1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 1997, 2010, Oracle and/or its affiliates. All rights reserved.
23 */
24
25 #include <sys/cred.h>
26 #include <sys/cmn_err.h>
27 #include <sys/debug.h>
28 #include <sys/systm.h>
29 #include <sys/kmem.h>
30 #include <sys/disp.h>
31 #include <sys/atomic.h>
32 #include <rpc/types.h>
33 #include <nfs/nfs.h>
34 #include <nfs/nfssys.h>
35 #include <nfs/export.h>
36 #include <nfs/rnode.h>
37 #include <rpc/auth.h>
38 #include <rpc/svc.h>
39 #include <rpc/xdr.h>
40 #include <rpc/clnt.h>
41 #include <nfs/nfs_log.h>
42
43 #define NUM_RECORDS_TO_WRITE 256
44 #define NUM_BYTES_TO_WRITE 65536
45
46 extern krwlock_t exported_lock;
47
48 static int nfslog_num_records_to_write = NUM_RECORDS_TO_WRITE;
49 static int nfslog_num_bytes_to_write = NUM_BYTES_TO_WRITE;
50
51 /*
52 * This struct is used to 'hide' the details of managing the log
53 * records internally to the logging code. Allocation routines
54 * are used to obtain pieces of memory for XDR encoding. This struct
55 * is a 'header' to those areas and a opaque cookie is used to pass
56 * this data structure between the allocating function and the put
57 * function.
58 */
59 struct lr_alloc {
60 struct lr_alloc *next; /* links for write queuing */
61 struct lr_alloc *prev;
62 #define LR_ALLOC_NOFREE 0x1 /* not present, call free */
63 int lr_flags;
64 caddr_t log_record; /* address to XDR encoding */
65 size_t size; /* final size of encoding */
66 struct kmem_cache *alloc_cache; /* keep track of cache ptr */
67 struct exportinfo *exi; /* who are we related to? */
68 struct log_buffer *lb;
69 };
70
71 struct flush_thread_params {
72 struct nfsl_flush_args tp_args;
73 int tp_error;
74 };
75
76 static int log_file_create(caddr_t, struct log_file **);
77 static void log_file_rele(struct log_file *);
78 static struct log_buffer *log_buffer_create(caddr_t);
79 static void log_buffer_rele(struct log_buffer *);
80 static int nfslog_record_append2all(struct lr_alloc *);
81 static int nfslog_logbuffer_rename(struct log_buffer *);
82 static void nfslog_logfile_wait(struct log_file *);
83 static int nfslog_logfile_rename(char *, char *);
84 static void nfslog_do_flush(struct flush_thread_params *);
85 static void create_buffer_header(caddr_t *, size_t *, size_t *);
86
87 static int nfslog_write_logrecords(struct log_file *, struct lr_alloc *, int);
88 static void nfslog_free_logrecords(struct lr_alloc *);
89 static int nfslog_records_flush_to_disk(struct log_buffer *);
90 static int nfslog_records_flush_to_disk_nolock(struct log_buffer *);
91
92 /*
93 * Read/Write lock that protects 'nfslog_buffer_list'.
94 * This lock must be held when searching or modifying 'nfslog_buffer_list'.
95 */
96 static krwlock_t nfslog_buffer_list_lock;
97
98 /*
99 * The list of "log_buffer" structures.
100 */
101 struct log_buffer *nfslog_buffer_list = NULL;
102
103
104 #define LOG_BUFFER_HOLD(lbp) { \
105 mutex_enter(&(lbp)->lb_lock); \
106 (lbp)->lb_refcnt++; \
107 mutex_exit(&(lbp)->lb_lock); \
108 }
109
110 #define LOG_FILE_HOLD(lfp) { \
111 mutex_enter(&(lfp)->lf_lock); \
112 (lfp)->lf_refcnt++; \
113 mutex_exit(&(lfp)->lf_lock); \
114 }
115
116 #define LOG_FILE_RELE(lfp) { \
117 log_file_rele(lfp); \
118 }
119
120 /*
121 * These two macros are used to prep a logfile data structure and
122 * associated file for writing data. Note that the lf_lock is
123 * held as a result of the call to the first macro. This is used
124 * for serialization correctness between the logbuffer struct and
125 * the logfile struct.
126 */
127 #define LOG_FILE_LOCK_TO_WRITE(lfp) { \
128 mutex_enter(&(lfp)->lf_lock); \
129 (lfp)->lf_refcnt++; \
130 (lfp)->lf_writers++; \
131 }
132
133 #define LOG_FILE_UNLOCK_FROM_WRITE(lfp) { \
134 (lfp)->lf_writers--; \
135 if ((lfp)->lf_writers == 0 && ((lfp)->lf_flags & L_WAITING)) { \
136 (lfp)->lf_flags &= ~L_WAITING; \
137 cv_broadcast(&(lfp)->lf_cv_waiters); \
138 } \
139 mutex_exit(&(lfp)->lf_lock); \
140 log_file_rele(lfp); \
141 }
142
143 int rfsl_log_buffer = 0;
144 static int rfsl_log_file = 0;
145
146 /* This array is used for memory allocation of record encoding spaces */
147 static struct {
148 int size;
149 struct kmem_cache *mem_cache;
150 char *cache_name;
151 } nfslog_mem_alloc[] = {
152 #define SMALL_INDX 0
153 { NFSLOG_SMALL_RECORD_SIZE - sizeof (struct lr_alloc),
154 NULL, NFSLOG_SMALL_REC_NAME },
155 #define MEDIUM_INDX 1
156 { NFSLOG_MEDIUM_RECORD_SIZE - sizeof (struct lr_alloc),
157 NULL, NFSLOG_MEDIUM_REC_NAME },
158 #define LARGE_INDX 2
159 { NFSLOG_LARGE_RECORD_SIZE - sizeof (struct lr_alloc),
160 NULL, NFSLOG_LARGE_REC_NAME },
161 { (-1), NULL }
162 };
163
164 /* Used to calculate the 'real' allocation size */
165 #define ALLOC_SIZE(index) \
166 (nfslog_mem_alloc[index].size + sizeof (struct lr_alloc))
167
168 /*
169 * Initialize logging data buffer cache
170 */
171 void
172 nfslog_init()
173 {
174 int indx;
175
176 rw_init(&nfslog_buffer_list_lock, NULL, RW_DEFAULT, NULL);
177
178 /*
179 * Initialize the kmem caches for encoding
180 */
181 for (indx = 0; nfslog_mem_alloc[indx].size != (-1); indx++) {
182 nfslog_mem_alloc[indx].mem_cache =
183 kmem_cache_create(nfslog_mem_alloc[indx].cache_name,
184 ALLOC_SIZE(indx), 0, NULL, NULL, NULL, NULL, NULL, 0);
185 }
186 }
187
188 /*
189 * Sets up the necessary log file and related buffers to enable logging
190 * on the given export point.
191 * Returns 0 on success, non-zero on failure.
192 */
193 int
194 nfslog_setup(struct exportinfo *exi)
195 {
196 struct exportdata *kex;
197 struct log_buffer *lbp;
198 struct log_buffer *nlbp;
199
200 kex = &exi->exi_export;
201 ASSERT(kex->ex_flags & EX_LOG);
202
203 /*
204 * Logging is enabled for the new export point, check
205 * the existing log_buffer structures to see if the
206 * desired buffer has already been opened. If so, point
207 * the new exportinfo's exi_logbuffer to the existing
208 * one.
209 */
210 rw_enter(&nfslog_buffer_list_lock, RW_READER);
211 for (lbp = nfslog_buffer_list; lbp != NULL; lbp = lbp->lb_next) {
212 LOGGING_DPRINT((10,
213 "searching for buffer... found log_buffer '%s'\n",
214 lbp->lb_path));
215 if (strcmp(lbp->lb_path, kex->ex_log_buffer) == 0) {
216 /* Found our match. Ref it and return */
217 LOG_BUFFER_HOLD(lbp);
218 exi->exi_logbuffer = lbp;
219 LOGGING_DPRINT((10, "\tfound log_buffer for '%s'\n",
220 kex->ex_log_buffer));
221 rw_exit(&nfslog_buffer_list_lock);
222 return (0);
223 }
224 }
225 rw_exit(&nfslog_buffer_list_lock);
226
227 /*
228 * New buffer needed, allocate it.
229 * The buffer list lock has been dropped so we will need to search
230 * the list again to ensure that another thread has not added
231 * a matching buffer.
232 */
233 if ((nlbp = log_buffer_create(kex->ex_log_buffer)) == NULL) {
234 /*
235 * Failed the buffer creation for some reason so we
236 * will need to return.
237 */
238 return (EIO);
239 }
240
241 rw_enter(&nfslog_buffer_list_lock, RW_WRITER);
242 for (lbp = nfslog_buffer_list; lbp != NULL;
243 lbp = lbp->lb_next) {
244 if (strcmp(lbp->lb_path, kex->ex_log_buffer) == 0) {
245 /*
246 * A log_buffer already exists for the
247 * indicated buffer, use it instead.
248 */
249 LOG_BUFFER_HOLD(lbp);
250
251 exi->exi_logbuffer = lbp;
252
253 LOGGING_DPRINT((10, "found log_buffer for '%s' "
254 "after allocation\n", kex->ex_log_buffer));
255
256 rw_exit(&nfslog_buffer_list_lock);
257
258 log_buffer_rele(nlbp);
259
260 return (0);
261 }
262 }
263 /*
264 * Didn't find an existing log_buffer for this buffer,
265 * use the the newly created one, and add to list. We
266 * increment the reference count because the node is
267 * entered into the global list.
268 */
269 LOGGING_DPRINT((10, "exportfs: adding nlbp=%p to list\n",
270 (void *)nlbp));
271
272 nlbp->lb_next = nfslog_buffer_list;
273 nfslog_buffer_list = nlbp;
274
275 LOG_BUFFER_HOLD(nlbp); /* hold is for export entry */
276 exi->exi_logbuffer = nlbp;
277
278 rw_exit(&nfslog_buffer_list_lock);
279
280 return (0);
281 }
282
283 /*
284 * Disables logging for the given export point.
285 */
286 void
287 nfslog_disable(struct exportinfo *exi)
288 {
289 log_buffer_rele(exi->exi_logbuffer);
290 }
291
292 /*
293 * Creates the corresponding log_buffer and log_file structures
294 * for the the buffer named 'name'.
295 * Returns a pointer to the log_buffer structure with reference one.
296 */
297 static struct log_buffer *
298 log_buffer_create(caddr_t name)
299 {
300 struct log_buffer *buffer;
301 struct log_file *logfile;
302 int namelen = strlen(name);
303
304 LOGGING_DPRINT((10, "log_buffer_create: %s\n", name));
305 if (log_file_create(name, &logfile))
306 return (NULL);
307
308 buffer = (struct log_buffer *)kmem_alloc(sizeof (*buffer), KM_SLEEP);
309 buffer->lb_refcnt = 1;
310 buffer->lb_rec_id = 0;
311 buffer->lb_path = (caddr_t)kmem_alloc(namelen + 1, KM_SLEEP);
312 bcopy(name, buffer->lb_path, namelen + 1);
313 buffer->lb_logfile = logfile;
314 buffer->lb_records = NULL;
315 buffer->lb_num_recs = 0;
316 buffer->lb_size_queued = 0;
317 mutex_init(&buffer->lb_lock, NULL, MUTEX_DEFAULT, NULL);
318 rfsl_log_buffer++;
319
320 return (buffer);
321 }
322
323 /*
324 * Release a log_buffer structure
325 */
326 static void
327 log_buffer_rele(struct log_buffer *lbp)
328 {
329 int len;
330
331 mutex_enter(&lbp->lb_lock);
332 if (--lbp->lb_refcnt > 1) {
333 mutex_exit(&lbp->lb_lock);
334 return;
335 }
336
337 if (lbp->lb_refcnt < 0) {
338 panic("log_rele: log_buffer refcnt < 0");
339 /*NOTREACHED*/
340 }
341
342 /*
343 * Need to drop the lb_lock before acquiring the
344 * nfslog_buffer_list_lock. To avoid double free we need
345 * to hold an additional reference to the log buffer.
346 * This will ensure that no two threads will simultaneously
347 * be trying to free the same log buffer.
348 */
349
350 if (lbp->lb_refcnt == 1) {
351
352 /*
353 * If the ref count is 1, then the last
354 * unshare/reference has been given up and we need to
355 * clean up the buffer and remove it from the buffer
356 * list.
357 */
358 LOGGING_DPRINT((10,
359 "log_buffer_rele lbp=%p disconnecting\n", (void *)lbp));
360 /*
361 * Hold additional reference before dropping the lb_lock
362 */
363
364 lbp->lb_refcnt++;
365 mutex_exit(&lbp->lb_lock);
366
367 /*
368 * Make sure that all of the buffered records are written.
369 * Don't bother checking the write return value since there
370 * isn't much we can do at this point.
371 */
372 (void) nfslog_records_flush_to_disk(lbp);
373
374 rw_enter(&nfslog_buffer_list_lock, RW_WRITER);
375 mutex_enter(&lbp->lb_lock);
376 /*
377 * Drop the reference count held above.
378 * If the ref count is still > 1 then someone has
379 * stepped in to use this log buffer. unlock and return.
380 */
381 if (--lbp->lb_refcnt > 1) {
382 mutex_exit(&lbp->lb_lock);
383 rw_exit(&nfslog_buffer_list_lock);
384 return;
385 }
386
387 if (lbp == nfslog_buffer_list) {
388 nfslog_buffer_list = lbp->lb_next;
389 } else {
390 struct log_buffer *tlbp;
391
392 /* Drop the log_buffer from the master list */
393 for (tlbp = nfslog_buffer_list; tlbp->lb_next != NULL;
394 tlbp = tlbp->lb_next) {
395 if (tlbp->lb_next == lbp) {
396 tlbp->lb_next = lbp->lb_next;
397 break;
398 }
399 }
400 }
401
402 mutex_exit(&lbp->lb_lock);
403 rw_exit(&nfslog_buffer_list_lock);
404 }
405 /*
406 * ref count zero; finish clean up.
407 */
408 LOGGING_DPRINT((10, "log_buffer_rele lbp=%p freeing\n", (void *)lbp));
409
410 log_file_rele(lbp->lb_logfile);
411 len = strlen(lbp->lb_path) + 1;
412 kmem_free(lbp->lb_path, len);
413 kmem_free(lbp, sizeof (*lbp));
414 rfsl_log_buffer--;
415 }
416
417 /*
418 * Creates the corresponding log_file structure for the buffer
419 * named 'log_file_name'.
420 * 'log_file_name' is created by concatenating 'origname' and LOG_INPROG_STRING.
421 * 'logfile' is set to be the log_file structure with reference one.
422 */
423 static int
424 log_file_create(caddr_t origname, struct log_file **lfpp)
425 {
426 vnode_t *vp = NULL;
427 char *name;
428 int namelen;
429 int error;
430 struct log_file *logfile = NULL;
431 vattr_t va;
432 caddr_t loghdr = NULL;
433 size_t loghdr_len = 0;
434 size_t loghdr_free = 0;
435
436 namelen = strlen(origname) + strlen(LOG_INPROG_STRING);
437 name = (caddr_t)kmem_alloc(namelen + 1, KM_SLEEP);
438 (void) sprintf(name, "%s%s", origname, LOG_INPROG_STRING);
439
440 LOGGING_DPRINT((3, "log_file_create: %s\n", name));
441 if (error = vn_open(name, UIO_SYSSPACE, FCREAT|FWRITE|FOFFMAX,
442 LOG_MODE, &vp, CRCREAT, 0)) {
443 nfs_cmn_err(error, CE_WARN,
444 "log_file_create: Can not open %s - error %m", name);
445 goto out;
446 }
447 LOGGING_DPRINT((3, "log_file_create: %s vp=%p v_count=%d\n",
448 name, (void *)vp, vp->v_count));
449
450 logfile = (struct log_file *)kmem_zalloc(sizeof (*logfile), KM_SLEEP);
451 logfile->lf_path = name;
452 /*
453 * No need to bump the vnode reference count since it is set
454 * to one by vn_open().
455 */
456 logfile->lf_vp = vp;
457 logfile->lf_refcnt = 1;
458 mutex_init(&logfile->lf_lock, NULL, MUTEX_DEFAULT, NULL);
459 rfsl_log_file++;
460
461 va.va_mask = AT_SIZE;
462 error = VOP_GETATTR(vp, &va, 0, CRED(), NULL);
463 if (error) {
464 nfs_cmn_err(error, CE_WARN,
465 "log_file_create: Can not stat %s - error = %m", name);
466 goto out;
467 }
468
469 if (va.va_size == 0) {
470 struct lr_alloc lr;
471
472 /*
473 * Write Header.
474 */
475 create_buffer_header(&loghdr, &loghdr_len, &loghdr_free);
476 /*
477 * Dummy up a lr_alloc struct for the write
478 */
479 lr.next = lr.prev = &lr;
480 lr.lr_flags = 0;
481 lr.log_record = loghdr;
482 lr.size = loghdr_len;
483 lr.alloc_cache = NULL;
484 lr.exi = NULL;
485 lr.lb = NULL;
486
487 mutex_enter(&logfile->lf_lock);
488
489 error = nfslog_write_logrecords(logfile, &lr, 1);
490
491 mutex_exit(&logfile->lf_lock);
492
493 if (error != 0) {
494 nfs_cmn_err(error, CE_WARN,
495 "log_file_create: Can not write header "
496 "on %s - error = %m", name);
497 goto out;
498 }
499 }
500 *lfpp = logfile;
501
502 if (loghdr != NULL)
503 kmem_free(loghdr, loghdr_free);
504
505 return (0);
506
507 out:
508 if (vp != NULL) {
509 int error1;
510 error1 = VOP_CLOSE(vp, FCREAT|FWRITE|FOFFMAX, 1, (offset_t)0,
511 CRED(), NULL);
512 if (error1) {
513 nfs_cmn_err(error1, CE_WARN,
514 "log_file_create: Can not close %s - "
515 "error = %m", name);
516 }
517 VN_RELE(vp);
518 }
519
520 kmem_free(name, namelen + 1);
521 if (logfile != NULL) {
522 mutex_destroy(&logfile->lf_lock);
523 kmem_free(logfile, sizeof (*logfile));
524 rfsl_log_file--;
525 }
526 if (loghdr != NULL)
527 kmem_free(loghdr, loghdr_free);
528
529 return (error);
530 }
531
532 /*
533 * Release a log_file structure
534 */
535 static void
536 log_file_rele(struct log_file *lfp)
537 {
538 int len;
539 int error;
540
541 mutex_enter(&lfp->lf_lock);
542 if (--lfp->lf_refcnt > 0) {
543 LOGGING_DPRINT((10,
544 "log_file_rele lfp=%p decremented refcnt to %d\n",
545 (void *)lfp, lfp->lf_refcnt));
546 mutex_exit(&lfp->lf_lock);
547 return;
548 }
549 if (lfp->lf_refcnt < 0) {
550 panic("log_file_rele: log_file refcnt < 0");
551 /*NOTREACHED*/
552 }
553
554 LOGGING_DPRINT((10, "log_file_rele lfp=%p freeing node\n",
555 (void *)lfp));
556
557 lfp->lf_flags &= ~(L_PRINTED | L_ERROR);
558
559 ASSERT(lfp->lf_flags == 0);
560 ASSERT(lfp->lf_writers == 0);
561
562 if (error = VOP_CLOSE(lfp->lf_vp, FCREAT|FWRITE|FOFFMAX, 1, (offset_t)0,
563 CRED(), NULL)) {
564 nfs_cmn_err(error, CE_WARN,
565 "NFS: Could not close log buffer %s - error = %m",
566 lfp->lf_path);
567 #ifdef DEBUG
568 } else {
569 LOGGING_DPRINT((3,
570 "log_file_rele: %s has been closed vp=%p v_count=%d\n",
571 lfp->lf_path, (void *)lfp->lf_vp, lfp->lf_vp->v_count));
572 #endif
573 }
574 VN_RELE(lfp->lf_vp);
575
576 len = strlen(lfp->lf_path) + 1;
577 kmem_free(lfp->lf_path, len);
578 kmem_free(lfp, sizeof (*lfp));
579 rfsl_log_file--;
580 }
581
582 /*
583 * Allocates a record of the size specified.
584 * 'exi' identifies the exportinfo structure being logged.
585 * 'size' indicates how much memory should be allocated
586 * 'cookie' is used to store an opaque value for the caller for later use
587 * 'flags' currently ignored.
588 *
589 * Returns a pointer to the beginning of the allocated memory.
590 * 'cookie' is a pointer to the 'lr_alloc' struct; this will be used
591 * to keep track of the encoded record and contains all the info
592 * for enqueuing the record on the log buffer for later writing.
593 *
594 * nfslog_record_put() must be used to 'free' this record or allocation.
595 */
596 /* ARGSUSED */
597 void *
598 nfslog_record_alloc(
599 struct exportinfo *exi,
600 int alloc_indx,
601 void **cookie,
602 int flags)
603 {
604 struct lr_alloc *lrp;
605
606 lrp = (struct lr_alloc *)
607 kmem_cache_alloc(nfslog_mem_alloc[alloc_indx].mem_cache,
608 KM_NOSLEEP);
609
610 if (lrp == NULL) {
611 *cookie = NULL;
612 return (NULL);
613 }
614
615 lrp->next = lrp;
616 lrp->prev = lrp;
617 lrp->lr_flags = 0;
618
619 lrp->log_record = (caddr_t)((uintptr_t)lrp +
620 (uintptr_t)sizeof (struct lr_alloc));
621 lrp->size = nfslog_mem_alloc[alloc_indx].size;
622 lrp->alloc_cache = nfslog_mem_alloc[alloc_indx].mem_cache;
623 lrp->exi = exi;
624
625 if (exi->exi_export.ex_flags & EX_LOG) {
626 LOG_BUFFER_HOLD(exi->exi_logbuffer);
627 lrp->lb = exi->exi_logbuffer;
628 } else {
629 lrp->lb = NULL;
630 }
631
632 *cookie = (void *)lrp;
633
634 LOGGING_DPRINT((3,
635 "nfslog_record_alloc(log_buffer=%p mem=%p size=%lu)\n",
636 (void *)exi->exi_logbuffer, (void *)lrp->log_record, lrp->size));
637 return (lrp->log_record);
638 }
639
640 /*
641 * After the above nfslog_record_alloc() has been called and a record
642 * encoded into the buffer that was returned, this function is called
643 * to handle appropriate disposition of the newly created record.
644 * The cookie value is the one that was returned from nfslog_record_alloc().
645 * Size is the actual size of the record that was encoded. This is
646 * passed in because the size used for the alloc was just an approximation.
647 * The sync parameter is used to tell us if we need to force this record
648 * to disk and if not it will be queued for later writing.
649 *
650 * Note that if the size parameter has a value of 0, then the record is
651 * not written to the log and the associated data structures are released.
652 */
653 void
654 nfslog_record_put(void *cookie, size_t size, bool_t sync,
655 unsigned int which_buffers)
656 {
657 struct lr_alloc *lrp = (struct lr_alloc *)cookie;
658 struct log_buffer *lbp = lrp->lb;
659
660 /*
661 * If the caller has nothing to write or if there is
662 * an apparent error, rele the buffer and free.
663 */
664 if (size == 0 || size > lrp->size) {
665 nfslog_free_logrecords(lrp);
666 return;
667 }
668
669 /*
670 * Reset the size to what actually needs to be written
671 * This is used later on when the iovec is built for
672 * writing the records to the log file.
673 */
674 lrp->size = size;
675
676 /* append to all if public exi */
677 if (which_buffers == NFSLOG_ALL_BUFFERS) {
678 (void) nfslog_record_append2all(lrp);
679 nfslog_free_logrecords(lrp);
680 return;
681 }
682
683 /* Insert the record on the list to be written */
684 mutex_enter(&lbp->lb_lock);
685 if (lbp->lb_records == NULL) {
686 lbp->lb_records = (caddr_t)lrp;
687 lbp->lb_num_recs = 1;
688 lbp->lb_size_queued = lrp->size;
689 } else {
690 insque(lrp, ((struct lr_alloc *)lbp->lb_records)->prev);
691 lbp->lb_num_recs++;
692 lbp->lb_size_queued += lrp->size;
693 }
694
695 /*
696 * Determine if the queue for this log buffer should be flushed.
697 * This is done by either the number of records queued, the total
698 * size of all records queued or by the request of the caller
699 * via the sync parameter.
700 */
701 if (lbp->lb_size_queued >= nfslog_num_bytes_to_write ||
702 lbp->lb_num_recs > nfslog_num_records_to_write || sync == TRUE) {
703 mutex_exit(&lbp->lb_lock);
704 (void) nfslog_records_flush_to_disk(lbp);
705 } else {
706 mutex_exit(&lbp->lb_lock);
707 }
708
709 }
710
711 /*
712 * Examine the log_buffer struct to see if there are queue log records
713 * that need to be written to disk. If some exist, pull them off of
714 * the log buffer and write them to the log file.
715 */
716 static int
717 nfslog_records_flush_to_disk(struct log_buffer *lbp)
718 {
719
720 mutex_enter(&lbp->lb_lock);
721
722 if (lbp->lb_records == NULL) {
723 mutex_exit(&lbp->lb_lock);
724 return (0);
725 }
726 return (nfslog_records_flush_to_disk_nolock(lbp));
727 }
728
729 /*
730 * Function requires that the caller holds lb_lock.
731 * Function flushes any records in the log buffer to the disk.
732 * Function drops the lb_lock on return.
733 */
734
735 static int
736 nfslog_records_flush_to_disk_nolock(struct log_buffer *lbp)
737 {
738 struct log_file *lfp = NULL;
739 struct lr_alloc *lrp_writers;
740 int num_recs;
741 int error = 0;
742
743 ASSERT(MUTEX_HELD(&lbp->lb_lock));
744
745 lfp = lbp->lb_logfile;
746
747 LOG_FILE_LOCK_TO_WRITE(lfp);
748 ASSERT(lbp->lb_records != NULL);
749
750 lrp_writers = (struct lr_alloc *)lbp->lb_records;
751 lbp->lb_records = NULL;
752 num_recs = lbp->lb_num_recs;
753 lbp->lb_num_recs = 0;
754 lbp->lb_size_queued = 0;
755 mutex_exit(&lbp->lb_lock);
756 error = nfslog_write_logrecords(lfp, lrp_writers, num_recs);
757
758 LOG_FILE_UNLOCK_FROM_WRITE(lfp);
759
760 nfslog_free_logrecords(lrp_writers);
761 return (error);
762 }
763
764
765 /*
766 * Take care of writing the provided log record(s) to the log file.
767 * We group the log records with an iovec and use VOP_WRITE to append
768 * them to the end of the log file.
769 */
770 static int
771 nfslog_write_logrecords(struct log_file *lfp,
772 struct lr_alloc *lrp_writers, int num_recs)
773 {
774 struct uio uio;
775 struct iovec *iovp;
776 int size_iovecs;
777 vnode_t *vp;
778 struct vattr va;
779 struct lr_alloc *lrp;
780 int i;
781 ssize_t len;
782 int ioflag = FAPPEND;
783 int error = 0;
784
785 ASSERT(MUTEX_HELD(&lfp->lf_lock));
786
787 vp = lfp->lf_vp;
788
789 size_iovecs = sizeof (struct iovec) * num_recs;
790 iovp = (struct iovec *)kmem_alloc(size_iovecs, KM_NOSLEEP);
791
792 if (iovp == NULL) {
793 error = ENOMEM;
794 goto out;
795 }
796
797 /* Build the iovec based on the list of log records */
798 i = 0;
799 len = 0;
800 lrp = lrp_writers;
801 do {
802 iovp[i].iov_base = lrp->log_record;
803 iovp[i].iov_len = lrp->size;
804 len += lrp->size;
805 lrp = lrp->next;
806 i++;
807 } while (lrp != lrp_writers);
808
809 ASSERT(i == num_recs);
810
811 uio.uio_iov = iovp;
812 uio.uio_iovcnt = num_recs;
813 uio.uio_loffset = 0;
814 uio.uio_segflg = (short)UIO_SYSSPACE;
815 uio.uio_resid = len;
816 uio.uio_llimit = (rlim64_t)MAXOFFSET_T;
817 uio.uio_fmode = FWRITE;
818 uio.uio_extflg = UIO_COPY_DEFAULT;
819
820 /*
821 * Save the size. If the write fails, reset the size to avoid
822 * corrupted log buffer files.
823 */
824 va.va_mask = AT_SIZE;
825
826 (void) VOP_RWLOCK(vp, V_WRITELOCK_TRUE, NULL); /* UIO_WRITE */
827 if ((error = VOP_GETATTR(vp, &va, 0, CRED(), NULL)) == 0) {
828 if ((len + va.va_size) < (MAXOFF32_T)) {
829 error = VOP_WRITE(vp, &uio, ioflag, CRED(), NULL);
830 VOP_RWUNLOCK(vp, V_WRITELOCK_TRUE, NULL);
831 if (uio.uio_resid)
832 error = ENOSPC;
833 if (error)
834 (void) VOP_SETATTR(vp, &va, 0, CRED(), NULL);
835 } else {
836 VOP_RWUNLOCK(vp, V_WRITELOCK_TRUE, NULL);
837 if (!(lfp->lf_flags & L_PRINTED)) {
838 cmn_err(CE_WARN,
839 "NFS Logging: buffer file %s exceeds 2GB; "
840 "stopped writing buffer \n", lfp->lf_path);
841 }
842 error = ENOSPC;
843 }
844 } else {
845 VOP_RWUNLOCK(vp, V_WRITELOCK_TRUE, NULL);
846 }
847
848 kmem_free(iovp, size_iovecs);
849
850 out:
851 if (error) {
852 if (!(lfp->lf_flags & L_PRINTED)) {
853 nfs_cmn_err(error, CE_WARN,
854 "NFS Logging disabled for buffer %s - "
855 "write error = %m\n", lfp->lf_path);
856 lfp->lf_flags |= L_PRINTED;
857 }
858 } else if (lfp->lf_flags & (L_ERROR | L_PRINTED)) {
859 lfp->lf_flags &= ~(L_ERROR | L_PRINTED);
860 cmn_err(CE_WARN,
861 "NFS Logging re-enabled for buffer %s\n", lfp->lf_path);
862 }
863
864 return (error);
865 }
866
867 static void
868 nfslog_free_logrecords(struct lr_alloc *lrp_writers)
869 {
870 struct lr_alloc *lrp = lrp_writers;
871 struct lr_alloc *lrp_free;
872
873 do {
874 lrp_free = lrp;
875
876 lrp = lrp->next;
877
878 /*
879 * Check to see if we are supposed to free this structure
880 * and relese the log_buffer ref count.
881 * It may be the case that the caller does not want this
882 * structure and its record contents freed just yet.
883 */
884 if ((lrp_free->lr_flags & LR_ALLOC_NOFREE) == 0) {
885 if (lrp_free->lb != NULL)
886 log_buffer_rele(lrp_free->lb);
887 if (lrp_free->alloc_cache) /* double check */
888 kmem_cache_free(lrp_free->alloc_cache,
889 (void *)lrp_free);
890 } else {
891 /*
892 * after being pulled from the list the
893 * pointers need to be reinitialized.
894 */
895 lrp_free->next = lrp_free;
896 lrp_free->prev = lrp_free;
897 }
898
899 } while (lrp != lrp_writers);
900 }
901
902 /*
903 * Rename lbp->lb_logfile to reflect the true name requested by 'share'
904 */
905 static int
906 nfslog_logbuffer_rename(struct log_buffer *lbp)
907 {
908 struct log_file *lf;
909 int error;
910 struct log_file *logfile;
911
912 /*
913 * Try our best to get the cache records into the log file
914 * before the rename occurs.
915 */
916 (void) nfslog_records_flush_to_disk(lbp);
917
918 /*
919 * Hold lb_lock before retrieving
920 * lb_logfile.
921 * Hold a reference to the
922 * "lf" structure. this is
923 * same as LOG_FILE_HOLD()
924 */
925 mutex_enter(&(lbp)->lb_lock);
926 lf = lbp->lb_logfile;
927 mutex_enter(&(lf)->lf_lock);
928 mutex_exit(&(lbp)->lb_lock);
929 lf->lf_refcnt++;
930 mutex_exit(&(lf)->lf_lock);
931
932 LOGGING_DPRINT((10, "nfslog_logbuffer_rename: renaming %s to %s\n",
933 lf->lf_path, lbp->lb_path));
934
935 /*
936 * rename the current buffer to what the daemon expects
937 */
938 if (error = nfslog_logfile_rename(lf->lf_path, lbp->lb_path))
939 goto out;
940
941 /*
942 * Create a new working buffer file and have all new data sent there.
943 */
944 if (error = log_file_create(lbp->lb_path, &logfile)) {
945 /* Attempt to rename to original */
946 (void) nfslog_logfile_rename(lbp->lb_path, lf->lf_path);
947 goto out;
948 }
949
950 /*
951 * Hold the lb_lock here, this will make
952 * all the threads trying to access lb->logfile block
953 * and get a new logfile structure instead of old one.
954 */
955 mutex_enter(&(lbp)->lb_lock);
956 lbp->lb_logfile = logfile;
957 mutex_exit(&(lbp)->lb_lock);
958
959 LOG_FILE_RELE(lf); /* release log_buffer's reference */
960
961 /*
962 * Wait for log_file to be in a quiescent state before we
963 * return to our caller to let it proceed with the reading of
964 * this file.
965 */
966 nfslog_logfile_wait(lf);
967
968 out:
969 /*
970 * Release our reference on "lf" in two different cases.
971 * 1. Error condition, release only the reference
972 * that we held at the begining of this
973 * routine on "lf" structure.
974 * 2. Fall through condition, no errors but the old
975 * logfile structure "lf" has been replaced with
976 * the new "logfile" structure, so release the
977 * reference that was part of the creation of
978 * "lf" structure to free up the resources.
979 */
980
981 LOG_FILE_RELE(lf);
982
983 return (error);
984 }
985
986 /*
987 * Renames the 'from' file to 'new'.
988 */
989 static int
990 nfslog_logfile_rename(char *from, char *new)
991 {
992 int error;
993
994 if (error = vn_rename(from, new, UIO_SYSSPACE)) {
995 cmn_err(CE_WARN,
996 "nfslog_logfile_rename: couldn't rename %s to %s\n",
997 from, new);
998 }
999 return (error);
1000 }
1001
1002 /*
1003 * Wait for the log_file writers to finish before returning
1004 */
1005 static void
1006 nfslog_logfile_wait(struct log_file *lf)
1007 {
1008 mutex_enter(&lf->lf_lock);
1009 while (lf->lf_writers > 0) {
1010 lf->lf_flags |= L_WAITING;
1011 (void) cv_wait_sig(&lf->lf_cv_waiters, &lf->lf_lock);
1012 }
1013 mutex_exit(&lf->lf_lock);
1014 }
1015
1016 static int
1017 nfslog_record_append2all(struct lr_alloc *lrp)
1018 {
1019 struct log_buffer *lbp, *nlbp;
1020 int error, ret_error = 0;
1021 int lr_flags = lrp->lr_flags;
1022
1023 rw_enter(&nfslog_buffer_list_lock, RW_READER);
1024 if ((lbp = nfslog_buffer_list) != NULL)
1025 LOG_BUFFER_HOLD(lbp);
1026 for (nlbp = NULL; lbp != NULL; lbp = nlbp) {
1027 if ((nlbp = lbp->lb_next) != NULL) {
1028 /*
1029 * Remember next element in the list
1030 */
1031 LOG_BUFFER_HOLD(nlbp);
1032 }
1033 rw_exit(&nfslog_buffer_list_lock);
1034
1035 /*
1036 * Insert the record on the buffer's list to be written
1037 * and then flush the records to the log file.
1038 * Make sure to set the no free flag so that the
1039 * record can be used for the next write
1040 */
1041 lrp->lr_flags = LR_ALLOC_NOFREE;
1042
1043 ASSERT(lbp != NULL);
1044 mutex_enter(&lbp->lb_lock);
1045 if (lbp->lb_records == NULL) {
1046 lbp->lb_records = (caddr_t)lrp;
1047 lbp->lb_num_recs = 1;
1048 lbp->lb_size_queued = lrp->size;
1049 } else {
1050 insque(lrp, ((struct lr_alloc *)lbp->lb_records)->prev);
1051 lbp->lb_num_recs++;
1052 lbp->lb_size_queued += lrp->size;
1053 }
1054
1055 /*
1056 * Flush log records to disk.
1057 * Function is called with lb_lock held.
1058 * Function drops the lb_lock on return.
1059 */
1060 error = nfslog_records_flush_to_disk_nolock(lbp);
1061
1062 if (error) {
1063 ret_error = -1;
1064 nfs_cmn_err(error, CE_WARN,
1065 "rfsl_log_pubfh: could not append record to "
1066 "\"%s\" error = %m\n", lbp->lb_path);
1067 }
1068 log_buffer_rele(lbp);
1069 rw_enter(&nfslog_buffer_list_lock, RW_READER);
1070 }
1071 rw_exit(&nfslog_buffer_list_lock);
1072
1073 lrp->lr_flags = lr_flags;
1074
1075 return (ret_error);
1076 }
1077
1078 #ifdef DEBUG
1079 static int logging_debug = 0;
1080
1081 /*
1082 * 0) no debugging
1083 * 3) current test software
1084 * 10) random stuff
1085 */
1086 void
1087 nfslog_dprint(const int level, const char *fmt, ...)
1088 {
1089 va_list args;
1090
1091 if (logging_debug == level ||
1092 (logging_debug > 10 && (logging_debug - 10) >= level)) {
1093 va_start(args, fmt);
1094 (void) vprintf(fmt, args);
1095 va_end(args);
1096 }
1097 }
1098
1099 #endif /* DEBUG */
1100
1101 /*
1102 * NFS Log Flush system call
1103 * Caller must check privileges.
1104 */
1105 /* ARGSUSED */
1106 int
1107 nfsl_flush(struct nfsl_flush_args *args, model_t model)
1108 {
1109 struct flush_thread_params *tparams;
1110 struct nfsl_flush_args *nfsl_args;
1111 int error = 0;
1112 ulong_t buffer_len;
1113 STRUCT_HANDLE(nfsl_flush_args, uap);
1114
1115 STRUCT_SET_HANDLE(uap, model, args);
1116
1117 tparams = (struct flush_thread_params *)
1118 kmem_zalloc(sizeof (*tparams), KM_SLEEP);
1119
1120 nfsl_args = &tparams->tp_args;
1121 nfsl_args->version = STRUCT_FGET(uap, version);
1122 if (nfsl_args->version != NFSL_FLUSH_ARGS_VERS) {
1123 cmn_err(CE_WARN, "nfsl_flush: exected version %d, got %d",
1124 NFSL_FLUSH_ARGS_VERS, nfsl_args->version);
1125 return (EIO);
1126 }
1127
1128 nfsl_args->directive = STRUCT_FGET(uap, directive);
1129 if ((nfsl_args->directive & NFSL_ALL) == 0) {
1130 /*
1131 * Process a specific buffer
1132 */
1133 nfsl_args->buff_len = STRUCT_FGET(uap, buff_len);
1134
1135 nfsl_args->buff = (char *)
1136 kmem_alloc(nfsl_args->buff_len, KM_NOSLEEP);
1137 if (nfsl_args->buff == NULL)
1138 return (ENOMEM);
1139
1140 error = copyinstr((const char *)STRUCT_FGETP(uap, buff),
1141 nfsl_args->buff, nfsl_args->buff_len, &buffer_len);
1142 if (error)
1143 return (EFAULT);
1144
1145 if (nfsl_args->buff_len != buffer_len)
1146 return (EFAULT);
1147 }
1148
1149 LOGGING_DPRINT((10, "nfsl_flush: Flushing %s buffer(s)\n",
1150 nfsl_args->directive & NFSL_ALL ? "all" : nfsl_args->buff));
1151
1152 if (nfsl_args->directive & NFSL_SYNC) {
1153 /*
1154 * Do the work synchronously
1155 */
1156 nfslog_do_flush(tparams);
1157 error = tparams->tp_error;
1158 kmem_free(nfsl_args->buff, nfsl_args->buff_len);
1159 kmem_free(tparams, sizeof (*tparams));
1160 } else {
1161 /*
1162 * Do the work asynchronously
1163 */
1164 (void) thread_create(NULL, 0, nfslog_do_flush,
1165 tparams, 0, &p0, TS_RUN, minclsyspri);
1166 }
1167
1168 return (error);
1169 }
1170
1171 /*
1172 * This is where buffer flushing would occur, but there is no buffering
1173 * at this time.
1174 * Possibly rename the log buffer for processing.
1175 * Sets tparams->ta_error equal to the value of the error that occurred,
1176 * 0 otherwise.
1177 * Returns ENOENT if the buffer is not found.
1178 */
1179 static void
1180 nfslog_do_flush(struct flush_thread_params *tparams)
1181 {
1182 struct nfsl_flush_args *args;
1183 struct log_buffer *lbp, *nlbp;
1184 int error = ENOENT;
1185 int found = 0;
1186 char *buf_inprog; /* name of buff in progress */
1187 int buf_inprog_len;
1188
1189 /*
1190 * Sanity check on the arguments.
1191 */
1192 if (!tparams)
1193 return;
1194 args = &tparams->tp_args;
1195 if (!args)
1196 return;
1197
1198 rw_enter(&nfslog_buffer_list_lock, RW_READER);
1199 if ((lbp = nfslog_buffer_list) != NULL) {
1200 LOG_BUFFER_HOLD(lbp);
1201 }
1202 for (nlbp = NULL; lbp != NULL; lbp = nlbp) {
1203 if ((nlbp = lbp->lb_next) != NULL) {
1204 LOG_BUFFER_HOLD(nlbp);
1205 }
1206 rw_exit(&nfslog_buffer_list_lock);
1207 if (args->directive & NFSL_ALL) {
1208 (void) nfslog_records_flush_to_disk(lbp);
1209 } else {
1210 if ((strcmp(lbp->lb_path, args->buff) == 0) &&
1211 (args->directive & NFSL_RENAME)) {
1212 error = nfslog_logbuffer_rename(lbp);
1213 found++;
1214 if (nlbp != NULL)
1215 log_buffer_rele(nlbp);
1216 log_buffer_rele(lbp);
1217 break;
1218 }
1219 }
1220 log_buffer_rele(lbp);
1221 rw_enter(&nfslog_buffer_list_lock, RW_READER);
1222 }
1223 if (!found)
1224 rw_exit(&nfslog_buffer_list_lock);
1225
1226 if (!found && ((args->directive & NFSL_ALL) == 0) &&
1227 (args->directive & NFSL_RENAME)) {
1228 /*
1229 * The specified buffer is not currently in use,
1230 * simply rename the file indicated.
1231 */
1232 buf_inprog_len = strlen(args->buff) +
1233 strlen(LOG_INPROG_STRING) + 1;
1234 buf_inprog = (caddr_t)kmem_alloc(buf_inprog_len, KM_SLEEP);
1235 (void) sprintf(buf_inprog, "%s%s",
1236 args->buff, LOG_INPROG_STRING);
1237
1238 error = nfslog_logfile_rename(buf_inprog, args->buff);
1239
1240 kmem_free(buf_inprog, buf_inprog_len);
1241 }
1242
1243 out:
1244 if ((args->directive & NFSL_SYNC) == 0) {
1245 /*
1246 * Work was performed asynchronously, the caller is
1247 * no longer waiting for us.
1248 * Free the thread arguments and exit.
1249 */
1250 kmem_free(args->buff, args->buff_len);
1251 kmem_free(tparams, sizeof (*tparams));
1252 thread_exit();
1253 /* NOTREACHED */
1254 }
1255
1256 tparams->tp_error = error;
1257 }
1258
1259 /*
1260 * Generate buffer_header.
1261 * 'loghdr' points the the buffer_header, and *reclen
1262 * contains the length of the buffer.
1263 */
1264 static void
1265 create_buffer_header(caddr_t *loghdr, size_t *reclen, size_t *freesize)
1266 {
1267 timestruc_t now;
1268 nfslog_buffer_header lh;
1269 XDR xdrs;
1270 unsigned int final_size;
1271
1272
1273 /* pick some size that will hold the buffer_header */
1274 *freesize = NFSLOG_SMALL_RECORD_SIZE;
1275
1276 /*
1277 * Fill header
1278 */
1279 lh.bh_length = 0; /* don't know yet how large it will be */
1280 lh.bh_version = NFSLOG_BUF_VERSION;
1281 lh.bh_flags = 0;
1282 lh.bh_offset = 0;
1283 gethrestime(&now);
1284 TIMESPEC_TO_TIMESPEC32(&lh.bh_timestamp, &now);
1285
1286 /*
1287 * Encode the header
1288 */
1289 *loghdr = (caddr_t)kmem_alloc(*freesize, KM_SLEEP);
1290 xdrmem_create(&xdrs, *loghdr, *freesize, XDR_ENCODE);
1291
1292 (void) xdr_nfslog_buffer_header(&xdrs, &lh);
1293
1294 /*
1295 * Reset with final size of the encoded data
1296 */
1297 final_size = xdr_getpos(&xdrs);
1298 xdr_setpos(&xdrs, 0);
1299 (void) xdr_u_int(&xdrs, &final_size);
1300
1301 *reclen = (size_t)final_size;
1302 }
1303
1304 /*
1305 * ****************************************************************
1306 * RPC dispatch table for logging
1307 * Indexed by program, version, proc
1308 * Based on NFS dispatch table.
1309 */
1310 struct nfslog_proc_disp {
1311 bool_t (*xdrargs)();
1312 bool_t (*xdrres)();
1313 bool_t affects_transactions; /* Operation affects transaction */
1314 /* processing */
1315 };
1316
1317 struct nfslog_vers_disp {
1318 int nfslog_dis_nprocs; /* number of procs */
1319 struct nfslog_proc_disp *nfslog_dis_proc_table; /* proc array */
1320 };
1321
1322 struct nfslog_prog_disp {
1323 int nfslog_dis_prog; /* program number */
1324 int nfslog_dis_versmin; /* Minimum version value */
1325 int nfslog_dis_nvers; /* Number of version values */
1326 struct nfslog_vers_disp *nfslog_dis_vers_table; /* versions array */
1327 };
1328
1329 static int rfs_log_bad = 0; /* incremented on bad log attempts */
1330 static int rfs_log_good = 0; /* incremented on successful log attempts */
1331
1332 /*
1333 * Define the actions taken per prog/vers/proc:
1334 *
1335 * In some cases, the nl types are the same as the nfs types and a simple
1336 * bcopy should suffice. Rather that define tens of identical procedures,
1337 * simply define these to bcopy. Similarly this takes care of different
1338 * procs that use same parameter struct.
1339 */
1340
1341 static struct nfslog_proc_disp nfslog_proc_v2[] = {
1342 /*
1343 * NFS VERSION 2
1344 */
1345
1346 /* RFS_NULL = 0 */
1347 {xdr_void, xdr_void, FALSE},
1348
1349 /* RFS_GETATTR = 1 */
1350 {xdr_fhandle, xdr_nfslog_getattrres, FALSE},
1351
1352 /* RFS_SETATTR = 2 */
1353 {xdr_nfslog_setattrargs, xdr_nfsstat, TRUE},
1354
1355 /* RFS_ROOT = 3 *** NO LONGER SUPPORTED *** */
1356 {xdr_void, xdr_void, FALSE},
1357
1358 /* RFS_LOOKUP = 4 */
1359 {xdr_nfslog_diropargs, xdr_nfslog_diropres, TRUE},
1360
1361 /* RFS_READLINK = 5 */
1362 {xdr_fhandle, xdr_nfslog_rdlnres, FALSE},
1363
1364 /* RFS_READ = 6 */
1365 {xdr_nfslog_nfsreadargs, xdr_nfslog_rdresult, TRUE},
1366
1367 /* RFS_WRITECACHE = 7 *** NO LONGER SUPPORTED *** */
1368 {xdr_void, xdr_void, FALSE},
1369
1370 /* RFS_WRITE = 8 */
1371 {xdr_nfslog_writeargs, xdr_nfslog_writeresult, TRUE},
1372
1373 /* RFS_CREATE = 9 */
1374 {xdr_nfslog_createargs, xdr_nfslog_diropres, TRUE},
1375
1376 /* RFS_REMOVE = 10 */
1377 {xdr_nfslog_diropargs, xdr_nfsstat, TRUE},
1378
1379 /* RFS_RENAME = 11 */
1380 {xdr_nfslog_rnmargs, xdr_nfsstat, TRUE},
1381
1382 /* RFS_LINK = 12 */
1383 {xdr_nfslog_linkargs, xdr_nfsstat, TRUE},
1384
1385 /* RFS_SYMLINK = 13 */
1386 {xdr_nfslog_symlinkargs, xdr_nfsstat, TRUE},
1387
1388 /* RFS_MKDIR = 14 */
1389 {xdr_nfslog_createargs, xdr_nfslog_diropres, TRUE},
1390
1391 /* RFS_RMDIR = 15 */
1392 {xdr_nfslog_diropargs, xdr_nfsstat, TRUE},
1393
1394 /* RFS_READDIR = 16 */
1395 {xdr_nfslog_rddirargs, xdr_nfslog_rddirres, TRUE},
1396
1397 /* RFS_STATFS = 17 */
1398 {xdr_fhandle, xdr_nfslog_statfs, FALSE},
1399 };
1400
1401
1402 /*
1403 * NFS VERSION 3
1404 */
1405
1406 static struct nfslog_proc_disp nfslog_proc_v3[] = {
1407
1408 /* NFSPROC3_NULL = 0 */
1409 {xdr_void, xdr_void, FALSE},
1410
1411 /* NFSPROC3_GETATTR = 1 */
1412 {xdr_nfslog_nfs_fh3, xdr_nfslog_GETATTR3res, FALSE},
1413
1414 /* NFSPROC3_SETATTR = 2 */
1415 {xdr_nfslog_SETATTR3args, xdr_nfslog_SETATTR3res, TRUE},
1416
1417 /* NFSPROC3_LOOKUP = 3 */
1418 {xdr_nfslog_diropargs3, xdr_nfslog_LOOKUP3res, TRUE},
1419
1420 /* NFSPROC3_ACCESS = 4 */
1421 {xdr_nfslog_ACCESS3args, xdr_nfslog_ACCESS3res, FALSE},
1422
1423 /* NFSPROC3_READLINK = 5 */
1424 {xdr_nfslog_nfs_fh3, xdr_nfslog_READLINK3res, FALSE},
1425
1426 /* NFSPROC3_READ = 6 */
1427 {xdr_nfslog_READ3args, xdr_nfslog_READ3res, TRUE},
1428
1429 /* NFSPROC3_WRITE = 7 */
1430 {xdr_nfslog_WRITE3args, xdr_nfslog_WRITE3res, TRUE},
1431
1432 /* NFSPROC3_CREATE = 8 */
1433 {xdr_nfslog_CREATE3args, xdr_nfslog_CREATE3res, TRUE},
1434
1435 /* NFSPROC3_MKDIR = 9 */
1436 {xdr_nfslog_MKDIR3args, xdr_nfslog_MKDIR3res, TRUE},
1437
1438 /* NFSPROC3_SYMLINK = 10 */
1439 {xdr_nfslog_SYMLINK3args, xdr_nfslog_SYMLINK3res, TRUE},
1440
1441 /* NFSPROC3_MKNOD = 11 */
1442 {xdr_nfslog_MKNOD3args, xdr_nfslog_MKNOD3res, TRUE},
1443
1444 /* NFSPROC3_REMOVE = 12 */
1445 {xdr_nfslog_REMOVE3args, xdr_nfslog_REMOVE3res, TRUE},
1446
1447 /* NFSPROC3_RMDIR = 13 */
1448 {xdr_nfslog_RMDIR3args, xdr_nfslog_RMDIR3res, TRUE},
1449
1450 /* NFSPROC3_RENAME = 14 */
1451 {xdr_nfslog_RENAME3args, xdr_nfslog_RENAME3res, TRUE},
1452
1453 /* NFSPROC3_LINK = 15 */
1454 {xdr_nfslog_LINK3args, xdr_nfslog_LINK3res, TRUE},
1455
1456 /* NFSPROC3_READDIR = 16 */
1457 {xdr_nfslog_READDIR3args, xdr_nfslog_READDIR3res, TRUE},
1458
1459 /* NFSPROC3_READDIRPLUS = 17 */
1460 {xdr_nfslog_READDIRPLUS3args, xdr_nfslog_READDIRPLUS3res, TRUE},
1461
1462 /* NFSPROC3_FSSTAT = 18 */
1463 {xdr_nfslog_FSSTAT3args, xdr_nfslog_FSSTAT3res, FALSE},
1464
1465 /* NFSPROC3_FSINFO = 19 */
1466 {xdr_nfslog_FSINFO3args, xdr_nfslog_FSINFO3res, FALSE},
1467
1468 /* NFSPROC3_PATHCONF = 20 */
1469 {xdr_nfslog_PATHCONF3args, xdr_nfslog_PATHCONF3res, FALSE},
1470
1471 /* NFSPROC3_COMMIT = 21 */
1472 {xdr_nfslog_COMMIT3args, xdr_nfslog_COMMIT3res, FALSE},
1473 };
1474
1475 static struct nfslog_proc_disp nfslog_proc_v1[] = {
1476 /*
1477 * NFSLOG VERSION 1
1478 */
1479
1480 /* NFSLOG_NULL = 0 */
1481 {xdr_void, xdr_void, TRUE},
1482
1483 /* NFSLOG_SHARE = 1 */
1484 {xdr_nfslog_sharefsargs, xdr_nfslog_sharefsres, TRUE},
1485
1486 /* NFSLOG_UNSHARE = 2 */
1487 {xdr_nfslog_sharefsargs, xdr_nfslog_sharefsres, TRUE},
1488
1489 /* NFSLOG_LOOKUP = 3 */
1490 {xdr_nfslog_diropargs3, xdr_nfslog_LOOKUP3res, TRUE},
1491
1492 /* NFSLOG_GETFH = 4 */
1493 {xdr_nfslog_getfhargs, xdr_nfsstat, TRUE},
1494 };
1495
1496 static struct nfslog_vers_disp nfslog_vers_disptable[] = {
1497 {sizeof (nfslog_proc_v2) / sizeof (nfslog_proc_v2[0]),
1498 nfslog_proc_v2},
1499 {sizeof (nfslog_proc_v3) / sizeof (nfslog_proc_v3[0]),
1500 nfslog_proc_v3},
1501 };
1502
1503 static struct nfslog_vers_disp nfslog_nfslog_vers_disptable[] = {
1504 {sizeof (nfslog_proc_v1) / sizeof (nfslog_proc_v1[0]),
1505 nfslog_proc_v1},
1506 };
1507
1508 static struct nfslog_prog_disp nfslog_dispatch_table[] = {
1509 {NFS_PROGRAM, NFS_VERSMIN,
1510 (sizeof (nfslog_vers_disptable) /
1511 sizeof (nfslog_vers_disptable[0])),
1512 nfslog_vers_disptable},
1513
1514 {NFSLOG_PROGRAM, NFSLOG_VERSMIN,
1515 (sizeof (nfslog_nfslog_vers_disptable) /
1516 sizeof (nfslog_nfslog_vers_disptable[0])),
1517 nfslog_nfslog_vers_disptable},
1518 };
1519
1520 static int nfslog_dispatch_table_arglen = sizeof (nfslog_dispatch_table) /
1521 sizeof (nfslog_dispatch_table[0]);
1522
1523 /*
1524 * This function will determine the appropriate export info struct to use
1525 * and allocate a record id to be used in the written log buffer.
1526 * Usually this is a straightforward operation but the existence of the
1527 * multicomponent lookup and its semantics of crossing file system
1528 * boundaries add to the complexity. See the comments below...
1529 */
1530 struct exportinfo *
1531 nfslog_get_exi(
1532 struct exportinfo *exi,
1533 struct svc_req *req,
1534 caddr_t res,
1535 unsigned int *nfslog_rec_id)
1536 {
1537 struct log_buffer *lb;
1538 struct exportinfo *exi_ret = NULL;
1539 fhandle_t *fh;
1540 nfs_fh3 *fh3;
1541
1542 if (exi == NULL)
1543 return (NULL);
1544
1545 /*
1546 * If the exi is marked for logging, allocate a record id and return
1547 */
1548 if (exi->exi_export.ex_flags & EX_LOG) {
1549 lb = exi->exi_logbuffer;
1550
1551 /* obtain the unique record id for the caller */
1552 *nfslog_rec_id = atomic_add_32_nv(&lb->lb_rec_id, (int32_t)1);
1553
1554 /*
1555 * The caller will expect to be able to exi_rele() it,
1556 * so exi->exi_count must be incremented before it can
1557 * be returned, to make it uniform with exi_ret->exi_count
1558 */
1559 exi_hold(exi);
1560 return (exi);
1561 }
1562
1563 if (exi != exi_public)
1564 return (NULL);
1565
1566 /*
1567 * Here we have an exi that is not marked for logging.
1568 * It is possible that this request is a multicomponent lookup
1569 * that was done from the public file handle (not logged) and
1570 * the resulting file handle being returned to the client exists
1571 * in a file system that is being logged. If this is the case
1572 * we need to log this multicomponent lookup to the appropriate
1573 * log buffer. This will allow for the appropriate path name
1574 * mapping to occur at user level.
1575 */
1576 if (req->rq_prog == NFS_PROGRAM) {
1577 switch (req->rq_vers) {
1578 case NFS_V3:
1579 if ((req->rq_proc == NFSPROC3_LOOKUP) &&
1580 (((LOOKUP3res *)res)->status == NFS3_OK)) {
1581 fh3 = &((LOOKUP3res *)res)->res_u.ok.object;
1582 exi_ret = checkexport(&fh3->fh3_fsid,
1583 FH3TOXFIDP(fh3), NULL);
1584 }
1585 break;
1586
1587 case NFS_VERSION:
1588 if ((req->rq_proc == RFS_LOOKUP) &&
1589 (((struct nfsdiropres *)
1590 res)->dr_status == NFS_OK)) {
1591 fh = &((struct nfsdiropres *)res)->
1592 dr_u.dr_drok_u.drok_fhandle;
1593 exi_ret = checkexport(&fh->fh_fsid,
1594 (fid_t *)&fh->fh_xlen, NULL);
1595 }
1596 break;
1597 default:
1598 break;
1599 }
1600 }
1601
1602 if (exi_ret != NULL && exi_ret->exi_export.ex_flags & EX_LOG) {
1603 lb = exi_ret->exi_logbuffer;
1604 /* obtain the unique record id for the caller */
1605 *nfslog_rec_id = atomic_add_32_nv(&lb->lb_rec_id, (int32_t)1);
1606
1607 return (exi_ret);
1608 }
1609 return (NULL);
1610 }
1611
1612 #ifdef DEBUG
1613 static long long rfslog_records_ignored = 0;
1614 #endif
1615
1616 /*
1617 * nfslog_write_record - Fill in the record buffer for writing out.
1618 * If logrecp is null, log it, otherwise, malloc the record and return it.
1619 *
1620 * It is the responsibility of the caller to check whether this exportinfo
1621 * has logging enabled.
1622 * Note that nfslog_share_public_record() only needs to check for the
1623 * existence of at least one logbuffer to which the public filehandle record
1624 * needs to be logged.
1625 */
1626 void
1627 nfslog_write_record(struct exportinfo *exi, struct svc_req *req,
1628 caddr_t args, caddr_t res, cred_t *cr, struct netbuf *pnb,
1629 unsigned int record_id, unsigned int which_buffers)
1630 {
1631 struct nfslog_prog_disp *progtable; /* prog struct */
1632 struct nfslog_vers_disp *verstable; /* version struct */
1633 struct nfslog_proc_disp *disp = NULL; /* proc struct */
1634 int i, vers;
1635 void *log_cookie; /* for logrecord if */
1636 caddr_t buffer;
1637 XDR xdrs;
1638 unsigned int final_size;
1639 int encode_ok;
1640 int alloc_indx;
1641
1642 ASSERT(exi != NULL); ASSERT(req != NULL); ASSERT(args != NULL);
1643 ASSERT(res != NULL); ASSERT(cr != NULL);
1644
1645 /*
1646 * Find program element
1647 * Search the list since program can not be used as index
1648 */
1649 for (i = 0; (i < nfslog_dispatch_table_arglen); i++) {
1650 if (req->rq_prog == nfslog_dispatch_table[i].nfslog_dis_prog)
1651 break;
1652 }
1653 if (i >= nfslog_dispatch_table_arglen) { /* program not logged */
1654 /* not an error */
1655 return;
1656 }
1657
1658 /*
1659 * Extract the dispatch functions based on program/version
1660 */
1661 progtable = &nfslog_dispatch_table[i];
1662 vers = req->rq_vers - progtable->nfslog_dis_versmin;
1663 verstable = &progtable->nfslog_dis_vers_table[vers];
1664 disp = &verstable->nfslog_dis_proc_table[req->rq_proc];
1665
1666 if (!(exi->exi_export.ex_flags & EX_LOG_ALLOPS) &&
1667 !disp->affects_transactions) {
1668 /*
1669 * Only interested in logging operations affecting
1670 * transaction generation. This is not one of them.
1671 */
1672 #ifdef DEBUG
1673 rfslog_records_ignored++;
1674 #endif
1675 return;
1676 }
1677
1678 switch (req->rq_prog) {
1679 case NFS_PROGRAM:
1680 switch (req->rq_vers) {
1681 case NFS_V3:
1682 switch (req->rq_proc) {
1683 case NFSPROC3_READDIRPLUS:
1684 alloc_indx = MEDIUM_INDX;
1685 break;
1686 default:
1687 alloc_indx = SMALL_INDX;
1688 break;
1689 }
1690 break;
1691 default:
1692 alloc_indx = SMALL_INDX;
1693 break;
1694 }
1695 break;
1696 case NFSLOG_PROGRAM:
1697 alloc_indx = MEDIUM_INDX;
1698 break;
1699 default:
1700 alloc_indx = SMALL_INDX;
1701 break;
1702 }
1703
1704 do {
1705 encode_ok = FALSE;
1706
1707 /* Pick the size to alloc; end of the road - return */
1708 if (nfslog_mem_alloc[alloc_indx].size == (-1)) {
1709 cmn_err(CE_WARN,
1710 "NFSLOG: unable to encode record - prog=%d "
1711 "proc = %d", req->rq_prog, req->rq_proc);
1712 return;
1713 }
1714
1715 buffer = nfslog_record_alloc(exi, alloc_indx, &log_cookie, 0);
1716 if (buffer == NULL) {
1717 /* Error processing - no space alloced */
1718 rfs_log_bad++;
1719 cmn_err(CE_WARN, "NFSLOG: can't get record");
1720 return;
1721 }
1722
1723 xdrmem_create(&xdrs, buffer,
1724 nfslog_mem_alloc[alloc_indx].size, XDR_ENCODE);
1725
1726 /*
1727 * Encode the header, args and results of the record
1728 */
1729 if (xdr_nfslog_request_record(&xdrs, exi, req, cr, pnb,
1730 nfslog_mem_alloc[alloc_indx].size, record_id) &&
1731 (*disp->xdrargs)(&xdrs, args) &&
1732 (*disp->xdrres)(&xdrs, res)) {
1733 encode_ok = TRUE;
1734
1735 rfs_log_good++;
1736 /*
1737 * Get the final size of the encoded
1738 * data and insert that length at the
1739 * beginning.
1740 */
1741 final_size = xdr_getpos(&xdrs);
1742 xdr_setpos(&xdrs, 0);
1743 (void) xdr_u_int(&xdrs, &final_size);
1744 } else {
1745 /* Oops, the encode failed so we need to free memory */
1746 nfslog_record_put(log_cookie, 0, FALSE, which_buffers);
1747 alloc_indx++;
1748 }
1749
1750 } while (encode_ok == FALSE);
1751
1752
1753 /*
1754 * Take the final log record and put it in the log file.
1755 * This may be queued to the file internally and written
1756 * later unless the last parameter is TRUE.
1757 * If the record_id is 0 then this is most likely a share/unshare
1758 * request and it should be written synchronously to the log file.
1759 */
1760 nfslog_record_put(log_cookie,
1761 final_size, (record_id == 0), which_buffers);
1762 }
1763
1764 static char *
1765 get_publicfh_path(int *alloc_length)
1766 {
1767 extern struct exportinfo *exi_public;
1768 char *pubpath;
1769
1770 rw_enter(&exported_lock, RW_READER);
1771
1772 *alloc_length = exi_public->exi_export.ex_pathlen + 1;
1773 pubpath = kmem_alloc(*alloc_length, KM_SLEEP);
1774
1775 (void) strcpy(pubpath, exi_public->exi_export.ex_path);
1776
1777 rw_exit(&exported_lock);
1778
1779 return (pubpath);
1780 }
1781
1782 static void
1783 log_public_record(struct exportinfo *exi, cred_t *cr)
1784 {
1785 struct svc_req req;
1786 struct netbuf nb = {0, 0, NULL};
1787 int free_length = 0;
1788 diropargs3 args;
1789 LOOKUP3res res;
1790
1791 bzero(&req, sizeof (req));
1792 req.rq_prog = NFSLOG_PROGRAM;
1793 req.rq_vers = NFSLOG_VERSION;
1794 req.rq_proc = NFSLOG_LOOKUP;
1795 req.rq_cred.oa_flavor = AUTH_NONE;
1796
1797 bzero(&args, sizeof (diropargs3));
1798 bzero(&res, sizeof (LOOKUP3res));
1799
1800 args.dir.fh3_length = 0;
1801 if ((args.name = get_publicfh_path(&free_length)) == NULL)
1802 return;
1803 args.dirp = &args.dir;
1804
1805 res.status = NFS3_OK;
1806 res.res_u.ok.object.fh3_length = 0;
1807
1808 /*
1809 * Calling this function with the exi_public
1810 * will have the effect of appending the record
1811 * to each of the open log buffers
1812 */
1813 nfslog_write_record(exi, &req,
1814 (caddr_t)&args, (caddr_t)&res, cr, &nb, 0, NFSLOG_ALL_BUFFERS);
1815
1816 kmem_free(args.name, free_length);
1817 }
1818
1819 /*
1820 * nfslog_share_record - logs a share request.
1821 * This is not an NFS request, but we pretend here...
1822 */
1823 void
1824 nfslog_share_record(struct exportinfo *exi, cred_t *cr)
1825 {
1826 struct svc_req req;
1827 int res = 0;
1828 struct netbuf nb = {0, 0, NULL};
1829
1830 ASSERT(exi != NULL);
1831
1832 if (nfslog_buffer_list == NULL)
1833 return;
1834
1835 if (exi->exi_export.ex_flags & EX_LOG) {
1836 bzero(&req, sizeof (req));
1837 req.rq_prog = NFSLOG_PROGRAM;
1838 req.rq_vers = NFSLOG_VERSION;
1839 req.rq_proc = NFSLOG_SHARE;
1840 req.rq_cred.oa_flavor = AUTH_NONE;
1841 nfslog_write_record(exi, &req, (caddr_t)exi, (caddr_t)&res, cr,
1842 &nb, 0, NFSLOG_ONE_BUFFER);
1843 }
1844
1845 log_public_record(exi, cr);
1846 }
1847
1848 /*
1849 * nfslog_unshare_record - logs an unshare request.
1850 * This is not an NFS request, but we pretend here...
1851 */
1852 void
1853 nfslog_unshare_record(struct exportinfo *exi, cred_t *cr)
1854 {
1855 struct svc_req req;
1856 int res = 0;
1857 struct netbuf nb = {0, 0, NULL};
1858
1859 ASSERT(exi != NULL);
1860 ASSERT(exi->exi_export.ex_flags & EX_LOG);
1861
1862 bzero(&req, sizeof (req));
1863 req.rq_prog = NFSLOG_PROGRAM;
1864 req.rq_vers = NFSLOG_VERSION;
1865 req.rq_proc = NFSLOG_UNSHARE;
1866 req.rq_cred.oa_flavor = AUTH_NONE;
1867 nfslog_write_record(exi, &req,
1868 (caddr_t)exi, (caddr_t)&res, cr, &nb, 0, NFSLOG_ONE_BUFFER);
1869 }
1870
1871
1872 void
1873 nfslog_getfh(struct exportinfo *exi,
1874 fhandle *fh,
1875 char *fname,
1876 enum uio_seg seg,
1877 cred_t *cr)
1878 {
1879 struct svc_req req;
1880 int res = 0;
1881 struct netbuf nb = {0, 0, NULL};
1882 int error = 0;
1883 char *namebuf;
1884 size_t len;
1885 nfslog_getfhargs gfh;
1886
1887 ASSERT(exi != NULL);
1888 ASSERT(exi->exi_export.ex_flags & EX_LOG);
1889
1890 bzero(&req, sizeof (req));
1891 req.rq_prog = NFSLOG_PROGRAM;
1892 req.rq_vers = NFSLOG_VERSION;
1893 req.rq_proc = NFSLOG_GETFH;
1894 req.rq_cred.oa_flavor = AUTH_NONE;
1895
1896 namebuf = kmem_alloc(MAXPATHLEN + 4, KM_SLEEP);
1897 if (seg == UIO_USERSPACE) {
1898 error = copyinstr(fname, namebuf, MAXPATHLEN, &len);
1899 } else {
1900 error = copystr(fname, namebuf, MAXPATHLEN, &len);
1901 }
1902
1903 if (!error) {
1904 gfh.gfh_fh_buf = *fh;
1905 gfh.gfh_path = namebuf;
1906
1907 nfslog_write_record(exi, &req, (caddr_t)&gfh, (caddr_t)&res,
1908 cr, &nb, 0, NFSLOG_ONE_BUFFER);
1909 }
1910 kmem_free(namebuf, MAXPATHLEN + 4);
1911 }