1*5b995484Sbostic# @(#)TODO 8.1 (Berkeley) 06/11/93 235c1176cSbostic 335c1176cSbosticNOTE: Changed the lookup on a page of inodes to search from the back 435c1176cSbosticin case the same inode gets written twice on the same page. 535c1176cSbostic 635c1176cSbosticMake sure that if you are writing a file, but not all the blocks 735c1176cSbosticmake it into a single segment, that you do not write the inode in 835c1176cSbosticthat segment. 935c1176cSbostic 1038aaa36dSbosticKeith: 1159051aa2Sbostic Why not delete the lfs_bmapv call, just mark everything dirty 1259051aa2Sbostic that isn't deleted/truncated? Get some numbers about 1359051aa2Sbostic what percentage of the stuff that the cleaner thinks 1459051aa2Sbostic might be live is live. If it's high, get rid of lfs_bmapv. 15f69fae42Sbostic 16f69fae42Sbostic There is a nasty problem in that it may take *more* room to write 17f69fae42Sbostic the data to clean a segment than is returned by the new segment 18f69fae42Sbostic because of indirect blocks in segment 2 being dirtied by the data 19f69fae42Sbostic being copied into the log from segment 1. The suggested solution 20f69fae42Sbostic at this point is to detect it when we have no space left on the 21f69fae42Sbostic filesystem, write the extra data into the last segment (leaving 22f69fae42Sbostic no clean ones), make it a checkpoint and shut down the file system 23f69fae42Sbostic for fixing by a utility reading the raw partition. Argument is 24f69fae42Sbostic that this should never happen and is practically impossible to fix 25f69fae42Sbostic since the cleaner would have to theoretically build a model of the 26f69fae42Sbostic entire filesystem in memory to detect the condition occurring. 27f69fae42Sbostic A file coalescing cleaner will help avoid the problem, and one 28f69fae42Sbostic that reads/writes from the raw disk could fix it. 29f69fae42Sbostic 301666c792SbosticDONE Currently, inodes are being flushed to disk synchronously upon 3159051aa2Sbostic creation -- see ufs_makeinode. However, only the inode 3259051aa2Sbostic is flushed, the directory "name" is written using VOP_BWRITE, 3359051aa2Sbostic so it's not synchronous. Possible solutions: 1: get some 3459051aa2Sbostic ordering in the writes so that inode/directory entries get 3559051aa2Sbostic stuffed into the same segment. 2: do both synchronously 3659051aa2Sbostic 3: add Mendel's information into the stream so we log 3759051aa2Sbostic creation/deletion of inodes. 4: do some form of partial 3859051aa2Sbostic segment when changing the inode (creation/deletion/rename). 391666c792SbosticDONE Fix i_block increment for indirect blocks. 4054a53d58Sbostic If the file system is tar'd, extracted on top of another LFS, the 4154a53d58Sbostic IFILE ain't worth diddly. Is the cleaner writing the IFILE? 4254a53d58Sbostic If not, let's make it read-only. 431666c792SbosticDONE Delete unnecessary source from utils in main-line source tree. 441666c792SbosticDONE Make sure that we're counting meta blocks in the inode i_block count. 4554a53d58Sbostic Overlap the version and nextfree fields in the IFILE 461666c792SbosticDONE Vinvalbuf (Kirk): 4754a53d58Sbostic Why writing blocks that are no longer useful? 4854a53d58Sbostic Are the semantics of close such that blocks have to be flushed? 4954a53d58Sbostic How specify in the buf chain the blocks that don't need 5054a53d58Sbostic to be written? (Different numbering of indirect blocks.) 5138aaa36dSbostic 5238aaa36dSbosticMargo: 53c0d3060eSbostic Change so that only search one sector of inode block file for the 54c0d3060eSbostic inode by using sector addresses in the ifile instead of 55c0d3060eSbostic logical disk addresses. 56c0d3060eSbostic Fix the use of the ifile version field to use the generation 57c0d3060eSbostic number instead. 581666c792SbosticDONE Unmount; not doing a bgetvp (VHOLD) in lfs_newbuf call. 591666c792SbosticDONE Document in the README file where the checkpoint information is 6037a202d6Sbostic on disk. 6138aaa36dSbostic Variable block sizes (Margo/Keith). 6238aaa36dSbostic Switch the byte accounting to sector accounting. 631666c792SbosticDONE Check lfs.h and make sure that the #defines/structures are all 6438aaa36dSbostic actually needed. 651666c792SbosticDONE Add a check in lfs_segment.c so that if the segment is empty, 661666c792Sbostic we don't write it. 6738aaa36dSbostic Need to keep vnode v_numoutput up to date for pending writes? 681666c792SbosticDONE USENIX paper (Carl/Margo). 6938aaa36dSbostic 701666c792Sbostic 711666c792SbosticEvelyn: 7238aaa36dSbostic lfsck: If delete a file that's being executed, the version number 7338aaa36dSbostic isn't updated, and lfsck has to figure this out; case is the same as if have an inode that no directory references, 7438aaa36dSbostic so the file should be reattached into lost+found. 751666c792Sbostic Recovery/fsck. 761666c792Sbostic 771666c792SbosticCarl: 7838aaa36dSbostic Investigate: clustering of reads (if blocks in the segment are ordered, 7938aaa36dSbostic should read them all) and writes (McVoy paper). 8038aaa36dSbostic Investigate: should the access time be part of the IFILE: 8138aaa36dSbostic pro: theoretically, saves disk writes 8238aaa36dSbostic con: cacheing inodes should obviate this advantage 8338aaa36dSbostic the IFILE is already humongous 8438aaa36dSbostic Cleaner. 8538aaa36dSbostic Port to OSF/1 (Carl/Keith). 8638aaa36dSbostic Currently there's no notion of write error checking. 8738aaa36dSbostic + Failed data/inode writes should be rescheduled (kernel level 8838aaa36dSbostic bad blocking). 8938aaa36dSbostic + Failed superblock writes should cause selection of new 9038aaa36dSbostic superblock for checkpointing. 9138aaa36dSbostic 9238aaa36dSbosticFUTURE FANTASIES: ============ 9338aaa36dSbostic 941666c792Sbostic+ unrm, versioning 9538aaa36dSbostic+ transactions 961666c792Sbostic+ extended cleaner policies (hot/cold data, data placement) 9738aaa36dSbostic 9838aaa36dSbostic============================== 9938aaa36dSbosticProblem with the concept of multiple buffer headers referencing the segment: 10038aaa36dSbosticPositives: 10138aaa36dSbostic Don't lock down 1 segment per file system of physical memory. 10238aaa36dSbostic Don't copy from buffers to segment memory. 10338aaa36dSbostic Don't tie down the bus to transfer 1M. 10438aaa36dSbostic Works on controllers supporting less than large transfers. 10538aaa36dSbostic Disk can start writing immediately instead of waiting 1/2 rotation 10638aaa36dSbostic and the full transfer. 10738aaa36dSbosticNegatives: 10838aaa36dSbostic Have to do segment write then segment summary write, since the latter 10938aaa36dSbostic is what verifies that the segment is okay. (Is there another way 11038aaa36dSbostic to do this?) 11138aaa36dSbostic============================== 11238aaa36dSbostic 11338aaa36dSbosticThe algorithm for selecting the disk addresses of the super-blocks 11438aaa36dSbostichas to be available to the user program which checks the file system. 11538aaa36dSbostic 11638aaa36dSbostic(Currently in newfs, becomes a common subroutine.) 117