| 1 |
/*------------------------------------------------------------------------- |
| 2 |
* |
| 3 |
* be-fsstubs.c |
| 4 |
* Builtin functions for open/close/read/write operations on large objects |
| 5 |
* |
| 6 |
* Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group |
| 7 |
* Portions Copyright (c) 1994, Regents of the University of California |
| 8 |
* |
| 9 |
* |
| 10 |
* IDENTIFICATION |
| 11 |
* src/backend/libpq/be-fsstubs.c |
| 12 |
* |
| 13 |
* NOTES |
| 14 |
* This should be moved to a more appropriate place. It is here |
| 15 |
* for lack of a better place. |
| 16 |
* |
| 17 |
* These functions store LargeObjectDesc structs in a private MemoryContext, |
| 18 |
* which means that large object descriptors hang around until we destroy |
| 19 |
* the context at transaction end. It'd be possible to prolong the lifetime |
| 20 |
* of the context so that LO FDs are good across transactions (for example, |
| 21 |
* we could release the context only if we see that no FDs remain open). |
| 22 |
* But we'd need additional state in order to do the right thing at the |
| 23 |
* end of an aborted transaction. FDs opened during an aborted xact would |
| 24 |
* still need to be closed, since they might not be pointing at valid |
| 25 |
* relations at all. Locking semantics are also an interesting problem |
| 26 |
* if LOs stay open across transactions. For now, we'll stick with the |
| 27 |
* existing documented semantics of LO FDs: they're only good within a |
| 28 |
* transaction. |
| 29 |
* |
| 30 |
* As of PostgreSQL 8.0, much of the angst expressed above is no longer |
| 31 |
* relevant, and in fact it'd be pretty easy to allow LO FDs to stay |
| 32 |
* open across transactions. (Snapshot relevancy would still be an issue.) |
| 33 |
* However backwards compatibility suggests that we should stick to the |
| 34 |
* status quo. |
| 35 |
* |
| 36 |
*------------------------------------------------------------------------- |
| 37 |
*/ |
| 38 |
|
| 39 |
#include "postgres.h" |
| 40 |
|
| 41 |
#include <fcntl.h> |
| 42 |
#include <sys/stat.h> |
| 43 |
#include <unistd.h> |
| 44 |
|
| 45 |
#include "access/xact.h" |
| 46 |
#include "catalog/pg_largeobject.h" |
| 47 |
#include "libpq/be-fsstubs.h" |
| 48 |
#include "libpq/libpq-fs.h" |
| 49 |
#include "miscadmin.h" |
| 50 |
#include "storage/fd.h" |
| 51 |
#include "storage/large_object.h" |
| 52 |
#include "utils/acl.h" |
| 53 |
#include "utils/builtins.h" |
| 54 |
#include "utils/memutils.h" |
| 55 |
#include "utils/snapmgr.h" |
| 56 |
#include "varatt.h" |
| 57 |
|
| 58 |
/* define this to enable debug logging */ |
| 59 |
/* #define FSDB 1 */ |
| 60 |
/* chunk size for lo_import/lo_export transfers */ |
| 61 |
#define BUFSIZE 8192 |
| 62 |
|
| 63 |
/* |
| 64 |
* LO "FD"s are indexes into the cookies array. |
| 65 |
* |
| 66 |
* A non-null entry is a pointer to a LargeObjectDesc allocated in the |
| 67 |
* LO private memory context "fscxt". The cookies array itself is also |
| 68 |
* dynamically allocated in that context. Its current allocated size is |
| 69 |
* cookies_size entries, of which any unused entries will be NULL. |
| 70 |
*/ |
| 71 |
static LargeObjectDesc **cookies = NULL; |
| 72 |
static int cookies_size = 0; |
| 73 |
|
| 74 |
static bool lo_cleanup_needed = false; |
| 75 |
static MemoryContext fscxt = NULL; |
| 76 |
|
| 77 |
static int newLOfd(void); |
| 78 |
static void closeLOfd(int fd); |
| 79 |
static Oid lo_import_internal(text *filename, Oid lobjOid); |
| 80 |
|
| 81 |
|
| 82 |
/***************************************************************************** |
| 83 |
* File Interfaces for Large Objects |
| 84 |
*****************************************************************************/ |
| 85 |
|
| 86 |
Datum |
| 87 |
be_lo_open(PG_FUNCTION_ARGS) |
| 88 |
{ |
| 89 |
Oid lobjId = PG_GETARG_OID(0); |
| 90 |
int32 mode = PG_GETARG_INT32(1); |
| 91 |
LargeObjectDesc *lobjDesc; |
| 92 |
int fd; |
| 93 |
|
| 94 |
#ifdef FSDB |
| 95 |
elog(DEBUG4, "lo_open(%u,%d)", lobjId, mode); |
| 96 |
#endif |
| 97 |
|
| 98 |
if (mode & INV_WRITE) |
| 99 |
PreventCommandIfReadOnly("lo_open(INV_WRITE)"); |
| 100 |
|
| 101 |
/* |
| 102 |
* Allocate a large object descriptor first. This will also create |
| 103 |
* 'fscxt' if this is the first LO opened in this transaction. |
| 104 |
*/ |
| 105 |
fd = newLOfd(); |
| 106 |
|
| 107 |
lobjDesc = inv_open(lobjId, mode, fscxt); |
| 108 |
lobjDesc->subid = GetCurrentSubTransactionId(); |
| 109 |
|
| 110 |
/* |
| 111 |
* We must register the snapshot in TopTransaction's resowner so that it |
| 112 |
* stays alive until the LO is closed rather than until the current portal |
| 113 |
* shuts down. |
| 114 |
*/ |
| 115 |
if (lobjDesc->snapshot) |
| 116 |
lobjDesc->snapshot = RegisterSnapshotOnOwner(lobjDesc->snapshot, |
| 117 |
TopTransactionResourceOwner); |
| 118 |
|
| 119 |
Assert(cookies[fd] == NULL); |
| 120 |
cookies[fd] = lobjDesc; |
| 121 |
|
| 122 |
PG_RETURN_INT32(fd); |
| 123 |
} |
| 124 |
|
| 125 |
Datum |
| 126 |
be_lo_close(PG_FUNCTION_ARGS) |
| 127 |
{ |
| 128 |
int32 fd = PG_GETARG_INT32(0); |
| 129 |
|
| 130 |
if (fd < 0 || fd >= cookies_size || cookies[fd] == NULL) |
| 131 |
ereport(ERROR, |
| 132 |
(errcode(ERRCODE_UNDEFINED_OBJECT), |
| 133 |
errmsg("invalid large-object descriptor: %d", fd))); |
| 134 |
|
| 135 |
#ifdef FSDB |
| 136 |
elog(DEBUG4, "lo_close(%d)", fd); |
| 137 |
#endif |
| 138 |
|
| 139 |
closeLOfd(fd); |
| 140 |
|
| 141 |
PG_RETURN_INT32(0); |
| 142 |
} |
| 143 |
|
| 144 |
|
| 145 |
/***************************************************************************** |
| 146 |
* Bare Read/Write operations --- these are not fmgr-callable! |
| 147 |
* |
| 148 |
* We assume the large object supports byte oriented reads and seeks so |
| 149 |
* that our work is easier. |
| 150 |
* |
| 151 |
*****************************************************************************/ |
| 152 |
|
| 153 |
int |
| 154 |
lo_read(int fd, char *buf, int len) |
| 155 |
{ |
| 156 |
int status; |
| 157 |
LargeObjectDesc *lobj; |
| 158 |
|
| 159 |
if (fd < 0 || fd >= cookies_size || cookies[fd] == NULL) |
| 160 |
ereport(ERROR, |
| 161 |
(errcode(ERRCODE_UNDEFINED_OBJECT), |
| 162 |
errmsg("invalid large-object descriptor: %d", fd))); |
| 163 |
lobj = cookies[fd]; |
| 164 |
|
| 165 |
/* |
| 166 |
* Check state. inv_read() would throw an error anyway, but we want the |
| 167 |
* error to be about the FD's state not the underlying privilege; it might |
| 168 |
* be that the privilege exists but user forgot to ask for read mode. |
| 169 |
*/ |
| 170 |
if ((lobj->flags & IFS_RDLOCK) == 0) |
| 171 |
ereport(ERROR, |
| 172 |
(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), |
| 173 |
errmsg("large object descriptor %d was not opened for reading", |
| 174 |
fd))); |
| 175 |
|
| 176 |
status = inv_read(lobj, buf, len); |
| 177 |
|
| 178 |
return status; |
| 179 |
} |
| 180 |
|
| 181 |
int |
| 182 |
lo_write(int fd, const char *buf, int len) |
| 183 |
{ |
| 184 |
int status; |
| 185 |
LargeObjectDesc *lobj; |
| 186 |
|
| 187 |
if (fd < 0 || fd >= cookies_size || cookies[fd] == NULL) |
| 188 |
ereport(ERROR, |
| 189 |
(errcode(ERRCODE_UNDEFINED_OBJECT), |
| 190 |
errmsg("invalid large-object descriptor: %d", fd))); |
| 191 |
lobj = cookies[fd]; |
| 192 |
|
| 193 |
/* see comment in lo_read() */ |
| 194 |
if ((lobj->flags & IFS_WRLOCK) == 0) |
| 195 |
ereport(ERROR, |
| 196 |
(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), |
| 197 |
errmsg("large object descriptor %d was not opened for writing", |
| 198 |
fd))); |
| 199 |
|
| 200 |
status = inv_write(lobj, buf, len); |
| 201 |
|
| 202 |
return status; |
| 203 |
} |
| 204 |
|
| 205 |
Datum |
| 206 |
be_lo_lseek(PG_FUNCTION_ARGS) |
| 207 |
{ |
| 208 |
int32 fd = PG_GETARG_INT32(0); |
| 209 |
int32 offset = PG_GETARG_INT32(1); |
| 210 |
int32 whence = PG_GETARG_INT32(2); |
| 211 |
int64 status; |
| 212 |
|
| 213 |
if (fd < 0 || fd >= cookies_size || cookies[fd] == NULL) |
| 214 |
ereport(ERROR, |
| 215 |
(errcode(ERRCODE_UNDEFINED_OBJECT), |
| 216 |
errmsg("invalid large-object descriptor: %d", fd))); |
| 217 |
|
| 218 |
status = inv_seek(cookies[fd], offset, whence); |
| 219 |
|
| 220 |
/* guard against result overflow */ |
| 221 |
if (status != (int32) status) |
| 222 |
ereport(ERROR, |
| 223 |
(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE), |
| 224 |
errmsg("lo_lseek result out of range for large-object descriptor %d", |
| 225 |
fd))); |
| 226 |
|
| 227 |
PG_RETURN_INT32((int32) status); |
| 228 |
} |
| 229 |
|
| 230 |
Datum |
| 231 |
be_lo_lseek64(PG_FUNCTION_ARGS) |
| 232 |
{ |
| 233 |
int32 fd = PG_GETARG_INT32(0); |
| 234 |
int64 offset = PG_GETARG_INT64(1); |
| 235 |
int32 whence = PG_GETARG_INT32(2); |
| 236 |
int64 status; |
| 237 |
|
| 238 |
if (fd < 0 || fd >= cookies_size || cookies[fd] == NULL) |
| 239 |
ereport(ERROR, |
| 240 |
(errcode(ERRCODE_UNDEFINED_OBJECT), |
| 241 |
errmsg("invalid large-object descriptor: %d", fd))); |
| 242 |
|
| 243 |
status = inv_seek(cookies[fd], offset, whence); |
| 244 |
|
| 245 |
PG_RETURN_INT64(status); |
| 246 |
} |
| 247 |
|
| 248 |
Datum |
| 249 |
be_lo_creat(PG_FUNCTION_ARGS) |
| 250 |
{ |
| 251 |
Oid lobjId; |
| 252 |
|
| 253 |
PreventCommandIfReadOnly("lo_creat()"); |
| 254 |
|
| 255 |
lo_cleanup_needed = true; |
| 256 |
lobjId = inv_create(InvalidOid); |
| 257 |
|
| 258 |
PG_RETURN_OID(lobjId); |
| 259 |
} |
| 260 |
|
| 261 |
Datum |
| 262 |
be_lo_create(PG_FUNCTION_ARGS) |
| 263 |
{ |
| 264 |
Oid lobjId = PG_GETARG_OID(0); |
| 265 |
|
| 266 |
PreventCommandIfReadOnly("lo_create()"); |
| 267 |
|
| 268 |
lo_cleanup_needed = true; |
| 269 |
lobjId = inv_create(lobjId); |
| 270 |
|
| 271 |
PG_RETURN_OID(lobjId); |
| 272 |
} |
| 273 |
|
| 274 |
Datum |
| 275 |
be_lo_tell(PG_FUNCTION_ARGS) |
| 276 |
{ |
| 277 |
int32 fd = PG_GETARG_INT32(0); |
| 278 |
int64 offset; |
| 279 |
|
| 280 |
if (fd < 0 || fd >= cookies_size || cookies[fd] == NULL) |
| 281 |
ereport(ERROR, |
| 282 |
(errcode(ERRCODE_UNDEFINED_OBJECT), |
| 283 |
errmsg("invalid large-object descriptor: %d", fd))); |
| 284 |
|
| 285 |
offset = inv_tell(cookies[fd]); |
| 286 |
|
| 287 |
/* guard against result overflow */ |
| 288 |
if (offset != (int32) offset) |
| 289 |
ereport(ERROR, |
| 290 |
(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE), |
| 291 |
errmsg("lo_tell result out of range for large-object descriptor %d", |
| 292 |
fd))); |
| 293 |
|
| 294 |
PG_RETURN_INT32((int32) offset); |
| 295 |
} |
| 296 |
|
| 297 |
Datum |
| 298 |
be_lo_tell64(PG_FUNCTION_ARGS) |
| 299 |
{ |
| 300 |
int32 fd = PG_GETARG_INT32(0); |
| 301 |
int64 offset; |
| 302 |
|
| 303 |
if (fd < 0 || fd >= cookies_size || cookies[fd] == NULL) |
| 304 |
ereport(ERROR, |
| 305 |
(errcode(ERRCODE_UNDEFINED_OBJECT), |
| 306 |
errmsg("invalid large-object descriptor: %d", fd))); |
| 307 |
|
| 308 |
offset = inv_tell(cookies[fd]); |
| 309 |
|
| 310 |
PG_RETURN_INT64(offset); |
| 311 |
} |
| 312 |
|
| 313 |
Datum |
| 314 |
be_lo_unlink(PG_FUNCTION_ARGS) |
| 315 |
{ |
| 316 |
Oid lobjId = PG_GETARG_OID(0); |
| 317 |
|
| 318 |
PreventCommandIfReadOnly("lo_unlink()"); |
| 319 |
|
| 320 |
if (!LargeObjectExists(lobjId)) |
| 321 |
ereport(ERROR, |
| 322 |
(errcode(ERRCODE_UNDEFINED_OBJECT), |
| 323 |
errmsg("large object %u does not exist", lobjId))); |
| 324 |
|
| 325 |
/* |
| 326 |
* Must be owner of the large object. It would be cleaner to check this |
| 327 |
* in inv_drop(), but we want to throw the error before not after closing |
| 328 |
* relevant FDs. |
| 329 |
*/ |
| 330 |
if (!lo_compat_privileges && |
| 331 |
!object_ownercheck(LargeObjectRelationId, lobjId, GetUserId())) |
| 332 |
ereport(ERROR, |
| 333 |
(errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), |
| 334 |
errmsg("must be owner of large object %u", lobjId))); |
| 335 |
|
| 336 |
/* |
| 337 |
* If there are any open LO FDs referencing that ID, close 'em. |
| 338 |
*/ |
| 339 |
if (fscxt != NULL) |
| 340 |
{ |
| 341 |
int i; |
| 342 |
|
| 343 |
for (i = 0; i < cookies_size; i++) |
| 344 |
{ |
| 345 |
if (cookies[i] != NULL && cookies[i]->id == lobjId) |
| 346 |
closeLOfd(i); |
| 347 |
} |
| 348 |
} |
| 349 |
|
| 350 |
/* |
| 351 |
* inv_drop does not create a need for end-of-transaction cleanup and |
| 352 |
* hence we don't need to set lo_cleanup_needed. |
| 353 |
*/ |
| 354 |
PG_RETURN_INT32(inv_drop(lobjId)); |
| 355 |
} |
| 356 |
|
| 357 |
/***************************************************************************** |
| 358 |
* Read/Write using bytea |
| 359 |
*****************************************************************************/ |
| 360 |
|
| 361 |
Datum |
| 362 |
be_loread(PG_FUNCTION_ARGS) |
| 363 |
{ |
| 364 |
int32 fd = PG_GETARG_INT32(0); |
| 365 |
int32 len = PG_GETARG_INT32(1); |
| 366 |
bytea *retval; |
| 367 |
int totalread; |
| 368 |
|
| 369 |
if (len < 0) |
| 370 |
len = 0; |
| 371 |
|
| 372 |
retval = (bytea *) palloc(VARHDRSZ + len); |
| 373 |
totalread = lo_read(fd, VARDATA(retval), len); |
| 374 |
SET_VARSIZE(retval, totalread + VARHDRSZ); |
| 375 |
|
| 376 |
PG_RETURN_BYTEA_P(retval); |
| 377 |
} |
| 378 |
|
| 379 |
Datum |
| 380 |
be_lowrite(PG_FUNCTION_ARGS) |
| 381 |
{ |
| 382 |
int32 fd = PG_GETARG_INT32(0); |
| 383 |
bytea *wbuf = PG_GETARG_BYTEA_PP(1); |
| 384 |
int bytestowrite; |
| 385 |
int totalwritten; |
| 386 |
|
| 387 |
PreventCommandIfReadOnly("lowrite()"); |
| 388 |
|
| 389 |
bytestowrite = VARSIZE_ANY_EXHDR(wbuf); |
| 390 |
totalwritten = lo_write(fd, VARDATA_ANY(wbuf), bytestowrite); |
| 391 |
PG_RETURN_INT32(totalwritten); |
| 392 |
} |
| 393 |
|
| 394 |
/***************************************************************************** |
| 395 |
* Import/Export of Large Object |
| 396 |
*****************************************************************************/ |
| 397 |
|
| 398 |
/* |
| 399 |
* lo_import - |
| 400 |
* imports a file as an (inversion) large object. |
| 401 |
*/ |
| 402 |
Datum |
| 403 |
be_lo_import(PG_FUNCTION_ARGS) |
| 404 |
{ |
| 405 |
text *filename = PG_GETARG_TEXT_PP(0); |
| 406 |
|
| 407 |
PG_RETURN_OID(lo_import_internal(filename, InvalidOid)); |
| 408 |
} |
| 409 |
|
| 410 |
/* |
| 411 |
* lo_import_with_oid - |
| 412 |
* imports a file as an (inversion) large object specifying oid. |
| 413 |
*/ |
| 414 |
Datum |
| 415 |
be_lo_import_with_oid(PG_FUNCTION_ARGS) |
| 416 |
{ |
| 417 |
text *filename = PG_GETARG_TEXT_PP(0); |
| 418 |
Oid oid = PG_GETARG_OID(1); |
| 419 |
|
| 420 |
PG_RETURN_OID(lo_import_internal(filename, oid)); |
| 421 |
} |
| 422 |
|
| 423 |
static Oid |
| 424 |
lo_import_internal(text *filename, Oid lobjOid) |
| 425 |
{ |
| 426 |
int fd; |
| 427 |
int nbytes, |
| 428 |
tmp PG_USED_FOR_ASSERTS_ONLY; |
| 429 |
char buf[BUFSIZE]; |
| 430 |
char fnamebuf[MAXPGPATH]; |
| 431 |
LargeObjectDesc *lobj; |
| 432 |
Oid oid; |
| 433 |
|
| 434 |
PreventCommandIfReadOnly("lo_import()"); |
| 435 |
|
| 436 |
/* |
| 437 |
* open the file to be read in |
| 438 |
*/ |
| 439 |
text_to_cstring_buffer(filename, fnamebuf, sizeof(fnamebuf)); |
| 440 |
fd = OpenTransientFile(fnamebuf, O_RDONLY | PG_BINARY); |
| 441 |
if (fd < 0) |
| 442 |
ereport(ERROR, |
| 443 |
(errcode_for_file_access(), |
| 444 |
errmsg("could not open server file \"%s\": %m", |
| 445 |
fnamebuf))); |
| 446 |
|
| 447 |
/* |
| 448 |
* create an inversion object |
| 449 |
*/ |
| 450 |
lo_cleanup_needed = true; |
| 451 |
oid = inv_create(lobjOid); |
| 452 |
|
| 453 |
/* |
| 454 |
* read in from the filesystem and write to the inversion object |
| 455 |
*/ |
| 456 |
lobj = inv_open(oid, INV_WRITE, CurrentMemoryContext); |
| 457 |
|
| 458 |
while ((nbytes = read(fd, buf, BUFSIZE)) > 0) |
| 459 |
{ |
| 460 |
tmp = inv_write(lobj, buf, nbytes); |
| 461 |
Assert(tmp == nbytes); |
| 462 |
} |
| 463 |
|
| 464 |
if (nbytes < 0) |
| 465 |
ereport(ERROR, |
| 466 |
(errcode_for_file_access(), |
| 467 |
errmsg("could not read server file \"%s\": %m", |
| 468 |
fnamebuf))); |
| 469 |
|
| 470 |
inv_close(lobj); |
| 471 |
|
| 472 |
if (CloseTransientFile(fd) != 0) |
| 473 |
ereport(ERROR, |
| 474 |
(errcode_for_file_access(), |
| 475 |
errmsg("could not close file \"%s\": %m", |
| 476 |
fnamebuf))); |
| 477 |
|
| 478 |
return oid; |
| 479 |
} |
| 480 |
|
| 481 |
/* |
| 482 |
* lo_export - |
| 483 |
* exports an (inversion) large object. |
| 484 |
*/ |
| 485 |
Datum |
| 486 |
be_lo_export(PG_FUNCTION_ARGS) |
| 487 |
{ |
| 488 |
Oid lobjId = PG_GETARG_OID(0); |
| 489 |
text *filename = PG_GETARG_TEXT_PP(1); |
| 490 |
int fd; |
| 491 |
int nbytes, |
| 492 |
tmp; |
| 493 |
char buf[BUFSIZE]; |
| 494 |
char fnamebuf[MAXPGPATH]; |
| 495 |
LargeObjectDesc *lobj; |
| 496 |
mode_t oumask; |
| 497 |
|
| 498 |
/* |
| 499 |
* open the inversion object (no need to test for failure) |
| 500 |
*/ |
| 501 |
lo_cleanup_needed = true; |
| 502 |
lobj = inv_open(lobjId, INV_READ, CurrentMemoryContext); |
| 503 |
|
| 504 |
/* |
| 505 |
* open the file to be written to |
| 506 |
* |
| 507 |
* Note: we reduce backend's normal 077 umask to the slightly friendlier |
| 508 |
* 022. This code used to drop it all the way to 0, but creating |
| 509 |
* world-writable export files doesn't seem wise. |
| 510 |
*/ |
| 511 |
text_to_cstring_buffer(filename, fnamebuf, sizeof(fnamebuf)); |
| 512 |
oumask = umask(S_IWGRP | S_IWOTH); |
| 513 |
PG_TRY(); |
| 514 |
{ |
| 515 |
fd = OpenTransientFilePerm(fnamebuf, O_CREAT | O_WRONLY | O_TRUNC | PG_BINARY, |
| 516 |
S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH); |
| 517 |
} |
| 518 |
PG_FINALLY(); |
| 519 |
{ |
| 520 |
umask(oumask); |
| 521 |
} |
| 522 |
PG_END_TRY(); |
| 523 |
if (fd < 0) |
| 524 |
ereport(ERROR, |
| 525 |
(errcode_for_file_access(), |
| 526 |
errmsg("could not create server file \"%s\": %m", |
| 527 |
fnamebuf))); |
| 528 |
|
| 529 |
/* |
| 530 |
* read in from the inversion file and write to the filesystem |
| 531 |
*/ |
| 532 |
while ((nbytes = inv_read(lobj, buf, BUFSIZE)) > 0) |
| 533 |
{ |
| 534 |
tmp = write(fd, buf, nbytes); |
| 535 |
if (tmp != nbytes) |
| 536 |
ereport(ERROR, |
| 537 |
(errcode_for_file_access(), |
| 538 |
errmsg("could not write server file \"%s\": %m", |
| 539 |
fnamebuf))); |
| 540 |
} |
| 541 |
|
| 542 |
if (CloseTransientFile(fd) != 0) |
| 543 |
ereport(ERROR, |
| 544 |
(errcode_for_file_access(), |
| 545 |
errmsg("could not close file \"%s\": %m", |
| 546 |
fnamebuf))); |
| 547 |
|
| 548 |
inv_close(lobj); |
| 549 |
|
| 550 |
PG_RETURN_INT32(1); |
| 551 |
} |
| 552 |
|
| 553 |
/* |
| 554 |
* lo_truncate - |
| 555 |
* truncate a large object to a specified length |
| 556 |
*/ |
| 557 |
static void |
| 558 |
lo_truncate_internal(int32 fd, int64 len) |
| 559 |
{ |
| 560 |
LargeObjectDesc *lobj; |
| 561 |
|
| 562 |
if (fd < 0 || fd >= cookies_size || cookies[fd] == NULL) |
| 563 |
ereport(ERROR, |
| 564 |
(errcode(ERRCODE_UNDEFINED_OBJECT), |
| 565 |
errmsg("invalid large-object descriptor: %d", fd))); |
| 566 |
lobj = cookies[fd]; |
| 567 |
|
| 568 |
/* see comment in lo_read() */ |
| 569 |
if ((lobj->flags & IFS_WRLOCK) == 0) |
| 570 |
ereport(ERROR, |
| 571 |
(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), |
| 572 |
errmsg("large object descriptor %d was not opened for writing", |
| 573 |
fd))); |
| 574 |
|
| 575 |
inv_truncate(lobj, len); |
| 576 |
} |
| 577 |
|
| 578 |
Datum |
| 579 |
be_lo_truncate(PG_FUNCTION_ARGS) |
| 580 |
{ |
| 581 |
int32 fd = PG_GETARG_INT32(0); |
| 582 |
int32 len = PG_GETARG_INT32(1); |
| 583 |
|
| 584 |
PreventCommandIfReadOnly("lo_truncate()"); |
| 585 |
|
| 586 |
lo_truncate_internal(fd, len); |
| 587 |
PG_RETURN_INT32(0); |
| 588 |
} |
| 589 |
|
| 590 |
Datum |
| 591 |
be_lo_truncate64(PG_FUNCTION_ARGS) |
| 592 |
{ |
| 593 |
int32 fd = PG_GETARG_INT32(0); |
| 594 |
int64 len = PG_GETARG_INT64(1); |
| 595 |
|
| 596 |
PreventCommandIfReadOnly("lo_truncate64()"); |
| 597 |
|
| 598 |
lo_truncate_internal(fd, len); |
| 599 |
PG_RETURN_INT32(0); |
| 600 |
} |
| 601 |
|
| 602 |
/* |
| 603 |
* AtEOXact_LargeObject - |
| 604 |
* prepares large objects for transaction commit |
| 605 |
*/ |
| 606 |
void |
| 607 |
AtEOXact_LargeObject(bool isCommit) |
| 608 |
{ |
| 609 |
int i; |
| 610 |
|
| 611 |
if (!lo_cleanup_needed) |
| 612 |
return; /* no LO operations in this xact */ |
| 613 |
|
| 614 |
/* |
| 615 |
* Close LO fds and clear cookies array so that LO fds are no longer good. |
| 616 |
* The memory context and resource owner holding them are going away at |
| 617 |
* the end-of-transaction anyway, but on commit, we need to close them to |
| 618 |
* avoid warnings about leaked resources at commit. On abort we can skip |
| 619 |
* this step. |
| 620 |
*/ |
| 621 |
if (isCommit) |
| 622 |
{ |
| 623 |
for (i = 0; i < cookies_size; i++) |
| 624 |
{ |
| 625 |
if (cookies[i] != NULL) |
| 626 |
closeLOfd(i); |
| 627 |
} |
| 628 |
} |
| 629 |
|
| 630 |
/* Needn't actually pfree since we're about to zap context */ |
| 631 |
cookies = NULL; |
| 632 |
cookies_size = 0; |
| 633 |
|
| 634 |
/* Release the LO memory context to prevent permanent memory leaks. */ |
| 635 |
if (fscxt) |
| 636 |
MemoryContextDelete(fscxt); |
| 637 |
fscxt = NULL; |
| 638 |
|
| 639 |
/* Give inv_api.c a chance to clean up, too */ |
| 640 |
close_lo_relation(isCommit); |
| 641 |
|
| 642 |
lo_cleanup_needed = false; |
| 643 |
} |
| 644 |
|
| 645 |
/* |
| 646 |
* AtEOSubXact_LargeObject |
| 647 |
* Take care of large objects at subtransaction commit/abort |
| 648 |
* |
| 649 |
* Reassign LOs created/opened during a committing subtransaction |
| 650 |
* to the parent subtransaction. On abort, just close them. |
| 651 |
*/ |
| 652 |
void |
| 653 |
AtEOSubXact_LargeObject(bool isCommit, SubTransactionId mySubid, |
| 654 |
SubTransactionId parentSubid) |
| 655 |
{ |
| 656 |
int i; |
| 657 |
|
| 658 |
if (fscxt == NULL) /* no LO operations in this xact */ |
| 659 |
return; |
| 660 |
|
| 661 |
for (i = 0; i < cookies_size; i++) |
| 662 |
{ |
| 663 |
LargeObjectDesc *lo = cookies[i]; |
| 664 |
|
| 665 |
if (lo != NULL && lo->subid == mySubid) |
| 666 |
{ |
| 667 |
if (isCommit) |
| 668 |
lo->subid = parentSubid; |
| 669 |
else |
| 670 |
closeLOfd(i); |
| 671 |
} |
| 672 |
} |
| 673 |
} |
| 674 |
|
| 675 |
/***************************************************************************** |
| 676 |
* Support routines for this file |
| 677 |
*****************************************************************************/ |
| 678 |
|
| 679 |
static int |
| 680 |
newLOfd(void) |
| 681 |
{ |
| 682 |
int i, |
| 683 |
newsize; |
| 684 |
|
| 685 |
lo_cleanup_needed = true; |
| 686 |
if (fscxt == NULL) |
| 687 |
fscxt = AllocSetContextCreate(TopMemoryContext, |
| 688 |
"Filesystem", |
| 689 |
ALLOCSET_DEFAULT_SIZES); |
| 690 |
|
| 691 |
/* Try to find a free slot */ |
| 692 |
for (i = 0; i < cookies_size; i++) |
| 693 |
{ |
| 694 |
if (cookies[i] == NULL) |
| 695 |
return i; |
| 696 |
} |
| 697 |
|
| 698 |
/* No free slot, so make the array bigger */ |
| 699 |
if (cookies_size <= 0) |
| 700 |
{ |
| 701 |
/* First time through, arbitrarily make 64-element array */ |
| 702 |
i = 0; |
| 703 |
newsize = 64; |
| 704 |
cookies = (LargeObjectDesc **) |
| 705 |
MemoryContextAllocZero(fscxt, newsize * sizeof(LargeObjectDesc *)); |
| 706 |
} |
| 707 |
else |
| 708 |
{ |
| 709 |
/* Double size of array */ |
| 710 |
i = cookies_size; |
| 711 |
newsize = cookies_size * 2; |
| 712 |
cookies = |
| 713 |
repalloc0_array(cookies, LargeObjectDesc *, cookies_size, newsize); |
| 714 |
} |
| 715 |
cookies_size = newsize; |
| 716 |
|
| 717 |
return i; |
| 718 |
} |
| 719 |
|
| 720 |
static void |
| 721 |
closeLOfd(int fd) |
| 722 |
{ |
| 723 |
LargeObjectDesc *lobj; |
| 724 |
|
| 725 |
/* |
| 726 |
* Make sure we do not try to free twice if this errors out for some |
| 727 |
* reason. Better a leak than a crash. |
| 728 |
*/ |
| 729 |
lobj = cookies[fd]; |
| 730 |
cookies[fd] = NULL; |
| 731 |
|
| 732 |
if (lobj->snapshot) |
| 733 |
UnregisterSnapshotFromOwner(lobj->snapshot, |
| 734 |
TopTransactionResourceOwner); |
| 735 |
inv_close(lobj); |
| 736 |
} |
| 737 |
|
| 738 |
/***************************************************************************** |
| 739 |
* Wrappers oriented toward SQL callers |
| 740 |
*****************************************************************************/ |
| 741 |
|
| 742 |
/* |
| 743 |
* Read [offset, offset+nbytes) within LO; when nbytes is -1, read to end. |
| 744 |
*/ |
| 745 |
static bytea * |
| 746 |
lo_get_fragment_internal(Oid loOid, int64 offset, int32 nbytes) |
| 747 |
{ |
| 748 |
LargeObjectDesc *loDesc; |
| 749 |
int64 loSize; |
| 750 |
int64 result_length; |
| 751 |
int total_read PG_USED_FOR_ASSERTS_ONLY; |
| 752 |
bytea *result = NULL; |
| 753 |
|
| 754 |
lo_cleanup_needed = true; |
| 755 |
loDesc = inv_open(loOid, INV_READ, CurrentMemoryContext); |
| 756 |
|
| 757 |
/* |
| 758 |
* Compute number of bytes we'll actually read, accommodating nbytes == -1 |
| 759 |
* and reads beyond the end of the LO. |
| 760 |
*/ |
| 761 |
loSize = inv_seek(loDesc, 0, SEEK_END); |
| 762 |
if (loSize > offset) |
| 763 |
{ |
| 764 |
if (nbytes >= 0 && nbytes <= loSize - offset) |
| 765 |
result_length = nbytes; /* request is wholly inside LO */ |
| 766 |
else |
| 767 |
result_length = loSize - offset; /* adjust to end of LO */ |
| 768 |
} |
| 769 |
else |
| 770 |
result_length = 0; /* request is wholly outside LO */ |
| 771 |
|
| 772 |
/* |
| 773 |
* A result_length calculated from loSize may not fit in a size_t. Check |
| 774 |
* that the size will satisfy this and subsequently-enforced size limits. |
| 775 |
*/ |
| 776 |
if (result_length > MaxAllocSize - VARHDRSZ) |
| 777 |
ereport(ERROR, |
| 778 |
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED), |
| 779 |
errmsg("large object read request is too large"))); |
| 780 |
|
| 781 |
result = (bytea *) palloc(VARHDRSZ + result_length); |
| 782 |
|
| 783 |
inv_seek(loDesc, offset, SEEK_SET); |
| 784 |
total_read = inv_read(loDesc, VARDATA(result), result_length); |
| 785 |
Assert(total_read == result_length); |
| 786 |
SET_VARSIZE(result, result_length + VARHDRSZ); |
| 787 |
|
| 788 |
inv_close(loDesc); |
| 789 |
|
| 790 |
return result; |
| 791 |
} |
| 792 |
|
| 793 |
/* |
| 794 |
* Read entire LO |
| 795 |
*/ |
| 796 |
Datum |
| 797 |
be_lo_get(PG_FUNCTION_ARGS) |
| 798 |
{ |
| 799 |
Oid loOid = PG_GETARG_OID(0); |
| 800 |
bytea *result; |
| 801 |
|
| 802 |
result = lo_get_fragment_internal(loOid, 0, -1); |
| 803 |
|
| 804 |
PG_RETURN_BYTEA_P(result); |
| 805 |
} |
| 806 |
|
| 807 |
/* |
| 808 |
* Read range within LO |
| 809 |
*/ |
| 810 |
Datum |
| 811 |
be_lo_get_fragment(PG_FUNCTION_ARGS) |
| 812 |
{ |
| 813 |
Oid loOid = PG_GETARG_OID(0); |
| 814 |
int64 offset = PG_GETARG_INT64(1); |
| 815 |
int32 nbytes = PG_GETARG_INT32(2); |
| 816 |
bytea *result; |
| 817 |
|
| 818 |
if (nbytes < 0) |
| 819 |
ereport(ERROR, |
| 820 |
(errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
| 821 |
errmsg("requested length cannot be negative"))); |
| 822 |
|
| 823 |
result = lo_get_fragment_internal(loOid, offset, nbytes); |
| 824 |
|
| 825 |
PG_RETURN_BYTEA_P(result); |
| 826 |
} |
| 827 |
|
| 828 |
/* |
| 829 |
* Create LO with initial contents given by a bytea argument |
| 830 |
*/ |
| 831 |
Datum |
| 832 |
be_lo_from_bytea(PG_FUNCTION_ARGS) |
| 833 |
{ |
| 834 |
Oid loOid = PG_GETARG_OID(0); |
| 835 |
bytea *str = PG_GETARG_BYTEA_PP(1); |
| 836 |
LargeObjectDesc *loDesc; |
| 837 |
int written PG_USED_FOR_ASSERTS_ONLY; |
| 838 |
|
| 839 |
PreventCommandIfReadOnly("lo_from_bytea()"); |
| 840 |
|
| 841 |
lo_cleanup_needed = true; |
| 842 |
loOid = inv_create(loOid); |
| 843 |
loDesc = inv_open(loOid, INV_WRITE, CurrentMemoryContext); |
| 844 |
written = inv_write(loDesc, VARDATA_ANY(str), VARSIZE_ANY_EXHDR(str)); |
| 845 |
Assert(written == VARSIZE_ANY_EXHDR(str)); |
| 846 |
inv_close(loDesc); |
| 847 |
|
| 848 |
PG_RETURN_OID(loOid); |
| 849 |
} |
| 850 |
|
| 851 |
/* |
| 852 |
* Update range within LO |
| 853 |
*/ |
| 854 |
Datum |
| 855 |
be_lo_put(PG_FUNCTION_ARGS) |
| 856 |
{ |
| 857 |
Oid loOid = PG_GETARG_OID(0); |
| 858 |
int64 offset = PG_GETARG_INT64(1); |
| 859 |
bytea *str = PG_GETARG_BYTEA_PP(2); |
| 860 |
LargeObjectDesc *loDesc; |
| 861 |
int written PG_USED_FOR_ASSERTS_ONLY; |
| 862 |
|
| 863 |
PreventCommandIfReadOnly("lo_put()"); |
| 864 |
|
| 865 |
lo_cleanup_needed = true; |
| 866 |
loDesc = inv_open(loOid, INV_WRITE, CurrentMemoryContext); |
| 867 |
inv_seek(loDesc, offset, SEEK_SET); |
| 868 |
written = inv_write(loDesc, VARDATA_ANY(str), VARSIZE_ANY_EXHDR(str)); |
| 869 |
Assert(written == VARSIZE_ANY_EXHDR(str)); |
| 870 |
inv_close(loDesc); |
| 871 |
|
| 872 |
PG_RETURN_VOID(); |
| 873 |
} |