Properly assert the state of the MP lock in the async syscall message
[dragonfly.git] / sys / sys / buf.h
CommitLineData
984263bc
MD
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 $
16523a43 40 * $DragonFly: src/sys/sys/buf.h,v 1.21 2005/11/19 17:19:48 dillon Exp $
984263bc
MD
41 */
42
43#ifndef _SYS_BUF_H_
44#define _SYS_BUF_H_
45
335dda38 46#ifndef _SYS_QUEUE_H_
984263bc 47#include <sys/queue.h>
335dda38
MD
48#endif
49#ifndef _SYS_LOCK_H_
984263bc 50#include <sys/lock.h>
335dda38
MD
51#endif
52#ifndef _SYS_DEVICE_H_
53#include <sys/device.h>
54#endif
984263bc 55
54f51aeb
HP
56#ifndef _SYS_XIO_H_
57#include <sys/xio.h>
58#endif
6bae6177
MD
59#ifndef _SYS_TREE_H_
60#include <sys/tree.h>
61#endif
1eabbf32
HP
62#ifndef _SYS_BIO_H_
63#include <sys/bio.h>
64#endif
16523a43
MD
65#ifndef _SYS_SPINLOCK_H_
66#include <sys/spinlock.h>
67#endif
1eabbf32 68
984263bc
MD
69struct buf;
70struct mount;
71struct vnode;
54f51aeb 72struct xio;
984263bc 73
6bae6177
MD
74struct buf_rb_tree;
75RB_PROTOTYPE(buf_rb_tree, buf, b_rbnode, rb_buf_compare);
76
984263bc
MD
77/*
78 * To avoid including <ufs/ffs/softdep.h>
79 */
80LIST_HEAD(workhead, worklist);
81/*
82 * These are currently used only by the soft dependency code, hence
83 * are stored once in a global variable. If other subsystems wanted
84 * to use these hooks, a pointer to a set of bio_ops could be added
85 * to each buffer.
86 */
87extern struct bio_ops {
b153f746
RG
88 void (*io_start) (struct buf *);
89 void (*io_complete) (struct buf *);
90 void (*io_deallocate) (struct buf *);
1e5a0c40
HP
91 int (*io_fsync) (struct vnode *);
92 int (*io_sync) (struct mount *);
b153f746 93 void (*io_movedeps) (struct buf *, struct buf *);
1e5a0c40 94 int (*io_countdeps) (struct buf *, int);
984263bc
MD
95} bioops;
96
97struct iodone_chain {
98 long ic_prev_flags;
b153f746 99 void (*ic_prev_iodone) (struct buf *);
984263bc
MD
100 void *ic_prev_iodone_chain;
101 struct {
102 long ia_long;
103 void *ia_ptr;
1e5a0c40 104 } ic_args[5];
984263bc
MD
105};
106
107/*
108 * The buffer header describes an I/O operation in the kernel.
109 *
110 * NOTES:
111 * b_bufsize, b_bcount. b_bufsize is the allocation size of the
112 * buffer, either DEV_BSIZE or PAGE_SIZE aligned. b_bcount is the
113 * originally requested buffer size and can serve as a bounds check
114 * against EOF. For most, but not all uses, b_bcount == b_bufsize.
115 *
116 * b_dirtyoff, b_dirtyend. Buffers support piecemeal, unaligned
117 * ranges of dirty data that need to be written to backing store.
118 * The range is typically clipped at b_bcount ( not b_bufsize ).
119 *
120 * b_resid. Number of bytes remaining in I/O. After an I/O operation
121 * completes, b_resid is usually 0 indicating 100% success.
122 */
123struct buf {
124 LIST_ENTRY(buf) b_hash; /* Hash chain. */
6bae6177 125 RB_ENTRY(buf) b_rbnode; /* Red-Black node in vnode RB tree */
984263bc 126 TAILQ_ENTRY(buf) b_freelist; /* Free list position if not active. */
1eabbf32
HP
127 TAILQ_ENTRY(buf) b_act; /* driver queue when active. *new* */
128 struct bio b_bio; /* Underlying I/O */
984263bc
MD
129 long b_flags; /* B_* flags. */
130 unsigned short b_qindex; /* buffer queue index */
131 unsigned char b_xflags; /* extra flags */
132 struct lock b_lock; /* Buffer lock */
984263bc
MD
133 long b_bufsize; /* Allocated buffer size. */
134 long b_runningbufspace; /* when I/O is running, pipelining */
135 long b_bcount; /* Valid bytes in buffer. */
984263bc
MD
136 caddr_t b_data; /* Memory, superblocks, indirect etc. */
137 caddr_t b_kvabase; /* base kva for buffer */
138 int b_kvasize; /* size of kva for buffer */
106a9f09 139 daddr_t b_lblkno; /* Logical block number. */
984263bc 140 off_t b_offset; /* Offset into file */
984263bc
MD
141 /* For nested b_iodone's. */
142 struct iodone_chain *b_iodone_chain;
143 struct vnode *b_vp; /* Device vnode. */
144 int b_dirtyoff; /* Offset in buffer of dirty region. */
145 int b_dirtyend; /* Offset of end of dirty region. */
984263bc 146 void *b_saveaddr; /* Original b_addr for physio. */
984263bc
MD
147 union pager_info {
148 void *pg_spc;
149 int pg_reqpage;
150 } b_pager;
151 union cluster_info {
152 TAILQ_HEAD(cluster_list_head, buf) cluster_head;
153 TAILQ_ENTRY(buf) cluster_entry;
154 } b_cluster;
837ddd46 155 struct xio b_xio; /* page list management for buffer head. */
984263bc
MD
156 struct workhead b_dep; /* List of filesystem dependencies. */
157 struct chain_info { /* buffer chaining */
158 struct buf *parent;
159 int count;
160 } b_chain;
161};
162
1eabbf32
HP
163#define b_dev b_bio.bio_dev
164#define b_resid b_bio.bio_resid
165#define b_error b_bio.bio_error
1eabbf32
HP
166#define b_blkno b_bio.bio_blkno
167#define b_pblkno b_bio.bio_pblkno
168#define b_driver1 b_bio.bio_driver_ctx
169#define b_caller1 b_bio.bio_caller_ctx
170#define b_iodone b_bio.bio_done
171#define b_spc b_pager.pg_spc
984263bc
MD
172
173/*
174 * These flags are kept in b_flags.
175 *
176 * Notes:
177 *
178 * B_ASYNC VOP calls on bp's are usually async whether or not
179 * B_ASYNC is set, but some subsystems, such as NFS, like
180 * to know what is best for the caller so they can
181 * optimize the I/O.
182 *
183 * B_PAGING Indicates that bp is being used by the paging system or
184 * some paging system and that the bp is not linked into
185 * the b_vp's clean/dirty linked lists or ref counts.
186 * Buffer vp reassignments are illegal in this case.
187 *
188 * B_CACHE This may only be set if the buffer is entirely valid.
189 * The situation where B_DELWRI is set and B_CACHE is
190 * clear MUST be committed to disk by getblk() so
191 * B_DELWRI can also be cleared. See the comments for
192 * getblk() in kern/vfs_bio.c. If B_CACHE is clear,
193 * the caller is expected to clear B_ERROR|B_INVAL,
194 * set B_READ, and initiate an I/O.
195 *
196 * The 'entire buffer' is defined to be the range from
197 * 0 through b_bcount.
198 *
199 * B_MALLOC Request that the buffer be allocated from the malloc
200 * pool, DEV_BSIZE aligned instead of PAGE_SIZE aligned.
201 *
202 * B_CLUSTEROK This flag is typically set for B_DELWRI buffers
203 * by filesystems that allow clustering when the buffer
204 * is fully dirty and indicates that it may be clustered
205 * with other adjacent dirty buffers. Note the clustering
206 * may not be used with the stage 1 data write under NFS
207 * but may be used for the commit rpc portion.
208 *
209 * B_VMIO Indicates that the buffer is tied into an VM object.
210 * The buffer's data is always PAGE_SIZE aligned even
211 * if b_bufsize and b_bcount are not. ( b_bufsize is
212 * always at least DEV_BSIZE aligned, though ).
213 *
214 * B_DIRECT Hint that we should attempt to completely free
215 * the pages underlying the buffer. B_DIRECT is
216 * sticky until the buffer is released and typically
217 * only has an effect when B_RELBUF is also set.
218 *
219 * B_NOWDRAIN This flag should be set when a device (like VN)
220 * does a turn-around VOP_WRITE from its strategy
221 * routine. This flag prevents bwrite() from blocking
222 * in wdrain, avoiding a deadlock situation.
223 */
224
225#define B_AGE 0x00000001 /* Move to age queue when I/O done. */
226#define B_NEEDCOMMIT 0x00000002 /* Append-write in progress. */
227#define B_ASYNC 0x00000004 /* Start I/O, do not wait. */
228#define B_DIRECT 0x00000008 /* direct I/O flag (pls free vmio) */
229#define B_DEFERRED 0x00000010 /* Skipped over for cleaning */
230#define B_CACHE 0x00000020 /* Bread found us in the cache. */
1e5a0c40 231#define B_UNUSED40 0x00000040 /* Unused */
984263bc
MD
232#define B_DELWRI 0x00000080 /* Delay I/O until buffer reused. */
233#define B_FREEBUF 0x00000100 /* Instruct driver: free blocks */
234#define B_DONE 0x00000200 /* I/O completed. */
235#define B_EINTR 0x00000400 /* I/O was interrupted */
236#define B_ERROR 0x00000800 /* I/O error occurred. */
71bc1d95 237#define B_UNUSED1000 0x00001000 /* Unused */
984263bc
MD
238#define B_INVAL 0x00002000 /* Does not contain valid info. */
239#define B_LOCKED 0x00004000 /* Locked in core (not reusable). */
240#define B_NOCACHE 0x00008000 /* Do not cache block after use. */
241#define B_MALLOC 0x00010000 /* malloced b_data */
242#define B_CLUSTEROK 0x00020000 /* Pagein op, so swap() can count it. */
243#define B_PHYS 0x00040000 /* I/O to user memory. */
244#define B_RAW 0x00080000 /* Set by physio for raw transfers. */
245#define B_READ 0x00100000 /* Read buffer. */
246#define B_DIRTY 0x00200000 /* Needs writing later. */
247#define B_RELBUF 0x00400000 /* Release VMIO buffer. */
248#define B_WANT 0x00800000 /* Used by vm_pager.c */
249#define B_WRITE 0x00000000 /* Write buffer (pseudo flag). */
71bc1d95 250#define B_UNUSED1000000 0x01000000 /* Unused */
984263bc
MD
251#define B_XXX 0x02000000 /* Debugging flag. */
252#define B_PAGING 0x04000000 /* volatile paging I/O -- bypass VMIO */
253#define B_ORDERED 0x08000000 /* Must guarantee I/O ordering */
254#define B_RAM 0x10000000 /* Read ahead mark (flag) */
255#define B_VMIO 0x20000000 /* VMIO flag */
256#define B_CLUSTER 0x40000000 /* pagein op, so swap() can count it */
257#define B_NOWDRAIN 0x80000000 /* Avoid wdrain deadlock */
258
259#define PRINT_BUF_FLAGS "\20\40nowdrain\37cluster\36vmio\35ram\34ordered" \
260 "\33paging\32xxx\31writeinprog\30want\27relbuf\26dirty" \
261 "\25read\24raw\23phys\22clusterok\21malloc\20nocache" \
262 "\17locked\16inval\15scanned\14error\13eintr\12done\11freebuf" \
263 "\10delwri\7call\6cache\4direct\3async\2needcommit\1age"
264
265/*
266 * These flags are kept in b_xflags.
267 */
268#define BX_VNDIRTY 0x00000001 /* On vnode dirty list */
269#define BX_VNCLEAN 0x00000002 /* On vnode clean list */
270#define BX_BKGRDWRITE 0x00000004 /* Do writes in background */
271#define BX_BKGRDINPROG 0x00000008 /* Background write in progress */
272#define BX_BKGRDWAIT 0x00000010 /* Background write waiting */
273#define BX_AUTOCHAINDONE 0x00000020 /* pager I/O chain auto mode */
274
275#define NOOFFSET (-1LL) /* No buffer offset calculated yet */
276
277#ifdef _KERNEL
278/*
3020e3be 279 * Buffer locking. See sys/buf2.h for inline functions.
984263bc 280 */
16523a43 281extern struct spinlock buftimespinlock; /* Interlock on setting prio and timo */
984263bc
MD
282extern char *buf_wmesg; /* Default buffer lock message */
283#define BUF_WMESG "bufwait"
984263bc
MD
284
285#endif /* _KERNEL */
286
287struct buf_queue_head {
288 TAILQ_HEAD(buf_queue, buf) queue;
289 daddr_t last_pblkno;
290 struct buf *insert_point;
291 struct buf *switch_point;
292};
293
294/*
295 * This structure describes a clustered I/O. It is stored in the b_saveaddr
296 * field of the buffer on which I/O is done. At I/O completion, cluster
297 * callback uses the structure to parcel I/O's to individual buffers, and
298 * then free's this structure.
299 */
300struct cluster_save {
301 long bs_bcount; /* Saved b_bcount. */
302 long bs_bufsize; /* Saved b_bufsize. */
303 void *bs_saveaddr; /* Saved b_addr. */
304 int bs_nchildren; /* Number of associated buffers. */
305 struct buf **bs_children; /* List of associated buffers. */
306};
307
984263bc
MD
308/*
309 * Zero out the buffer's data area.
310 */
311#define clrbuf(bp) { \
312 bzero((bp)->b_data, (u_int)(bp)->b_bcount); \
313 (bp)->b_resid = 0; \
314}
315
316/*
317 * Flags to low-level bitmap allocation routines (balloc).
318 *
319 * Note: sequential_heuristic() in kern/vfs_vnops.c limits the count
320 * to 127.
321 */
322#define B_SEQMASK 0x7F000000 /* Sequential heuristic mask. */
323#define B_SEQSHIFT 24 /* Sequential heuristic shift. */
324#define B_SEQMAX 0x7F
325#define B_CLRBUF 0x01 /* Cleared invalid areas of buffer. */
326#define B_SYNC 0x02 /* Do all allocations synchronously. */
327
328#ifdef _KERNEL
329extern int nbuf; /* The number of buffer headers */
330extern int maxswzone; /* Max KVA for swap structures */
331extern int maxbcache; /* Max KVA for buffer cache */
332extern int runningbufspace;
333extern int buf_maxio; /* nominal maximum I/O for buffer */
334extern struct buf *buf; /* The buffer headers. */
335extern char *buffers; /* The buffer contents. */
336extern int bufpages; /* Number of memory pages in the buffer pool. */
337extern struct buf *swbuf; /* Swap I/O buffer headers. */
338extern int nswbuf; /* Number of swap I/O buffer headers. */
984263bc
MD
339
340struct uio;
341
b153f746
RG
342caddr_t bufhashinit (caddr_t);
343void bufinit (void);
344void bwillwrite (void);
345int buf_dirty_count_severe (void);
346void bremfree (struct buf *);
347int bread (struct vnode *, daddr_t, int, struct buf **);
348int breadn (struct vnode *, daddr_t, int, daddr_t *, int *, int,
349 struct buf **);
350int bwrite (struct buf *);
351void bdwrite (struct buf *);
352void bawrite (struct buf *);
353void bdirty (struct buf *);
354void bundirty (struct buf *);
355int bowrite (struct buf *);
356void brelse (struct buf *);
357void bqrelse (struct buf *);
358int vfs_bio_awrite (struct buf *);
359struct buf * getpbuf (int *);
360struct buf *incore (struct vnode *, daddr_t);
361struct buf *gbincore (struct vnode *, daddr_t);
362int inmem (struct vnode *, daddr_t);
363struct buf *getblk (struct vnode *, daddr_t, int, int, int);
364struct buf *geteblk (int);
365int biowait (struct buf *);
366void biodone (struct buf *);
984263bc 367
b153f746
RG
368void cluster_callback (struct buf *);
369int cluster_read (struct vnode *, u_quad_t, daddr_t, long,
370 long, int, struct buf **);
371int cluster_wbuild (struct vnode *, long, daddr_t, int);
372void cluster_write (struct buf *, u_quad_t, int);
373int physio (dev_t dev, struct uio *uio, int ioflag);
984263bc
MD
374#define physread physio
375#define physwrite physio
b153f746
RG
376void vfs_bio_set_validclean (struct buf *, int base, int size);
377void vfs_bio_clrbuf (struct buf *);
378void vfs_busy_pages (struct buf *, int clear_modify);
379void vfs_unbusy_pages (struct buf *);
380void vwakeup (struct buf *);
381int vmapbuf (struct buf *);
382void vunmapbuf (struct buf *);
383void relpbuf (struct buf *, int *);
384void brelvp (struct buf *);
385void bgetvp (struct vnode *, struct buf *);
386void pbgetvp (struct vnode *, struct buf *);
387void pbrelvp (struct buf *);
388int allocbuf (struct buf *bp, int size);
389void reassignbuf (struct buf *, struct vnode *);
390void pbreassignbuf (struct buf *, struct vnode *);
391struct buf *trypbuf (int *);
984263bc
MD
392
393#endif /* _KERNEL */
394
395#endif /* !_SYS_BUF_H_ */