1# $NetBSD: TODO,v 1.7 2003/02/23 00:22:33 perseant Exp $ 2 3- Lock audit. Need to check locking for multiprocessor case in particular. 4 5- Get rid of the syscalls: make them into ioctl calls instead. This would 6 allow LFS to be loaded as a module. We would then ideally have an 7 in-kernel cleaner that runs if no userland cleaner has asserted itself. 8 9- Get rid of lfs_segclean(); the kernel should clean a dirty segment IFF it 10 has passed two checkpoints containing zero live bytes. 11 12- Now that our cache is basically all of physical memory, we need to make 13 sure that segwrite is not starving other important things. Need a way 14 to prioritize which blocks are most important to write, and write only 15 those before giving up the seglock to do the rest. How does this change 16 our notion of what a checkpoint is? 17 18- Investigate alternate inode locking strategy: Inode locks are useful 19 for locking against simultaneous changes to inode size (balloc, 20 truncate, write) but because the assignment of disk blocks is also 21 covered by the segment lock, we don't really need to pay attention to 22 the inode lock when writing a segment, right? If this is true, the 23 locking problem in lfs_{bmapv,markv} goes away and lfs_reserve can go, 24 too. 25 26- Fully working fsck_lfs. (Really, need a general-purpose external 27 partial-segment writer.) 28 29- Get rid of DEV_BSIZE, pay attention to the media block size at mount time. 30 31- More fs ops need to call lfs_imtime. Which ones? (Blackwell et al., 1995) 32 33- lfs_vunref_head exists so that vnodes loaded solely for cleaning can 34 be put back on the *head* of the vnode free list. Make sure we 35 actually do this, since we now take IN_CLEANING off during segment write. 36 37- The cleaner could be enhanced to be controlled from other processes, 38 and possibly perform additional tasks: 39 40 - Backups. At a minimum, turn the cleaner off and on to allow 41 effective live backups. More aggressively, the cleaner itself could 42 be the backup agent, and dump_lfs would merely be a controller. 43 44 - Cleaning time policies. Be able to tweak the cleaner's thresholds 45 to allow more thorough cleaning during policy-determined idle 46 periods (regardless of actual idleness) or put off until later 47 during short, intensive write periods. 48 49 - File coalescing and placement. During periods we expect to be idle, 50 coalesce fragmented files into one place on disk for better read 51 performance. Ideally, move files that have not been accessed in a 52 while to the extremes of the disk, thereby shortening seek times for 53 files that are accessed more frequently (though how the cleaner 54 should communicate "please put this near the beginning or end of the 55 disk" to the kernel is a very good question; flags to lfs_markv?). 56 57 - Versioning. When it cleans a segment it could write data for files 58 that were less than n versions old to tape or elsewhere. Perhaps it 59 could even write them back onto the disk, although that requires 60 more thought (and kernel mods). 61 62- Move lfs_countlocked() into vfs_bio.c, to replace count_locked_queue; 63 perhaps keep the name, replace the function. Could it count referenced 64 vnodes as well, if it was in vfs_subr.c instead? 65 66- Why not delete the lfs_bmapv call, just mark everything dirty that 67 isn't deleted/truncated? Get some numbers about what percentage of 68 the stuff that the cleaner thinks might be live is live. If it's 69 high, get rid of lfs_bmapv. 70 71- There is a nasty problem in that it may take *more* room to write the 72 data to clean a segment than is returned by the new segment because of 73 indirect blocks in segment 2 being dirtied by the data being copied 74 into the log from segment 1. The suggested solution at this point is 75 to detect it when we have no space left on the filesystem, write the 76 extra data into the last segment (leaving no clean ones), make it a 77 checkpoint and shut down the file system for fixing by a utility 78 reading the raw partition. Argument is that this should never happen 79 and is practically impossible to fix since the cleaner would have to 80 theoretically build a model of the entire filesystem in memory to 81 detect the condition occurring. A file coalescing cleaner will help 82 avoid the problem, and one that reads/writes from the raw disk could 83 fix it. 84 85- Need to keep vnode v_numoutput up to date for pending writes? 86 87- If delete a file that's being executed, the version number isn't 88 updated, and fsck_lfs has to figure this out; case is the same as if 89 have an inode that no directory references, so the file should be 90 reattached into lost+found. 91 92- Currently there's no notion of write error checking. 93 + Failed data/inode writes should be rescheduled (kernel level bad blocking). 94 + Failed superblock writes should cause selection of new superblock 95 for checkpointing. 96 97- Future fantasies: 98 - unrm, versioning 99 - transactions 100 - extended cleaner policies (hot/cold data, data placement) 101 102- Problem with the concept of multiple buffer headers referencing the segment: 103 Positives: 104 Don't lock down 1 segment per file system of physical memory. 105 Don't copy from buffers to segment memory. 106 Don't tie down the bus to transfer 1M. 107 Works on controllers supporting less than large transfers. 108 Disk can start writing immediately instead of waiting 1/2 rotation 109 and the full transfer. 110 Negatives: 111 Have to do segment write then segment summary write, since the latter 112 is what verifies that the segment is okay. (Is there another way 113 to do this?) 114 115- The algorithm for selecting the disk addresses of the super-blocks 116 has to be available to the user program which checks the file system. 117