docs: post-5.7 toxic extent handling

Signed-off-by: Zygo Blaxell <bees@furryterror.org>

docs/btrfs-kernel.md

@@ -67,8 +67,10 @@ These bugs are particularly popular among bees users, though not all are specifi
 | - | 5.18 | wrong superblock num_devices makes filesystem unmountable | 4.14.283, 4.19.247, 5.4.198, 5.10.121, 5.15.46, 5.17.14, 5.18.3, 5.19 and later | d201238ccd2f btrfs: repair super block num_devices automatically
 | 5.18 | 5.19 | parent transid verify failed during log tree replay after a crash during a rename operation | 5.18.18, 5.19.2, 6.0 and later | 723df2bcc9e1 btrfs: join running log transaction when logging new name
 | 5.12 | 6.0 | space cache corruption and potential double allocations | 5.15.65, 5.19.6, 6.0 and later | ced8ecf026fd btrfs: fix space cache corruption and potential double allocations
+| 6.0 | 6.5 | suboptimal allocation in multi-device filesystems due to chunk allocator regression | 6.1.60, 6.5.9, 6.6 and later | 8a540e990d7d btrfs: fix stripe length calculation for non-zoned data chunk allocation
 | 6.3, backported to 5.15.107, 6.1.24, 6.2.11 | 6.3 | vmalloc error, failed to allocate pages | 6.3.10, 6.4 and later.  Bug (f349b15e183d "mm: vmalloc: avoid warn_alloc noise caused by fatal signal" in v6.3-rc6) backported to 6.1.24, 6.2.11, and 5.15.107. | 95a301eefa82 mm/vmalloc: do not output a spurious warning when huge vmalloc() fails
 | 6.2 | 6.3 | `IGNORE_OFFSET` flag ignored in `LOGICAL_INO` ioctl | 6.2.16, 6.3.3, 6.4 and later | 0cad8f14d70c btrfs: fix backref walking not returning all inode refs
+| 6.10 | 6.11 | `adding refs to an existing tree ref`, `failed to run delayed ref`, then read-only | 6.11.10, 6.12 and later | 7d493a5ecc26 btrfs: fix incorrect comparison for delayed refs
 | 5.4 | - | kernel hang when multiple threads are running `LOGICAL_INO` and dedupe ioctl on the same extent | - | workaround: avoid doing that
 
 "Last bad kernel" refers to that version's last stable update from

docs/config.md

@@ -98,27 +98,72 @@ code files over and over, so it will need a smaller hash table than a
 backup server which has to refer to the oldest data on the filesystem
 every time a new client machine's data is added to the server.
 
-Scanning modes for multiple subvols
+Scanning modes
------------------------------------
+--------------
 
-The `--scan-mode` option affects how bees schedules worker threads
+The `--scan-mode` option affects how bees iterates over the filesystem,
-between subvolumes.  Scan modes are an experimental feature and will
+schedules extents for scanning, and tracks progress.
-likely be deprecated in favor of a better solution.
 
-Scan mode can be changed at any time by restarting bees with a different
+There are now two kinds of scan mode:  the legacy **subvol** scan modes,
-mode option.  Scan state tracking is the same for all of the currently
+and the new **extent** scan mode.
-implemented modes.  The difference between the modes is the order in
-which subvols are selected.
 
-If a filesystem has only one subvolume with data in it, then the
+Scan mode can be changed by restarting bees with a different scan mode
-`--scan-mode` option has no effect.  In this case, there is only one
+option.
-subvolume to scan, so worker threads will all scan that one.
 
-Within a subvol, there is a single optimal scan order:  files are scanned
+Extent scan mode:
-in ascending numerical inode order.  Each worker will scan a different
+
-inode to avoid having the threads contend with each other for locks.
+ * Works with 4.15 and later kernels.
-File data is read sequentially and in order, but old blocks from earlier
+ * Can estimate progress and provide an ETA.
-scans are skipped.
+ * Can optimize scanning order to dedupe large extents first.
+ * Cannot avoid modifying read-only subvols.
+ * Can keep up with frequent creation and deletion of snapshots.
+
+Subvol scan modes:
+
+ * Work with 4.14 and earlier kernels.
+ * Cannot estimate or report progress.
+ * Cannot optimize scanning order by extent size.
+ * Can avoid modifying read-only subvols (for `btrfs send` workaround).
+ * Have problems keeping up with snapshots created during a scan.
+
+The default scan mode is 4, "extent".
+
+If you are using bees for the first time on a filesystem with many
+existing snapshots, you should read about [snapshot gotchas](gotchas.md).
+
+Subvol scan modes
+-----------------
+
+Subvol scan modes are maintained for compatibility with existing
+installations, but will not be developed further.  New installations
+should use extent scan mode instead.
+
+The _quantity_ of text below detailing the shortcomings of each subvol
+scan mode should be informative all by itself.
+
+Subvol scan modes work on any kernel version supported by bees.  They
+are the only scan modes usable on kernel 4.14 and earlier.
+
+The difference between the subvol scan modes is the order in which the
+files from different subvols are fed into the scanner.  They all scan
+files in inode number order, from low to high offset within each inode,
+the same way that a program like `cat` would read files (but skipping
+over old data from earlier btrfs transactions).
+
+If a filesystem has only one subvolume with data in it, then all of
+the subvol scan modes are equivalent.  In this case, there is only one
+subvolume to scan, so every possible ordering of subvols is the same.
+
+The `--workaround-btrfs-send` option pauses scanning subvols that are
+read-only.  If the subvol is made read-write (e.g. with `btrfs prop set
+$subvol ro false`), or if the `--workaround-btrfs-send` option is removed,
+then the scan of that subvol is unpaused and dedupe proceeds normally.
+Space will only be recovered when the last read-only subvol is deleted.
+
+Subvol scan modes cannot efficiently or accurately calculate an ETA for
+completion or estimate progress through the data.  They simply request
+"the next new inode" from btrfs, and they are completed when btrfs says
+there is no next new inode.
 
 Between subvols, there are several scheduling algorithms with different
 trade-offs:
@@ -126,53 +171,99 @@ trade-offs:
 Scan mode 0, "lockstep", scans the same inode number in each subvol at
 close to the same time.  This is useful if the subvols are snapshots
 with a common ancestor, since the same inode number in each subvol will
-have similar or identical contents.  This maximizes the likelihood
+have similar or identical contents.  This maximizes the likelihood that
-that all of the references to a snapshot of a file are scanned at
+all of the references to a snapshot of a file are scanned at close to
-close to the same time, improving dedupe hit rate and possibly taking
+the same time, improving dedupe hit rate.  If the subvols are unrelated
-advantage of VFS caching in the Linux kernel.  If the subvols are
+(i.e. not snapshots of a single subvol) then this mode does not provide
-unrelated (i.e. not snapshots of a single subvol) then this mode does
+any significant advantage.  This mode uses smaller amounts of temporary
-not provide significant benefit over random selection.  This mode uses
+space for shorter periods of time when most subvols are snapshots.  When a
-smaller amounts of temporary space for shorter periods of time when most
+new snapshot is created, this mode will stop scanning other subvols and
-subvols are snapshots.  When a new snapshot is created, this mode will
+scan the new snapshot until the same inode number is reached in each
-stop scanning other subvols and scan the new snapshot until the same
+subvol, which will effectively stop dedupe temporarily as this data has
-inode number is reached in each subvol, which will effectively stop
+already been scanned and deduped in the other snapshots.
-dedupe temporarily as this data has already been scanned and deduped
-in the other snapshots.
 
-Scan mode 1, "independent", scans the next inode with new data in each
+Scan mode 1, "independent", scans the next inode with new data in
-subvol.  Each subvol's scanner shares inodes uniformly with all other
+each subvol.  There is no coordination between the subvols, other than
-subvol scanners until the subvol has no new inodes left.  This mode makes
+round-robin distribution of files from each subvol to each worker thread.
-continuous forward progress across the filesystem and provides average
+This mode makes continuous forward progress in all subvols.  When a new
-performance across a variety of workloads, but is slow to respond to new
+snapshot is created, previous subvol scans continue as before, but the
-data, and may spend a lot of time deduping short-lived subvols that will
+worker threads are now divided among one more subvol.
-soon be deleted when it is preferable to dedupe long-lived subvols that
-will be the origin of future snapshots.  When a new snapshot is created,
-previous subvol scans continue as before, but the time is now divided
-among one more subvol.
 
 Scan mode 2, "sequential", scans one subvol at a time, in numerical subvol
-ID order, processing each subvol completely before proceeding to the
+ID order, processing each subvol completely before proceeding to the next
-next subvol.  This avoids spending time scanning short-lived snapshots
+subvol.  This avoids spending time scanning short-lived snapshots that
-that will be deleted before they can be fully deduped (e.g. those used
+will be deleted before they can be fully deduped (e.g. those used for
-for `btrfs send`).  Scanning is concentrated on older subvols that are
+`btrfs send`).  Scanning starts on older subvols that are more likely
-more likely to be origin subvols for future snapshots, eliminating the
+to be origin subvols for future snapshots, eliminating the need to
-need to dedupe future snapshots separately.  This mode uses the largest
+dedupe future snapshots separately.  This mode uses the largest amount
-amount of temporary space for the longest time, and typically requires
+of temporary space for the longest time, and typically requires a larger
-a larger hash table to maintain dedupe hit rate.
+hash table to maintain dedupe hit rate.
 
 Scan mode 3, "recent", scans the subvols with the highest `min_transid`
 value first (i.e. the ones that were most recently completely scanned),
 then falls back to "independent" mode to break ties.  This interrupts
-long scans of old subvols to give a rapid dedupe response to new data,
+long scans of old subvols to give a rapid dedupe response to new data
-then returns to the old subvols after the new data is scanned.  It is
+in previously scanned subvols, then returns to the old subvols after
-useful for large filesystems with multiple active subvols and rotating
+the new data is scanned.
-snapshots, where the first-pass scan can take months, but new duplicate
-data appears every day.
 
-The default scan mode is 1, "independent".
+Extent scan mode
+----------------
 
-If you are using bees for the first time on a filesystem with many
+Scan mode 4, "extent", scans the extent tree instead of the subvol trees.
-existing snapshots, you should read about [snapshot gotchas](gotchas.md).
+Extent scan mode reads each extent once, regardless of the number of
+reflinks or snapshots.  It adapts to the creation of new snapshots
+immediately, without having to revisit old data.
+
+In the extent scan mode, extents are separated into multiple size tiers
+to prioritize large extents over small ones.  Deduping large extents
+keeps the metadata update cost low per block saved, resulting in faster
+dedupe at the start of a scan cycle.  This is important for maximizing
+performance in use cases where bees runs for a limited time, such as
+during an overnight maintenance window.
+
+Once the larger size tiers are completed, dedupe space recovery speeds
+slow down significantly.  It may be desirable to stop bees running once
+the larger size tiers are finished, then start bees running some time
+later after new data has appeared.
+
+Each extent is mapped in physical address order, and all extent references
+are submitted to the scanner at the same time, resulting in much better
+cache behavior and dedupe performance compared to the subvol scan modes.
+
+The "extent" scan mode is not usable on kernels before 4.15 because
+it relies on the `LOGICAL_INO_V2` ioctl added in that kernel release.
+When using bees with an older kernel, only subvol scan modes will work.
+
+Extents are divided into virtual subvols by size, using reserved btrfs
+subvol IDs 250..255.  The size tier groups are:
+ * 250: 32M+1 and larger
+ * 251: 8M+1..32M
+ * 252: 2M+1..8M
+ * 253: 512K+1..2M
+ * 254: 128K+1..512K
+ * 255: 128K and smaller (includes all compressed extents)
+
+Extent scan mode can efficiently calculate dedupe progress within
+the filesystem and estimate an ETA for completion within each size
+tier; however, the accuracy of the ETA can be questionable due to the
+non-uniform distribution of block addresses in a typical user filesystem.
+
+Older versions of bees do not recognize the virtual subvols, so running
+an old bees version after running a new bees version will reset the
+"extent" scan mode's progress in `beescrawl.dat` to the beginning.
+This may change in future bees releases, i.e. extent scans will store
+their checkpoint data somewhere else.
+
+The `--workaround-btrfs-send` option behaves differently in extent
+scan modes:  In extent scan mode, dedupe proceeds on all subvols that are
+read-write, but all subvols that are read-only are excluded from dedupe.
+Space will only be recovered when the last read-only subvol is deleted.
+
+During `btrfs send` all duplicate extents in the sent subvol will not be
+removed (the kernel will reject dedupe commands while send is active,
+and bees currently will not re-issue them after the send is complete).
+It may be preferable to terminate the bees process while running `btrfs
+send` in extent scan mode, and restart bees after the `send` is complete.
 
 Threads and load management
 ---------------------------

docs/event-counters.md

@@ -120,9 +120,12 @@ The `crawl` event group consists of operations related to scanning btrfs trees t
 
  * `crawl_again`: An inode crawl was restarted because the extent was already locked by another running crawl.
  * `crawl_blacklisted`: An extent was not scanned because it belongs to a blacklisted file.
- * `crawl_create`: A new subvol crawler was created.
+ * `crawl_create`: A new subvol or extent crawler was created.
- * `crawl_done`: One pass over all subvols on the filesystem was completed.
+ * `crawl_deferred_inode`: Two tasks attempted to scan the same inode at the same time, so one was deferred.
+ * `crawl_done`: One pass over a subvol was completed.
+ * `crawl_discard`: An extent that didn't match the crawler's size tier was discarded.
  * `crawl_empty`: A `TREE_SEARCH_V2` ioctl call failed or returned an empty set (usually because all data in the subvol was scanned).
+ * `crawl_extent`: The extent crawler queued all references to an extent for processing.
  * `crawl_fail`: A `TREE_SEARCH_V2` ioctl call failed.
  * `crawl_gen_high`: An extent item in the search results refers to an extent that is newer than the current crawl's `max_transid` allows.
  * `crawl_gen_low`: An extent item in the search results refers to an extent that is older than the current crawl's `min_transid` allows.
@@ -136,7 +139,10 @@ The `crawl` event group consists of operations related to scanning btrfs trees t
  * `crawl_push`: An extent item in the search results is suitable for scanning and deduplication.
  * `crawl_scan`: An extent item in the search results is submitted to `BeesContext::scan_forward` for scanning and deduplication.
  * `crawl_search`: A `TREE_SEARCH_V2` ioctl call was successful.
+ * `crawl_throttled`: Extent scan created too many work queue items and was prevented from creating any more.
+ * `crawl_tree_block`: Extent scan found and skipped a metadata tree block.
  * `crawl_unknown`: An extent item in the search results has an unrecognized type.
+ * `crawl_unthrottled`: Extent scan allowed to create work queue items again.
 
 dedup
 -----
@@ -162,6 +168,25 @@ The `exception` event group consists of C++ exceptions.  C++ exceptions are thro
  * `exception_caught`: Total number of C++ exceptions thrown and caught by a generic exception handler.
  * `exception_caught_silent`: Total number of "silent" C++ exceptions thrown and caught by a generic exception handler.  These are exceptions which are part of the correct and normal operation of bees.  The exceptions are logged at a lower log level.
 
+extent
+------
+
+The `extent` event group consists of events that occur within the extent scanner.
+
+ * `extent_deferred_inode`: A lock conflict was detected when two worker threads attempted to manipulate the same inode at the same time.
+ * `extent_empty`: A complete list of references to an extent was created but the list was empty, e.g. because all refs are in deleted inodes or snapshots.
+ * `extent_fail`: An ioctl call to `LOGICAL_INO` failed.
+ * `extent_forward`: An extent reference was submitted for scanning.
+ * `extent_mapped`: A complete map of references to an extent was created and added to the crawl queue.
+ * `extent_ok`: An ioctl call to `LOGICAL_INO` completed successfully.
+ * `extent_overflow`: A complete map of references to an extent exceeded `BEES_MAX_EXTENT_REF_COUNT`, so the extent was dropped.
+ * `extent_ref_missing`: An extent reference reported by `LOGICAL_INO` was not found by later `TREE_SEARCH_V2` calls.
+ * `extent_ref_ok`: One extent reference was queued for scanning.
+ * `extent_restart`: An extent reference was requeued to be scanned again after an active extent lock is released.
+ * `extent_retry`: An extent reference was requeued to be scanned again after an active inode lock is released.
+ * `extent_skip`: A 4K extent with more than 1000 refs was skipped.
+ * `extent_zero`: An ioctl call to `LOGICAL_INO` succeeded, but reported an empty list of extents.
+
 hash
 ----
 
@@ -180,24 +205,6 @@ The `hash` event group consists of operations related to the bees hash table.
  * `hash_insert`: A `(hash, address)` pair was inserted by `BeesHashTable::push_random_hash_addr`.
  * `hash_lookup`: The hash table was searched for `(hash, address)` pairs matching a given `hash`.
 
-inserted
---------
-
-The `inserted` event group consists of operations related to storing hash and address data in the hash table (i.e. the hash table client).
-
- * `inserted_block`: Total number of data block references scanned and inserted into the hash table.
- * `inserted_clobbered`: Total number of data block references scanned and eliminated from the filesystem.
-
-matched
--------
-
-The `matched` event group consists of events related to matching incoming data blocks against existing hash table entries.
-
- * `matched_0`: A data block was scanned, hash table entries found, but no matching data blocks on the filesytem located.
- * `matched_1_or_more`: A data block was scanned, hash table entries found, and one or more matching data blocks on the filesystem located.
- * `matched_2_or_more`: A data block was scanned, hash table entries found, and two or more matching data blocks on the filesystem located.
- * `matched_3_or_more`: A data block was scanned, hash table entries found, and three or more matching data blocks on the filesystem located.
-
 open
 ----
 
@@ -259,12 +266,26 @@ The `pairforward` event group consists of events related to extending matching b
  * `pairforward_try`: Started extending a pair of matching block ranges forward.
  * `pairforward_zero`: A pair of matching block ranges could not be extended backward by one block because the src block contained all zeros and was not compressed.
 
+progress
+--------
+
+The `progress` event group consists of events related to progress estimation.
+
+ * `progress_no_data_bg`: Failed to retrieve any data block groups from the filesystem.
+ * `progress_not_created`: A crawler for one size tier had not been created for the extent scanner.
+ * `progress_complete`: A crawler for one size tier has completed a scan.
+ * `progress_not_found`: The extent position for a crawler does not correspond to any block group.
+ * `progress_out_of_bg`: The extent position for a crawler does not correspond to any data block group.
+ * `progress_ok`: Table of progress and ETA created successfully.
+
 readahead
 ---------
 
 The `readahead` event group consists of events related to calls to `posix_fadvise`.
 
- * `readahead_ms`: Total time spent running `posix_fadvise(..., POSIX_FADV_WILLNEED)` aka `readahead()`.
+ * `readahead_clear`: Number of times the duplicate read cache was cleared.
+ * `readahead_skip`: Number of times a duplicate read was identified in the cache and skipped.
+ * `readahead_ms`: Total time spent emulating readahead in user-space (kernel readahead is not measured).
  * `readahead_unread_ms`: Total time spent running `posix_fadvise(..., POSIX_FADV_DONTNEED)`.
 
 replacedst
@@ -301,7 +322,7 @@ The `resolve` event group consists of operations related to translating a btrfs
  * `resolve_large`: The `LOGICAL_INO` ioctl returned more than 2730 results (the limit of the v1 ioctl).
  * `resolve_ms`: Total time spent in the `LOGICAL_INO` ioctl (i.e. wallclock time, not kernel CPU time).
  * `resolve_ok`: The `LOGICAL_INO` ioctl returned success.
- * `resolve_overflow`: The `LOGICAL_INO` ioctl returned more than 655050 extents (the limit of the v2 ioctl).
+ * `resolve_overflow`: The `LOGICAL_INO` ioctl returned 9999 or more extents (the limit configured in `bees.h`).
  * `resolve_toxic`: The `LOGICAL_INO` ioctl took more than 0.1 seconds of kernel CPU time.
 
 root
@@ -329,35 +350,38 @@ The `scan` event group consists of operations related to scanning incoming data.
 
  * `scan_blacklisted`: A blacklisted extent was passed to `scan_forward` and dropped.
  * `scan_block`: A block of data was scanned.
- * `scan_bump`: After deduping a block range, the scan pointer had to be moved past the end of the deduped byte range.
+ * `scan_compressed_no_dedup`: An extent that was compressed contained non-zero, non-duplicate data.
- * `scan_dup_block`: Number of duplicate blocks deduped.
+ * `scan_dup_block`: Number of duplicate block references deduped.
- * `scan_dup_hit`: A pair of duplicate block ranges was found and removed.
+ * `scan_dup_hit`: A pair of duplicate block ranges was found.
  * `scan_dup_miss`: A pair of duplicate blocks was found in the hash table but not in the filesystem.
- * `scan_eof`: Scan past EOF was attempted.
- * `scan_erase_redundant`: Blocks in the hash table were removed because they were removed from the filesystem by dedupe.
  * `scan_extent`: An extent was scanned (`scan_one_extent`).
- * `scan_extent_tiny`: An extent below 128K that was not the beginning or end of a file was scanned.  No action is currently taken for these--they are merely counted.
  * `scan_forward`: A logical byte range was scanned (`scan_forward`).
  * `scan_found`: An entry was found in the hash table matching a scanned block from the filesystem.
  * `scan_hash_hit`: A block was found on the filesystem corresponding to a block found in the hash table.
  * `scan_hash_miss`: A block was not found on the filesystem corresponding to a block found in the hash table.
- * `scan_hash_preinsert`: A block was prepared for insertion into the hash table.
+ * `scan_hash_preinsert`: A non-zero data block's hash was prepared for possible insertion into the hash table.
+ * `scan_hash_insert`: A non-zero data block's hash was inserted into the hash table.
  * `scan_hole`: A hole extent was found during scan and ignored.
  * `scan_interesting`: An extent had flags that were not recognized by bees and was ignored.
  * `scan_lookup`: A hash was looked up in the hash table.
  * `scan_malign`: A block being scanned matched a hash at EOF in the hash table, but the EOF was not aligned to a block boundary and the two blocks did not have the same length.
- * `scan_no_fd`: References to a block from the hash table were found, but a FD could not be opened.
- * `scan_no_rewrite`: All blocks in an extent were removed by dedupe (i.e. no copies).
  * `scan_push_front`: An entry in the hash table matched a duplicate block, so the entry was moved to the head of its LRU list.
  * `scan_reinsert`: A copied block's hash and block address was inserted into the hash table.
  * `scan_resolve_hit`: A block address in the hash table was successfully resolved to an open FD and offset pair.
  * `scan_resolve_zero`: A block address in the hash table was not resolved to any subvol/inode pair, so the corresponding hash table entry was removed.
  * `scan_rewrite`: A range of bytes in a file was copied, then the copy deduped over the original data.
+ * `scan_root_dead`: A deleted subvol was detected.
+ * `scan_seen_clear`: The list of recently scanned extents reached maximum size and was cleared.
+ * `scan_seen_erase`: An extent reference was modified by scan, so all future references to the extent must be scanned.
+ * `scan_seen_hit`: A scan was skipped because the same extent had recently been scanned.
+ * `scan_seen_insert`: An extent reference was not modified by scan and its hashes have been inserted into the hash table, so all future references to the extent can be ignored.
+ * `scan_seen_miss`: A scan was not skipped because the same extent had not recently been scanned (i.e. the extent was scanned normally).
+ * `scan_skip_bytes`: Nuisance dedupe or hole-punching would save less than half of the data in an extent.
+ * `scan_skip_ops`: Nuisance dedupe or hole-punching would require too many dedupe/copy/hole-punch operations in an extent.
  * `scan_toxic_hash`: A scanned block has the same hash as a hash table entry that is marked toxic.
  * `scan_toxic_match`: A hash table entry points to a block that is discovered to be toxic.
  * `scan_twice`: Two references to the same block have been found in the hash table.
- * `scan_zero_compressed`: An extent that was compressed and contained only zero bytes was found.
+ * `scan_zero`: A data block containing only zero bytes was detected.
- * `scan_zero_uncompressed`: A block that contained only zero bytes was found in an uncompressed extent.
 
 scanf
 -----
@@ -365,9 +389,10 @@ scanf
 The `scanf` event group consists of operations related to `BeesContext::scan_forward`.  This is the entry point where `crawl` schedules new data for scanning.
 
  * `scanf_deferred_extent`: Two tasks attempted to scan the same extent at the same time, so one was deferred.
- * `scanf_deferred_inode`: Two tasks attempted to scan the same inode at the same time, so one was deferred.
+ * `scanf_eof`: Scan past EOF was attempted.
  * `scanf_extent`: A btrfs extent item was scanned.
  * `scanf_extent_ms`: Total thread-seconds spent scanning btrfs extent items.
+ * `scanf_no_fd`: References to a block from the hash table were found, but a FD could not be opened.
  * `scanf_total`: A logical byte range of a file was scanned.
  * `scanf_total_ms`: Total thread-seconds spent scanning logical byte ranges.
 

docs/gotchas.md

@@ -205,7 +205,7 @@ Other Gotchas
 
 * bees avoids the [slow backrefs kernel bug](btrfs-kernel.md) by
   measuring the time required to perform `LOGICAL_INO` operations.
-  If an extent requires over 0.1 kernel CPU seconds to perform a
+  If an extent requires over 5.0 kernel CPU seconds to perform a
   `LOGICAL_INO` ioctl, then bees blacklists the extent and avoids
   referencing it in future operations.  In most cases, fewer than 0.1%
   of extents in a filesystem must be avoided this way.  This results

docs/missing.md

@@ -15,16 +15,9 @@ specific files (patches welcome).
 * PREALLOC extents and extents containing blocks filled with zeros will
 be replaced by holes.  There is no way to turn this off.
 
-* Consecutive runs of duplicate blocks that are less than 12K in length
+* The fundamental unit of deduplication is the extent _reference_, when
-can take 30% of the processing time while saving only 3% of the disk
+it should be the _extent_ itself.  This is an architectural limitation
-space.  There should be an option to just not bother with those, but it's
+that results in excess reads of extent data, even in the Extent scan mode.
-complicated by the btrfs requirement to always dedupe complete extents.
-
-* There is a lot of duplicate reading of blocks in snapshots.  bees will
-scan all snapshots at close to the same time to try to get better
-performance by caching, but really fixing this requires rewriting the
-crawler to scan the btrfs extent tree directly instead of the subvol
-FS trees.
 
 * Block reads are currently more allocation- and CPU-intensive than they
 should be, especially for filesystems on SSD where the IO overhead is
@@ -33,8 +26,9 @@ much smaller.  This is a problem for CPU-power-constrained environments
 
 * bees can currently fragment extents when required to remove duplicate
 blocks, but has no defragmentation capability yet.  When possible, bees
-will attempt to work with existing extent boundaries, but it will not
+will attempt to work with existing extent boundaries and choose the
-aggregate blocks together from multiple extents to create larger ones.
+largest fragments available, but it will not aggregate blocks together
+from multiple extents to create larger ones.
 
 * When bees fragments an extent, the copied data is compressed.  There
 is currently no way (other than by modifying the source) to select a

docs/options.md

@@ -47,6 +47,7 @@
   * Mode 1: independent
   * Mode 2: sequential
   * Mode 3: recent
+  * Mode 4: extent
 
  For details of the different scanning modes and the default value of
  this option, see [bees configuration](config.md).

include/crucible/btrfs-tree.h

@@ -64,11 +64,13 @@ namespace crucible {
 		/// @{ Extent items (EXTENT_ITEM)
 		uint64_t extent_begin() const;
 		uint64_t extent_end() const;
+		uint64_t extent_flags() const;
 		uint64_t extent_generation() const;
 		/// @}
 
 		/// @{ Root items
 		uint64_t root_flags() const;
+		uint64_t root_refs() const;
 		/// @}
 
 		/// @{ Root backref items.
@@ -108,7 +110,9 @@ namespace crucible {
 		virtual ~BtrfsTreeFetcher() = default;
 		BtrfsTreeFetcher(Fd new_fd);
 		void type(uint8_t type);
+		uint8_t type();
 		void tree(uint64_t tree);
+		uint64_t tree();
 		void transid(uint64_t min_transid, uint64_t max_transid = numeric_limits<uint64_t>::max());
 		/// Block size (sectorsize) of filesystem
 		uint64_t block_size() const;

include/crucible/fs.h

@@ -197,6 +197,10 @@ namespace crucible {
 
 		size_t m_buf_size;
 		set<BtrfsIoctlSearchHeader> m_result;
+
+		static thread_local size_t s_calls;
+		static thread_local size_t s_loops;
+		static thread_local size_t s_loops_empty;
 	};
 
 	ostream & operator<<(ostream &os, const btrfs_ioctl_search_key &key);

include/crucible/multilock.h

@@ -14,6 +14,7 @@ namespace crucible {
 		mutex m_mutex;
 		condition_variable m_cv;
 		map<string, size_t> m_counters;
+		bool m_do_locking = true;
 
 		class LockHandle {
 			const string m_type;
@@ -33,6 +34,7 @@ namespace crucible {
 		shared_ptr<LockHandle> get_lock_private(const string &type);
 	public:
 		static shared_ptr<LockHandle> get_lock(const string &type);
+		static void enable_locking(bool enabled);
 	};
 
 }

include/crucible/table.h

@@ -0,0 +1,106 @@
+#ifndef CRUCIBLE_TABLE_H
+#define CRUCIBLE_TABLE_H
+
+#include <functional>
+#include <limits>
+#include <map>
+#include <memory>
+#include <ostream>
+#include <sstream>
+#include <string>
+#include <vector>
+
+namespace crucible {
+	namespace Table {
+		using namespace std;
+
+		using Content = function<string(size_t width, size_t height)>;
+		const size_t endpos = numeric_limits<size_t>::max();
+
+		Content Fill(const char c);
+		Content Text(const string& s);
+
+		template <class T>
+		Content Number(const T& num)
+		{
+			ostringstream oss;
+			oss << num;
+			return Text(oss.str());
+		}
+
+		class Cell {
+			Content m_content;
+		public:
+			Cell(const Content &fn = [](size_t, size_t) { return string(); } );
+			Cell& operator=(const Content &fn);
+			string text(size_t width, size_t height) const;
+		};
+
+		class Dimension {
+			size_t m_next_pos = 0;
+			vector<size_t> m_elements;
+		friend class Table;
+			size_t at(size_t) const;
+		public:
+			size_t size() const;
+			size_t insert(size_t pos);
+			void erase(size_t pos);
+		};
+
+		class Table {
+			Dimension m_rows, m_cols;
+			map<pair<size_t, size_t>, Cell> m_cells;
+			string m_left = "|";
+			string m_mid = "|";
+			string m_right = "|";
+		public:
+			Dimension &rows();
+			const Dimension& rows() const;
+			Dimension &cols();
+			const Dimension& cols() const;
+			Cell& at(size_t row, size_t col);
+			const Cell& at(size_t row, size_t col) const;
+			template <class T> void insert_row(size_t pos, const T& container);
+			template <class T> void insert_col(size_t pos, const T& container);
+			void left(const string &s);
+			void mid(const string &s);
+			void right(const string &s);
+			const string& left() const;
+			const string& mid() const;
+			const string& right() const;
+		};
+
+		ostream& operator<<(ostream &os, const Table &table);
+
+		template <class T>
+		void
+		Table::insert_row(size_t pos, const T& container)
+		{
+			const auto new_pos = m_rows.insert(pos);
+			size_t col = 0;
+			for (const auto &i : container) {
+				if (col >= cols().size()) {
+					cols().insert(col);
+				}
+				at(new_pos, col++) = i;
+			}
+		}
+
+		template <class T>
+		void
+		Table::insert_col(size_t pos, const T& container)
+		{
+			const auto new_pos = m_cols.insert(pos);
+			size_t row = 0;
+			for (const auto &i : container) {
+				if (row >= rows().size()) {
+					rows().insert(row);
+				}
+				at(row++, new_pos) = i;
+			}
+		}
+
+	}
+}
+
+#endif // CRUCIBLE_TABLE_H

include/crucible/task.h

@@ -40,6 +40,9 @@ namespace crucible {
 		/// after the current instance exits.
 		void run() const;
 
+		/// Schedule task to run when no other Task is available.
+		void idle() const;
+
 		/// Schedule Task to run after this Task has run or
 		/// been destroyed.
 		void append(const Task &task) const;
@@ -163,9 +166,9 @@ namespace crucible {
 		/// (it is the ExclusionLock that owns the lock, so it can
 		/// be passed to other Tasks or threads, but this is not
 		/// recommended practice).
-		/// If not successful, current Task is appended to the
+		/// If not successful, the argument Task is appended to the
 		/// task that currently holds the lock.  Current task is
-		/// expected to release any other ExclusionLock
+		/// expected to immediately release any other ExclusionLock
 		/// objects it holds, and exit its Task function.
 		ExclusionLock try_lock(const Task &task);
 

lib/Makefile

@@ -17,6 +17,7 @@ CRUCIBLE_OBJS = \
 	path.o \
 	process.o \
 	string.o \
+	table.o \
 	task.o \
 	time.o \
 	uname.o \

lib/btrfs-tree.cc

@@ -22,6 +22,13 @@ namespace crucible {
 		return m_objectid + m_offset;
 	}
 
+	uint64_t
+	BtrfsTreeItem::extent_flags() const
+	{
+		THROW_CHECK1(invalid_argument, btrfs_search_type_ntoa(m_type), m_type == BTRFS_EXTENT_ITEM_KEY);
+		return btrfs_get_member(&btrfs_extent_item::flags, m_data);
+	}
+
 	uint64_t
 	BtrfsTreeItem::extent_generation() const
 	{
@@ -61,6 +68,13 @@ namespace crucible {
 		return btrfs_get_member(&btrfs_root_item::flags, m_data);
 	}
 
+	uint64_t
+	BtrfsTreeItem::root_refs() const
+	{
+		THROW_CHECK1(invalid_argument, btrfs_search_type_ntoa(m_type), m_type == BTRFS_ROOT_ITEM_KEY);
+		return btrfs_get_member(&btrfs_root_item::refs, m_data);
+	}
+
 	ostream &
 	operator<<(ostream &os, const BtrfsTreeItem &bti)
 	{
@@ -269,12 +283,24 @@ namespace crucible {
 		m_type = type;
 	}
 
+	uint8_t
+	BtrfsTreeFetcher::type()
+	{
+		return m_type;
+	}
+
 	void
 	BtrfsTreeFetcher::tree(uint64_t tree)
 	{
 		m_tree = tree;
 	}
 
+	uint64_t
+	BtrfsTreeFetcher::tree()
+	{
+		return m_tree;
+	}
+
 	void
 	BtrfsTreeFetcher::transid(uint64_t min_transid, uint64_t max_transid)
 	{

lib/fs.cc

@@ -159,12 +159,13 @@ namespace crucible {
 	{
 		THROW_CHECK1(invalid_argument, src_length, src_length > 0);
 		while (src_length > 0) {
-			off_t length = min(off_t(BTRFS_MAX_DEDUPE_LEN), src_length);
+			BtrfsExtentSame bes(src_fd, src_offset, src_length);
-			BtrfsExtentSame bes(src_fd, src_offset, length);
 			bes.add(dst_fd, dst_offset);
 			bes.do_ioctl();
-			auto status = bes.m_info.at(0).status;
+			const auto status = bes.m_info.at(0).status;
 			if (status == 0) {
+				const off_t length = bes.m_info.at(0).bytes_deduped;
+				THROW_CHECK0(invalid_argument, length > 0);
 				src_offset += length;
 				dst_offset += length;
 				src_length -= length;
@@ -333,7 +334,7 @@ namespace crucible {
 		btrfs_ioctl_logical_ino_args args = (btrfs_ioctl_logical_ino_args) {
 			.logical = m_logical,
 			.size = m_container_size,
-			.inodes = reinterpret_cast<uint64_t>(m_container.prepare(m_container_size)),
+			.inodes = reinterpret_cast<uintptr_t>(m_container.prepare(m_container_size)),
 		};
 		// We are still supporting building with old headers that don't have .flags yet
 		*(&args.reserved[0] + 3) = m_flags;
@@ -416,7 +417,7 @@ namespace crucible {
 	{
 		btrfs_ioctl_ino_path_args *p = static_cast<btrfs_ioctl_ino_path_args *>(this);
 		BtrfsDataContainer container(m_container_size);
-		fspath = reinterpret_cast<uint64_t>(container.prepare(m_container_size));
+		fspath = reinterpret_cast<uintptr_t>(container.prepare(m_container_size));
 		size = container.get_size();
 
 		m_paths.clear();
@@ -753,6 +754,10 @@ namespace crucible {
 		return offset + len;
 	}
 
+	thread_local size_t BtrfsIoctlSearchKey::s_calls = 0;
+	thread_local size_t BtrfsIoctlSearchKey::s_loops = 0;
+	thread_local size_t BtrfsIoctlSearchKey::s_loops_empty = 0;
+
 	bool
 	BtrfsIoctlSearchKey::do_ioctl_nothrow(int fd)
 	{
@@ -773,6 +778,12 @@ namespace crucible {
 			ioctl_ptr->buf_size = buf_size;
 			// Don't bother supporting V1.  Kernels that old have other problems.
 			int rv = ioctl(fd, BTRFS_IOC_TREE_SEARCH_V2, ioctl_arg.data());
+			++s_calls;
+			if (rv != 0 && errno == ENOENT) {
+				// If we are searching a tree that is deleted or no longer exists, just return an empty list
+				nr_items = 0;
+				break;
+			}
 			if (rv != 0 && errno != EOVERFLOW) {
 				return false;
 			}
@@ -794,6 +805,10 @@ namespace crucible {
 				buf_size *= 2;
 			}
 			// don't automatically raise the buf size higher than 64K, the largest possible btrfs item
+			++s_loops;
+			if (ioctl_ptr->key.nr_items == 0) {
+				++s_loops_empty;
+			}
 		} while (buf_size < 65536);
 
 		// ioctl changes nr_items, this has to be copied back

lib/multilock.cc

@@ -62,11 +62,22 @@ namespace crucible {
 		return rv;
 	}
 
+	static MultiLocker s_process_instance;
+
 	shared_ptr<MultiLocker::LockHandle>
 	MultiLocker::get_lock(const string &type)
 	{
-		static MultiLocker s_process_instance;
+		if (s_process_instance.m_do_locking) {
-		return s_process_instance.get_lock_private(type);
+			return s_process_instance.get_lock_private(type);
+		} else {
+			return shared_ptr<MultiLocker::LockHandle>();
+		}
+	}
+
+	void
+	MultiLocker::enable_locking(const bool enabled)
+	{
+		s_process_instance.m_do_locking = enabled;
 	}
 
 }

lib/table.cc

@@ -0,0 +1,254 @@
+#include "crucible/table.h"
+
+#include "crucible/string.h"
+
+namespace crucible {
+	namespace Table {
+		using namespace std;
+
+		Content
+		Fill(const char c)
+		{
+			return [=](size_t width, size_t height) -> string {
+				string rv;
+				while (height--) {
+					rv += string(width, c);
+					if (height) {
+						rv += "\n";
+					}
+				}
+				return rv;
+			};
+		}
+
+		Content
+		Text(const string &s)
+		{
+			return [=](size_t width, size_t height) -> string {
+				const auto lines = split("\n", s);
+				string rv;
+				size_t line_count = 0;
+				for (const auto &i : lines) {
+					if (line_count++) {
+						rv += "\n";
+					}
+					if (i.length() < width) {
+						rv += string(width - i.length(), ' ');
+					}
+					rv += i;
+				}
+				while (line_count < height) {
+					if (line_count++) {
+						rv += "\n";
+					}
+					rv += string(width, ' ');
+				}
+				return rv;
+			};
+		}
+
+		Content
+		Number(const string &s)
+		{
+			return [=](size_t width, size_t height) -> string {
+				const auto lines = split("\n", s);
+				string rv;
+				size_t line_count = 0;
+				for (const auto &i : lines) {
+					if (line_count++) {
+						rv += "\n";
+					}
+					if (i.length() < width) {
+						rv += string(width - i.length(), ' ');
+					}
+					rv += i;
+				}
+				while (line_count < height) {
+					if (line_count++) {
+						rv += "\n";
+					}
+					rv += string(width, ' ');
+				}
+				return rv;
+			};
+		}
+
+		Cell::Cell(const Content &fn) :
+			m_content(fn)
+		{
+		}
+
+		Cell&
+		Cell::operator=(const Content &fn)
+		{
+			m_content = fn;
+			return *this;
+		}
+
+		string
+		Cell::text(size_t width, size_t height) const
+		{
+			return m_content(width, height);
+		}
+
+		size_t
+		Dimension::size() const
+		{
+			return m_elements.size();
+		}
+
+		size_t
+		Dimension::insert(size_t pos)
+		{
+			++m_next_pos;
+			const auto insert_pos = min(m_elements.size(), pos);
+			const auto it = m_elements.begin() + insert_pos;
+			m_elements.insert(it, m_next_pos);
+			return insert_pos;
+		}
+
+		void
+		Dimension::erase(size_t pos)
+		{
+			const auto it = m_elements.begin() + min(m_elements.size(), pos);
+			m_elements.erase(it);
+		}
+
+		size_t
+		Dimension::at(size_t pos) const
+		{
+			return m_elements.at(pos);
+		}
+
+		Dimension&
+		Table::rows()
+		{
+			return m_rows;
+		};
+
+		const Dimension&
+		Table::rows() const
+		{
+			return m_rows;
+		};
+
+		Dimension&
+		Table::cols()
+		{
+			return m_cols;
+		};
+
+		const Dimension&
+		Table::cols() const
+		{
+			return m_cols;
+		};
+
+		const Cell&
+		Table::at(size_t row, size_t col) const
+		{
+			const auto row_idx = m_rows.at(row);
+			const auto col_idx = m_cols.at(col);
+			const auto found = m_cells.find(make_pair(row_idx, col_idx));
+			if (found == m_cells.end()) {
+				static const Cell s_empty(Fill('.'));
+				return s_empty;
+			}
+			return found->second;
+		};
+
+		Cell&
+		Table::at(size_t row, size_t col)
+		{
+			const auto row_idx = m_rows.at(row);
+			const auto col_idx = m_cols.at(col);
+			return m_cells[make_pair(row_idx, col_idx)];
+		};
+
+		static
+		pair<size_t, size_t>
+		text_size(const string &s)
+		{
+			const auto s_split = split("\n", s);
+			size_t width = 0;
+			for (const auto &i : s_split) {
+				width = max(width, i.length());
+			}
+			return make_pair(width, s_split.size());
+		}
+
+		ostream& operator<<(ostream &os, const Table &table)
+		{
+			const auto rows = table.rows().size();
+			const auto cols = table.cols().size();
+			vector<size_t> row_heights(rows, 1);
+			vector<size_t> col_widths(cols, 1);
+			// Get the size of all fixed- and minimum-sized content cells
+			for (size_t row = 0; row < table.rows().size(); ++row) {
+				vector<string> col_text;
+				for (size_t col = 0; col < table.cols().size(); ++col) {
+					col_text.push_back(table.at(row, col).text(0, 0));
+					const auto tsize = text_size(*col_text.rbegin());
+					row_heights[row] = max(row_heights[row], tsize.second);
+					col_widths[col] = max(col_widths[col], tsize.first);
+				}
+			}
+			// Render the table
+			for (size_t row = 0; row < table.rows().size(); ++row) {
+				vector<string> lines(row_heights[row], "");
+				for (size_t col = 0; col < table.cols().size(); ++col) {
+					const auto& table_cell = table.at(row, col);
+					const auto table_text = table_cell.text(col_widths[col], row_heights[row]);
+					auto col_lines = split("\n", table_text);
+					col_lines.resize(row_heights[row], "");
+					for (size_t line = 0; line < row_heights[row]; ++line) {
+						if (col > 0) {
+							lines[line] += table.mid();
+						}
+						lines[line] += col_lines[line];
+					}
+				}
+				for (const auto &line : lines) {
+					os << table.left() << line << table.right() << "\n";
+				}
+			}
+			return os;
+		}
+
+		void
+		Table::left(const string &s)
+		{
+			m_left = s;
+		}
+
+		void
+		Table::mid(const string &s)
+		{
+			m_mid = s;
+		}
+
+		void
+		Table::right(const string &s)
+		{
+			m_right = s;
+		}
+
+		const string&
+		Table::left() const
+		{
+			return m_left;
+		}
+
+		const string&
+		Table::mid() const
+		{
+			return m_mid;
+		}
+
+		const string&
+		Table::right() const
+		{
+			return m_right;
+		}
+	}
+}

lib/task.cc

@@ -124,6 +124,9 @@ namespace crucible {
 		/// instance at the end of TaskMaster's global queue.
 		void run();
 
+		/// Run the task when there are no more Tasks on the main queue.
+		void idle();
+
 		/// Execute task immediately in current thread if it is not already
 		/// executing in another thread; otherwise, append the current task
 		/// to itself to be executed immediately in the other thread.
@@ -150,6 +153,7 @@ namespace crucible {
 		mutex 					m_mutex;
 		condition_variable 			m_condvar;
 		TaskQueue				m_queue;
+		TaskQueue				m_idle_queue;
 		size_t					m_thread_max;
 		size_t					m_thread_min = 0;
 		set<TaskConsumerPtr>			m_threads;
@@ -184,6 +188,7 @@ namespace crucible {
 		TaskMasterState(size_t thread_max = thread::hardware_concurrency());
 
 		static void push_back(const TaskStatePtr &task);
+		static void push_back_idle(const TaskStatePtr &task);
 		static void push_front(TaskQueue &queue);
 		size_t get_queue_count();
 		size_t get_thread_count();
@@ -367,6 +372,17 @@ namespace crucible {
 		TaskMasterState::push_back(shared_from_this());
 	}
 
+	void
+	TaskState::idle()
+	{
+		unique_lock<mutex> lock(m_mutex);
+		if (m_run_now) {
+			return;
+		}
+		m_run_now = true;
+		TaskMasterState::push_back_idle(shared_from_this());
+	}
+
 	TaskMasterState::TaskMasterState(size_t thread_max) :
 		m_thread_max(thread_max),
 		m_configured_thread_max(thread_max),
@@ -410,6 +426,20 @@ namespace crucible {
 		s_tms->start_threads_nolock();
 	}
 
+	void
+	TaskMasterState::push_back_idle(const TaskStatePtr &task)
+	{
+		THROW_CHECK0(runtime_error, task);
+		unique_lock<mutex> lock(s_tms->m_mutex);
+		if (s_tms->m_cancelled) {
+			task->clear();
+			return;
+		}
+		s_tms->m_idle_queue.push_back(task);
+		s_tms->m_condvar.notify_all();
+		s_tms->start_threads_nolock();
+	}
+
 	void
 	TaskMasterState::push_front(TaskQueue &queue)
 	{
@@ -456,12 +486,26 @@ namespace crucible {
 	TaskMaster::print_queue(ostream &os)
 	{
 		unique_lock<mutex> lock(s_tms->m_mutex);
-		os << "Queue (size " << s_tms->m_queue.size() << "):" << endl;
+		auto queue_copy = s_tms->m_queue;
+		lock.unlock();
+		os << "Queue (size " << queue_copy.size() << "):" << endl;
 		size_t counter = 0;
-		for (auto i : s_tms->m_queue) {
+		for (auto i : queue_copy) {
 			os << "Queue #" << ++counter << " Task ID " << i->id() << " " << i->title() << endl;
 		}
-		return os << "Queue End" << endl;
+		os << "Queue End" << endl;
+
+		lock.lock();
+		queue_copy = s_tms->m_idle_queue;
+		lock.unlock();
+		os << "Idle (size " << queue_copy.size() << "):" << endl;
+		counter = 0;
+		for (const auto &i : queue_copy) {
+			os << "Idle #" << ++counter << " Task ID " << i->id() << " " << i->title() << endl;
+		}
+		os << "Idle End" << endl;
+
+		return os;
 	}
 
 	ostream &
@@ -583,6 +627,7 @@ namespace crucible {
 		m_cancelled = true;
 		decltype(m_queue) empty_queue;
 		m_queue.swap(empty_queue);
+		empty_queue.splice(empty_queue.end(), m_idle_queue);
 		m_condvar.notify_all();
 		lock.unlock();
 		TaskState::clear_queue(empty_queue);
@@ -682,6 +727,13 @@ namespace crucible {
 		m_task_state->run();
 	}
 
+	void
+	Task::idle() const
+	{
+		THROW_CHECK0(runtime_error, m_task_state);
+		m_task_state->idle();
+	}
+
 	void
 	Task::append(const Task &that) const
 	{
@@ -772,6 +824,9 @@ namespace crucible {
 			} else if (!master_copy->m_queue.empty()) {
 				m_current_task = *master_copy->m_queue.begin();
 				master_copy->m_queue.pop_front();
+			} else if (!master_copy->m_idle_queue.empty()) {
+				m_current_task = *master_copy->m_idle_queue.begin();
+				master_copy->m_idle_queue.pop_front();
 			} else {
 				master_copy->m_condvar.wait(lock);
 				continue;

src/bees-context.cc

@@ -98,6 +98,8 @@ BeesContext::dump_status()
 		TaskMaster::print_queue(ofs);
 #endif
 
+		ofs << get_progress();
+
 		ofs.close();
 
 		BEESNOTE("renaming status file '" << status_file << "'");
@@ -112,6 +114,20 @@ BeesContext::dump_status()
 	}
 }
 
+void
+BeesContext::set_progress(const string &str)
+{
+	unique_lock<mutex> lock(m_progress_mtx);
+	m_progress_str = str;
+}
+
+string
+BeesContext::get_progress()
+{
+	unique_lock<mutex> lock(m_progress_mtx);
+	return m_progress_str;
+}
+
 void
 BeesContext::show_progress()
 {
@@ -159,6 +175,8 @@ BeesContext::show_progress()
 			BEESLOGINFO("\ttid " << t.first << ": " << t.second);
 		}
 
+		// No need to log progress here, it is logged when set
+
 		lastStats = thisStats;
 	}
 }
@@ -182,7 +200,7 @@ BeesContext::home_fd()
 }
 
 bool
-BeesContext::is_root_ro(uint64_t root)
+BeesContext::is_root_ro(uint64_t const root)
 {
 	return roots()->is_root_ro(root);
 }
@@ -264,6 +282,7 @@ BeesContext::rewrite_file_range(const BeesFileRange &bfr)
 	// BEESLOG("BeesResolver br(..., " << bfr << ")");
 	BEESTRACE("BeesContext::rewrite_file_range calling BeesResolver " << bfr);
 	BeesResolver br(m_ctx, BeesAddress(bfr.fd(), bfr.begin()));
+	BEESTRACE("BeesContext::rewrite_file_range calling replace_src " << dup_bbd);
 	// BEESLOG("\treplace_src " << dup_bbd);
 	br.replace_src(dup_bbd);
 	BEESCOUNT(scan_rewrite);
@@ -291,24 +310,37 @@ BeesContext::rewrite_file_range(const BeesFileRange &bfr)
 	}
 }
 
-BeesFileRange
+struct BeesSeenRange {
+	uint64_t bytenr;
+	off_t offset;
+	off_t length;
+};
+
+static
+bool
+operator<(const BeesSeenRange &bsr1, const BeesSeenRange &bsr2)
+{
+	return tie(bsr1.bytenr, bsr1.offset, bsr1.length) < tie(bsr2.bytenr, bsr2.offset, bsr2.length);
+}
+
+static
+__attribute__((unused))
+ostream&
+operator<<(ostream &os, const BeesSeenRange &tup)
+{
+	return os << "BeesSeenRange { " << to_hex(tup.bytenr) << ", " << to_hex(tup.offset) << "+" << pretty(tup.length) << " }";
+}
+
+void
 BeesContext::scan_one_extent(const BeesFileRange &bfr, const Extent &e)
 {
 	BEESNOTE("Scanning " << pretty(e.size()) << " "
 		<< to_hex(e.begin()) << ".." << to_hex(e.end())
 		<< " " << name_fd(bfr.fd()) );
 	BEESTRACE("scan extent " << e);
+	BEESTRACE("scan bfr " << bfr);
 	BEESCOUNT(scan_extent);
 
-	// EXPERIMENT:  Don't bother with tiny extents unless they are the entire file.
-	// We'll take a tiny extent at BOF or EOF but not in between.
-	if (e.begin() && e.size() < 128 * 1024 && e.end() != Stat(bfr.fd()).st_size) {
-		BEESCOUNT(scan_extent_tiny);
-		// This doesn't work properly with the current architecture,
-		// so we don't do an early return here.
-		// return bfr;
-	}
-
 	// We keep moving this method around
 	auto m_ctx = shared_from_this();
 
@@ -328,7 +360,7 @@ BeesContext::scan_one_extent(const BeesFileRange &bfr, const Extent &e)
 	if (e.flags() & Extent::HOLE) {
 		// Nothing here, dispose of this early
 		BEESCOUNT(scan_hole);
-		return bfr;
+		return;
 	}
 
 	if (e.flags() & Extent::PREALLOC) {
@@ -347,38 +379,57 @@ BeesContext::scan_one_extent(const BeesFileRange &bfr, const Extent &e)
 		if (m_ctx->dedup(brp)) {
 			BEESCOUNT(dedup_prealloc_hit);
 			BEESCOUNTADD(dedup_prealloc_bytes, e.size());
-			return bfr;
+			return;
 		} else {
 			BEESCOUNT(dedup_prealloc_miss);
 		}
 	}
 
+	// If we already read this extent and inserted it into the hash table, no need to read it again
+	static mutex s_seen_mutex;
+	unique_lock<mutex> lock_seen(s_seen_mutex);
+	const BeesSeenRange tup = {
+		.bytenr = e.bytenr(),
+		.offset = e.offset(),
+		.length = e.size(),
+	};
+	static set<BeesSeenRange> s_seen;
+	if (s_seen.size() > BEES_MAX_EXTENT_REF_COUNT) {
+		s_seen.clear();
+		BEESCOUNT(scan_seen_clear);
+	}
+	const auto seen_rv = s_seen.find(tup) != s_seen.end();
+	if (!seen_rv) {
+		BEESCOUNT(scan_seen_miss);
+	} else {
+		// BEESLOGDEBUG("Skip " << tup << " " << e);
+		BEESCOUNT(scan_seen_hit);
+		return;
+	}
+	lock_seen.unlock();
+
 	// OK we need to read extent now
 	bees_readahead(bfr.fd(), bfr.begin(), bfr.size());
 
 	map<off_t, pair<BeesHash, BeesAddress>> insert_map;
-	set<off_t> noinsert_set;
+	set<off_t> dedupe_set;
-
+	set<off_t> zero_set;
-	// Hole handling
-	bool extent_compressed = e.flags() & FIEMAP_EXTENT_ENCODED;
-	bool extent_contains_zero = false;
-	bool extent_contains_nonzero = false;
-
-	// Need to replace extent
-	bool rewrite_extent = false;
 
 	// Pretty graphs
 	off_t block_count = ((e.size() + BLOCK_MASK_SUMS) & ~BLOCK_MASK_SUMS) / BLOCK_SIZE_SUMS;
 	BEESTRACE(e << " block_count " << block_count);
 	string bar(block_count, '#');
 
-	for (off_t next_p = e.begin(); next_p < e.end(); ) {
+	// List of dedupes found
+	list<BeesRangePair> dedupe_list;
+	list<BeesFileRange> copy_list;
+	list<pair<BeesHash, BeesAddress>> front_hash_list;
+	list<uint64_t> invalidate_addr_list;
 
-		// Guarantee forward progress
+	off_t next_p = e.begin();
-		off_t p = next_p;
+	for (off_t p = e.begin(); p < e.end(); p += BLOCK_SIZE_SUMS) {
-		next_p += BLOCK_SIZE_SUMS;
 
-		off_t bar_p = (p - e.begin()) / BLOCK_SIZE_SUMS;
+		const off_t bar_p = (p - e.begin()) / BLOCK_SIZE_SUMS;
 		BeesAddress addr(e, p);
 
 		// This extent should consist entirely of non-magic blocks
@@ -393,69 +444,68 @@ BeesContext::scan_one_extent(const BeesFileRange &bfr, const Extent &e)
 
 		// Calculate the hash first because it lets us shortcut on is_data_zero
 		BEESNOTE("scan hash " << bbd);
-		BeesHash hash = bbd.hash();
+		const BeesHash hash = bbd.hash();
+
+		// Weed out zero blocks
+		BEESNOTE("is_data_zero " << bbd);
+		const bool data_is_zero = bbd.is_data_zero();
+		if (data_is_zero) {
+			bar.at(bar_p) = '0';
+			zero_set.insert(p);
+			BEESCOUNT(scan_zero);
+			continue;
+		}
 
 		// Schedule this block for insertion if we decide to keep this extent.
 		BEESCOUNT(scan_hash_preinsert);
 		BEESTRACE("Pushing hash " << hash << " addr " << addr << " bbd " << bbd);
 		insert_map.insert(make_pair(p, make_pair(hash, addr)));
-		bar.at(bar_p) = 'R';
+		bar.at(bar_p) = 'i';
 
-		// Weed out zero blocks
+		// Ensure we fill in the entire insert_map without skipping any non-zero blocks
-		BEESNOTE("is_data_zero " << bbd);
+		if (p < next_p) continue;
-		bool extent_is_zero = bbd.is_data_zero();
-		if (extent_is_zero) {
-			bar.at(bar_p) = '0';
-			if (extent_compressed) {
-				if (!extent_contains_zero) {
-					// BEESLOG("compressed zero bbd " << bbd << "\n\tin extent " << e);
-				}
-				extent_contains_zero = true;
-				// Do not attempt to lookup hash of zero block
-				continue;
-			} else {
-				BEESLOGINFO("zero bbd " << bbd << "\n\tin extent " << e);
-				BEESCOUNT(scan_zero_uncompressed);
-				rewrite_extent = true;
-				break;
-			}
-		} else {
-			if (extent_contains_zero && !extent_contains_nonzero) {
-				// BEESLOG("compressed nonzero bbd " << bbd << "\n\tin extent " << e);
-			}
-			extent_contains_nonzero = true;
-		}
 
 		BEESNOTE("lookup hash " << bbd);
-		auto found = hash_table->find_cell(hash);
+		const auto found = hash_table->find_cell(hash);
 		BEESCOUNT(scan_lookup);
 
-		set<BeesResolver> resolved_addrs;
 		set<BeesAddress> found_addrs;
+		list<BeesAddress> ordered_addrs;
 
-		// We know that there is at least one copy of the data and where it is,
+		for (const auto &i : found) {
-		// but we don't want to do expensive LOGICAL_INO operations unless there
-		// are at least two distinct addresses to look at.
-		found_addrs.insert(addr);
-
-		for (auto i : found) {
 			BEESTRACE("found (hash, address): " << i);
 			BEESCOUNT(scan_found);
 
 			// Hash has to match
 			THROW_CHECK2(runtime_error, i.e_hash, hash, i.e_hash == hash);
 
+			// We know that there is at least one copy of the data and where it is.
+			// Filter out anything that can't possibly match before we pull out the
+			// LOGICAL_INO hammer.
 			BeesAddress found_addr(i.e_addr);
 
-#if 0
 			// If address already in hash table, move on to next extent.
-			// We've already seen this block and may have made additional references to it.
+			// Only extents that are scanned but not modified are inserted, so if there's
-			// The current extent is effectively "pinned" and can't be modified any more.
+			// a matching hash:address pair in the hash table:
+			// 1.  We have already scanned this extent.
+			// 2.  We may have already created references to this extent.
+			// 3.  We won't scan this extent again.
+			// The current extent is effectively "pinned" and can't be modified
+			// without rescanning all the existing references.
 			if (found_addr.get_physical_or_zero() == addr.get_physical_or_zero()) {
+				// No log message because this happens to many thousands of blocks
+				// when bees is interrupted.
+				// BEESLOGDEBUG("Found matching hash " << hash << " at same address " << addr << ", skipping " << bfr);
 				BEESCOUNT(scan_already);
-				return bfr;
+				return;
+			}
+
+			// Address is a duplicate.
+			// Check this early so we don't have duplicate counts.
+			if (!found_addrs.insert(found_addr).second) {
+				BEESCOUNT(scan_twice);
+				continue;
 			}
-#endif
 
 			// Block must have matching EOF alignment
 			if (found_addr.is_unaligned_eof() != addr.is_unaligned_eof()) {
@@ -463,12 +513,6 @@ BeesContext::scan_one_extent(const BeesFileRange &bfr, const Extent &e)
 				continue;
 			}
 
-			// Address is a duplicate
-			if (!found_addrs.insert(found_addr).second) {
-				BEESCOUNT(scan_twice);
-				continue;
-			}
-
 			// Hash is toxic
 			if (found_addr.is_toxic()) {
 				BEESLOGWARN("WORKAROUND: abandoned toxic match for hash " << hash << " addr " << found_addr << " matching bbd " << bbd);
@@ -476,201 +520,345 @@ BeesContext::scan_one_extent(const BeesFileRange &bfr, const Extent &e)
 				// Extents may become non-toxic so give them a chance to expire.
 				// hash_table->push_front_hash_addr(hash, found_addr);
 				BEESCOUNT(scan_toxic_hash);
-				return bfr;
+				return;
 			}
 
-			// Distinct address, go resolve it
+			// Put this address in the list without changing hash table order
-			bool abandon_extent = false;
+			ordered_addrs.push_back(found_addr);
-			catch_all([&]() {
+		}
-				BEESNOTE("resolving " << found_addr << " matched " << bbd);
-				BEESTRACE("resolving " << found_addr << " matched " << bbd);
-				BEESTRACE("BeesContext::scan_one_extent calling BeesResolver " << found_addr);
-				BeesResolver resolved(m_ctx, found_addr);
-				// Toxic extents are really toxic
-				if (resolved.is_toxic()) {
-					BEESLOGWARN("WORKAROUND: discovered toxic match at found_addr " << found_addr << " matching bbd " << bbd);
-					BEESCOUNT(scan_toxic_match);
-					// Make sure we never see this hash again.
-					// It has become toxic since it was inserted into the hash table.
-					found_addr.set_toxic();
-					hash_table->push_front_hash_addr(hash, found_addr);
-					abandon_extent = true;
-				} else if (!resolved.count()) {
-					BEESCOUNT(scan_resolve_zero);
-					// Didn't find anything, address is dead
-					BEESTRACE("matched hash " << hash << " addr " << addr << " count zero");
-					hash_table->erase_hash_addr(hash, found_addr);
-				} else {
-					resolved_addrs.insert(resolved);
-					BEESCOUNT(scan_resolve_hit);
-				}
-			});
 
-			if (abandon_extent) {
+		// Cheap filtering is now out of the way, now for some heavy lifting
-				return bfr;
+		for (auto found_addr : ordered_addrs) {
+			// Hash table says there's a matching block on the filesystem.
+			// Go find refs to it.
+			BEESNOTE("resolving " << found_addr << " matched " << bbd);
+			BEESTRACE("resolving " << found_addr << " matched " << bbd);
+			BEESTRACE("BeesContext::scan_one_extent calling BeesResolver " << found_addr);
+			BeesResolver resolved(m_ctx, found_addr);
+			// Toxic extents are really toxic
+			if (resolved.is_toxic()) {
+				BEESLOGWARN("WORKAROUND: discovered toxic match at found_addr " << found_addr << " matching bbd " << bbd);
+				BEESCOUNT(scan_toxic_match);
+				// Make sure we never see this hash again.
+				// It has become toxic since it was inserted into the hash table.
+				found_addr.set_toxic();
+				hash_table->push_front_hash_addr(hash, found_addr);
+				return;
+			} else if (!resolved.count()) {
+				BEESCOUNT(scan_resolve_zero);
+				// Didn't find a block at the table address, address is dead
+				BEESLOGDEBUG("Erasing stale addr " << addr << " hash " << hash);
+				hash_table->erase_hash_addr(hash, found_addr);
+				continue;
+			} else {
+				BEESCOUNT(scan_resolve_hit);
 			}
-		}
 
-		// This shouldn't happen (often), so let's count it separately
+			// `resolved` contains references to a block on the filesystem that still exists.
-		if (resolved_addrs.size() > 2) {
-			BEESCOUNT(matched_3_or_more);
-		}
-		if (resolved_addrs.size() > 1) {
-			BEESCOUNT(matched_2_or_more);
-		}
-
-		// No need to do all this unless there are two or more distinct matches
-		if (!resolved_addrs.empty()) {
 			bar.at(bar_p) = 'M';
-			BEESCOUNT(matched_1_or_more);
-			BEESTRACE("resolved_addrs.size() = " << resolved_addrs.size());
-			BEESNOTE("resolving " << resolved_addrs.size() << " matches for hash " << hash);
 
-			BeesFileRange replaced_bfr;
+			BEESNOTE("finding one match (out of " << resolved.count() << ") at " << resolved.addr() << " for " << bbd);
+			BEESTRACE("finding one match (out of " << resolved.count() << ") at " << resolved.addr() << " for " << bbd);
+			auto replaced_brp = resolved.replace_dst(bbd);
+			BeesFileRange &replaced_bfr = replaced_brp.second;
+			BEESTRACE("next_p " << to_hex(next_p) << " -> replaced_bfr " << replaced_bfr);
 
-			BeesAddress last_replaced_addr;
+			// If we did find a block, but not this hash, correct the hash table and move on
-			for (auto it = resolved_addrs.begin(); it != resolved_addrs.end(); ++it) {
+			if (resolved.found_hash()) {
-				// FIXME:  Need to terminate this loop on replace_dst exception condition
+				BEESCOUNT(scan_hash_hit);
-				// catch_all([&]() {
+			} else {
-					auto it_copy = *it;
+				BEESLOGDEBUG("Erasing stale hash " << hash << " addr " << resolved.addr());
-					BEESNOTE("finding one match (out of " << it_copy.count() << ") at " << it_copy.addr() << " for " << bbd);
+				hash_table->erase_hash_addr(hash, resolved.addr());
-					BEESTRACE("finding one match (out of " << it_copy.count() << ") at " << it_copy.addr() << " for " << bbd);
+				BEESCOUNT(scan_hash_miss);
-					replaced_bfr = it_copy.replace_dst(bbd);
+				continue;
-					BEESTRACE("next_p " << to_hex(next_p) << " -> replaced_bfr " << replaced_bfr);
-
-					// If we didn't find this hash where the hash table said it would be,
-					// correct the hash table.
-					if (it_copy.found_hash()) {
-						BEESCOUNT(scan_hash_hit);
-					} else {
-						// BEESLOGDEBUG("erase src hash " << hash << " addr " << it_copy.addr());
-						BEESCOUNT(scan_hash_miss);
-						hash_table->erase_hash_addr(hash, it_copy.addr());
-					}
-
-					if (it_copy.found_dup()) {
-						BEESCOUNT(scan_dup_hit);
-
-						// FIXME:  we will thrash if we let multiple references to identical blocks
-						// exist in the hash table.  Erase all but the last one.
-						if (last_replaced_addr) {
-							BEESLOGINFO("Erasing redundant hash " << hash << " addr " << last_replaced_addr);
-							hash_table->erase_hash_addr(hash, last_replaced_addr);
-							BEESCOUNT(scan_erase_redundant);
-						}
-						last_replaced_addr = it_copy.addr();
-
-						// Invalidate resolve cache so we can count refs correctly
-						m_ctx->invalidate_addr(it_copy.addr());
-						m_ctx->invalidate_addr(bbd.addr());
-
-						// Remove deduped blocks from insert map
-						THROW_CHECK0(runtime_error, replaced_bfr);
-						for (off_t ip = replaced_bfr.begin(); ip < replaced_bfr.end(); ip += BLOCK_SIZE_SUMS) {
-							BEESCOUNT(scan_dup_block);
-							noinsert_set.insert(ip);
-							if (ip >= e.begin() && ip < e.end()) {
-								off_t bar_p = (ip - e.begin()) / BLOCK_SIZE_SUMS;
-								bar.at(bar_p) = 'd';
-							}
-						}
-
-						// next_p may be past EOF so check p only
-						THROW_CHECK2(runtime_error, p, replaced_bfr, p < replaced_bfr.end());
-
-						BEESCOUNT(scan_bump);
-						next_p = replaced_bfr.end();
-					} else {
-						BEESCOUNT(scan_dup_miss);
-					}
-				// });
 			}
-			if (last_replaced_addr) {
+
-				// If we replaced extents containing the incoming addr,
+			// We found a block and it was a duplicate
-				// push the addr we kept to the front of the hash LRU.
+			if (resolved.found_dup()) {
-				hash_table->push_front_hash_addr(hash, last_replaced_addr);
+				THROW_CHECK0(runtime_error, replaced_bfr);
-				BEESCOUNT(scan_push_front);
+				BEESCOUNT(scan_dup_hit);
+
+				// Save this match.  If a better match is found later,
+				// it will be replaced.
+				dedupe_list.push_back(replaced_brp);
+
+				// Push matching block to front of LRU
+				front_hash_list.push_back(make_pair(hash, resolved.addr()));
+
+				// This is the block that matched in the replaced bfr
+				bar.at(bar_p) = '=';
+
+				// Invalidate resolve cache so we can count refs correctly
+				invalidate_addr_list.push_back(resolved.addr());
+				invalidate_addr_list.push_back(bbd.addr());
+
+				// next_p may be past EOF so check p only
+				THROW_CHECK2(runtime_error, p, replaced_bfr, p < replaced_bfr.end());
+
+				// We may find duplicate ranges of various lengths, so make sure
+				// we don't pick a smaller one
+				next_p = max(next_p, replaced_bfr.end());
+
+				// Stop after one dedupe is found.  If there's a longer matching range
+				// out there, we'll find a matching block after the end of this range,
+				// since the longer range is longer than this one.
+				break;
+			} else {
+				BEESCOUNT(scan_dup_miss);
 			}
-		} else {
-			BEESCOUNT(matched_0);
 		}
 	}
 
-	// If the extent was compressed and all zeros, nuke entire thing
+	bool force_insert = false;
-	if (!rewrite_extent && (extent_contains_zero && !extent_contains_nonzero)) {
+
-		rewrite_extent = true;
+	// We don't want to punch holes into compressed extents, unless:
-		BEESCOUNT(scan_zero_compressed);
+	// 1.  There was dedupe of non-zero blocks, so we always have to copy the rest of the extent
+	// 2.  The entire extent is zero and the whole thing can be replaced with a single hole
+	const bool extent_compressed = e.flags() & FIEMAP_EXTENT_ENCODED;
+	if (extent_compressed && dedupe_list.empty() && !insert_map.empty()) {
+		// BEESLOGDEBUG("Compressed extent with non-zero data and no dedupe, skipping");
+		BEESCOUNT(scan_compressed_no_dedup);
+		force_insert = true;
 	}
 
-	// If we deduped any blocks then we must rewrite the remainder of the extent
+	// FIXME:  dedupe_list contains a lot of overlapping matches.  Get rid of all but one.
-	if (!noinsert_set.empty()) {
+	list<BeesRangePair> dedupe_list_out;
-		rewrite_extent = true;
+	dedupe_list.sort([](const BeesRangePair &a, const BeesRangePair &b) {
+		return b.second.size() < a.second.size();
+	});
+	// Shorten each dedupe brp by removing any overlap with earlier (longer) extents in list
+	for (auto i : dedupe_list) {
+		bool insert_i = true;
+		BEESTRACE("i = " << i << " insert_i " << insert_i);
+		for (const auto &j : dedupe_list_out) {
+			BEESTRACE("j = " << j);
+			// No overlap, try next one
+			if (j.second.end() <= i.second.begin() || j.second.begin() >= i.second.end()) {
+				continue;
+			}
+			// j fully overlaps or is the same as i, drop i
+			if (j.second.begin() <= i.second.begin() && j.second.end() >= i.second.end()) {
+				insert_i = false;
+				break;
+			}
+			// i begins outside j, i ends inside j, remove the end of i
+			if (i.second.end() > j.second.begin() && i.second.begin() <= j.second.begin()) {
+				const auto delta = i.second.end() - j.second.begin();
+				if (delta == i.second.size()) {
+					insert_i = false;
+					break;
+				}
+				i.shrink_end(delta);
+				continue;
+			}
+			// i begins inside j, ends outside j, remove the begin of i
+			if (i.second.begin() < j.second.end() && i.second.end() >= j.second.end()) {
+				const auto delta = j.second.end() - i.second.begin();
+				if (delta == i.second.size()) {
+					insert_i = false;
+					break;
+				}
+				i.shrink_begin(delta);
+				continue;
+			}
+			// i fully overlaps j, split i into two parts, push the other part onto dedupe_list
+			if (j.second.begin() > i.second.begin() && j.second.end() < i.second.end()) {
+				auto other_i = i;
+				const auto end_left_delta = i.second.end() - j.second.begin();
+				const auto begin_right_delta = i.second.begin() - j.second.end();
+				i.shrink_end(end_left_delta);
+				other_i.shrink_begin(begin_right_delta);
+				dedupe_list.push_back(other_i);
+				continue;
+			}
+			// None of the sbove.  Oops!
+			THROW_CHECK0(runtime_error, false);
+		}
+		if (insert_i) {
+			dedupe_list_out.push_back(i);
+		}
+	}
+	dedupe_list = dedupe_list_out;
+	dedupe_list_out.clear();
+
+	// Count total dedupes
+	uint64_t bytes_deduped = 0;
+	for (const auto &i : dedupe_list) {
+		// Remove deduped blocks from insert map and zero map
+		for (off_t ip = i.second.begin(); ip < i.second.end(); ip += BLOCK_SIZE_SUMS) {
+			BEESCOUNT(scan_dup_block);
+			dedupe_set.insert(ip);
+			zero_set.erase(ip);
+		}
+		bytes_deduped += i.second.size();
 	}
 
-	// If we need to replace part of the extent, rewrite all instances of it
+	// Copy all blocks of the extent that were not deduped or zero, but don't copy an entire extent
-	if (rewrite_extent) {
+	uint64_t bytes_zeroed = 0;
-		bool blocks_rewritten = false;
+	if (!force_insert) {
 		BEESTRACE("Rewriting extent " << e);
 		off_t last_p = e.begin();
 		off_t p = last_p;
-		off_t next_p;
+		off_t next_p = last_p;
 		BEESTRACE("next_p " << to_hex(next_p) << " p " << to_hex(p) << " last_p " << to_hex(last_p));
 		for (next_p = e.begin(); next_p < e.end(); ) {
 			p = next_p;
-			next_p += BLOCK_SIZE_SUMS;
+			next_p = min(next_p + BLOCK_SIZE_SUMS, e.end());
 
-			// BEESLOG("noinsert_set.count(" << to_hex(p) << ") " << noinsert_set.count(p));
+			// Can't be both dedupe and zero
-			if (noinsert_set.count(p)) {
+			THROW_CHECK2(runtime_error, zero_set.count(p), dedupe_set.count(p), zero_set.count(p) + dedupe_set.count(p) < 2);
+			if (zero_set.count(p)) {
+				bytes_zeroed += next_p - p;
+			}
+			// BEESLOG("dedupe_set.count(" << to_hex(p) << ") " << dedupe_set.count(p));
+			if (dedupe_set.count(p)) {
 				if (p - last_p > 0) {
-					rewrite_file_range(BeesFileRange(bfr.fd(), last_p, p));
+					THROW_CHECK2(runtime_error, p, e.end(), p <= e.end());
-					blocks_rewritten = true;
+					copy_list.push_back(BeesFileRange(bfr.fd(), last_p, p));
 				}
 				last_p = next_p;
-			} else {
-				off_t bar_p = (p - e.begin()) / BLOCK_SIZE_SUMS;
-				bar.at(bar_p) = '+';
 			}
 		}
 		BEESTRACE("last");
-		if (next_p - last_p > 0) {
+		if (next_p > last_p) {
-			rewrite_file_range(BeesFileRange(bfr.fd(), last_p, next_p));
+			THROW_CHECK2(runtime_error, next_p, e.end(), next_p <= e.end());
-			blocks_rewritten = true;
+			copy_list.push_back(BeesFileRange(bfr.fd(), last_p, next_p));
-		}
-		if (blocks_rewritten) {
-			// Nothing left to insert, all blocks clobbered
-			insert_map.clear();
-		} else {
-			// BEESLOG("No blocks rewritten");
-			BEESCOUNT(scan_no_rewrite);
 		}
 	}
 
-	// We did not rewrite the extent and it contained data, so insert it.
+	// Don't copy an entire extent
-	for (auto i : insert_map) {
+	if (!bytes_zeroed && copy_list.size() == 1 && copy_list.begin()->size() == e.size()) {
-		off_t bar_p = (i.first - e.begin()) / BLOCK_SIZE_SUMS;
+		copy_list.clear();
-		BEESTRACE("e " << e << "bar_p = " << bar_p << " i.first-e.begin() " << i.first - e.begin() << " i.second " << i.second.first << ", " << i.second.second);
+	}
-		if (noinsert_set.count(i.first)) {
+
-			// FIXME:  we removed one reference to this copy.  Avoid thrashing?
+	// Count total copies
-			hash_table->erase_hash_addr(i.second.first, i.second.second);
+	uint64_t bytes_copied = 0;
-			// Block was clobbered, do not insert
+	for (const auto &i : copy_list) {
-			// Will look like 'Ddddd' because we skip deduped blocks
+		bytes_copied += i.size();
-			bar.at(bar_p) = 'D';
+	}
-			BEESCOUNT(inserted_clobbered);
+
+	BEESTRACE("bar: " << bar);
+
+	// Don't do nuisance dedupes part 1:  free more blocks than we create
+	THROW_CHECK3(runtime_error, bytes_copied, bytes_zeroed, bytes_deduped, bytes_copied >= bytes_zeroed);
+	const auto cost_copy = bytes_copied - bytes_zeroed;
+	const auto gain_dedupe = bytes_deduped + bytes_zeroed;
+	if (cost_copy > gain_dedupe) {
+		BEESLOGDEBUG("Too many bytes copied (" << pretty(bytes_copied) << ") for bytes deduped (" << pretty(bytes_deduped) << ") and holes punched (" << pretty(bytes_zeroed) << "), skipping extent");
+		BEESCOUNT(scan_skip_bytes);
+		force_insert = true;
+	}
+
+	// Don't do nuisance dedupes part 2:  nobody needs more than 100 dedupe/copy ops in one extent
+	if (dedupe_list.size() + copy_list.size() > 100) {
+		BEESLOGDEBUG("Too many dedupe (" << dedupe_list.size() << ") and copy (" << copy_list.size() << ") operations, skipping extent");
+		BEESCOUNT(scan_skip_ops);
+		force_insert = true;
+	}
+
+	// Track whether we rewrote anything
+	bool extent_modified = false;
+
+	// If we didn't delete the dedupe list, do the dedupes now
+	for (const auto &i : dedupe_list) {
+		BEESNOTE("dedup " << i);
+		if (force_insert || m_ctx->dedup(i)) {
+			BEESCOUNT(replacedst_dedup_hit);
+			THROW_CHECK0(runtime_error, i.second);
+			for (off_t ip = i.second.begin(); ip < i.second.end(); ip += BLOCK_SIZE_SUMS) {
+				if (ip >= e.begin() && ip < e.end()) {
+					off_t bar_p = (ip - e.begin()) / BLOCK_SIZE_SUMS;
+					if (bar.at(bar_p) != '=') {
+						if (ip == i.second.begin()) {
+							bar.at(bar_p) = '<';
+						} else if (ip + BLOCK_SIZE_SUMS >= i.second.end()) {
+							bar.at(bar_p) = '>';
+						} else {
+							bar.at(bar_p) = 'd';
+						}
+					}
+				}
+			}
+			extent_modified = !force_insert;
 		} else {
+			BEESLOGINFO("dedup failed: " << i);
+			BEESCOUNT(replacedst_dedup_miss);
+			// User data changed while we were looking up the extent, or we have a bug.
+			// We can't fix this, but we can immediately stop wasting effort.
+			return;
+		}
+	}
+
+	// Then the copy/rewrites
+	for (const auto &i : copy_list) {
+		if (!force_insert) {
+			rewrite_file_range(i);
+			extent_modified = true;
+		}
+		for (auto p = i.begin(); p < i.end(); p += BLOCK_SIZE_SUMS) {
+			off_t bar_p = (p - e.begin()) / BLOCK_SIZE_SUMS;
+			// Leave zeros as-is because they aren't really copies
+			if (bar.at(bar_p) != '0') {
+				bar.at(bar_p) = '+';
+			}
+		}
+	}
+
+	if (!force_insert) {
+		// Push matched hashes to front
+		for (const auto &i : front_hash_list) {
+			hash_table->push_front_hash_addr(i.first, i.second);
+			BEESCOUNT(scan_push_front);
+		}
+		// Invalidate cached resolves
+		for (const auto &i : invalidate_addr_list) {
+			m_ctx->invalidate_addr(i);
+		}
+	}
+
+	// Don't insert hashes pointing to an extent we just deleted
+	if (!extent_modified) {
+		// We did not rewrite the extent and it contained data, so insert it.
+		// BEESLOGDEBUG("Inserting " << insert_map.size() << " hashes from " << bfr);
+		for (const auto &i : insert_map) {
 			hash_table->push_random_hash_addr(i.second.first, i.second.second);
-			bar.at(bar_p) = '.';
+			off_t bar_p = (i.first - e.begin()) / BLOCK_SIZE_SUMS;
-			BEESCOUNT(inserted_block);
+			if (bar.at(bar_p) == 'i') {
+				bar.at(bar_p) = '.';
+			}
+			BEESCOUNT(scan_hash_insert);
 		}
 	}
 
 	// Visualize
 	if (bar != string(block_count, '.')) {
-		BEESLOGINFO("scan: " << pretty(e.size()) << " " << to_hex(e.begin()) << " [" << bar << "] " << to_hex(e.end()) << ' ' << name_fd(bfr.fd()));
+		BEESLOGINFO(
+			(force_insert ? "skip" : "scan") << ": "
+			<< pretty(e.size()) << " "
+			<< dedupe_list.size() << "d" << copy_list.size() << "c"
+			<< ((bytes_zeroed + BLOCK_SIZE_SUMS - 1) / BLOCK_SIZE_SUMS) << "p"
+			<< (extent_compressed ? "z {" : " {")
+			<< to_hex(e.bytenr()) << "+" << to_hex(e.offset()) << "} "
+			<< to_hex(e.begin()) << " [" << bar << "] " << to_hex(e.end())
+			<< ' ' << name_fd(bfr.fd())
+		);
 	}
 
-	// Costs 10% on benchmarks
+	// Put this extent into the recently seen list if we didn't rewrite it,
+	// and remove it if we did.
+	lock_seen.lock();
+	if (extent_modified) {
+		s_seen.erase(tup);
+		BEESCOUNT(scan_seen_erase);
+	} else {
+		// BEESLOGDEBUG("Seen " << tup << " " << e);
+		s_seen.insert(tup);
+		BEESCOUNT(scan_seen_insert);
+	}
+	lock_seen.unlock();
+
+	// Now causes 75% loss of performance in benchmarks
 	// bees_unreadahead(bfr.fd(), bfr.begin(), bfr.size());
-	return bfr;
 }
 
 shared_ptr<Exclusion>
@@ -703,14 +891,14 @@ BeesContext::scan_forward(const BeesFileRange &bfr_in)
 	// No FD?  Well, that was quick.
 	if (!bfr.fd()) {
 		// BEESLOGINFO("No FD in " << root_path() << " for " << bfr);
-		BEESCOUNT(scan_no_fd);
+		BEESCOUNT(scanf_no_fd);
 		return false;
 	}
 
 	// Sanity check
 	if (bfr.begin() >= bfr.file_size()) {
 		BEESLOGWARN("past EOF: " << bfr);
-		BEESCOUNT(scan_eof);
+		BEESCOUNT(scanf_eof);
 		return false;
 	}
 
@@ -730,9 +918,11 @@ BeesContext::scan_forward(const BeesFileRange &bfr_in)
 					// BEESLOGDEBUG("Deferring extent bytenr " << to_hex(extent_bytenr) << " from " << bfr);
 					BEESCOUNT(scanf_deferred_extent);
 					start_over = true;
+					return; // from closure
 				}
 				Timer one_extent_timer;
 				scan_one_extent(bfr, e);
+				// BEESLOGDEBUG("Scanned " << e << " " << bfr);
 				BEESCOUNTADD(scanf_extent_ms, one_extent_timer.age() * 1000);
 				BEESCOUNT(scanf_extent);
 			});
@@ -925,7 +1115,8 @@ BeesContext::start()
 		return make_shared<BeesTempFile>(shared_from_this());
 	});
 	m_logical_ino_pool.generator([]() {
-		return make_shared<BtrfsIoctlLogicalInoArgs>(0);
+		const auto extent_ref_size = sizeof(uint64_t) * 3;
+		return make_shared<BtrfsIoctlLogicalInoArgs>(0, BEES_MAX_EXTENT_REF_COUNT * extent_ref_size + sizeof(btrfs_data_container));
 	});
 	m_tmpfile_pool.checkin([](const shared_ptr<BeesTempFile> &btf) {
 		catch_all([&](){

src/bees-hash.cc

@@ -356,6 +356,8 @@ BeesHashTable::prefetch_loop()
 		auto avg_rates = thisStats / m_ctx->total_timer().age();
 		graph_blob << "\t" << avg_rates << "\n";
 
+		graph_blob << m_ctx->get_progress();
+
 		BEESLOGINFO(graph_blob.str());
 		catch_all([&]() {
 			m_stats_file.write(graph_blob.str());
@@ -446,10 +448,38 @@ BeesHashTable::fetch_missing_extent_by_index(uint64_t extent_index)
 
 		// If we are in prefetch, give the kernel a hint about the next extent
 		if (m_prefetch_running) {
-			// XXX: don't call this if bees_readahead is implemented by pread()
+			// Use the kernel readahead here, because it might work for this use case
-			bees_readahead(m_fd, dirty_extent_offset + dirty_extent_size, dirty_extent_size);
+			readahead(m_fd, dirty_extent_offset + dirty_extent_size, dirty_extent_size);
 		}
 	});
+
+	Cell *cell     = m_extent_ptr[extent_index    ].p_buckets[0].p_cells;
+	Cell *cell_end = m_extent_ptr[extent_index + 1].p_buckets[0].p_cells;
+	size_t toxic_cleared_count = 0;
+	set<BeesHashTable::Cell> seen_it(cell, cell_end);
+	while (cell < cell_end) {
+		if (cell->e_addr & BeesAddress::c_toxic_mask) {
+			++toxic_cleared_count;
+			cell->e_addr &= ~BeesAddress::c_toxic_mask;
+			// Clearing the toxic bit might mean we now have a duplicate.
+			// This could be due to a race between two
+			// inserts, one finds the extent toxic while the
+			// other does not.  That's arguably a bug elsewhere,
+			// but we should rewrite the whole extent lookup/insert
+			// loop, not spend time fixing code that will be
+			// thrown out later anyway.
+			// If there is a cell that is identical to this one
+			// except for the toxic bit, then we don't need this one.
+			if (seen_it.count(*cell)) {
+				cell->e_addr = 0;
+				cell->e_hash = 0;
+			}
+		}
+		++cell;
+	}
+	if (toxic_cleared_count) {
+		BEESLOGDEBUG("Cleared " << toxic_cleared_count << " hashes while fetching hash table extent " << extent_index);
+	}
 }
 
 void

src/bees-resolve.cc

@@ -384,7 +384,7 @@ BeesResolver::for_each_extent_ref(BeesBlockData bbd, function<bool(const BeesFil
 	return stop_now;
 }
 
-BeesFileRange
+BeesRangePair
 BeesResolver::replace_dst(const BeesFileRange &dst_bfr_in)
 {
 	BEESTRACE("replace_dst dst_bfr " << dst_bfr_in);
@@ -400,6 +400,7 @@ BeesResolver::replace_dst(const BeesFileRange &dst_bfr_in)
 	BEESTRACE("overlap_bfr " << overlap_bfr);
 
 	BeesBlockData bbd(dst_bfr);
+	BeesRangePair rv = { BeesFileRange(), BeesFileRange() };
 
 	for_each_extent_ref(bbd, [&](const BeesFileRange &src_bfr_in) -> bool {
 		// Open src
@@ -436,21 +437,12 @@ BeesResolver::replace_dst(const BeesFileRange &dst_bfr_in)
 			BEESCOUNT(replacedst_grown);
 		}
 
-		// Dedup
+		rv = brp;
-		BEESNOTE("dedup " << brp);
+		m_found_dup = true;
-		if (m_ctx->dedup(brp)) {
+		return true;
-			BEESCOUNT(replacedst_dedup_hit);
-			m_found_dup = true;
-			overlap_bfr = brp.second;
-			// FIXME:  find best range first, then dedupe that
-			return true; // i.e. break
-		} else {
-			BEESCOUNT(replacedst_dedup_miss);
-			return false; // i.e. continue
-		}
 	});
 	// BEESLOG("overlap_bfr after " << overlap_bfr);
-	return overlap_bfr.copy_closed();
+	return rv;
 }
 
 BeesFileRange

src/bees-types.cc

@@ -183,6 +183,24 @@ BeesFileRange::grow_begin(off_t delta)
 	return m_begin;
 }
 
+off_t
+BeesFileRange::shrink_begin(off_t delta)
+{
+	THROW_CHECK1(invalid_argument, delta, delta > 0);
+	THROW_CHECK3(invalid_argument, delta, m_begin, m_end, delta + m_begin < m_end);
+	m_begin += delta;
+	return m_begin;
+}
+
+off_t
+BeesFileRange::shrink_end(off_t delta)
+{
+	THROW_CHECK1(invalid_argument, delta, delta > 0);
+	THROW_CHECK2(invalid_argument, delta, m_end, m_end >= delta);
+	m_end -= delta;
+	return m_end;
+}
+
 BeesFileRange::BeesFileRange(const BeesBlockData &bbd) :
 	m_fd(bbd.fd()),
 	m_begin(bbd.begin()),
@@ -349,8 +367,8 @@ BeesRangePair::grow(shared_ptr<BeesContext> ctx, bool constrained)
 	BEESTRACE("e_second " << e_second);
 
 	// Preread entire extent
-	bees_readahead(second.fd(), e_second.begin(), e_second.size());
+	bees_readahead_pair(second.fd(), e_second.begin(), e_second.size(),
-	bees_readahead(first.fd(), e_second.begin() + first.begin() - second.begin(), e_second.size());
+			    first.fd(), e_second.begin() + first.begin() - second.begin(), e_second.size());
 
 	auto hash_table = ctx->hash_table();
 
@@ -388,17 +406,6 @@ BeesRangePair::grow(shared_ptr<BeesContext> ctx, bool constrained)
 			break;
 		}
 
-		// Source extent cannot be toxic
-		BeesAddress first_addr(first.fd(), new_first.begin());
-		if (!first_addr.is_magic()) {
-			auto first_resolved = ctx->resolve_addr(first_addr);
-			if (first_resolved.is_toxic()) {
-				BEESLOGWARN("WORKAROUND: not growing matching pair backward because src addr is toxic:\n" << *this);
-				BEESCOUNT(pairbackward_toxic_addr);
-				break;
-			}
-		}
-
 		// Extend second range.  If we hit BOF we can go no further.
 		BeesFileRange new_second = second;
 		BEESTRACE("new_second = " << new_second);
@@ -434,6 +441,7 @@ BeesRangePair::grow(shared_ptr<BeesContext> ctx, bool constrained)
 		}
 
 		// Source block cannot be zero in a non-compressed non-magic extent
+		BeesAddress first_addr(first.fd(), new_first.begin());
 		if (first_bbd.is_data_zero() && !first_addr.is_magic() && !first_addr.is_compressed()) {
 			BEESCOUNT(pairbackward_zero);
 			break;
@@ -491,17 +499,6 @@ BeesRangePair::grow(shared_ptr<BeesContext> ctx, bool constrained)
 			break;
 		}
 
-		// Source extent cannot be toxic
-		BeesAddress first_addr(first.fd(), new_first.begin());
-		if (!first_addr.is_magic()) {
-			auto first_resolved = ctx->resolve_addr(first_addr);
-			if (first_resolved.is_toxic()) {
-				BEESLOGWARN("WORKAROUND: not growing matching pair forward because src is toxic:\n" << *this);
-				BEESCOUNT(pairforward_toxic);
-				break;
-			}
-		}
-
 		// Extend second range.  If we hit EOF we can go no further.
 		BeesFileRange new_second = second;
 		BEESTRACE("new_second = " << new_second);
@@ -545,6 +542,7 @@ BeesRangePair::grow(shared_ptr<BeesContext> ctx, bool constrained)
 		}
 
 		// Source block cannot be zero in a non-compressed non-magic extent
+		BeesAddress first_addr(first.fd(), new_first.begin());
 		if (first_bbd.is_data_zero() && !first_addr.is_magic() && !first_addr.is_compressed()) {
 			BEESCOUNT(pairforward_zero);
 			break;
@@ -589,6 +587,22 @@ BeesRangePair::copy_closed() const
 	return BeesRangePair(first.copy_closed(), second.copy_closed());
 }
 
+void
+BeesRangePair::shrink_begin(off_t const delta)
+{
+	first.shrink_begin(delta);
+	second.shrink_begin(delta);
+	THROW_CHECK2(runtime_error, first.size(), second.size(), first.size() == second.size());
+}
+
+void
+BeesRangePair::shrink_end(off_t const delta)
+{
+	first.shrink_end(delta);
+	second.shrink_end(delta);
+	THROW_CHECK2(runtime_error, first.size(), second.size(), first.size() == second.size());
+}
+
 ostream &
 operator<<(ostream &os, const BeesAddress &ba)
 {

src/bees-usage.txt

@@ -14,7 +14,7 @@ Load management options:
     -g, --loadavg-target  Target load average for worker threads (default none)
 
 Filesystem tree traversal options:
-    -m, --scan-mode       Scanning mode (0..2, default 0)
+    -m, --scan-mode       Scanning mode (0..4, default 4)
 
 Workarounds:
     -a, --workaround-btrfs-send    Workaround for btrfs send

src/bees.cc

@@ -214,9 +214,35 @@ BeesTooLong::operator=(const func_type &f)
 	return *this;
 }
 
-void
+static
-bees_readahead(int const fd, const off_t offset, const size_t size)
+bool
+bees_readahead_check(int const fd, off_t const offset, size_t const size)
 {
+	// FIXME: the rest of the code calls this function more often than necessary,
+	// usually back-to-back calls on the same range in a loop.
+	// Simply discard requests that are identical to recent requests from the same thread.
+	const Stat stat_rv(fd);
+	auto tup = make_tuple(offset, size, stat_rv.st_dev, stat_rv.st_ino);
+	static mutex s_recent_mutex;
+	static set<decltype(tup)> s_recent;
+	unique_lock<mutex> lock(s_recent_mutex);
+	if (s_recent.size() > BEES_MAX_EXTENT_REF_COUNT) {
+		s_recent.clear();
+		BEESCOUNT(readahead_clear);
+	}
+	const auto rv = s_recent.insert(tup);
+	// If we recently did this readahead, we're done here
+	if (!rv.second) {
+		BEESCOUNT(readahead_skip);
+	}
+	return rv.second;
+}
+
+static
+void
+bees_readahead_nolock(int const fd, const off_t offset, const size_t size)
+{
+	if (!bees_readahead_check(fd, size, offset)) return;
 	Timer readahead_timer;
 	BEESNOTE("readahead " << name_fd(fd) << " offset " << to_hex(offset) << " len " << pretty(size));
 	BEESTOOLONG("readahead " << name_fd(fd) << " offset " << to_hex(offset) << " len " << pretty(size));
@@ -225,10 +251,8 @@ bees_readahead(int const fd, const off_t offset, const size_t size)
 	DIE_IF_NON_ZERO(readahead(fd, offset, size));
 #else
 	// Make sure this data is in page cache by brute force
-	// This isn't necessary and it might even be slower,
+	// The btrfs kernel code does readahead with lower ioprio
-	// but the btrfs kernel code does readahead with lower ioprio
+	// and might discard the readahead request entirely.
-	// and might discard the readahead request entirely,
-	// so it's maybe, *maybe*, worth doing both.
 	BEESNOTE("emulating readahead " << name_fd(fd) << " offset " << to_hex(offset) << " len " << pretty(size));
 	auto working_size = size;
 	auto working_offset = offset;
@@ -249,6 +273,28 @@ bees_readahead(int const fd, const off_t offset, const size_t size)
 	BEESCOUNTADD(readahead_ms, readahead_timer.age() * 1000);
 }
 
+static mutex s_only_one;
+
+void
+bees_readahead_pair(int fd, off_t offset, size_t size, int fd2, off_t offset2, size_t size2)
+{
+	if (!bees_readahead_check(fd, size, offset) && !bees_readahead_check(fd2, offset2, size2)) return;
+	BEESNOTE("waiting to readahead " << name_fd(fd) << " offset " << to_hex(offset) << " len " << pretty(size) << ","
+		<< "\n\t" << name_fd(fd2) << " offset " << to_hex(offset2) << " len " << pretty(size2));
+	unique_lock<mutex> m_lock(s_only_one);
+	bees_readahead_nolock(fd, offset, size);
+	bees_readahead_nolock(fd2, offset2, size2);
+}
+
+void
+bees_readahead(int const fd, const off_t offset, const size_t size)
+{
+	if (!bees_readahead_check(fd, size, offset)) return;
+	BEESNOTE("waiting to readahead " << name_fd(fd) << " offset " << to_hex(offset) << " len " << pretty(size));
+	unique_lock<mutex> m_lock(s_only_one);
+	bees_readahead_nolock(fd, offset, size);
+}
+
 void
 bees_unreadahead(int const fd, off_t offset, size_t size)
 {
@@ -603,7 +649,7 @@ bees_main(int argc, char *argv[])
 	unsigned thread_min = 0;
 	double load_target = 0;
 	bool workaround_btrfs_send = false;
-	BeesRoots::ScanMode root_scan_mode = BeesRoots::SCAN_MODE_INDEPENDENT;
+	BeesRoots::ScanMode root_scan_mode = BeesRoots::SCAN_MODE_EXTENT;
 
 	// Configure getopt_long
 	static const struct option long_options[] = {
@@ -751,6 +797,13 @@ bees_main(int argc, char *argv[])
 	// Set root scan mode
 	bc->roots()->set_scan_mode(root_scan_mode);
 
+	if (root_scan_mode == BeesRoots::SCAN_MODE_EXTENT) {
+		MultiLocker::enable_locking(false);
+	} else {
+		// Workaround for a kernel bug that the subvol-based crawlers keep triggering
+		MultiLocker::enable_locking(true);
+	}
+
 	// Start crawlers
 	bc->start();
 

src/bees.h

@@ -78,13 +78,13 @@ const int BEES_PROGRESS_INTERVAL = BEES_STATS_INTERVAL;
 const int BEES_STATUS_INTERVAL = 1;
 
 // Number of file FDs to cache when not in active use
-const size_t BEES_FILE_FD_CACHE_SIZE = 4096;
+const size_t BEES_FILE_FD_CACHE_SIZE = 32768;
 
 // Number of root FDs to cache when not in active use
-const size_t BEES_ROOT_FD_CACHE_SIZE = 1024;
+const size_t BEES_ROOT_FD_CACHE_SIZE = 4096;
 
 // Number of FDs to open (rlimit)
-const size_t BEES_OPEN_FILE_LIMIT = (BEES_FILE_FD_CACHE_SIZE + BEES_ROOT_FD_CACHE_SIZE) * 2 + 100;
+const size_t BEES_OPEN_FILE_LIMIT = BEES_FILE_FD_CACHE_SIZE + BEES_ROOT_FD_CACHE_SIZE + 100;
 
 // Worker thread factor (multiplied by detected number of CPU cores)
 const double BEES_DEFAULT_THREAD_FACTOR = 1.0;
@@ -93,10 +93,11 @@ const double BEES_DEFAULT_THREAD_FACTOR = 1.0;
 const double BEES_TOO_LONG = 5.0;
 
 // Avoid any extent where LOGICAL_INO takes this much kernel CPU time
-const double BEES_TOXIC_SYS_DURATION = 0.1;
+const double BEES_TOXIC_SYS_DURATION = 5.0;
 
-// Maximum number of refs to a single extent
+// Maximum number of refs to a single extent before we have other problems
-const size_t BEES_MAX_EXTENT_REF_COUNT = (16 * 1024 * 1024 / 24) - 1;
+// If we have more than 10K refs to an extent, adding another will save 0.01% space
+const size_t BEES_MAX_EXTENT_REF_COUNT = 9999; // (16 * 1024 * 1024 / 24);
 
 // How long between hash table histograms
 const double BEES_HASH_TABLE_ANALYZE_INTERVAL = BEES_STATS_INTERVAL;
@@ -123,7 +124,7 @@ const int FLAGS_OPEN_FANOTIFY = O_RDWR | O_NOATIME | O_CLOEXEC | O_LARGEFILE;
 #define BEESLOG(lv,x)   do { if (lv < bees_log_level) { Chatter __chatter(lv, BeesNote::get_name()); __chatter << x; } } while (0)
 #define BEESLOGTRACE(x) do { BEESLOG(LOG_DEBUG, x); BeesTracer::trace_now(); } while (0)
 
-#define BEESTRACE(x)   BeesTracer  SRSLY_WTF_C(beesTracer_,  __LINE__) ([&]()                 { BEESLOG(LOG_ERR, x);   })
+#define BEESTRACE(x)   BeesTracer  SRSLY_WTF_C(beesTracer_,  __LINE__) ([&]()                 { BEESLOG(LOG_ERR, x << " at " << __FILE__ << ":" << __LINE__);   })
 #define BEESTOOLONG(x) BeesTooLong SRSLY_WTF_C(beesTooLong_, __LINE__) ([&](ostream &_btl_os) { _btl_os << x; })
 #define BEESNOTE(x)    BeesNote    SRSLY_WTF_C(beesNote_,    __LINE__) ([&](ostream &_btl_os) { _btl_os << x; })
 
@@ -299,6 +300,11 @@ public:
 	off_t grow_begin(off_t delta);
 	/// @}
 
+	/// @{ Make range smaller
+	off_t shrink_end(off_t delta);
+	off_t shrink_begin(off_t delta);
+	/// @}
+
 friend ostream & operator<<(ostream &os, const BeesFileRange &bfr);
 };
 
@@ -515,7 +521,7 @@ class BeesCrawl {
 
 	bool fetch_extents();
 	void fetch_extents_harder();
-	bool next_transid();
+	bool restart_crawl();
 	BeesFileRange bti_to_bfr(const BtrfsTreeItem &bti) const;
 
 public:
@@ -527,6 +533,8 @@ public:
 	BeesCrawlState get_state_end() const;
 	void set_state(const BeesCrawlState &bcs);
 	void deferred(bool def_setting);
+	bool deferred() const;
+	bool finished() const;
 };
 
 class BeesScanMode;
@@ -555,16 +563,12 @@ class BeesRoots : public enable_shared_from_this<BeesRoots> {
 	bool					m_stop_requested = false;
 
 	void insert_new_crawl();
-	void insert_root(const BeesCrawlState &bcs);
 	Fd open_root_nocache(uint64_t root);
 	Fd open_root_ino_nocache(uint64_t root, uint64_t ino);
-	uint64_t transid_min();
-	uint64_t transid_max();
 	uint64_t transid_max_nocache();
 	void state_load();
 	ostream &state_to_stream(ostream &os);
 	void state_save();
-	bool crawl_roots();
 	string crawl_state_filename() const;
 	void crawl_state_set_dirty();
 	void crawl_state_erase(const BeesCrawlState &bcs);
@@ -576,6 +580,9 @@ class BeesRoots : public enable_shared_from_this<BeesRoots> {
 	bool crawl_batch(shared_ptr<BeesCrawl> crawl);
 	void clear_caches();
 
+friend class BeesScanModeExtent;
+	shared_ptr<BeesCrawl> insert_root(const BeesCrawlState &bcs);
+
 friend class BeesCrawl;
 friend class BeesFdCache;
 friend class BeesScanMode;
@@ -594,17 +601,20 @@ public:
 	Fd open_root_ino(const BeesFileId &bfi) { return open_root_ino(bfi.root(), bfi.ino()); }
 	bool is_root_ro(uint64_t root);
 
-	// TODO:  do extent-tree scans instead
 	enum ScanMode {
 		SCAN_MODE_LOCKSTEP,
 		SCAN_MODE_INDEPENDENT,
 		SCAN_MODE_SEQUENTIAL,
 		SCAN_MODE_RECENT,
+		SCAN_MODE_EXTENT,
 		SCAN_MODE_COUNT, // must be last
 	};
 
 	void set_scan_mode(ScanMode new_mode);
 	void set_workaround_btrfs_send(bool do_avoid);
+
+	uint64_t transid_min();
+	uint64_t transid_max();
 };
 
 struct BeesHash {
@@ -664,6 +674,8 @@ class BeesRangePair : public pair<BeesFileRange, BeesFileRange> {
 public:
 	BeesRangePair(const BeesFileRange &src, const BeesFileRange &dst);
 	bool grow(shared_ptr<BeesContext> ctx, bool constrained);
+	void shrink_begin(const off_t delta);
+	void shrink_end(const off_t delta);
 	BeesRangePair copy_closed() const;
 	bool operator<(const BeesRangePair &that) const;
 friend ostream & operator<<(ostream &os, const BeesRangePair &brp);
@@ -737,11 +749,14 @@ class BeesContext : public enable_shared_from_this<BeesContext> {
 	shared_ptr<BeesThread>				m_progress_thread;
 	shared_ptr<BeesThread>				m_status_thread;
 
+	mutex						m_progress_mtx;
+	string						m_progress_str;
+
 	void set_root_fd(Fd fd);
 
 	BeesResolveAddrResult resolve_addr_uncached(BeesAddress addr);
 
-	BeesFileRange scan_one_extent(const BeesFileRange &bfr, const Extent &e);
+	void scan_one_extent(const BeesFileRange &bfr, const Extent &e);
 	void rewrite_file_range(const BeesFileRange &bfr);
 
 public:
@@ -772,6 +787,8 @@ public:
 
 	void dump_status();
 	void show_progress();
+	void set_progress(const string &str);
+	string get_progress();
 
 	void start();
 	void stop();
@@ -834,7 +851,7 @@ public:
 	BeesFileRange find_one_match(BeesHash hash);
 
 	void replace_src(const BeesFileRange &src_bfr);
-	BeesFileRange replace_dst(const BeesFileRange &dst_bfr);
+	BeesRangePair replace_dst(const BeesFileRange &dst_bfr);
 
 	bool found_addr() const { return m_found_addr; }
 	bool found_data() const { return m_found_data; }
@@ -868,6 +885,7 @@ extern const char *BEES_VERSION;
 extern thread_local default_random_engine bees_generator;
 string pretty(double d);
 void bees_readahead(int fd, off_t offset, size_t size);
+void bees_readahead_pair(int fd, off_t offset, size_t size, int fd2, off_t offset2, size_t size2);
 void bees_unreadahead(int fd, off_t offset, size_t size);
 string format_time(time_t t);
 

test/Makefile

@@ -8,6 +8,7 @@ PROGRAMS = \
 	process \
 	progress \
 	seeker \
+	table \
 	task \
 
 all: test

test/table.cc

@@ -0,0 +1,63 @@
+#include "tests.h"
+
+#include "crucible/table.h"
+
+using namespace crucible;
+using namespace std;
+
+void
+print_table(const Table::Table& t)
+{
+	cerr << "BEGIN TABLE\n";
+	cerr << t;
+	cerr << "END TABLE\n";
+	cerr << endl;
+}
+
+void
+test_table()
+{
+	Table::Table t;
+	t.insert_row(Table::endpos, vector<Table::Content> {
+		Table::Text("Hello, World!"),
+		Table::Text("2"),
+		Table::Text("3"),
+		Table::Text("4"),
+	});
+	print_table(t);
+	t.insert_row(Table::endpos, vector<Table::Content> {
+		Table::Text("Greeting"),
+		Table::Text("two"),
+		Table::Text("three"),
+		Table::Text("four"),
+	});
+	print_table(t);
+	t.insert_row(Table::endpos, vector<Table::Content> {
+		Table::Fill('-'),
+		Table::Text("ii"),
+		Table::Text("iii"),
+		Table::Text("iv"),
+	});
+	print_table(t);
+	t.mid(" | ");
+	t.left("| ");
+	t.right(" |");
+	print_table(t);
+	t.insert_col(1, vector<Table::Content> {
+		Table::Text("1"),
+		Table::Text("one"),
+		Table::Text("i"),
+		Table::Text("I"),
+	});
+	print_table(t);
+	t.at(2, 1) = Table::Text("Two\nLines");
+	print_table(t);
+}
+
+int
+main(int, char**)
+{
+	RUN_A_TEST(test_table());
+
+	exit(EXIT_SUCCESS);
+}