From bb413489288e4e457353bac513fddb6330d245ca Mon Sep 17 00:00:00 2001 From: David Howells Date: Fri, 12 Jun 2020 00:15:13 +0100 Subject: afs: Fix non-setting of mtime when writing into mmap The mtime on an inode needs to be updated when a write is made into an mmap'ed section. There are three ways in which this could be done: update it when page_mkwrite is called, update it when a page is changed from dirty to writeback or leave it to the server and fix the mtime up from the reply to the StoreData RPC. Found with the generic/215 xfstest. Fixes: 1cf7a1518aef ("afs: Implement shared-writeable mmap") Signed-off-by: David Howells --- fs/afs/write.c | 1 + 1 file changed, 1 insertion(+) (limited to 'fs/afs/write.c') diff --git a/fs/afs/write.c b/fs/afs/write.c index 768497f82aee..9270bb01be67 100644 --- a/fs/afs/write.c +++ b/fs/afs/write.c @@ -844,6 +844,7 @@ vm_fault_t afs_page_mkwrite(struct vm_fault *vmf) vmf->page->index, priv); SetPagePrivate(vmf->page); set_page_private(vmf->page, priv); + file_update_time(file); sb_end_pagefault(inode->i_sb); return VM_FAULT_LOCKED; -- cgit From 1f32ef79897052ef7d3d154610d8d6af95abde83 Mon Sep 17 00:00:00 2001 From: David Howells Date: Fri, 12 Jun 2020 23:58:51 +0100 Subject: afs: afs_write_end() should change i_size under the right lock Fix afs_write_end() to change i_size under vnode->cb_lock rather than ->wb_lock so that it doesn't race with afs_vnode_commit_status() and afs_getattr(). The ->wb_lock is only meant to guard access to ->wb_keys which isn't accessed by that piece of code. Fixes: 4343d00872e1 ("afs: Get rid of the afs_writeback record") Signed-off-by: David Howells --- fs/afs/write.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'fs/afs/write.c') diff --git a/fs/afs/write.c b/fs/afs/write.c index 9270bb01be67..a55cb73e0449 100644 --- a/fs/afs/write.c +++ b/fs/afs/write.c @@ -194,11 +194,11 @@ int afs_write_end(struct file *file, struct address_space *mapping, i_size = i_size_read(&vnode->vfs_inode); if (maybe_i_size > i_size) { - spin_lock(&vnode->wb_lock); + write_seqlock(&vnode->cb_lock); i_size = i_size_read(&vnode->vfs_inode); if (maybe_i_size > i_size) i_size_write(&vnode->vfs_inode, maybe_i_size); - spin_unlock(&vnode->wb_lock); + write_sequnlock(&vnode->cb_lock); } if (!PageUptodate(page)) { -- cgit From da8d07551275abb3a38fae2d16e02bc9cc7396b2 Mon Sep 17 00:00:00 2001 From: David Howells Date: Sat, 13 Jun 2020 19:34:59 +0100 Subject: afs: Concoct ctimes The in-kernel afs filesystem ignores ctime because the AFS fileserver protocol doesn't support ctimes. This, however, causes various xfstests to fail. Work around this by: (1) Setting ctime to attr->ia_ctime in afs_setattr(). (2) Not ignoring ATTR_MTIME_SET, ATTR_TIMES_SET and ATTR_TOUCH settings. (3) Setting the ctime from the server mtime when on the target file when creating a hard link to it. (4) Setting the ctime on directories from their revised mtimes when renaming/moving a file. Found by the generic/221 and generic/309 xfstests. Signed-off-by: David Howells --- fs/afs/write.c | 1 + 1 file changed, 1 insertion(+) (limited to 'fs/afs/write.c') diff --git a/fs/afs/write.c b/fs/afs/write.c index a55cb73e0449..2003d7ee9e43 100644 --- a/fs/afs/write.c +++ b/fs/afs/write.c @@ -393,6 +393,7 @@ static void afs_store_data_success(struct afs_operation *op) { struct afs_vnode *vnode = op->file[0].vnode; + op->ctime = op->file[0].scb.status.mtime_client; afs_vnode_commit_status(op, &op->file[0]); if (op->error == 0) { afs_pages_written_back(vnode, op->store.first, op->store.last); -- cgit From 793fe82ee33aab1023cf023cd7d744af19a3dff9 Mon Sep 17 00:00:00 2001 From: David Howells Date: Fri, 12 Jun 2020 16:13:52 +0100 Subject: afs: Fix truncation issues and mmap writeback size Fix the following issues: (1) Fix writeback to reduce the size of a store operation to i_size, effectively discarding the extra data. The problem comes when afs_page_mkwrite() records that a page is about to be modified by mmap(). It doesn't know what bits of the page are going to be modified, so it records the whole page as being dirty (this is stored in page->private as start and end offsets). Without this, the marshalling for the store to the server extends the size of the file to the end of the page (in afs_fs_store_data() and yfs_fs_store_data()). (2) Fix setattr to actually truncate the pagecache, thereby clearing the discarded part of a file. (3) Fix setattr to check that the new size is okay and to disable ATTR_SIZE if i_size wouldn't change. (4) Force i_size to be updated as the result of a truncate. (5) Don't truncate if ATTR_SIZE is not set. (6) Call pagecache_isize_extended() if the file was enlarged. Note that truncate_set_size() isn't used because the setting of i_size is done inside afs_vnode_commit_status() under the vnode->cb_lock. Found with the generic/029 and generic/393 xfstests. Fixes: 31143d5d515e ("AFS: implement basic file write support") Fixes: 4343d00872e1 ("afs: Get rid of the afs_writeback record") Signed-off-by: David Howells --- fs/afs/write.c | 6 ++++++ 1 file changed, 6 insertions(+) (limited to 'fs/afs/write.c') diff --git a/fs/afs/write.c b/fs/afs/write.c index 2003d7ee9e43..7437806332d9 100644 --- a/fs/afs/write.c +++ b/fs/afs/write.c @@ -492,6 +492,7 @@ static int afs_write_back_from_locked_page(struct address_space *mapping, unsigned long count, priv; unsigned n, offset, to, f, t; pgoff_t start, first, last; + loff_t i_size, end; int loop, ret; _enter(",%lx", primary_page->index); @@ -592,7 +593,12 @@ no_more: first = primary_page->index; last = first + count - 1; + end = (loff_t)last * PAGE_SIZE + to; + i_size = i_size_read(&vnode->vfs_inode); + _debug("write back %lx[%u..] to %lx[..%u]", first, offset, last, to); + if (end > i_size) + to = i_size & ~PAGE_MASK; ret = afs_store_data(mapping, first, last, offset, to); switch (ret) { -- cgit From 811f04bac15181a3351ef1d1aaa377954056e93b Mon Sep 17 00:00:00 2001 From: David Howells Date: Wed, 8 Jul 2020 09:27:07 +0100 Subject: afs: Fix interruption of operations The afs filesystem driver allows unstarted operations to be cancelled by signal, but most of these can easily be restarted (mkdir for example). The primary culprits for reproducing this are those applications that use SIGALRM to display a progress counter. File lock-extension operation is marked uninterruptible as we have a limited time in which to do it, and the release op is marked uninterruptible also as if we fail to unlock a file, we'll have to wait 20 mins before anyone can lock it again. The store operation logs a warning if it gets interruption, e.g.: kAFS: Unexpected error from FS.StoreData -4 because it's run from the background - but it can also be run from fdatasync()-type things. However, store options aren't marked interruptible at the moment. Fix this in the following ways: (1) Mark store operations as uninterruptible. It might make sense to relax this for certain situations, but I'm not sure how to make sure that background store ops aren't affected by signals to foreground processes that happen to trigger them. (2) In afs_get_io_locks(), where we're getting the serialisation lock for talking to the fileserver, return ERESTARTSYS rather than EINTR because a lot of the operations (e.g. mkdir) are restartable if we haven't yet started sending the op to the server. Fixes: e49c7b2f6de7 ("afs: Build an abstraction around an "operation" concept") Signed-off-by: David Howells Signed-off-by: Linus Torvalds --- fs/afs/write.c | 1 + 1 file changed, 1 insertion(+) (limited to 'fs/afs/write.c') diff --git a/fs/afs/write.c b/fs/afs/write.c index 7437806332d9..a121c247d95a 100644 --- a/fs/afs/write.c +++ b/fs/afs/write.c @@ -449,6 +449,7 @@ static int afs_store_data(struct address_space *mapping, op->store.first_offset = offset; op->store.last_to = to; op->mtime = vnode->vfs_inode.i_mtime; + op->flags |= AFS_OPERATION_UNINTR; op->ops = &afs_store_data_operation; try_next_key: -- cgit