xref: /qemu/block/blklogwrites.c (revision c23a9563)
1 /*
2  * Write logging blk driver based on blkverify and blkdebug.
3  *
4  * Copyright (c) 2017 Tuomas Tynkkynen <tuomas@tuxera.com>
5  * Copyright (c) 2018 Aapo Vienamo <aapo@tuxera.com>
6  * Copyright (c) 2018 Ari Sundholm <ari@tuxera.com>
7  *
8  * This work is licensed under the terms of the GNU GPL, version 2 or later.
9  * See the COPYING file in the top-level directory.
10  */
11 
12 #include "qemu/osdep.h"
13 #include "qapi/error.h"
14 #include "qemu/sockets.h" /* for EINPROGRESS on Windows */
15 #include "block/block_int.h"
16 #include "qapi/qmp/qdict.h"
17 #include "qapi/qmp/qstring.h"
18 #include "qemu/cutils.h"
19 #include "qemu/module.h"
20 #include "qemu/option.h"
21 
22 /* Disk format stuff - taken from Linux drivers/md/dm-log-writes.c */
23 
24 #define LOG_FLUSH_FLAG   (1 << 0)
25 #define LOG_FUA_FLAG     (1 << 1)
26 #define LOG_DISCARD_FLAG (1 << 2)
27 #define LOG_MARK_FLAG    (1 << 3)
28 #define LOG_FLAG_MASK    (LOG_FLUSH_FLAG \
29                          | LOG_FUA_FLAG \
30                          | LOG_DISCARD_FLAG \
31                          | LOG_MARK_FLAG)
32 
33 #define WRITE_LOG_VERSION 1ULL
34 #define WRITE_LOG_MAGIC 0x6a736677736872ULL
35 
36 /* All fields are little-endian. */
37 struct log_write_super {
38     uint64_t magic;
39     uint64_t version;
40     uint64_t nr_entries;
41     uint32_t sectorsize;
42 } QEMU_PACKED;
43 
44 struct log_write_entry {
45     uint64_t sector;
46     uint64_t nr_sectors;
47     uint64_t flags;
48     uint64_t data_len;
49 } QEMU_PACKED;
50 
51 /* End of disk format structures. */
52 
53 typedef struct {
54     BdrvChild *log_file;
55     uint32_t sectorsize;
56     uint32_t sectorbits;
57     uint64_t cur_log_sector;
58     uint64_t nr_entries;
59     uint64_t update_interval;
60 } BDRVBlkLogWritesState;
61 
62 static QemuOptsList runtime_opts = {
63     .name = "blklogwrites",
64     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
65     .desc = {
66         {
67             .name = "log-append",
68             .type = QEMU_OPT_BOOL,
69             .help = "Append to an existing log",
70         },
71         {
72             .name = "log-sector-size",
73             .type = QEMU_OPT_SIZE,
74             .help = "Log sector size",
75         },
76         {
77             .name = "log-super-update-interval",
78             .type = QEMU_OPT_NUMBER,
79             .help = "Log superblock update interval (# of write requests)",
80         },
81         { /* end of list */ }
82     },
83 };
84 
85 static inline uint32_t blk_log_writes_log2(uint32_t value)
86 {
87     assert(value > 0);
88     return 31 - clz32(value);
89 }
90 
91 static inline bool blk_log_writes_sector_size_valid(uint32_t sector_size)
92 {
93     return is_power_of_2(sector_size) &&
94         sector_size >= sizeof(struct log_write_super) &&
95         sector_size >= sizeof(struct log_write_entry) &&
96         sector_size < (1ull << 24);
97 }
98 
99 static uint64_t blk_log_writes_find_cur_log_sector(BdrvChild *log,
100                                                    uint32_t sector_size,
101                                                    uint64_t nr_entries,
102                                                    Error **errp)
103 {
104     uint64_t cur_sector = 1;
105     uint64_t cur_idx = 0;
106     uint32_t sector_bits = blk_log_writes_log2(sector_size);
107     struct log_write_entry cur_entry;
108 
109     while (cur_idx < nr_entries) {
110         int read_ret = bdrv_pread(log, cur_sector << sector_bits,
111                                   sizeof(cur_entry), &cur_entry, 0);
112         if (read_ret < 0) {
113             error_setg_errno(errp, -read_ret,
114                              "Failed to read log entry %"PRIu64, cur_idx);
115             return (uint64_t)-1ull;
116         }
117 
118         if (cur_entry.flags & ~cpu_to_le64(LOG_FLAG_MASK)) {
119             error_setg(errp, "Invalid flags 0x%"PRIx64" in log entry %"PRIu64,
120                        le64_to_cpu(cur_entry.flags), cur_idx);
121             return (uint64_t)-1ull;
122         }
123 
124         /* Account for the sector of the entry itself */
125         ++cur_sector;
126 
127         /*
128          * Account for the data of the write.
129          * For discards, this data is not present.
130          */
131         if (!(cur_entry.flags & cpu_to_le64(LOG_DISCARD_FLAG))) {
132             cur_sector += le64_to_cpu(cur_entry.nr_sectors);
133         }
134 
135         ++cur_idx;
136     }
137 
138     return cur_sector;
139 }
140 
141 static int blk_log_writes_open(BlockDriverState *bs, QDict *options, int flags,
142                                Error **errp)
143 {
144     BDRVBlkLogWritesState *s = bs->opaque;
145     QemuOpts *opts;
146     Error *local_err = NULL;
147     int ret;
148     uint64_t log_sector_size;
149     bool log_append;
150 
151     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
152     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
153         ret = -EINVAL;
154         goto fail;
155     }
156 
157     /* Open the file */
158     ret = bdrv_open_file_child(NULL, options, "file", bs, errp);
159     if (ret < 0) {
160         goto fail;
161     }
162 
163     /* Open the log file */
164     s->log_file = bdrv_open_child(NULL, options, "log", bs, &child_of_bds,
165                                   BDRV_CHILD_METADATA, false, errp);
166     if (!s->log_file) {
167         ret = -EINVAL;
168         goto fail;
169     }
170 
171     log_append = qemu_opt_get_bool(opts, "log-append", false);
172 
173     if (log_append) {
174         struct log_write_super log_sb = { 0, 0, 0, 0 };
175 
176         if (qemu_opt_find(opts, "log-sector-size")) {
177             ret = -EINVAL;
178             error_setg(errp, "log-append and log-sector-size are mutually "
179                        "exclusive");
180             goto fail_log;
181         }
182 
183         /* Read log superblock or fake one for an empty log */
184         if (!bdrv_getlength(s->log_file->bs)) {
185             log_sb.magic      = cpu_to_le64(WRITE_LOG_MAGIC);
186             log_sb.version    = cpu_to_le64(WRITE_LOG_VERSION);
187             log_sb.nr_entries = cpu_to_le64(0);
188             log_sb.sectorsize = cpu_to_le32(BDRV_SECTOR_SIZE);
189         } else {
190             ret = bdrv_pread(s->log_file, 0, sizeof(log_sb), &log_sb, 0);
191             if (ret < 0) {
192                 error_setg_errno(errp, -ret, "Could not read log superblock");
193                 goto fail_log;
194             }
195         }
196 
197         if (log_sb.magic != cpu_to_le64(WRITE_LOG_MAGIC)) {
198             ret = -EINVAL;
199             error_setg(errp, "Invalid log superblock magic");
200             goto fail_log;
201         }
202 
203         if (log_sb.version != cpu_to_le64(WRITE_LOG_VERSION)) {
204             ret = -EINVAL;
205             error_setg(errp, "Unsupported log version %"PRIu64,
206                        le64_to_cpu(log_sb.version));
207             goto fail_log;
208         }
209 
210         log_sector_size = le32_to_cpu(log_sb.sectorsize);
211         s->cur_log_sector = 1;
212         s->nr_entries = 0;
213 
214         if (blk_log_writes_sector_size_valid(log_sector_size)) {
215             s->cur_log_sector =
216                 blk_log_writes_find_cur_log_sector(s->log_file, log_sector_size,
217                                     le64_to_cpu(log_sb.nr_entries), &local_err);
218             if (local_err) {
219                 ret = -EINVAL;
220                 error_propagate(errp, local_err);
221                 goto fail_log;
222             }
223 
224             s->nr_entries = le64_to_cpu(log_sb.nr_entries);
225         }
226     } else {
227         log_sector_size = qemu_opt_get_size(opts, "log-sector-size",
228                                             BDRV_SECTOR_SIZE);
229         s->cur_log_sector = 1;
230         s->nr_entries = 0;
231     }
232 
233     if (!blk_log_writes_sector_size_valid(log_sector_size)) {
234         ret = -EINVAL;
235         error_setg(errp, "Invalid log sector size %"PRIu64, log_sector_size);
236         goto fail_log;
237     }
238 
239     s->sectorsize = log_sector_size;
240     s->sectorbits = blk_log_writes_log2(log_sector_size);
241     s->update_interval = qemu_opt_get_number(opts, "log-super-update-interval",
242                                              4096);
243     if (!s->update_interval) {
244         ret = -EINVAL;
245         error_setg(errp, "Invalid log superblock update interval %"PRIu64,
246                    s->update_interval);
247         goto fail_log;
248     }
249 
250     ret = 0;
251 fail_log:
252     if (ret < 0) {
253         bdrv_unref_child(bs, s->log_file);
254         s->log_file = NULL;
255     }
256 fail:
257     qemu_opts_del(opts);
258     return ret;
259 }
260 
261 static void blk_log_writes_close(BlockDriverState *bs)
262 {
263     BDRVBlkLogWritesState *s = bs->opaque;
264 
265     bdrv_unref_child(bs, s->log_file);
266     s->log_file = NULL;
267 }
268 
269 static int64_t blk_log_writes_getlength(BlockDriverState *bs)
270 {
271     return bdrv_getlength(bs->file->bs);
272 }
273 
274 static void blk_log_writes_child_perm(BlockDriverState *bs, BdrvChild *c,
275                                       BdrvChildRole role,
276                                       BlockReopenQueue *ro_q,
277                                       uint64_t perm, uint64_t shrd,
278                                       uint64_t *nperm, uint64_t *nshrd)
279 {
280     if (!c) {
281         *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
282         *nshrd = (shrd & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
283         return;
284     }
285 
286     bdrv_default_perms(bs, c, role, ro_q, perm, shrd,
287                        nperm, nshrd);
288 }
289 
290 static void blk_log_writes_refresh_limits(BlockDriverState *bs, Error **errp)
291 {
292     BDRVBlkLogWritesState *s = bs->opaque;
293     bs->bl.request_alignment = s->sectorsize;
294 }
295 
296 static int coroutine_fn
297 blk_log_writes_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes,
298                          QEMUIOVector *qiov, BdrvRequestFlags flags)
299 {
300     return bdrv_co_preadv(bs->file, offset, bytes, qiov, flags);
301 }
302 
303 typedef struct BlkLogWritesFileReq {
304     BlockDriverState *bs;
305     uint64_t offset;
306     uint64_t bytes;
307     int file_flags;
308     QEMUIOVector *qiov;
309     int (*func)(struct BlkLogWritesFileReq *r);
310     int file_ret;
311 } BlkLogWritesFileReq;
312 
313 typedef struct {
314     BlockDriverState *bs;
315     QEMUIOVector *qiov;
316     struct log_write_entry entry;
317     uint64_t zero_size;
318     int log_ret;
319 } BlkLogWritesLogReq;
320 
321 static void coroutine_fn blk_log_writes_co_do_log(BlkLogWritesLogReq *lr)
322 {
323     BDRVBlkLogWritesState *s = lr->bs->opaque;
324     uint64_t cur_log_offset = s->cur_log_sector << s->sectorbits;
325 
326     s->nr_entries++;
327     s->cur_log_sector +=
328             ROUND_UP(lr->qiov->size, s->sectorsize) >> s->sectorbits;
329 
330     lr->log_ret = bdrv_co_pwritev(s->log_file, cur_log_offset, lr->qiov->size,
331                                   lr->qiov, 0);
332 
333     /* Logging for the "write zeroes" operation */
334     if (lr->log_ret == 0 && lr->zero_size) {
335         cur_log_offset = s->cur_log_sector << s->sectorbits;
336         s->cur_log_sector +=
337                 ROUND_UP(lr->zero_size, s->sectorsize) >> s->sectorbits;
338 
339         lr->log_ret = bdrv_co_pwrite_zeroes(s->log_file, cur_log_offset,
340                                             lr->zero_size, 0);
341     }
342 
343     /* Update super block on flush or every update interval */
344     if (lr->log_ret == 0 && ((lr->entry.flags & LOG_FLUSH_FLAG)
345         || (s->nr_entries % s->update_interval == 0)))
346     {
347         struct log_write_super super = {
348             .magic      = cpu_to_le64(WRITE_LOG_MAGIC),
349             .version    = cpu_to_le64(WRITE_LOG_VERSION),
350             .nr_entries = cpu_to_le64(s->nr_entries),
351             .sectorsize = cpu_to_le32(s->sectorsize),
352         };
353         void *zeroes = g_malloc0(s->sectorsize - sizeof(super));
354         QEMUIOVector qiov;
355 
356         qemu_iovec_init(&qiov, 2);
357         qemu_iovec_add(&qiov, &super, sizeof(super));
358         qemu_iovec_add(&qiov, zeroes, s->sectorsize - sizeof(super));
359 
360         lr->log_ret =
361             bdrv_co_pwritev(s->log_file, 0, s->sectorsize, &qiov, 0);
362         if (lr->log_ret == 0) {
363             lr->log_ret = bdrv_co_flush(s->log_file->bs);
364         }
365         qemu_iovec_destroy(&qiov);
366         g_free(zeroes);
367     }
368 }
369 
370 static void coroutine_fn blk_log_writes_co_do_file(BlkLogWritesFileReq *fr)
371 {
372     fr->file_ret = fr->func(fr);
373 }
374 
375 static int coroutine_fn
376 blk_log_writes_co_log(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
377                       QEMUIOVector *qiov, int flags,
378                       int (*file_func)(BlkLogWritesFileReq *r),
379                       uint64_t entry_flags, bool is_zero_write)
380 {
381     QEMUIOVector log_qiov;
382     size_t niov = qiov ? qiov->niov : 0;
383     BDRVBlkLogWritesState *s = bs->opaque;
384     BlkLogWritesFileReq fr = {
385         .bs         = bs,
386         .offset     = offset,
387         .bytes      = bytes,
388         .file_flags = flags,
389         .qiov       = qiov,
390         .func       = file_func,
391     };
392     BlkLogWritesLogReq lr = {
393         .bs             = bs,
394         .qiov           = &log_qiov,
395         .entry = {
396             .sector     = cpu_to_le64(offset >> s->sectorbits),
397             .nr_sectors = cpu_to_le64(bytes >> s->sectorbits),
398             .flags      = cpu_to_le64(entry_flags),
399             .data_len   = 0,
400         },
401         .zero_size = is_zero_write ? bytes : 0,
402     };
403     void *zeroes = g_malloc0(s->sectorsize - sizeof(lr.entry));
404 
405     assert((1 << s->sectorbits) == s->sectorsize);
406     assert(bs->bl.request_alignment == s->sectorsize);
407     assert(QEMU_IS_ALIGNED(offset, bs->bl.request_alignment));
408     assert(QEMU_IS_ALIGNED(bytes, bs->bl.request_alignment));
409 
410     qemu_iovec_init(&log_qiov, niov + 2);
411     qemu_iovec_add(&log_qiov, &lr.entry, sizeof(lr.entry));
412     qemu_iovec_add(&log_qiov, zeroes, s->sectorsize - sizeof(lr.entry));
413     if (qiov) {
414         qemu_iovec_concat(&log_qiov, qiov, 0, qiov->size);
415     }
416 
417     blk_log_writes_co_do_file(&fr);
418     blk_log_writes_co_do_log(&lr);
419 
420     qemu_iovec_destroy(&log_qiov);
421     g_free(zeroes);
422 
423     if (lr.log_ret < 0) {
424         return lr.log_ret;
425     }
426 
427     return fr.file_ret;
428 }
429 
430 static int coroutine_fn
431 blk_log_writes_co_do_file_pwritev(BlkLogWritesFileReq *fr)
432 {
433     return bdrv_co_pwritev(fr->bs->file, fr->offset, fr->bytes,
434                            fr->qiov, fr->file_flags);
435 }
436 
437 static int coroutine_fn
438 blk_log_writes_co_do_file_pwrite_zeroes(BlkLogWritesFileReq *fr)
439 {
440     return bdrv_co_pwrite_zeroes(fr->bs->file, fr->offset, fr->bytes,
441                                  fr->file_flags);
442 }
443 
444 static int coroutine_fn blk_log_writes_co_do_file_flush(BlkLogWritesFileReq *fr)
445 {
446     return bdrv_co_flush(fr->bs->file->bs);
447 }
448 
449 static int coroutine_fn
450 blk_log_writes_co_do_file_pdiscard(BlkLogWritesFileReq *fr)
451 {
452     return bdrv_co_pdiscard(fr->bs->file, fr->offset, fr->bytes);
453 }
454 
455 static int coroutine_fn
456 blk_log_writes_co_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes,
457                           QEMUIOVector *qiov, BdrvRequestFlags flags)
458 {
459     return blk_log_writes_co_log(bs, offset, bytes, qiov, flags,
460                                  blk_log_writes_co_do_file_pwritev, 0, false);
461 }
462 
463 static int coroutine_fn
464 blk_log_writes_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
465                                 int64_t bytes, BdrvRequestFlags flags)
466 {
467     return blk_log_writes_co_log(bs, offset, bytes, NULL, flags,
468                                  blk_log_writes_co_do_file_pwrite_zeroes, 0,
469                                  true);
470 }
471 
472 static int coroutine_fn blk_log_writes_co_flush_to_disk(BlockDriverState *bs)
473 {
474     return blk_log_writes_co_log(bs, 0, 0, NULL, 0,
475                                  blk_log_writes_co_do_file_flush,
476                                  LOG_FLUSH_FLAG, false);
477 }
478 
479 static int coroutine_fn
480 blk_log_writes_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes)
481 {
482     return blk_log_writes_co_log(bs, offset, bytes, NULL, 0,
483                                  blk_log_writes_co_do_file_pdiscard,
484                                  LOG_DISCARD_FLAG, false);
485 }
486 
487 static const char *const blk_log_writes_strong_runtime_opts[] = {
488     "log-append",
489     "log-sector-size",
490 
491     NULL
492 };
493 
494 static BlockDriver bdrv_blk_log_writes = {
495     .format_name            = "blklogwrites",
496     .instance_size          = sizeof(BDRVBlkLogWritesState),
497 
498     .bdrv_open              = blk_log_writes_open,
499     .bdrv_close             = blk_log_writes_close,
500     .bdrv_getlength         = blk_log_writes_getlength,
501     .bdrv_child_perm        = blk_log_writes_child_perm,
502     .bdrv_refresh_limits    = blk_log_writes_refresh_limits,
503 
504     .bdrv_co_preadv         = blk_log_writes_co_preadv,
505     .bdrv_co_pwritev        = blk_log_writes_co_pwritev,
506     .bdrv_co_pwrite_zeroes  = blk_log_writes_co_pwrite_zeroes,
507     .bdrv_co_flush_to_disk  = blk_log_writes_co_flush_to_disk,
508     .bdrv_co_pdiscard       = blk_log_writes_co_pdiscard,
509 
510     .is_filter              = true,
511     .strong_runtime_opts    = blk_log_writes_strong_runtime_opts,
512 };
513 
514 static void bdrv_blk_log_writes_init(void)
515 {
516     bdrv_register(&bdrv_blk_log_writes);
517 }
518 
519 block_init(bdrv_blk_log_writes_init);
520