27  *   1) space_info.  This is the ultimate arbiter of how much space we can use.
30  *   reservations we care about total_bytes - SUM(space_info->bytes_) when
35  *   metadata reservation we have.  You can see the comment in the block_rsv
39  *   3) btrfs_calc*_size.  These are the worst case calculations we used based
40  *   on the number of items we will want to modify.  We have one for changing
41  *   items, and one for inserting new items.  Generally we use these helpers to
47  *   We call into either btrfs_reserve_data_bytes() or
48  *   btrfs_reserve_metadata_bytes(), depending on which we're looking for, with
49  *   num_bytes we want to reserve.
66  *   Assume we are unable to simply make the reservation because we do not have
89  *     Check if ->bytes == 0, if it does we got our reservation and we can carry
90  *     on, if not return the appropriate error (ENOSPC, but can be EINTR if we
95  *   Same as the above, except we add ourselves to the
96  *   space_info->priority_tickets, and we do not use ticket->wait, we simply
102  *   Generally speaking we will have two cases for each state, a "nice" state
103  *   and a "ALL THE THINGS" state.  In btrfs we delay a lot of work in order to
107  *   reclaim space so we can make new reservations.
111  *     for example, we would update the inode item at write time to update the
113  *     isize or bytes.  We keep these delayed items to coalesce these operations
119  *     for delayed allocation.  We can reclaim some of this space simply by
120  *     running delalloc, but usually we need to wait for ordered extents to
124  *     We have a block reserve for the outstanding delayed refs space, and every
126  *     to reclaim space, but we want to hold this until the end because COW can
127  *     churn a lot and we can avoid making some extent tree modifications if we
131  *     We will skip this the first time through space reservation, because of
132  *     overcommit and we don't want to have a lot of useless metadata space when
136  *     If we're freeing inodes we're likely freeing checksums, file extent
141  *     This will commit the transaction.  Historically we had a lot of logic
142  *     surrounding whether or not we'd commit the transaction, but this waits born
143  *     out of a pre-tickets era where we could end up committing the transaction
145  *     ticketing system we know if we're not making progress and can error
151  *   Because we hold so many reservations for metadata we will allow you to
156  *   You can see the current logic for when we allow overcommit in
177  * after adding space to the filesystem, we need to clear the full flags
355 	 * "optimal" chunk size based on the fs size.  However when we actually  in calc_effective_data_chunk_size()
356 	 * allocate the chunk we will strip this down further, making it no  in calc_effective_data_chunk_size()
359 	 * On the zoned mode, we need to use zone_size (= data_sinfo->chunk_size)  in calc_effective_data_chunk_size()
387 	 * If we have dup, raid1 or raid10 then only half of the free  in calc_available_free_space()
389 	 * doesn't include the parity drive, so we don't have to  in calc_available_free_space()
401 	 * reservation, because we assume that data reservations will == actual  in calc_available_free_space()
402 	 * usage, we could potentially overcommit and then immediately have that  in calc_available_free_space()
404 	 * bind when we get close to filling the file system.  in calc_available_free_space()
407 	 * space.  If we are relatively empty this won't affect our ability to  in calc_available_free_space()
408 	 * overcommit much, and if we're very close to full it'll keep us from  in calc_available_free_space()
409 	 * getting into a position where we've given ourselves very little  in calc_available_free_space()
417 	 * If we aren't flushing all things, let us overcommit up to  in calc_available_free_space()
418 	 * 1/2th of the space. If we can flush, don't let us overcommit  in calc_available_free_space()
427 	 * On the zoned mode, we always allocate one zone as one chunk.  in calc_available_free_space()
469  * This is for space we already have accounted in space_info->bytes_may_use, so
470  * basically when we're returning space from block_rsv's.
637 	/* Calc the number of the pages we need flush for space reservation */  in shrink_delalloc()
642 		 * to_reclaim is set to however much metadata we need to  in shrink_delalloc()
644 		 * exactly.  What we really want to do is reclaim full inode's  in shrink_delalloc()
646 		 * here.  We will take a fraction of the delalloc bytes for our  in shrink_delalloc()
648 		 * the amount we write to cover an entire dirty extent, which  in shrink_delalloc()
660 	 * If we are doing more ordered than delalloc we need to just wait on  in shrink_delalloc()
661 	 * ordered extents, otherwise we'll waste time trying to flush delalloc  in shrink_delalloc()
662 	 * that likely won't give us the space back we need.  in shrink_delalloc()
676 		 * We need to make sure any outstanding async pages are now  in shrink_delalloc()
677 		 * processed before we continue.  This is because things like  in shrink_delalloc()
679 		 * marked clean.  We don't use filemap_fwrite for flushing  in shrink_delalloc()
680 		 * because we want to control how many pages we write out at a  in shrink_delalloc()
681 		 * time, thus this is the only safe way to make sure we've  in shrink_delalloc()
685 		 * This exists because we do not want to wait for each  in shrink_delalloc()
686 		 * individual inode to finish its async work, we simply want to  in shrink_delalloc()
688 		 * for all of the async work to catch up.  Once we're done with  in shrink_delalloc()
689 		 * that we know we'll have ordered extents for everything and we  in shrink_delalloc()
690 		 * can decide if we wait for that or not.  in shrink_delalloc()
692 		 * If we choose to replace this in the future, make absolutely  in shrink_delalloc()
701 		 * We don't want to wait forever, if we wrote less pages in this  in shrink_delalloc()
702 		 * loop than we have outstanding, only wait for that number of  in shrink_delalloc()
703 		 * pages, otherwise we can wait for all async pages to finish  in shrink_delalloc()
724 		 * If we are for preemption we just want a one-shot of delalloc  in shrink_delalloc()
725 		 * flushing so we can stop flushing if we decide we don't need  in shrink_delalloc()
819 		 * If we have pending delayed iputs then we could free up a  in flush_space()
820 		 * bunch of pinned space, so make sure we run the iputs before  in flush_space()
821 		 * we do our pinned bytes check below.  in flush_space()
829 		 * We don't want to start a new transaction, just attach to the  in flush_space()
831 		 * happening at the moment. Note: we don't use a nostart join  in flush_space()
861 	 * We may be flushing because suddenly we have less space than we had  in btrfs_calc_reclaim_metadata_size()
862 	 * before, and now we're well over-committed based on our current free  in btrfs_calc_reclaim_metadata_size()
863 	 * space.  If that's the case add in our overage so we make sure to put  in btrfs_calc_reclaim_metadata_size()
884 	/* If we're just plain full then async reclaim just slows us down. */  in need_preemptive_reclaim()
896 	 * 128MiB is 1/4 of the maximum global rsv size.  If we have less than  in need_preemptive_reclaim()
898 	 * we don't have a lot of things that need flushing.  in need_preemptive_reclaim()
904 	 * We have tickets queued, bail so we don't compete with the async  in need_preemptive_reclaim()
911 	 * If we have over half of the free space occupied by reservations or  in need_preemptive_reclaim()
912 	 * pinned then we want to start flushing.  in need_preemptive_reclaim()
914 	 * We do not do the traditional thing here, which is to say  in need_preemptive_reclaim()
919 	 * because this doesn't quite work how we want.  If we had more than 50%  in need_preemptive_reclaim()
920 	 * of the space_info used by bytes_used and we had 0 available we'd just  in need_preemptive_reclaim()
921 	 * constantly run the background flusher.  Instead we want it to kick in  in need_preemptive_reclaim()
936 	 * much delalloc we need for the background flusher to kick in.  in need_preemptive_reclaim()
950 	 * If we have more ordered bytes than delalloc bytes then we're either  in need_preemptive_reclaim()
951 	 * doing a lot of DIO, or we simply don't have a lot of delalloc waiting  in need_preemptive_reclaim()
955 	 * nothing, if our reservations are tied up in ordered extents we'll  in need_preemptive_reclaim()
960 	 * block reserves that we would actually be able to directly reclaim  in need_preemptive_reclaim()
961 	 * from.  In this case if we're heavy on metadata operations this will  in need_preemptive_reclaim()
966 	 * We want to make sure we truly are maxed out on ordered however, so  in need_preemptive_reclaim()
967 	 * cut ordered in half, and if it's still higher than delalloc then we  in need_preemptive_reclaim()
968 	 * can keep flushing.  This is to avoid the case where we start  in need_preemptive_reclaim()
969 	 * flushing, and now delalloc == ordered and we stop preemptively  in need_preemptive_reclaim()
970 	 * flushing when we could still have several gigs of delalloc to flush.  in need_preemptive_reclaim()
1016  * We've exhausted our flushing, start failing tickets.
1019  * @space_info - the space info we were flushing
1021  * We call this when we've exhausted our flushing ability and haven't made
1023  * order, so if there is a large ticket first and then smaller ones we could
1065 		 * We're just throwing tickets away, so more flushing may not  in maybe_fail_all_tickets()
1066 		 * trip over btrfs_try_granting_tickets, so we need to call it  in maybe_fail_all_tickets()
1067 		 * here to see if we can make progress with the next ticket in  in maybe_fail_all_tickets()
1077  * This is for normal flushers, we can wait all goddamned day if we want to.  We
1078  * will loop and continuously try to flush as long as we are making progress.
1079  * We count progress as clearing off tickets each time we have to loop.
1124 		 * We do not want to empty the system of delalloc unless we're  in btrfs_async_reclaim_metadata_space()
1126 		 * logic before we start doing a FLUSH_DELALLOC_FULL.  in btrfs_async_reclaim_metadata_space()
1132 		 * We don't want to force a chunk allocation until we've tried  in btrfs_async_reclaim_metadata_space()
1133 		 * pretty hard to reclaim space.  Think of the case where we  in btrfs_async_reclaim_metadata_space()
1135 		 * to reclaim.  We would rather use that than possibly create a  in btrfs_async_reclaim_metadata_space()
1139 		 * around then we can force a chunk allocation.  in btrfs_async_reclaim_metadata_space()
1162  * This handles pre-flushing of metadata space before we get to the point that
1163  * we need to start blocking threads on tickets.  The logic here is different
1165  * much we need to flush, instead it attempts to keep us below the 80% full
1197 		 * We don't have a precise counter for the metadata being  in btrfs_preempt_reclaim_metadata_space()
1198 		 * reserved for delalloc, so we'll approximate it by subtracting  in btrfs_preempt_reclaim_metadata_space()
1200 		 * amount is higher than the individual reserves, then we can  in btrfs_preempt_reclaim_metadata_space()
1211 		 * We don't want to include the global_rsv in our calculation,  in btrfs_preempt_reclaim_metadata_space()
1212 		 * because that's space we can't touch.  Subtract it from the  in btrfs_preempt_reclaim_metadata_space()
1218 		 * We really want to avoid flushing delalloc too much, as it  in btrfs_preempt_reclaim_metadata_space()
1242 		 * We don't want to reclaim everything, just a portion, so scale  in btrfs_preempt_reclaim_metadata_space()
1254 	/* We only went through once, back off our clamping. */  in btrfs_preempt_reclaim_metadata_space()
1265  *   1) compression is on and we allocate less space than we reserved
1266  *   2) we are overwriting existing space
1273  *   For #2 this is trickier.  Once the ordered extent runs we will drop the
1274  *   extent in the range we are overwriting, which creates a delayed ref for
1279  *   If we are freeing inodes, we want to make sure all delayed iputs have
1286  *   This is where we reclaim all of the pinned space generated by running the
1290  *   For data we start with alloc chunk force, however we could have been full
1292  *   so if we now have space to allocate do the force chunk allocation.
1417 	 * because we may have only satisfied the priority tickets and still  in priority_reclaim_metadata_space()
1418 	 * left non priority tickets on the list.  We would then have  in priority_reclaim_metadata_space()
1439 	 * Attempt to steal from the global rsv if we can, except if the fs was  in priority_reclaim_metadata_space()
1442 	 * success to the caller if we can steal from the global rsv - this is  in priority_reclaim_metadata_space()
1455 	 * We must run try_granting_tickets here because we could be a large  in priority_reclaim_metadata_space()
1469 	/* We could have been granted before we got here. */  in priority_reclaim_data_space()
1504 			 * Delete us from the list. After we unlock the space  in wait_reserve_ticket()
1505 			 * info, we don't want the async reclaim job to reserve  in wait_reserve_ticket()
1533  * @flush:      how much we can flush
1573 	 * Check that we can't have an error set if the reservation succeeded,  in handle_reserve_ticket()
1601 	 * If we're heavy on ordered operations then clamping won't help us.  We  in maybe_clamp_preempt()
1605 	 * delayed nodes.  If we're already more ordered than delalloc then  in maybe_clamp_preempt()
1606 	 * we're keeping up, otherwise we aren't and should probably clamp.  in maybe_clamp_preempt()
1632  * @space_info: space info we want to allocate from
1633  * @orig_bytes: number of bytes we want
1634  * @flush:      whether or not we can flush to make our reservation
1658 	 * BTRFS_RESERVE_FLUSH_EVICT, as we could deadlock because those  in __reserve_bytes()
1677 	 * We don't want NO_FLUSH allocations to jump everybody, they can  in __reserve_bytes()
1688 	 * Carry on if we have enough space (short-circuit) OR call  in __reserve_bytes()
1689 	 * can_overcommit() to ensure we can overcommit to continue.  in __reserve_bytes()
1700 	 * Things are dire, we need to make a reservation so we don't abort.  We  in __reserve_bytes()
1701 	 * will let this reservation go through as long as we have actual space  in __reserve_bytes()
1714 	 * If we couldn't make a reservation then setup our reservation ticket  in __reserve_bytes()
1717 	 * If we are a priority flusher then we just need to add our ticket to  in __reserve_bytes()
1718 	 * the list and we will do our own flushing further down.  in __reserve_bytes()
1735 				 * We were forced to add a reserve ticket, so  in __reserve_bytes()
1756 		 * We will do the space reservation dance during log replay,  in __reserve_bytes()
1757 		 * which means we won't have fs_info->fs_root set, so don't do  in __reserve_bytes()
1758 		 * the async reclaim as we will panic.  in __reserve_bytes()
1781  * @space_info: the space_info we're allocating for
1782  * @orig_bytes: number of bytes we want
1783  * @flush:      whether or not we can flush to make our reservation
1814  * @bytes:   number of bytes we need
1815  * @flush:   how we are allowed to flush
1818  * space then we will attempt to flush space as specified by flush.
1841 /* Dump all the space infos when we abort a transaction due to ENOSPC. */
1865 	/* It's df, we don't care if it's racy */  in btrfs_account_ro_block_groups_free_space()
1910  * If we claw this back repeatedly, we can still achieve efficient
1932  * Furthermore, we want to avoid doing too much reclaim even if there are good
1934  * holes with writes. So we want to do just enough reclaim to try and stay
1939  * - ratchet up the intensity of reclaim depending on how far we are from
1956 	/* If we have no unused space, don't bother, it won't work anyway. */  in calc_dynamic_reclaim_threshold()
1974  * Under "urgent" reclaim, we will reclaim even fresh block groups that have
1975  * recently seen successful allocations, as we are desperate to reclaim
1976  * whatever we can to avoid ENOSPC in a transaction leading to a readonly fs.
2021 	 * In situations where we are very motivated to reclaim (low unalloc)  in do_reclaim_sweep()
2024 	 * If we have any staler groups, we don't touch the fresher ones, but if we  in do_reclaim_sweep()