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. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)buf.h 8.9 (Berkeley) 3/30/95 35 * $FreeBSD: src/sys/sys/buf.h,v 1.88.2.10 2003/01/25 19:02:23 dillon Exp $ 36 * $DragonFly: src/sys/sys/buf.h,v 1.54 2008/08/29 20:08:37 dillon Exp $ 37 */ 38 39 #ifndef _SYS_BUF_H_ 40 #define _SYS_BUF_H_ 41 42 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES) 43 44 #ifndef _SYS_QUEUE_H_ 45 #include <sys/queue.h> 46 #endif 47 #ifndef _SYS_LOCK_H_ 48 #include <sys/lock.h> 49 #endif 50 #ifndef _SYS_DEVICE_H_ 51 #include <sys/device.h> 52 #endif 53 54 #ifndef _SYS_XIO_H_ 55 #include <sys/xio.h> 56 #endif 57 #ifndef _SYS_TREE_H_ 58 #include <sys/tree.h> 59 #endif 60 #ifndef _SYS_BIO_H_ 61 #include <sys/bio.h> 62 #endif 63 #ifndef _SYS_SPINLOCK_H_ 64 #include <sys/spinlock.h> 65 #endif 66 67 struct buf; 68 struct bio; 69 struct mount; 70 struct vnode; 71 struct xio; 72 73 #define NBUF_BIO 6 74 75 struct buf_rb_tree; 76 struct buf_rb_hash; 77 RB_PROTOTYPE2(buf_rb_tree, buf, b_rbnode, rb_buf_compare, off_t); 78 RB_PROTOTYPE2(buf_rb_hash, buf, b_rbhash, rb_buf_compare, off_t); 79 80 /* 81 * To avoid including <ufs/ffs/softdep.h> 82 */ 83 LIST_HEAD(workhead, worklist); 84 85 #endif 86 87 typedef enum buf_cmd { 88 BUF_CMD_DONE = 0, 89 BUF_CMD_READ, 90 BUF_CMD_WRITE, 91 BUF_CMD_FREEBLKS, 92 BUF_CMD_FORMAT, 93 BUF_CMD_FLUSH 94 } buf_cmd_t; 95 96 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES) 97 98 /* 99 * The buffer header describes an I/O operation in the kernel. 100 * 101 * NOTES: 102 * b_bufsize represents the filesystem block size (for this particular 103 * block) and/or the allocation size or original request size. This 104 * field is NOT USED by lower device layers. VNode and device 105 * strategy routines WILL NEVER ACCESS THIS FIELD. 106 * 107 * b_bcount represents the I/O request size. Unless B_NOBCLIP is set, 108 * the device chain is allowed to clip b_bcount to accomodate the device 109 * EOF. Note that this is different from the byte oriented file EOF. 110 * If B_NOBCLIP is set, the device chain is required to generate an 111 * error if it would othrewise have to clip the request. Buffers 112 * obtained via getblk() automatically set B_NOBCLIP. It is important 113 * to note that EOF clipping via b_bcount is different from EOF clipping 114 * via returning a b_actual < b_bcount. B_NOBCLIP only effects block 115 * oriented EOF clipping (b_bcount modifications). 116 * 117 * b_actual represents the number of bytes of I/O that actually occured, 118 * whether an error occured or not. b_actual must be initialized to 0 119 * prior to initiating I/O as the device drivers will assume it to 120 * start at 0. 121 * 122 * b_dirtyoff, b_dirtyend. Buffers support piecemeal, unaligned 123 * ranges of dirty data that need to be written to backing store. 124 * The range is typically clipped at b_bcount (not b_bufsize). 125 * 126 * b_bio1 and b_bio2 represent the two primary I/O layers. Additional 127 * I/O layers are allocated out of the object cache and may also exist. 128 * 129 * b_bio1 is the logical layer and contains offset or block number 130 * data for the primary vnode, b_vp. I/O operations are almost 131 * universally initiated from the logical layer, so you will often 132 * see things like: vn_strategy(bp->b_vp, &bp->b_bio1). 133 * 134 * b_bio2 is the first physical layer (typically the slice-relative 135 * layer) and contains the translated offset or block number for 136 * the block device underlying a filesystem. Filesystems such as UFS 137 * will maintain cached translations and you may see them initiate 138 * a 'physical' I/O using vn_strategy(devvp, &bp->b_bio2). BUT, 139 * remember that the layering is relative to bp->b_vp, so the 140 * device-relative block numbers for buffer cache operations that occur 141 * directly on a block device will be in the first BIO layer. 142 * 143 * b_ops - initialized if a buffer has a bio_ops 144 * 145 * NOTE!!! Only the BIO subsystem accesses b_bio1 and b_bio2 directly. 146 * ALL STRATEGY LAYERS FOR BOTH VNODES AND DEVICES ONLY ACCESS THE BIO 147 * PASSED TO THEM, AND WILL PUSH ANOTHER BIO LAYER IF FORWARDING THE 148 * I/O DEEPER. In particular, a vn_strategy() or dev_dstrategy() 149 * call should not ever access buf->b_vp as this vnode may be totally 150 * unrelated to the vnode/device whos strategy routine was called. 151 */ 152 struct buf { 153 RB_ENTRY(buf) b_rbnode; /* RB node in vnode clean/dirty tree */ 154 RB_ENTRY(buf) b_rbhash; /* RB node in vnode hash tree */ 155 TAILQ_ENTRY(buf) b_freelist; /* Free list position if not active. */ 156 struct buf *b_cluster_next; /* Next buffer (cluster code) */ 157 struct vnode *b_vp; /* (vp, loffset) index */ 158 struct bio b_bio_array[NBUF_BIO]; /* BIO translation layers */ 159 u_int32_t b_flags; /* B_* flags. */ 160 unsigned int b_qindex; /* buffer queue index */ 161 unsigned int b_qcpu; /* buffer queue cpu */ 162 unsigned char b_act_count; /* similar to vm_page act_count */ 163 unsigned char b_unused01; 164 struct lock b_lock; /* Buffer lock */ 165 buf_cmd_t b_cmd; /* I/O command */ 166 int b_bufsize; /* Allocated buffer size. */ 167 int b_runningbufspace; /* when I/O is running, pipelining */ 168 int b_bcount; /* Valid bytes in buffer. */ 169 int b_resid; /* Remaining I/O */ 170 int b_error; /* Error return */ 171 caddr_t b_data; /* Memory, superblocks, indirect etc. */ 172 caddr_t b_kvabase; /* base kva for buffer */ 173 int b_kvasize; /* size of kva for buffer */ 174 int b_dirtyoff; /* Offset in buffer of dirty region. */ 175 int b_dirtyend; /* Offset of end of dirty region. */ 176 int b_refs; /* FINDBLK_REF/bqhold()/bqdrop() */ 177 struct xio b_xio; /* data buffer page list management */ 178 struct bio_ops *b_ops; /* bio_ops used w/ b_dep */ 179 struct workhead b_dep; /* List of filesystem dependencies. */ 180 }; 181 182 /* 183 * XXX temporary 184 */ 185 #define b_bio1 b_bio_array[0] /* logical layer */ 186 #define b_bio2 b_bio_array[1] /* (typically) the disk layer */ 187 #define b_loffset b_bio1.bio_offset 188 189 190 /* 191 * Flags passed to getblk() 192 * 193 * GETBLK_PCATCH - Allow signals to be caught. getblk() is allowed to return 194 * NULL if this flag is passed. 195 * 196 * GETBLK_BHEAVY - This is a heavy weight buffer, meaning that resolving 197 * writes can require additional buffers. 198 * 199 * GETBLK_SZMATCH- blksize must match pre-existing b_bcount. getblk() can 200 * return NULL. 201 * 202 * GETBLK_NOWAIT - Do not use a blocking lock. getblk() can return NULL. 203 */ 204 #define GETBLK_PCATCH 0x0001 /* catch signals */ 205 #define GETBLK_BHEAVY 0x0002 /* heavy weight buffer */ 206 #define GETBLK_SZMATCH 0x0004 /* pre-existing buffer must match */ 207 #define GETBLK_NOWAIT 0x0008 /* non-blocking */ 208 209 #define FINDBLK_TEST 0x0010 /* test only, do not lock */ 210 #define FINDBLK_NBLOCK 0x0020 /* use non-blocking lock, can return NULL */ 211 #define FINDBLK_REF 0x0040 /* ref the buf to prevent reuse */ 212 213 /* 214 * These flags are kept in b_flags. 215 * 216 * Notes: 217 * 218 * B_PAGING Indicates that bp is being used by the paging system or 219 * some paging system and that the bp is not linked into 220 * the b_vp's clean/dirty linked lists or ref counts. 221 * Buffer vp reassignments are illegal in this case. 222 * 223 * B_CACHE This may only be set if the buffer is entirely valid. 224 * The situation where B_DELWRI is set and B_CACHE is 225 * clear MUST be committed to disk by getblk() so 226 * B_DELWRI can also be cleared. See the comments for 227 * getblk() in kern/vfs_bio.c. If B_CACHE is clear, 228 * the caller is expected to clear B_ERROR|B_INVAL, 229 * set BUF_CMD_READ, and initiate an I/O. 230 * 231 * The 'entire buffer' is defined to be the range from 232 * 0 through b_bcount. 233 * 234 * B_MALLOC Request that the buffer be allocated from the malloc 235 * pool, DEV_BSIZE aligned instead of PAGE_SIZE aligned. 236 * 237 * B_CLUSTEROK This flag is typically set for B_DELWRI buffers 238 * by filesystems that allow clustering when the buffer 239 * is fully dirty and indicates that it may be clustered 240 * with other adjacent dirty buffers. Note the clustering 241 * may not be used with the stage 1 data write under NFS 242 * but may be used for the commit rpc portion. 243 * 244 * B_VMIO Indicates that the buffer is tied into an VM object. 245 * The buffer's data is always PAGE_SIZE aligned even 246 * if b_bufsize and b_bcount are not. ( b_bufsize is 247 * always at least DEV_BSIZE aligned, though ). 248 * 249 * B_DIRECT Hint that we should attempt to completely free 250 * the pages underlying the buffer. B_DIRECT is 251 * sticky until the buffer is released and typically 252 * only has an effect when B_RELBUF is also set. 253 * 254 * B_LOCKED The buffer will be released to the locked queue 255 * regardless of its current state. Note that 256 * if B_DELWRI is set, no I/O occurs until the caller 257 * acquires the buffer, clears B_LOCKED, then releases 258 * it again. 259 * 260 * B_AGE When allocating a new buffer any buffer encountered 261 * with B_AGE set will be reallocated more quickly then 262 * buffers encountered without it set. B_AGE is set 263 * as part of the loop so idle buffers should eventually 264 * wind up with B_AGE set. B_AGE explicitly does NOT 265 * cause the buffer to be instantly reallocated for 266 * other purposes. 267 * 268 * Standard buffer flushing routines leave B_AGE intact 269 * through the DIRTY queue and into the CLEAN queue. 270 * Setting B_AGE on a dirty buffer will not cause it 271 * to be flushed more quickly but will cause it to be 272 * reallocated more quickly after having been flushed. 273 * 274 * B_NOCACHE Request that the buffer and backing store be 275 * destroyed on completion. If B_DELWRI is set and the 276 * write fails, the buffer remains intact. 277 * 278 * B_NOTMETA May be set on block device buffers representing 279 * file data (i.e. which aren't really meta-data), 280 * which will cause the buffer cache to set PG_NOTMETA 281 * in the VM pages when releasing them and the 282 * swapcache to not try to cache them. 283 * 284 * B_MARKER Special marker buf, always skip. 285 */ 286 287 #define B_AGE 0x00000001 /* Reuse more quickly */ 288 #define B_NEEDCOMMIT 0x00000002 /* Append-write in progress. */ 289 #define B_NOTMETA 0x00000004 /* This really isn't metadata */ 290 #define B_DIRECT 0x00000008 /* direct I/O flag (pls free vmio) */ 291 #define B_DEFERRED 0x00000010 /* vfs-controlled deferment */ 292 #define B_CACHE 0x00000020 /* Bread found us in the cache. */ 293 #define B_HASHED 0x00000040 /* Indexed via v_rbhash_tree */ 294 #define B_DELWRI 0x00000080 /* Delay I/O until buffer reused. */ 295 #define B_BNOCLIP 0x00000100 /* EOF clipping b_bcount not allowed */ 296 #define B_HASBOGUS 0x00000200 /* Contains bogus pages */ 297 #define B_EINTR 0x00000400 /* I/O was interrupted */ 298 #define B_ERROR 0x00000800 /* I/O error occurred. */ 299 #define B_IODEBUG 0x00001000 /* (Debugging only bread) */ 300 #define B_INVAL 0x00002000 /* Does not contain valid info. */ 301 #define B_LOCKED 0x00004000 /* Locked in core (not reusable). */ 302 #define B_NOCACHE 0x00008000 /* Destroy buffer AND backing store */ 303 #define B_MALLOC 0x00010000 /* malloced b_data */ 304 #define B_CLUSTEROK 0x00020000 /* Pagein op, so swap() can count it. */ 305 #define B_MARKER 0x00040000 /* Special marker buf in queue */ 306 #define B_RAW 0x00080000 /* Set by physio for raw transfers. */ 307 #define B_HEAVY 0x00100000 /* Heavy-weight buffer */ 308 #define B_DIRTY 0x00200000 /* Needs writing later. */ 309 #define B_RELBUF 0x00400000 /* Release VMIO buffer. */ 310 #define B_FAILONDIS 0x00800000 /* Fail on disconnect */ 311 #define B_VNCLEAN 0x01000000 /* On vnode clean list */ 312 #define B_VNDIRTY 0x02000000 /* On vnode dirty list */ 313 #define B_PAGING 0x04000000 /* volatile paging I/O -- bypass VMIO */ 314 #define B_ORDERED 0x08000000 /* Must guarantee I/O ordering */ 315 #define B_RAM 0x10000000 /* Read ahead mark (flag) */ 316 #define B_VMIO 0x20000000 /* VMIO flag */ 317 #define B_CLUSTER 0x40000000 /* pagein op, so swap() can count it */ 318 #define B_VFSFLAG1 0x80000000 /* VFSs can set this flag */ 319 320 #define PRINT_BUF_FLAGS "\20" \ 321 "\40unused31\37cluster\36vmio\35ram\34ordered" \ 322 "\33paging\32vndirty\31vnclean\30unused23\27relbuf\26dirty" \ 323 "\25unused20\24raw\23unused18\22clusterok\21malloc\20nocache" \ 324 "\17locked\16inval\15unused12\14error\13eintr\12unused9\11bnoclip" \ 325 "\10delwri\7hashed\6cache\5deferred\4direct\3unused2\2needcommit\1age" 326 327 #define NOOFFSET (-1LL) /* No buffer offset calculated yet */ 328 329 #ifdef _KERNEL 330 /* 331 * Buffer locking. See sys/buf2.h for inline functions. 332 */ 333 extern char *buf_wmesg; /* Default buffer lock message */ 334 #define BUF_WMESG "bufwait" 335 336 #endif /* _KERNEL */ 337 338 struct bio_queue_head { 339 TAILQ_HEAD(bio_queue, bio) queue; 340 off_t off_unused; 341 int reorder; 342 struct bio *transition; 343 struct bio *bio_unused; 344 }; 345 346 /* 347 * This structure describes a clustered I/O. 348 */ 349 struct cluster_save { 350 int bs_nchildren; /* Number of associated buffers. */ 351 struct buf **bs_children; /* List of associated buffers. */ 352 }; 353 354 /* 355 * Zero out the buffer's data area. 356 */ 357 #define clrbuf(bp) { \ 358 bzero((bp)->b_data, (u_int)(bp)->b_bcount); \ 359 (bp)->b_resid = 0; \ 360 } 361 362 /* 363 * Flags to low-level bitmap allocation routines (balloc). 364 * 365 * Note: sequential_heuristic() in kern/vfs_vnops.c limits the count 366 * to 127. 367 */ 368 #define B_SEQMASK 0x7F000000 /* Sequential heuristic mask. */ 369 #define B_SEQSHIFT 24 /* Sequential heuristic shift. */ 370 #define B_SEQMAX 0x7F 371 #define B_CLRBUF 0x01 /* Cleared invalid areas of buffer. */ 372 #define B_SYNC 0x02 /* Do all allocations synchronously. */ 373 374 #ifdef _KERNEL 375 extern long nbuf; /* The number of buffer headers */ 376 extern long maxswzone; /* Max KVA for swap structures */ 377 extern long maxbcache; /* Max KVA for buffer cache */ 378 extern long hidirtybufspace; 379 extern int buf_maxio; /* nominal maximum I/O for buffer */ 380 extern struct buf *buf; /* The buffer headers. */ 381 extern char *buffers; /* The buffer contents. */ 382 extern int bufpages; /* Number of memory pages in the buffer pool. */ 383 extern struct buf *swbuf; /* Swap I/O buffer headers. */ 384 extern long nswbuf; /* Number of swap I/O buffer headers. */ 385 extern int bioq_reorder_burst_interval; 386 extern int bioq_reorder_burst_bytes; 387 extern int bioq_reorder_minor_interval; 388 extern int bioq_reorder_minor_bytes; 389 390 struct uio; 391 struct devstat; 392 393 long bd_heatup (void); 394 void bd_wait (long count); 395 void waitrunningbufspace(void); 396 int buf_dirty_count_severe (void); 397 int buf_runningbufspace_severe (void); 398 void initbufbio(struct buf *); 399 void uninitbufbio(struct buf *); 400 void reinitbufbio(struct buf *); 401 void clearbiocache(struct bio *); 402 void bremfree (struct buf *); 403 int breadx (struct vnode *, off_t, int, struct buf **); 404 int breadnx (struct vnode *, off_t, int, off_t *, int *, int, 405 struct buf **); 406 void breadcb(struct vnode *, off_t, int, 407 void (*)(struct bio *), void *); 408 int bwrite (struct buf *); 409 void bdwrite (struct buf *); 410 void buwrite (struct buf *); 411 void bawrite (struct buf *); 412 void bdirty (struct buf *); 413 void bheavy (struct buf *); 414 void bundirty (struct buf *); 415 int bowrite (struct buf *); 416 void brelse (struct buf *); 417 void bqrelse (struct buf *); 418 int cluster_awrite (struct buf *); 419 struct buf *getpbuf (int *); 420 struct buf *getpbuf_kva (int *); 421 int inmem (struct vnode *, off_t); 422 struct buf *findblk (struct vnode *, off_t, int); 423 struct buf *getblk (struct vnode *, off_t, int, int, int); 424 struct buf *getcacheblk (struct vnode *, off_t, int, int); 425 struct buf *geteblk (int); 426 struct buf *getnewbuf(int, int, int, int); 427 void bqhold(struct buf *bp); 428 void bqdrop(struct buf *bp); 429 void regetblk(struct buf *bp); 430 struct bio *push_bio(struct bio *); 431 struct bio *pop_bio(struct bio *); 432 int biowait (struct bio *, const char *); 433 int biowait_timeout (struct bio *, const char *, int); 434 void bpdone (struct buf *, int); 435 void biodone (struct bio *); 436 void biodone_sync (struct bio *); 437 void pbuf_adjcount(int *pfreecnt, int n); 438 439 void cluster_append(struct bio *, struct buf *); 440 int cluster_readx (struct vnode *, off_t, off_t, int, 441 size_t, size_t, struct buf **); 442 void cluster_readcb (struct vnode *, off_t, off_t, int, 443 size_t, size_t, void (*func)(struct bio *), void *arg); 444 void cluster_write (struct buf *, off_t, int, int); 445 int physread (struct dev_read_args *); 446 int physwrite (struct dev_write_args *); 447 void vfs_bio_clrbuf (struct buf *); 448 void vfs_busy_pages (struct vnode *, struct buf *); 449 void vfs_unbusy_pages (struct buf *); 450 int vmapbuf (struct buf *, caddr_t, int); 451 void vunmapbuf (struct buf *); 452 void relpbuf (struct buf *, int *); 453 void brelvp (struct buf *); 454 int bgetvp (struct vnode *, struct buf *, int); 455 void bsetrunningbufspace(struct buf *, int); 456 int allocbuf (struct buf *bp, int size); 457 int scan_all_buffers (int (*)(struct buf *, void *), void *); 458 void reassignbuf (struct buf *); 459 struct buf *trypbuf (int *); 460 struct buf *trypbuf_kva (int *); 461 void bio_ops_sync(struct mount *mp); 462 void vm_hold_free_pages(struct buf *bp, vm_offset_t from, vm_offset_t to); 463 void vm_hold_load_pages(struct buf *bp, vm_offset_t from, vm_offset_t to); 464 void nestiobuf_done(struct bio *mbio, int donebytes, int error, struct devstat *stats); 465 void nestiobuf_init(struct bio *mbio); 466 void nestiobuf_add(struct bio *mbio, struct buf *bp, int off, size_t size, struct devstat *stats); 467 void nestiobuf_start(struct bio *mbio); 468 void nestiobuf_error(struct bio *mbio, int error); 469 #endif /* _KERNEL */ 470 #endif /* _KERNEL || _KERNEL_STRUCTURES */ 471 #endif /* !_SYS_BUF_H_ */ 472