1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2019 The FreeBSD Foundation 5 * 6 * This software was developed by BFF Storage Systems, LLC under sponsorship 7 * from the FreeBSD Foundation. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28 * SUCH DAMAGE. 29 * 30 * $FreeBSD$ 31 */ 32 33 extern "C" { 34 #include <sys/types.h> 35 #include <sys/mman.h> 36 #include <sys/sysctl.h> 37 38 #include <fcntl.h> 39 #include <stdlib.h> 40 #include <unistd.h> 41 } 42 43 #include "mockfs.hh" 44 #include "utils.hh" 45 46 /* 47 * For testing I/O like fsx does, but deterministically and without a real 48 * underlying file system 49 */ 50 51 using namespace testing; 52 53 const char FULLPATH[] = "mountpoint/some_file.txt"; 54 const char RELPATH[] = "some_file.txt"; 55 const uint64_t ino = 42; 56 57 static void compare(const void *tbuf, const void *controlbuf, off_t baseofs, 58 ssize_t size) 59 { 60 int i; 61 62 for (i = 0; i < size; i++) { 63 if (((const char*)tbuf)[i] != ((const char*)controlbuf)[i]) { 64 off_t ofs = baseofs + i; 65 FAIL() << "miscompare at offset " 66 << std::hex 67 << std::showbase 68 << ofs 69 << ". expected = " 70 << std::setw(2) 71 << (unsigned)((const uint8_t*)controlbuf)[i] 72 << " got = " 73 << (unsigned)((const uint8_t*)tbuf)[i]; 74 } 75 } 76 } 77 78 typedef tuple<bool, uint32_t, cache_mode> IoParam; 79 80 class Io: public FuseTest, public WithParamInterface<IoParam> { 81 public: 82 int m_backing_fd, m_control_fd, m_test_fd; 83 off_t m_filesize; 84 bool m_direct_io; 85 86 Io(): m_backing_fd(-1), m_control_fd(-1), m_test_fd(-1), m_filesize(0), 87 m_direct_io(false) {}; 88 89 void SetUp() 90 { 91 m_backing_fd = open("backing_file", O_RDWR | O_CREAT | O_TRUNC, 0644); 92 if (m_backing_fd < 0) 93 FAIL() << strerror(errno); 94 m_control_fd = open("control", O_RDWR | O_CREAT | O_TRUNC, 0644); 95 if (m_control_fd < 0) 96 FAIL() << strerror(errno); 97 srandom(22'9'1982); // Seed with my birthday 98 99 if (get<0>(GetParam())) 100 m_init_flags |= FUSE_ASYNC_READ; 101 m_maxwrite = get<1>(GetParam()); 102 switch (get<2>(GetParam())) { 103 case Uncached: 104 m_direct_io = true; 105 break; 106 case WritebackAsync: 107 m_async = true; 108 /* FALLTHROUGH */ 109 case Writeback: 110 m_init_flags |= FUSE_WRITEBACK_CACHE; 111 /* FALLTHROUGH */ 112 case Writethrough: 113 break; 114 default: 115 FAIL() << "Unknown cache mode"; 116 } 117 m_noatime = true; // To prevent SETATTR for atime on close 118 119 FuseTest::SetUp(); 120 if (IsSkipped()) 121 return; 122 123 if (verbosity > 0) { 124 printf("Test Parameters: init_flags=%#x maxwrite=%#x " 125 "%sasync cache=%s\n", 126 m_init_flags, m_maxwrite, m_async? "" : "no", 127 cache_mode_to_s(get<2>(GetParam()))); 128 } 129 130 expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); 131 expect_open(ino, m_direct_io ? FOPEN_DIRECT_IO : 0, 1); 132 EXPECT_CALL(*m_mock, process( 133 ResultOf([=](auto in) { 134 return (in.header.opcode == FUSE_WRITE && 135 in.header.nodeid == ino); 136 }, Eq(true)), 137 _) 138 ).WillRepeatedly(Invoke(ReturnImmediate([=](auto in, auto& out) { 139 const char *buf = (const char*)in.body.bytes + 140 sizeof(struct fuse_write_in); 141 ssize_t isize = in.body.write.size; 142 off_t iofs = in.body.write.offset; 143 144 ASSERT_EQ(isize, pwrite(m_backing_fd, buf, isize, iofs)) 145 << strerror(errno); 146 SET_OUT_HEADER_LEN(out, write); 147 out.body.write.size = isize; 148 }))); 149 EXPECT_CALL(*m_mock, process( 150 ResultOf([=](auto in) { 151 return (in.header.opcode == FUSE_READ && 152 in.header.nodeid == ino); 153 }, Eq(true)), 154 _) 155 ).WillRepeatedly(Invoke(ReturnImmediate([=](auto in, auto& out) { 156 ssize_t isize = in.body.write.size; 157 off_t iofs = in.body.write.offset; 158 void *buf = out.body.bytes; 159 ssize_t osize; 160 161 osize = pread(m_backing_fd, buf, isize, iofs); 162 ASSERT_LE(0, osize) << strerror(errno); 163 out.header.len = sizeof(struct fuse_out_header) + osize; 164 }))); 165 EXPECT_CALL(*m_mock, process( 166 ResultOf([=](auto in) { 167 return (in.header.opcode == FUSE_SETATTR && 168 in.header.nodeid == ino && 169 (in.body.setattr.valid & FATTR_SIZE)); 170 171 }, Eq(true)), 172 _) 173 ).WillRepeatedly(Invoke(ReturnImmediate([=](auto in, auto& out) { 174 ASSERT_EQ(0, ftruncate(m_backing_fd, in.body.setattr.size)) 175 << strerror(errno); 176 SET_OUT_HEADER_LEN(out, attr); 177 out.body.attr.attr.ino = ino; 178 out.body.attr.attr.mode = S_IFREG | 0755; 179 out.body.attr.attr.size = in.body.setattr.size; 180 out.body.attr.attr_valid = UINT64_MAX; 181 }))); 182 /* Any test that close()s will send FUSE_FLUSH and FUSE_RELEASE */ 183 EXPECT_CALL(*m_mock, process( 184 ResultOf([=](auto in) { 185 return (in.header.opcode == FUSE_FLUSH && 186 in.header.nodeid == ino); 187 }, Eq(true)), 188 _) 189 ).WillRepeatedly(Invoke(ReturnErrno(0))); 190 EXPECT_CALL(*m_mock, process( 191 ResultOf([=](auto in) { 192 return (in.header.opcode == FUSE_RELEASE && 193 in.header.nodeid == ino); 194 }, Eq(true)), 195 _) 196 ).WillRepeatedly(Invoke(ReturnErrno(0))); 197 198 m_test_fd = open(FULLPATH, O_RDWR ); 199 EXPECT_LE(0, m_test_fd) << strerror(errno); 200 } 201 202 void TearDown() 203 { 204 if (m_test_fd >= 0) 205 close(m_test_fd); 206 if (m_backing_fd >= 0) 207 close(m_backing_fd); 208 if (m_control_fd >= 0) 209 close(m_control_fd); 210 FuseTest::TearDown(); 211 leak(m_test_fd); 212 } 213 214 void do_closeopen() 215 { 216 ASSERT_EQ(0, close(m_test_fd)) << strerror(errno); 217 m_test_fd = open("backing_file", O_RDWR); 218 ASSERT_LE(0, m_test_fd) << strerror(errno); 219 220 ASSERT_EQ(0, close(m_control_fd)) << strerror(errno); 221 m_control_fd = open("control", O_RDWR); 222 ASSERT_LE(0, m_control_fd) << strerror(errno); 223 } 224 225 void do_ftruncate(off_t offs) 226 { 227 ASSERT_EQ(0, ftruncate(m_test_fd, offs)) << strerror(errno); 228 ASSERT_EQ(0, ftruncate(m_control_fd, offs)) << strerror(errno); 229 m_filesize = offs; 230 } 231 232 void do_mapread(ssize_t size, off_t offs) 233 { 234 void *control_buf, *p; 235 off_t pg_offset, page_mask; 236 size_t map_size; 237 238 page_mask = getpagesize() - 1; 239 pg_offset = offs & page_mask; 240 map_size = pg_offset + size; 241 242 p = mmap(NULL, map_size, PROT_READ, MAP_FILE | MAP_SHARED, m_test_fd, 243 offs - pg_offset); 244 ASSERT_NE(p, MAP_FAILED) << strerror(errno); 245 246 control_buf = malloc(size); 247 ASSERT_NE(nullptr, control_buf) << strerror(errno); 248 249 ASSERT_EQ(size, pread(m_control_fd, control_buf, size, offs)) 250 << strerror(errno); 251 252 compare((void*)((char*)p + pg_offset), control_buf, offs, size); 253 254 ASSERT_EQ(0, munmap(p, map_size)) << strerror(errno); 255 free(control_buf); 256 } 257 258 void do_read(ssize_t size, off_t offs) 259 { 260 void *test_buf, *control_buf; 261 ssize_t r; 262 263 test_buf = malloc(size); 264 ASSERT_NE(nullptr, test_buf) << strerror(errno); 265 control_buf = malloc(size); 266 ASSERT_NE(nullptr, control_buf) << strerror(errno); 267 268 errno = 0; 269 r = pread(m_test_fd, test_buf, size, offs); 270 ASSERT_NE(-1, r) << strerror(errno); 271 ASSERT_EQ(size, r) << "unexpected short read"; 272 r = pread(m_control_fd, control_buf, size, offs); 273 ASSERT_NE(-1, r) << strerror(errno); 274 ASSERT_EQ(size, r) << "unexpected short read"; 275 276 compare(test_buf, control_buf, offs, size); 277 278 free(control_buf); 279 free(test_buf); 280 } 281 282 void do_mapwrite(ssize_t size, off_t offs) 283 { 284 char *buf; 285 void *p; 286 off_t pg_offset, page_mask; 287 size_t map_size; 288 long i; 289 290 page_mask = getpagesize() - 1; 291 pg_offset = offs & page_mask; 292 map_size = pg_offset + size; 293 294 buf = (char*)malloc(size); 295 ASSERT_NE(nullptr, buf) << strerror(errno); 296 for (i=0; i < size; i++) 297 buf[i] = random(); 298 299 if (offs + size > m_filesize) { 300 /* 301 * Must manually extend. vm_mmap_vnode will not implicitly 302 * extend a vnode 303 */ 304 do_ftruncate(offs + size); 305 } 306 307 p = mmap(NULL, map_size, PROT_READ | PROT_WRITE, 308 MAP_FILE | MAP_SHARED, m_test_fd, offs - pg_offset); 309 ASSERT_NE(p, MAP_FAILED) << strerror(errno); 310 311 bcopy(buf, (char*)p + pg_offset, size); 312 ASSERT_EQ(size, pwrite(m_control_fd, buf, size, offs)) 313 << strerror(errno); 314 315 free(buf); 316 ASSERT_EQ(0, munmap(p, map_size)) << strerror(errno); 317 } 318 319 void do_write(ssize_t size, off_t offs) 320 { 321 char *buf; 322 long i; 323 324 buf = (char*)malloc(size); 325 ASSERT_NE(nullptr, buf) << strerror(errno); 326 for (i=0; i < size; i++) 327 buf[i] = random(); 328 329 ASSERT_EQ(size, pwrite(m_test_fd, buf, size, offs )) 330 << strerror(errno); 331 ASSERT_EQ(size, pwrite(m_control_fd, buf, size, offs)) 332 << strerror(errno); 333 m_filesize = std::max(m_filesize, offs + size); 334 335 free(buf); 336 } 337 338 }; 339 340 class IoCacheable: public Io { 341 public: 342 virtual void SetUp() { 343 Io::SetUp(); 344 } 345 }; 346 347 /* 348 * Extend a file with dirty data in the last page of the last block. 349 * 350 * fsx -WR -P /tmp -S8 -N3 fsx.bin 351 */ 352 TEST_P(Io, extend_from_dirty_page) 353 { 354 off_t wofs = 0x21a0; 355 ssize_t wsize = 0xf0a8; 356 off_t rofs = 0xb284; 357 ssize_t rsize = 0x9b22; 358 off_t truncsize = 0x28702; 359 360 do_write(wsize, wofs); 361 do_ftruncate(truncsize); 362 do_read(rsize, rofs); 363 } 364 365 /* 366 * mapwrite into a newly extended part of a file. 367 * 368 * fsx -c 100 -i 100 -l 524288 -o 131072 -N5 -P /tmp -S19 fsx.bin 369 */ 370 TEST_P(IoCacheable, extend_by_mapwrite) 371 { 372 do_mapwrite(0x849e, 0x29a3a); /* [0x29a3a, 0x31ed7] */ 373 do_mapwrite(0x3994, 0x3c7d8); /* [0x3c7d8, 0x4016b] */ 374 do_read(0xf556, 0x30c16); /* [0x30c16, 0x4016b] */ 375 } 376 377 /* 378 * When writing the last page of a file, it must be written synchronously. 379 * Otherwise the cached page can become invalid by a subsequent extend 380 * operation. 381 * 382 * fsx -WR -P /tmp -S642 -N3 fsx.bin 383 */ 384 TEST_P(Io, last_page) 385 { 386 do_write(0xcc77, 0x1134f); /* [0x1134f, 0x1dfc5] */ 387 do_write(0xdfa7, 0x2096a); /* [0x2096a, 0x2e910] */ 388 do_read(0xb5b7, 0x1a3aa); /* [0x1a3aa, 0x25960] */ 389 } 390 391 /* 392 * Read a hole using mmap 393 * 394 * fsx -c 100 -i 100 -l 524288 -o 131072 -N11 -P /tmp -S14 fsx.bin 395 */ 396 TEST_P(IoCacheable, mapread_hole) 397 { 398 do_write(0x123b7, 0xf205); /* [0xf205, 0x215bb] */ 399 do_mapread(0xeeea, 0x2f4c); /* [0x2f4c, 0x11e35] */ 400 } 401 402 /* 403 * Read a hole from a block that contains some cached data. 404 * 405 * fsx -WR -P /tmp -S55 fsx.bin 406 */ 407 TEST_P(Io, read_hole_from_cached_block) 408 { 409 off_t wofs = 0x160c5; 410 ssize_t wsize = 0xa996; 411 off_t rofs = 0x472e; 412 ssize_t rsize = 0xd8d5; 413 414 do_write(wsize, wofs); 415 do_read(rsize, rofs); 416 } 417 418 /* 419 * Truncating a file into a dirty buffer should not causing anything untoward 420 * to happen when that buffer is eventually flushed. 421 * 422 * fsx -WR -P /tmp -S839 -d -N6 fsx.bin 423 */ 424 TEST_P(Io, truncate_into_dirty_buffer) 425 { 426 off_t wofs0 = 0x3bad7; 427 ssize_t wsize0 = 0x4529; 428 off_t wofs1 = 0xc30d; 429 ssize_t wsize1 = 0x5f77; 430 off_t truncsize0 = 0x10916; 431 off_t rofs = 0xdf17; 432 ssize_t rsize = 0x29ff; 433 off_t truncsize1 = 0x152b4; 434 435 do_write(wsize0, wofs0); 436 do_write(wsize1, wofs1); 437 do_ftruncate(truncsize0); 438 do_read(rsize, rofs); 439 do_ftruncate(truncsize1); 440 close(m_test_fd); 441 } 442 443 /* 444 * Truncating a file into a dirty buffer should not causing anything untoward 445 * to happen when that buffer is eventually flushed, even when the buffer's 446 * dirty_off is > 0. 447 * 448 * Based on this command with a few steps removed: 449 * fsx -WR -P /tmp -S677 -d -N8 fsx.bin 450 */ 451 TEST_P(Io, truncate_into_dirty_buffer2) 452 { 453 off_t truncsize0 = 0x344f3; 454 off_t wofs = 0x2790c; 455 ssize_t wsize = 0xd86a; 456 off_t truncsize1 = 0x2de38; 457 off_t rofs2 = 0x1fd7a; 458 ssize_t rsize2 = 0xc594; 459 off_t truncsize2 = 0x31e71; 460 461 /* Sets the file size to something larger than the next write */ 462 do_ftruncate(truncsize0); 463 /* 464 * Creates a dirty buffer. The part in lbn 2 doesn't flush 465 * synchronously. 466 */ 467 do_write(wsize, wofs); 468 /* Truncates part of the dirty buffer created in step 2 */ 469 do_ftruncate(truncsize1); 470 /* XXX ?I don't know why this is necessary? */ 471 do_read(rsize2, rofs2); 472 /* Truncates the dirty buffer */ 473 do_ftruncate(truncsize2); 474 close(m_test_fd); 475 } 476 477 /* 478 * Regression test for a bug introduced in r348931 479 * 480 * Sequence of operations: 481 * 1) The first write reads lbn so it can modify it 482 * 2) The first write flushes lbn 3 immediately because it's the end of file 483 * 3) The first write then flushes lbn 4 because it's the end of the file 484 * 4) The second write modifies the cached versions of lbn 3 and 4 485 * 5) The third write's getblkx invalidates lbn 4's B_CACHE because it's 486 * extending the buffer. Then it flushes lbn 4 because B_DELWRI was set but 487 * B_CACHE was clear. 488 * 6) fuse_write_biobackend erroneously called vfs_bio_clrbuf, putting the 489 * buffer into a weird write-only state. All read operations would return 490 * 0. Writes were apparently still processed, because the buffer's contents 491 * were correct when examined in a core dump. 492 * 7) The third write reads lbn 4 because cache is clear 493 * 9) uiomove dutifully copies new data into the buffer 494 * 10) The buffer's dirty is flushed to lbn 4 495 * 11) The read returns all zeros because of step 6. 496 * 497 * Based on: 498 * fsx -WR -l 524388 -o 131072 -P /tmp -S6456 -q fsx.bin 499 */ 500 TEST_P(Io, resize_a_valid_buffer_while_extending) 501 { 502 do_write(0x14530, 0x36ee6); /* [0x36ee6, 0x4b415] */ 503 do_write(0x1507c, 0x33256); /* [0x33256, 0x482d1] */ 504 do_write(0x175c, 0x4c03d); /* [0x4c03d, 0x4d798] */ 505 do_read(0xe277, 0x3599c); /* [0x3599c, 0x43c12] */ 506 close(m_test_fd); 507 } 508 509 INSTANTIATE_TEST_CASE_P(Io, Io, 510 Combine(Bool(), /* async read */ 511 Values(0x1000, 0x10000, 0x20000), /* m_maxwrite */ 512 Values(Uncached, Writethrough, Writeback, WritebackAsync) 513 ) 514 ); 515 516 INSTANTIATE_TEST_CASE_P(Io, IoCacheable, 517 Combine(Bool(), /* async read */ 518 Values(0x1000, 0x10000, 0x20000), /* m_maxwrite */ 519 Values(Writethrough, Writeback, WritebackAsync) 520 ) 521 ); 522