xref: /netbsd-src/share/examples/puffs/pgfs/pgfs_subs.c (revision 86c517f33d938ded3c0e66e70e266a02847fdf87)
1 /*	$NetBSD: pgfs_subs.c,v 1.5 2012/04/11 14:28:18 yamt Exp $	*/
2 
3 /*-
4  * Copyright (c)2010,2011 YAMAMOTO Takashi,
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 /*
30  * a file system server which stores the data in a PostgreSQL database.
31  */
32 
33 /*
34  * we use large objects to store file contents.  there are a few XXXs wrt it.
35  *
36  * - large objects don't obey the normal transaction semantics.
37  *
38  * - we use large object server-side functions directly (instead of via the
39  *   libpq large object api) because:
40  *	- we want to use asynchronous (in the sense of PQsendFoo) operations
41  *	  which is not available with the libpq large object api.
42  *	- with the libpq large object api, there's no way to know details of
43  *	  an error because PGresult is freed in the library without saving
44  *	  PG_DIAG_SQLSTATE etc.
45  */
46 
47 #include <sys/cdefs.h>
48 #ifndef lint
49 __RCSID("$NetBSD: pgfs_subs.c,v 1.5 2012/04/11 14:28:18 yamt Exp $");
50 #endif /* not lint */
51 
52 #include <assert.h>
53 #include <err.h>
54 #include <errno.h>
55 #include <puffs.h>
56 #include <inttypes.h>
57 #include <stdarg.h>
58 #include <stdbool.h>
59 #include <stdio.h>
60 #include <stdlib.h>
61 #include <time.h>
62 #include <util.h>
63 
64 #include <libpq-fe.h>
65 #include <libpq/libpq-fs.h>	/* INV_* */
66 
67 #include "pgfs.h"
68 #include "pgfs_db.h"
69 #include "pgfs_debug.h"
70 #include "pgfs_waitq.h"
71 #include "pgfs_subs.h"
72 
73 const char * const vtype_table[] = {
74 	[VREG] = "regular",
75 	[VDIR] = "directory",
76 	[VLNK] = "link",
77 };
78 
79 static unsigned int
tovtype(const char * type)80 tovtype(const char *type)
81 {
82 	unsigned int i;
83 
84 	for (i = 0; i < __arraycount(vtype_table); i++) {
85 		if (vtype_table[i] == NULL) {
86 			continue;
87 		}
88 		if (!strcmp(type, vtype_table[i])) {
89 			return i;
90 		}
91 	}
92 	assert(0);
93 	return 0;
94 }
95 
96 static const char *
fromvtype(enum vtype vtype)97 fromvtype(enum vtype vtype)
98 {
99 
100 	if (vtype < __arraycount(vtype_table)) {
101 		assert(vtype_table[vtype] != NULL);
102 		return vtype_table[vtype];
103 	}
104 	return NULL;
105 }
106 
107 /*
108  * fileid_lock stuff below is to keep ordering of operations for a file.
109  * it is a workaround for the lack of operation barriers in the puffs
110  * protocol.
111  *
112  * currently we do this locking only for SETATTR, GETATTR, and WRITE as
113  * they are known to be reorder-unsafe.  they are sensitive to the file
114  * attributes, mainly the file size.  note that as the kernel issues async
115  * SETATTR/WRITE requests, vnode lock doesn't prevent GETATTR from seeing
116  * the stale attributes.
117  *
118  * we are relying on waiton/wakeup being a FIFO.
119  */
120 
121 struct fileid_lock_handle {
122 	TAILQ_ENTRY(fileid_lock_handle) list;
123 	fileid_t fileid;
124 	struct puffs_cc *owner;	/* diagnostic only */
125 	struct waitq waitq;
126 };
127 
128 TAILQ_HEAD(, fileid_lock_handle) fileid_lock_list =
129     TAILQ_HEAD_INITIALIZER(fileid_lock_list);
130 struct waitq fileid_lock_waitq = TAILQ_HEAD_INITIALIZER(fileid_lock_waitq);
131 
132 /*
133  * fileid_lock: serialize requests for the fileid.
134  *
135  * this function should be the first yieldable point in a puffs callback.
136  */
137 
138 struct fileid_lock_handle *
fileid_lock(fileid_t fileid,struct puffs_cc * cc)139 fileid_lock(fileid_t fileid, struct puffs_cc *cc)
140 {
141 	struct fileid_lock_handle *lock;
142 
143 	TAILQ_FOREACH(lock, &fileid_lock_list, list) {
144 		if (lock->fileid == fileid) {
145 			DPRINTF("fileid wait %" PRIu64 " cc %p\n", fileid, cc);
146 			assert(lock->owner != cc);
147 			waiton(&lock->waitq, cc);	/* enter FIFO */
148 			assert(lock->owner == cc);
149 			return lock;
150 		}
151 	}
152 	lock = emalloc(sizeof(*lock));
153 	lock->fileid = fileid;
154 	lock->owner = cc;
155 	DPRINTF("fileid lock %" PRIu64 " cc %p\n", lock->fileid, cc);
156 	waitq_init(&lock->waitq);
157 	TAILQ_INSERT_HEAD(&fileid_lock_list, lock, list);
158 	return lock;
159 }
160 
161 void
fileid_unlock(struct fileid_lock_handle * lock)162 fileid_unlock(struct fileid_lock_handle *lock)
163 {
164 
165 	DPRINTF("fileid unlock %" PRIu64 "\n", lock->fileid);
166 	assert(lock != NULL);
167 	assert(lock->owner != NULL);
168 	/*
169 	 * perform direct-handoff to the first waiter.
170 	 *
171 	 * a handoff is essential to keep the order of requests.
172 	 */
173 	lock->owner = wakeup_one(&lock->waitq);
174 	if (lock->owner != NULL) {
175 		return;
176 	}
177 	/*
178 	 * no one is waiting this fileid.
179 	 */
180 	TAILQ_REMOVE(&fileid_lock_list, lock, list);
181 	free(lock);
182 }
183 
184 /*
185  * timespec_to_pgtimestamp: create a text representation of timestamp which
186  * can be recognized by the database server.
187  *
188  * it's caller's responsibility to free(3) the result.
189  */
190 
191 int
timespec_to_pgtimestamp(const struct timespec * tv,char ** resultp)192 timespec_to_pgtimestamp(const struct timespec *tv, char **resultp)
193 {
194 	/*
195 	 * XXX is there any smarter way?
196 	 */
197 	char buf1[1024];
198 	char buf2[1024];
199 	struct tm tm_store;
200 	struct tm *tm;
201 
202 	tm = gmtime_r(&tv->tv_sec, &tm_store);
203 	if (tm == NULL) {
204 		assert(errno != 0);
205 		return errno;
206 	}
207 	strftime(buf1, sizeof(buf1), "%Y%m%dT%H%M%S", tm);
208 	snprintf(buf2, sizeof(buf2), "%s.%ju", buf1,
209 	    (uintmax_t)tv->tv_nsec / 1000);
210 	*resultp = estrdup(buf2);
211 	return 0;
212 }
213 
214 int
my_lo_truncate(struct Xconn * xc,int32_t fd,int32_t size)215 my_lo_truncate(struct Xconn *xc, int32_t fd, int32_t size)
216 {
217 	static struct cmd *c;
218 	int32_t ret;
219 	int error;
220 
221 	CREATECMD(c, "SELECT lo_truncate($1, $2)", INT4OID, INT4OID);
222 	error = sendcmd(xc, c, fd, size);
223 	if (error != 0) {
224 		return error;
225 	}
226 	error = simplefetch(xc, INT4OID, &ret);
227 	if (error != 0) {
228 		if (error == EEXIST) {
229 			/*
230 			 * probably the insertion of the new-sized page
231 			 * caused a duplicated key error.  retry.
232 			 */
233 			DPRINTF("map EEXIST to EAGAIN\n");
234 			error = EAGAIN;
235 		}
236 		return error;
237 	}
238 	assert(ret == 0);
239 	return 0;
240 }
241 
242 int
my_lo_lseek(struct Xconn * xc,int32_t fd,int32_t offset,int32_t whence,int32_t * retp)243 my_lo_lseek(struct Xconn *xc, int32_t fd, int32_t offset, int32_t whence,
244     int32_t *retp)
245 {
246 	static struct cmd *c;
247 	int32_t ret;
248 	int error;
249 
250 	CREATECMD(c, "SELECT lo_lseek($1, $2, $3)", INT4OID, INT4OID, INT4OID);
251 	error = sendcmd(xc, c, fd, offset, whence);
252 	if (error != 0) {
253 		return error;
254 	}
255 	error = simplefetch(xc, INT4OID, &ret);
256 	if (error != 0) {
257 		return error;
258 	}
259 	if (retp != NULL) {
260 		*retp = ret;
261 	}
262 	return 0;
263 }
264 
265 int
my_lo_read(struct Xconn * xc,int32_t fd,void * buf,size_t size,size_t * resultsizep)266 my_lo_read(struct Xconn *xc, int32_t fd, void *buf, size_t size,
267     size_t *resultsizep)
268 {
269 	static struct cmd *c;
270 	size_t resultsize;
271 	int error;
272 
273 	CREATECMD(c, "SELECT loread($1, $2)", INT4OID, INT4OID);
274 	error = sendcmdx(xc, 1, c, fd, (int32_t)size);
275 	if (error != 0) {
276 		return error;
277 	}
278 	error = simplefetch(xc, BYTEA, buf, &resultsize);
279 	if (error != 0) {
280 		return error;
281 	}
282 	*resultsizep = resultsize;
283 	if (size != resultsize) {
284 		DPRINTF("shortread? %zu != %zu\n", size, resultsize);
285 	}
286 	return 0;
287 }
288 
289 int
my_lo_write(struct Xconn * xc,int32_t fd,const void * buf,size_t size,size_t * resultsizep)290 my_lo_write(struct Xconn *xc, int32_t fd, const void *buf, size_t size,
291     size_t *resultsizep)
292 {
293 	static struct cmd *c;
294 	int32_t resultsize;
295 	int error;
296 
297 	CREATECMD(c, "SELECT lowrite($1, $2)", INT4OID, BYTEA);
298 	error = sendcmd(xc, c, fd, buf, (int32_t)size);
299 	if (error != 0) {
300 		return error;
301 	}
302 	error = simplefetch(xc, INT4OID, &resultsize);
303 	if (error != 0) {
304 		if (error == EEXIST) {
305 			/*
306 			 * probably the insertion of the new data page
307 			 * caused a duplicated key error.  retry.
308 			 */
309 			DPRINTF("map EEXIST to EAGAIN\n");
310 			error = EAGAIN;
311 		}
312 		return error;
313 	}
314 	*resultsizep = resultsize;
315 	if (size != (size_t)resultsize) {
316 		DPRINTF("shortwrite? %zu != %zu\n", size, (size_t)resultsize);
317 	}
318 	return 0;
319 }
320 
321 int
my_lo_open(struct Xconn * xc,Oid loid,int32_t mode,int32_t * fdp)322 my_lo_open(struct Xconn *xc, Oid loid, int32_t mode, int32_t *fdp)
323 {
324 	static struct cmd *c;
325 	int error;
326 
327 	CREATECMD(c, "SELECT lo_open($1, $2)", OIDOID, INT4OID);
328 	error = sendcmd(xc, c, loid, mode);
329 	if (error != 0) {
330 		return error;
331 	}
332 	return simplefetch(xc, INT4OID, fdp);
333 }
334 
335 int
my_lo_close(struct Xconn * xc,int32_t fd)336 my_lo_close(struct Xconn *xc, int32_t fd)
337 {
338 #if 1
339 	/*
340 	 * do nothing.
341 	 *
342 	 * LO handles are automatically closed at the end of transactions.
343 	 * our transactions are small enough.
344 	 */
345 #else
346 	static struct cmd *c;
347 	int32_t ret;
348 	int error;
349 
350 	CREATECMD(c, "SELECT lo_close($1)", INT4OID);
351 	error = sendcmd(xc, c, fd);
352 	if (error != 0) {
353 		return error;
354 	}
355 	error = simplefetch(xc, INT4OID, &ret);
356 	if (error != 0) {
357 		return error;
358 	}
359 	assert(ret == 0);
360 #endif
361 	return 0;
362 }
363 
364 static int
lo_lookup_by_fileid(struct Xconn * xc,fileid_t fileid,Oid * idp)365 lo_lookup_by_fileid(struct Xconn *xc, fileid_t fileid, Oid *idp)
366 {
367 	static struct cmd *c;
368 	static const Oid types[] = { OIDOID, };
369 	struct fetchstatus s;
370 	int error;
371 
372 	CREATECMD(c, "SELECT loid FROM datafork WHERE fileid = $1", INT8OID);
373 	error = sendcmd(xc, c, fileid);
374 	if (error != 0) {
375 		return error;
376 	}
377 	fetchinit(&s, xc);
378 	error = FETCHNEXT(&s, types, idp);
379 	fetchdone(&s);
380 	DPRINTF("error %d\n", error);
381 	return error;
382 }
383 
384 int
lo_open_by_fileid(struct Xconn * xc,fileid_t fileid,int mode,int * fdp)385 lo_open_by_fileid(struct Xconn *xc, fileid_t fileid, int mode, int *fdp)
386 {
387 	Oid loid;
388 	int fd;
389 	int error;
390 
391 	error = lo_lookup_by_fileid(xc, fileid, &loid);
392 	if (error != 0) {
393 		return error;
394 	}
395 	error = my_lo_open(xc, loid, mode, &fd);
396 	if (error != 0) {
397 		return error;
398 	}
399 	*fdp = fd;
400 	return 0;
401 }
402 
403 static int
getsize(struct Xconn * xc,fileid_t fileid,int * resultp)404 getsize(struct Xconn *xc, fileid_t fileid, int *resultp)
405 {
406 	int32_t size;
407 	int fd;
408 	int error;
409 
410 	error = lo_open_by_fileid(xc, fileid, INV_READ, &fd);
411 	if (error != 0) {
412 		return error;
413 	}
414 	error = my_lo_lseek(xc, fd, 0, SEEK_END, &size);
415 	if (error != 0) {
416 		return error;
417 	}
418 	error = my_lo_close(xc, fd);
419 	if (error != 0) {
420 		return error;
421 	}
422 	*resultp = size;
423 	return 0;
424 }
425 
426 #define	GETATTR_TYPE	0x00000001
427 #define	GETATTR_NLINK	0x00000002
428 #define	GETATTR_SIZE	0x00000004
429 #define	GETATTR_MODE	0x00000008
430 #define	GETATTR_UID	0x00000010
431 #define	GETATTR_GID	0x00000020
432 #define	GETATTR_TIME	0x00000040
433 #define	GETATTR_ALL	\
434 	(GETATTR_TYPE|GETATTR_NLINK|GETATTR_SIZE|GETATTR_MODE| \
435 	GETATTR_UID|GETATTR_GID|GETATTR_TIME)
436 
437 int
getattr(struct Xconn * xc,fileid_t fileid,struct vattr * va,unsigned int mask)438 getattr(struct Xconn *xc, fileid_t fileid, struct vattr *va, unsigned int mask)
439 {
440 	char *type;
441 	long long atime_s;
442 	long long atime_us;
443 	long long ctime_s;
444 	long long ctime_us;
445 	long long mtime_s;
446 	long long mtime_us;
447 	long long btime_s;
448 	long long btime_us;
449 	uint64_t mode;
450 	long long uid;
451 	long long gid;
452 	long long nlink;
453 	long long rev;
454 	struct fetchstatus s;
455 	int error;
456 
457 	if (mask == 0) {
458 		return 0;
459 	}
460 	/*
461 	 * unless explicitly requested, avoid fetching timestamps as they
462 	 * are a little more expensive than other simple attributes.
463 	 */
464 	if ((mask & GETATTR_TIME) != 0) {
465 		static struct cmd *c;
466 		static const Oid types[] = {
467 			TEXTOID,
468 			INT8OID,
469 			INT8OID,
470 			INT8OID,
471 			INT8OID,
472 			INT8OID,
473 			INT8OID,
474 			INT8OID,
475 			INT8OID,
476 			INT8OID,
477 			INT8OID,
478 			INT8OID,
479 			INT8OID,
480 			INT8OID,
481 		};
482 
483 		CREATECMD(c, "SELECT type::text, mode, uid, gid, nlink, rev, "
484 		    "extract(epoch from date_trunc('second', atime))::int8, "
485 		    "extract(microseconds from atime)::int8, "
486 		    "extract(epoch from date_trunc('second', ctime))::int8, "
487 		    "extract(microseconds from ctime)::int8, "
488 		    "extract(epoch from date_trunc('second', mtime))::int8, "
489 		    "extract(microseconds from mtime)::int8, "
490 		    "extract(epoch from date_trunc('second', btime))::int8, "
491 		    "extract(microseconds from btime)::int8 "
492 		    "FROM file "
493 		    "WHERE fileid = $1", INT8OID);
494 		error = sendcmd(xc, c, fileid);
495 		if (error != 0) {
496 			return error;
497 		}
498 		fetchinit(&s, xc);
499 		error = FETCHNEXT(&s, types, &type, &mode, &uid, &gid, &nlink,
500 		    &rev,
501 		    &atime_s, &atime_us,
502 		    &ctime_s, &ctime_us,
503 		    &mtime_s, &mtime_us,
504 		    &btime_s, &btime_us);
505 	} else {
506 		static struct cmd *c;
507 		static const Oid types[] = {
508 			TEXTOID,
509 			INT8OID,
510 			INT8OID,
511 			INT8OID,
512 			INT8OID,
513 			INT8OID,
514 		};
515 
516 		CREATECMD(c, "SELECT type::text, mode, uid, gid, nlink, rev "
517 		    "FROM file "
518 		    "WHERE fileid = $1", INT8OID);
519 		error = sendcmd(xc, c, fileid);
520 		if (error != 0) {
521 			return error;
522 		}
523 		fetchinit(&s, xc);
524 		error = FETCHNEXT(&s, types, &type, &mode, &uid, &gid, &nlink,
525 		    &rev);
526 	}
527 	fetchdone(&s);
528 	if (error != 0) {
529 		return error;
530 	}
531 	memset(va, 0xaa, sizeof(*va)); /* fill with garbage for debug */
532 	va->va_type = tovtype(type);
533 	free(type);
534 	va->va_mode = mode;
535 	va->va_uid = uid;
536 	va->va_gid = gid;
537 	if (nlink > 0 && va->va_type == VDIR) {
538 		nlink++; /* "." */
539 	}
540 	va->va_nlink = nlink;
541 	va->va_fileid = fileid;
542 	va->va_atime.tv_sec = atime_s;
543 	va->va_atime.tv_nsec = atime_us * 1000;
544 	va->va_ctime.tv_sec = ctime_s;
545 	va->va_ctime.tv_nsec = ctime_us * 1000;
546 	va->va_mtime.tv_sec = mtime_s;
547 	va->va_mtime.tv_nsec = mtime_us * 1000;
548 	va->va_birthtime.tv_sec = btime_s;
549 	va->va_birthtime.tv_nsec = btime_us * 1000;
550 	va->va_blocksize = LOBLKSIZE;
551 	va->va_gen = 1;
552 	va->va_filerev = rev;
553 	if ((mask & GETATTR_SIZE) != 0) {
554 		int size;
555 
556 		size = 0;
557 		if (va->va_type == VREG || va->va_type == VLNK) {
558 			error = getsize(xc, fileid, &size);
559 			if (error != 0) {
560 				return error;
561 			}
562 		} else if (va->va_type == VDIR) {
563 			size = 100; /* XXX */
564 		}
565 		va->va_size = size;
566 	}
567 	/*
568 	 * XXX va_bytes: likely wrong due to toast compression.
569 	 * there's no cheap way to get the compressed size of LO.
570 	 */
571 	va->va_bytes = va->va_size;
572 	va->va_flags = 0;
573 	return 0;
574 }
575 
576 int
update_mctime(struct Xconn * xc,fileid_t fileid)577 update_mctime(struct Xconn *xc, fileid_t fileid)
578 {
579 	static struct cmd *c;
580 
581 	CREATECMD(c,
582 	    "UPDATE file "
583 	    "SET mtime = current_timestamp, ctime = current_timestamp, "
584 		"rev = rev + 1 "
585 	    "WHERE fileid = $1", INT8OID);
586 	return simplecmd(xc, c, fileid);
587 }
588 
589 int
update_atime(struct Xconn * xc,fileid_t fileid)590 update_atime(struct Xconn *xc, fileid_t fileid)
591 {
592 	static struct cmd *c;
593 
594 	CREATECMD(c,
595 	    "UPDATE file SET atime = current_timestamp WHERE fileid = $1",
596 	    INT8OID);
597 	return simplecmd(xc, c, fileid);
598 }
599 
600 int
update_mtime(struct Xconn * xc,fileid_t fileid)601 update_mtime(struct Xconn *xc, fileid_t fileid)
602 {
603 	static struct cmd *c;
604 
605 	CREATECMD(c,
606 	    "UPDATE file "
607 	    "SET mtime = current_timestamp, rev = rev + 1 "
608 	    "WHERE fileid = $1", INT8OID);
609 	return simplecmd(xc, c, fileid);
610 }
611 
612 int
update_ctime(struct Xconn * xc,fileid_t fileid)613 update_ctime(struct Xconn *xc, fileid_t fileid)
614 {
615 	static struct cmd *c;
616 
617 	CREATECMD(c,
618 	    "UPDATE file SET ctime = current_timestamp WHERE fileid = $1",
619 	    INT8OID);
620 	return simplecmd(xc, c, fileid);
621 }
622 
623 int
update_nlink(struct Xconn * xc,fileid_t fileid,int delta)624 update_nlink(struct Xconn *xc, fileid_t fileid, int delta)
625 {
626 	static struct cmd *c;
627 
628 	CREATECMD(c,
629 	    "UPDATE file "
630 	    "SET nlink = nlink + $1 "
631 	    "WHERE fileid = $2",
632 	    INT8OID, INT8OID);
633 	return simplecmd(xc, c, (int64_t)delta, fileid);
634 }
635 
636 int
lookupp(struct Xconn * xc,fileid_t fileid,fileid_t * parent)637 lookupp(struct Xconn *xc, fileid_t fileid, fileid_t *parent)
638 {
639 	static struct cmd *c;
640 	static const Oid types[] = { INT8OID, };
641 	struct fetchstatus s;
642 	int error;
643 
644 	CREATECMD(c, "SELECT parent_fileid FROM dirent "
645 		"WHERE child_fileid = $1 LIMIT 1", INT8OID);
646 	error = sendcmd(xc, c, fileid);
647 	if (error != 0) {
648 		return error;
649 	}
650 	fetchinit(&s, xc);
651 	error = FETCHNEXT(&s, types, parent);
652 	fetchdone(&s);
653 	if (error != 0) {
654 		return error;
655 	}
656 	return 0;
657 }
658 
659 int
mkfile(struct Xconn * xc,enum vtype vtype,mode_t mode,uid_t uid,gid_t gid,fileid_t * idp)660 mkfile(struct Xconn *xc, enum vtype vtype, mode_t mode, uid_t uid, gid_t gid,
661     fileid_t *idp)
662 {
663 	static struct cmd *c;
664 	const char *type;
665 	int error;
666 
667 	type = fromvtype(vtype);
668 	if (type == NULL) {
669 		return EOPNOTSUPP;
670 	}
671 	CREATECMD(c,
672 		"INSERT INTO file "
673 		"(fileid, type, mode, uid, gid, nlink, rev, "
674 		"atime, ctime, mtime, btime) "
675 		"VALUES(nextval('fileid_seq'), $1::filetype, $2, $3, $4, 0, 0, "
676 		"current_timestamp, "
677 		"current_timestamp, "
678 		"current_timestamp, "
679 		"current_timestamp) "
680 		"RETURNING fileid", TEXTOID, INT8OID, INT8OID, INT8OID);
681 	error = sendcmd(xc, c, type, (uint64_t)mode, (uint64_t)uid,
682 	    (uint64_t)gid);
683 	if (error != 0) {
684 		return error;
685 	}
686 	return simplefetch(xc, INT8OID, idp);
687 }
688 
689 int
linkfile(struct Xconn * xc,fileid_t parent,const char * name,fileid_t child)690 linkfile(struct Xconn *xc, fileid_t parent, const char *name, fileid_t child)
691 {
692 	static struct cmd *c;
693 	int error;
694 
695 	CREATECMD(c,
696 		"INSERT INTO dirent "
697 		"(parent_fileid, name, child_fileid) "
698 		"VALUES($1, $2, $3)", INT8OID, TEXTOID, INT8OID);
699 	error = simplecmd(xc, c, parent, name, child);
700 	if (error != 0) {
701 		return error;
702 	}
703 	error = update_nlink(xc, child, 1);
704 	if (error != 0) {
705 		return error;
706 	}
707 	return update_mtime(xc, parent);
708 }
709 
710 int
unlinkfile(struct Xconn * xc,fileid_t parent,const char * name,fileid_t child)711 unlinkfile(struct Xconn *xc, fileid_t parent, const char *name, fileid_t child)
712 {
713 	static struct cmd *c;
714 	int error;
715 
716 	/*
717 	 * in addition to the primary key, we check child_fileid as well here
718 	 * to avoid removing an entry which was appeared after our VOP_LOOKUP.
719 	 */
720 	CREATECMD(c,
721 		"DELETE FROM dirent "
722 		"WHERE parent_fileid = $1 AND name = $2 AND child_fileid = $3",
723 		INT8OID, TEXTOID, INT8OID);
724 	error = simplecmd(xc, c, parent, name, child);
725 	if (error != 0) {
726 		return error;
727 	}
728 	error = update_nlink(xc, child, -1);
729 	if (error != 0) {
730 		return error;
731 	}
732 	error = update_mtime(xc, parent);
733 	if (error != 0) {
734 		return error;
735 	}
736 	return update_ctime(xc, child);
737 }
738 
739 int
mklinkfile(struct Xconn * xc,fileid_t parent,const char * name,enum vtype vtype,mode_t mode,uid_t uid,gid_t gid,fileid_t * idp)740 mklinkfile(struct Xconn *xc, fileid_t parent, const char *name,
741     enum vtype vtype, mode_t mode, uid_t uid, gid_t gid, fileid_t *idp)
742 {
743 	fileid_t fileid;
744 	int error;
745 
746 	error = mkfile(xc, vtype, mode, uid, gid, &fileid);
747 	if (error != 0) {
748 		return error;
749 	}
750 	error = linkfile(xc, parent, name, fileid);
751 	if (error != 0) {
752 		return error;
753 	}
754 	if (idp != NULL) {
755 		*idp = fileid;
756 	}
757 	return 0;
758 }
759 
760 int
mklinkfile_lo(struct Xconn * xc,fileid_t parent_fileid,const char * name,enum vtype vtype,mode_t mode,uid_t uid,gid_t gid,fileid_t * fileidp,int * loidp)761 mklinkfile_lo(struct Xconn *xc, fileid_t parent_fileid, const char *name,
762     enum vtype vtype, mode_t mode, uid_t uid, gid_t gid, fileid_t *fileidp,
763     int *loidp)
764 {
765 	static struct cmd *c;
766 	fileid_t new_fileid;
767 	int loid;
768 	int error;
769 
770 	error = mklinkfile(xc, parent_fileid, name, vtype, mode, uid, gid,
771 	    &new_fileid);
772 	if (error != 0) {
773 		return error;
774 	}
775 	CREATECMD(c,
776 		"INSERT INTO datafork (fileid, loid) "
777 		"VALUES($1, lo_creat(-1)) "
778 		"RETURNING loid", INT8OID);
779 	error = sendcmd(xc, c, new_fileid);
780 	if (error != 0) {
781 		return error;
782 	}
783 	error = simplefetch(xc, OIDOID, &loid);
784 	if (error != 0) {
785 		return error;
786 	}
787 	if (fileidp != NULL) {
788 		*fileidp = new_fileid;
789 	}
790 	if (loidp != NULL) {
791 		*loidp = loid;
792 	}
793 	return 0;
794 }
795 
796 int
cleanupfile(struct Xconn * xc,fileid_t fileid)797 cleanupfile(struct Xconn *xc, fileid_t fileid)
798 {
799 	static struct cmd *c;
800 	char *type;
801 	unsigned int vtype;
802 	int error;
803 
804 	CREATECMD(c, "DELETE FROM file WHERE fileid = $1 AND nlink = 0 "
805 		"RETURNING type::text", INT8OID);
806 	error = sendcmd(xc, c, fileid);
807 	if (error != 0) {
808 		return error;
809 	}
810 	error = simplefetch(xc, TEXTOID, &type);
811 	if (error == ENOENT) {
812 		return 0; /* probably nlink > 0 */
813 	}
814 	if (error != 0) {
815 		return error;
816 	}
817 	vtype = tovtype(type);
818 	free(type);
819 	if (vtype == VREG || vtype == VLNK) {
820 		static struct cmd *c_datafork;
821 		int32_t ret;
822 
823 		CREATECMD(c_datafork,
824 			"WITH loids AS (DELETE FROM datafork WHERE fileid = $1 "
825 			"RETURNING loid) SELECT lo_unlink(loid) FROM loids",
826 			INT8OID);
827 		error = sendcmd(xc, c_datafork, fileid);
828 		if (error != 0) {
829 			return error;
830 		}
831 		error = simplefetch(xc, INT4OID, &ret);
832 		if (error != 0) {
833 			return error;
834 		}
835 		if (ret != 1) {
836 			return EIO; /* lo_unlink failed */
837 		}
838 	}
839 	return 0;
840 }
841 
842 /*
843  * check_path: do locking and check to prevent a rename from creating loop.
844  *
845  * lock the dirents between child_fileid and the root directory.
846  * if gate_fileid is appeared in the path, return EINVAL.
847  * caller should ensure that child_fileid is of VDIR beforehand.
848  *
849  * we uses FOR SHARE row level locks as poor man's predicate locks.
850  *
851  * the following is an example to show why we need to lock the path.
852  *
853  * consider:
854  * "mkdir -p /a/b/c/d/e/f && mkdir -p /1/2/3/4/5/6"
855  * and then
856  * thread 1 is doing "mv /a/b /1/2/3/4/5/6"
857  * thread 2 is doing "mv /1/2 /a/b/c/d/e/f"
858  *
859  * a possible consequence:
860  *	thread 1: check_path -> success
861  *	thread 2: check_path -> success
862  *	thread 1: modify directories -> block on row-level lock
863  *	thread 2: modify directories -> block on row-level lock
864  *			-> deadlock detected
865  *			-> rollback and retry
866  *
867  * another possible consequence:
868  *	thread 1: check_path -> success
869  *	thread 1: modify directory entries -> success
870  *	thread 2: check_path -> block on row-level lock
871  *	thread 1: commit
872  *	thread 2: acquire the lock and notices the row is updated
873  *			-> serialization error
874  *			-> rollback and retry
875  *
876  * XXX it might be better to use real serializable transactions,
877  * which will be available for PostgreSQL 9.1
878  */
879 
880 int
check_path(struct Xconn * xc,fileid_t gate_fileid,fileid_t child_fileid)881 check_path(struct Xconn *xc, fileid_t gate_fileid, fileid_t child_fileid)
882 {
883 	static struct cmd *c;
884 	fileid_t parent_fileid;
885 	struct fetchstatus s;
886 	int error;
887 
888 	CREATECMD(c,
889 		"WITH RECURSIVE r AS "
890 		"( "
891 				"SELECT parent_fileid, cookie, child_fileid "
892 				"FROM dirent "
893 				"WHERE child_fileid = $1 "
894 			"UNION ALL "
895 				"SELECT d.parent_fileid, d.cookie, "
896 				"d.child_fileid "
897 				"FROM dirent AS d INNER JOIN r "
898 				"ON d.child_fileid = r.parent_fileid "
899 		") "
900 		"SELECT d.parent_fileid "
901 		"FROM dirent d "
902 		"JOIN r "
903 		"ON d.cookie = r.cookie "
904 		"FOR SHARE", INT8OID);
905 	error = sendcmd(xc, c, child_fileid);
906 	if (error != 0) {
907 		return error;
908 	}
909 	fetchinit(&s, xc);
910 	do {
911 		static const Oid types[] = { INT8OID, };
912 
913 		error = FETCHNEXT(&s, types, &parent_fileid);
914 		if (error == ENOENT) {
915 			fetchdone(&s);
916 			return 0;
917 		}
918 		if (error != 0) {
919 			fetchdone(&s);
920 			return error;
921 		}
922 	} while (gate_fileid != parent_fileid);
923 	fetchdone(&s);
924 	return EINVAL;
925 }
926 
927 int
isempty(struct Xconn * xc,fileid_t fileid,bool * emptyp)928 isempty(struct Xconn *xc, fileid_t fileid, bool *emptyp)
929 {
930 	int32_t dummy;
931 	static struct cmd *c;
932 	int error;
933 
934 	CREATECMD(c,
935 		"SELECT 1 FROM dirent "
936 		"WHERE parent_fileid = $1 LIMIT 1", INT8OID);
937 	error = sendcmd(xc, c, fileid);
938 	if (error != 0) {
939 		return error;
940 	}
941 	error = simplefetch(xc, INT4OID, &dummy);
942 	assert(error != 0 || dummy == 1);
943 	if (error == ENOENT) {
944 		*emptyp = true;
945 		error = 0;
946 	} else {
947 		*emptyp = false;
948 	}
949 	return error;
950 }
951