Print this page
OS-881 To workaround OS-580 add support to only invalidate mappings from a single process
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/sys/buf.h
+++ new/usr/src/uts/common/sys/buf.h
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
|
↓ open down ↓ |
13 lines elided |
↑ open up ↑ |
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21 /*
22 22 * Copyright 2008 Sun Microsystems, Inc. All rights reserved.
23 23 * Use is subject to license terms.
24 + * Copyright 2012 Joyent, Inc. All rights reserved.
24 25 */
25 26
26 27 /* Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */
27 28 /* All Rights Reserved */
28 29
29 30 /*
30 31 * University Copyright- Copyright (c) 1982, 1986, 1988
31 32 * The Regents of the University of California
32 33 * All Rights Reserved
33 34 *
34 35 * University Acknowledgment- Portions of this document are derived from
35 36 * software developed by the University of California, Berkeley, and its
36 37 * contributors.
37 38 */
38 39
39 40 #ifndef _SYS_BUF_H
40 41 #define _SYS_BUF_H
41 42
42 43 #include <sys/types32.h>
43 44 #include <sys/t_lock.h>
44 45 #include <sys/kstat.h>
45 46
46 47 #ifdef __cplusplus
47 48 extern "C" {
48 49 #endif
49 50
50 51 /*
51 52 * Each buffer in the pool is usually doubly linked into 2 lists:
52 53 * the device with which it is currently associated (always)
53 54 * and also on a list of blocks available for allocation
54 55 * for other use (usually).
55 56 * The latter list is kept in last-used order, and the two
56 57 * lists are doubly linked to make it easy to remove
57 58 * a buffer from one list when it was found by
58 59 * looking through the other.
59 60 * A buffer is on the available list, and is liable
60 61 * to be reassigned to another disk block, if and only
61 62 * if it is not marked BUSY. When a buffer is busy, the
62 63 * available-list pointers can be used for other purposes.
63 64 * Most drivers use the forward ptr as a link in their I/O active queue.
64 65 * A buffer header contains all the information required to perform I/O.
65 66 * Most of the routines which manipulate these things are in bio.c.
66 67 *
67 68 * There are a number of locks associated with the buffer management
68 69 * system.
69 70 * hbuf.b_lock: protects hash chains, buffer hdr freelists
70 71 * and delayed write freelist
71 72 * bfree_lock; protects the bfreelist structure
72 73 * bhdr_lock: protects the free header list
73 74 * blist_lock: protects b_list fields
74 75 * buf.b_sem: protects all remaining members in the buf struct
75 76 * buf.b_io: I/O synchronization variable
76 77 *
77 78 * A buffer header is never "locked" (b_sem) when it is on
78 79 * a "freelist" (bhdrlist or bfreelist avail lists).
79 80 */
80 81 typedef struct buf {
81 82 int b_flags; /* see defines below */
82 83 struct buf *b_forw; /* headed by d_tab of conf.c */
83 84 struct buf *b_back; /* " */
84 85 struct buf *av_forw; /* position on free list, */
85 86 struct buf *av_back; /* if not BUSY */
86 87 o_dev_t b_dev; /* OLD major+minor device name */
87 88 size_t b_bcount; /* transfer count */
88 89 union {
89 90 caddr_t b_addr; /* low order core address */
90 91 struct fs *b_fs; /* superblocks */
91 92 struct cg *b_cg; /* UFS cylinder group block */
92 93 struct dinode *b_dino; /* UFS ilist */
93 94 daddr32_t *b_daddr; /* disk blocks */
94 95 } b_un;
95 96
96 97 lldaddr_t _b_blkno; /* block # on device (union) */
97 98 #define b_lblkno _b_blkno._f
98 99 #ifdef _LP64
99 100 #define b_blkno _b_blkno._f
100 101 #else
101 102 #define b_blkno _b_blkno._p._l
102 103 #endif /* _LP64 */
103 104
104 105 char b_obs1; /* obsolete */
105 106 size_t b_resid; /* words not transferred after error */
106 107 clock_t b_start; /* request start time */
107 108 struct proc *b_proc; /* process doing physical or swap I/O */
108 109 struct page *b_pages; /* page list for PAGEIO */
109 110 clock_t b_obs2; /* obsolete */
110 111 /* Begin new stuff */
111 112 #define b_actf av_forw
112 113 #define b_actl av_back
113 114 #define b_active b_bcount
114 115 #define b_errcnt b_resid
115 116 size_t b_bufsize; /* size of allocated buffer */
116 117 int (*b_iodone)(struct buf *); /* function called by iodone */
117 118 struct vnode *b_vp; /* vnode associated with block */
118 119 struct buf *b_chain; /* chain together all buffers here */
119 120 int b_obs3; /* obsolete */
120 121 int b_error; /* expanded error field */
121 122 void *b_private; /* "opaque" driver private area */
122 123 dev_t b_edev; /* expanded dev field */
123 124 ksema_t b_sem; /* Exclusive access to buf */
124 125 ksema_t b_io; /* I/O Synchronization */
125 126 struct buf *b_list; /* List of potential B_DELWRI bufs */
126 127 struct page **b_shadow; /* shadow page list */
127 128 void *b_dip; /* device info pointer */
128 129 struct vnode *b_file; /* file associated with this buffer */
129 130 offset_t b_offset; /* offset in file assoc. with buffer */
130 131 } buf_t;
131 132
132 133 /*
133 134 * Bufhd structures used at the head of the hashed buffer queues.
134 135 * We only need seven words for this, so this abbreviated
135 136 * definition saves some space.
136 137 */
137 138 struct diskhd {
138 139 int b_flags; /* not used, needed for consistency */
139 140 struct buf *b_forw, *b_back; /* queue of unit queues */
140 141 struct buf *av_forw, *av_back; /* queue of bufs for this unit */
141 142 o_dev_t b_dev; /* OLD major+minor device name */
142 143 size_t b_bcount; /* transfer count */
143 144 };
144 145
145 146
146 147 /*
147 148 * Statistics on the buffer cache
148 149 */
149 150 struct biostats {
150 151 kstat_named_t bio_lookup; /* requests to assign buffer */
151 152 kstat_named_t bio_hit; /* buffer already associated with blk */
152 153 kstat_named_t bio_bufwant; /* kmem_allocs NOSLEEP failed new buf */
153 154 kstat_named_t bio_bufwait; /* kmem_allocs with KM_SLEEP for buf */
154 155 kstat_named_t bio_bufbusy; /* buffer locked by someone else */
155 156 kstat_named_t bio_bufdup; /* duplicate buffer found for block */
156 157 };
157 158
158 159 /*
159 160 * These flags are kept in b_flags.
160 161 * The first group is part of the DDI
161 162 */
162 163 #define B_BUSY 0x0001 /* not on av_forw/back list */
163 164 #define B_DONE 0x0002 /* transaction finished */
164 165 #define B_ERROR 0x0004 /* transaction aborted */
165 166 #define B_PAGEIO 0x0010 /* do I/O to pages on bp->p_pages */
166 167 #define B_PHYS 0x0020 /* Physical IO potentially using UNIBUS map */
167 168 #define B_READ 0x0040 /* read when I/O occurs */
168 169 #define B_WRITE 0x0100 /* non-read pseudo-flag */
169 170
170 171 /* Not part of the DDI */
171 172 #define B_WANTED 0x0080 /* issue wakeup when BUSY goes off */
172 173 #define B_AGE 0x000200 /* delayed write for correct aging */
173 174 #define B_ASYNC 0x000400 /* don't wait for I/O completion */
174 175 #define B_DELWRI 0x000800 /* delayed write-wait til buf needed */
175 176 #define B_STALE 0x001000 /* on av_* list; invalid contents */
176 177 #define B_DONTNEED 0x002000 /* after write, need not be cached */
177 178 #define B_REMAPPED 0x004000 /* buffer is kernel addressable */
178 179 #define B_FREE 0x008000 /* free page when done */
|
↓ open down ↓ |
145 lines elided |
↑ open up ↑ |
179 180 #define B_INVAL 0x010000 /* destroy page when done */
180 181 #define B_FORCE 0x020000 /* semi-permanent removal from cache */
181 182 #define B_NOCACHE 0x080000 /* don't cache block when released */
182 183 #define B_TRUNC 0x100000 /* truncate page without I/O */
183 184 #define B_SHADOW 0x200000 /* is b_shadow field valid? */
184 185 #define B_RETRYWRI 0x400000 /* retry write til works or bfinval */
185 186 #define B_FAILFAST 0x1000000 /* Fail promptly if device goes away */
186 187 #define B_STARTED 0x2000000 /* io:::start probe called for buf */
187 188 #define B_ABRWRITE 0x4000000 /* Application based recovery active */
188 189 #define B_PAGE_NOWAIT 0x8000000 /* Skip the page if it is locked */
190 +#define B_INVALCURONLY 0x10000000 /* invalidate only for curproc */
189 191
190 192 /*
191 193 * There is some confusion over the meaning of B_FREE and B_INVAL and what
192 194 * the use of one over the other implies.
193 195 *
194 196 * In both cases, when we are done with the page (buffer) we want to free
195 197 * up the page. In the case of B_FREE, the page will go to the cachelist.
196 198 * In the case of B_INVAL, the page will be destroyed (hashed out of it's
197 199 * vnode) and placed on the freelist. Beyond this, there is no difference
198 200 * between the sole use of these two flags. In both cases, IO will be done
199 201 * if the page is not yet committed to storage.
200 202 *
203 + * The B_INVALCURONLY flag modifies the behavior of the B_INVAL flag and is
204 + * intended to be used in conjunction with B_INVAL. B_INVALCURONLY has no
205 + * meaning on its own. When both B_INVALCURONLY and B_INVAL are set, then
206 + * the mapping for the page is only invalidated for the current process.
207 + * In this case, the page is not destroyed unless this was the final mapping.
208 + *
201 209 * In order to discard pages without writing them back, (B_INVAL | B_TRUNC)
202 210 * should be used.
203 211 *
204 212 * Use (B_INVAL | B_FORCE) to force the page to be destroyed even if we
205 213 * could not successfuly write out the page.
206 214 */
207 215
208 216 /*
209 217 * Insq/Remq for the buffer hash lists.
210 218 */
211 219 #define bremhash(bp) { \
212 220 ASSERT((bp)->b_forw != NULL); \
213 221 ASSERT((bp)->b_back != NULL); \
214 222 (bp)->b_back->b_forw = (bp)->b_forw; \
215 223 (bp)->b_forw->b_back = (bp)->b_back; \
216 224 (bp)->b_forw = (bp)->b_back = NULL; \
217 225 }
218 226 #define binshash(bp, dp) { \
219 227 ASSERT((bp)->b_forw == NULL); \
220 228 ASSERT((bp)->b_back == NULL); \
221 229 ASSERT((dp)->b_forw != NULL); \
222 230 ASSERT((dp)->b_back != NULL); \
223 231 (bp)->b_forw = (dp)->b_forw; \
224 232 (bp)->b_back = (dp); \
225 233 (dp)->b_forw->b_back = (bp); \
226 234 (dp)->b_forw = (bp); \
227 235 }
228 236
229 237
230 238 /*
231 239 * The hash structure maintains two lists:
232 240 *
233 241 * 1) The hash list of buffers (b_forw & b_back)
234 242 * 2) The LRU free list of buffers on this hash bucket (av_forw & av_back)
235 243 *
236 244 * The dwbuf structure keeps a list of delayed write buffers per hash bucket
237 245 * hence there are exactly the same number of dwbuf structures as there are
238 246 * the hash buckets (hbuf structures) in the system.
239 247 *
240 248 * The number of buffers on the freelist may not be equal to the number of
241 249 * buffers on the hash list. That is because when buffers are busy they are
242 250 * taken off the freelist but not off the hash list. "b_length" field keeps
243 251 * track of the number of free buffers (including delayed writes ones) on
244 252 * the hash bucket. The "b_lock" mutex protects the free list as well as
245 253 * the hash list. It also protects the counter "b_length".
246 254 *
247 255 * Enties b_forw, b_back, av_forw & av_back must be at the same offset
248 256 * as the ones in buf structure.
249 257 */
250 258 struct hbuf {
251 259 int b_flags;
252 260
253 261 struct buf *b_forw; /* hash list forw pointer */
254 262 struct buf *b_back; /* hash list back pointer */
255 263
256 264 struct buf *av_forw; /* free list forw pointer */
257 265 struct buf *av_back; /* free list back pointer */
258 266
259 267 int b_length; /* # of entries on free list */
260 268 kmutex_t b_lock; /* lock to protect this structure */
261 269 };
262 270
263 271
264 272 /*
265 273 * The delayed list pointer entries should match with the buf strcuture.
266 274 */
267 275 struct dwbuf {
268 276 int b_flags; /* not used */
269 277
270 278 struct buf *b_forw; /* not used */
271 279 struct buf *b_back; /* not used */
272 280
273 281 struct buf *av_forw; /* delayed write forw pointer */
274 282 struct buf *av_back; /* delayed write back pointer */
275 283 };
276 284
277 285
278 286 /*
279 287 * Unlink a buffer from the available (free or delayed write) list and mark
280 288 * it busy (internal interface).
281 289 */
282 290 #define notavail(bp) \
283 291 {\
284 292 ASSERT(SEMA_HELD(&bp->b_sem)); \
285 293 ASSERT((bp)->av_forw != NULL); \
286 294 ASSERT((bp)->av_back != NULL); \
287 295 ASSERT((bp)->av_forw != (bp)); \
288 296 ASSERT((bp)->av_back != (bp)); \
289 297 (bp)->av_back->av_forw = (bp)->av_forw; \
290 298 (bp)->av_forw->av_back = (bp)->av_back; \
291 299 (bp)->b_flags |= B_BUSY; \
292 300 (bp)->av_forw = (bp)->av_back = NULL; \
293 301 }
294 302
295 303 #if defined(_KERNEL)
296 304 /*
297 305 * Macros to avoid the extra function call needed for binary compat.
298 306 *
299 307 * B_RETRYWRI is not included in clear_flags for BWRITE(), BWRITE2(),
300 308 * or brwrite() so that the retry operation is persistent until the
301 309 * write either succeeds or the buffer is bfinval()'d.
302 310 *
303 311 */
304 312 #define BREAD(dev, blkno, bsize) \
305 313 bread_common(/* ufsvfsp */ NULL, dev, blkno, bsize)
306 314
307 315 #define BWRITE(bp) \
308 316 bwrite_common(/* ufsvfsp */ NULL, bp, /* force_wait */ 0, \
309 317 /* do_relse */ 1, \
310 318 /* clear_flags */ (B_READ | B_DONE | B_ERROR | B_DELWRI))
311 319
312 320 #define BWRITE2(bp) \
313 321 bwrite_common(/* ufsvfsp */ NULL, bp, /* force_wait */ 1, \
314 322 /* do_relse */ 0, \
315 323 /* clear_flags */ (B_READ | B_DONE | B_ERROR | B_DELWRI))
316 324
317 325 #define GETBLK(dev, blkno, bsize) \
318 326 getblk_common(/* ufsvfsp */ NULL, dev, blkno, bsize, /* errflg */ 0)
319 327
320 328
321 329 /*
322 330 * Macros for new retry write interfaces.
323 331 */
324 332
325 333 /*
326 334 * Same as bdwrite() except write failures are retried.
327 335 */
328 336 #define bdrwrite(bp) { \
329 337 (bp)->b_flags |= B_RETRYWRI; \
330 338 bdwrite((bp)); \
331 339 }
332 340
333 341 /*
334 342 * Same as bwrite() except write failures are retried.
335 343 */
336 344 #define brwrite(bp) { \
337 345 (bp)->b_flags |= B_RETRYWRI; \
338 346 bwrite_common((bp), /* force_wait */ 0, /* do_relse */ 1, \
339 347 /* clear_flags */ (B_READ | B_DONE | B_ERROR | B_DELWRI)); \
340 348 }
341 349
342 350 extern struct hbuf *hbuf; /* Hash table */
343 351 extern struct dwbuf *dwbuf; /* delayed write hash table */
344 352 extern struct buf *buf; /* The buffer pool itself */
345 353 extern struct buf bfreelist; /* head of available list */
346 354
347 355 extern void (*bio_lufs_strategy)(void *, buf_t *); /* UFS Logging */
348 356 extern void (*bio_snapshot_strategy)(void *, buf_t *); /* UFS snapshots */
349 357
350 358 int bcheck(dev_t, struct buf *);
351 359 int iowait(struct buf *);
352 360 int hash2ints(int x, int y);
353 361 int bio_busy(int);
354 362 int biowait(struct buf *);
355 363 int biomodified(struct buf *);
356 364 int geterror(struct buf *);
357 365 void minphys(struct buf *);
358 366 /*
359 367 * ufsvfsp is declared as a void * to avoid having everyone that uses
360 368 * this header file include sys/fs/ufs_inode.h.
361 369 */
362 370 void bwrite_common(void *ufsvfsp, struct buf *, int force_wait,
363 371 int do_relse, int clear_flags);
364 372 void bwrite(struct buf *);
365 373 void bwrite2(struct buf *);
366 374 void bdwrite(struct buf *);
367 375 void bawrite(struct buf *);
368 376 void brelse(struct buf *);
369 377 void iodone(struct buf *);
370 378 void clrbuf(struct buf *);
371 379 void bflush(dev_t);
372 380 void blkflush(dev_t, daddr_t);
373 381 void binval(dev_t);
374 382 int bfinval(dev_t, int);
375 383 void binit(void);
376 384 void biodone(struct buf *);
377 385 void bioinit(struct buf *);
378 386 void biofini(struct buf *);
379 387 void bp_mapin(struct buf *);
380 388 void *bp_mapin_common(struct buf *, int);
381 389 void bp_mapout(struct buf *);
382 390 int bp_copyin(struct buf *, void *, offset_t, size_t);
383 391 int bp_copyout(void *, struct buf *, offset_t, size_t);
384 392 void bp_init(size_t, uint_t);
385 393 int bp_color(struct buf *);
386 394 void pageio_done(struct buf *);
387 395 struct buf *bread(dev_t, daddr_t, long);
388 396 struct buf *bread_common(void *, dev_t, daddr_t, long);
389 397 struct buf *breada(dev_t, daddr_t, daddr_t, long);
390 398 struct buf *getblk(dev_t, daddr_t, long);
391 399 struct buf *getblk_common(void *, dev_t, daddr_t, long, int);
392 400 struct buf *ngeteblk(long);
393 401 struct buf *geteblk(void);
394 402 struct buf *pageio_setup(struct page *, size_t, struct vnode *, int);
395 403 void bioerror(struct buf *bp, int error);
396 404 void bioreset(struct buf *bp);
397 405 struct buf *bioclone(struct buf *, off_t, size_t, dev_t, daddr_t,
398 406 int (*)(struct buf *), struct buf *, int);
399 407 size_t biosize(void);
400 408 #endif /* defined(_KERNEL) */
401 409
402 410 #ifdef __cplusplus
403 411 }
404 412 #endif
405 413
406 414 #endif /* _SYS_BUF_H */
|
↓ open down ↓ |
196 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX