1*0Sstevel@tonic-gate /*-
2*0Sstevel@tonic-gate * See the file LICENSE for redistribution information.
3*0Sstevel@tonic-gate *
4*0Sstevel@tonic-gate * Copyright (c) 1996, 1997, 1998
5*0Sstevel@tonic-gate * Sleepycat Software. All rights reserved.
6*0Sstevel@tonic-gate */
7*0Sstevel@tonic-gate #include "config.h"
8*0Sstevel@tonic-gate
9*0Sstevel@tonic-gate #ifndef lint
10*0Sstevel@tonic-gate static const char sccsid[] = "@(#)mp_bh.c 10.45 (Sleepycat) 11/25/98";
11*0Sstevel@tonic-gate #endif /* not lint */
12*0Sstevel@tonic-gate
13*0Sstevel@tonic-gate #ifndef NO_SYSTEM_INCLUDES
14*0Sstevel@tonic-gate #include <sys/types.h>
15*0Sstevel@tonic-gate
16*0Sstevel@tonic-gate #include <errno.h>
17*0Sstevel@tonic-gate #include <string.h>
18*0Sstevel@tonic-gate #include <unistd.h>
19*0Sstevel@tonic-gate #endif
20*0Sstevel@tonic-gate
21*0Sstevel@tonic-gate #include "db_int.h"
22*0Sstevel@tonic-gate #include "shqueue.h"
23*0Sstevel@tonic-gate #include "db_shash.h"
24*0Sstevel@tonic-gate #include "mp.h"
25*0Sstevel@tonic-gate #include "common_ext.h"
26*0Sstevel@tonic-gate
27*0Sstevel@tonic-gate static int __memp_upgrade __P((DB_MPOOL *, DB_MPOOLFILE *, MPOOLFILE *));
28*0Sstevel@tonic-gate
29*0Sstevel@tonic-gate /*
30*0Sstevel@tonic-gate * __memp_bhwrite --
31*0Sstevel@tonic-gate * Write the page associated with a given bucket header.
32*0Sstevel@tonic-gate *
33*0Sstevel@tonic-gate * PUBLIC: int __memp_bhwrite
34*0Sstevel@tonic-gate * PUBLIC: __P((DB_MPOOL *, MPOOLFILE *, BH *, int *, int *));
35*0Sstevel@tonic-gate */
36*0Sstevel@tonic-gate int
__memp_bhwrite(dbmp,mfp,bhp,restartp,wrotep)37*0Sstevel@tonic-gate __memp_bhwrite(dbmp, mfp, bhp, restartp, wrotep)
38*0Sstevel@tonic-gate DB_MPOOL *dbmp;
39*0Sstevel@tonic-gate MPOOLFILE *mfp;
40*0Sstevel@tonic-gate BH *bhp;
41*0Sstevel@tonic-gate int *restartp, *wrotep;
42*0Sstevel@tonic-gate {
43*0Sstevel@tonic-gate DB_MPOOLFILE *dbmfp;
44*0Sstevel@tonic-gate DB_MPREG *mpreg;
45*0Sstevel@tonic-gate int incremented, ret;
46*0Sstevel@tonic-gate
47*0Sstevel@tonic-gate if (restartp != NULL)
48*0Sstevel@tonic-gate *restartp = 0;
49*0Sstevel@tonic-gate if (wrotep != NULL)
50*0Sstevel@tonic-gate *wrotep = 0;
51*0Sstevel@tonic-gate incremented = 0;
52*0Sstevel@tonic-gate
53*0Sstevel@tonic-gate /*
54*0Sstevel@tonic-gate * Walk the process' DB_MPOOLFILE list and find a file descriptor for
55*0Sstevel@tonic-gate * the file. We also check that the descriptor is open for writing.
56*0Sstevel@tonic-gate * If we find a descriptor on the file that's not open for writing, we
57*0Sstevel@tonic-gate * try and upgrade it to make it writeable. If that fails, we're done.
58*0Sstevel@tonic-gate */
59*0Sstevel@tonic-gate LOCKHANDLE(dbmp, dbmp->mutexp);
60*0Sstevel@tonic-gate for (dbmfp = TAILQ_FIRST(&dbmp->dbmfq);
61*0Sstevel@tonic-gate dbmfp != NULL; dbmfp = TAILQ_NEXT(dbmfp, q))
62*0Sstevel@tonic-gate if (dbmfp->mfp == mfp) {
63*0Sstevel@tonic-gate if (F_ISSET(dbmfp, MP_READONLY) &&
64*0Sstevel@tonic-gate __memp_upgrade(dbmp, dbmfp, mfp)) {
65*0Sstevel@tonic-gate UNLOCKHANDLE(dbmp, dbmp->mutexp);
66*0Sstevel@tonic-gate return (0);
67*0Sstevel@tonic-gate }
68*0Sstevel@tonic-gate
69*0Sstevel@tonic-gate /*
70*0Sstevel@tonic-gate * Increment the reference count -- see the comment in
71*0Sstevel@tonic-gate * memp_fclose().
72*0Sstevel@tonic-gate */
73*0Sstevel@tonic-gate ++dbmfp->ref;
74*0Sstevel@tonic-gate incremented = 1;
75*0Sstevel@tonic-gate break;
76*0Sstevel@tonic-gate }
77*0Sstevel@tonic-gate UNLOCKHANDLE(dbmp, dbmp->mutexp);
78*0Sstevel@tonic-gate if (dbmfp != NULL)
79*0Sstevel@tonic-gate goto found;
80*0Sstevel@tonic-gate
81*0Sstevel@tonic-gate /*
82*0Sstevel@tonic-gate * It's not a page from a file we've opened. If the file requires
83*0Sstevel@tonic-gate * input/output processing, see if this process has ever registered
84*0Sstevel@tonic-gate * information as to how to write this type of file. If not, there's
85*0Sstevel@tonic-gate * nothing we can do.
86*0Sstevel@tonic-gate */
87*0Sstevel@tonic-gate if (mfp->ftype != 0) {
88*0Sstevel@tonic-gate LOCKHANDLE(dbmp, dbmp->mutexp);
89*0Sstevel@tonic-gate for (mpreg = LIST_FIRST(&dbmp->dbregq);
90*0Sstevel@tonic-gate mpreg != NULL; mpreg = LIST_NEXT(mpreg, q))
91*0Sstevel@tonic-gate if (mpreg->ftype == mfp->ftype)
92*0Sstevel@tonic-gate break;
93*0Sstevel@tonic-gate UNLOCKHANDLE(dbmp, dbmp->mutexp);
94*0Sstevel@tonic-gate if (mpreg == NULL)
95*0Sstevel@tonic-gate return (0);
96*0Sstevel@tonic-gate }
97*0Sstevel@tonic-gate
98*0Sstevel@tonic-gate /*
99*0Sstevel@tonic-gate * Try and open the file, attaching to the underlying shared area.
100*0Sstevel@tonic-gate *
101*0Sstevel@tonic-gate * XXX
102*0Sstevel@tonic-gate * Don't try to attach to temporary files. There are two problems in
103*0Sstevel@tonic-gate * trying to do that. First, if we have different privileges than the
104*0Sstevel@tonic-gate * process that "owns" the temporary file, we might create the backing
105*0Sstevel@tonic-gate * disk file such that the owning process couldn't read/write its own
106*0Sstevel@tonic-gate * buffers, e.g., memp_trickle() running as root creating a file owned
107*0Sstevel@tonic-gate * as root, mode 600. Second, if the temporary file has already been
108*0Sstevel@tonic-gate * created, we don't have any way of finding out what its real name is,
109*0Sstevel@tonic-gate * and, even if we did, it was already unlinked (so that it won't be
110*0Sstevel@tonic-gate * left if the process dies horribly). This decision causes a problem,
111*0Sstevel@tonic-gate * however: if the temporary file consumes the entire buffer cache,
112*0Sstevel@tonic-gate * and the owner doesn't flush the buffers to disk, we could end up
113*0Sstevel@tonic-gate * with resource starvation, and the memp_trickle() thread couldn't do
114*0Sstevel@tonic-gate * anything about it. That's a pretty unlikely scenario, though.
115*0Sstevel@tonic-gate *
116*0Sstevel@tonic-gate * XXX
117*0Sstevel@tonic-gate * There's no negative cache, so we may repeatedly try and open files
118*0Sstevel@tonic-gate * that we have previously tried (and failed) to open.
119*0Sstevel@tonic-gate *
120*0Sstevel@tonic-gate * Ignore any error, assume it's a permissions problem.
121*0Sstevel@tonic-gate */
122*0Sstevel@tonic-gate if (F_ISSET(mfp, MP_TEMP))
123*0Sstevel@tonic-gate return (0);
124*0Sstevel@tonic-gate
125*0Sstevel@tonic-gate if (__memp_fopen(dbmp, mfp, R_ADDR(dbmp, mfp->path_off),
126*0Sstevel@tonic-gate 0, 0, mfp->stat.st_pagesize, 0, NULL, &dbmfp) != 0)
127*0Sstevel@tonic-gate return (0);
128*0Sstevel@tonic-gate
129*0Sstevel@tonic-gate found: ret = __memp_pgwrite(dbmfp, bhp, restartp, wrotep);
130*0Sstevel@tonic-gate
131*0Sstevel@tonic-gate if (incremented) {
132*0Sstevel@tonic-gate LOCKHANDLE(dbmp, dbmp->mutexp);
133*0Sstevel@tonic-gate --dbmfp->ref;
134*0Sstevel@tonic-gate UNLOCKHANDLE(dbmp, dbmp->mutexp);
135*0Sstevel@tonic-gate }
136*0Sstevel@tonic-gate
137*0Sstevel@tonic-gate return (ret);
138*0Sstevel@tonic-gate }
139*0Sstevel@tonic-gate
140*0Sstevel@tonic-gate /*
141*0Sstevel@tonic-gate * __memp_pgread --
142*0Sstevel@tonic-gate * Read a page from a file.
143*0Sstevel@tonic-gate *
144*0Sstevel@tonic-gate * PUBLIC: int __memp_pgread __P((DB_MPOOLFILE *, BH *, int));
145*0Sstevel@tonic-gate */
146*0Sstevel@tonic-gate int
__memp_pgread(dbmfp,bhp,can_create)147*0Sstevel@tonic-gate __memp_pgread(dbmfp, bhp, can_create)
148*0Sstevel@tonic-gate DB_MPOOLFILE *dbmfp;
149*0Sstevel@tonic-gate BH *bhp;
150*0Sstevel@tonic-gate int can_create;
151*0Sstevel@tonic-gate {
152*0Sstevel@tonic-gate DB_IO db_io;
153*0Sstevel@tonic-gate DB_MPOOL *dbmp;
154*0Sstevel@tonic-gate MPOOLFILE *mfp;
155*0Sstevel@tonic-gate size_t len, pagesize;
156*0Sstevel@tonic-gate ssize_t nr;
157*0Sstevel@tonic-gate int created, ret;
158*0Sstevel@tonic-gate
159*0Sstevel@tonic-gate dbmp = dbmfp->dbmp;
160*0Sstevel@tonic-gate mfp = dbmfp->mfp;
161*0Sstevel@tonic-gate pagesize = mfp->stat.st_pagesize;
162*0Sstevel@tonic-gate
163*0Sstevel@tonic-gate F_SET(bhp, BH_LOCKED | BH_TRASH);
164*0Sstevel@tonic-gate LOCKBUFFER(dbmp, bhp);
165*0Sstevel@tonic-gate UNLOCKREGION(dbmp);
166*0Sstevel@tonic-gate
167*0Sstevel@tonic-gate /*
168*0Sstevel@tonic-gate * Temporary files may not yet have been created. We don't create
169*0Sstevel@tonic-gate * them now, we create them when the pages have to be flushed.
170*0Sstevel@tonic-gate */
171*0Sstevel@tonic-gate nr = 0;
172*0Sstevel@tonic-gate if (dbmfp->fd == -1)
173*0Sstevel@tonic-gate ret = 0;
174*0Sstevel@tonic-gate else {
175*0Sstevel@tonic-gate /*
176*0Sstevel@tonic-gate * Ignore read errors if we have permission to create the page.
177*0Sstevel@tonic-gate * Assume that the page doesn't exist, and that we'll create it
178*0Sstevel@tonic-gate * when we write it out.
179*0Sstevel@tonic-gate */
180*0Sstevel@tonic-gate db_io.fd_io = dbmfp->fd;
181*0Sstevel@tonic-gate db_io.fd_lock = dbmp->reginfo.fd;
182*0Sstevel@tonic-gate db_io.mutexp =
183*0Sstevel@tonic-gate F_ISSET(dbmp, MP_LOCKHANDLE) ? dbmfp->mutexp : NULL;
184*0Sstevel@tonic-gate db_io.pagesize = db_io.bytes = pagesize;
185*0Sstevel@tonic-gate db_io.pgno = bhp->pgno;
186*0Sstevel@tonic-gate db_io.buf = bhp->buf;
187*0Sstevel@tonic-gate
188*0Sstevel@tonic-gate ret = __os_io(&db_io, DB_IO_READ, &nr);
189*0Sstevel@tonic-gate }
190*0Sstevel@tonic-gate
191*0Sstevel@tonic-gate created = 0;
192*0Sstevel@tonic-gate if (nr < (ssize_t)pagesize)
193*0Sstevel@tonic-gate if (can_create)
194*0Sstevel@tonic-gate created = 1;
195*0Sstevel@tonic-gate else {
196*0Sstevel@tonic-gate /* If we had a short read, ret may be 0. */
197*0Sstevel@tonic-gate if (ret == 0)
198*0Sstevel@tonic-gate ret = EIO;
199*0Sstevel@tonic-gate __db_err(dbmp->dbenv,
200*0Sstevel@tonic-gate "%s: page %lu doesn't exist, create flag not set",
201*0Sstevel@tonic-gate __memp_fn(dbmfp), (u_long)bhp->pgno);
202*0Sstevel@tonic-gate goto err;
203*0Sstevel@tonic-gate }
204*0Sstevel@tonic-gate
205*0Sstevel@tonic-gate /*
206*0Sstevel@tonic-gate * Clear any bytes we didn't read that need to be cleared. If we're
207*0Sstevel@tonic-gate * running in diagnostic mode, smash any bytes on the page that are
208*0Sstevel@tonic-gate * unknown quantities for the caller.
209*0Sstevel@tonic-gate */
210*0Sstevel@tonic-gate if (nr != (ssize_t)pagesize) {
211*0Sstevel@tonic-gate len = mfp->clear_len == 0 ? pagesize : mfp->clear_len;
212*0Sstevel@tonic-gate if (nr < (ssize_t)len)
213*0Sstevel@tonic-gate memset(bhp->buf + nr, 0, len - nr);
214*0Sstevel@tonic-gate #ifdef DIAGNOSTIC
215*0Sstevel@tonic-gate if (nr > (ssize_t)len)
216*0Sstevel@tonic-gate len = nr;
217*0Sstevel@tonic-gate if (len < pagesize)
218*0Sstevel@tonic-gate memset(bhp->buf + len, 0xdb, pagesize - len);
219*0Sstevel@tonic-gate #endif
220*0Sstevel@tonic-gate }
221*0Sstevel@tonic-gate
222*0Sstevel@tonic-gate /* Call any pgin function. */
223*0Sstevel@tonic-gate ret = mfp->ftype == 0 ? 0 : __memp_pg(dbmfp, bhp, 1);
224*0Sstevel@tonic-gate
225*0Sstevel@tonic-gate /* Unlock the buffer and reacquire the region lock. */
226*0Sstevel@tonic-gate err: UNLOCKBUFFER(dbmp, bhp);
227*0Sstevel@tonic-gate LOCKREGION(dbmp);
228*0Sstevel@tonic-gate
229*0Sstevel@tonic-gate /*
230*0Sstevel@tonic-gate * If no errors occurred, the data is now valid, clear the BH_TRASH
231*0Sstevel@tonic-gate * flag; regardless, clear the lock bit and let other threads proceed.
232*0Sstevel@tonic-gate */
233*0Sstevel@tonic-gate F_CLR(bhp, BH_LOCKED);
234*0Sstevel@tonic-gate if (ret == 0) {
235*0Sstevel@tonic-gate F_CLR(bhp, BH_TRASH);
236*0Sstevel@tonic-gate
237*0Sstevel@tonic-gate /* Update the statistics. */
238*0Sstevel@tonic-gate if (created) {
239*0Sstevel@tonic-gate ++dbmp->mp->stat.st_page_create;
240*0Sstevel@tonic-gate ++mfp->stat.st_page_create;
241*0Sstevel@tonic-gate } else {
242*0Sstevel@tonic-gate ++dbmp->mp->stat.st_page_in;
243*0Sstevel@tonic-gate ++mfp->stat.st_page_in;
244*0Sstevel@tonic-gate }
245*0Sstevel@tonic-gate }
246*0Sstevel@tonic-gate
247*0Sstevel@tonic-gate return (ret);
248*0Sstevel@tonic-gate }
249*0Sstevel@tonic-gate
250*0Sstevel@tonic-gate /*
251*0Sstevel@tonic-gate * __memp_pgwrite --
252*0Sstevel@tonic-gate * Write a page to a file.
253*0Sstevel@tonic-gate *
254*0Sstevel@tonic-gate * PUBLIC: int __memp_pgwrite __P((DB_MPOOLFILE *, BH *, int *, int *));
255*0Sstevel@tonic-gate */
256*0Sstevel@tonic-gate int
__memp_pgwrite(dbmfp,bhp,restartp,wrotep)257*0Sstevel@tonic-gate __memp_pgwrite(dbmfp, bhp, restartp, wrotep)
258*0Sstevel@tonic-gate DB_MPOOLFILE *dbmfp;
259*0Sstevel@tonic-gate BH *bhp;
260*0Sstevel@tonic-gate int *restartp, *wrotep;
261*0Sstevel@tonic-gate {
262*0Sstevel@tonic-gate DB_ENV *dbenv;
263*0Sstevel@tonic-gate DB_IO db_io;
264*0Sstevel@tonic-gate DB_LOG *lg_info;
265*0Sstevel@tonic-gate DB_LSN lsn;
266*0Sstevel@tonic-gate DB_MPOOL *dbmp;
267*0Sstevel@tonic-gate MPOOL *mp;
268*0Sstevel@tonic-gate MPOOLFILE *mfp;
269*0Sstevel@tonic-gate ssize_t nw;
270*0Sstevel@tonic-gate int callpgin, dosync, ret, syncfail;
271*0Sstevel@tonic-gate const char *fail;
272*0Sstevel@tonic-gate
273*0Sstevel@tonic-gate dbmp = dbmfp->dbmp;
274*0Sstevel@tonic-gate dbenv = dbmp->dbenv;
275*0Sstevel@tonic-gate mp = dbmp->mp;
276*0Sstevel@tonic-gate mfp = dbmfp->mfp;
277*0Sstevel@tonic-gate
278*0Sstevel@tonic-gate if (restartp != NULL)
279*0Sstevel@tonic-gate *restartp = 0;
280*0Sstevel@tonic-gate if (wrotep != NULL)
281*0Sstevel@tonic-gate *wrotep = 0;
282*0Sstevel@tonic-gate callpgin = 0;
283*0Sstevel@tonic-gate
284*0Sstevel@tonic-gate /*
285*0Sstevel@tonic-gate * Check the dirty bit -- this buffer may have been written since we
286*0Sstevel@tonic-gate * decided to write it.
287*0Sstevel@tonic-gate */
288*0Sstevel@tonic-gate if (!F_ISSET(bhp, BH_DIRTY)) {
289*0Sstevel@tonic-gate if (wrotep != NULL)
290*0Sstevel@tonic-gate *wrotep = 1;
291*0Sstevel@tonic-gate return (0);
292*0Sstevel@tonic-gate }
293*0Sstevel@tonic-gate
294*0Sstevel@tonic-gate LOCKBUFFER(dbmp, bhp);
295*0Sstevel@tonic-gate
296*0Sstevel@tonic-gate /*
297*0Sstevel@tonic-gate * If there were two writers, we may have just been waiting while the
298*0Sstevel@tonic-gate * other writer completed I/O on this buffer. Check the dirty bit one
299*0Sstevel@tonic-gate * more time.
300*0Sstevel@tonic-gate */
301*0Sstevel@tonic-gate if (!F_ISSET(bhp, BH_DIRTY)) {
302*0Sstevel@tonic-gate UNLOCKBUFFER(dbmp, bhp);
303*0Sstevel@tonic-gate
304*0Sstevel@tonic-gate if (wrotep != NULL)
305*0Sstevel@tonic-gate *wrotep = 1;
306*0Sstevel@tonic-gate return (0);
307*0Sstevel@tonic-gate }
308*0Sstevel@tonic-gate
309*0Sstevel@tonic-gate F_SET(bhp, BH_LOCKED);
310*0Sstevel@tonic-gate UNLOCKREGION(dbmp);
311*0Sstevel@tonic-gate
312*0Sstevel@tonic-gate if (restartp != NULL)
313*0Sstevel@tonic-gate *restartp = 1;
314*0Sstevel@tonic-gate
315*0Sstevel@tonic-gate /* Copy the LSN off the page if we're going to need it. */
316*0Sstevel@tonic-gate lg_info = dbenv->lg_info;
317*0Sstevel@tonic-gate if (lg_info != NULL || F_ISSET(bhp, BH_WRITE))
318*0Sstevel@tonic-gate memcpy(&lsn, bhp->buf + mfp->lsn_off, sizeof(DB_LSN));
319*0Sstevel@tonic-gate
320*0Sstevel@tonic-gate /* Ensure the appropriate log records are on disk. */
321*0Sstevel@tonic-gate if (lg_info != NULL && (ret = log_flush(lg_info, &lsn)) != 0)
322*0Sstevel@tonic-gate goto err;
323*0Sstevel@tonic-gate
324*0Sstevel@tonic-gate /*
325*0Sstevel@tonic-gate * Call any pgout function. We set the callpgin flag so that we flag
326*0Sstevel@tonic-gate * that the contents of the buffer will need to be passed through pgin
327*0Sstevel@tonic-gate * before they are reused.
328*0Sstevel@tonic-gate */
329*0Sstevel@tonic-gate if (mfp->ftype == 0)
330*0Sstevel@tonic-gate ret = 0;
331*0Sstevel@tonic-gate else {
332*0Sstevel@tonic-gate callpgin = 1;
333*0Sstevel@tonic-gate if ((ret = __memp_pg(dbmfp, bhp, 0)) != 0)
334*0Sstevel@tonic-gate goto err;
335*0Sstevel@tonic-gate }
336*0Sstevel@tonic-gate
337*0Sstevel@tonic-gate /* Temporary files may not yet have been created. */
338*0Sstevel@tonic-gate if (dbmfp->fd == -1) {
339*0Sstevel@tonic-gate LOCKHANDLE(dbmp, dbmfp->mutexp);
340*0Sstevel@tonic-gate if (dbmfp->fd == -1 && ((ret = __db_appname(dbenv,
341*0Sstevel@tonic-gate DB_APP_TMP, NULL, NULL, DB_CREATE | DB_EXCL | DB_TEMPORARY,
342*0Sstevel@tonic-gate &dbmfp->fd, NULL)) != 0 || dbmfp->fd == -1)) {
343*0Sstevel@tonic-gate UNLOCKHANDLE(dbmp, dbmfp->mutexp);
344*0Sstevel@tonic-gate __db_err(dbenv,
345*0Sstevel@tonic-gate "unable to create temporary backing file");
346*0Sstevel@tonic-gate goto err;
347*0Sstevel@tonic-gate }
348*0Sstevel@tonic-gate UNLOCKHANDLE(dbmp, dbmfp->mutexp);
349*0Sstevel@tonic-gate }
350*0Sstevel@tonic-gate
351*0Sstevel@tonic-gate /* Write the page. */
352*0Sstevel@tonic-gate db_io.fd_io = dbmfp->fd;
353*0Sstevel@tonic-gate db_io.fd_lock = dbmp->reginfo.fd;
354*0Sstevel@tonic-gate db_io.mutexp = F_ISSET(dbmp, MP_LOCKHANDLE) ? dbmfp->mutexp : NULL;
355*0Sstevel@tonic-gate db_io.pagesize = db_io.bytes = mfp->stat.st_pagesize;
356*0Sstevel@tonic-gate db_io.pgno = bhp->pgno;
357*0Sstevel@tonic-gate db_io.buf = bhp->buf;
358*0Sstevel@tonic-gate if ((ret = __os_io(&db_io, DB_IO_WRITE, &nw)) != 0) {
359*0Sstevel@tonic-gate __db_panic(dbenv, ret);
360*0Sstevel@tonic-gate fail = "write";
361*0Sstevel@tonic-gate goto syserr;
362*0Sstevel@tonic-gate }
363*0Sstevel@tonic-gate if (nw != (ssize_t)mfp->stat.st_pagesize) {
364*0Sstevel@tonic-gate ret = EIO;
365*0Sstevel@tonic-gate fail = "write";
366*0Sstevel@tonic-gate goto syserr;
367*0Sstevel@tonic-gate }
368*0Sstevel@tonic-gate
369*0Sstevel@tonic-gate if (wrotep != NULL)
370*0Sstevel@tonic-gate *wrotep = 1;
371*0Sstevel@tonic-gate
372*0Sstevel@tonic-gate /* Unlock the buffer and reacquire the region lock. */
373*0Sstevel@tonic-gate UNLOCKBUFFER(dbmp, bhp);
374*0Sstevel@tonic-gate LOCKREGION(dbmp);
375*0Sstevel@tonic-gate
376*0Sstevel@tonic-gate /*
377*0Sstevel@tonic-gate * Clean up the flags based on a successful write.
378*0Sstevel@tonic-gate *
379*0Sstevel@tonic-gate * If we rewrote the page, it will need processing by the pgin
380*0Sstevel@tonic-gate * routine before reuse.
381*0Sstevel@tonic-gate */
382*0Sstevel@tonic-gate if (callpgin)
383*0Sstevel@tonic-gate F_SET(bhp, BH_CALLPGIN);
384*0Sstevel@tonic-gate F_CLR(bhp, BH_DIRTY | BH_LOCKED);
385*0Sstevel@tonic-gate
386*0Sstevel@tonic-gate /*
387*0Sstevel@tonic-gate * If we write a buffer for which a checkpoint is waiting, update
388*0Sstevel@tonic-gate * the count of pending buffers (both in the mpool as a whole and
389*0Sstevel@tonic-gate * for this file). If the count for this file goes to zero, set a
390*0Sstevel@tonic-gate * flag so we flush the writes.
391*0Sstevel@tonic-gate */
392*0Sstevel@tonic-gate if (F_ISSET(bhp, BH_WRITE)) {
393*0Sstevel@tonic-gate F_CLR(bhp, BH_WRITE);
394*0Sstevel@tonic-gate
395*0Sstevel@tonic-gate --mp->lsn_cnt;
396*0Sstevel@tonic-gate dosync = --mfp->lsn_cnt == 0 ? 1 : 0;
397*0Sstevel@tonic-gate } else
398*0Sstevel@tonic-gate dosync = 0;
399*0Sstevel@tonic-gate
400*0Sstevel@tonic-gate /* Update the page clean/dirty statistics. */
401*0Sstevel@tonic-gate ++mp->stat.st_page_clean;
402*0Sstevel@tonic-gate --mp->stat.st_page_dirty;
403*0Sstevel@tonic-gate
404*0Sstevel@tonic-gate /* Update I/O statistics. */
405*0Sstevel@tonic-gate ++mp->stat.st_page_out;
406*0Sstevel@tonic-gate ++mfp->stat.st_page_out;
407*0Sstevel@tonic-gate
408*0Sstevel@tonic-gate /*
409*0Sstevel@tonic-gate * Do the sync after everything else has been updated, so any incoming
410*0Sstevel@tonic-gate * checkpoint doesn't see inconsistent information.
411*0Sstevel@tonic-gate *
412*0Sstevel@tonic-gate * XXX:
413*0Sstevel@tonic-gate * Don't lock the region around the sync, fsync(2) has no atomicity
414*0Sstevel@tonic-gate * issues.
415*0Sstevel@tonic-gate *
416*0Sstevel@tonic-gate * XXX:
417*0Sstevel@tonic-gate * We ignore errors from the sync -- it makes no sense to return an
418*0Sstevel@tonic-gate * error to the calling process, so set a flag causing the checkpoint
419*0Sstevel@tonic-gate * to be retried later. There is a possibility, of course, that a
420*0Sstevel@tonic-gate * subsequent checkpoint was started and that we're going to force it
421*0Sstevel@tonic-gate * to fail. That should be unlikely, and fixing it would be difficult.
422*0Sstevel@tonic-gate */
423*0Sstevel@tonic-gate if (dosync) {
424*0Sstevel@tonic-gate UNLOCKREGION(dbmp);
425*0Sstevel@tonic-gate syncfail = __os_fsync(dbmfp->fd) != 0;
426*0Sstevel@tonic-gate LOCKREGION(dbmp);
427*0Sstevel@tonic-gate if (syncfail)
428*0Sstevel@tonic-gate F_SET(mp, MP_LSN_RETRY);
429*0Sstevel@tonic-gate }
430*0Sstevel@tonic-gate
431*0Sstevel@tonic-gate return (0);
432*0Sstevel@tonic-gate
433*0Sstevel@tonic-gate syserr: __db_err(dbenv, "%s: %s failed for page %lu",
434*0Sstevel@tonic-gate __memp_fn(dbmfp), fail, (u_long)bhp->pgno);
435*0Sstevel@tonic-gate
436*0Sstevel@tonic-gate err: /* Unlock the buffer and reacquire the region lock. */
437*0Sstevel@tonic-gate UNLOCKBUFFER(dbmp, bhp);
438*0Sstevel@tonic-gate LOCKREGION(dbmp);
439*0Sstevel@tonic-gate
440*0Sstevel@tonic-gate /*
441*0Sstevel@tonic-gate * Clean up the flags based on a failure.
442*0Sstevel@tonic-gate *
443*0Sstevel@tonic-gate * The page remains dirty but we remove our lock. If we rewrote the
444*0Sstevel@tonic-gate * page, it will need processing by the pgin routine before reuse.
445*0Sstevel@tonic-gate */
446*0Sstevel@tonic-gate if (callpgin)
447*0Sstevel@tonic-gate F_SET(bhp, BH_CALLPGIN);
448*0Sstevel@tonic-gate F_CLR(bhp, BH_LOCKED);
449*0Sstevel@tonic-gate
450*0Sstevel@tonic-gate return (ret);
451*0Sstevel@tonic-gate }
452*0Sstevel@tonic-gate
453*0Sstevel@tonic-gate /*
454*0Sstevel@tonic-gate * __memp_pg --
455*0Sstevel@tonic-gate * Call the pgin/pgout routine.
456*0Sstevel@tonic-gate *
457*0Sstevel@tonic-gate * PUBLIC: int __memp_pg __P((DB_MPOOLFILE *, BH *, int));
458*0Sstevel@tonic-gate */
459*0Sstevel@tonic-gate int
__memp_pg(dbmfp,bhp,is_pgin)460*0Sstevel@tonic-gate __memp_pg(dbmfp, bhp, is_pgin)
461*0Sstevel@tonic-gate DB_MPOOLFILE *dbmfp;
462*0Sstevel@tonic-gate BH *bhp;
463*0Sstevel@tonic-gate int is_pgin;
464*0Sstevel@tonic-gate {
465*0Sstevel@tonic-gate DBT dbt, *dbtp;
466*0Sstevel@tonic-gate DB_MPOOL *dbmp;
467*0Sstevel@tonic-gate DB_MPREG *mpreg;
468*0Sstevel@tonic-gate MPOOLFILE *mfp;
469*0Sstevel@tonic-gate int ftype, ret;
470*0Sstevel@tonic-gate
471*0Sstevel@tonic-gate dbmp = dbmfp->dbmp;
472*0Sstevel@tonic-gate mfp = dbmfp->mfp;
473*0Sstevel@tonic-gate
474*0Sstevel@tonic-gate LOCKHANDLE(dbmp, dbmp->mutexp);
475*0Sstevel@tonic-gate
476*0Sstevel@tonic-gate ftype = mfp->ftype;
477*0Sstevel@tonic-gate for (mpreg = LIST_FIRST(&dbmp->dbregq);
478*0Sstevel@tonic-gate mpreg != NULL; mpreg = LIST_NEXT(mpreg, q)) {
479*0Sstevel@tonic-gate if (ftype != mpreg->ftype)
480*0Sstevel@tonic-gate continue;
481*0Sstevel@tonic-gate if (mfp->pgcookie_len == 0)
482*0Sstevel@tonic-gate dbtp = NULL;
483*0Sstevel@tonic-gate else {
484*0Sstevel@tonic-gate dbt.size = mfp->pgcookie_len;
485*0Sstevel@tonic-gate dbt.data = R_ADDR(dbmp, mfp->pgcookie_off);
486*0Sstevel@tonic-gate dbtp = &dbt;
487*0Sstevel@tonic-gate }
488*0Sstevel@tonic-gate UNLOCKHANDLE(dbmp, dbmp->mutexp);
489*0Sstevel@tonic-gate
490*0Sstevel@tonic-gate if (is_pgin) {
491*0Sstevel@tonic-gate if (mpreg->pgin != NULL && (ret =
492*0Sstevel@tonic-gate mpreg->pgin(bhp->pgno, bhp->buf, dbtp)) != 0)
493*0Sstevel@tonic-gate goto err;
494*0Sstevel@tonic-gate } else
495*0Sstevel@tonic-gate if (mpreg->pgout != NULL && (ret =
496*0Sstevel@tonic-gate mpreg->pgout(bhp->pgno, bhp->buf, dbtp)) != 0)
497*0Sstevel@tonic-gate goto err;
498*0Sstevel@tonic-gate break;
499*0Sstevel@tonic-gate }
500*0Sstevel@tonic-gate
501*0Sstevel@tonic-gate if (mpreg == NULL)
502*0Sstevel@tonic-gate UNLOCKHANDLE(dbmp, dbmp->mutexp);
503*0Sstevel@tonic-gate
504*0Sstevel@tonic-gate return (0);
505*0Sstevel@tonic-gate
506*0Sstevel@tonic-gate err: UNLOCKHANDLE(dbmp, dbmp->mutexp);
507*0Sstevel@tonic-gate __db_err(dbmp->dbenv, "%s: %s failed for page %lu",
508*0Sstevel@tonic-gate __memp_fn(dbmfp), is_pgin ? "pgin" : "pgout", (u_long)bhp->pgno);
509*0Sstevel@tonic-gate return (ret);
510*0Sstevel@tonic-gate }
511*0Sstevel@tonic-gate
512*0Sstevel@tonic-gate /*
513*0Sstevel@tonic-gate * __memp_bhfree --
514*0Sstevel@tonic-gate * Free a bucket header and its referenced data.
515*0Sstevel@tonic-gate *
516*0Sstevel@tonic-gate * PUBLIC: void __memp_bhfree __P((DB_MPOOL *, MPOOLFILE *, BH *, int));
517*0Sstevel@tonic-gate */
518*0Sstevel@tonic-gate void
__memp_bhfree(dbmp,mfp,bhp,free_mem)519*0Sstevel@tonic-gate __memp_bhfree(dbmp, mfp, bhp, free_mem)
520*0Sstevel@tonic-gate DB_MPOOL *dbmp;
521*0Sstevel@tonic-gate MPOOLFILE *mfp;
522*0Sstevel@tonic-gate BH *bhp;
523*0Sstevel@tonic-gate int free_mem;
524*0Sstevel@tonic-gate {
525*0Sstevel@tonic-gate size_t off;
526*0Sstevel@tonic-gate
527*0Sstevel@tonic-gate /* Delete the buffer header from the hash bucket queue. */
528*0Sstevel@tonic-gate off = BUCKET(dbmp->mp, R_OFFSET(dbmp, mfp), bhp->pgno);
529*0Sstevel@tonic-gate SH_TAILQ_REMOVE(&dbmp->htab[off], bhp, hq, __bh);
530*0Sstevel@tonic-gate
531*0Sstevel@tonic-gate /* Delete the buffer header from the LRU queue. */
532*0Sstevel@tonic-gate SH_TAILQ_REMOVE(&dbmp->mp->bhq, bhp, q, __bh);
533*0Sstevel@tonic-gate
534*0Sstevel@tonic-gate /*
535*0Sstevel@tonic-gate * If we're not reusing it immediately, free the buffer header
536*0Sstevel@tonic-gate * and data for real.
537*0Sstevel@tonic-gate */
538*0Sstevel@tonic-gate if (free_mem) {
539*0Sstevel@tonic-gate __db_shalloc_free(dbmp->addr, bhp);
540*0Sstevel@tonic-gate --dbmp->mp->stat.st_page_clean;
541*0Sstevel@tonic-gate }
542*0Sstevel@tonic-gate }
543*0Sstevel@tonic-gate
544*0Sstevel@tonic-gate /*
545*0Sstevel@tonic-gate * __memp_upgrade --
546*0Sstevel@tonic-gate * Upgrade a file descriptor from readonly to readwrite.
547*0Sstevel@tonic-gate */
548*0Sstevel@tonic-gate static int
__memp_upgrade(dbmp,dbmfp,mfp)549*0Sstevel@tonic-gate __memp_upgrade(dbmp, dbmfp, mfp)
550*0Sstevel@tonic-gate DB_MPOOL *dbmp;
551*0Sstevel@tonic-gate DB_MPOOLFILE *dbmfp;
552*0Sstevel@tonic-gate MPOOLFILE *mfp;
553*0Sstevel@tonic-gate {
554*0Sstevel@tonic-gate int fd, ret;
555*0Sstevel@tonic-gate char *rpath;
556*0Sstevel@tonic-gate
557*0Sstevel@tonic-gate /*
558*0Sstevel@tonic-gate * !!!
559*0Sstevel@tonic-gate * We expect the handle to already be locked.
560*0Sstevel@tonic-gate */
561*0Sstevel@tonic-gate
562*0Sstevel@tonic-gate /* Check to see if we've already upgraded. */
563*0Sstevel@tonic-gate if (F_ISSET(dbmfp, MP_UPGRADE))
564*0Sstevel@tonic-gate return (0);
565*0Sstevel@tonic-gate
566*0Sstevel@tonic-gate /* Check to see if we've already failed. */
567*0Sstevel@tonic-gate if (F_ISSET(dbmfp, MP_UPGRADE_FAIL))
568*0Sstevel@tonic-gate return (1);
569*0Sstevel@tonic-gate
570*0Sstevel@tonic-gate /*
571*0Sstevel@tonic-gate * Calculate the real name for this file and try to open it read/write.
572*0Sstevel@tonic-gate * We know we have a valid pathname for the file because it's the only
573*0Sstevel@tonic-gate * way we could have gotten a file descriptor of any kind.
574*0Sstevel@tonic-gate */
575*0Sstevel@tonic-gate if ((ret = __db_appname(dbmp->dbenv, DB_APP_DATA,
576*0Sstevel@tonic-gate NULL, R_ADDR(dbmp, mfp->path_off), 0, NULL, &rpath)) != 0)
577*0Sstevel@tonic-gate return (ret);
578*0Sstevel@tonic-gate if (__db_open(rpath, 0, 0, 0, &fd) != 0) {
579*0Sstevel@tonic-gate F_SET(dbmfp, MP_UPGRADE_FAIL);
580*0Sstevel@tonic-gate ret = 1;
581*0Sstevel@tonic-gate } else {
582*0Sstevel@tonic-gate /* Swap the descriptors and set the upgrade flag. */
583*0Sstevel@tonic-gate (void)__os_close(dbmfp->fd);
584*0Sstevel@tonic-gate dbmfp->fd = fd;
585*0Sstevel@tonic-gate F_SET(dbmfp, MP_UPGRADE);
586*0Sstevel@tonic-gate ret = 0;
587*0Sstevel@tonic-gate }
588*0Sstevel@tonic-gate __os_freestr(rpath);
589*0Sstevel@tonic-gate return (ret);
590*0Sstevel@tonic-gate }
591