct wapbl_replay *wr) 2376 1.2 simonb { 2377 1.2 simonb 2378 1.2 simonb return wapbl_replay_isopen(wr); 2379 1.2 simonb } 2380 1.2 simonb 2381 1.2 simonb static int 2382 1.2 simonb wapbl_replay_prescan(struct wapbl_replay *wr) 2383 1.2 simonb { 2384 1.2 simonb off_t off; 2385 1.2 simonb struct wapbl_wc_header *wch = &wr->wr_wc_header; 2386 1.2 simonb int error; 2387 1.2 simonb 2388 1.2 simonb int logblklen = 1<<wch->wc_log_dev_bshift; 2389 1.2 simonb int fsblklen = 1<<wch->wc_fs_dev_bshift; 2390 1.2 simonb 2391 1.2 simonb wapbl_blkhash_clear(wr); 2392 1.2 simonb 2393 1.2 simonb off = wch->wc_tail; 2394 1.2 simonb while (off != wch->wc_head) { 2395 1.2 simonb struct wapbl_wc_null *wcn; 2396 1.2 simonb off_t saveoff = off; 2397 1.2 simonb error = wapbl_circ_read(wr, wr->wr_scratch, logblklen, &off); 2398 1.2 simonb if (error) 2399 1.2 simonb goto errout; 2400 1.2 simonb wcn = (struct wapbl_wc_null *)wr->wr_scratch; 2401 1.2 simonb switch (wcn->wc_type) { 2402 1.2 simonb case WAPBL_WC_BLOCKS: 2403 1.2 simonb { 2404 1.2 simonb struct wapbl_wc_blocklist *wc = 2405 1.2 simonb (struct wapbl_wc_blocklist *)wr->wr_scratch; 2406 1.2 simonb int i; 2407 1.2 simonb for (i = 0; i < wc->wc_blkcount; i++) { 2408 1.2 simonb int j, n; 2409 1.2 simonb /* 2410 1.2 simonb * Enter each physical block into the 2411 1.2 simonb * hashtable independently 2412 1.2 simonb */ 2413 1.2 simonb n = wc->wc_blocks[i].wc_dlen >> 2414 1.2 simonb wch->wc_fs_dev_bshift; 2415 1.2 simonb for (j = 0; j < n; j++) { 2416 1.2 simonb wapbl_blkhash_ins(wr, 2417 1.2 simonb wc->wc_blocks[i].wc_daddr + j, 2418 1.2 simonb off); 2419 1.2 simonb wapbl_circ_advance(wr, 2420 1.2 simonb fsblklen, &off); 2421 1.2 simonb } 2422 1.2 simonb } 2423 1.2 simonb } 2424 1.2 simonb break; 2425 1.2 simonb 2426 1.2 simonb case WAPBL_WC_REVOCATIONS: 2427 1.2 simonb { 2428 1.2 simonb struct wapbl_wc_blocklist *wc = 2429 1.2 simonb (struct wapbl_wc_blocklist *)wr->wr_scratch; 2430 1.2 simonb int i; 2431 1.2 simonb for (i = 0; i < wc->wc_blkcount; i++) { 2432 1.2 simonb int j, n; 2433 1.2 simonb /* 2434 1.2 simonb * Remove any blocks found from the 2435 1.2 simonb * hashtable 2436 1.2 simonb */ 2437 1.2 simonb n = wc->wc_blocks[i].wc_dlen >> 2438 1.2 simonb wch->wc_fs_dev_bshift; 2439 1.2 simonb for (j = 0; j < n; j++) { 2440 1.2 simonb wapbl_blkhash_rem(wr, 2441 1.2 simonb wc->wc_blocks[i].wc_daddr + j); 2442 1.2 simonb } 2443 1.2 simonb } 2444 1.2 simonb } 2445 1.2 simonb break; 2446 1.2 simonb 2447 1.2 simonb case WAPBL_WC_INODES: 2448 1.2 simonb { 2449 1.2 simonb struct wapbl_wc_inodelist *wc = 2450 1.2 simonb (struct wapbl_wc_inodelist *)wr->wr_scratch; 2451 1.2 simonb /* 2452 1.2 simonb * Keep track of where we found this so we 2453 1.2 simonb * can use it later 2454 1.2 simonb */ 2455 1.2 simonb if (wc->wc_clear) { 2456 1.2 simonb wr->wr_inodestail = saveoff; 2457 1.2 simonb wr->wr_inodescnt = 0; 2458 1.2 simonb } 2459 1.2 simonb if (wr->wr_inodestail) 2460 1.2 simonb wr->wr_inodeshead = off; 2461 1.2 simonb wr->wr_inodescnt += wc->wc_inocnt; 2462 1.2 simonb } 2463 1.2 simonb break; 2464 1.2 simonb default: 2465 1.2 simonb printf("Unrecognized wapbl type: 0x%08x\n", 2466 1.2 simonb wcn->wc_type); 2467 1.2 simonb error = EFTYPE; 2468 1.2 simonb goto errout; 2469 1.2 simonb } 2470 1.2 simonb wapbl_circ_advance(wr, wcn->wc_len, &saveoff); 2471 1.2 simonb if (off != saveoff) { 2472 1.2 simonb printf("wapbl_replay: corrupted records\n"); 2473 1.2 simonb error = EFTYPE; 2474 1.2 simonb goto errout; 2475 1.2 simonb } 2476 1.2 simonb } 2477 1.2 simonb return 0; 2478 1.2 simonb 2479 1.2 simonb errout: 2480 1.2 simonb wapbl_blkhash_clear(wr); 2481 1.2 simonb return error; 2482 1.2 simonb } 2483 1.2 simonb 2484 1.2 simonb static int 2485 1.2 simonb wapbl_replay_get_inodes(struct wapbl_replay *wr) 2486 1.2 simonb { 2487 1.2 simonb off_t off; 2488 1.2 simonb struct wapbl_wc_header *wch = &wr->wr_wc_header; 2489 1.2 simonb int logblklen = 1<<wch->wc_log_dev_bshift; 2490 1.2 simonb int cnt= 0; 2491 1.2 simonb 2492 1.2 simonb KDASSERT(wapbl_replay_isopen(wr)); 2493 1.2 simonb 2494 1.2 simonb if (wr->wr_inodescnt == 0) 2495 1.2 simonb return 0; 2496 1.2 simonb 2497 1.2 simonb KASSERT(!wr->wr_inodes); 2498 1.2 simonb 2499 1.2 simonb wr->wr_inodes = wapbl_malloc(wr->wr_inodescnt*sizeof(wr->wr_inodes[0])); 2500 1.2 simonb 2501 1.2 simonb off = wr->wr_inodestail; 2502 1.2 simonb 2503 1.2 simonb while (off != wr->wr_inodeshead) { 2504 1.2 simonb struct wapbl_wc_null *wcn; 2505 1.2 simonb int error; 2506 1.2 simonb off_t saveoff = off; 2507 1.2 simonb error = wapbl_circ_read(wr, wr->wr_scratch, logblklen, &off); 2508 1.2 simonb if (error) { 2509 1.2 simonb wapbl_free(wr->wr_inodes); 2510 1.2 simonb wr->wr_inodes = 0; 2511 1.2 simonb return error; 2512 1.2 simonb } 2513 1.2 simonb wcn = (struct wapbl_wc_null *)wr->wr_scratch; 2514 1.2 simonb switch (wcn->wc_type) { 2515 1.2 simonb case WAPBL_WC_BLOCKS: 2516 1.2 simonb case WAPBL_WC_REVOCATIONS: 2517 1.2 simonb break; 2518 1.2 simonb case WAPBL_WC_INODES: 2519 1.2 simonb { 2520 1.2 simonb struct wapbl_wc_inodelist *wc = 2521 1.2 simonb (struct wapbl_wc_inodelist *)wr->wr_scratch; 2522 1.2 simonb /* 2523 1.2 simonb * Keep track of where we found this so we 2524 1.2 simonb * can use it later 2525 1.2 simonb */ 2526 1.2 simonb if (wc->wc_clear) { 2527 1.2 simonb cnt = 0; 2528 1.2 simonb } 2529 1.2 simonb /* This memcpy assumes that wr_inodes is 2530 1.2 simonb * laid out the same as wc_inodes. */ 2531 1.2 simonb memcpy(&wr->wr_inodes[cnt], wc->wc_inodes, 2532 1.2 simonb wc->wc_inocnt*sizeof(wc->wc_inodes[0])); 2533 1.2 simonb cnt += wc->wc_inocnt; 2534 1.2 simonb } 2535 1.2 simonb break; 2536 1.2 simonb default: 2537 1.2 simonb KASSERT(0); 2538 1.2 simonb } 2539 1.2 simonb off = saveoff; 2540 1.2 simonb wapbl_circ_advance(wr, wcn->wc_len, &off); 2541 1.2 simonb } 2542 1.2 simonb KASSERT(cnt == wr->wr_inodescnt); 2543 1.2 simonb return 0; 2544 1.2 simonb } 2545 1.2 simonb 2546 1.2 simonb #ifdef DEBUG 2547 1.2 simonb int 2548 1.2 simonb wapbl_replay_verify(struct wapbl_replay *wr, struct vnode *fsdevvp) 2549 1.2 simonb { 2550 1.2 simonb off_t off; 2551 1.2 simonb struct wapbl_wc_header *wch = &wr->wr_wc_header; 2552 1.2 simonb int mismatchcnt = 0; 2553 1.2 simonb int logblklen = 1<<wch->wc_log_dev_bshift; 2554 1.2 simonb int fsblklen = 1<<wch->wc_fs_dev_bshift; 2555 1.2 simonb void *scratch1 = wapbl_malloc(MAXBSIZE); 2556 1.2 simonb void *scratch2 = wapbl_malloc(MAXBSIZE); 2557 1.2 simonb int error = 0; 2558 1.2 simonb 2559 1.2 simonb KDASSERT(wapbl_replay_isopen(wr)); 2560 1.2 simonb 2561 1.2 simonb off = wch->wc_tail; 2562 1.2 simonb while (off != wch->wc_head) { 2563 1.2 simonb struct wapbl_wc_null *wcn; 2564 1.2 simonb #ifdef DEBUG 2565 1.2 simonb off_t saveoff = off; 2566 1.2 simonb #endif 2567 1.2 simonb error = wapbl_circ_read(wr, wr->wr_scratch, logblklen, &off); 2568 1.2 simonb if (error) 2569 1.2 simonb goto out; 2570 1.2 simonb wcn = (struct wapbl_wc_null *)wr->wr_scratch; 2571 1.2 simonb switch (wcn->wc_type) { 2572 1.2 simonb case WAPBL_WC_BLOCKS: 2573 1.2 simonb { 2574 1.2 simonb struct wapbl_wc_blocklist *wc = 2575 1.2 simonb (struct wapbl_wc_blocklist *)wr->wr_scratch; 2576 1.2 simonb int i; 2577 1.2 simonb for (i = 0; i < wc->wc_blkcount; i++) { 2578 1.2 simonb int foundcnt = 0; 2579 1.2 simonb int dirtycnt = 0; 2580 1.2 simonb int j, n; 2581 1.2 simonb /* 2582 1.2 simonb * Check each physical block into the 2583 1.2 simonb * hashtable independently 2584 1.2 simonb */ 2585 1.2 simonb n = wc->wc_blocks[i].wc_dlen >> 2586 1.2 simonb wch->wc_fs_dev_bshift; 2587 1.2 simonb for (j = 0; j < n; j++) { 2588 1.2 simonb struct wapbl_blk *wb = 2589 1.2 simonb wapbl_blkhash_get(wr, 2590 1.2 simonb wc->wc_blocks[i].wc_daddr + j); 2591 1.2 simonb if (wb && (wb->wb_off == off)) { 2592 1.2 simonb foundcnt++; 2593 1.2 simonb error = 2594 1.2 simonb wapbl_circ_read(wr, 2595 1.2 simonb scratch1, fsblklen, 2596 1.2 simonb &off); 2597 1.2 simonb if (error) 2598 1.2 simonb goto out; 2599 1.2 simonb error = 2600 1.2 simonb wapbl_read(scratch2, 2601 1.2 simonb fsblklen, fsdevvp, 2602 1.2 simonb wb->wb_blk); 2603 1.2 simonb if (error) 2604 1.2 simonb goto out; 2605 1.2 simonb if (memcmp(scratch1, 2606 1.2 simonb scratch2, 2607 1.2 simonb fsblklen)) { 2608 1.2 simonb printf( 2609 1.2 simonb "wapbl_verify: mismatch block %"PRId64" at off %"PRIdMAX"\n", 2610 1.2 simonb wb->wb_blk, (intmax_t)off); 2611 1.2 simonb dirtycnt++; 2612 1.2 simonb mismatchcnt++; 2613 1.2 simonb } 2614 1.2 simonb } else { 2615 1.2 simonb wapbl_circ_advance(wr, 2616 1.2 simonb fsblklen, &off); 2617 1.2 simonb } 2618 1.2 simonb } 2619 1.2 simonb #if 0 2620 1.2 simonb /* 2621 1.2 simonb * If all of the blocks in an entry 2622 1.2 simonb * are clean, then remove all of its 2623 1.2 simonb * blocks from the hashtable since they 2624 1.2 simonb * never will need replay. 2625 1.2 simonb */ 2626 1.2 simonb if ((foundcnt != 0) && 2627 1.2 simonb (dirtycnt == 0)) { 2628 1.2 simonb off = saveoff; 2629 1.2 simonb wapbl_circ_advance(wr, 2630 1.2 simonb logblklen, &off); 2631 1.2 simonb for (j = 0; j < n; j++) { 2632 1.2 simonb struct wapbl_blk *wb = 2633 1.2 simonb wapbl_blkhash_get(wr, 2634 1.2 simonb wc->wc_blocks[i].wc_daddr + j); 2635 1.2 simonb if (wb && 2636 1.2 simonb (wb->wb_off == off)) { 2637 1.2 simonb wapbl_blkhash_rem(wr, wb->wb_blk); 2638 1.2 simonb } 2639 1.2 simonb wapbl_circ_advance(wr, 2640 1.2 simonb fsblklen, &off); 2641 1.2 simonb } 2642 1.2 simonb } 2643 1.2 simonb #endif 2644 1.2 simonb } 2645 1.2 simonb } 2646 1.2 simonb break; 2647 1.2 simonb case WAPBL_WC_REVOCATIONS: 2648 1.2 simonb case WAPBL_WC_INODES: 2649 1.2 simonb break; 2650 1.2 simonb default: 2651 1.2 simonb KASSERT(0); 2652 1.2 simonb } 2653 1.2 simonb #ifdef DEBUG 2654 1.2 simonb wapbl_circ_advance(wr, wcn->wc_len, &saveoff); 2655 1.2 simonb KASSERT(off == saveoff); 2656 1.2 simonb #endif 2657 1.2 simonb } 2658 1.2 simonb out: 2659 1.2 simonb wapbl_free(scratch1); 2660 1.2 simonb wapbl_free(scratch2); 2661 1.2 simonb if (!error && mismatchcnt) 2662 1.2 simonb error = EFTYPE; 2663 1.2 simonb return error; 2664 1.2 simonb } 2665 1.2 simonb #endif 2666 1.2 simonb 2667 1.2 simonb int 2668 1.2 simonb wapbl_replay_write(struct wapbl_replay *wr, struct vnode *fsdevvp) 2669 1.2 simonb { 2670 1.2 simonb off_t off; 2671 1.2 simonb struct wapbl_wc_header *wch = &wr->wr_wc_header; 2672 1.2 simonb int logblklen = 1<<wch->wc_log_dev_bshift; 2673 1.2 simonb int fsblklen = 1<<wch->wc_fs_dev_bshift; 2674 1.2 simonb void *scratch1 = wapbl_malloc(MAXBSIZE); 2675 1.2 simonb int error = 0; 2676 1.2 simonb 2677 1.2 simonb KDASSERT(wapbl_replay_isopen(wr)); 2678 1.2 simonb 2679 1.2 simonb /* 2680 1.2 simonb * This parses the journal for replay, although it could 2681 1.2 simonb * just as easily walk the hashtable instead. 2682 1.2 simonb */ 2683 1.2 simonb 2684 1.2 simonb off = wch->wc_tail; 2685 1.2 simonb while (off != wch->wc_head) { 2686 1.2 simonb struct wapbl_wc_null *wcn; 2687 1.2 simonb #ifdef DEBUG 2688 1.2 simonb off_t saveoff = off; 2689 1.2 simonb #endif 2690 1.2 simonb error = wapbl_circ_read(wr, wr->wr_scratch, logblklen, &off); 2691 1.2 simonb if (error) 2692 1.2 simonb goto out; 2693 1.2 simonb wcn = (struct wapbl_wc_null *)wr->wr_scratch; 2694 1.2 simonb switch (wcn->wc_type) { 2695 1.2 simonb case WAPBL_WC_BLOCKS: 2696 1.2 simonb { 2697 1.2 simonb struct wapbl_wc_blocklist *wc = 2698 1.2 simonb (struct wapbl_wc_blocklist *)wr->wr_scratch; 2699 1.2 simonb int i; 2700 1.2 simonb for (i = 0; i < wc->wc_blkcount; i++) { 2701 1.2 simonb int j, n; 2702 1.2 simonb /* 2703 1.2 simonb * Check each physical block against 2704 1.2 simonb * the hashtable independently 2705 1.2 simonb */ 2706 1.2 simonb n = wc->wc_blocks[i].wc_dlen >> 2707 1.2 simonb wch->wc_fs_dev_bshift; 2708 1.2 simonb for (j = 0; j < n; j++) { 2709 1.2 simonb struct wapbl_blk *wb = 2710 1.2 simonb wapbl_blkhash_get(wr, 2711 1.2 simonb wc->wc_blocks[i].wc_daddr + j); 2712 1.2 simonb if (wb && (wb->wb_off == off)) { 2713 1.2 simonb error = wapbl_circ_read( 2714 1.2 simonb wr, scratch1, 2715 1.2 simonb fsblklen, &off); 2716 1.2 simonb if (error) 2717 1.2 simonb goto out; 2718 1.2 simonb error = 2719 1.2 simonb wapbl_write(scratch1, 2720 1.2 simonb fsblklen, fsdevvp, 2721 1.2 simonb wb->wb_blk); 2722 1.2 simonb if (error) 2723 1.2 simonb goto out; 2724 1.2 simonb } else { 2725 1.2 simonb wapbl_circ_advance(wr, 2726 1.2 simonb fsblklen, &off); 2727 1.2 simonb } 2728 1.2 simonb } 2729 1.2 simonb } 2730 1.2 simonb } 2731 1.2 simonb break; 2732 1.2 simonb case WAPBL_WC_REVOCATIONS: 2733 1.2 simonb case WAPBL_WC_INODES: 2734 1.2 simonb break; 2735 1.2 simonb default: 2736 1.2 simonb KASSERT(0); 2737 1.2 simonb } 2738 1.2 simonb #ifdef DEBUG 2739 1.2 simonb wapbl_circ_advance(wr, wcn->wc_len, &saveoff); 2740 1.2 simonb KASSERT(off == saveoff); 2741 1.2 simonb #endif 2742 1.2 simonb } 2743 1.2 simonb out: 2744 1.2 simonb wapbl_free(scratch1); 2745 1.2 simonb return error; 2746 1.2 simonb } 2747 1.2 simonb 2748 1.2 simonb int 2749 1.2 simonb wapbl_replay_read(struct wapbl_replay *wr, void *data, daddr_t blk, long len) 2750 1.2 simonb { 2751 1.2 simonb struct wapbl_wc_header *wch = &wr->wr_wc_header; 2752 1.2 simonb int fsblklen = 1<<wch->wc_fs_dev_bshift; 2753 1.2 simonb 2754 1.2 simonb KDASSERT(wapbl_replay_isopen(wr)); 2755 1.2 simonb 2756 1.2 simonb KASSERT((len % fsblklen) == 0); 2757 1.2 simonb 2758 1.2 simonb while (len != 0) { 2759 1.2 simonb struct wapbl_blk *wb = wapbl_blkhash_get(wr, blk); 2760 1.2 simonb if (wb) { 2761 1.2 simonb off_t off = wb->wb_off; 2762 1.2 simonb int error; 2763 1.2 simonb error = wapbl_circ_read(wr, data, fsblklen, &off); 2764 1.2 simonb if (error) 2765 1.2 simonb return error; 2766 1.2 simonb } 2767 1.2 simonb data = (uint8_t *)data + fsblklen; 2768 1.2 simonb len -= fsblklen; 2769 1.2 simonb blk++; 2770 1.2 simonb } 2771 1.2 simonb return 0; 2772 1.2 simonb } 2773
Indexes created Fri Oct 03 02:09:57 GMT 2025