4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or https://opensource.org/licenses/CDDL-1.0.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2018, 2019 by Delphix. All rights reserved.
26 #include <sys/dmu_objset.h>
27 #include <sys/metaslab.h>
28 #include <sys/metaslab_impl.h>
30 #include <sys/spa_impl.h>
31 #include <sys/spa_log_spacemap.h>
32 #include <sys/vdev_impl.h>
38 * Log space maps are an optimization in ZFS metadata allocations for pools
39 * whose workloads are primarily random-writes. Random-write workloads are also
40 * typically random-free, meaning that they are freeing from locations scattered
41 * throughout the pool. This means that each TXG we will have to append some
42 * FREE records to almost every metaslab. With log space maps, we hold their
43 * changes in memory and log them altogether in one pool-wide space map on-disk
44 * for persistence. As more blocks are accumulated in the log space maps and
45 * more unflushed changes are accounted in memory, we flush a selected group
46 * of metaslabs every TXG to relieve memory pressure and potential overheads
47 * when loading the pool. Flushing a metaslab to disk relieves memory as we
48 * flush any unflushed changes from memory to disk (i.e. the metaslab's space
49 * map) and saves import time by making old log space maps obsolete and
50 * eventually destroying them. [A log space map is said to be obsolete when all
51 * its entries have made it to their corresponding metaslab space maps].
53 * == On disk data structures used ==
55 * - The pool has a new feature flag and a new entry in the MOS. The feature
56 * is activated when we create the first log space map and remains active
57 * for the lifetime of the pool. The new entry in the MOS Directory [refer
58 * to DMU_POOL_LOG_SPACEMAP_ZAP] is populated with a ZAP whose key-value
59 * pairs are of the form <key: txg, value: log space map object for that txg>.
60 * This entry is our on-disk reference of the log space maps that exist in
61 * the pool for each TXG and it is used during import to load all the
62 * metaslab unflushed changes in memory. To see how this structure is first
63 * created and later populated refer to spa_generate_syncing_log_sm(). To see
64 * how it is used during import time refer to spa_ld_log_sm_metadata().
66 * - Each vdev has a new entry in its vdev_top_zap (see field
67 * VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS) which holds the msp_unflushed_txg of
68 * each metaslab in this vdev. This field is the on-disk counterpart of the
69 * in-memory field ms_unflushed_txg which tells us from which TXG and onwards
70 * the metaslab haven't had its changes flushed. During import, we use this
71 * to ignore any entries in the space map log that are for this metaslab but
72 * from a TXG before msp_unflushed_txg. At that point, we also populate its
73 * in-memory counterpart and from there both fields are updated every time
74 * we flush that metaslab.
76 * - A space map is created every TXG and, during that TXG, it is used to log
77 * all incoming changes (the log space map). When created, the log space map
78 * is referenced in memory by spa_syncing_log_sm and its object ID is inserted
79 * to the space map ZAP mentioned above. The log space map is closed at the
80 * end of the TXG and will be destroyed when it becomes fully obsolete. We
81 * know when a log space map has become obsolete by looking at the oldest
82 * (and smallest) ms_unflushed_txg in the pool. If the value of that is bigger
83 * than the log space map's TXG, then it means that there is no metaslab who
84 * doesn't have the changes from that log and we can therefore destroy it.
85 * [see spa_cleanup_old_sm_logs()].
87 * == Important in-memory structures ==
89 * - The per-spa field spa_metaslabs_by_flushed sorts all the metaslabs in
90 * the pool by their ms_unflushed_txg field. It is primarily used for three
91 * reasons. First of all, it is used during flushing where we try to flush
92 * metaslabs in-order from the oldest-flushed to the most recently flushed
93 * every TXG. Secondly, it helps us to lookup the ms_unflushed_txg of the
94 * oldest flushed metaslab to distinguish which log space maps have become
95 * obsolete and which ones are still relevant. Finally it tells us which
96 * metaslabs have unflushed changes in a pool where this feature was just
97 * enabled, as we don't immediately add all of the pool's metaslabs but we
98 * add them over time as they go through metaslab_sync(). The reason that
99 * we do that is to ease these pools into the behavior of the flushing
100 * algorithm (described later on).
102 * - The per-spa field spa_sm_logs_by_txg can be thought as the in-memory
103 * counterpart of the space map ZAP mentioned above. It's an AVL tree whose
104 * nodes represent the log space maps in the pool. This in-memory
105 * representation of log space maps in the pool sorts the log space maps by
106 * the TXG that they were created (which is also the TXG of their unflushed
107 * changes). It also contains the following extra information for each
109 * [1] The number of metaslabs that were last flushed on that TXG. This is
110 * important because if that counter is zero and this is the oldest
111 * log then it means that it is also obsolete.
112 * [2] The number of blocks of that space map. This field is used by the
113 * block heuristic of our flushing algorithm (described later on).
114 * It represents how many blocks of metadata changes ZFS had to write
115 * to disk for that TXG.
117 * - The per-spa field spa_log_summary is a list of entries that summarizes
118 * the metaslab and block counts of all the nodes of the spa_sm_logs_by_txg
119 * AVL tree mentioned above. The reason this exists is that our flushing
120 * algorithm (described later) tries to estimate how many metaslabs to flush
121 * in each TXG by iterating over all the log space maps and looking at their
122 * block counts. Summarizing that information means that don't have to
123 * iterate through each space map, minimizing the runtime overhead of the
124 * flushing algorithm which would be induced in syncing context. In terms of
125 * implementation the log summary is used as a queue:
126 * * we modify or pop entries from its head when we flush metaslabs
127 * * we modify or append entries to its tail when we sync changes.
129 * - Each metaslab has two new range trees that hold its unflushed changes,
130 * ms_unflushed_allocs and ms_unflushed_frees. These are always disjoint.
132 * == Flushing algorithm ==
134 * The decision of how many metaslabs to flush on a give TXG is guided by
137 * [1] The memory heuristic -
138 * We keep track of the memory used by the unflushed trees from all the
139 * metaslabs [see sus_memused of spa_unflushed_stats] and we ensure that it
140 * stays below a certain threshold which is determined by an arbitrary hard
141 * limit and an arbitrary percentage of the system's memory [see
142 * spa_log_exceeds_memlimit()]. When we see that the memory usage of the
143 * unflushed changes are passing that threshold, we flush metaslabs, which
144 * empties their unflushed range trees, reducing the memory used.
146 * [2] The block heuristic -
147 * We try to keep the total number of blocks in the log space maps in check
148 * so the log doesn't grow indefinitely and we don't induce a lot of overhead
149 * when loading the pool. At the same time we don't want to flush a lot of
150 * metaslabs too often as this would defeat the purpose of the log space map.
151 * As a result we set a limit in the amount of blocks that we think it's
152 * acceptable for the log space maps to have and try not to cross it.
153 * [see sus_blocklimit from spa_unflushed_stats].
155 * In order to stay below the block limit every TXG we have to estimate how
156 * many metaslabs we need to flush based on the current rate of incoming blocks
157 * and our history of log space map blocks. The main idea here is to answer
158 * the question of how many metaslabs do we need to flush in order to get rid
159 * at least an X amount of log space map blocks. We can answer this question
160 * by iterating backwards from the oldest log space map to the newest one
161 * and looking at their metaslab and block counts. At this point the log summary
162 * mentioned above comes handy as it reduces the amount of things that we have
163 * to iterate (even though it may reduce the preciseness of our estimates due
164 * to its aggregation of data). So with that in mind, we project the incoming
165 * rate of the current TXG into the future and attempt to approximate how many
166 * metaslabs would we need to flush from now in order to avoid exceeding our
167 * block limit in different points in the future (granted that we would keep
168 * flushing the same number of metaslabs for every TXG). Then we take the
169 * maximum number from all these estimates to be on the safe side. For the
170 * exact implementation details of algorithm refer to
171 * spa_estimate_metaslabs_to_flush.
175 * This is used as the block size for the space maps used for the
176 * log space map feature. These space maps benefit from a bigger
177 * block size as we expect to be writing a lot of data to them at
180 static const unsigned long zfs_log_sm_blksz = 1ULL << 17;
183 * Percentage of the overall system's memory that ZFS allows to be
184 * used for unflushed changes (e.g. the sum of size of all the nodes
185 * in the unflushed trees).
187 * Note that this value is calculated over 1000000 for finer granularity
188 * (thus the _ppm suffix; reads as "parts per million"). As an example,
189 * the default of 1000 allows 0.1% of memory to be used.
191 static uint64_t zfs_unflushed_max_mem_ppm = 1000;
194 * Specific hard-limit in memory that ZFS allows to be used for
197 static uint64_t zfs_unflushed_max_mem_amt = 1ULL << 30;
200 * The following tunable determines the number of blocks that can be used for
201 * the log space maps. It is expressed as a percentage of the total number of
202 * metaslabs in the pool (i.e. the default of 400 means that the number of log
203 * blocks is capped at 4 times the number of metaslabs).
205 * This value exists to tune our flushing algorithm, with higher values
206 * flushing metaslabs less often (doing less I/Os) per TXG versus lower values
207 * flushing metaslabs more aggressively with the upside of saving overheads
208 * when loading the pool. Another factor in this tradeoff is that flushing
209 * less often can potentially lead to better utilization of the metaslab space
210 * map's block size as we accumulate more changes per flush.
212 * Given that this tunable indirectly controls the flush rate (metaslabs
213 * flushed per txg) and that's why making it a percentage in terms of the
214 * number of metaslabs in the pool makes sense here.
216 * As a rule of thumb we default this tunable to 400% based on the following:
218 * 1] Assuming a constant flush rate and a constant incoming rate of log blocks
219 * it is reasonable to expect that the amount of obsolete entries changes
220 * linearly from txg to txg (e.g. the oldest log should have the most
221 * obsolete entries, and the most recent one the least). With this we could
222 * say that, at any given time, about half of the entries in the whole space
223 * map log are obsolete. Thus for every two entries for a metaslab in the
224 * log space map, only one of them is valid and actually makes it to the
225 * metaslab's space map.
227 * 2] Each entry in the log space map is guaranteed to be two words while
228 * entries in metaslab space maps are generally single-word.
229 * [an extra factor of 2 - 400% overall]
230 * 3] Even if [1] and [2] are slightly less than 2 each, we haven't taken into
231 * account any consolidation of segments from the log space map to the
232 * unflushed range trees nor their history (e.g. a segment being allocated,
233 * then freed, then allocated again means 3 log space map entries but 0
234 * metaslab space map entries). Depending on the workload, we've seen ~1.8
235 * non-obsolete log space map entries per metaslab entry, for a total of
236 * ~600%. Since most of these estimates though are workload dependent, we
237 * default on 400% to be conservative.
239 * Thus we could say that even in the worst
240 * case of [1] and [2], the factor should end up being 4.
242 * That said, regardless of the number of metaslabs in the pool we need to
243 * provide upper and lower bounds for the log block limit.
244 * [see zfs_unflushed_log_block_{min,max}]
246 static uint_t zfs_unflushed_log_block_pct = 400;
249 * If the number of metaslabs is small and our incoming rate is high, we could
250 * get into a situation that we are flushing all our metaslabs every TXG. Thus
251 * we always allow at least this many log blocks.
253 static uint64_t zfs_unflushed_log_block_min = 1000;
256 * If the log becomes too big, the import time of the pool can take a hit in
257 * terms of performance. Thus we have a hard limit in the size of the log in
260 static uint64_t zfs_unflushed_log_block_max = (1ULL << 17);
263 * Also we have a hard limit in the size of the log in terms of dirty TXGs.
265 static uint64_t zfs_unflushed_log_txg_max = 1000;
268 * Max # of rows allowed for the log_summary. The tradeoff here is accuracy and
269 * stability of the flushing algorithm (longer summary) vs its runtime overhead
270 * (smaller summary is faster to traverse).
272 static uint64_t zfs_max_logsm_summary_length = 10;
275 * Tunable that sets the lower bound on the metaslabs to flush every TXG.
277 * Setting this to 0 has no effect since if the pool is idle we won't even be
278 * creating log space maps and therefore we won't be flushing. On the other
279 * hand if the pool has any incoming workload our block heuristic will start
280 * flushing metaslabs anyway.
282 * The point of this tunable is to be used in extreme cases where we really
283 * want to flush more metaslabs than our adaptable heuristic plans to flush.
285 static uint64_t zfs_min_metaslabs_to_flush = 1;
288 * Tunable that specifies how far in the past do we want to look when trying to
289 * estimate the incoming log blocks for the current TXG.
291 * Setting this too high may not only increase runtime but also minimize the
292 * effect of the incoming rates from the most recent TXGs as we take the
293 * average over all the blocks that we walk
294 * [see spa_estimate_incoming_log_blocks].
296 static uint64_t zfs_max_log_walking = 5;
299 * This tunable exists solely for testing purposes. It ensures that the log
300 * spacemaps are not flushed and destroyed during export in order for the
301 * relevant log spacemap import code paths to be tested (effectively simulating
304 int zfs_keep_log_spacemaps_at_export = 0;
307 spa_estimate_incoming_log_blocks(spa_t *spa)
309 ASSERT3U(spa_sync_pass(spa), ==, 1);
310 uint64_t steps = 0, sum = 0;
311 for (spa_log_sm_t *sls = avl_last(&spa->spa_sm_logs_by_txg);
312 sls != NULL && steps < zfs_max_log_walking;
313 sls = AVL_PREV(&spa->spa_sm_logs_by_txg, sls)) {
314 if (sls->sls_txg == spa_syncing_txg(spa)) {
316 * skip the log created in this TXG as this would
317 * make our estimations inaccurate.
321 sum += sls->sls_nblocks;
324 return ((steps > 0) ? DIV_ROUND_UP(sum, steps) : 0);
328 spa_log_sm_blocklimit(spa_t *spa)
330 return (spa->spa_unflushed_stats.sus_blocklimit);
334 spa_log_sm_set_blocklimit(spa_t *spa)
336 if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) {
337 ASSERT0(spa_log_sm_blocklimit(spa));
341 uint64_t msdcount = 0;
342 for (log_summary_entry_t *e = list_head(&spa->spa_log_summary);
343 e; e = list_next(&spa->spa_log_summary, e))
344 msdcount += e->lse_msdcount;
346 uint64_t limit = msdcount * zfs_unflushed_log_block_pct / 100;
347 spa->spa_unflushed_stats.sus_blocklimit = MIN(MAX(limit,
348 zfs_unflushed_log_block_min), zfs_unflushed_log_block_max);
352 spa_log_sm_nblocks(spa_t *spa)
354 return (spa->spa_unflushed_stats.sus_nblocks);
358 * Ensure that the in-memory log space map structures and the summary
359 * have the same block and metaslab counts.
362 spa_log_summary_verify_counts(spa_t *spa)
364 ASSERT(spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP));
366 if ((zfs_flags & ZFS_DEBUG_LOG_SPACEMAP) == 0)
369 uint64_t ms_in_avl = avl_numnodes(&spa->spa_metaslabs_by_flushed);
371 uint64_t ms_in_summary = 0, blk_in_summary = 0;
372 for (log_summary_entry_t *e = list_head(&spa->spa_log_summary);
373 e; e = list_next(&spa->spa_log_summary, e)) {
374 ms_in_summary += e->lse_mscount;
375 blk_in_summary += e->lse_blkcount;
378 uint64_t ms_in_logs = 0, blk_in_logs = 0;
379 for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
380 sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls)) {
381 ms_in_logs += sls->sls_mscount;
382 blk_in_logs += sls->sls_nblocks;
385 VERIFY3U(ms_in_logs, ==, ms_in_summary);
386 VERIFY3U(ms_in_logs, ==, ms_in_avl);
387 VERIFY3U(blk_in_logs, ==, blk_in_summary);
388 VERIFY3U(blk_in_logs, ==, spa_log_sm_nblocks(spa));
392 summary_entry_is_full(spa_t *spa, log_summary_entry_t *e, uint64_t txg)
394 if (e->lse_end == txg)
396 if (e->lse_txgcount >= DIV_ROUND_UP(zfs_unflushed_log_txg_max,
397 zfs_max_logsm_summary_length))
399 uint64_t blocks_per_row = MAX(1,
400 DIV_ROUND_UP(spa_log_sm_blocklimit(spa),
401 zfs_max_logsm_summary_length));
402 return (blocks_per_row <= e->lse_blkcount);
406 * Update the log summary information to reflect the fact that a metaslab
407 * was flushed or destroyed (e.g due to device removal or pool export/destroy).
409 * We typically flush the oldest flushed metaslab so the first (and oldest)
410 * entry of the summary is updated. However if that metaslab is getting loaded
411 * we may flush the second oldest one which may be part of an entry later in
412 * the summary. Moreover, if we call into this function from metaslab_fini()
413 * the metaslabs probably won't be ordered by ms_unflushed_txg. Thus we ask
414 * for a txg as an argument so we can locate the appropriate summary entry for
418 spa_log_summary_decrement_mscount(spa_t *spa, uint64_t txg, boolean_t dirty)
421 * We don't track summary data for read-only pools and this function
422 * can be called from metaslab_fini(). In that case return immediately.
424 if (!spa_writeable(spa))
427 log_summary_entry_t *target = NULL;
428 for (log_summary_entry_t *e = list_head(&spa->spa_log_summary);
429 e != NULL; e = list_next(&spa->spa_log_summary, e)) {
430 if (e->lse_start > txg)
435 if (target == NULL || target->lse_mscount == 0) {
437 * We didn't find a summary entry for this metaslab. We must be
438 * at the teardown of a spa_load() attempt that got an error
439 * while reading the log space maps.
441 VERIFY3S(spa_load_state(spa), ==, SPA_LOAD_ERROR);
445 target->lse_mscount--;
447 target->lse_msdcount--;
451 * Update the log summary information to reflect the fact that we destroyed
452 * old log space maps. Since we can only destroy the oldest log space maps,
453 * we decrement the block count of the oldest summary entry and potentially
454 * destroy it when that count hits 0.
456 * This function is called after a metaslab is flushed and typically that
457 * metaslab is the oldest flushed, which means that this function will
458 * typically decrement the block count of the first entry of the summary and
459 * potentially free it if the block count gets to zero (its metaslab count
460 * should be zero too at that point).
462 * There are certain scenarios though that don't work exactly like that so we
463 * need to account for them:
465 * Scenario [1]: It is possible that after we flushed the oldest flushed
466 * metaslab and we destroyed the oldest log space map, more recent logs had 0
467 * metaslabs pointing to them so we got rid of them too. This can happen due
468 * to metaslabs being destroyed through device removal, or because the oldest
469 * flushed metaslab was loading but we kept flushing more recently flushed
470 * metaslabs due to the memory pressure of unflushed changes. Because of that,
471 * we always iterate from the beginning of the summary and if blocks_gone is
472 * bigger than the block_count of the current entry we free that entry (we
473 * expect its metaslab count to be zero), we decrement blocks_gone and on to
474 * the next entry repeating this procedure until blocks_gone gets decremented
475 * to 0. Doing this also works for the typical case mentioned above.
477 * Scenario [2]: The oldest flushed metaslab isn't necessarily accounted by
478 * the first (and oldest) entry in the summary. If the first few entries of
479 * the summary were only accounting metaslabs from a device that was just
480 * removed, then the current oldest flushed metaslab could be accounted by an
481 * entry somewhere in the middle of the summary. Moreover flushing that
482 * metaslab will destroy all the log space maps older than its ms_unflushed_txg
483 * because they became obsolete after the removal. Thus, iterating as we did
484 * for scenario [1] works out for this case too.
486 * Scenario [3]: At times we decide to flush all the metaslabs in the pool
487 * in one TXG (either because we are exporting the pool or because our flushing
488 * heuristics decided to do so). When that happens all the log space maps get
489 * destroyed except the one created for the current TXG which doesn't have
490 * any log blocks yet. As log space maps get destroyed with every metaslab that
491 * we flush, entries in the summary are also destroyed. This brings a weird
492 * corner-case when we flush the last metaslab and the log space map of the
493 * current TXG is in the same summary entry with other log space maps that
494 * are older. When that happens we are eventually left with this one last
495 * summary entry whose blocks are gone (blocks_gone equals the entry's block
496 * count) but its metaslab count is non-zero (because it accounts all the
497 * metaslabs in the pool as they all got flushed). Under this scenario we can't
498 * free this last summary entry as it's referencing all the metaslabs in the
499 * pool and its block count will get incremented at the end of this sync (when
500 * we close the syncing log space map). Thus we just decrement its current
501 * block count and leave it alone. In the case that the pool gets exported,
502 * its metaslab count will be decremented over time as we call metaslab_fini()
503 * for all the metaslabs in the pool and the entry will be freed at
504 * spa_unload_log_sm_metadata().
507 spa_log_summary_decrement_blkcount(spa_t *spa, uint64_t blocks_gone)
509 log_summary_entry_t *e = list_head(&spa->spa_log_summary);
510 ASSERT3P(e, !=, NULL);
511 if (e->lse_txgcount > 0)
513 for (; e != NULL; e = list_head(&spa->spa_log_summary)) {
514 if (e->lse_blkcount > blocks_gone) {
515 e->lse_blkcount -= blocks_gone;
518 } else if (e->lse_mscount == 0) {
519 /* remove obsolete entry */
520 blocks_gone -= e->lse_blkcount;
521 list_remove(&spa->spa_log_summary, e);
522 kmem_free(e, sizeof (log_summary_entry_t));
524 /* Verify that this is scenario [3] mentioned above. */
525 VERIFY3U(blocks_gone, ==, e->lse_blkcount);
528 * Assert that this is scenario [3] further by ensuring
529 * that this is the only entry in the summary.
531 VERIFY3P(e, ==, list_tail(&spa->spa_log_summary));
532 ASSERT3P(e, ==, list_head(&spa->spa_log_summary));
534 blocks_gone = e->lse_blkcount = 0;
540 * Ensure that there is no way we are trying to remove more blocks
541 * than the # of blocks in the summary.
543 ASSERT0(blocks_gone);
547 spa_log_sm_decrement_mscount(spa_t *spa, uint64_t txg)
549 spa_log_sm_t target = { .sls_txg = txg };
550 spa_log_sm_t *sls = avl_find(&spa->spa_sm_logs_by_txg,
555 * We must be at the teardown of a spa_load() attempt that
556 * got an error while reading the log space maps.
558 VERIFY3S(spa_load_state(spa), ==, SPA_LOAD_ERROR);
562 ASSERT(sls->sls_mscount > 0);
567 spa_log_sm_increment_current_mscount(spa_t *spa)
569 spa_log_sm_t *last_sls = avl_last(&spa->spa_sm_logs_by_txg);
570 ASSERT3U(last_sls->sls_txg, ==, spa_syncing_txg(spa));
571 last_sls->sls_mscount++;
575 summary_add_data(spa_t *spa, uint64_t txg, uint64_t metaslabs_flushed,
576 uint64_t metaslabs_dirty, uint64_t nblocks)
578 log_summary_entry_t *e = list_tail(&spa->spa_log_summary);
580 if (e == NULL || summary_entry_is_full(spa, e, txg)) {
581 e = kmem_zalloc(sizeof (log_summary_entry_t), KM_SLEEP);
582 e->lse_start = e->lse_end = txg;
584 list_insert_tail(&spa->spa_log_summary, e);
587 ASSERT3U(e->lse_start, <=, txg);
588 if (e->lse_end < txg) {
592 e->lse_mscount += metaslabs_flushed;
593 e->lse_msdcount += metaslabs_dirty;
594 e->lse_blkcount += nblocks;
598 spa_log_summary_add_incoming_blocks(spa_t *spa, uint64_t nblocks)
600 summary_add_data(spa, spa_syncing_txg(spa), 0, 0, nblocks);
604 spa_log_summary_add_flushed_metaslab(spa_t *spa, boolean_t dirty)
606 summary_add_data(spa, spa_syncing_txg(spa), 1, dirty ? 1 : 0, 0);
610 spa_log_summary_dirty_flushed_metaslab(spa_t *spa, uint64_t txg)
612 log_summary_entry_t *target = NULL;
613 for (log_summary_entry_t *e = list_head(&spa->spa_log_summary);
614 e != NULL; e = list_next(&spa->spa_log_summary, e)) {
615 if (e->lse_start > txg)
619 ASSERT3P(target, !=, NULL);
620 ASSERT3U(target->lse_mscount, !=, 0);
621 target->lse_msdcount++;
625 * This function attempts to estimate how many metaslabs should
626 * we flush to satisfy our block heuristic for the log spacemap
627 * for the upcoming TXGs.
629 * Specifically, it first tries to estimate the number of incoming
630 * blocks in this TXG. Then by projecting that incoming rate to
631 * future TXGs and using the log summary, it figures out how many
632 * flushes we would need to do for future TXGs individually to
633 * stay below our block limit and returns the maximum number of
634 * flushes from those estimates.
637 spa_estimate_metaslabs_to_flush(spa_t *spa)
639 ASSERT(spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP));
640 ASSERT3U(spa_sync_pass(spa), ==, 1);
641 ASSERT(spa_log_sm_blocklimit(spa) != 0);
644 * This variable contains the incoming rate that will be projected
645 * and used for our flushing estimates in the future.
647 uint64_t incoming = spa_estimate_incoming_log_blocks(spa);
650 * At any point in time this variable tells us how many
651 * TXGs in the future we are so we can make our estimations.
653 uint64_t txgs_in_future = 1;
656 * This variable tells us how much room do we have until we hit
657 * our limit. When it goes negative, it means that we've exceeded
658 * our limit and we need to flush.
660 * Note that since we start at the first TXG in the future (i.e.
661 * txgs_in_future starts from 1) we already decrement this
662 * variable by the incoming rate.
664 int64_t available_blocks =
665 spa_log_sm_blocklimit(spa) - spa_log_sm_nblocks(spa) - incoming;
667 int64_t available_txgs = zfs_unflushed_log_txg_max;
668 for (log_summary_entry_t *e = list_head(&spa->spa_log_summary);
669 e; e = list_next(&spa->spa_log_summary, e))
670 available_txgs -= e->lse_txgcount;
673 * This variable tells us the total number of flushes needed to
674 * keep the log size within the limit when we reach txgs_in_future.
676 uint64_t total_flushes = 0;
678 /* Holds the current maximum of our estimates so far. */
679 uint64_t max_flushes_pertxg = zfs_min_metaslabs_to_flush;
682 * For our estimations we only look as far in the future
683 * as the summary allows us.
685 for (log_summary_entry_t *e = list_head(&spa->spa_log_summary);
686 e; e = list_next(&spa->spa_log_summary, e)) {
689 * If there is still room before we exceed our limit
690 * then keep skipping TXGs accumulating more blocks
691 * based on the incoming rate until we exceed it.
693 if (available_blocks >= 0 && available_txgs >= 0) {
694 uint64_t skip_txgs = (incoming == 0) ?
695 available_txgs + 1 : MIN(available_txgs + 1,
696 (available_blocks / incoming) + 1);
697 available_blocks -= (skip_txgs * incoming);
698 available_txgs -= skip_txgs;
699 txgs_in_future += skip_txgs;
700 ASSERT3S(available_blocks, >=, -incoming);
701 ASSERT3S(available_txgs, >=, -1);
705 * At this point we're far enough into the future where
706 * the limit was just exceeded and we flush metaslabs
707 * based on the current entry in the summary, updating
708 * our available_blocks.
710 ASSERT(available_blocks < 0 || available_txgs < 0);
711 available_blocks += e->lse_blkcount;
712 available_txgs += e->lse_txgcount;
713 total_flushes += e->lse_msdcount;
716 * Keep the running maximum of the total_flushes that
717 * we've done so far over the number of TXGs in the
718 * future that we are. The idea here is to estimate
719 * the average number of flushes that we should do
720 * every TXG so that when we are that many TXGs in the
721 * future we stay under the limit.
723 max_flushes_pertxg = MAX(max_flushes_pertxg,
724 DIV_ROUND_UP(total_flushes, txgs_in_future));
726 return (max_flushes_pertxg);
730 spa_log_sm_memused(spa_t *spa)
732 return (spa->spa_unflushed_stats.sus_memused);
736 spa_log_exceeds_memlimit(spa_t *spa)
738 if (spa_log_sm_memused(spa) > zfs_unflushed_max_mem_amt)
741 uint64_t system_mem_allowed = ((physmem * PAGESIZE) *
742 zfs_unflushed_max_mem_ppm) / 1000000;
743 if (spa_log_sm_memused(spa) > system_mem_allowed)
750 spa_flush_all_logs_requested(spa_t *spa)
752 return (spa->spa_log_flushall_txg != 0);
756 spa_flush_metaslabs(spa_t *spa, dmu_tx_t *tx)
758 uint64_t txg = dmu_tx_get_txg(tx);
760 if (spa_sync_pass(spa) != 1)
763 if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
767 * If we don't have any metaslabs with unflushed changes
768 * return immediately.
770 if (avl_numnodes(&spa->spa_metaslabs_by_flushed) == 0)
774 * During SPA export we leave a few empty TXGs to go by [see
775 * spa_final_dirty_txg() to understand why]. For this specific
776 * case, it is important to not flush any metaslabs as that
777 * would dirty this TXG.
779 * That said, during one of these dirty TXGs that is less or
780 * equal to spa_final_dirty(), spa_unload() will request that
781 * we try to flush all the metaslabs for that TXG before
782 * exporting the pool, thus we ensure that we didn't get a
783 * request of flushing everything before we attempt to return
786 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg &&
787 !dmu_objset_is_dirty(spa_meta_objset(spa), txg) &&
788 !spa_flush_all_logs_requested(spa))
792 * We need to generate a log space map before flushing because this
793 * will set up the in-memory data (i.e. node in spa_sm_logs_by_txg)
794 * for this TXG's flushed metaslab count (aka sls_mscount which is
795 * manipulated in many ways down the metaslab_flush() codepath).
797 * That is not to say that we may generate a log space map when we
798 * don't need it. If we are flushing metaslabs, that means that we
799 * were going to write changes to disk anyway, so even if we were
800 * not flushing, a log space map would have been created anyway in
803 spa_generate_syncing_log_sm(spa, tx);
806 * This variable tells us how many metaslabs we want to flush based
807 * on the block-heuristic of our flushing algorithm (see block comment
808 * of log space map feature). We also decrement this as we flush
809 * metaslabs and attempt to destroy old log space maps.
811 uint64_t want_to_flush;
812 if (spa_flush_all_logs_requested(spa)) {
813 ASSERT3S(spa_state(spa), ==, POOL_STATE_EXPORTED);
814 want_to_flush = UINT64_MAX;
816 want_to_flush = spa_estimate_metaslabs_to_flush(spa);
819 /* Used purely for verification purposes */
820 uint64_t visited = 0;
823 * Ideally we would only iterate through spa_metaslabs_by_flushed
824 * using only one variable (curr). We can't do that because
825 * metaslab_flush() mutates position of curr in the AVL when
826 * it flushes that metaslab by moving it to the end of the tree.
827 * Thus we always keep track of the original next node of the
828 * current node (curr) in another variable (next).
830 metaslab_t *next = NULL;
831 for (metaslab_t *curr = avl_first(&spa->spa_metaslabs_by_flushed);
832 curr != NULL; curr = next) {
833 next = AVL_NEXT(&spa->spa_metaslabs_by_flushed, curr);
836 * If this metaslab has been flushed this txg then we've done
837 * a full circle over the metaslabs.
839 if (metaslab_unflushed_txg(curr) == txg)
843 * If we are done flushing for the block heuristic and the
844 * unflushed changes don't exceed the memory limit just stop.
846 if (want_to_flush == 0 && !spa_log_exceeds_memlimit(spa))
849 if (metaslab_unflushed_dirty(curr)) {
850 mutex_enter(&curr->ms_sync_lock);
851 mutex_enter(&curr->ms_lock);
852 metaslab_flush(curr, tx);
853 mutex_exit(&curr->ms_lock);
854 mutex_exit(&curr->ms_sync_lock);
855 if (want_to_flush > 0)
858 metaslab_unflushed_bump(curr, tx, B_FALSE);
862 ASSERT3U(avl_numnodes(&spa->spa_metaslabs_by_flushed), >=, visited);
864 spa_log_sm_set_blocklimit(spa);
868 * Close the log space map for this TXG and update the block counts
869 * for the log's in-memory structure and the summary.
872 spa_sync_close_syncing_log_sm(spa_t *spa)
874 if (spa_syncing_log_sm(spa) == NULL)
876 ASSERT(spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP));
878 spa_log_sm_t *sls = avl_last(&spa->spa_sm_logs_by_txg);
879 ASSERT3U(sls->sls_txg, ==, spa_syncing_txg(spa));
881 sls->sls_nblocks = space_map_nblocks(spa_syncing_log_sm(spa));
882 spa->spa_unflushed_stats.sus_nblocks += sls->sls_nblocks;
885 * Note that we can't assert that sls_mscount is not 0,
886 * because there is the case where the first metaslab
887 * in spa_metaslabs_by_flushed is loading and we were
888 * not able to flush any metaslabs the current TXG.
890 ASSERT(sls->sls_nblocks != 0);
892 spa_log_summary_add_incoming_blocks(spa, sls->sls_nblocks);
893 spa_log_summary_verify_counts(spa);
895 space_map_close(spa->spa_syncing_log_sm);
896 spa->spa_syncing_log_sm = NULL;
899 * At this point we tried to flush as many metaslabs as we
900 * can as the pool is getting exported. Reset the "flush all"
901 * so the last few TXGs before closing the pool can be empty
904 if (spa_flush_all_logs_requested(spa)) {
905 ASSERT3S(spa_state(spa), ==, POOL_STATE_EXPORTED);
906 spa->spa_log_flushall_txg = 0;
911 spa_cleanup_old_sm_logs(spa_t *spa, dmu_tx_t *tx)
913 objset_t *mos = spa_meta_objset(spa);
915 uint64_t spacemap_zap;
916 int error = zap_lookup(mos, DMU_POOL_DIRECTORY_OBJECT,
917 DMU_POOL_LOG_SPACEMAP_ZAP, sizeof (spacemap_zap), 1, &spacemap_zap);
918 if (error == ENOENT) {
919 ASSERT(avl_is_empty(&spa->spa_sm_logs_by_txg));
924 metaslab_t *oldest = avl_first(&spa->spa_metaslabs_by_flushed);
925 uint64_t oldest_flushed_txg = metaslab_unflushed_txg(oldest);
927 /* Free all log space maps older than the oldest_flushed_txg. */
928 for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
929 sls && sls->sls_txg < oldest_flushed_txg;
930 sls = avl_first(&spa->spa_sm_logs_by_txg)) {
931 ASSERT0(sls->sls_mscount);
932 avl_remove(&spa->spa_sm_logs_by_txg, sls);
933 space_map_free_obj(mos, sls->sls_sm_obj, tx);
934 VERIFY0(zap_remove_int(mos, spacemap_zap, sls->sls_txg, tx));
935 spa_log_summary_decrement_blkcount(spa, sls->sls_nblocks);
936 spa->spa_unflushed_stats.sus_nblocks -= sls->sls_nblocks;
937 kmem_free(sls, sizeof (spa_log_sm_t));
941 static spa_log_sm_t *
942 spa_log_sm_alloc(uint64_t sm_obj, uint64_t txg)
944 spa_log_sm_t *sls = kmem_zalloc(sizeof (*sls), KM_SLEEP);
945 sls->sls_sm_obj = sm_obj;
951 spa_generate_syncing_log_sm(spa_t *spa, dmu_tx_t *tx)
953 uint64_t txg = dmu_tx_get_txg(tx);
954 objset_t *mos = spa_meta_objset(spa);
956 if (spa_syncing_log_sm(spa) != NULL)
959 if (!spa_feature_is_enabled(spa, SPA_FEATURE_LOG_SPACEMAP))
962 uint64_t spacemap_zap;
963 int error = zap_lookup(mos, DMU_POOL_DIRECTORY_OBJECT,
964 DMU_POOL_LOG_SPACEMAP_ZAP, sizeof (spacemap_zap), 1, &spacemap_zap);
965 if (error == ENOENT) {
966 ASSERT(avl_is_empty(&spa->spa_sm_logs_by_txg));
969 spacemap_zap = zap_create(mos,
970 DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx);
971 VERIFY0(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
972 DMU_POOL_LOG_SPACEMAP_ZAP, sizeof (spacemap_zap), 1,
974 spa_feature_incr(spa, SPA_FEATURE_LOG_SPACEMAP, tx);
979 ASSERT3U(zap_lookup_int_key(mos, spacemap_zap, txg, &sm_obj),
981 sm_obj = space_map_alloc(mos, zfs_log_sm_blksz, tx);
982 VERIFY0(zap_add_int_key(mos, spacemap_zap, txg, sm_obj, tx));
983 avl_add(&spa->spa_sm_logs_by_txg, spa_log_sm_alloc(sm_obj, txg));
986 * We pass UINT64_MAX as the space map's representation size
987 * and SPA_MINBLOCKSHIFT as the shift, to make the space map
988 * accept any sorts of segments since there's no real advantage
989 * to being more restrictive (given that we're already going
990 * to be using 2-word entries).
992 VERIFY0(space_map_open(&spa->spa_syncing_log_sm, mos, sm_obj,
993 0, UINT64_MAX, SPA_MINBLOCKSHIFT));
995 spa_log_sm_set_blocklimit(spa);
999 * Find all the log space maps stored in the space map ZAP and sort
1000 * them by their TXG in spa_sm_logs_by_txg.
1003 spa_ld_log_sm_metadata(spa_t *spa)
1006 uint64_t spacemap_zap;
1008 ASSERT(avl_is_empty(&spa->spa_sm_logs_by_txg));
1010 error = zap_lookup(spa_meta_objset(spa), DMU_POOL_DIRECTORY_OBJECT,
1011 DMU_POOL_LOG_SPACEMAP_ZAP, sizeof (spacemap_zap), 1, &spacemap_zap);
1012 if (error == ENOENT) {
1013 /* the space map ZAP doesn't exist yet */
1015 } else if (error != 0) {
1016 spa_load_failed(spa, "spa_ld_log_sm_metadata(): failed at "
1017 "zap_lookup(DMU_POOL_DIRECTORY_OBJECT) [error %d]",
1024 for (zap_cursor_init(&zc, spa_meta_objset(spa), spacemap_zap);
1025 (error = zap_cursor_retrieve(&zc, &za)) == 0;
1026 zap_cursor_advance(&zc)) {
1027 uint64_t log_txg = zfs_strtonum(za.za_name, NULL);
1029 spa_log_sm_alloc(za.za_first_integer, log_txg);
1030 avl_add(&spa->spa_sm_logs_by_txg, sls);
1032 zap_cursor_fini(&zc);
1033 if (error != ENOENT) {
1034 spa_load_failed(spa, "spa_ld_log_sm_metadata(): failed at "
1035 "zap_cursor_retrieve(spacemap_zap) [error %d]",
1040 for (metaslab_t *m = avl_first(&spa->spa_metaslabs_by_flushed);
1041 m; m = AVL_NEXT(&spa->spa_metaslabs_by_flushed, m)) {
1042 spa_log_sm_t target = { .sls_txg = metaslab_unflushed_txg(m) };
1043 spa_log_sm_t *sls = avl_find(&spa->spa_sm_logs_by_txg,
1047 * At this point if sls is zero it means that a bug occurred
1048 * in ZFS the last time the pool was open or earlier in the
1049 * import code path. In general, we would have placed a
1050 * VERIFY() here or in this case just let the kernel panic
1051 * with NULL pointer dereference when incrementing sls_mscount,
1052 * but since this is the import code path we can be a bit more
1053 * lenient. Thus, for DEBUG bits we always cause a panic, while
1054 * in production we log the error and just fail the import.
1056 ASSERT(sls != NULL);
1058 spa_load_failed(spa, "spa_ld_log_sm_metadata(): bug "
1059 "encountered: could not find log spacemap for "
1060 "TXG %llu [error %d]",
1061 (u_longlong_t)metaslab_unflushed_txg(m), ENOENT);
1070 typedef struct spa_ld_log_sm_arg {
1073 } spa_ld_log_sm_arg_t;
1076 spa_ld_log_sm_cb(space_map_entry_t *sme, void *arg)
1078 uint64_t offset = sme->sme_offset;
1079 uint64_t size = sme->sme_run;
1080 uint32_t vdev_id = sme->sme_vdev;
1082 spa_ld_log_sm_arg_t *slls = arg;
1083 spa_t *spa = slls->slls_spa;
1085 vdev_t *vd = vdev_lookup_top(spa, vdev_id);
1088 * If the vdev has been removed (i.e. it is indirect or a hole)
1089 * skip this entry. The contents of this vdev have already moved
1092 if (!vdev_is_concrete(vd))
1095 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
1096 ASSERT(!ms->ms_loaded);
1099 * If we have already flushed entries for this TXG to this
1100 * metaslab's space map, then ignore it. Note that we flush
1101 * before processing any allocations/frees for that TXG, so
1102 * the metaslab's space map only has entries from *before*
1103 * the unflushed TXG.
1105 if (slls->slls_txg < metaslab_unflushed_txg(ms))
1108 switch (sme->sme_type) {
1110 range_tree_remove_xor_add_segment(offset, offset + size,
1111 ms->ms_unflushed_frees, ms->ms_unflushed_allocs);
1114 range_tree_remove_xor_add_segment(offset, offset + size,
1115 ms->ms_unflushed_allocs, ms->ms_unflushed_frees);
1118 panic("invalid maptype_t");
1121 if (!metaslab_unflushed_dirty(ms)) {
1122 metaslab_set_unflushed_dirty(ms, B_TRUE);
1123 spa_log_summary_dirty_flushed_metaslab(spa,
1124 metaslab_unflushed_txg(ms));
1130 spa_ld_log_sm_data(spa_t *spa)
1132 spa_log_sm_t *sls, *psls;
1136 * If we are not going to do any writes there is no need
1137 * to read the log space maps.
1139 if (!spa_writeable(spa))
1142 ASSERT0(spa->spa_unflushed_stats.sus_nblocks);
1143 ASSERT0(spa->spa_unflushed_stats.sus_memused);
1145 hrtime_t read_logs_starttime = gethrtime();
1147 /* Prefetch log spacemaps dnodes. */
1148 for (sls = avl_first(&spa->spa_sm_logs_by_txg); sls;
1149 sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls)) {
1150 dmu_prefetch(spa_meta_objset(spa), sls->sls_sm_obj,
1151 0, 0, 0, ZIO_PRIORITY_SYNC_READ);
1156 psls = sls = avl_first(&spa->spa_sm_logs_by_txg);
1157 while (sls != NULL) {
1158 /* Prefetch log spacemaps up to 16 TXGs or MBs ahead. */
1159 if (psls != NULL && pn < 16 &&
1160 (pn < 2 || ps < 2 * dmu_prefetch_max)) {
1161 error = space_map_open(&psls->sls_sm,
1162 spa_meta_objset(spa), psls->sls_sm_obj, 0,
1163 UINT64_MAX, SPA_MINBLOCKSHIFT);
1165 spa_load_failed(spa, "spa_ld_log_sm_data(): "
1166 "failed at space_map_open(obj=%llu) "
1168 (u_longlong_t)sls->sls_sm_obj, error);
1171 dmu_prefetch(spa_meta_objset(spa), psls->sls_sm_obj,
1172 0, 0, space_map_length(psls->sls_sm),
1173 ZIO_PRIORITY_ASYNC_READ);
1175 ps += space_map_length(psls->sls_sm);
1176 psls = AVL_NEXT(&spa->spa_sm_logs_by_txg, psls);
1180 /* Load TXG log spacemap into ms_unflushed_allocs/frees. */
1181 kpreempt(KPREEMPT_SYNC);
1182 ASSERT0(sls->sls_nblocks);
1183 sls->sls_nblocks = space_map_nblocks(sls->sls_sm);
1184 spa->spa_unflushed_stats.sus_nblocks += sls->sls_nblocks;
1185 summary_add_data(spa, sls->sls_txg,
1186 sls->sls_mscount, 0, sls->sls_nblocks);
1188 struct spa_ld_log_sm_arg vla = {
1190 .slls_txg = sls->sls_txg
1192 error = space_map_iterate(sls->sls_sm,
1193 space_map_length(sls->sls_sm), spa_ld_log_sm_cb, &vla);
1195 spa_load_failed(spa, "spa_ld_log_sm_data(): failed "
1196 "at space_map_iterate(obj=%llu) [error %d]",
1197 (u_longlong_t)sls->sls_sm_obj, error);
1202 ps -= space_map_length(sls->sls_sm);
1203 space_map_close(sls->sls_sm);
1205 sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls);
1207 /* Update log block limits considering just loaded. */
1208 spa_log_sm_set_blocklimit(spa);
1211 hrtime_t read_logs_endtime = gethrtime();
1213 "read %llu log space maps (%llu total blocks - blksz = %llu bytes) "
1214 "in %lld ms", (u_longlong_t)avl_numnodes(&spa->spa_sm_logs_by_txg),
1215 (u_longlong_t)spa_log_sm_nblocks(spa),
1216 (u_longlong_t)zfs_log_sm_blksz,
1217 (longlong_t)((read_logs_endtime - read_logs_starttime) / 1000000));
1221 for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
1222 sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls)) {
1224 space_map_close(sls->sls_sm);
1233 * Now that the metaslabs contain their unflushed changes:
1234 * [1] recalculate their actual allocated space
1235 * [2] recalculate their weights
1236 * [3] sum up the memory usage of their unflushed range trees
1237 * [4] optionally load them, if debug_load is set
1239 * Note that even in the case where we get here because of an
1240 * error (e.g. error != 0), we still want to update the fields
1241 * below in order to have a proper teardown in spa_unload().
1243 for (metaslab_t *m = avl_first(&spa->spa_metaslabs_by_flushed);
1244 m != NULL; m = AVL_NEXT(&spa->spa_metaslabs_by_flushed, m)) {
1245 mutex_enter(&m->ms_lock);
1246 m->ms_allocated_space = space_map_allocated(m->ms_sm) +
1247 range_tree_space(m->ms_unflushed_allocs) -
1248 range_tree_space(m->ms_unflushed_frees);
1250 vdev_t *vd = m->ms_group->mg_vd;
1251 metaslab_space_update(vd, m->ms_group->mg_class,
1252 range_tree_space(m->ms_unflushed_allocs), 0, 0);
1253 metaslab_space_update(vd, m->ms_group->mg_class,
1254 -range_tree_space(m->ms_unflushed_frees), 0, 0);
1256 ASSERT0(m->ms_weight & METASLAB_ACTIVE_MASK);
1257 metaslab_recalculate_weight_and_sort(m);
1259 spa->spa_unflushed_stats.sus_memused +=
1260 metaslab_unflushed_changes_memused(m);
1262 if (metaslab_debug_load && m->ms_sm != NULL) {
1263 VERIFY0(metaslab_load(m));
1264 metaslab_set_selected_txg(m, 0);
1266 mutex_exit(&m->ms_lock);
1273 spa_ld_unflushed_txgs(vdev_t *vd)
1275 spa_t *spa = vd->vdev_spa;
1276 objset_t *mos = spa_meta_objset(spa);
1278 if (vd->vdev_top_zap == 0)
1281 uint64_t object = 0;
1282 int error = zap_lookup(mos, vd->vdev_top_zap,
1283 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS,
1284 sizeof (uint64_t), 1, &object);
1285 if (error == ENOENT)
1287 else if (error != 0) {
1288 spa_load_failed(spa, "spa_ld_unflushed_txgs(): failed at "
1289 "zap_lookup(vdev_top_zap=%llu) [error %d]",
1290 (u_longlong_t)vd->vdev_top_zap, error);
1294 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
1295 metaslab_t *ms = vd->vdev_ms[m];
1298 metaslab_unflushed_phys_t entry;
1299 uint64_t entry_size = sizeof (entry);
1300 uint64_t entry_offset = ms->ms_id * entry_size;
1302 error = dmu_read(mos, object,
1303 entry_offset, entry_size, &entry, 0);
1305 spa_load_failed(spa, "spa_ld_unflushed_txgs(): "
1306 "failed at dmu_read(obj=%llu) [error %d]",
1307 (u_longlong_t)object, error);
1311 ms->ms_unflushed_txg = entry.msp_unflushed_txg;
1312 ms->ms_unflushed_dirty = B_FALSE;
1313 ASSERT(range_tree_is_empty(ms->ms_unflushed_allocs));
1314 ASSERT(range_tree_is_empty(ms->ms_unflushed_frees));
1315 if (ms->ms_unflushed_txg != 0) {
1316 mutex_enter(&spa->spa_flushed_ms_lock);
1317 avl_add(&spa->spa_metaslabs_by_flushed, ms);
1318 mutex_exit(&spa->spa_flushed_ms_lock);
1325 * Read all the log space map entries into their respective
1326 * metaslab unflushed trees and keep them sorted by TXG in the
1327 * SPA's metadata. In addition, setup all the metadata for the
1328 * memory and the block heuristics.
1331 spa_ld_log_spacemaps(spa_t *spa)
1335 spa_log_sm_set_blocklimit(spa);
1337 for (uint64_t c = 0; c < spa->spa_root_vdev->vdev_children; c++) {
1338 vdev_t *vd = spa->spa_root_vdev->vdev_child[c];
1339 error = spa_ld_unflushed_txgs(vd);
1344 error = spa_ld_log_sm_metadata(spa);
1349 * Note: we don't actually expect anything to change at this point
1350 * but we grab the config lock so we don't fail any assertions
1351 * when using vdev_lookup_top().
1353 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
1354 error = spa_ld_log_sm_data(spa);
1355 spa_config_exit(spa, SCL_CONFIG, FTAG);
1361 ZFS_MODULE_PARAM(zfs, zfs_, unflushed_max_mem_amt, U64, ZMOD_RW,
1362 "Specific hard-limit in memory that ZFS allows to be used for "
1363 "unflushed changes");
1365 ZFS_MODULE_PARAM(zfs, zfs_, unflushed_max_mem_ppm, U64, ZMOD_RW,
1366 "Percentage of the overall system memory that ZFS allows to be "
1367 "used for unflushed changes (value is calculated over 1000000 for "
1368 "finer granularity)");
1370 ZFS_MODULE_PARAM(zfs, zfs_, unflushed_log_block_max, U64, ZMOD_RW,
1371 "Hard limit (upper-bound) in the size of the space map log "
1372 "in terms of blocks.");
1374 ZFS_MODULE_PARAM(zfs, zfs_, unflushed_log_block_min, U64, ZMOD_RW,
1375 "Lower-bound limit for the maximum amount of blocks allowed in "
1376 "log spacemap (see zfs_unflushed_log_block_max)");
1378 ZFS_MODULE_PARAM(zfs, zfs_, unflushed_log_txg_max, U64, ZMOD_RW,
1379 "Hard limit (upper-bound) in the size of the space map log "
1380 "in terms of dirty TXGs.");
1382 ZFS_MODULE_PARAM(zfs, zfs_, unflushed_log_block_pct, UINT, ZMOD_RW,
1383 "Tunable used to determine the number of blocks that can be used for "
1384 "the spacemap log, expressed as a percentage of the total number of "
1385 "metaslabs in the pool (e.g. 400 means the number of log blocks is "
1386 "capped at 4 times the number of metaslabs)");
1388 ZFS_MODULE_PARAM(zfs, zfs_, max_log_walking, U64, ZMOD_RW,
1389 "The number of past TXGs that the flushing algorithm of the log "
1390 "spacemap feature uses to estimate incoming log blocks");
1392 ZFS_MODULE_PARAM(zfs, zfs_, keep_log_spacemaps_at_export, INT, ZMOD_RW,
1393 "Prevent the log spacemaps from being flushed and destroyed "
1394 "during pool export/destroy");
1397 ZFS_MODULE_PARAM(zfs, zfs_, max_logsm_summary_length, U64, ZMOD_RW,
1398 "Maximum number of rows allowed in the summary of the spacemap log");
1400 ZFS_MODULE_PARAM(zfs, zfs_, min_metaslabs_to_flush, U64, ZMOD_RW,
1401 "Minimum number of metaslabs to flush per dirty TXG");