maildir-sync.c revision 8153fdec343e40e2a78f5c12353e89b994b28f74
409N/A/* Copyright (C) 2004 Timo Sirainen */ 26N/A Here's a description of how we handle Maildir synchronization and 26N/A We want to be as efficient as we can. The most efficient way to 26N/A check if changes have occured is to stat() the new/ and cur/ 26N/A directories and uidlist file - if their mtimes haven't changed, 26N/A there's no changes and we don't need to do anything. 26N/A Problem 1: Multiple changes can happen within a single second - 26N/A nothing guarantees that once we synced it, someone else didn't just 26N/A then make a modification. Such modifications wouldn't get noticed 26N/A until a new modification occured later. 26N/A Problem 2: Syncing cur/ directory is much more costly than syncing 26N/A new/. Moving mails from new/ to cur/ will always change mtime of 26N/A cur/ causing us to sync it as well. 26N/A Problem 3: We may not be able to move mail from new/ to cur/ 26N/A because we're out of quota, or simply because we're accessing a 576N/A Several checks below use MAILDIR_SYNC_SECS, which should be maximum 576N/A clock drift between all computers accessing the maildir (eg. via 576N/A NFS), rounded up to next second. Our default is 1 second, since 576N/A everyone should be using NTP. 576N/A Note that setting it to 0 works only if there's only one computer 576N/A accessing the maildir. It's practically impossible to make two 696N/A clocks _exactly_ synchronized. 1218N/A It might be possible to only use file server's clock by looking at 1218N/A the atime field, but I don't know how well that would actually work. 1218N/A We have dirty_cur_time variable which is set to cur/ directory's 1218N/A mtime when it's >= time() - MAILDIR_SYNC_SECS and we _think_ we have 1218N/A synchronized the directory. 1218N/A When dirty_cur_time is non-zero, we don't synchronize the cur/ 1218N/A b) opening a mail fails with ENOENT 1218N/A c) time() > dirty_cur_time + MAILDIR_SYNC_SECS 1218N/A This allows us to modify the maildir multiple times without having 1218N/A to sync it at every change. The sync will eventually be done to 1218N/A make sure we didn't miss any external changes. 1218N/A The dirty_cur_time is set when: 1218N/A - we move mail from new/ to cur/ 1218N/A - we sync cur/ directory and it's mtime is >= time() - MAILDIR_SYNC_SECS 1218N/A It's unset when we do the final syncing, ie. when mtime is 1218N/A older than time() - MAILDIR_SYNC_SECS. If new/'s mtime is >= time() - MAILDIR_SYNC_SECS, always synchronize it. dirty_cur_time-like feature might save us a few syncs, but that might break a client which saves a mail in one connection and tries to fetch it in another one. new/ directory is almost always empty, so syncing it should be very fast anyway. Actually this can still happen if we sync only new/ dir while another client is also moving mails from it to cur/ - it takes us a while to see them. That's pretty unlikely to happen however, and only way to fix it would be to always synchronize cur/ after new/. Normally we move all mails from new/ to cur/ whenever we sync it. If it's not possible for some reason, we mark the mail with "probably exists in new/ directory" flag. If rename() still fails because of ENOSPC or EDQUOT, we still save the flag changes in index with dirty-flag on. When moving the mail to cur/ directory, or when we notice it's already moved there, we apply the flag changes to the filename, rename it and remove the dirty flag. If there's dirty flags, this should be tried every time after expunge or when closing the mailbox. This file contains UID <-> filename mappings. It's updated only when new mail arrives, so it may contain filenames that have already been deleted. Updating is done by getting uidlist.lock file, writing the whole uidlist into it and rename()ing it over the old uidlist. This means there's no need to lock the file for reading. Whenever uidlist is rewritten, it's mtime must be larger than the old one's. Use utime() before rename() if needed. Note that inode checking wouldn't have been sufficient as inode numbers can be reused. This file is usually read the first time you need to know filename for given UID. After that it's not re-read unless new mails come that we Originally the middle identifier in Maildir filename was specified only as <process id>_<delivery counter>. That however created a problem with randomized PIDs which made it possible that the same PID was reused within one second. So if within one second a mail was delivered, MUA moved it to cur/ and another mail was delivered by a new process using same PID as the first one, we likely ended up overwriting the first mail when the second mail was moved over it. Nowadays everyone should be giving a bit more specific identifier, for example include microseconds in it which Dovecot does. There's a simple way to prevent this from happening in some cases: Don't move the mail from new/ to cur/ if it's mtime is >= time() - MAILDIR_SYNC_SECS. The second delivery's link() call then fails because the file is already in new/, and it will then use a different filename. There's a few problems with this however: - it requires extra stat() call which is unneeded extra I/O - another MUA might still move the mail to cur/ - if first file's flags are modified by either Dovecot or another MUA, it's moved to cur/ (you _could_ just do the dirty-flagging Because this is useful only for very few people and it requires extra I/O, I decided not to implement this. It should be however quite easy to do since we need to be able to deal with files in new/ It's also possible to never accidentally overwrite a mail by using link() + unlink() rather than rename(). This however isn't very good idea as it introduces potential race conditions when multiple clients are accessing the mailbox: Trying to move the same mail from new/ to cur/ at the same time: a) Client 1 uses slightly different filename than client 2, for example one sets read-flag on but the other doesn't. You have the same mail duplicated now. b) Client 3 sees the mail between Client 1's and 2's link() calls and changes it's flag. You have the same mail duplicated now. And it gets worse when they're unlink()ing in cur/ directory: c) Client 1 changes mails's flag and client 2 changes it back between 1's link() and unlink(). The mail is now expunged. d) If you try to deal with the duplicates by unlink()ing another one of them, you might end up unlinking both of them. So, what should we do then if we notice a duplicate? First of all, it might not be a duplicate at all, readdir() might have just returned it twice because it was just renamed. What we should do is create a completely new base name for it and rename() it to that. If the call fails with ENOENT, it only means that it wasn't a "unlink(%s) failed: %m",
path);
/* make it go through sequences to avoid looping through huge /* flag isn't dirty anymore */ "opendir(%s) failed: %m",
dir);
/* new file and we couldn't lock uidlist, check this /* we moved it - it's \Recent for us */ /* someone else moved it already */ /* not enough disk space, leave here */ "rename(%s, %s) failed: %m",
/* possibly duplicate - try fixing it */ "closedir(%s) failed: %m",
dir);
/* cur stamp is kept in index, we don't have to sync if someone else has done it and updated the index. */ /* check if the index has been updated.. */ /* cur/ changed, or delayed cur/ check */ /* uidvalidity changed and mailbox isn't being initialized, "Maildir %s sync: UIDVALIDITY changed (%u -> %u)",
/* mail is recent for next session as well */ /* most likely a race condition: we read the maildir, then someone else expunged messages and committed changes to index. so, this message shouldn't actually exist. mark it racy and check in next sync. the difference between this and the later check is that this one happens when messages are expunged from the end */ /* most likely a race condition: we read the maildir, then someone else expunged messages and committed changes to index. so, this message shouldn't actually exist. mark it racy and check "UID inserted in the middle of mailbox " /* we haven't been able to update maildir with this record's flag changes. don't sync them. */ /* FIXME: this is wrong if there's pending changes in transaction log already. it gets fixed in next sync /* just remove recent flag */ // FIXME: update keywords /* now, sync the index */ /* get the initial uidvalidity */ Locking, locking, locking.. Wasn't maildir supposed to be lockless? We can get here either as beginning a real maildir sync, or when committing changes to maildir but a file was lost (maybe renamed). So, we're going to need two locks. One for index and one for uidlist. To avoid deadlocking do the index lock first. uidlist is needed only for figuring out UIDs for newly seen files, so theoretically we wouldn't need to lock it unless there are new files. It has a few problems though, assuming the index lock didn't already protect it (eg. in-memory indexes): 1. Just because you see a new file which doesn't exist in uidlist file, doesn't mean that the file really exists anymore, or that your readdir() lists all new files. Meaning that this is possible: A: opendir(), readdir() -> new file ... -- new files are written to the maildir -- B: opendir(), readdir() -> new file, lock uidlist, readdir() -> another new file, rewrite uidlist, unlock A: ... lock uidlist, readdir() -> nothing left, rewrite uidlist, The second time running A didn't see the two new files. To handle this correctly, it must not remove the new unseen files from uidlist. This is possible to do, but adds extra complexity. 2. If another process is rename()ing files while we are readdir()ing, it's possible that readdir() never lists some files, causing Dovecot to assume they were expunged. In next sync they would show up again, but client could have already been notified of that and they would show up under new UIDs, so the damage is Both of the problems can be avoided if we simply lock the uidlist before syncing and keep it until sync is finished. Typically this would happen in any case, as there is the index lock.. The second case is still a problem with external changes though, because maildir doesn't require any kind of locking. Luckily this problem rarely happens except under high amount of modifications. /* failure / timeout. if forced is TRUE, we could still go forward and check only for renamed files, but is it worth /* finish uidlist syncing, but keep it still locked */