1 /* 2 * Copyright (c) 1982, 1986, 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)buf.h 8.9 (Berkeley) 3/30/95 39 * $FreeBSD: src/sys/sys/buf.h,v 1.88.2.10 2003/01/25 19:02:23 dillon Exp $ 40 * $DragonFly: src/sys/sys/buf.h,v 1.43 2008/02/05 07:58:41 dillon Exp $ 41 */ 42 43 #ifndef _SYS_BUF_H_ 44 #define _SYS_BUF_H_ 45 46 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES) 47 48 #ifndef _SYS_QUEUE_H_ 49 #include <sys/queue.h> 50 #endif 51 #ifndef _SYS_LOCK_H_ 52 #include <sys/lock.h> 53 #endif 54 #ifndef _SYS_DEVICE_H_ 55 #include <sys/device.h> 56 #endif 57 58 #ifndef _SYS_XIO_H_ 59 #include <sys/xio.h> 60 #endif 61 #ifndef _SYS_TREE_H_ 62 #include <sys/tree.h> 63 #endif 64 #ifndef _SYS_BIO_H_ 65 #include <sys/bio.h> 66 #endif 67 #ifndef _SYS_SPINLOCK_H_ 68 #include <sys/spinlock.h> 69 #endif 70 71 struct buf; 72 struct bio; 73 struct mount; 74 struct vnode; 75 struct xio; 76 77 #define NBUF_BIO 4 78 79 struct buf_rb_tree; 80 struct buf_rb_hash; 81 RB_PROTOTYPE2(buf_rb_tree, buf, b_rbnode, rb_buf_compare, off_t); 82 RB_PROTOTYPE2(buf_rb_hash, buf, b_rbhash, rb_buf_compare, off_t); 83 84 /* 85 * To avoid including <ufs/ffs/softdep.h> 86 */ 87 LIST_HEAD(workhead, worklist); 88 89 typedef enum buf_cmd { 90 BUF_CMD_DONE = 0, 91 BUF_CMD_READ, 92 BUF_CMD_WRITE, 93 BUF_CMD_FREEBLKS, 94 BUF_CMD_FORMAT 95 } buf_cmd_t; 96 97 /* 98 * The buffer header describes an I/O operation in the kernel. 99 * 100 * NOTES: 101 * b_bufsize represents the filesystem block size (for this particular 102 * block) and/or the allocation size or original request size. This 103 * field is NOT USED by lower device layers. VNode and device 104 * strategy routines WILL NEVER ACCESS THIS FIELD. 105 * 106 * b_bcount represents the I/O request size. Unless B_NOBCLIP is set, 107 * the device chain is allowed to clip b_bcount to accomodate the device 108 * EOF. Note that this is different from the byte oriented file EOF. 109 * If B_NOBCLIP is set, the device chain is required to generate an 110 * error if it would othrewise have to clip the request. Buffers 111 * obtained via getblk() automatically set B_NOBCLIP. It is important 112 * to note that EOF clipping via b_bcount is different from EOF clipping 113 * via returning a b_actual < b_bcount. B_NOBCLIP only effects block 114 * oriented EOF clipping (b_bcount modifications). 115 * 116 * b_actual represents the number of bytes of I/O that actually occured, 117 * whether an error occured or not. b_actual must be initialized to 0 118 * prior to initiating I/O as the device drivers will assume it to 119 * start at 0. 120 * 121 * b_dirtyoff, b_dirtyend. Buffers support piecemeal, unaligned 122 * ranges of dirty data that need to be written to backing store. 123 * The range is typically clipped at b_bcount (not b_bufsize). 124 * 125 * b_bio1 and b_bio2 represent the two primary I/O layers. Additional 126 * I/O layers are allocated out of the object cache and may also exist. 127 * 128 * b_bio1 is the logical layer and contains offset or block number 129 * data for the primary vnode, b_vp. I/O operations are almost 130 * universally initiated from the logical layer, so you will often 131 * see things like: vn_strategy(bp->b_vp, &bp->b_bio1). 132 * 133 * b_bio2 is the first physical layer (typically the slice-relative 134 * layer) and contains the translated offset or block number for 135 * the block device underlying a filesystem. Filesystems such as UFS 136 * will maintain cached translations and you may see them initiate 137 * a 'physical' I/O using vn_strategy(devvp, &bp->b_bio2). BUT, 138 * remember that the layering is relative to bp->b_vp, so the 139 * device-relative block numbers for buffer cache operations that occur 140 * directly on a block device will be in the first BIO layer. 141 * 142 * b_ops - initialized if a buffer has a bio_ops 143 * 144 * NOTE!!! Only the BIO subsystem accesses b_bio1 and b_bio2 directly. 145 * ALL STRATEGY LAYERS FOR BOTH VNODES AND DEVICES ONLY ACCESS THE BIO 146 * PASSED TO THEM, AND WILL PUSH ANOTHER BIO LAYER IF FORWARDING THE 147 * I/O DEEPER. In particular, a vn_strategy() or dev_dstrategy() 148 * call should not ever access buf->b_vp as this vnode may be totally 149 * unrelated to the vnode/device whos strategy routine was called. 150 */ 151 struct buf { 152 RB_ENTRY(buf) b_rbnode; /* RB node in vnode clean/dirty tree */ 153 RB_ENTRY(buf) b_rbhash; /* RB node in vnode hash tree */ 154 TAILQ_ENTRY(buf) b_freelist; /* Free list position if not active. */ 155 struct buf *b_cluster_next; /* Next buffer (cluster code) */ 156 struct vnode *b_vp; /* (vp, loffset) index */ 157 struct bio b_bio_array[NBUF_BIO]; /* BIO translation layers */ 158 u_int32_t b_flags; /* B_* flags. */ 159 unsigned short b_qindex; /* buffer queue index */ 160 unsigned short b_unused01; 161 struct lock b_lock; /* Buffer lock */ 162 buf_cmd_t b_cmd; /* I/O command */ 163 int b_bufsize; /* Allocated buffer size. */ 164 int b_runningbufspace; /* when I/O is running, pipelining */ 165 int b_bcount; /* Valid bytes in buffer. */ 166 int b_resid; /* Remaining I/O */ 167 int b_error; /* Error return */ 168 caddr_t b_data; /* Memory, superblocks, indirect etc. */ 169 caddr_t b_kvabase; /* base kva for buffer */ 170 int b_kvasize; /* size of kva for buffer */ 171 int b_dirtyoff; /* Offset in buffer of dirty region. */ 172 int b_dirtyend; /* Offset of end of dirty region. */ 173 struct xio b_xio; /* data buffer page list management */ 174 struct bio_ops *b_ops; /* bio_ops used w/ b_dep */ 175 struct workhead b_dep; /* List of filesystem dependencies. */ 176 u_int64_t b_tid; /* Transaction id */ 177 }; 178 179 /* 180 * XXX temporary 181 */ 182 #define b_bio1 b_bio_array[0] /* logical layer */ 183 #define b_bio2 b_bio_array[1] /* (typically) the disk layer */ 184 #define b_loffset b_bio1.bio_offset 185 186 187 /* 188 * Flags passed to getblk() 189 * 190 * GETBLK_PCATCH - Allow signals to be caught. getblk() is allowed to return 191 * NULL if this flag is passed. 192 * 193 * GETBLK_BHEAVY - This is a heavy weight buffer, meaning that resolving 194 * writes can require additional buffers. 195 */ 196 #define GETBLK_PCATCH 0x0001 /* catch signals */ 197 #define GETBLK_BHEAVY 0x0002 /* heavy weight buffer */ 198 199 /* 200 * These flags are kept in b_flags. 201 * 202 * Notes: 203 * 204 * B_ASYNC VOP calls on bp's are usually async whether or not 205 * B_ASYNC is set, but some subsystems, such as NFS, like 206 * to know what is best for the caller so they can 207 * optimize the I/O. 208 * 209 * B_PAGING Indicates that bp is being used by the paging system or 210 * some paging system and that the bp is not linked into 211 * the b_vp's clean/dirty linked lists or ref counts. 212 * Buffer vp reassignments are illegal in this case. 213 * 214 * B_CACHE This may only be set if the buffer is entirely valid. 215 * The situation where B_DELWRI is set and B_CACHE is 216 * clear MUST be committed to disk by getblk() so 217 * B_DELWRI can also be cleared. See the comments for 218 * getblk() in kern/vfs_bio.c. If B_CACHE is clear, 219 * the caller is expected to clear B_ERROR|B_INVAL, 220 * set BUF_CMD_READ, and initiate an I/O. 221 * 222 * The 'entire buffer' is defined to be the range from 223 * 0 through b_bcount. 224 * 225 * B_MALLOC Request that the buffer be allocated from the malloc 226 * pool, DEV_BSIZE aligned instead of PAGE_SIZE aligned. 227 * 228 * B_CLUSTEROK This flag is typically set for B_DELWRI buffers 229 * by filesystems that allow clustering when the buffer 230 * is fully dirty and indicates that it may be clustered 231 * with other adjacent dirty buffers. Note the clustering 232 * may not be used with the stage 1 data write under NFS 233 * but may be used for the commit rpc portion. 234 * 235 * B_VMIO Indicates that the buffer is tied into an VM object. 236 * The buffer's data is always PAGE_SIZE aligned even 237 * if b_bufsize and b_bcount are not. ( b_bufsize is 238 * always at least DEV_BSIZE aligned, though ). 239 * 240 * B_DIRECT Hint that we should attempt to completely free 241 * the pages underlying the buffer. B_DIRECT is 242 * sticky until the buffer is released and typically 243 * only has an effect when B_RELBUF is also set. 244 * 245 * B_NOWDRAIN This flag should be set when a device (like VN) 246 * does a turn-around VOP_WRITE from its strategy 247 * routine. This flag prevents bwrite() from blocking 248 * in wdrain, avoiding a deadlock situation. 249 * 250 * B_LOCKED The buffer will be released to the locked queue 251 * regardless of its current state. Note that 252 * if B_DELWRI is set, no I/O occurs until the caller 253 * acquires the buffer, clears B_LOCKED, then releases 254 * it again. 255 */ 256 257 #define B_AGE 0x00000001 /* Move to age queue when I/O done. */ 258 #define B_NEEDCOMMIT 0x00000002 /* Append-write in progress. */ 259 #define B_ASYNC 0x00000004 /* Start I/O, do not wait. */ 260 #define B_DIRECT 0x00000008 /* direct I/O flag (pls free vmio) */ 261 #define B_DEFERRED 0x00000010 /* Skipped over for cleaning */ 262 #define B_CACHE 0x00000020 /* Bread found us in the cache. */ 263 #define B_HASHED 0x00000040 /* Indexed via v_rbhash_tree */ 264 #define B_DELWRI 0x00000080 /* Delay I/O until buffer reused. */ 265 #define B_BNOCLIP 0x00000100 /* EOF clipping b_bcount not allowed */ 266 #define B_UNUSED0200 0x00000200 267 #define B_EINTR 0x00000400 /* I/O was interrupted */ 268 #define B_ERROR 0x00000800 /* I/O error occurred. */ 269 #define B_UNUSED1000 0x00001000 /* Unused */ 270 #define B_INVAL 0x00002000 /* Does not contain valid info. */ 271 #define B_LOCKED 0x00004000 /* Locked in core (not reusable). */ 272 #define B_NOCACHE 0x00008000 /* Destroy buffer AND backing store */ 273 #define B_MALLOC 0x00010000 /* malloced b_data */ 274 #define B_CLUSTEROK 0x00020000 /* Pagein op, so swap() can count it. */ 275 #define B_UNUSED40000 0x00040000 276 #define B_RAW 0x00080000 /* Set by physio for raw transfers. */ 277 #define B_HEAVY 0x00100000 /* Heavy-weight buffer */ 278 #define B_DIRTY 0x00200000 /* Needs writing later. */ 279 #define B_RELBUF 0x00400000 /* Release VMIO buffer. */ 280 #define B_WANT 0x00800000 /* Used by vm_pager.c */ 281 #define B_VNCLEAN 0x01000000 /* On vnode clean list */ 282 #define B_VNDIRTY 0x02000000 /* On vnode dirty list */ 283 #define B_PAGING 0x04000000 /* volatile paging I/O -- bypass VMIO */ 284 #define B_ORDERED 0x08000000 /* Must guarantee I/O ordering */ 285 #define B_RAM 0x10000000 /* Read ahead mark (flag) */ 286 #define B_VMIO 0x20000000 /* VMIO flag */ 287 #define B_CLUSTER 0x40000000 /* pagein op, so swap() can count it */ 288 #define B_NOWDRAIN 0x80000000 /* Avoid wdrain deadlock */ 289 290 #define PRINT_BUF_FLAGS "\20" \ 291 "\40nowdrain\37cluster\36vmio\35ram\34ordered" \ 292 "\33paging\32vndirty\31vnclean\30want\27relbuf\26dirty" \ 293 "\25unused20\24raw\23unused18\22clusterok\21malloc\20nocache" \ 294 "\17locked\16inval\15unused12\14error\13eintr\12unused9\11unused8" \ 295 "\10delwri\7hashed\6cache\5deferred\4direct\3async\2needcommit\1age" 296 297 #define NOOFFSET (-1LL) /* No buffer offset calculated yet */ 298 299 #ifdef _KERNEL 300 /* 301 * Buffer locking. See sys/buf2.h for inline functions. 302 */ 303 extern char *buf_wmesg; /* Default buffer lock message */ 304 #define BUF_WMESG "bufwait" 305 306 #endif /* _KERNEL */ 307 308 struct bio_queue_head { 309 TAILQ_HEAD(bio_queue, bio) queue; 310 off_t last_offset; 311 struct bio *insert_point; 312 struct bio *switch_point; 313 }; 314 315 /* 316 * This structure describes a clustered I/O. 317 */ 318 struct cluster_save { 319 int bs_nchildren; /* Number of associated buffers. */ 320 struct buf **bs_children; /* List of associated buffers. */ 321 }; 322 323 /* 324 * Zero out the buffer's data area. 325 */ 326 #define clrbuf(bp) { \ 327 bzero((bp)->b_data, (u_int)(bp)->b_bcount); \ 328 (bp)->b_resid = 0; \ 329 } 330 331 /* 332 * Flags to low-level bitmap allocation routines (balloc). 333 * 334 * Note: sequential_heuristic() in kern/vfs_vnops.c limits the count 335 * to 127. 336 */ 337 #define B_SEQMASK 0x7F000000 /* Sequential heuristic mask. */ 338 #define B_SEQSHIFT 24 /* Sequential heuristic shift. */ 339 #define B_SEQMAX 0x7F 340 #define B_CLRBUF 0x01 /* Cleared invalid areas of buffer. */ 341 #define B_SYNC 0x02 /* Do all allocations synchronously. */ 342 343 #ifdef _KERNEL 344 extern int nbuf; /* The number of buffer headers */ 345 extern int maxswzone; /* Max KVA for swap structures */ 346 extern int maxbcache; /* Max KVA for buffer cache */ 347 extern int runningbufspace; 348 extern int buf_maxio; /* nominal maximum I/O for buffer */ 349 extern struct buf *buf; /* The buffer headers. */ 350 extern char *buffers; /* The buffer contents. */ 351 extern int bufpages; /* Number of memory pages in the buffer pool. */ 352 extern struct buf *swbuf; /* Swap I/O buffer headers. */ 353 extern int nswbuf; /* Number of swap I/O buffer headers. */ 354 355 struct uio; 356 357 void bufinit (void); 358 void bwillwrite (void); 359 int buf_dirty_count_severe (void); 360 void initbufbio(struct buf *); 361 void reinitbufbio(struct buf *); 362 void clearbiocache(struct bio *); 363 void bremfree (struct buf *); 364 int bread (struct vnode *, off_t, int, struct buf **); 365 int breadn (struct vnode *, off_t, int, off_t *, int *, int, 366 struct buf **); 367 int bwrite (struct buf *); 368 void bdwrite (struct buf *); 369 void bawrite (struct buf *); 370 void bdirty (struct buf *); 371 void bheavy (struct buf *); 372 void bundirty (struct buf *); 373 int bowrite (struct buf *); 374 void brelse (struct buf *); 375 void bqrelse (struct buf *); 376 int vfs_bio_awrite (struct buf *); 377 struct buf *getpbuf (int *); 378 int inmem (struct vnode *, off_t); 379 struct buf *findblk (struct vnode *, off_t); 380 struct buf *getblk (struct vnode *, off_t, int, int, int); 381 struct buf *geteblk (int); 382 void regetblk(struct buf *bp); 383 struct bio *push_bio(struct bio *); 384 void pop_bio(struct bio *); 385 int biowait (struct buf *); 386 void biodone (struct bio *); 387 388 void cluster_append(struct bio *, struct buf *); 389 int cluster_read (struct vnode *, off_t, off_t, int, 390 int, int, struct buf **); 391 int cluster_wbuild (struct vnode *, int, off_t, int); 392 void cluster_write (struct buf *, off_t, int); 393 int physread (struct dev_read_args *); 394 int physwrite (struct dev_write_args *); 395 void vfs_bio_set_validclean (struct buf *, int base, int size); 396 void vfs_bio_clrbuf (struct buf *); 397 void vfs_busy_pages (struct vnode *, struct buf *); 398 void vfs_unbusy_pages (struct buf *); 399 int vmapbuf (struct buf *, caddr_t, int); 400 void vunmapbuf (struct buf *); 401 void relpbuf (struct buf *, int *); 402 void brelvp (struct buf *); 403 void bgetvp (struct vnode *, struct buf *); 404 int allocbuf (struct buf *bp, int size); 405 int scan_all_buffers (int (*)(struct buf *, void *), void *); 406 void reassignbuf (struct buf *); 407 struct buf *trypbuf (int *); 408 void bio_ops_sync(struct mount *mp); 409 410 #endif /* _KERNEL */ 411 #endif /* _KERNEL || _KERNEL_STRUCTURES */ 412 #endif /* !_SYS_BUF_H_ */ 413