]>
Commit | Line | Data |
---|---|---|
7c673cae FG |
1 | // -*- mode:C++; tab-width:8; c-basic-offset:8; indent-tabs-mode:t -*- |
2 | // vim: ts=8 sw=8 smarttab | |
3 | /* | |
4 | * Copyright (C) 1991, NeXT Computer, Inc. All Rights Reserverd. | |
5 | * | |
6 | * File: fsx.cc | |
7 | * Author: Avadis Tevanian, Jr. | |
8 | * | |
9 | * File system exerciser. | |
10 | * | |
11 | * Rewritten 8/98 by Conrad Minshall. | |
12 | * | |
13 | * Small changes to work under Linux -- davej. | |
14 | * | |
15 | * Checks for mmap last-page zero fill. | |
16 | */ | |
17 | ||
18 | #include <sys/types.h> | |
19 | #include <unistd.h> | |
11fdf7f2 | 20 | #include <getopt.h> |
7c673cae | 21 | #include <limits.h> |
7c673cae | 22 | #include <strings.h> |
11fdf7f2 TL |
23 | #if defined(__FreeBSD__) |
24 | #include <sys/disk.h> | |
25 | #endif | |
7c673cae FG |
26 | #include <sys/file.h> |
27 | #include <sys/stat.h> | |
28 | #include <sys/mman.h> | |
11fdf7f2 | 29 | #if defined(__linux__) |
7c673cae | 30 | #include <linux/fs.h> |
11fdf7f2 | 31 | #endif |
7c673cae FG |
32 | #include <sys/ioctl.h> |
33 | #ifdef HAVE_ERR_H | |
34 | #include <err.h> | |
35 | #endif | |
36 | #include <signal.h> | |
7c673cae FG |
37 | #include <stddef.h> |
38 | #include <stdio.h> | |
39 | #include <stdlib.h> | |
40 | #include <string.h> | |
41 | #include <stdarg.h> | |
42 | #include <assert.h> | |
43 | #include <errno.h> | |
44 | #include <math.h> | |
45 | #include <fcntl.h> | |
46 | #include <random> | |
47 | ||
11fdf7f2 | 48 | #include "include/compat.h" |
7c673cae | 49 | #include "include/intarith.h" |
11fdf7f2 | 50 | #if defined(WITH_KRBD) |
7c673cae | 51 | #include "include/krbd.h" |
11fdf7f2 | 52 | #endif |
7c673cae FG |
53 | #include "include/rados/librados.h" |
54 | #include "include/rados/librados.hpp" | |
55 | #include "include/rbd/librbd.h" | |
56 | #include "include/rbd/librbd.hpp" | |
57 | #include "common/Cond.h" | |
58 | #include "common/SubProcess.h" | |
59 | #include "common/safe_io.h" | |
60 | #include "journal/Journaler.h" | |
61 | #include "journal/ReplayEntry.h" | |
62 | #include "journal/ReplayHandler.h" | |
63 | #include "journal/Settings.h" | |
64 | ||
65 | #include <boost/scope_exit.hpp> | |
66 | ||
67 | #define NUMPRINTCOLUMNS 32 /* # columns of data to print on each line */ | |
68 | ||
69 | /* | |
70 | * A log entry is an operation and a bunch of arguments. | |
71 | */ | |
72 | ||
73 | struct log_entry { | |
74 | int operation; | |
75 | int args[3]; | |
76 | }; | |
77 | ||
78 | #define LOGSIZE 1000 | |
79 | ||
80 | struct log_entry oplog[LOGSIZE]; /* the log */ | |
81 | int logptr = 0; /* current position in log */ | |
82 | int logcount = 0; /* total ops */ | |
83 | ||
84 | /* | |
85 | * The operation matrix is complex due to conditional execution of different | |
86 | * features. Hence when we come to deciding what operation to run, we need to | |
87 | * be careful in how we select the different operations. The active operations | |
88 | * are mapped to numbers as follows: | |
89 | * | |
90 | * lite !lite | |
91 | * READ: 0 0 | |
92 | * WRITE: 1 1 | |
93 | * MAPREAD: 2 2 | |
94 | * MAPWRITE: 3 3 | |
95 | * TRUNCATE: - 4 | |
96 | * FALLOCATE: - 5 | |
97 | * PUNCH HOLE: - 6 | |
98 | * WRITESAME: - 7 | |
c07f9fc5 | 99 | * COMPAREANDWRITE: - 8 |
7c673cae FG |
100 | * |
101 | * When mapped read/writes are disabled, they are simply converted to normal | |
102 | * reads and writes. When fallocate/fpunch calls are disabled, they are | |
103 | * converted to OP_SKIPPED. Hence OP_SKIPPED needs to have a number higher than | |
104 | * the operation selction matrix, as does the OP_CLOSEOPEN which is an | |
105 | * operation modifier rather than an operation in itself. | |
106 | * | |
107 | * Because of the "lite" version, we also need to have different "maximum | |
108 | * operation" defines to allow the ops to be selected correctly based on the | |
109 | * mode being run. | |
110 | */ | |
111 | ||
112 | /* common operations */ | |
113 | #define OP_READ 0 | |
114 | #define OP_WRITE 1 | |
115 | #define OP_MAPREAD 2 | |
116 | #define OP_MAPWRITE 3 | |
117 | #define OP_MAX_LITE 4 | |
118 | ||
119 | /* !lite operations */ | |
120 | #define OP_TRUNCATE 4 | |
121 | #define OP_FALLOCATE 5 | |
122 | #define OP_PUNCH_HOLE 6 | |
123 | #define OP_WRITESAME 7 | |
c07f9fc5 | 124 | #define OP_COMPARE_AND_WRITE 8 |
7c673cae | 125 | /* rbd-specific operations */ |
c07f9fc5 FG |
126 | #define OP_CLONE 9 |
127 | #define OP_FLATTEN 10 | |
128 | #define OP_MAX_FULL 11 | |
7c673cae FG |
129 | |
130 | /* operation modifiers */ | |
131 | #define OP_CLOSEOPEN 100 | |
132 | #define OP_SKIPPED 101 | |
133 | ||
134 | #undef PAGE_SIZE | |
135 | #define PAGE_SIZE getpagesize() | |
136 | #undef PAGE_MASK | |
137 | #define PAGE_MASK (PAGE_SIZE - 1) | |
138 | ||
139 | ||
140 | char *original_buf; /* a pointer to the original data */ | |
141 | char *good_buf; /* a pointer to the correct data */ | |
142 | char *temp_buf; /* a pointer to the current data */ | |
143 | ||
144 | char dirpath[1024]; | |
145 | ||
146 | off_t file_size = 0; | |
147 | off_t biggest = 0; | |
148 | unsigned long testcalls = 0; /* calls to function "test" */ | |
149 | ||
11fdf7f2 TL |
150 | const char* cluster_name = "ceph"; /* --cluster optional */ |
151 | const char* client_id = "admin"; /* --id optional */ | |
152 | ||
7c673cae FG |
153 | unsigned long simulatedopcount = 0; /* -b flag */ |
154 | int closeprob = 0; /* -c flag */ | |
155 | int debug = 0; /* -d flag */ | |
156 | unsigned long debugstart = 0; /* -D flag */ | |
157 | int flush_enabled = 0; /* -f flag */ | |
11fdf7f2 | 158 | int deep_copy = 0; /* -g flag */ |
7c673cae FG |
159 | int holebdy = 1; /* -h flag */ |
160 | bool journal_replay = false; /* -j flah */ | |
161 | int keep_on_success = 0; /* -k flag */ | |
162 | int do_fsync = 0; /* -y flag */ | |
163 | unsigned long maxfilelen = 256 * 1024; /* -l flag */ | |
164 | int sizechecks = 1; /* -n flag disables them */ | |
165 | int maxoplen = 64 * 1024; /* -o flag */ | |
166 | int quiet = 0; /* -q flag */ | |
167 | unsigned long progressinterval = 0; /* -p flag */ | |
168 | int readbdy = 1; /* -r flag */ | |
169 | int style = 0; /* -s flag */ | |
170 | int prealloc = 0; /* -x flag */ | |
171 | int truncbdy = 1; /* -t flag */ | |
172 | int writebdy = 1; /* -w flag */ | |
173 | long monitorstart = -1; /* -m flag */ | |
174 | long monitorend = -1; /* -m flag */ | |
175 | int lite = 0; /* -L flag */ | |
176 | long numops = -1; /* -N flag */ | |
177 | int randomoplen = 1; /* -O flag disables it */ | |
178 | int seed = 1; /* -S flag */ | |
179 | int mapped_writes = 0; /* -W flag disables */ | |
180 | int fallocate_calls = 0; /* -F flag disables */ | |
181 | int punch_hole_calls = 1; /* -H flag disables */ | |
182 | int clone_calls = 1; /* -C flag disables */ | |
183 | int randomize_striping = 1; /* -U flag disables */ | |
184 | int randomize_parent_overlap = 1; | |
185 | int mapped_reads = 0; /* -R flag disables it */ | |
186 | int fsxgoodfd = 0; | |
187 | int o_direct = 0; /* -Z flag */ | |
188 | ||
189 | int num_clones = 0; | |
190 | ||
191 | int page_size; | |
192 | int page_mask; | |
193 | int mmap_mask; | |
194 | ||
195 | FILE * fsxlogf = NULL; | |
196 | int badoff = -1; | |
197 | int closeopen = 0; | |
198 | ||
199 | void | |
200 | vwarnc(int code, const char *fmt, va_list ap) { | |
201 | fprintf(stderr, "fsx: "); | |
202 | if (fmt != NULL) { | |
203 | vfprintf(stderr, fmt, ap); | |
204 | fprintf(stderr, ": "); | |
205 | } | |
206 | fprintf(stderr, "%s\n", strerror(code)); | |
207 | } | |
208 | ||
209 | void | |
210 | warn(const char * fmt, ...) { | |
211 | va_list ap; | |
212 | va_start(ap, fmt); | |
213 | vwarnc(errno, fmt, ap); | |
214 | va_end(ap); | |
215 | } | |
216 | ||
217 | #define BUF_SIZE 1024 | |
218 | ||
219 | void | |
220 | prt(const char *fmt, ...) | |
221 | { | |
222 | va_list args; | |
223 | char buffer[BUF_SIZE]; | |
224 | ||
225 | va_start(args, fmt); | |
226 | vsnprintf(buffer, BUF_SIZE, fmt, args); | |
227 | va_end(args); | |
228 | fprintf(stdout, "%s", buffer); | |
229 | if (fsxlogf) | |
230 | fprintf(fsxlogf, "%s", buffer); | |
231 | } | |
232 | ||
233 | void | |
234 | prterr(const char *prefix) | |
235 | { | |
236 | prt("%s%s%s\n", prefix, prefix ? ": " : "", strerror(errno)); | |
237 | } | |
238 | ||
239 | void | |
240 | prterrcode(const char *prefix, int code) | |
241 | { | |
242 | prt("%s%s%s\n", prefix, prefix ? ": " : "", strerror(-code)); | |
243 | } | |
244 | ||
245 | void | |
246 | simple_err(const char *msg, int err) | |
247 | { | |
248 | fprintf(stderr, "%s: %s\n", msg, strerror(-err)); | |
249 | } | |
250 | ||
251 | /* | |
252 | * random | |
253 | */ | |
254 | std::mt19937 random_generator; | |
255 | ||
256 | uint_fast32_t | |
257 | get_random(void) | |
258 | { | |
259 | return random_generator(); | |
260 | } | |
261 | ||
11fdf7f2 | 262 | int get_features(uint64_t* features); |
7c673cae FG |
263 | void replay_imagename(char *buf, size_t len, int clones); |
264 | ||
265 | namespace { | |
266 | ||
267 | static const std::string JOURNAL_CLIENT_ID("fsx"); | |
268 | ||
269 | struct ReplayHandler : public journal::ReplayHandler { | |
270 | journal::Journaler *journaler; | |
271 | journal::Journaler *replay_journaler; | |
272 | Context *on_finish; | |
273 | ||
274 | ReplayHandler(journal::Journaler *journaler, | |
275 | journal::Journaler *replay_journaler, Context *on_finish) | |
276 | : journaler(journaler), replay_journaler(replay_journaler), | |
277 | on_finish(on_finish) { | |
278 | } | |
279 | ||
280 | void get() override { | |
281 | } | |
282 | void put() override { | |
283 | } | |
284 | ||
285 | void handle_entries_available() override { | |
286 | while (true) { | |
287 | journal::ReplayEntry replay_entry; | |
288 | if (!journaler->try_pop_front(&replay_entry)) { | |
289 | return; | |
290 | } | |
291 | ||
292 | replay_journaler->append(0, replay_entry.get_data()); | |
293 | } | |
294 | } | |
295 | ||
296 | void handle_complete(int r) override { | |
297 | on_finish->complete(r); | |
298 | } | |
299 | }; | |
300 | ||
301 | int get_image_id(librados::IoCtx &io_ctx, const char *image_name, | |
302 | std::string *image_id) { | |
303 | librbd::RBD rbd; | |
304 | librbd::Image image; | |
305 | int r = rbd.open(io_ctx, image, image_name); | |
306 | if (r < 0) { | |
307 | simple_err("failed to open image", r); | |
308 | return r; | |
309 | } | |
310 | ||
311 | rbd_image_info_t info; | |
312 | r = image.stat(info, sizeof(info)); | |
313 | if (r < 0) { | |
314 | simple_err("failed to stat image", r); | |
315 | return r; | |
316 | } | |
317 | ||
318 | *image_id = std::string(&info.block_name_prefix[strlen(RBD_DATA_PREFIX)]); | |
319 | return 0; | |
320 | } | |
321 | ||
322 | int register_journal(rados_ioctx_t ioctx, const char *image_name) { | |
323 | librados::IoCtx io_ctx; | |
324 | librados::IoCtx::from_rados_ioctx_t(ioctx, io_ctx); | |
325 | ||
326 | std::string image_id; | |
327 | int r = get_image_id(io_ctx, image_name, &image_id); | |
328 | if (r < 0) { | |
329 | return r; | |
330 | } | |
331 | ||
332 | journal::Journaler journaler(io_ctx, image_id, JOURNAL_CLIENT_ID, {}); | |
333 | r = journaler.register_client(bufferlist()); | |
334 | if (r < 0) { | |
335 | simple_err("failed to register journal client", r); | |
336 | return r; | |
337 | } | |
338 | return 0; | |
339 | } | |
340 | ||
341 | int unregister_journal(rados_ioctx_t ioctx, const char *image_name) { | |
342 | librados::IoCtx io_ctx; | |
343 | librados::IoCtx::from_rados_ioctx_t(ioctx, io_ctx); | |
344 | ||
345 | std::string image_id; | |
346 | int r = get_image_id(io_ctx, image_name, &image_id); | |
347 | if (r < 0) { | |
348 | return r; | |
349 | } | |
350 | ||
351 | journal::Journaler journaler(io_ctx, image_id, JOURNAL_CLIENT_ID, {}); | |
352 | r = journaler.unregister_client(); | |
353 | if (r < 0) { | |
354 | simple_err("failed to unregister journal client", r); | |
355 | return r; | |
356 | } | |
357 | return 0; | |
358 | } | |
359 | ||
360 | int create_replay_image(rados_ioctx_t ioctx, int order, | |
361 | uint64_t stripe_unit, int stripe_count, | |
362 | const char *replay_image_name, | |
363 | const char *last_replay_image_name) { | |
364 | librados::IoCtx io_ctx; | |
365 | librados::IoCtx::from_rados_ioctx_t(ioctx, io_ctx); | |
366 | ||
11fdf7f2 TL |
367 | uint64_t features; |
368 | int r = get_features(&features); | |
369 | if (r < 0) { | |
370 | return r; | |
371 | } | |
372 | ||
7c673cae FG |
373 | librbd::RBD rbd; |
374 | if (last_replay_image_name == nullptr) { | |
11fdf7f2 | 375 | r = rbd.create2(io_ctx, replay_image_name, 0, features, &order); |
7c673cae FG |
376 | } else { |
377 | r = rbd.clone2(io_ctx, last_replay_image_name, "snap", | |
11fdf7f2 TL |
378 | io_ctx, replay_image_name, features, &order, |
379 | stripe_unit, stripe_count); | |
7c673cae FG |
380 | } |
381 | ||
382 | if (r < 0) { | |
383 | simple_err("failed to create replay image", r); | |
384 | return r; | |
385 | } | |
386 | ||
387 | return 0; | |
388 | } | |
389 | ||
390 | int replay_journal(rados_ioctx_t ioctx, const char *image_name, | |
391 | const char *replay_image_name) { | |
392 | librados::IoCtx io_ctx; | |
393 | librados::IoCtx::from_rados_ioctx_t(ioctx, io_ctx); | |
394 | ||
395 | std::string image_id; | |
396 | int r = get_image_id(io_ctx, image_name, &image_id); | |
397 | if (r < 0) { | |
398 | return r; | |
399 | } | |
400 | ||
401 | std::string replay_image_id; | |
402 | r = get_image_id(io_ctx, replay_image_name, &replay_image_id); | |
403 | if (r < 0) { | |
404 | return r; | |
405 | } | |
406 | ||
407 | journal::Journaler journaler(io_ctx, image_id, JOURNAL_CLIENT_ID, {}); | |
408 | C_SaferCond init_ctx; | |
409 | journaler.init(&init_ctx); | |
410 | BOOST_SCOPE_EXIT_ALL( (&journaler) ) { | |
411 | journaler.shut_down(); | |
412 | }; | |
413 | ||
414 | r = init_ctx.wait(); | |
415 | if (r < 0) { | |
416 | simple_err("failed to initialize journal", r); | |
417 | return r; | |
418 | } | |
419 | ||
420 | journal::Journaler replay_journaler(io_ctx, replay_image_id, "", {}); | |
421 | ||
422 | C_SaferCond replay_init_ctx; | |
423 | replay_journaler.init(&replay_init_ctx); | |
424 | BOOST_SCOPE_EXIT_ALL( (&replay_journaler) ) { | |
425 | replay_journaler.shut_down(); | |
426 | }; | |
427 | ||
428 | r = replay_init_ctx.wait(); | |
429 | if (r < 0) { | |
430 | simple_err("failed to initialize replay journal", r); | |
431 | return r; | |
432 | } | |
433 | ||
494da23a | 434 | replay_journaler.start_append(0); |
7c673cae FG |
435 | |
436 | C_SaferCond replay_ctx; | |
437 | ReplayHandler replay_handler(&journaler, &replay_journaler, | |
438 | &replay_ctx); | |
439 | ||
440 | // copy journal events from source image to replay image | |
441 | journaler.start_replay(&replay_handler); | |
442 | r = replay_ctx.wait(); | |
443 | ||
444 | journaler.stop_replay(); | |
445 | ||
446 | C_SaferCond stop_ctx; | |
447 | replay_journaler.stop_append(&stop_ctx); | |
448 | int stop_r = stop_ctx.wait(); | |
449 | if (r == 0 && stop_r < 0) { | |
450 | r = stop_r; | |
451 | } | |
452 | ||
453 | if (r < 0) { | |
454 | simple_err("failed to replay journal", r); | |
455 | return r; | |
456 | } | |
457 | ||
458 | librbd::RBD rbd; | |
459 | librbd::Image image; | |
460 | r = rbd.open(io_ctx, image, replay_image_name); | |
461 | if (r < 0) { | |
462 | simple_err("failed to open replay image", r); | |
463 | return r; | |
464 | } | |
465 | ||
466 | // perform an IO op to initiate the journal replay | |
467 | bufferlist bl; | |
468 | r = static_cast<ssize_t>(image.write(0, 0, bl)); | |
469 | if (r < 0) { | |
470 | simple_err("failed to write to replay image", r); | |
471 | return r; | |
472 | } | |
473 | return 0; | |
474 | } | |
475 | ||
476 | int finalize_journal(rados_ioctx_t ioctx, const char *imagename, int clones, | |
477 | int order, uint64_t stripe_unit, int stripe_count) { | |
478 | char replayimagename[1024]; | |
479 | replay_imagename(replayimagename, sizeof(replayimagename), clones); | |
480 | ||
481 | char lastreplayimagename[1024]; | |
482 | if (clones > 0) { | |
483 | replay_imagename(lastreplayimagename, | |
484 | sizeof(lastreplayimagename), clones - 1); | |
485 | } | |
486 | ||
487 | int ret = create_replay_image(ioctx, order, stripe_unit, | |
488 | stripe_count, replayimagename, | |
489 | clones > 0 ? lastreplayimagename : | |
490 | nullptr); | |
491 | if (ret < 0) { | |
492 | exit(EXIT_FAILURE); | |
493 | } | |
494 | ||
495 | ret = replay_journal(ioctx, imagename, replayimagename); | |
496 | if (ret < 0) { | |
497 | exit(EXIT_FAILURE); | |
498 | } | |
499 | return 0; | |
500 | } | |
501 | ||
502 | } // anonymous namespace | |
503 | ||
504 | /* | |
505 | * rbd | |
506 | */ | |
507 | ||
508 | struct rbd_ctx { | |
509 | const char *name; /* image name */ | |
510 | rbd_image_t image; /* image handle */ | |
511 | const char *krbd_name; /* image /dev/rbd<id> name */ /* reused for nbd test */ | |
512 | int krbd_fd; /* image /dev/rbd<id> fd */ /* reused for nbd test */ | |
513 | }; | |
514 | ||
515 | #define RBD_CTX_INIT (struct rbd_ctx) { NULL, NULL, NULL, -1} | |
516 | ||
517 | struct rbd_operations { | |
518 | int (*open)(const char *name, struct rbd_ctx *ctx); | |
519 | int (*close)(struct rbd_ctx *ctx); | |
520 | ssize_t (*read)(struct rbd_ctx *ctx, uint64_t off, size_t len, char *buf); | |
521 | ssize_t (*write)(struct rbd_ctx *ctx, uint64_t off, size_t len, const char *buf); | |
522 | int (*flush)(struct rbd_ctx *ctx); | |
523 | int (*discard)(struct rbd_ctx *ctx, uint64_t off, uint64_t len); | |
524 | int (*get_size)(struct rbd_ctx *ctx, uint64_t *size); | |
525 | int (*resize)(struct rbd_ctx *ctx, uint64_t size); | |
526 | int (*clone)(struct rbd_ctx *ctx, const char *src_snapname, | |
527 | const char *dst_imagename, int *order, int stripe_unit, | |
528 | int stripe_count); | |
529 | int (*flatten)(struct rbd_ctx *ctx); | |
530 | ssize_t (*writesame)(struct rbd_ctx *ctx, uint64_t off, size_t len, | |
531 | const char *buf, size_t data_len); | |
c07f9fc5 FG |
532 | ssize_t (*compare_and_write)(struct rbd_ctx *ctx, uint64_t off, size_t len, |
533 | const char *cmp_buf, const char *buf); | |
7c673cae FG |
534 | }; |
535 | ||
536 | char *pool; /* name of the pool our test image is in */ | |
537 | char *iname; /* name of our test image */ | |
538 | rados_t cluster; /* handle for our test cluster */ | |
539 | rados_ioctx_t ioctx; /* handle for our test pool */ | |
11fdf7f2 | 540 | #if defined(WITH_KRBD) |
7c673cae | 541 | struct krbd_ctx *krbd; /* handle for libkrbd */ |
11fdf7f2 | 542 | #endif |
7c673cae FG |
543 | bool skip_partial_discard; /* rbd_skip_partial_discard config value*/ |
544 | ||
11fdf7f2 TL |
545 | int get_features(uint64_t* features) { |
546 | char buf[1024]; | |
547 | int r = rados_conf_get(cluster, "rbd_default_features", buf, | |
548 | sizeof(buf)); | |
549 | if (r < 0) { | |
550 | simple_err("Could not get rbd_default_features value", r); | |
551 | return r; | |
552 | } | |
553 | ||
554 | *features = strtol(buf, NULL, 0); | |
555 | ||
556 | if (clone_calls) { | |
557 | *features |= RBD_FEATURE_LAYERING; | |
558 | } | |
559 | if (journal_replay) { | |
560 | *features |= (RBD_FEATURE_EXCLUSIVE_LOCK | | |
561 | RBD_FEATURE_JOURNALING); | |
562 | } | |
563 | return 0; | |
564 | } | |
565 | ||
7c673cae FG |
566 | /* |
567 | * librbd/krbd rbd_operations handlers. Given the rest of fsx.c, no | |
568 | * attempt to do error handling is made in these handlers. | |
569 | */ | |
570 | ||
571 | int | |
572 | __librbd_open(const char *name, struct rbd_ctx *ctx) | |
573 | { | |
574 | rbd_image_t image; | |
575 | int ret; | |
576 | ||
11fdf7f2 | 577 | ceph_assert(!ctx->name && !ctx->image && |
7c673cae FG |
578 | !ctx->krbd_name && ctx->krbd_fd < 0); |
579 | ||
580 | ret = rbd_open(ioctx, name, &image, NULL); | |
581 | if (ret < 0) { | |
582 | prt("rbd_open(%s) failed\n", name); | |
583 | return ret; | |
584 | } | |
585 | ||
586 | ctx->name = strdup(name); | |
587 | ctx->image = image; | |
588 | ctx->krbd_name = NULL; | |
589 | ctx->krbd_fd = -1; | |
590 | ||
591 | return 0; | |
592 | } | |
593 | ||
594 | int | |
595 | librbd_open(const char *name, struct rbd_ctx *ctx) | |
596 | { | |
597 | return __librbd_open(name, ctx); | |
598 | } | |
599 | ||
600 | int | |
601 | __librbd_close(struct rbd_ctx *ctx) | |
602 | { | |
603 | int ret; | |
604 | ||
11fdf7f2 | 605 | ceph_assert(ctx->name && ctx->image); |
7c673cae FG |
606 | |
607 | ret = rbd_close(ctx->image); | |
608 | if (ret < 0) { | |
609 | prt("rbd_close(%s) failed\n", ctx->name); | |
610 | return ret; | |
611 | } | |
612 | ||
613 | free((void *)ctx->name); | |
614 | ||
615 | ctx->name = NULL; | |
616 | ctx->image = NULL; | |
617 | ||
618 | return 0; | |
619 | } | |
620 | ||
621 | int | |
622 | librbd_close(struct rbd_ctx *ctx) | |
623 | { | |
624 | return __librbd_close(ctx); | |
625 | } | |
626 | ||
627 | int | |
628 | librbd_verify_object_map(struct rbd_ctx *ctx) | |
629 | { | |
630 | int n; | |
631 | uint64_t flags; | |
632 | n = rbd_get_flags(ctx->image, &flags); | |
633 | if (n < 0) { | |
634 | prt("rbd_get_flags() failed\n"); | |
635 | return n; | |
636 | } | |
637 | ||
638 | if ((flags & RBD_FLAG_OBJECT_MAP_INVALID) != 0) { | |
639 | prt("rbd_get_flags() indicates object map is invalid\n"); | |
640 | return -EINVAL; | |
641 | } | |
642 | return 0; | |
643 | } | |
644 | ||
645 | ssize_t | |
646 | librbd_read(struct rbd_ctx *ctx, uint64_t off, size_t len, char *buf) | |
647 | { | |
648 | ssize_t n; | |
649 | ||
650 | n = rbd_read(ctx->image, off, len, buf); | |
651 | if (n < 0) | |
652 | prt("rbd_read(%llu, %zu) failed\n", off, len); | |
653 | ||
654 | return n; | |
655 | } | |
656 | ||
657 | ssize_t | |
658 | librbd_write(struct rbd_ctx *ctx, uint64_t off, size_t len, const char *buf) | |
659 | { | |
660 | ssize_t n; | |
661 | int ret; | |
662 | ||
663 | n = rbd_write(ctx->image, off, len, buf); | |
664 | if (n < 0) { | |
665 | prt("rbd_write(%llu, %zu) failed\n", off, len); | |
666 | return n; | |
667 | } | |
668 | ||
669 | ret = librbd_verify_object_map(ctx); | |
670 | if (ret < 0) { | |
671 | return ret; | |
672 | } | |
673 | return n; | |
674 | } | |
675 | ||
676 | int | |
677 | librbd_flush(struct rbd_ctx *ctx) | |
678 | { | |
679 | int ret; | |
680 | ||
681 | ret = rbd_flush(ctx->image); | |
682 | if (ret < 0) { | |
683 | prt("rbd_flush failed\n"); | |
684 | return ret; | |
685 | } | |
686 | ||
687 | return librbd_verify_object_map(ctx); | |
688 | } | |
689 | ||
690 | int | |
691 | librbd_discard(struct rbd_ctx *ctx, uint64_t off, uint64_t len) | |
692 | { | |
693 | int ret; | |
694 | ||
695 | ret = rbd_discard(ctx->image, off, len); | |
696 | if (ret < 0) { | |
697 | prt("rbd_discard(%llu, %llu) failed\n", off, len); | |
698 | return ret; | |
699 | } | |
700 | ||
701 | return librbd_verify_object_map(ctx); | |
702 | } | |
703 | ||
704 | ssize_t | |
705 | librbd_writesame(struct rbd_ctx *ctx, uint64_t off, size_t len, | |
706 | const char *buf, size_t data_len) | |
707 | { | |
708 | ssize_t n; | |
709 | int ret; | |
710 | ||
711 | n = rbd_writesame(ctx->image, off, len, buf, data_len, 0); | |
712 | if (n < 0) { | |
713 | prt("rbd_writesame(%llu, %zu) failed\n", off, len); | |
714 | return n; | |
715 | } | |
716 | ||
717 | ret = librbd_verify_object_map(ctx); | |
718 | if (ret < 0) { | |
719 | return ret; | |
720 | } | |
721 | return n; | |
722 | } | |
723 | ||
c07f9fc5 FG |
724 | ssize_t |
725 | librbd_compare_and_write(struct rbd_ctx *ctx, uint64_t off, size_t len, | |
726 | const char *cmp_buf, const char *buf) | |
727 | { | |
728 | ssize_t n; | |
729 | int ret; | |
730 | uint64_t mismatch_off = 0; | |
731 | ||
732 | n = rbd_compare_and_write(ctx->image, off, len, cmp_buf, buf, &mismatch_off, 0); | |
733 | if (n == -EINVAL) { | |
734 | return n; | |
735 | } else if (n < 0) { | |
736 | prt("rbd_compare_and_write mismatch(%llu, %zu, %llu) failed\n", | |
737 | off, len, mismatch_off); | |
738 | return n; | |
739 | } | |
740 | ||
741 | ret = librbd_verify_object_map(ctx); | |
742 | if (ret < 0) { | |
743 | return ret; | |
744 | } | |
745 | return n; | |
746 | ||
747 | } | |
748 | ||
7c673cae FG |
749 | int |
750 | librbd_get_size(struct rbd_ctx *ctx, uint64_t *size) | |
751 | { | |
7c673cae FG |
752 | int ret; |
753 | ||
11fdf7f2 | 754 | ret = rbd_get_size(ctx->image, size); |
7c673cae | 755 | if (ret < 0) { |
11fdf7f2 | 756 | prt("rbd_get_size failed\n"); |
7c673cae FG |
757 | return ret; |
758 | } | |
759 | ||
7c673cae FG |
760 | return 0; |
761 | } | |
762 | ||
763 | int | |
764 | __librbd_resize(struct rbd_ctx *ctx, uint64_t size) | |
765 | { | |
766 | int ret; | |
767 | ||
768 | ret = rbd_resize(ctx->image, size); | |
769 | if (ret < 0) { | |
770 | prt("rbd_resize(%llu) failed\n", size); | |
771 | return ret; | |
772 | } | |
773 | ||
774 | return librbd_verify_object_map(ctx); | |
775 | } | |
776 | ||
777 | int | |
778 | librbd_resize(struct rbd_ctx *ctx, uint64_t size) | |
779 | { | |
780 | return __librbd_resize(ctx, size); | |
781 | } | |
782 | ||
11fdf7f2 TL |
783 | int |
784 | __librbd_deep_copy(struct rbd_ctx *ctx, const char *src_snapname, | |
785 | const char *dst_imagename, uint64_t features, int *order, | |
786 | int stripe_unit, int stripe_count) { | |
787 | int ret; | |
788 | ||
789 | rbd_image_options_t opts; | |
790 | rbd_image_options_create(&opts); | |
791 | BOOST_SCOPE_EXIT_ALL( (&opts) ) { | |
792 | rbd_image_options_destroy(opts); | |
793 | }; | |
794 | ret = rbd_image_options_set_uint64(opts, RBD_IMAGE_OPTION_FEATURES, | |
795 | features); | |
796 | ceph_assert(ret == 0); | |
797 | ret = rbd_image_options_set_uint64(opts, RBD_IMAGE_OPTION_ORDER, | |
798 | *order); | |
799 | ceph_assert(ret == 0); | |
800 | ret = rbd_image_options_set_uint64(opts, RBD_IMAGE_OPTION_STRIPE_UNIT, | |
801 | stripe_unit); | |
802 | ceph_assert(ret == 0); | |
803 | ret = rbd_image_options_set_uint64(opts, RBD_IMAGE_OPTION_STRIPE_COUNT, | |
804 | stripe_count); | |
805 | ceph_assert(ret == 0); | |
806 | ||
807 | ret = rbd_snap_set(ctx->image, src_snapname); | |
808 | if (ret < 0) { | |
809 | prt("rbd_snap_set(%s@%s) failed\n", ctx->name, src_snapname); | |
810 | return ret; | |
811 | } | |
812 | ||
813 | ret = rbd_deep_copy(ctx->image, ioctx, dst_imagename, opts); | |
814 | if (ret < 0) { | |
815 | prt("rbd_deep_copy(%s@%s -> %s) failed\n", | |
816 | ctx->name, src_snapname, dst_imagename); | |
817 | return ret; | |
818 | } | |
819 | ||
820 | ret = rbd_snap_set(ctx->image, ""); | |
821 | if (ret < 0) { | |
822 | prt("rbd_snap_set(%s@) failed\n", ctx->name); | |
823 | return ret; | |
824 | } | |
825 | ||
826 | rbd_image_t image; | |
827 | ret = rbd_open(ioctx, dst_imagename, &image, nullptr); | |
828 | if (ret < 0) { | |
829 | prt("rbd_open(%s) failed\n", dst_imagename); | |
830 | return ret; | |
831 | } | |
832 | ||
833 | ret = rbd_snap_unprotect(image, src_snapname); | |
834 | if (ret < 0) { | |
835 | prt("rbd_snap_unprotect(%s@%s) failed\n", dst_imagename, | |
836 | src_snapname); | |
837 | return ret; | |
838 | } | |
839 | ||
840 | ret = rbd_snap_remove(image, src_snapname); | |
841 | if (ret < 0) { | |
842 | prt("rbd_snap_remove(%s@%s) failed\n", dst_imagename, | |
843 | src_snapname); | |
844 | return ret; | |
845 | } | |
846 | ||
847 | ret = rbd_close(image); | |
848 | if (ret < 0) { | |
849 | prt("rbd_close(%s) failed\n", dst_imagename); | |
850 | return ret; | |
851 | } | |
852 | ||
853 | return 0; | |
854 | } | |
855 | ||
7c673cae FG |
856 | int |
857 | __librbd_clone(struct rbd_ctx *ctx, const char *src_snapname, | |
858 | const char *dst_imagename, int *order, int stripe_unit, | |
859 | int stripe_count, bool krbd) | |
860 | { | |
861 | int ret; | |
862 | ||
863 | ret = rbd_snap_create(ctx->image, src_snapname); | |
864 | if (ret < 0) { | |
865 | prt("rbd_snap_create(%s@%s) failed\n", ctx->name, | |
866 | src_snapname); | |
867 | return ret; | |
868 | } | |
869 | ||
870 | ret = rbd_snap_protect(ctx->image, src_snapname); | |
871 | if (ret < 0) { | |
872 | prt("rbd_snap_protect(%s@%s) failed\n", ctx->name, | |
873 | src_snapname); | |
874 | return ret; | |
875 | } | |
876 | ||
11fdf7f2 TL |
877 | uint64_t features; |
878 | ret = get_features(&features); | |
879 | if (ret < 0) { | |
880 | return ret; | |
881 | } | |
882 | ||
7c673cae FG |
883 | if (krbd) { |
884 | features &= ~(RBD_FEATURE_OBJECT_MAP | | |
885 | RBD_FEATURE_FAST_DIFF | | |
886 | RBD_FEATURE_DEEP_FLATTEN | | |
887 | RBD_FEATURE_JOURNALING); | |
888 | } | |
11fdf7f2 TL |
889 | if (deep_copy) { |
890 | ret = __librbd_deep_copy(ctx, src_snapname, dst_imagename, features, | |
891 | order, stripe_unit, stripe_count); | |
892 | if (ret < 0) { | |
893 | prt("deep_copy(%s@%s -> %s) failed\n", ctx->name, | |
894 | src_snapname, dst_imagename); | |
895 | return ret; | |
896 | } | |
897 | } else { | |
898 | ret = rbd_clone2(ioctx, ctx->name, src_snapname, ioctx, | |
899 | dst_imagename, features, order, | |
900 | stripe_unit, stripe_count); | |
901 | if (ret < 0) { | |
902 | prt("rbd_clone2(%s@%s -> %s) failed\n", ctx->name, | |
903 | src_snapname, dst_imagename); | |
904 | return ret; | |
905 | } | |
7c673cae FG |
906 | } |
907 | ||
908 | return 0; | |
909 | } | |
910 | ||
911 | int | |
912 | librbd_clone(struct rbd_ctx *ctx, const char *src_snapname, | |
913 | const char *dst_imagename, int *order, int stripe_unit, | |
914 | int stripe_count) | |
915 | { | |
916 | return __librbd_clone(ctx, src_snapname, dst_imagename, order, | |
917 | stripe_unit, stripe_count, false); | |
918 | } | |
919 | ||
920 | int | |
921 | __librbd_flatten(struct rbd_ctx *ctx) | |
922 | { | |
923 | int ret; | |
924 | ||
925 | ret = rbd_flatten(ctx->image); | |
926 | if (ret < 0) { | |
927 | prt("rbd_flatten failed\n"); | |
928 | return ret; | |
929 | } | |
930 | ||
931 | return librbd_verify_object_map(ctx); | |
932 | } | |
933 | ||
934 | int | |
935 | librbd_flatten(struct rbd_ctx *ctx) | |
936 | { | |
937 | return __librbd_flatten(ctx); | |
938 | } | |
939 | ||
940 | const struct rbd_operations librbd_operations = { | |
941 | librbd_open, | |
942 | librbd_close, | |
943 | librbd_read, | |
944 | librbd_write, | |
945 | librbd_flush, | |
946 | librbd_discard, | |
947 | librbd_get_size, | |
948 | librbd_resize, | |
949 | librbd_clone, | |
950 | librbd_flatten, | |
951 | librbd_writesame, | |
c07f9fc5 | 952 | librbd_compare_and_write, |
7c673cae FG |
953 | }; |
954 | ||
11fdf7f2 | 955 | #if defined(WITH_KRBD) |
7c673cae FG |
956 | int |
957 | krbd_open(const char *name, struct rbd_ctx *ctx) | |
958 | { | |
959 | char *devnode; | |
960 | int fd; | |
961 | int ret; | |
962 | ||
963 | ret = __librbd_open(name, ctx); | |
964 | if (ret < 0) | |
965 | return ret; | |
966 | ||
11fdf7f2 | 967 | ret = krbd_map(krbd, pool, "", name, "", "", &devnode); |
7c673cae FG |
968 | if (ret < 0) { |
969 | prt("krbd_map(%s) failed\n", name); | |
970 | return ret; | |
971 | } | |
972 | ||
973 | fd = open(devnode, O_RDWR | o_direct); | |
974 | if (fd < 0) { | |
975 | ret = -errno; | |
976 | prt("open(%s) failed\n", devnode); | |
977 | return ret; | |
978 | } | |
979 | ||
980 | ctx->krbd_name = devnode; | |
981 | ctx->krbd_fd = fd; | |
982 | ||
983 | return 0; | |
984 | } | |
985 | ||
986 | int | |
987 | krbd_close(struct rbd_ctx *ctx) | |
988 | { | |
989 | int ret; | |
990 | ||
11fdf7f2 | 991 | ceph_assert(ctx->krbd_name && ctx->krbd_fd >= 0); |
7c673cae FG |
992 | |
993 | if (close(ctx->krbd_fd) < 0) { | |
994 | ret = -errno; | |
995 | prt("close(%s) failed\n", ctx->krbd_name); | |
996 | return ret; | |
997 | } | |
998 | ||
999 | ret = krbd_unmap(krbd, ctx->krbd_name, ""); | |
1000 | if (ret < 0) { | |
1001 | prt("krbd_unmap(%s) failed\n", ctx->krbd_name); | |
1002 | return ret; | |
1003 | } | |
1004 | ||
1005 | free((void *)ctx->krbd_name); | |
1006 | ||
1007 | ctx->krbd_name = NULL; | |
1008 | ctx->krbd_fd = -1; | |
1009 | ||
1010 | return __librbd_close(ctx); | |
1011 | } | |
11fdf7f2 | 1012 | #endif // WITH_KRBD |
7c673cae | 1013 | |
11fdf7f2 | 1014 | #if defined(__linux__) |
7c673cae FG |
1015 | ssize_t |
1016 | krbd_read(struct rbd_ctx *ctx, uint64_t off, size_t len, char *buf) | |
1017 | { | |
1018 | ssize_t n; | |
1019 | ||
1020 | n = pread(ctx->krbd_fd, buf, len, off); | |
1021 | if (n < 0) { | |
1022 | n = -errno; | |
1023 | prt("pread(%llu, %zu) failed\n", off, len); | |
1024 | return n; | |
1025 | } | |
1026 | ||
1027 | return n; | |
1028 | } | |
1029 | ||
1030 | ssize_t | |
1031 | krbd_write(struct rbd_ctx *ctx, uint64_t off, size_t len, const char *buf) | |
1032 | { | |
1033 | ssize_t n; | |
1034 | ||
1035 | n = pwrite(ctx->krbd_fd, buf, len, off); | |
1036 | if (n < 0) { | |
1037 | n = -errno; | |
1038 | prt("pwrite(%llu, %zu) failed\n", off, len); | |
1039 | return n; | |
1040 | } | |
1041 | ||
1042 | return n; | |
1043 | } | |
1044 | ||
1045 | int | |
1046 | __krbd_flush(struct rbd_ctx *ctx, bool invalidate) | |
1047 | { | |
1048 | int ret; | |
1049 | ||
1050 | if (o_direct) | |
1051 | return 0; | |
1052 | ||
1053 | /* | |
1054 | * BLKFLSBUF will sync the filesystem on top of the device (we | |
1055 | * don't care about that here, since we write directly to it), | |
1056 | * write out any dirty buffers and invalidate the buffer cache. | |
1057 | * It won't do a hardware cache flush. | |
1058 | * | |
1059 | * fsync() will write out any dirty buffers and do a hardware | |
1060 | * cache flush (which we don't care about either, because for | |
1061 | * krbd it's a noop). It won't try to empty the buffer cache | |
1062 | * nor poke the filesystem before writing out. | |
1063 | * | |
1064 | * Given that, for our purposes, fsync is a flush, while | |
1065 | * BLKFLSBUF is a flush+invalidate. | |
1066 | */ | |
1067 | if (invalidate) | |
1068 | ret = ioctl(ctx->krbd_fd, BLKFLSBUF, NULL); | |
1069 | else | |
1070 | ret = fsync(ctx->krbd_fd); | |
1071 | if (ret < 0) { | |
1072 | ret = -errno; | |
1073 | prt("%s failed\n", invalidate ? "BLKFLSBUF" : "fsync"); | |
1074 | return ret; | |
1075 | } | |
1076 | ||
1077 | return 0; | |
1078 | } | |
1079 | ||
1080 | int | |
1081 | krbd_flush(struct rbd_ctx *ctx) | |
1082 | { | |
1083 | return __krbd_flush(ctx, false); | |
1084 | } | |
1085 | ||
1086 | int | |
1087 | krbd_discard(struct rbd_ctx *ctx, uint64_t off, uint64_t len) | |
1088 | { | |
1089 | uint64_t range[2] = { off, len }; | |
1090 | int ret; | |
1091 | ||
1092 | /* | |
a8e16298 | 1093 | * BLKZEROOUT goes straight to disk and doesn't do anything |
7c673cae FG |
1094 | * about dirty buffers. This means we need to flush so that |
1095 | * | |
1096 | * write 0..3M | |
1097 | * discard 1..2M | |
1098 | * | |
1099 | * results in "data 0000 data" rather than "data data data" on | |
1100 | * disk and invalidate so that | |
1101 | * | |
1102 | * discard 1..2M | |
1103 | * read 0..3M | |
1104 | * | |
1105 | * returns "data 0000 data" rather than "data data data" in | |
1106 | * case 1..2M was cached. | |
a8e16298 TL |
1107 | * |
1108 | * Note: These cache coherency issues are supposed to be fixed | |
1109 | * in recent kernels. | |
7c673cae FG |
1110 | */ |
1111 | ret = __krbd_flush(ctx, true); | |
1112 | if (ret < 0) | |
1113 | return ret; | |
1114 | ||
1115 | /* | |
a8e16298 | 1116 | * off and len must be 512-byte aligned, otherwise BLKZEROOUT |
7c673cae FG |
1117 | * will fail with -EINVAL. This means that -K (enable krbd |
1118 | * mode) requires -h 512 or similar. | |
1119 | */ | |
a8e16298 | 1120 | if (ioctl(ctx->krbd_fd, BLKZEROOUT, &range) < 0) { |
7c673cae | 1121 | ret = -errno; |
a8e16298 | 1122 | prt("BLKZEROOUT(%llu, %llu) failed\n", off, len); |
7c673cae FG |
1123 | return ret; |
1124 | } | |
1125 | ||
1126 | return 0; | |
1127 | } | |
1128 | ||
1129 | int | |
1130 | krbd_get_size(struct rbd_ctx *ctx, uint64_t *size) | |
1131 | { | |
1132 | uint64_t bytes; | |
1133 | ||
1134 | if (ioctl(ctx->krbd_fd, BLKGETSIZE64, &bytes) < 0) { | |
1135 | int ret = -errno; | |
1136 | prt("BLKGETSIZE64 failed\n"); | |
1137 | return ret; | |
1138 | } | |
1139 | ||
1140 | *size = bytes; | |
1141 | ||
1142 | return 0; | |
1143 | } | |
1144 | ||
1145 | int | |
1146 | krbd_resize(struct rbd_ctx *ctx, uint64_t size) | |
1147 | { | |
1148 | int ret; | |
1149 | ||
11fdf7f2 | 1150 | ceph_assert(size % truncbdy == 0); |
7c673cae FG |
1151 | |
1152 | /* | |
1153 | * When krbd detects a size change, it calls revalidate_disk(), | |
1154 | * which ends up calling invalidate_bdev(), which invalidates | |
1155 | * clean pages and does nothing about dirty pages beyond the | |
1156 | * new size. The preceding cache flush makes sure those pages | |
1157 | * are invalidated, which is what we need on shrink so that | |
1158 | * | |
1159 | * write 0..1M | |
1160 | * resize 0 | |
1161 | * resize 2M | |
1162 | * read 0..2M | |
1163 | * | |
1164 | * returns "0000 0000" rather than "data 0000". | |
1165 | */ | |
1166 | ret = __krbd_flush(ctx, false); | |
1167 | if (ret < 0) | |
1168 | return ret; | |
1169 | ||
1170 | return __librbd_resize(ctx, size); | |
1171 | } | |
1172 | ||
1173 | int | |
1174 | krbd_clone(struct rbd_ctx *ctx, const char *src_snapname, | |
1175 | const char *dst_imagename, int *order, int stripe_unit, | |
1176 | int stripe_count) | |
1177 | { | |
1178 | int ret; | |
1179 | ||
1180 | ret = __krbd_flush(ctx, false); | |
1181 | if (ret < 0) | |
1182 | return ret; | |
1183 | ||
1184 | return __librbd_clone(ctx, src_snapname, dst_imagename, order, | |
1185 | stripe_unit, stripe_count, true); | |
1186 | } | |
1187 | ||
1188 | int | |
1189 | krbd_flatten(struct rbd_ctx *ctx) | |
1190 | { | |
1191 | int ret; | |
1192 | ||
1193 | ret = __krbd_flush(ctx, false); | |
1194 | if (ret < 0) | |
1195 | return ret; | |
1196 | ||
1197 | return __librbd_flatten(ctx); | |
1198 | } | |
11fdf7f2 | 1199 | #endif // __linux__ |
7c673cae | 1200 | |
11fdf7f2 | 1201 | #if defined(WITH_KRBD) |
7c673cae FG |
1202 | const struct rbd_operations krbd_operations = { |
1203 | krbd_open, | |
1204 | krbd_close, | |
1205 | krbd_read, | |
1206 | krbd_write, | |
1207 | krbd_flush, | |
1208 | krbd_discard, | |
1209 | krbd_get_size, | |
1210 | krbd_resize, | |
1211 | krbd_clone, | |
1212 | krbd_flatten, | |
1213 | NULL, | |
1214 | }; | |
11fdf7f2 | 1215 | #endif // WITH_KRBD |
7c673cae | 1216 | |
11fdf7f2 | 1217 | #if defined(__linux__) |
7c673cae FG |
1218 | int |
1219 | nbd_open(const char *name, struct rbd_ctx *ctx) | |
1220 | { | |
1221 | int r; | |
1222 | int fd; | |
1223 | char dev[4096]; | |
1224 | char *devnode; | |
1225 | ||
1226 | SubProcess process("rbd-nbd", SubProcess::KEEP, SubProcess::PIPE, | |
1227 | SubProcess::KEEP); | |
1228 | process.add_cmd_arg("map"); | |
eafe8130 | 1229 | process.add_cmd_arg("--timeout=600"); |
7c673cae FG |
1230 | std::string img; |
1231 | img.append(pool); | |
1232 | img.append("/"); | |
1233 | img.append(name); | |
1234 | process.add_cmd_arg(img.c_str()); | |
1235 | ||
1236 | r = __librbd_open(name, ctx); | |
1237 | if (r < 0) | |
1238 | return r; | |
1239 | ||
1240 | r = process.spawn(); | |
1241 | if (r < 0) { | |
1242 | prt("nbd_open failed to run rbd-nbd error: %s\n", process.err().c_str()); | |
1243 | return r; | |
1244 | } | |
1245 | r = safe_read(process.get_stdout(), dev, sizeof(dev)); | |
1246 | if (r < 0) { | |
1247 | prt("nbd_open failed to get nbd device path\n"); | |
1248 | return r; | |
1249 | } | |
1250 | for (int i = 0; i < r; ++i) | |
1251 | if (dev[i] == 10 || dev[i] == 13) | |
1252 | dev[i] = 0; | |
1253 | dev[r] = 0; | |
1254 | r = process.join(); | |
1255 | if (r) { | |
1256 | prt("rbd-nbd failed with error: %s", process.err().c_str()); | |
1257 | return -EINVAL; | |
1258 | } | |
1259 | ||
1260 | devnode = strdup(dev); | |
1261 | if (!devnode) | |
1262 | return -ENOMEM; | |
1263 | ||
1264 | fd = open(devnode, O_RDWR | o_direct); | |
1265 | if (fd < 0) { | |
1266 | r = -errno; | |
1267 | prt("open(%s) failed\n", devnode); | |
1268 | return r; | |
1269 | } | |
1270 | ||
1271 | ctx->krbd_name = devnode; | |
1272 | ctx->krbd_fd = fd; | |
1273 | ||
1274 | return 0; | |
1275 | } | |
1276 | ||
1277 | int | |
1278 | nbd_close(struct rbd_ctx *ctx) | |
1279 | { | |
1280 | int r; | |
1281 | ||
11fdf7f2 | 1282 | ceph_assert(ctx->krbd_name && ctx->krbd_fd >= 0); |
7c673cae FG |
1283 | |
1284 | if (close(ctx->krbd_fd) < 0) { | |
1285 | r = -errno; | |
1286 | prt("close(%s) failed\n", ctx->krbd_name); | |
1287 | return r; | |
1288 | } | |
1289 | ||
1290 | SubProcess process("rbd-nbd"); | |
1291 | process.add_cmd_arg("unmap"); | |
1292 | process.add_cmd_arg(ctx->krbd_name); | |
1293 | ||
1294 | r = process.spawn(); | |
1295 | if (r < 0) { | |
1296 | prt("nbd_close failed to run rbd-nbd error: %s\n", process.err().c_str()); | |
1297 | return r; | |
1298 | } | |
1299 | r = process.join(); | |
1300 | if (r) { | |
1301 | prt("rbd-nbd failed with error: %d", process.err().c_str()); | |
1302 | return -EINVAL; | |
1303 | } | |
1304 | ||
1305 | free((void *)ctx->krbd_name); | |
1306 | ||
1307 | ctx->krbd_name = NULL; | |
1308 | ctx->krbd_fd = -1; | |
1309 | ||
1310 | return __librbd_close(ctx); | |
1311 | } | |
1312 | ||
1313 | int | |
1314 | nbd_clone(struct rbd_ctx *ctx, const char *src_snapname, | |
1315 | const char *dst_imagename, int *order, int stripe_unit, | |
1316 | int stripe_count) | |
1317 | { | |
1318 | int ret; | |
1319 | ||
1320 | ret = __krbd_flush(ctx, false); | |
1321 | if (ret < 0) | |
1322 | return ret; | |
1323 | ||
1324 | return __librbd_clone(ctx, src_snapname, dst_imagename, order, | |
1325 | stripe_unit, stripe_count, false); | |
1326 | } | |
1327 | ||
1328 | const struct rbd_operations nbd_operations = { | |
1329 | nbd_open, | |
1330 | nbd_close, | |
1331 | krbd_read, | |
1332 | krbd_write, | |
1333 | krbd_flush, | |
1334 | krbd_discard, | |
1335 | krbd_get_size, | |
1336 | krbd_resize, | |
1337 | nbd_clone, | |
1338 | krbd_flatten, | |
1339 | NULL, | |
1340 | }; | |
11fdf7f2 TL |
1341 | #endif // __linux__ |
1342 | ||
1343 | #if defined(__FreeBSD__) | |
1344 | int | |
1345 | ggate_open(const char *name, struct rbd_ctx *ctx) | |
1346 | { | |
1347 | int r; | |
1348 | int fd; | |
1349 | char dev[4096]; | |
1350 | char *devnode; | |
1351 | ||
1352 | SubProcess process("rbd-ggate", SubProcess::KEEP, SubProcess::PIPE, | |
1353 | SubProcess::KEEP); | |
1354 | process.add_cmd_arg("map"); | |
1355 | std::string img; | |
1356 | img.append(pool); | |
1357 | img.append("/"); | |
1358 | img.append(name); | |
1359 | process.add_cmd_arg(img.c_str()); | |
1360 | ||
1361 | r = __librbd_open(name, ctx); | |
1362 | if (r < 0) { | |
1363 | return r; | |
1364 | } | |
1365 | ||
1366 | r = process.spawn(); | |
1367 | if (r < 0) { | |
1368 | prt("ggate_open failed to run rbd-ggate: %s\n", | |
1369 | process.err().c_str()); | |
1370 | return r; | |
1371 | } | |
1372 | r = safe_read(process.get_stdout(), dev, sizeof(dev)); | |
1373 | if (r < 0) { | |
1374 | prt("ggate_open failed to get ggate device path\n"); | |
1375 | return r; | |
1376 | } | |
1377 | for (int i = 0; i < r; ++i) { | |
1378 | if (dev[i] == '\r' || dev[i] == '\n') { | |
1379 | dev[i] = 0; | |
1380 | } | |
1381 | } | |
1382 | dev[r] = 0; | |
1383 | r = process.join(); | |
1384 | if (r) { | |
1385 | prt("rbd-ggate failed with error: %s", process.err().c_str()); | |
1386 | return -EINVAL; | |
1387 | } | |
1388 | ||
1389 | devnode = strdup(dev); | |
1390 | if (!devnode) { | |
1391 | return -ENOMEM; | |
1392 | } | |
1393 | ||
1394 | for (int i = 0; i < 100; i++) { | |
1395 | fd = open(devnode, O_RDWR | o_direct); | |
1396 | if (fd >= 0 || errno != ENOENT) { | |
1397 | break; | |
1398 | } | |
1399 | usleep(100000); | |
1400 | } | |
1401 | if (fd < 0) { | |
1402 | r = -errno; | |
1403 | prt("open(%s) failed\n", devnode); | |
1404 | return r; | |
1405 | } | |
1406 | ||
1407 | ctx->krbd_name = devnode; | |
1408 | ctx->krbd_fd = fd; | |
1409 | ||
1410 | return 0; | |
1411 | } | |
1412 | ||
1413 | int | |
1414 | ggate_close(struct rbd_ctx *ctx) | |
1415 | { | |
1416 | int r; | |
1417 | ||
1418 | ceph_assert(ctx->krbd_name && ctx->krbd_fd >= 0); | |
1419 | ||
1420 | if (close(ctx->krbd_fd) < 0) { | |
1421 | r = -errno; | |
1422 | prt("close(%s) failed\n", ctx->krbd_name); | |
1423 | return r; | |
1424 | } | |
1425 | ||
1426 | SubProcess process("rbd-ggate"); | |
1427 | process.add_cmd_arg("unmap"); | |
1428 | process.add_cmd_arg(ctx->krbd_name); | |
1429 | ||
1430 | r = process.spawn(); | |
1431 | if (r < 0) { | |
1432 | prt("ggate_close failed to run rbd-nbd: %s\n", | |
1433 | process.err().c_str()); | |
1434 | return r; | |
1435 | } | |
1436 | r = process.join(); | |
1437 | if (r) { | |
1438 | prt("rbd-ggate failed with error: %d", process.err().c_str()); | |
1439 | return -EINVAL; | |
1440 | } | |
1441 | ||
1442 | free((void *)ctx->krbd_name); | |
1443 | ||
1444 | ctx->krbd_name = NULL; | |
1445 | ctx->krbd_fd = -1; | |
1446 | ||
1447 | return __librbd_close(ctx); | |
1448 | } | |
1449 | ||
1450 | ssize_t | |
1451 | ggate_read(struct rbd_ctx *ctx, uint64_t off, size_t len, char *buf) | |
1452 | { | |
1453 | ssize_t n; | |
1454 | ||
1455 | n = pread(ctx->krbd_fd, buf, len, off); | |
1456 | if (n < 0) { | |
1457 | n = -errno; | |
1458 | prt("pread(%llu, %zu) failed\n", off, len); | |
1459 | return n; | |
1460 | } | |
1461 | ||
1462 | return n; | |
1463 | } | |
1464 | ||
1465 | ssize_t | |
1466 | ggate_write(struct rbd_ctx *ctx, uint64_t off, size_t len, const char *buf) | |
1467 | { | |
1468 | ssize_t n; | |
1469 | ||
1470 | n = pwrite(ctx->krbd_fd, buf, len, off); | |
1471 | if (n < 0) { | |
1472 | n = -errno; | |
1473 | prt("pwrite(%llu, %zu) failed\n", off, len); | |
1474 | return n; | |
1475 | } | |
1476 | ||
1477 | return n; | |
1478 | } | |
1479 | ||
1480 | int | |
1481 | __ggate_flush(struct rbd_ctx *ctx, bool invalidate) | |
1482 | { | |
1483 | int ret; | |
1484 | ||
1485 | if (o_direct) { | |
1486 | return 0; | |
1487 | } | |
1488 | ||
1489 | if (invalidate) { | |
1490 | ret = ioctl(ctx->krbd_fd, DIOCGFLUSH, NULL); | |
1491 | } else { | |
1492 | ret = fsync(ctx->krbd_fd); | |
1493 | } | |
1494 | if (ret < 0) { | |
1495 | ret = -errno; | |
1496 | prt("%s failed\n", invalidate ? "DIOCGFLUSH" : "fsync"); | |
1497 | return ret; | |
1498 | } | |
1499 | ||
1500 | return 0; | |
1501 | } | |
1502 | ||
1503 | int | |
1504 | ggate_flush(struct rbd_ctx *ctx) | |
1505 | { | |
1506 | return __ggate_flush(ctx, false); | |
1507 | } | |
1508 | ||
1509 | int | |
1510 | ggate_discard(struct rbd_ctx *ctx, uint64_t off, uint64_t len) | |
1511 | { | |
1512 | off_t range[2] = {static_cast<off_t>(off), static_cast<off_t>(len)}; | |
1513 | int ret; | |
1514 | ||
1515 | ret = __ggate_flush(ctx, true); | |
1516 | if (ret < 0) { | |
1517 | return ret; | |
1518 | } | |
1519 | ||
1520 | if (ioctl(ctx->krbd_fd, DIOCGDELETE, &range) < 0) { | |
1521 | ret = -errno; | |
1522 | prt("DIOCGDELETE(%llu, %llu) failed\n", off, len); | |
1523 | return ret; | |
1524 | } | |
1525 | ||
1526 | return 0; | |
1527 | } | |
1528 | ||
1529 | int | |
1530 | ggate_get_size(struct rbd_ctx *ctx, uint64_t *size) | |
1531 | { | |
1532 | off_t bytes; | |
1533 | ||
1534 | if (ioctl(ctx->krbd_fd, DIOCGMEDIASIZE, &bytes) < 0) { | |
1535 | int ret = -errno; | |
1536 | prt("DIOCGMEDIASIZE failed\n"); | |
1537 | return ret; | |
1538 | } | |
1539 | ||
1540 | *size = bytes; | |
1541 | ||
1542 | return 0; | |
1543 | } | |
1544 | ||
1545 | int | |
1546 | ggate_resize(struct rbd_ctx *ctx, uint64_t size) | |
1547 | { | |
1548 | int ret; | |
1549 | ||
1550 | ceph_assert(size % truncbdy == 0); | |
1551 | ||
1552 | ret = __ggate_flush(ctx, false); | |
1553 | if (ret < 0) { | |
1554 | return ret; | |
1555 | } | |
1556 | ||
1557 | return __librbd_resize(ctx, size); | |
1558 | } | |
1559 | ||
1560 | int | |
1561 | ggate_clone(struct rbd_ctx *ctx, const char *src_snapname, | |
1562 | const char *dst_imagename, int *order, int stripe_unit, | |
1563 | int stripe_count) | |
1564 | { | |
1565 | int ret; | |
1566 | ||
1567 | ret = __ggate_flush(ctx, false); | |
1568 | if (ret < 0) { | |
1569 | return ret; | |
1570 | } | |
1571 | ||
1572 | return __librbd_clone(ctx, src_snapname, dst_imagename, order, | |
1573 | stripe_unit, stripe_count, false); | |
1574 | } | |
1575 | ||
1576 | int | |
1577 | ggate_flatten(struct rbd_ctx *ctx) | |
1578 | { | |
1579 | int ret; | |
1580 | ||
1581 | ret = __ggate_flush(ctx, false); | |
1582 | if (ret < 0) { | |
1583 | return ret; | |
1584 | } | |
1585 | ||
1586 | return __librbd_flatten(ctx); | |
1587 | } | |
1588 | ||
1589 | const struct rbd_operations ggate_operations = { | |
1590 | ggate_open, | |
1591 | ggate_close, | |
1592 | ggate_read, | |
1593 | ggate_write, | |
1594 | ggate_flush, | |
1595 | ggate_discard, | |
1596 | ggate_get_size, | |
1597 | ggate_resize, | |
1598 | ggate_clone, | |
1599 | ggate_flatten, | |
1600 | NULL, | |
1601 | }; | |
1602 | #endif // __FreeBSD__ | |
7c673cae FG |
1603 | |
1604 | struct rbd_ctx ctx = RBD_CTX_INIT; | |
1605 | const struct rbd_operations *ops = &librbd_operations; | |
1606 | ||
1607 | static bool rbd_image_has_parent(struct rbd_ctx *ctx) | |
1608 | { | |
1609 | int ret; | |
11fdf7f2 TL |
1610 | rbd_linked_image_spec_t parent_image; |
1611 | rbd_snap_spec_t parent_snap; | |
7c673cae | 1612 | |
11fdf7f2 TL |
1613 | ret = rbd_get_parent(ctx->image, &parent_image, &parent_snap); |
1614 | if (ret < 0 && ret != -ENOENT) { | |
7c673cae FG |
1615 | prterrcode("rbd_get_parent_info", ret); |
1616 | exit(1); | |
1617 | } | |
11fdf7f2 TL |
1618 | rbd_linked_image_spec_cleanup(&parent_image); |
1619 | rbd_snap_spec_cleanup(&parent_snap); | |
7c673cae FG |
1620 | |
1621 | return !ret; | |
1622 | } | |
1623 | ||
1624 | /* | |
1625 | * fsx | |
1626 | */ | |
1627 | ||
1628 | void | |
1629 | log4(int operation, int arg0, int arg1, int arg2) | |
1630 | { | |
1631 | struct log_entry *le; | |
1632 | ||
1633 | le = &oplog[logptr]; | |
1634 | le->operation = operation; | |
1635 | if (closeopen) | |
1636 | le->operation = ~ le->operation; | |
1637 | le->args[0] = arg0; | |
1638 | le->args[1] = arg1; | |
1639 | le->args[2] = arg2; | |
1640 | logptr++; | |
1641 | logcount++; | |
1642 | if (logptr >= LOGSIZE) | |
1643 | logptr = 0; | |
1644 | } | |
1645 | ||
1646 | void | |
1647 | logdump(void) | |
1648 | { | |
1649 | int i, count, down; | |
1650 | struct log_entry *lp; | |
1651 | const char *falloc_type[3] = {"PAST_EOF", "EXTENDING", "INTERIOR"}; | |
1652 | ||
1653 | prt("LOG DUMP (%d total operations):\n", logcount); | |
1654 | if (logcount < LOGSIZE) { | |
1655 | i = 0; | |
1656 | count = logcount; | |
1657 | } else { | |
1658 | i = logptr; | |
1659 | count = LOGSIZE; | |
1660 | } | |
1661 | for ( ; count > 0; count--) { | |
1662 | int opnum; | |
1663 | ||
1664 | opnum = i+1 + (logcount/LOGSIZE)*LOGSIZE; | |
1665 | prt("%d(%3d mod 256): ", opnum, opnum%256); | |
1666 | lp = &oplog[i]; | |
1667 | if ((closeopen = lp->operation < 0)) | |
1668 | lp->operation = ~ lp->operation; | |
1669 | ||
1670 | switch (lp->operation) { | |
1671 | case OP_MAPREAD: | |
1672 | prt("MAPREAD 0x%x thru 0x%x\t(0x%x bytes)", | |
1673 | lp->args[0], lp->args[0] + lp->args[1] - 1, | |
1674 | lp->args[1]); | |
1675 | if (badoff >= lp->args[0] && badoff < | |
1676 | lp->args[0] + lp->args[1]) | |
1677 | prt("\t***RRRR***"); | |
1678 | break; | |
1679 | case OP_MAPWRITE: | |
1680 | prt("MAPWRITE 0x%x thru 0x%x\t(0x%x bytes)", | |
1681 | lp->args[0], lp->args[0] + lp->args[1] - 1, | |
1682 | lp->args[1]); | |
1683 | if (badoff >= lp->args[0] && badoff < | |
1684 | lp->args[0] + lp->args[1]) | |
1685 | prt("\t******WWWW"); | |
1686 | break; | |
1687 | case OP_READ: | |
1688 | prt("READ 0x%x thru 0x%x\t(0x%x bytes)", | |
1689 | lp->args[0], lp->args[0] + lp->args[1] - 1, | |
1690 | lp->args[1]); | |
1691 | if (badoff >= lp->args[0] && | |
1692 | badoff < lp->args[0] + lp->args[1]) | |
1693 | prt("\t***RRRR***"); | |
1694 | break; | |
1695 | case OP_WRITE: | |
1696 | prt("WRITE 0x%x thru 0x%x\t(0x%x bytes)", | |
1697 | lp->args[0], lp->args[0] + lp->args[1] - 1, | |
1698 | lp->args[1]); | |
1699 | if (lp->args[0] > lp->args[2]) | |
1700 | prt(" HOLE"); | |
1701 | else if (lp->args[0] + lp->args[1] > lp->args[2]) | |
1702 | prt(" EXTEND"); | |
1703 | if ((badoff >= lp->args[0] || badoff >=lp->args[2]) && | |
1704 | badoff < lp->args[0] + lp->args[1]) | |
1705 | prt("\t***WWWW"); | |
1706 | break; | |
1707 | case OP_TRUNCATE: | |
1708 | down = lp->args[0] < lp->args[1]; | |
1709 | prt("TRUNCATE %s\tfrom 0x%x to 0x%x", | |
1710 | down ? "DOWN" : "UP", lp->args[1], lp->args[0]); | |
1711 | if (badoff >= lp->args[!down] && | |
1712 | badoff < lp->args[!!down]) | |
1713 | prt("\t******WWWW"); | |
1714 | break; | |
1715 | case OP_FALLOCATE: | |
1716 | /* 0: offset 1: length 2: where alloced */ | |
1717 | prt("FALLOC 0x%x thru 0x%x\t(0x%x bytes) %s", | |
1718 | lp->args[0], lp->args[0] + lp->args[1], | |
1719 | lp->args[1], falloc_type[lp->args[2]]); | |
1720 | if (badoff >= lp->args[0] && | |
1721 | badoff < lp->args[0] + lp->args[1]) | |
1722 | prt("\t******FFFF"); | |
1723 | break; | |
1724 | case OP_PUNCH_HOLE: | |
1725 | prt("PUNCH 0x%x thru 0x%x\t(0x%x bytes)", | |
1726 | lp->args[0], lp->args[0] + lp->args[1] - 1, | |
1727 | lp->args[1]); | |
1728 | if (badoff >= lp->args[0] && badoff < | |
1729 | lp->args[0] + lp->args[1]) | |
1730 | prt("\t******PPPP"); | |
1731 | break; | |
1732 | case OP_WRITESAME: | |
1733 | prt("WRITESAME 0x%x thru 0x%x\t(0x%x bytes) data_size 0x%x", | |
1734 | lp->args[0], lp->args[0] + lp->args[1] - 1, | |
1735 | lp->args[1], lp->args[2]); | |
1736 | if (badoff >= lp->args[0] && | |
1737 | badoff < lp->args[0] + lp->args[1]) | |
1738 | prt("\t***WSWSWSWS"); | |
1739 | break; | |
c07f9fc5 FG |
1740 | case OP_COMPARE_AND_WRITE: |
1741 | prt("COMPARE_AND_WRITE 0x%x thru 0x%x\t(0x%x bytes)", | |
1742 | lp->args[0], lp->args[0] + lp->args[1] - 1, | |
1743 | lp->args[1]); | |
1744 | if (lp->args[0] > lp->args[2]) | |
1745 | prt(" HOLE"); | |
1746 | else if (lp->args[0] + lp->args[1] > lp->args[2]) | |
1747 | prt(" EXTEND"); | |
1748 | if ((badoff >= lp->args[0] || badoff >=lp->args[2]) && | |
1749 | badoff < lp->args[0] + lp->args[1]) | |
1750 | prt("\t***WWWW"); | |
1751 | break; | |
7c673cae FG |
1752 | case OP_CLONE: |
1753 | prt("CLONE"); | |
1754 | break; | |
1755 | case OP_FLATTEN: | |
1756 | prt("FLATTEN"); | |
1757 | break; | |
1758 | case OP_SKIPPED: | |
1759 | prt("SKIPPED (no operation)"); | |
1760 | break; | |
1761 | default: | |
1762 | prt("BOGUS LOG ENTRY (operation code = %d)!", | |
1763 | lp->operation); | |
1764 | } | |
1765 | if (closeopen) | |
1766 | prt("\n\t\tCLOSE/OPEN"); | |
1767 | prt("\n"); | |
1768 | i++; | |
1769 | if (i == LOGSIZE) | |
1770 | i = 0; | |
1771 | } | |
1772 | } | |
1773 | ||
1774 | void | |
1775 | save_buffer(char *buffer, off_t bufferlength, int fd) | |
1776 | { | |
1777 | off_t ret; | |
1778 | ssize_t byteswritten; | |
1779 | ||
1780 | if (fd <= 0 || bufferlength == 0) | |
1781 | return; | |
1782 | ||
1783 | if (bufferlength > SSIZE_MAX) { | |
1784 | prt("fsx flaw: overflow in save_buffer\n"); | |
1785 | exit(67); | |
1786 | } | |
1787 | ||
1788 | ret = lseek(fd, (off_t)0, SEEK_SET); | |
1789 | if (ret == (off_t)-1) | |
1790 | prterr("save_buffer: lseek 0"); | |
1791 | ||
1792 | byteswritten = write(fd, buffer, (size_t)bufferlength); | |
1793 | if (byteswritten != bufferlength) { | |
1794 | if (byteswritten == -1) | |
1795 | prterr("save_buffer write"); | |
1796 | else | |
1797 | warn("save_buffer: short write, 0x%x bytes instead of 0x%llx\n", | |
1798 | (unsigned)byteswritten, | |
1799 | (unsigned long long)bufferlength); | |
1800 | } | |
1801 | } | |
1802 | ||
1803 | ||
1804 | void | |
1805 | report_failure(int status) | |
1806 | { | |
1807 | logdump(); | |
1808 | ||
1809 | if (fsxgoodfd) { | |
1810 | if (good_buf) { | |
1811 | save_buffer(good_buf, file_size, fsxgoodfd); | |
1812 | prt("Correct content saved for comparison\n"); | |
1813 | prt("(maybe hexdump \"%s\" vs \"%s.fsxgood\")\n", | |
1814 | iname, iname); | |
1815 | } | |
1816 | close(fsxgoodfd); | |
1817 | } | |
1818 | sleep(3); // so the log can flush to disk. KLUDGEY! | |
1819 | exit(status); | |
1820 | } | |
1821 | ||
1822 | #define short_at(cp) ((unsigned short)((*((unsigned char *)(cp)) << 8) | \ | |
1823 | *(((unsigned char *)(cp)) + 1))) | |
1824 | ||
1825 | int | |
1826 | fsxcmp(char *good_buf, char *temp_buf, unsigned size) | |
1827 | { | |
1828 | if (!skip_partial_discard) { | |
1829 | return memcmp(good_buf, temp_buf, size); | |
1830 | } | |
1831 | ||
1832 | for (unsigned i = 0; i < size; i++) { | |
1833 | if (good_buf[i] != temp_buf[i] && good_buf[i] != 0) { | |
1834 | return good_buf[i] - temp_buf[i]; | |
1835 | } | |
1836 | } | |
1837 | return 0; | |
1838 | } | |
1839 | ||
1840 | void | |
1841 | check_buffers(char *good_buf, char *temp_buf, unsigned offset, unsigned size) | |
1842 | { | |
1843 | if (fsxcmp(good_buf + offset, temp_buf, size) != 0) { | |
1844 | unsigned i = 0; | |
1845 | unsigned n = 0; | |
1846 | ||
1847 | prt("READ BAD DATA: offset = 0x%x, size = 0x%x, fname = %s\n", | |
1848 | offset, size, iname); | |
1849 | prt("OFFSET\tGOOD\tBAD\tRANGE\n"); | |
1850 | while (size > 0) { | |
1851 | unsigned char c = good_buf[offset]; | |
1852 | unsigned char t = temp_buf[i]; | |
1853 | if (c != t) { | |
1854 | if (n < 16) { | |
1855 | unsigned bad = short_at(&temp_buf[i]); | |
1856 | prt("0x%5x\t0x%04x\t0x%04x", offset, | |
1857 | short_at(&good_buf[offset]), bad); | |
1858 | unsigned op = temp_buf[(offset & 1) ? i+1 : i]; | |
1859 | prt("\t0x%5x\n", n); | |
1860 | if (op) | |
1861 | prt("operation# (mod 256) for " | |
1862 | "the bad data may be %u\n", | |
1863 | ((unsigned)op & 0xff)); | |
1864 | else | |
1865 | prt("operation# (mod 256) for " | |
1866 | "the bad data unknown, check" | |
1867 | " HOLE and EXTEND ops\n"); | |
1868 | } | |
1869 | n++; | |
1870 | badoff = offset; | |
1871 | } | |
1872 | offset++; | |
1873 | i++; | |
1874 | size--; | |
1875 | } | |
1876 | report_failure(110); | |
1877 | } | |
1878 | } | |
1879 | ||
1880 | ||
1881 | void | |
1882 | check_size(void) | |
1883 | { | |
1884 | uint64_t size; | |
1885 | int ret; | |
1886 | ||
1887 | ret = ops->get_size(&ctx, &size); | |
1888 | if (ret < 0) | |
1889 | prterrcode("check_size: ops->get_size", ret); | |
1890 | ||
1891 | if ((uint64_t)file_size != size) { | |
1892 | prt("Size error: expected 0x%llx stat 0x%llx\n", | |
1893 | (unsigned long long)file_size, | |
1894 | (unsigned long long)size); | |
1895 | report_failure(120); | |
1896 | } | |
1897 | } | |
1898 | ||
1899 | #define TRUNC_HACK_SIZE (200ULL << 9) /* 512-byte aligned for krbd */ | |
1900 | ||
1901 | void | |
1902 | check_trunc_hack(void) | |
1903 | { | |
1904 | uint64_t size; | |
1905 | int ret; | |
1906 | ||
1907 | ret = ops->resize(&ctx, 0ULL); | |
1908 | if (ret < 0) | |
1909 | prterrcode("check_trunc_hack: ops->resize pre", ret); | |
1910 | ||
1911 | ret = ops->resize(&ctx, TRUNC_HACK_SIZE); | |
1912 | if (ret < 0) | |
1913 | prterrcode("check_trunc_hack: ops->resize actual", ret); | |
1914 | ||
1915 | ret = ops->get_size(&ctx, &size); | |
1916 | if (ret < 0) | |
1917 | prterrcode("check_trunc_hack: ops->get_size", ret); | |
1918 | ||
1919 | if (size != TRUNC_HACK_SIZE) { | |
1920 | prt("no extend on truncate! not posix!\n"); | |
1921 | exit(130); | |
1922 | } | |
1923 | ||
1924 | ret = ops->resize(&ctx, 0ULL); | |
1925 | if (ret < 0) | |
1926 | prterrcode("check_trunc_hack: ops->resize post", ret); | |
1927 | } | |
1928 | ||
1929 | int | |
1930 | create_image() | |
1931 | { | |
1932 | int r; | |
1933 | int order = 0; | |
1934 | char buf[32]; | |
11fdf7f2 TL |
1935 | char client_name[256]; |
1936 | ||
1937 | sprintf(client_name, "client.%s", client_id); | |
7c673cae | 1938 | |
11fdf7f2 | 1939 | r = rados_create2(&cluster, cluster_name, client_name, 0); |
7c673cae FG |
1940 | if (r < 0) { |
1941 | simple_err("Could not create cluster handle", r); | |
1942 | return r; | |
1943 | } | |
1944 | rados_conf_parse_env(cluster, NULL); | |
1945 | r = rados_conf_read_file(cluster, NULL); | |
1946 | if (r < 0) { | |
1947 | simple_err("Error reading ceph config file", r); | |
1948 | goto failed_shutdown; | |
1949 | } | |
1950 | r = rados_connect(cluster); | |
1951 | if (r < 0) { | |
1952 | simple_err("Error connecting to cluster", r); | |
1953 | goto failed_shutdown; | |
1954 | } | |
11fdf7f2 | 1955 | #if defined(WITH_KRBD) |
7c673cae FG |
1956 | r = krbd_create_from_context(rados_cct(cluster), &krbd); |
1957 | if (r < 0) { | |
1958 | simple_err("Could not create libkrbd handle", r); | |
1959 | goto failed_shutdown; | |
1960 | } | |
11fdf7f2 | 1961 | #endif |
7c673cae FG |
1962 | |
1963 | r = rados_pool_create(cluster, pool); | |
1964 | if (r < 0 && r != -EEXIST) { | |
1965 | simple_err("Error creating pool", r); | |
1966 | goto failed_krbd; | |
1967 | } | |
1968 | r = rados_ioctx_create(cluster, pool, &ioctx); | |
1969 | if (r < 0) { | |
1970 | simple_err("Error creating ioctx", r); | |
1971 | goto failed_krbd; | |
1972 | } | |
c07f9fc5 FG |
1973 | rados_application_enable(ioctx, "rbd", 1); |
1974 | ||
7c673cae | 1975 | if (clone_calls || journal_replay) { |
11fdf7f2 TL |
1976 | uint64_t features; |
1977 | r = get_features(&features); | |
1978 | if (r < 0) { | |
1979 | goto failed_open; | |
7c673cae | 1980 | } |
11fdf7f2 TL |
1981 | |
1982 | r = rbd_create2(ioctx, iname, file_size, features, &order); | |
7c673cae | 1983 | } else { |
11fdf7f2 | 1984 | r = rbd_create(ioctx, iname, file_size, &order); |
7c673cae FG |
1985 | } |
1986 | if (r < 0) { | |
1987 | simple_err("Error creating image", r); | |
1988 | goto failed_open; | |
1989 | } | |
1990 | ||
1991 | if (journal_replay) { | |
1992 | r = register_journal(ioctx, iname); | |
1993 | if (r < 0) { | |
1994 | goto failed_open; | |
1995 | } | |
1996 | } | |
1997 | ||
1998 | r = rados_conf_get(cluster, "rbd_skip_partial_discard", buf, | |
1999 | sizeof(buf)); | |
2000 | if (r < 0) { | |
2001 | simple_err("Could not get rbd_skip_partial_discard value", r); | |
2002 | goto failed_open; | |
2003 | } | |
2004 | skip_partial_discard = (strcmp(buf, "true") == 0); | |
2005 | ||
2006 | return 0; | |
2007 | ||
2008 | failed_open: | |
2009 | rados_ioctx_destroy(ioctx); | |
2010 | failed_krbd: | |
11fdf7f2 | 2011 | #if defined(WITH_KRBD) |
7c673cae | 2012 | krbd_destroy(krbd); |
11fdf7f2 | 2013 | #endif |
7c673cae FG |
2014 | failed_shutdown: |
2015 | rados_shutdown(cluster); | |
2016 | return r; | |
2017 | } | |
2018 | ||
2019 | void | |
2020 | doflush(unsigned offset, unsigned size) | |
2021 | { | |
2022 | int ret; | |
2023 | ||
2024 | if (o_direct) | |
2025 | return; | |
2026 | ||
2027 | ret = ops->flush(&ctx); | |
2028 | if (ret < 0) | |
2029 | prterrcode("doflush: ops->flush", ret); | |
2030 | } | |
2031 | ||
2032 | void | |
2033 | doread(unsigned offset, unsigned size) | |
2034 | { | |
2035 | int ret; | |
2036 | ||
2037 | offset -= offset % readbdy; | |
2038 | if (o_direct) | |
2039 | size -= size % readbdy; | |
2040 | if (size == 0) { | |
2041 | if (!quiet && testcalls > simulatedopcount && !o_direct) | |
2042 | prt("skipping zero size read\n"); | |
2043 | log4(OP_SKIPPED, OP_READ, offset, size); | |
2044 | return; | |
2045 | } | |
2046 | if (size + offset > file_size) { | |
2047 | if (!quiet && testcalls > simulatedopcount) | |
2048 | prt("skipping seek/read past end of file\n"); | |
2049 | log4(OP_SKIPPED, OP_READ, offset, size); | |
2050 | return; | |
2051 | } | |
2052 | ||
2053 | log4(OP_READ, offset, size, 0); | |
2054 | ||
2055 | if (testcalls <= simulatedopcount) | |
2056 | return; | |
2057 | ||
2058 | if (!quiet && | |
2059 | ((progressinterval && testcalls % progressinterval == 0) || | |
2060 | (debug && | |
2061 | (monitorstart == -1 || | |
2062 | (static_cast<long>(offset + size) > monitorstart && | |
2063 | (monitorend == -1 || | |
2064 | static_cast<long>(offset) <= monitorend)))))) | |
2065 | prt("%lu read\t0x%x thru\t0x%x\t(0x%x bytes)\n", testcalls, | |
2066 | offset, offset + size - 1, size); | |
2067 | ||
2068 | ret = ops->read(&ctx, offset, size, temp_buf); | |
2069 | if (ret != (int)size) { | |
2070 | if (ret < 0) | |
2071 | prterrcode("doread: ops->read", ret); | |
2072 | else | |
2073 | prt("short read: 0x%x bytes instead of 0x%x\n", | |
2074 | ret, size); | |
2075 | report_failure(141); | |
2076 | } | |
2077 | ||
2078 | check_buffers(good_buf, temp_buf, offset, size); | |
2079 | } | |
2080 | ||
2081 | ||
2082 | void | |
2083 | check_eofpage(char *s, unsigned offset, char *p, int size) | |
2084 | { | |
2085 | unsigned long last_page, should_be_zero; | |
2086 | ||
2087 | if (offset + size <= (file_size & ~page_mask)) | |
2088 | return; | |
2089 | /* | |
2090 | * we landed in the last page of the file | |
2091 | * test to make sure the VM system provided 0's | |
2092 | * beyond the true end of the file mapping | |
2093 | * (as required by mmap def in 1996 posix 1003.1) | |
2094 | */ | |
2095 | last_page = ((unsigned long)p + (offset & page_mask) + size) & ~page_mask; | |
2096 | ||
2097 | for (should_be_zero = last_page + (file_size & page_mask); | |
2098 | should_be_zero < last_page + page_size; | |
2099 | should_be_zero++) | |
2100 | if (*(char *)should_be_zero) { | |
2101 | prt("Mapped %s: non-zero data past EOF (0x%llx) page offset 0x%x is 0x%04x\n", | |
2102 | s, file_size - 1, should_be_zero & page_mask, | |
2103 | short_at(should_be_zero)); | |
2104 | report_failure(205); | |
2105 | } | |
2106 | } | |
2107 | ||
2108 | ||
2109 | void | |
2110 | gendata(char *original_buf, char *good_buf, unsigned offset, unsigned size) | |
2111 | { | |
2112 | while (size--) { | |
2113 | good_buf[offset] = testcalls % 256; | |
2114 | if (offset % 2) | |
2115 | good_buf[offset] += original_buf[offset]; | |
2116 | offset++; | |
2117 | } | |
2118 | } | |
2119 | ||
2120 | ||
2121 | void | |
2122 | dowrite(unsigned offset, unsigned size) | |
2123 | { | |
2124 | ssize_t ret; | |
2125 | off_t newsize; | |
2126 | ||
2127 | offset -= offset % writebdy; | |
2128 | if (o_direct) | |
2129 | size -= size % writebdy; | |
2130 | if (size == 0) { | |
2131 | if (!quiet && testcalls > simulatedopcount && !o_direct) | |
2132 | prt("skipping zero size write\n"); | |
2133 | log4(OP_SKIPPED, OP_WRITE, offset, size); | |
2134 | return; | |
2135 | } | |
2136 | ||
2137 | log4(OP_WRITE, offset, size, file_size); | |
2138 | ||
2139 | gendata(original_buf, good_buf, offset, size); | |
2140 | if (file_size < offset + size) { | |
2141 | newsize = ceil(((double)offset + size) / truncbdy) * truncbdy; | |
2142 | if (file_size < newsize) | |
2143 | memset(good_buf + file_size, '\0', newsize - file_size); | |
2144 | file_size = newsize; | |
2145 | if (lite) { | |
2146 | warn("Lite file size bug in fsx!"); | |
2147 | report_failure(149); | |
2148 | } | |
2149 | ret = ops->resize(&ctx, newsize); | |
2150 | if (ret < 0) { | |
2151 | prterrcode("dowrite: ops->resize", ret); | |
2152 | report_failure(150); | |
2153 | } | |
2154 | } | |
2155 | ||
2156 | if (testcalls <= simulatedopcount) | |
2157 | return; | |
2158 | ||
2159 | if (!quiet && | |
2160 | ((progressinterval && testcalls % progressinterval == 0) || | |
2161 | (debug && | |
2162 | (monitorstart == -1 || | |
2163 | (static_cast<long>(offset + size) > monitorstart && | |
2164 | (monitorend == -1 || | |
2165 | static_cast<long>(offset) <= monitorend)))))) | |
2166 | prt("%lu write\t0x%x thru\t0x%x\t(0x%x bytes)\n", testcalls, | |
2167 | offset, offset + size - 1, size); | |
2168 | ||
2169 | ret = ops->write(&ctx, offset, size, good_buf + offset); | |
2170 | if (ret != (ssize_t)size) { | |
2171 | if (ret < 0) | |
2172 | prterrcode("dowrite: ops->write", ret); | |
2173 | else | |
2174 | prt("short write: 0x%x bytes instead of 0x%x\n", | |
2175 | ret, size); | |
2176 | report_failure(151); | |
2177 | } | |
2178 | ||
2179 | if (flush_enabled) | |
2180 | doflush(offset, size); | |
2181 | } | |
2182 | ||
2183 | ||
2184 | void | |
2185 | dotruncate(unsigned size) | |
2186 | { | |
2187 | int oldsize = file_size; | |
2188 | int ret; | |
2189 | ||
2190 | size -= size % truncbdy; | |
2191 | if (size > biggest) { | |
2192 | biggest = size; | |
2193 | if (!quiet && testcalls > simulatedopcount) | |
2194 | prt("truncating to largest ever: 0x%x\n", size); | |
2195 | } | |
2196 | ||
2197 | log4(OP_TRUNCATE, size, (unsigned)file_size, 0); | |
2198 | ||
2199 | if (size > file_size) | |
2200 | memset(good_buf + file_size, '\0', size - file_size); | |
2201 | else if (size < file_size) | |
2202 | memset(good_buf + size, '\0', file_size - size); | |
2203 | file_size = size; | |
2204 | ||
2205 | if (testcalls <= simulatedopcount) | |
2206 | return; | |
2207 | ||
2208 | if ((progressinterval && testcalls % progressinterval == 0) || | |
2209 | (debug && (monitorstart == -1 || monitorend == -1 || | |
2210 | (long)size <= monitorend))) | |
2211 | prt("%lu trunc\tfrom 0x%x to 0x%x\n", testcalls, oldsize, size); | |
2212 | ||
2213 | ret = ops->resize(&ctx, size); | |
2214 | if (ret < 0) { | |
2215 | prterrcode("dotruncate: ops->resize", ret); | |
2216 | report_failure(160); | |
2217 | } | |
2218 | } | |
2219 | ||
2220 | void | |
2221 | do_punch_hole(unsigned offset, unsigned length) | |
2222 | { | |
2223 | unsigned end_offset; | |
2224 | int max_offset = 0; | |
2225 | int max_len = 0; | |
2226 | int ret; | |
2227 | ||
2228 | offset -= offset % holebdy; | |
2229 | length -= length % holebdy; | |
2230 | if (length == 0) { | |
2231 | if (!quiet && testcalls > simulatedopcount) | |
2232 | prt("skipping zero length punch hole\n"); | |
2233 | log4(OP_SKIPPED, OP_PUNCH_HOLE, offset, length); | |
2234 | return; | |
2235 | } | |
2236 | ||
2237 | if (file_size <= (loff_t)offset) { | |
2238 | if (!quiet && testcalls > simulatedopcount) | |
2239 | prt("skipping hole punch off the end of the file\n"); | |
2240 | log4(OP_SKIPPED, OP_PUNCH_HOLE, offset, length); | |
2241 | return; | |
2242 | } | |
2243 | ||
2244 | end_offset = offset + length; | |
2245 | ||
2246 | log4(OP_PUNCH_HOLE, offset, length, 0); | |
2247 | ||
2248 | if (testcalls <= simulatedopcount) | |
2249 | return; | |
2250 | ||
2251 | if ((progressinterval && testcalls % progressinterval == 0) || | |
2252 | (debug && (monitorstart == -1 || monitorend == -1 || | |
2253 | (long)end_offset <= monitorend))) { | |
2254 | prt("%lu punch\tfrom 0x%x to 0x%x, (0x%x bytes)\n", testcalls, | |
2255 | offset, offset+length, length); | |
2256 | } | |
2257 | ||
2258 | ret = ops->discard(&ctx, (unsigned long long)offset, | |
2259 | (unsigned long long)length); | |
2260 | if (ret < 0) { | |
2261 | prterrcode("do_punch_hole: ops->discard", ret); | |
2262 | report_failure(161); | |
2263 | } | |
2264 | ||
2265 | max_offset = offset < file_size ? offset : file_size; | |
2266 | max_len = max_offset + length <= file_size ? length : | |
2267 | file_size - max_offset; | |
2268 | memset(good_buf + max_offset, '\0', max_len); | |
2269 | } | |
2270 | ||
2271 | unsigned get_data_size(unsigned size) | |
2272 | { | |
2273 | unsigned i; | |
2274 | unsigned hint; | |
2275 | unsigned max = sqrt((double)size) + 1; | |
2276 | unsigned good = 1; | |
2277 | unsigned curr = good; | |
2278 | ||
2279 | hint = get_random() % max; | |
2280 | ||
2281 | for (i = 1; i < max && curr < hint; i++) { | |
2282 | if (size % i == 0) { | |
2283 | good = curr; | |
2284 | curr = i; | |
2285 | } | |
2286 | } | |
2287 | ||
2288 | if (curr == hint) | |
2289 | good = curr; | |
2290 | ||
2291 | return good; | |
2292 | } | |
2293 | ||
2294 | void | |
2295 | dowritesame(unsigned offset, unsigned size) | |
2296 | { | |
2297 | ssize_t ret; | |
2298 | off_t newsize; | |
2299 | unsigned buf_off; | |
2300 | unsigned data_size; | |
2301 | int n; | |
2302 | ||
2303 | offset -= offset % writebdy; | |
2304 | if (o_direct) | |
2305 | size -= size % writebdy; | |
2306 | if (size == 0) { | |
2307 | if (!quiet && testcalls > simulatedopcount && !o_direct) | |
2308 | prt("skipping zero size writesame\n"); | |
2309 | log4(OP_SKIPPED, OP_WRITESAME, offset, size); | |
2310 | return; | |
2311 | } | |
2312 | ||
2313 | data_size = get_data_size(size); | |
2314 | ||
2315 | log4(OP_WRITESAME, offset, size, data_size); | |
2316 | ||
2317 | gendata(original_buf, good_buf, offset, data_size); | |
2318 | if (file_size < offset + size) { | |
2319 | newsize = ceil(((double)offset + size) / truncbdy) * truncbdy; | |
2320 | if (file_size < newsize) | |
2321 | memset(good_buf + file_size, '\0', newsize - file_size); | |
2322 | file_size = newsize; | |
2323 | if (lite) { | |
2324 | warn("Lite file size bug in fsx!"); | |
2325 | report_failure(162); | |
2326 | } | |
2327 | ret = ops->resize(&ctx, newsize); | |
2328 | if (ret < 0) { | |
2329 | prterrcode("dowritesame: ops->resize", ret); | |
2330 | report_failure(163); | |
2331 | } | |
2332 | } | |
2333 | ||
2334 | for (n = size / data_size, buf_off = data_size; n > 1; n--) { | |
2335 | memcpy(good_buf + offset + buf_off, good_buf + offset, data_size); | |
2336 | buf_off += data_size; | |
2337 | } | |
2338 | ||
2339 | if (testcalls <= simulatedopcount) | |
2340 | return; | |
2341 | ||
2342 | if (!quiet && | |
2343 | ((progressinterval && testcalls % progressinterval == 0) || | |
2344 | (debug && | |
2345 | (monitorstart == -1 || | |
2346 | (static_cast<long>(offset + size) > monitorstart && | |
2347 | (monitorend == -1 || | |
2348 | static_cast<long>(offset) <= monitorend)))))) | |
2349 | prt("%lu writesame\t0x%x thru\t0x%x\tdata_size\t0x%x(0x%x bytes)\n", testcalls, | |
2350 | offset, offset + size - 1, data_size, size); | |
2351 | ||
2352 | ret = ops->writesame(&ctx, offset, size, good_buf + offset, data_size); | |
2353 | if (ret != (ssize_t)size) { | |
2354 | if (ret < 0) | |
2355 | prterrcode("dowritesame: ops->writesame", ret); | |
2356 | else | |
2357 | prt("short writesame: 0x%x bytes instead of 0x%x\n", | |
2358 | ret, size); | |
2359 | report_failure(164); | |
2360 | } | |
2361 | ||
2362 | if (flush_enabled) | |
2363 | doflush(offset, size); | |
2364 | } | |
2365 | ||
c07f9fc5 FG |
2366 | void |
2367 | docompareandwrite(unsigned offset, unsigned size) | |
2368 | { | |
2369 | int ret; | |
2370 | ||
b32b8144 FG |
2371 | if (skip_partial_discard) { |
2372 | if (!quiet && testcalls > simulatedopcount) | |
2373 | prt("compare and write disabled\n"); | |
2374 | log4(OP_SKIPPED, OP_COMPARE_AND_WRITE, offset, size); | |
2375 | return; | |
2376 | } | |
2377 | ||
c07f9fc5 FG |
2378 | offset -= offset % writebdy; |
2379 | if (o_direct) | |
2380 | size -= size % writebdy; | |
2381 | ||
2382 | if (size == 0) { | |
2383 | if (!quiet && testcalls > simulatedopcount && !o_direct) | |
2384 | prt("skipping zero size read\n"); | |
2385 | log4(OP_SKIPPED, OP_READ, offset, size); | |
2386 | return; | |
2387 | } | |
2388 | ||
2389 | if (size + offset > file_size) { | |
2390 | if (!quiet && testcalls > simulatedopcount) | |
2391 | prt("skipping seek/compare past end of file\n"); | |
2392 | log4(OP_SKIPPED, OP_COMPARE_AND_WRITE, offset, size); | |
2393 | return; | |
2394 | } | |
2395 | ||
2396 | memcpy(temp_buf + offset, good_buf + offset, size); | |
2397 | gendata(original_buf, good_buf, offset, size); | |
2398 | log4(OP_COMPARE_AND_WRITE, offset, size, 0); | |
2399 | ||
2400 | if (testcalls <= simulatedopcount) | |
2401 | return; | |
2402 | ||
2403 | if (!quiet && | |
2404 | ((progressinterval && testcalls % progressinterval == 0) || | |
2405 | (debug && | |
2406 | (monitorstart == -1 || | |
2407 | (static_cast<long>(offset + size) > monitorstart && | |
2408 | (monitorend == -1 || | |
2409 | static_cast<long>(offset) <= monitorend)))))) | |
2410 | prt("%lu compareandwrite\t0x%x thru\t0x%x\t(0x%x bytes)\n", testcalls, | |
2411 | offset, offset + size - 1, size); | |
2412 | ||
2413 | ret = ops->compare_and_write(&ctx, offset, size, temp_buf + offset, | |
2414 | good_buf + offset); | |
2415 | if (ret != (ssize_t)size) { | |
2416 | if (ret == -EINVAL) { | |
2417 | memcpy(good_buf + offset, temp_buf + offset, size); | |
2418 | return; | |
2419 | } | |
2420 | if (ret < 0) | |
2421 | prterrcode("docompareandwrite: ops->compare_and_write", ret); | |
2422 | else | |
2423 | prt("short write: 0x%x bytes instead of 0x%x\n", ret, size); | |
2424 | report_failure(151); | |
2425 | return; | |
2426 | } | |
2427 | ||
2428 | if (flush_enabled) | |
2429 | doflush(offset, size); | |
2430 | } | |
2431 | ||
7c673cae FG |
2432 | void clone_filename(char *buf, size_t len, int clones) |
2433 | { | |
11fdf7f2 TL |
2434 | #if __GNUC__ && __GNUC__ >= 8 |
2435 | #pragma GCC diagnostic push | |
2436 | #pragma GCC diagnostic ignored "-Wformat-truncation" | |
2437 | #endif | |
7c673cae FG |
2438 | snprintf(buf, len, "%s/fsx-%s-parent%d", |
2439 | dirpath, iname, clones); | |
11fdf7f2 TL |
2440 | #if __GNUC__ && __GNUC__ >= 8 |
2441 | #pragma GCC diagnostic pop | |
2442 | #endif | |
7c673cae FG |
2443 | } |
2444 | ||
2445 | void clone_imagename(char *buf, size_t len, int clones) | |
2446 | { | |
2447 | if (clones > 0) | |
2448 | snprintf(buf, len, "%s-clone%d", iname, clones); | |
2449 | else | |
2450 | strncpy(buf, iname, len); | |
2451 | buf[len - 1] = '\0'; | |
2452 | } | |
2453 | ||
2454 | void replay_imagename(char *buf, size_t len, int clones) | |
2455 | { | |
2456 | clone_imagename(buf, len, clones); | |
2457 | strncat(buf, "-replay", len - strlen(buf)); | |
2458 | buf[len - 1] = '\0'; | |
2459 | } | |
2460 | ||
2461 | void check_clone(int clonenum, bool replay_image); | |
2462 | ||
2463 | void | |
2464 | do_clone() | |
2465 | { | |
2466 | char filename[1024]; | |
2467 | char imagename[1024]; | |
2468 | char lastimagename[1024]; | |
2469 | int ret, fd; | |
2470 | int order = 0, stripe_unit = 0, stripe_count = 0; | |
2471 | uint64_t newsize = file_size; | |
2472 | ||
2473 | log4(OP_CLONE, 0, 0, 0); | |
2474 | ++num_clones; | |
2475 | ||
2476 | if (randomize_striping) { | |
2477 | order = 18 + get_random() % 8; | |
2478 | stripe_unit = 1ull << (order - 1 - (get_random() % 8)); | |
2479 | stripe_count = 2 + get_random() % 14; | |
2480 | } | |
2481 | ||
2482 | prt("%lu clone\t%d order %d su %d sc %d\n", testcalls, num_clones, | |
2483 | order, stripe_unit, stripe_count); | |
2484 | ||
2485 | clone_imagename(imagename, sizeof(imagename), num_clones); | |
2486 | clone_imagename(lastimagename, sizeof(lastimagename), | |
2487 | num_clones - 1); | |
11fdf7f2 | 2488 | ceph_assert(strcmp(lastimagename, ctx.name) == 0); |
7c673cae FG |
2489 | |
2490 | ret = ops->clone(&ctx, "snap", imagename, &order, stripe_unit, | |
2491 | stripe_count); | |
2492 | if (ret < 0) { | |
2493 | prterrcode("do_clone: ops->clone", ret); | |
2494 | exit(165); | |
2495 | } | |
2496 | ||
2497 | if (randomize_parent_overlap && rbd_image_has_parent(&ctx)) { | |
2498 | int rand = get_random() % 16 + 1; // [1..16] | |
2499 | ||
2500 | if (rand < 13) { | |
2501 | uint64_t overlap; | |
2502 | ||
2503 | ret = rbd_get_overlap(ctx.image, &overlap); | |
2504 | if (ret < 0) { | |
2505 | prterrcode("do_clone: rbd_get_overlap", ret); | |
2506 | exit(1); | |
2507 | } | |
2508 | ||
2509 | if (rand < 10) { // 9/16 | |
2510 | newsize = overlap * ((double)rand / 10); | |
2511 | newsize -= newsize % truncbdy; | |
2512 | } else { // 3/16 | |
2513 | newsize = 0; | |
2514 | } | |
2515 | ||
11fdf7f2 | 2516 | ceph_assert(newsize != (uint64_t)file_size); |
7c673cae FG |
2517 | prt("truncating image %s from 0x%llx (overlap 0x%llx) to 0x%llx\n", |
2518 | ctx.name, file_size, overlap, newsize); | |
2519 | ||
2520 | ret = ops->resize(&ctx, newsize); | |
2521 | if (ret < 0) { | |
2522 | prterrcode("do_clone: ops->resize", ret); | |
2523 | exit(1); | |
2524 | } | |
2525 | } else if (rand < 15) { // 2/16 | |
2526 | prt("flattening image %s\n", ctx.name); | |
2527 | ||
2528 | ret = ops->flatten(&ctx); | |
2529 | if (ret < 0) { | |
2530 | prterrcode("do_clone: ops->flatten", ret); | |
2531 | exit(1); | |
2532 | } | |
2533 | } else { // 2/16 | |
2534 | prt("leaving image %s intact\n", ctx.name); | |
2535 | } | |
2536 | } | |
2537 | ||
2538 | clone_filename(filename, sizeof(filename), num_clones); | |
2539 | if ((fd = open(filename, O_WRONLY|O_CREAT|O_TRUNC, 0666)) < 0) { | |
2540 | simple_err("do_clone: open", -errno); | |
2541 | exit(162); | |
2542 | } | |
2543 | save_buffer(good_buf, newsize, fd); | |
2544 | if ((ret = close(fd)) < 0) { | |
2545 | simple_err("do_clone: close", -errno); | |
2546 | exit(163); | |
2547 | } | |
2548 | ||
2549 | /* | |
2550 | * Close parent. | |
2551 | */ | |
2552 | if ((ret = ops->close(&ctx)) < 0) { | |
2553 | prterrcode("do_clone: ops->close", ret); | |
2554 | exit(174); | |
2555 | } | |
2556 | ||
2557 | if (journal_replay) { | |
2558 | ret = finalize_journal(ioctx, lastimagename, num_clones - 1, | |
2559 | order, stripe_unit, stripe_count); | |
2560 | if (ret < 0) { | |
2561 | exit(EXIT_FAILURE); | |
2562 | } | |
2563 | ||
2564 | ret = register_journal(ioctx, imagename); | |
2565 | if (ret < 0) { | |
2566 | exit(EXIT_FAILURE); | |
2567 | } | |
2568 | } | |
2569 | ||
2570 | /* | |
2571 | * Open freshly made clone. | |
2572 | */ | |
2573 | if ((ret = ops->open(imagename, &ctx)) < 0) { | |
2574 | prterrcode("do_clone: ops->open", ret); | |
2575 | exit(166); | |
2576 | } | |
2577 | ||
2578 | if (num_clones > 1) { | |
2579 | if (journal_replay) { | |
2580 | check_clone(num_clones - 2, true); | |
2581 | } | |
2582 | check_clone(num_clones - 2, false); | |
2583 | } | |
2584 | } | |
2585 | ||
2586 | void | |
2587 | check_clone(int clonenum, bool replay_image) | |
2588 | { | |
2589 | char filename[128]; | |
2590 | char imagename[128]; | |
2591 | int ret, fd; | |
2592 | struct rbd_ctx cur_ctx = RBD_CTX_INIT; | |
2593 | struct stat file_info; | |
2594 | char *good_buf, *temp_buf; | |
2595 | ||
2596 | if (replay_image) { | |
2597 | replay_imagename(imagename, sizeof(imagename), clonenum); | |
2598 | } else { | |
2599 | clone_imagename(imagename, sizeof(imagename), clonenum); | |
2600 | } | |
2601 | ||
2602 | if ((ret = ops->open(imagename, &cur_ctx)) < 0) { | |
2603 | prterrcode("check_clone: ops->open", ret); | |
2604 | exit(167); | |
2605 | } | |
2606 | ||
2607 | clone_filename(filename, sizeof(filename), clonenum + 1); | |
2608 | if ((fd = open(filename, O_RDONLY)) < 0) { | |
2609 | simple_err("check_clone: open", -errno); | |
2610 | exit(168); | |
2611 | } | |
2612 | ||
2613 | prt("checking clone #%d, image %s against file %s\n", | |
2614 | clonenum, imagename, filename); | |
2615 | if ((ret = fstat(fd, &file_info)) < 0) { | |
2616 | simple_err("check_clone: fstat", -errno); | |
2617 | exit(169); | |
2618 | } | |
2619 | ||
2620 | good_buf = NULL; | |
2621 | ret = posix_memalign((void **)&good_buf, | |
11fdf7f2 | 2622 | std::max(writebdy, (int)sizeof(void *)), |
7c673cae FG |
2623 | file_info.st_size); |
2624 | if (ret > 0) { | |
2625 | prterrcode("check_clone: posix_memalign(good_buf)", -ret); | |
2626 | exit(96); | |
2627 | } | |
2628 | ||
2629 | temp_buf = NULL; | |
2630 | ret = posix_memalign((void **)&temp_buf, | |
11fdf7f2 | 2631 | std::max(readbdy, (int)sizeof(void *)), |
7c673cae FG |
2632 | file_info.st_size); |
2633 | if (ret > 0) { | |
2634 | prterrcode("check_clone: posix_memalign(temp_buf)", -ret); | |
2635 | exit(97); | |
2636 | } | |
2637 | ||
2638 | if ((ret = pread(fd, good_buf, file_info.st_size, 0)) < 0) { | |
2639 | simple_err("check_clone: pread", -errno); | |
2640 | exit(170); | |
2641 | } | |
2642 | if ((ret = ops->read(&cur_ctx, 0, file_info.st_size, temp_buf)) < 0) { | |
2643 | prterrcode("check_clone: ops->read", ret); | |
2644 | exit(171); | |
2645 | } | |
2646 | close(fd); | |
2647 | if ((ret = ops->close(&cur_ctx)) < 0) { | |
2648 | prterrcode("check_clone: ops->close", ret); | |
2649 | exit(174); | |
2650 | } | |
2651 | check_buffers(good_buf, temp_buf, 0, file_info.st_size); | |
2652 | ||
2653 | if (!replay_image) { | |
2654 | unlink(filename); | |
2655 | } | |
2656 | ||
2657 | free(good_buf); | |
2658 | free(temp_buf); | |
2659 | } | |
2660 | ||
2661 | void | |
2662 | writefileimage() | |
2663 | { | |
2664 | ssize_t ret; | |
2665 | ||
2666 | ret = ops->write(&ctx, 0, file_size, good_buf); | |
2667 | if (ret != file_size) { | |
2668 | if (ret < 0) | |
2669 | prterrcode("writefileimage: ops->write", ret); | |
2670 | else | |
2671 | prt("short write: 0x%x bytes instead of 0x%llx\n", | |
2672 | ret, (unsigned long long)file_size); | |
2673 | report_failure(172); | |
2674 | } | |
2675 | ||
2676 | if (!lite) { | |
2677 | ret = ops->resize(&ctx, file_size); | |
2678 | if (ret < 0) { | |
2679 | prterrcode("writefileimage: ops->resize", ret); | |
2680 | report_failure(173); | |
2681 | } | |
2682 | } | |
2683 | } | |
2684 | ||
2685 | void | |
2686 | do_flatten() | |
2687 | { | |
2688 | int ret; | |
2689 | ||
2690 | if (!rbd_image_has_parent(&ctx)) { | |
2691 | log4(OP_SKIPPED, OP_FLATTEN, 0, 0); | |
2692 | return; | |
2693 | } | |
2694 | log4(OP_FLATTEN, 0, 0, 0); | |
2695 | prt("%lu flatten\n", testcalls); | |
2696 | ||
2697 | ret = ops->flatten(&ctx); | |
2698 | if (ret < 0) { | |
2699 | prterrcode("writefileimage: ops->flatten", ret); | |
2700 | exit(177); | |
2701 | } | |
2702 | } | |
2703 | ||
2704 | void | |
2705 | docloseopen(void) | |
2706 | { | |
2707 | char *name; | |
2708 | int ret; | |
2709 | ||
2710 | if (testcalls <= simulatedopcount) | |
2711 | return; | |
2712 | ||
2713 | name = strdup(ctx.name); | |
2714 | ||
2715 | if (debug) | |
2716 | prt("%lu close/open\n", testcalls); | |
2717 | ||
2718 | ret = ops->close(&ctx); | |
2719 | if (ret < 0) { | |
2720 | prterrcode("docloseopen: ops->close", ret); | |
2721 | report_failure(180); | |
2722 | } | |
2723 | ||
2724 | ret = ops->open(name, &ctx); | |
2725 | if (ret < 0) { | |
2726 | prterrcode("docloseopen: ops->open", ret); | |
2727 | report_failure(181); | |
2728 | } | |
2729 | ||
2730 | free(name); | |
2731 | } | |
2732 | ||
2733 | #define TRIM_OFF_LEN(off, len, size) \ | |
2734 | do { \ | |
2735 | if (size) \ | |
2736 | (off) %= (size); \ | |
2737 | else \ | |
2738 | (off) = 0; \ | |
2739 | if ((unsigned)(off) + (unsigned)(len) > (unsigned)(size)) \ | |
2740 | (len) = (size) - (off); \ | |
2741 | } while (0) | |
2742 | ||
2743 | void | |
2744 | test(void) | |
2745 | { | |
2746 | unsigned long offset; | |
2747 | unsigned long size = maxoplen; | |
2748 | unsigned long rv = get_random(); | |
2749 | unsigned long op; | |
2750 | ||
2751 | if (simulatedopcount > 0 && testcalls == simulatedopcount) | |
2752 | writefileimage(); | |
2753 | ||
2754 | testcalls++; | |
2755 | ||
2756 | if (closeprob) | |
2757 | closeopen = (rv >> 3) < (1u << 28) / (unsigned)closeprob; | |
2758 | ||
2759 | if (debugstart > 0 && testcalls >= debugstart) | |
2760 | debug = 1; | |
2761 | ||
2762 | if (!quiet && testcalls < simulatedopcount && testcalls % 100000 == 0) | |
2763 | prt("%lu...\n", testcalls); | |
2764 | ||
2765 | offset = get_random(); | |
2766 | if (randomoplen) | |
2767 | size = get_random() % (maxoplen + 1); | |
2768 | ||
2769 | /* calculate appropriate op to run */ | |
2770 | if (lite) | |
2771 | op = rv % OP_MAX_LITE; | |
2772 | else | |
2773 | op = rv % OP_MAX_FULL; | |
2774 | ||
2775 | switch (op) { | |
2776 | case OP_MAPREAD: | |
2777 | if (!mapped_reads) | |
2778 | op = OP_READ; | |
2779 | break; | |
2780 | case OP_MAPWRITE: | |
2781 | if (!mapped_writes) | |
2782 | op = OP_WRITE; | |
2783 | break; | |
2784 | case OP_FALLOCATE: | |
2785 | if (!fallocate_calls) { | |
2786 | log4(OP_SKIPPED, OP_FALLOCATE, offset, size); | |
2787 | goto out; | |
2788 | } | |
2789 | break; | |
2790 | case OP_PUNCH_HOLE: | |
2791 | if (!punch_hole_calls) { | |
2792 | log4(OP_SKIPPED, OP_PUNCH_HOLE, offset, size); | |
2793 | goto out; | |
2794 | } | |
2795 | break; | |
2796 | case OP_CLONE: | |
2797 | /* clone, 8% chance */ | |
2798 | if (!clone_calls || file_size == 0 || get_random() % 100 >= 8) { | |
2799 | log4(OP_SKIPPED, OP_CLONE, 0, 0); | |
2800 | goto out; | |
2801 | } | |
2802 | break; | |
2803 | case OP_FLATTEN: | |
2804 | /* flatten four times as rarely as clone, 2% chance */ | |
2805 | if (get_random() % 100 >= 2) { | |
2806 | log4(OP_SKIPPED, OP_FLATTEN, 0, 0); | |
2807 | goto out; | |
2808 | } | |
2809 | break; | |
2810 | case OP_WRITESAME: | |
2811 | /* writesame not implemented */ | |
2812 | if (!ops->writesame) { | |
2813 | log4(OP_SKIPPED, OP_WRITESAME, offset, size); | |
2814 | goto out; | |
2815 | } | |
d2e6a577 | 2816 | break; |
c07f9fc5 FG |
2817 | case OP_COMPARE_AND_WRITE: |
2818 | /* compare_and_write not implemented */ | |
2819 | if (!ops->compare_and_write) { | |
2820 | log4(OP_SKIPPED, OP_COMPARE_AND_WRITE, offset, size); | |
2821 | goto out; | |
2822 | } | |
d2e6a577 | 2823 | break; |
7c673cae FG |
2824 | } |
2825 | ||
2826 | switch (op) { | |
2827 | case OP_READ: | |
2828 | TRIM_OFF_LEN(offset, size, file_size); | |
2829 | doread(offset, size); | |
2830 | break; | |
2831 | ||
2832 | case OP_WRITE: | |
2833 | TRIM_OFF_LEN(offset, size, maxfilelen); | |
2834 | dowrite(offset, size); | |
2835 | break; | |
2836 | ||
2837 | case OP_MAPREAD: | |
2838 | TRIM_OFF_LEN(offset, size, file_size); | |
2839 | exit(183); | |
2840 | break; | |
2841 | ||
2842 | case OP_MAPWRITE: | |
2843 | TRIM_OFF_LEN(offset, size, maxfilelen); | |
2844 | exit(182); | |
2845 | break; | |
2846 | ||
2847 | case OP_TRUNCATE: | |
2848 | if (!style) | |
2849 | size = get_random() % maxfilelen; | |
2850 | dotruncate(size); | |
2851 | break; | |
2852 | ||
2853 | case OP_PUNCH_HOLE: | |
2854 | TRIM_OFF_LEN(offset, size, file_size); | |
2855 | do_punch_hole(offset, size); | |
2856 | break; | |
2857 | ||
2858 | case OP_WRITESAME: | |
2859 | TRIM_OFF_LEN(offset, size, maxfilelen); | |
2860 | dowritesame(offset, size); | |
2861 | break; | |
c07f9fc5 FG |
2862 | case OP_COMPARE_AND_WRITE: |
2863 | TRIM_OFF_LEN(offset, size, file_size); | |
2864 | docompareandwrite(offset, size); | |
2865 | break; | |
7c673cae FG |
2866 | |
2867 | case OP_CLONE: | |
2868 | do_clone(); | |
2869 | break; | |
2870 | ||
2871 | case OP_FLATTEN: | |
2872 | do_flatten(); | |
2873 | break; | |
2874 | ||
2875 | default: | |
2876 | prterr("test: unknown operation"); | |
2877 | report_failure(42); | |
2878 | break; | |
2879 | } | |
2880 | ||
2881 | out: | |
2882 | if (sizechecks && testcalls > simulatedopcount) | |
2883 | check_size(); | |
2884 | if (closeopen) | |
2885 | docloseopen(); | |
2886 | } | |
2887 | ||
2888 | ||
2889 | void | |
2890 | cleanup(int sig) | |
2891 | { | |
2892 | if (sig) | |
2893 | prt("signal %d\n", sig); | |
2894 | prt("testcalls = %lu\n", testcalls); | |
2895 | exit(sig); | |
2896 | } | |
2897 | ||
2898 | ||
2899 | void | |
2900 | usage(void) | |
2901 | { | |
2902 | fprintf(stdout, "usage: %s", | |
2903 | "fsx [-dfjknqxyACFHKLORUWZ] [-b opnum] [-c Prob] [-h holebdy] [-l flen] [-m start:end] [-o oplen] [-p progressinterval] [-r readbdy] [-s style] [-t truncbdy] [-w writebdy] [-D startingop] [-N numops] [-P dirpath] [-S seed] pname iname\n\ | |
2904 | -b opnum: beginning operation number (default 1)\n\ | |
2905 | -c P: 1 in P chance of file close+open at each op (default infinity)\n\ | |
2906 | -d: debug output for all operations\n\ | |
2907 | -f: flush and invalidate cache after I/O\n\ | |
11fdf7f2 | 2908 | -g: deep copy instead of clone\n\ |
7c673cae FG |
2909 | -h holebdy: 4096 would make discards page aligned (default 1)\n\ |
2910 | -j: journal replay stress test\n\ | |
2911 | -k: keep data on success (default 0)\n\ | |
2912 | -l flen: the upper bound on file size (default 262144)\n\ | |
2913 | -m startop:endop: monitor (print debug output) specified byte range (default 0:infinity)\n\ | |
2914 | -n: no verifications of file size\n\ | |
2915 | -o oplen: the upper bound on operation size (default 65536)\n\ | |
2916 | -p progressinterval: debug output at specified operation interval\n\ | |
2917 | -q: quieter operation\n\ | |
2918 | -r readbdy: 4096 would make reads page aligned (default 1)\n\ | |
2919 | -s style: 1 gives smaller truncates (default 0)\n\ | |
2920 | -t truncbdy: 4096 would make truncates page aligned (default 1)\n\ | |
2921 | -w writebdy: 4096 would make writes page aligned (default 1)\n\ | |
2922 | -x: preallocate file space before starting, XFS only (default 0)\n\ | |
2923 | -y: synchronize changes to a file\n" | |
2924 | ||
2925 | " -C: do not use clone calls\n\ | |
2926 | -D startingop: debug output starting at specified operation\n" | |
2927 | #ifdef FALLOCATE | |
2928 | " -F: Do not use fallocate (preallocation) calls\n" | |
2929 | #endif | |
11fdf7f2 TL |
2930 | #if defined(__FreeBSD__) |
2931 | " -G: enable rbd-ggate mode (use -L, -r and -w too)\n" | |
2932 | #endif | |
2933 | " -H: do not use punch hole calls\n" | |
2934 | #if defined(WITH_KRBD) | |
2935 | " -K: enable krbd mode (use -t and -h too)\n" | |
2936 | #endif | |
2937 | #if defined(__linux__) | |
2938 | " -M: enable rbd-nbd mode (use -t and -h too)\n" | |
2939 | #endif | |
2940 | " -L: fsxLite - no file creations & no file size changes\n\ | |
7c673cae FG |
2941 | -N numops: total # operations to do (default infinity)\n\ |
2942 | -O: use oplen (see -o flag) for every op (default random)\n\ | |
2943 | -P dirpath: save .fsxlog and .fsxgood files in dirpath (default ./)\n\ | |
2944 | -R: read() system calls only (mapped reads disabled)\n\ | |
2945 | -S seed: for random # generator (default 1) 0 gets timestamp\n\ | |
2946 | -U: disable randomized striping\n\ | |
2947 | -W: mapped write operations DISabled\n\ | |
2948 | -Z: O_DIRECT (use -R, -W, -r and -w too)\n\ | |
2949 | poolname: this is REQUIRED (no default)\n\ | |
2950 | imagename: this is REQUIRED (no default)\n"); | |
2951 | exit(89); | |
2952 | } | |
2953 | ||
2954 | ||
2955 | int | |
2956 | getnum(char *s, char **e) | |
2957 | { | |
2958 | int ret; | |
2959 | ||
2960 | *e = (char *) 0; | |
2961 | ret = strtol(s, e, 0); | |
2962 | if (*e) | |
2963 | switch (**e) { | |
2964 | case 'b': | |
2965 | case 'B': | |
2966 | ret *= 512; | |
2967 | *e = *e + 1; | |
2968 | break; | |
2969 | case 'k': | |
2970 | case 'K': | |
2971 | ret *= 1024; | |
2972 | *e = *e + 1; | |
2973 | break; | |
2974 | case 'm': | |
2975 | case 'M': | |
2976 | ret *= 1024*1024; | |
2977 | *e = *e + 1; | |
2978 | break; | |
2979 | case 'w': | |
2980 | case 'W': | |
2981 | ret *= 4; | |
2982 | *e = *e + 1; | |
2983 | break; | |
2984 | } | |
2985 | return (ret); | |
2986 | } | |
2987 | ||
2988 | void | |
2989 | test_fallocate() | |
2990 | { | |
2991 | #ifdef FALLOCATE | |
2992 | if (!lite && fallocate_calls) { | |
2993 | if (fallocate(fd, 0, 0, 1) && errno == EOPNOTSUPP) { | |
2994 | if(!quiet) | |
2995 | warn("main: filesystem does not support fallocate, disabling\n"); | |
2996 | fallocate_calls = 0; | |
2997 | } else { | |
2998 | ftruncate(fd, 0); | |
2999 | } | |
3000 | } | |
3001 | #else /* ! FALLOCATE */ | |
3002 | fallocate_calls = 0; | |
3003 | #endif | |
3004 | ||
3005 | } | |
3006 | ||
3007 | void remove_image(rados_ioctx_t ioctx, char *imagename, bool remove_snap, | |
3008 | bool unregister) { | |
3009 | rbd_image_t image; | |
3010 | char errmsg[128]; | |
3011 | int ret; | |
3012 | ||
3013 | if ((ret = rbd_open(ioctx, imagename, &image, NULL)) < 0) { | |
3014 | sprintf(errmsg, "rbd_open %s", imagename); | |
3015 | prterrcode(errmsg, ret); | |
3016 | report_failure(101); | |
3017 | } | |
3018 | if (remove_snap) { | |
3019 | if ((ret = rbd_snap_unprotect(image, "snap")) < 0) { | |
3020 | sprintf(errmsg, "rbd_snap_unprotect %s@snap", | |
3021 | imagename); | |
3022 | prterrcode(errmsg, ret); | |
3023 | report_failure(102); | |
3024 | } | |
3025 | if ((ret = rbd_snap_remove(image, "snap")) < 0) { | |
3026 | sprintf(errmsg, "rbd_snap_remove %s@snap", | |
3027 | imagename); | |
3028 | prterrcode(errmsg, ret); | |
3029 | report_failure(103); | |
3030 | } | |
3031 | } | |
3032 | if ((ret = rbd_close(image)) < 0) { | |
3033 | sprintf(errmsg, "rbd_close %s", imagename); | |
3034 | prterrcode(errmsg, ret); | |
3035 | report_failure(104); | |
3036 | } | |
3037 | ||
3038 | if (unregister && | |
3039 | (ret = unregister_journal(ioctx, imagename)) < 0) { | |
3040 | report_failure(105); | |
3041 | } | |
3042 | ||
3043 | if ((ret = rbd_remove(ioctx, imagename)) < 0) { | |
3044 | sprintf(errmsg, "rbd_remove %s", imagename); | |
3045 | prterrcode(errmsg, ret); | |
3046 | report_failure(106); | |
3047 | } | |
3048 | } | |
3049 | ||
3050 | int | |
3051 | main(int argc, char **argv) | |
3052 | { | |
11fdf7f2 TL |
3053 | enum { |
3054 | LONG_OPT_CLUSTER = 1000, | |
3055 | LONG_OPT_ID = 1001 | |
3056 | }; | |
3057 | ||
7c673cae FG |
3058 | int i, style, ch, ret; |
3059 | char *endp; | |
3060 | char goodfile[1024]; | |
3061 | char logfile[1024]; | |
3062 | ||
11fdf7f2 TL |
3063 | const char* optstring = "b:c:dfgh:jkl:m:no:p:qr:s:t:w:xyCD:FGHKMLN:OP:RS:UWZ"; |
3064 | const struct option longopts[] = { | |
3065 | {"cluster", 1, NULL, LONG_OPT_CLUSTER}, | |
3066 | {"id", 1, NULL, LONG_OPT_ID}}; | |
3067 | ||
7c673cae FG |
3068 | goodfile[0] = 0; |
3069 | logfile[0] = 0; | |
3070 | ||
3071 | page_size = getpagesize(); | |
3072 | page_mask = page_size - 1; | |
3073 | mmap_mask = page_mask; | |
3074 | ||
3075 | setvbuf(stdout, (char *)0, _IOLBF, 0); /* line buffered stdout */ | |
3076 | ||
11fdf7f2 | 3077 | while ((ch = getopt_long(argc, argv, optstring, longopts, NULL)) != EOF) { |
7c673cae | 3078 | switch (ch) { |
11fdf7f2 TL |
3079 | case LONG_OPT_CLUSTER: |
3080 | cluster_name = optarg; | |
3081 | break; | |
3082 | case LONG_OPT_ID: | |
3083 | client_id = optarg; | |
3084 | break; | |
7c673cae FG |
3085 | case 'b': |
3086 | simulatedopcount = getnum(optarg, &endp); | |
3087 | if (!quiet) | |
3088 | fprintf(stdout, "Will begin at operation %lu\n", | |
3089 | simulatedopcount); | |
3090 | if (simulatedopcount == 0) | |
3091 | usage(); | |
3092 | simulatedopcount -= 1; | |
3093 | break; | |
3094 | case 'c': | |
3095 | closeprob = getnum(optarg, &endp); | |
3096 | if (!quiet) | |
3097 | fprintf(stdout, | |
3098 | "Chance of close/open is 1 in %d\n", | |
3099 | closeprob); | |
3100 | if (closeprob <= 0) | |
3101 | usage(); | |
3102 | break; | |
3103 | case 'd': | |
3104 | debug = 1; | |
3105 | break; | |
3106 | case 'f': | |
3107 | flush_enabled = 1; | |
3108 | break; | |
11fdf7f2 TL |
3109 | case 'g': |
3110 | deep_copy = 1; | |
3111 | break; | |
7c673cae FG |
3112 | case 'h': |
3113 | holebdy = getnum(optarg, &endp); | |
3114 | if (holebdy <= 0) | |
3115 | usage(); | |
3116 | break; | |
3117 | case 'j': | |
3118 | journal_replay = true; | |
3119 | break; | |
3120 | case 'k': | |
3121 | keep_on_success = 1; | |
3122 | break; | |
3123 | case 'l': | |
3124 | { | |
3125 | int _num = getnum(optarg, &endp); | |
3126 | if (_num <= 0) | |
3127 | usage(); | |
3128 | maxfilelen = _num; | |
3129 | } | |
3130 | break; | |
3131 | case 'm': | |
3132 | monitorstart = getnum(optarg, &endp); | |
3133 | if (monitorstart < 0) | |
3134 | usage(); | |
3135 | if (!endp || *endp++ != ':') | |
3136 | usage(); | |
3137 | monitorend = getnum(endp, &endp); | |
3138 | if (monitorend < 0) | |
3139 | usage(); | |
3140 | if (monitorend == 0) | |
3141 | monitorend = -1; /* aka infinity */ | |
3142 | debug = 1; | |
3143 | break; | |
3144 | case 'n': | |
3145 | sizechecks = 0; | |
3146 | break; | |
3147 | case 'o': | |
3148 | maxoplen = getnum(optarg, &endp); | |
3149 | if (maxoplen <= 0) | |
3150 | usage(); | |
3151 | break; | |
3152 | case 'p': | |
3153 | progressinterval = getnum(optarg, &endp); | |
3154 | if (progressinterval == 0) | |
3155 | usage(); | |
3156 | break; | |
3157 | case 'q': | |
3158 | quiet = 1; | |
3159 | break; | |
3160 | case 'r': | |
3161 | readbdy = getnum(optarg, &endp); | |
3162 | if (readbdy <= 0) | |
3163 | usage(); | |
3164 | break; | |
3165 | case 's': | |
3166 | style = getnum(optarg, &endp); | |
3167 | if (style < 0 || style > 1) | |
3168 | usage(); | |
3169 | break; | |
3170 | case 't': | |
3171 | truncbdy = getnum(optarg, &endp); | |
3172 | if (truncbdy <= 0) | |
3173 | usage(); | |
3174 | break; | |
3175 | case 'w': | |
3176 | writebdy = getnum(optarg, &endp); | |
3177 | if (writebdy <= 0) | |
3178 | usage(); | |
3179 | break; | |
3180 | case 'x': | |
3181 | prealloc = 1; | |
3182 | break; | |
3183 | case 'y': | |
3184 | do_fsync = 1; | |
3185 | break; | |
3186 | case 'C': | |
3187 | clone_calls = 0; | |
3188 | break; | |
3189 | case 'D': | |
3190 | debugstart = getnum(optarg, &endp); | |
3191 | if (debugstart < 1) | |
3192 | usage(); | |
3193 | break; | |
3194 | case 'F': | |
3195 | fallocate_calls = 0; | |
3196 | break; | |
11fdf7f2 TL |
3197 | #if defined(__FreeBSD__) |
3198 | case 'G': | |
3199 | prt("rbd-ggate mode enabled\n"); | |
3200 | ops = &ggate_operations; | |
3201 | break; | |
3202 | #endif | |
7c673cae FG |
3203 | case 'H': |
3204 | punch_hole_calls = 0; | |
3205 | break; | |
11fdf7f2 | 3206 | #if defined(WITH_KRBD) |
7c673cae FG |
3207 | case 'K': |
3208 | prt("krbd mode enabled\n"); | |
3209 | ops = &krbd_operations; | |
3210 | break; | |
11fdf7f2 TL |
3211 | #endif |
3212 | #if defined(__linux__) | |
7c673cae FG |
3213 | case 'M': |
3214 | prt("rbd-nbd mode enabled\n"); | |
3215 | ops = &nbd_operations; | |
3216 | break; | |
11fdf7f2 | 3217 | #endif |
7c673cae | 3218 | case 'L': |
11fdf7f2 | 3219 | lite = 1; |
7c673cae FG |
3220 | break; |
3221 | case 'N': | |
3222 | numops = getnum(optarg, &endp); | |
3223 | if (numops < 0) | |
3224 | usage(); | |
3225 | break; | |
3226 | case 'O': | |
3227 | randomoplen = 0; | |
3228 | break; | |
3229 | case 'P': | |
3230 | strncpy(dirpath, optarg, sizeof(dirpath)-1); | |
3231 | dirpath[sizeof(dirpath)-1] = '\0'; | |
3232 | strncpy(goodfile, dirpath, sizeof(goodfile)-1); | |
3233 | goodfile[sizeof(goodfile)-1] = '\0'; | |
3234 | if (strlen(goodfile) < sizeof(goodfile)-2) { | |
3235 | strcat(goodfile, "/"); | |
3236 | } else { | |
3237 | prt("file name to long\n"); | |
3238 | exit(1); | |
3239 | } | |
3240 | strncpy(logfile, dirpath, sizeof(logfile)-1); | |
3241 | logfile[sizeof(logfile)-1] = '\0'; | |
3242 | if (strlen(logfile) < sizeof(logfile)-2) { | |
3243 | strcat(logfile, "/"); | |
3244 | } else { | |
3245 | prt("file path to long\n"); | |
3246 | exit(1); | |
3247 | } | |
3248 | break; | |
3249 | case 'R': | |
3250 | mapped_reads = 0; | |
3251 | if (!quiet) | |
3252 | fprintf(stdout, "mapped reads DISABLED\n"); | |
3253 | break; | |
3254 | case 'S': | |
3255 | seed = getnum(optarg, &endp); | |
3256 | if (seed == 0) | |
11fdf7f2 | 3257 | seed = std::random_device()() % 10000; |
7c673cae FG |
3258 | if (!quiet) |
3259 | fprintf(stdout, "Seed set to %d\n", seed); | |
3260 | if (seed < 0) | |
3261 | usage(); | |
3262 | break; | |
3263 | case 'U': | |
3264 | randomize_striping = 0; | |
3265 | break; | |
3266 | case 'W': | |
3267 | mapped_writes = 0; | |
3268 | if (!quiet) | |
3269 | fprintf(stdout, "mapped writes DISABLED\n"); | |
3270 | break; | |
3271 | case 'Z': | |
3272 | o_direct = O_DIRECT; | |
3273 | break; | |
3274 | default: | |
3275 | usage(); | |
3276 | /* NOTREACHED */ | |
3277 | } | |
11fdf7f2 | 3278 | } |
7c673cae FG |
3279 | argc -= optind; |
3280 | argv += optind; | |
3281 | if (argc != 2) | |
3282 | usage(); | |
3283 | pool = argv[0]; | |
3284 | iname = argv[1]; | |
3285 | ||
3286 | signal(SIGHUP, cleanup); | |
3287 | signal(SIGINT, cleanup); | |
3288 | signal(SIGPIPE, cleanup); | |
3289 | signal(SIGALRM, cleanup); | |
3290 | signal(SIGTERM, cleanup); | |
3291 | signal(SIGXCPU, cleanup); | |
3292 | signal(SIGXFSZ, cleanup); | |
3293 | signal(SIGVTALRM, cleanup); | |
3294 | signal(SIGUSR1, cleanup); | |
3295 | signal(SIGUSR2, cleanup); | |
3296 | ||
3297 | random_generator.seed(seed); | |
3298 | ||
11fdf7f2 TL |
3299 | if (lite) { |
3300 | file_size = maxfilelen; | |
3301 | } | |
3302 | ||
7c673cae FG |
3303 | ret = create_image(); |
3304 | if (ret < 0) { | |
3305 | prterrcode(iname, ret); | |
3306 | exit(90); | |
3307 | } | |
3308 | ret = ops->open(iname, &ctx); | |
3309 | if (ret < 0) { | |
3310 | simple_err("Error opening image", ret); | |
3311 | exit(91); | |
3312 | } | |
3313 | if (!dirpath[0]) | |
3314 | strcat(dirpath, "."); | |
3315 | strncat(goodfile, iname, 256); | |
3316 | strcat (goodfile, ".fsxgood"); | |
3317 | fsxgoodfd = open(goodfile, O_RDWR|O_CREAT|O_TRUNC, 0666); | |
3318 | if (fsxgoodfd < 0) { | |
3319 | prterr(goodfile); | |
3320 | exit(92); | |
3321 | } | |
3322 | strncat(logfile, iname, 256); | |
3323 | strcat (logfile, ".fsxlog"); | |
3324 | fsxlogf = fopen(logfile, "w"); | |
3325 | if (fsxlogf == NULL) { | |
3326 | prterr(logfile); | |
3327 | exit(93); | |
3328 | } | |
3329 | ||
3330 | original_buf = (char *) malloc(maxfilelen); | |
3331 | for (i = 0; i < (int)maxfilelen; i++) | |
3332 | original_buf[i] = get_random() % 256; | |
3333 | ||
3334 | ret = posix_memalign((void **)&good_buf, | |
11fdf7f2 | 3335 | std::max(writebdy, (int)sizeof(void *)), maxfilelen); |
7c673cae FG |
3336 | if (ret > 0) { |
3337 | if (ret == EINVAL) | |
3338 | prt("writebdy is not a suitable power of two\n"); | |
3339 | else | |
3340 | prterrcode("main: posix_memalign(good_buf)", -ret); | |
3341 | exit(94); | |
3342 | } | |
3343 | memset(good_buf, '\0', maxfilelen); | |
3344 | ||
3345 | ret = posix_memalign((void **)&temp_buf, | |
11fdf7f2 | 3346 | std::max(readbdy, (int)sizeof(void *)), maxfilelen); |
7c673cae FG |
3347 | if (ret > 0) { |
3348 | if (ret == EINVAL) | |
3349 | prt("readbdy is not a suitable power of two\n"); | |
3350 | else | |
3351 | prterrcode("main: posix_memalign(temp_buf)", -ret); | |
3352 | exit(95); | |
3353 | } | |
3354 | memset(temp_buf, '\0', maxfilelen); | |
3355 | ||
3356 | if (lite) { /* zero entire existing file */ | |
3357 | ssize_t written; | |
3358 | ||
3359 | written = ops->write(&ctx, 0, (size_t)maxfilelen, good_buf); | |
3360 | if (written != (ssize_t)maxfilelen) { | |
3361 | if (written < 0) { | |
3362 | prterrcode(iname, written); | |
3363 | warn("main: error on write"); | |
3364 | } else | |
3365 | warn("main: short write, 0x%x bytes instead " | |
3366 | "of 0x%lx\n", | |
3367 | (unsigned)written, | |
3368 | maxfilelen); | |
3369 | exit(98); | |
3370 | } | |
3371 | } else | |
3372 | check_trunc_hack(); | |
3373 | ||
3374 | //test_fallocate(); | |
3375 | ||
3376 | while (numops == -1 || numops--) | |
3377 | test(); | |
3378 | ||
3379 | ret = ops->close(&ctx); | |
3380 | if (ret < 0) { | |
3381 | prterrcode("ops->close", ret); | |
3382 | report_failure(99); | |
3383 | } | |
3384 | ||
3385 | if (journal_replay) { | |
3386 | char imagename[1024]; | |
3387 | clone_imagename(imagename, sizeof(imagename), num_clones); | |
3388 | ret = finalize_journal(ioctx, imagename, num_clones, 0, 0, 0); | |
3389 | if (ret < 0) { | |
3390 | report_failure(100); | |
3391 | } | |
3392 | } | |
3393 | ||
3394 | if (num_clones > 0) { | |
3395 | if (journal_replay) { | |
3396 | check_clone(num_clones - 1, true); | |
3397 | } | |
3398 | check_clone(num_clones - 1, false); | |
3399 | } | |
3400 | ||
3401 | if (!keep_on_success) { | |
3402 | while (num_clones >= 0) { | |
3403 | static bool remove_snap = false; | |
3404 | ||
3405 | if (journal_replay) { | |
3406 | char replayimagename[1024]; | |
3407 | replay_imagename(replayimagename, | |
3408 | sizeof(replayimagename), | |
3409 | num_clones); | |
3410 | remove_image(ioctx, replayimagename, | |
3411 | remove_snap, | |
3412 | false); | |
3413 | } | |
3414 | ||
3415 | char clonename[128]; | |
3416 | clone_imagename(clonename, 128, num_clones); | |
3417 | remove_image(ioctx, clonename, remove_snap, | |
3418 | journal_replay); | |
3419 | ||
3420 | remove_snap = true; | |
3421 | num_clones--; | |
3422 | } | |
3423 | } | |
3424 | ||
3425 | prt("All operations completed A-OK!\n"); | |
3426 | fclose(fsxlogf); | |
3427 | ||
3428 | rados_ioctx_destroy(ioctx); | |
11fdf7f2 | 3429 | #if defined(WITH_KRBD) |
7c673cae | 3430 | krbd_destroy(krbd); |
11fdf7f2 | 3431 | #endif |
7c673cae FG |
3432 | rados_shutdown(cluster); |
3433 | ||
3434 | free(original_buf); | |
3435 | free(good_buf); | |
3436 | free(temp_buf); | |
3437 | ||
3438 | exit(0); | |
3439 | return 0; | |
3440 | } |