- 02 May, 2024 2 commits
-
-
ssh://gitolite.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fsChristian Brauner authored
Pull cifs netfs updates from David Howells: This ports cifs over to use the netfs library. * 'cifs-netfs' of ssh://gitolite.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs: cifs: Enable large folio support cifs: Remove some code that's no longer used, part 3 cifs: Remove some code that's no longer used, part 2 cifs: Remove some code that's no longer used, part 1 cifs: Cut over to using netfslib cifs: Implement netfslib hooks cifs: Make add_credits_and_wake_if() clear deducted credits cifs: Add mempools for cifs_io_request and cifs_io_subrequest structs cifs: Set zero_point in the copy_file_range() and remap_file_range() cifs: Move cifs_loose_read_iter() and cifs_file_write_iter() to file.c cifs: Replace the writedata replay bool with a netfs sreq flag cifs: Make wait_mtu_credits take size_t args cifs: Use more fields from netfs_io_subrequest cifs: Replace cifs_writedata with a wrapper around netfs_io_subrequest cifs: Replace cifs_readdata with a wrapper around netfs_io_subrequest cifs: Use alternative invalidation to using launder_folio Signed-off-by:
Christian Brauner <brauner@kernel.org>
-
Christian Brauner authored
Merge branch 'netfs-writeback' of ssh://gitolite.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs Pull netfs writeback updates from David Howells: The primary purpose of these patches is to rework the netfslib writeback implementation such that pages read from the cache are written to the cache through ->writepages(), thereby allowing the fscache page flag to be retired. The reworking also: (1) builds on top of the new writeback_iter() infrastructure; (2) makes it possible to use vectored write RPCs as discontiguous streams of pages can be accommodated; (3) makes it easier to do simultaneous content crypto and stream division. (4) provides support for retrying writes and re-dividing a stream; (5) replaces the ->launder_folio() op, so that ->writepages() is used instead; (6) uses mempools to allocate the netfs_io_request and netfs_io_subrequest structs to avoid allocation failure in the writeback path. Some code that uses the fscache page flag is retained for compatibility purposes with nfs and ceph. The code is switched to using the synonymous private_2 label instead and marked with deprecation comments. I have a separate set of patches that convert cifs to use this code. -~- In this new implementation, writeback_iter() is used to pump folios, progressively creating two parallel, but separate streams. Either or both streams can contain gaps, and the subrequests in each stream can be of variable size, don't need to align with each other and don't need to align with the folios. (Note that more streams can be added if we have multiple servers to duplicate data to). Indeed, subrequests can cross folio boundaries, may cover several folios or a folio may be spanned by multiple subrequests, e.g.: +---+---+-----+-----+---+----------+ Folios: | | | | | | | +---+---+-----+-----+---+----------+ +------+------+ +----+----+ Upload: | | |.....| | | +------+------+ +----+----+ +------+------+------+------+------+ Cache: | | | | | | +------+------+------+------+------+ Data that got read from the server that needs copying to the cache is stored in folios that are marked dirty and have folio->private set to a special value. The progressive subrequest construction permits the algorithm to be preparing both the next upload to the server and the next write to the cache whilst the previous ones are already in progress. Throttling can be applied to control the rate of production of subrequests - and, in any case, we probably want to write them to the server in ascending order, particularly if the file will be extended. Content crypto can also be prepared at the same time as the subrequests and run asynchronously, with the prepped requests being stalled until the crypto catches up with them. This might also be useful for transport crypto, but that happens at a lower layer, so probably would be harder to pull off. The algorithm is split into three parts: (1) The issuer. This walks through the data, packaging it up, encrypting it and creating subrequests. The part of this that generates subrequests only deals with file positions and spans and so is usable for DIO/unbuffered writes as well as buffered writes. (2) The collector. This asynchronously collects completed subrequests, unlocks folios, frees crypto buffers and performs any retries. This runs in a work queue so that the issuer can return to the caller for writeback (so that the VM can have its kswapd thread back) or async writes. Collection is slightly complex as the collector has to work out where discontiguities happen in the folio list so that it doesn't try and collect folios that weren't included in the write out. (3) The retryer. This pauses the issuer, waits for all outstanding subrequests to complete and then goes through the failed subrequests to reissue them. This may involve reprepping them (with cifs, the credits must be renegotiated and a subrequest may need splitting), and doing RMW for content crypto if there's a conflicting change on the server. * 'netfs-writeback' of ssh://gitolite.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs: (22 commits) netfs, afs: Use writeback retry to deal with alternate keys netfs: Miscellaneous tidy ups netfs: Remove the old writeback code netfs: Cut over to using new writeback code netfs, cachefiles: Implement helpers for new write code netfs, 9p: Implement helpers for new write code netfs, afs: Implement helpers for new write code netfs: Add some write-side stats and clean up some stat names netfs: New writeback implementation netfs: Switch to using unsigned long long rather than loff_t mm: Export writeback_iter() netfs: Use mempools for allocating requests and subrequests netfs: Remove ->launder_folio() support afs: Use alternative invalidation to using launder_folio 9p: Use alternative invalidation to using launder_folio mm: Provide a means of invalidation without using launder_folio netfs: Use subreq_counter to allocate subreq debug_index values netfs: Make netfs_io_request::subreq_counter an atomic_t netfs: Remove deprecated use of PG_private_2 as a second writeback flag mm: Remove the PG_fscache alias for PG_private_2 ... Signed-off-by:
-
- 01 May, 2024 32 commits
-
-
David Howells authored
Now that cifs is using netfslib for its VM interaction, it only sees I/O in terms of iov_iter iterators and does not see pages or folios. This makes large multipage folios transparent to cifs and so we can turn on multipage folios on regular files. Signed-off-by:David Howells <dhowells@redhat.com> cc: Steve French <sfrench@samba.org> cc: Shyam Prasad N <nspmangalore@gmail.com> cc: Rohith Surabattula <rohiths.msft@gmail.com> cc: Jeff Layton <jlayton@kernel.org> cc: linux-cifs@vger.kernel.org cc: netfs@lists.linux.dev cc: linux-fsdevel@vger.kernel.org cc: linux-mm@kvack.org
-
David Howells authored
Remove some code that was #if'd out with the netfslib conversion. This is split into parts for file.c as the diff generator otherwise produces a hard to read diff for part of it where a big chunk is cut out. Signed-off-by: -
David Howells authored
Remove some code that was #if'd out with the netfslib conversion. This is split into parts for file.c as the diff generator otherwise produces a hard to read diff for part of it where a big chunk is cut out. Signed-off-by: -
David Howells authored
Remove some code that was #if'd out with the netfslib conversion. This is split into parts for file.c as the diff generator otherwise produces a hard to read diff for part of it where a big chunk is cut out. Signed-off-by: -
David Howells authored
Make the cifs filesystem use netfslib to handle reading and writing on behalf of cifs. The changes include: (1) Various read_iter/write_iter type functions are turned into wrappers around netfslib API functions or are pointed directly at those functions: cifs_file_direct{,_nobrl}_ops switch to use netfs_unbuffered_read_iter and netfs_unbuffered_write_iter. Large pieces of code that will be removed are #if'd out and will be removed in subsequent patches. [?] Why does cifs mark the page dirty in the destination buffer of a DIO read? Should that happen automatically? Does netfs need to do that? Signed-off-by: -
David Howells authored
Provide implementation of the netfslib hooks that will be used by netfslib to ask cifs to set up and perform operations. Of particular note are (*) cifs_clamp_length() - This is used to negotiate the size of the next subrequest in a read request, taking into account the credit available and the rsize. The credits are attached to the subrequest. (*) cifs_req_issue_read() - This is used to issue a subrequest that has been set up and clamped. (*) cifs_prepare_write() - This prepares to fill a subrequest by picking a channel, reopening the file and requesting credits so that we can set the maximum size of the subrequest and also sets the maximum number of segments if we're doing RDMA. (*) cifs_issue_write() - This releases any unneeded credits and issues an asynchronous data write for the contiguous slice of file covered by the subrequest. This should possibly be folded in to all ->async_writev() ops and that called directly. (*) cifs_begin_writeback() - This gets the cached writable handle through which we do writeback (this does not affect writethrough, unbuffered or direct writes). At this point, cifs is not wired up to actually *use* netfslib; that will be done in a subsequent patch. Signed-off-by: -
David Howells authored
Make add_credits_and_wake_if() clear the amount of credits in the cifs_credits struct after it has returned them to the overall counter. This allows add_credits_and_wake_if() to be called multiple times during the error handling and cleanup without accidentally returning the credits again and again. Note that the wake_up() in add_credits_and_wake_if() may also be superfluous as ->add_credits() also does a wake on the request_q. Signed-off-by: -
David Howells authored
Add mempools for the allocation of cifs_io_request and cifs_io_subrequest structs for netfslib to use so that it can guarantee eventual allocation in writeback. Signed-off-by: -
David Howells authored
Set zero_point in the copy_file_range() and remap_file_range() implementations so that we don't skip reading data modified on a server-side copy. Signed-off-by: -
David Howells authored
Move cifs_loose_read_iter() and cifs_file_write_iter() to file.c so that they are colocated with similar functions rather than being split with cifsfs.c. Signed-off-by: -
David Howells authored
Replace the 'replay' bool in cifs_writedata (now cifs_io_subrequest) with a flag in the netfs_io_subrequest flags. Signed-off-by: -
David Howells authored
Make the wait_mtu_credits functions use size_t for the size and num arguments rather than unsigned int as netfslib uses size_t/ssize_t for arguments and return values to allow for extra capacity. Signed-off-by: -
David Howells authored
Use more fields from netfs_io_subrequest instead of those incorporated into cifs_io_subrequest from cifs_readdata and cifs_writedata. Signed-off-by: -
David Howells authored
Replace the cifs_writedata struct with the same wrapper around netfs_io_subrequest that was used to replace cifs_readdata. Signed-off-by: -
David Howells authored
Netfslib has a facility whereby the allocation for netfs_io_subrequest can be increased to so that filesystem-specific data can be tagged on the end. Prepare to use this by making a struct, cifs_io_subrequest, that wraps netfs_io_subrequest, and absorb struct cifs_readdata into it. Signed-off-by: -
David Howells authored
Use writepages-based flushing invalidation instead of invalidate_inode_pages2() and ->launder_folio(). This will allow ->launder_folio() to be removed eventually. Signed-off-by: -
David Howells authored
Use a hook in the new writeback code's retry algorithm to rotate the keys once all the outstanding subreqs have failed rather than doing it separately on each subreq. Signed-off-by:
Jeff Layton <jlayton@kernel.org> cc: Marc Dionne <marc.dionne@auristor.com> cc: linux-afs@lists.infradead.org cc: netfs@lists.linux.dev cc: linux-fsdevel@vger.kernel.org
-
David Howells authored
Do a couple of miscellaneous tidy ups: (1) Add a qualifier into a file banner comment. (2) Put the writeback folio traces back into alphabetical order. (3) Remove some unused folio traces. Signed-off-by:
-
David Howells authored
Remove the old writeback code. Signed-off-by: -
David Howells authored
Cut over to using the new writeback code. The old code is #ifdef'd out or otherwise removed from compilation to avoid conflicts and will be removed in a future patch. Signed-off-by:
-
David Howells authored
Implement the helpers for the new write code in cachefiles. There's now an optional ->prepare_write() that allows the filesystem to set the parameters for the next write, such as maximum size and maximum segment count, and an ->issue_write() that is called to initiate an (asynchronous) write operation. Signed-off-by:
-
David Howells authored
Implement the helpers for the new write code in 9p. There's now an optional ->prepare_write() that allows the filesystem to set the parameters for the next write, such as maximum size and maximum segment count, and an ->issue_write() that is called to initiate an (asynchronous) write operation. Signed-off-by:
-
David Howells authored
Implement the helpers for the new write code in afs. There's now an optional ->prepare_write() that allows the filesystem to set the parameters for the next write, such as maximum size and maximum segment count, and an ->issue_write() that is called to initiate an (asynchronous) write operation. Signed-off-by:
-
David Howells authored
Add some write-side stats to count buffered writes, buffered writethrough, and writepages calls. Whilst we're at it, clean up the naming on some of the existing stats counters and organise the output into two sets. Signed-off-by:
-
David Howells authored
The current netfslib writeback implementation creates writeback requests of contiguous folio data and then separately tiles subrequests over the space twice, once for the server and once for the cache. This creates a few issues: (1) Every time there's a discontiguity or a change between writing to only one destination or writing to both, it must create a new request. This makes it harder to do vectored writes. (2) The folios don't have the writeback mark removed until the end of the request - and a request could be hundreds of megabytes. (3) In future, I want to support a larger cache granularity, which will require aggregation of some folios that contain unmodified data (which only need to go to the cache) and some which contain modifications (which need to be uploaded and stored to the cache) - but, currently, these are treated as discontiguous. There's also a move to get everyone to use writeback_iter() to extract writable folios from the pagecache. That said, currently writeback_iter() has some issues that make it less than ideal: (1) there's no way to cancel the iteration, even if you find a "temporary" error that means the current folio and all subsequent folios are going to fail; (2) there's no way to filter the folios being written back - something that will impact Ceph with it's ordered snap system; (3) and if you get a folio you can't immediately deal with (say you need to flush the preceding writes), you are left with a folio hanging in the locked state for the duration, when really we should unlock it and relock it later. In this new implementation, I use writeback_iter() to pump folios, progressively creating two parallel, but separate streams and cleaning up the finished folios as the subrequests complete. Either or both streams can contain gaps, and the subrequests in each stream can be of variable size, don't need to align with each other and don't need to align with the folios. Indeed, subrequests can cross folio boundaries, may cover several folios or a folio may be spanned by multiple folios, e.g.: +---+---+-----+-----+---+----------+ Folios: | | | | | | | +---+---+-----+-----+---+----------+ +------+------+ +----+----+ Upload: | | |.....| | | +------+------+ +----+----+ +------+------+------+------+------+ Cache: | | | | | | +------+------+------+------+------+ The progressive subrequest construction permits the algorithm to be preparing both the next upload to the server and the next write to the cache whilst the previous ones are already in progress. Throttling can be applied to control the rate of production of subrequests - and, in any case, we probably want to write them to the server in ascending order, particularly if the file will be extended. Content crypto can also be prepared at the same time as the subrequests and run asynchronously, with the prepped requests being stalled until the crypto catches up with them. This might also be useful for transport crypto, but that happens at a lower layer, so probably would be harder to pull off. The algorithm is split into three parts: (1) The issuer. This walks through the data, packaging it up, encrypting it and creating subrequests. The part of this that generates subrequests only deals with file positions and spans and so is usable for DIO/unbuffered writes as well as buffered writes. (2) The collector. This asynchronously collects completed subrequests, unlocks folios, frees crypto buffers and performs any retries. This runs in a work queue so that the issuer can return to the caller for writeback (so that the VM can have its kswapd thread back) or async writes. (3) The retryer. This pauses the issuer, waits for all outstanding subrequests to complete and then goes through the failed subrequests to reissue them. This may involve reprepping them (with cifs, the credits must be renegotiated, and a subrequest may need splitting), and doing RMW for content crypto if there's a conflicting change on the server. [!] Note that some of the functions are prefixed with "new_" to avoid clashes with existing functions. These will be renamed in a later patch that cuts over to the new algorithm. Signed-off-by: -
David Howells authored
Switch to using unsigned long long rather than loff_t in netfslib to avoid problems with the sign flipping in the maths when we're dealing with the byte at position 0x7fffffffffffffff. Signed-off-by:
-
David Howells authored
Export writeback_iter() so that it can be used by netfslib as a module. Signed-off-by:
-
David Howells authored
Use mempools for allocating requests and subrequests in an effort to make sure that allocation always succeeds so that when performing writeback we can always make progress. Signed-off-by:
-
David Howells authored
Remove support for ->launder_folio() from netfslib and expect filesystems to use filemap_invalidate_inode() instead. netfs_launder_folio() can then be got rid of. Signed-off-by:
-
David Howells authored
Use writepages-based flushing invalidation instead of invalidate_inode_pages2() and ->launder_folio(). This will allow ->launder_folio() to be removed eventually. Signed-off-by: -
David Howells authored
Use writepages-based flushing invalidation instead of invalidate_inode_pages2() and ->launder_folio(). This will allow ->launder_folio() to be removed eventually. Signed-off-by: -
David Howells authored
Implement a replacement for launder_folio. The key feature of invalidate_inode_pages2() is that it locks each folio individually, unmaps it to prevent mmap'd accesses interfering and calls the ->launder_folio() address_space op to flush it. This has problems: firstly, each folio is written individually as one or more small writes; secondly, adjacent folios cannot be added so easily into the laundry; thirdly, it's yet another op to implement. Instead, use the invalidate lock to cause anyone wanting to add a folio to the inode to wait, then unmap all the folios if we have mmaps, then, conditionally, use ->writepages() to flush any dirty data back and then discard all pages. The invalidate lock prevents ->read_iter(), ->write_iter() and faulting through mmap all from adding pages for the duration. This is then used from netfslib to handle the flusing in unbuffered and direct writes. Signed-off-by:
-
- 29 Apr, 2024 6 commits
-
-
David Howells authored
Use the subreq_counter in netfs_io_request to allocate subrequest debug_index values in read ops as well as write ops. Signed-off-by: -
David Howells authored
Make the netfs_io_request::subreq_counter, used to generate values for netfs_io_subrequest::debug_index, into an atomic_t so that it can be called from the retry thread at the same time as the app thread issuing writes. Signed-off-by:
-
David Howells authored
Remove the deprecated use of PG_private_2 in netfslib. Signed-off-by:
-
David Howells authored
Remove the PG_fscache alias for PG_private_2 and use the latter directly. Use of this flag for marking pages undergoing writing to the cache should be considered deprecated and the folios should be marked dirty instead and the write done in ->writepages(). Note that PG_private_2 itself should be considered deprecated and up for future removal by the MM folks too. Signed-off-by:
-
David Howells authored
When dirty data is being written to the cache, setting/waiting on/clearing the fscache flag is always done in tandem with setting/waiting on/clearing the writeback flag. The netfslib buffered write routines wait on and set both flags and the write request cleanup clears both flags, so the fscache flag is almost superfluous. The reason it isn't superfluous is because the fscache flag is also used to indicate that data just read from the server is being written to the cache. The flag is used to prevent a race involving overlapping direct-I/O writes to the cache. Change this to indicate that a page is in need of being copied to the cache by placing a magic value in folio->private and marking the folios dirty. Then when the writeback code sees a folio marked in this way, it only writes it to the cache and not to the server. If a folio that has this magic value set is modified, the value is just replaced and the folio will then be uplodaded too. With this, PG_fscache is no longer required by the netfslib core, 9p and afs. Ceph and nfs, however, still need to use the old PG_fscache-based tracking. To deal with this, a flag, NETFS_ICTX_USE_PGPRIV2, now has to be set on the flags in the netfs_inode struct for those filesystems. This reenables the use of PG_fscache in that inode. 9p and afs use the netfslib write helpers so get switched over; cifs, for the moment, does page-by-page manual access to the cache, so doesn't use PG_fscache and is unaffected. Signed-off-by:
-
David Howells authored
Update i_blocks when i_size is updated when we finish making a write to the pagecache to reflect the amount of space we think will be consumed. This maintains cifs commit dbfdff40 ("smb3: update allocation size more accurately on write completion") which would otherwise be removed by the cifs part of the netfs writeback rewrite. Signed-off-by:
-
