Segfault in VFSs not setting block size in stat upon overwriting existing file

[zaytsev]

See #3406 for the details.

Reproducer:

1) Connect by SFTP
2) Copy local file to remote
3) Overwrite local file on remote
4) Observe segfault if unlucky

[zaytsev]

Andrew, so my guess is correct, what do we do about the segfault?

Obviously my dirty patch can be cleaned up (move the header to lib, remove disgusting copy-paste from sftp vfs, etc.), but I'm afraid we have a much bigger problem than that :-/ Basically any VFS that doesn't properly initialize st_blksize / st_blocks is broken after the optimization...

Do you think we should try to fix all sites where stat is being set, or rather add a wrapper around the block size detection thing?

[andrew_b]

Branch: 3749_vfs_blksize
Initial changeset:d49d3b906f1079de9ea55dfee01922e870cc0936

[zaytsev]

Do you think you could also customize blksize for ftpfs and fish? They seem to use 8k buffers which matched the default 8k buffer in the copy operation. It feels like 512 bytes will be... slow?

[andrew_b]

Replying to zaytsev:

Do you think you could also customize blksize for ftpfs and fish?

Done. Please review.

[zaytsev]

Travis doesn't like it:

Sources not indented

src/consaver/cons.saver.c
src/vfs/fish/fish.c
src/vfs/ftpfs/ftpfs.c

[zaytsev]

Now Travis is happy, and I left a couple of comments on GitHub?.

[zaytsev]

I like the last commit very much, is there any other work to do in this branch, or I can vote for it?

[andrew_b]

Replying to zaytsev:

is there any other work to do in this branch,

No.

or I can vote for it?

Please.

[andrew_b]

Merged to master: [9d735e02f17f0c77dd4916193de353b5ed266377].

git log --pretty=oneline 9478740..9d735e0