Skip to contentSkip to conversation

fscache,erofs: fscache-based on-demand read semantics

PATCH Superseded Jeffle Xu 4 participants 48 messages Apr 15, 2022 – Apr 22, 2022

Subsystems (6)

v9 of 11 Superseded Latest is v11 →

lore
Share & tools
lore

Jeffle Xu

author
#0 lore raw

v9 cover fscache,erofs: fscache-based on-demand read semantics

changes since v8:
- rebase to 5.18-rc2
- cachefiles: use object_id rather than anon_fd to uniquely identify a cachefile object to avoid potential issues when the user moves the anonymous fd around, e.g. through dup() (refer to commit message and cachefiles_ondemand_get_fd() of patch 2 for more details) (David Howells)
- cachefiles: add @unbind_pincount refcount to avoid the potential deadlock (refer to commit message of patch3 for more details)
- cachefiles: move the calling site of cachefiles_ondemand_read() from cachefiles_read() to cacehfiles_prep_read() (refer to commit message of patch 5 for more details)
- cachefiles: add tracepoints (patch 7) (David Howells)
- cachefiles: update documentation (patch 8) (David Howells)
- erofs: update Reviewed-by tag from Gao Xiang
- erofs: move the logic of initializing bdev/dax_dev in fscache mode out from patch 15/20. Instead move it into patch 9, so that patch 20 can focus on the mount option handling
- erofs: update the subject line and commit message of patch 12 (Gao Xiang)
- erofs: remove and fold erofs_fscache_get_folio() helper (patch 16) (Gao Xiang)
- erofs: change kmap() to kamp_loacl_folio(), and comment cleanup (patch
18) (Gao Xiang)
- update "advantage of fscache-based on-demand read" section of the cover letter
- we've finished a preliminary end-to-end on-demand download daemon in order to test the fscache on-demand kernel code as a real end-to-end workload for container use cases. The test user guide is added in the cover letter.
- Thanks Zichen Tian for testing

Kernel Patchset
---------------
Git tree:

https://github.com/lostjeffle/linux.git jingbo/dev-erofs-fscache-v9

Gitweb:

https://github.com/lostjeffle/linux/commits/jingbo/dev-erofs-fscache-v9

User Guide for E2E Container Use Case
-------------------------------------
User guide:

https://github.com/dragonflyoss/image-service/blob/fscache/docs/nydus-fscache.md

Video:

    https://youtu.be/F4IF2_DENXo

User Daemon for Quick Test
--------------------------
Git tree:

https://github.com/lostjeffle/demand-read-cachefilesd.git main

Gitweb:

https://github.com/lostjeffle/demand-read-cachefilesd

RFC: https://lore.kernel.org/all/[email protected]/t/
v1: https://lore.kernel.org/lkml/[email protected]/T/
v2: https://lore.kernel.org/all/[email protected]/t/
v3: https://lore.kernel.org/lkml/[email protected]/T/
v4: https://lore.kernel.org/lkml/[email protected]/T/#t
v5: https://lore.kernel.org/lkml/[email protected]/T/
v6: https://lore.kernel.org/lkml/[email protected]/T/
v7: https://lore.kernel.org/lkml/[email protected]/T/
v8: https://lore.kernel.org/all/[email protected]/T/

[Background]
============
Nydus [1] is an image distribution service especially optimized for distribution over network. Nydus is an excellent container image acceleration solution, since it only pulls data from remote when needed, a.k.a. on-demand reading and it also supports chunk-based deduplication, compression, etc.

erofs (Enhanced Read-Only File System) is a filesystem designed for read-only scenarios. (Documentation/filesystem/erofs.rst)

Over the past months we've been focusing on supporting Nydus image service with in-kernel erofs format[2]. In that case, each container image will be organized in one bootstrap (metadata) and (optional) multiple data blobs in erofs format. Massive container images will be stored on one machine.

To accelerate the container startup (fetching container images from remote and then start the container), we do hope that the bootstrap & blob files could support on-demand read. That is, erofs can be mounted and accessed even when the bootstrap/data blob files have not been fully downloaded. Then it'll have native performance after data is available locally.

That means we have to manage the cache state of the bootstrap/data blob files (if cache hit, read directly from the local cache; if cache miss, fetch the data somehow). It would be painful and may be dumb for erofs to implement the cache management itself. Thus we prefer fscache/cachefiles to do the cache management instead.

The fscache on-demand read feature aims to be implemented in a generic way so that it can benefit other use cases and/or filesystems if it's implemented in the fscache subsystem.

[1] https://nydus.dev
[2] https://sched.co/pcdL

[Overall Design]
================
Please refer to patch 7 ("cachefiles: document on-demand read mode") for more details.

When working in the original mode, cachefiles mainly serves as a local cache for remote networking fs, while in on-demand read mode, cachefiles can work in the scenario where on-demand read semantics is needed, e.g. container image distribution.

The essential difference between these two modes is that, in original mode, when cache miss, netfs itself will fetch data from remote, and then write the fetched data into cache file. While in on-demand read mode, a user daemon is responsible for fetching data and then feeds to the kernel fscache side.

The on-demand read mode relies on a simple protocol used for communication between kernel and user daemon.

The proposed implementation relies on the anonymous fd mechanism to avoid the dependence on the format of cache file. When a fscache cachefile is opened for the first time, an anon_fd associated with the cache file is sent to the user daemon. With the given anon_fd, user daemon could fetch and write data into the cache file in the background, even when kernel has not triggered the cache miss. Besides, the write() syscall to the anon_fd will finally call cachefiles kernel module, which will write data to cache file in the latest format of cache file.

1. cache miss
When cache miss, cachefiles kernel module will notify user daemon with the anon_fd, along with the requested file range. When notified, user daemon needs to fetch data of the requested file range, and then write the fetched data into cache file with the given anonymous fd. When finished processing the request, user daemon needs to notify the kernel.

After notifying the user daemon, the kernel read routine will hang there, until the request is handled by user daemon. When it's awaken by the notification from user daemon, i.e. the corresponding hole has been filled by the user daemon, it will retry to read from the same file range.

2. cache hit
Once data is already ready in cache file, netfs will read from cache file directly.

[Advantage of fscache-based on-demand read]
========================================
1. Asynchronous prefetch
In current mechanism, fscache is responsible for cache state management, while the data plane (fetching data from local/remote on cache miss) is done on the user daemon side even without any file system request driven. In addition, if cached data has already been available locally, fscache will use it instead of trapping to user space anymore.

Therefore, different from event-driven approaches, the fscache on-demand user daemon could also fetch data (from remote) asynchronously in the background just like most multi-threaded HTTP downloaders.

2. Flexible request amplification
Since the data plane can be independently controlled by the user daemon, the user daemon can also fetch more data from remote than that the file system actually requests for small I/O sizes. Then, fetched data in bulk will be available at once and fscache won't be trapped into the user daemon again.

3. Support massive blobs
This mechanism can naturally support a large amount of backing files, and thus can benefit the densely employed scenarios. In our use cases, one container image can be formed of one bootstrap (required) and multiple chunk-deduplicated data blobs (optional).

For example, one container image for node.js will correspond to ~20 files in total. In densely employed environment, there could be hundreds of containers and thus thousands of backing files on one machine.

Jeffle Xu (21):
cachefiles: extract write routine
cachefiles: notify user daemon when looking up cookie
cachefiles: unbind cachefiles gracefully in on-demand mode
cachefiles: notify user daemon when withdrawing cookie
cachefiles: implement on-demand read
cachefiles: enable on-demand read mode
cachefiles: add tracepoints for on-demand read mode
cachefiles: document on-demand read mode
erofs: make erofs_map_blocks() generally available
erofs: add fscache mode check helper
erofs: register fscache volume
erofs: add fscache context helper functions
erofs: add anonymous inode caching metadata for data blobs
erofs: add erofs_fscache_read_folios() helper
erofs: register fscache context for primary data blob
erofs: register fscache context for extra data blobs
erofs: implement fscache-based metadata read
erofs: implement fscache-based data read for non-inline layout
erofs: implement fscache-based data read for inline layout
erofs: implement fscache-based data readahead
erofs: add 'fsid' mount option

 .../filesystems/caching/cachefiles.rst        | 170 ++++++
 fs/cachefiles/Kconfig                         |  11 +
 fs/cachefiles/Makefile                        |   1 +
 fs/cachefiles/daemon.c                        | 116 +++-
 fs/cachefiles/interface.c                     |   2 +
 fs/cachefiles/internal.h                      |  74 +++
 fs/cachefiles/io.c                            |  76 ++-
 fs/cachefiles/namei.c                         |  16 +-
 fs/cachefiles/ondemand.c                      | 496 ++++++++++++++++++
 fs/erofs/Kconfig                              |  10 +
 fs/erofs/Makefile                             |   1 +
 fs/erofs/data.c                               |  26 +-
 fs/erofs/fscache.c                            | 365 +++++++++++++
 fs/erofs/inode.c                              |   4 +
 fs/erofs/internal.h                           |  49 ++
 fs/erofs/super.c                              | 105 +++-
 fs/erofs/sysfs.c                              |   4 +-
 include/linux/fscache.h                       |   1 +
 include/linux/netfs.h                         |   2 +
 include/trace/events/cachefiles.h             | 176 +++++++
 include/uapi/linux/cachefiles.h               |  68 +++
 21 files changed, 1694 insertions(+), 79 deletions(-)
 create mode 100644 fs/cachefiles/ondemand.c
 create mode 100644 fs/erofs/fscache.c
 create mode 100644 include/uapi/linux/cachefiles.h
Signature · 1 line
2.27.0
author
#1 lore raw

PATCH v9 1/21 cachefiles: extract write routine

Extract the generic routine of writing data to cache files, and make it generally available.

This will be used by the following patch implementing on-demand read mode. Since it's called inside cachefiles module in this case, make the interface generic and unrelated to netfs_cache_resources.

It is worth noting that, ki->inval_counter is not initialized after this cleanup. It shall not make any visible difference, since inval_counter is no longer used in the write completion routine, i.e. cachefiles_write_complete().

---
 fs/cachefiles/internal.h | 10 +++++++
 fs/cachefiles/io.c       | 61 +++++++++++++++++++++++-----------------
 2 files changed, 45 insertions(+), 26 deletions(-)
fs/cachefiles/internal.h +10 -0
index c793d33b0224..e80673d0ab97 100644
@@ -201,6 +201,16 @@ extern void cachefiles_put_object(struct cachefiles_object *object,
  */
 extern bool cachefiles_begin_operation(struct netfs_cache_resources *cres,
 				       enum fscache_want_state want_state);
+extern int __cachefiles_prepare_write(struct cachefiles_object *object,
+				      struct file *file,
+				      loff_t *_start, size_t *_len,
+				      bool no_space_allocated_yet);
+extern int __cachefiles_write(struct cachefiles_object *object,
+			      struct file *file,
+			      loff_t start_pos,
+			      struct iov_iter *iter,
+			      netfs_io_terminated_t term_func,
+			      void *term_func_priv);
 
 /*
  * key.c
fs/cachefiles/io.c +35 -26

109 lines not on this page · raw message

Signature · 1 line
2.27.0
maintainer Acked-by on v9 1/21
#2 lore raw

Re: [PATCH v9 01/21] cachefiles: extract write routine

Quoted 7 lines · Jeffle Xu

Extract the generic routine of writing data to cache files, and make it generally available.

This will be used by the following patch implementing on-demand read mode. Since it's called inside cachefiles module in this case, make the interface generic and unrelated to netfs_cache_resources.

Quoted 6 lines

It is worth noting that, ki->inval_counter is not initialized after this cleanup. It shall not make any visible difference, since inval_counter is no longer used in the write completion routine, i.e. cachefiles_write_complete().

author
#3 lore raw

PATCH v9 2/21 cachefiles: notify user daemon when looking up cookie

In reply to Jeffle Xu

Fscache/cachefiles used to serve as a local cache for a remote networking fs. This patch, along with the following patches, introduces a new on-demand read mode for cachefiles, which can boost the scenario where on-demand read semantics is needed, e.g. container image distribution.

The essential difference between these two modes is that, in original mode, when a cache miss occurs, the netfs will fetch the data from the remote server and then write it to the cache file. With on-demand read mode, however, fetching the data and writing it to the cache is delegated to a user daemon.

As the first step, notify the user daemon when looking up cookie. In this case, an anonymous fd is sent to the user daemon, through which the user daemon can write the fetched data to the cache file. Since the user daemon may move the anonymous fd around, e.g. through dup(), an object ID uniquely identifying the cache file is also attached.

Also add one advisory flag (FSCACHE_ADV_WANT_CACHE_SIZE) suggesting that cache file size shall be retrieved at runtime. This helps the scenario where one cache file contains multiple netfs files, e.g. for the purpose of deduplication. In this case, netfs itself has no idea the size of the cache file, whilst the user daemon needs to give the hint on it.

---
 fs/cachefiles/Kconfig             |  11 +
 fs/cachefiles/Makefile            |   1 +
 fs/cachefiles/daemon.c            |  79 +++++--
 fs/cachefiles/internal.h          |  47 ++++
 fs/cachefiles/namei.c             |  16 +-
 fs/cachefiles/ondemand.c          | 371 ++++++++++++++++++++++++++++++
 include/linux/fscache.h           |   1 +
 include/trace/events/cachefiles.h |   2 +
 include/uapi/linux/cachefiles.h   |  50 ++++
 9 files changed, 563 insertions(+), 15 deletions(-)
 create mode 100644 fs/cachefiles/ondemand.c
 create mode 100644 include/uapi/linux/cachefiles.h
fs/cachefiles/Kconfig +11 -0
index 719faeeda168..67371cb1eb08 100644
@@ -26,3 +26,14 @@ config CACHEFILES_ERROR_INJECTION
 	help
 	  This permits error injection to be enabled in cachefiles whilst a
 	  cache is in service.
+
+config CACHEFILES_ONDEMAND
+	bool "Support for on-demand read"
+	depends on CACHEFILES
+	default n
+	help
+	  This permits on-demand read mode of cachefiles.  In this mode, when
+	  cache miss, the cachefiles backend instead of netfs, is responsible
+	  for fetching data, e.g. through user daemon.
+
+	  If unsure, say N.
fs/cachefiles/Makefile +1 -0
index 16d811f1a2fa..c37a7a9af10b 100644
@@ -16,5 +16,6 @@ cachefiles-y := \
 	xattr.o
 
 cachefiles-$(CONFIG_CACHEFILES_ERROR_INJECTION) += error_inject.o
+cachefiles-$(CONFIG_CACHEFILES_ONDEMAND) += ondemand.o
 
 obj-$(CONFIG_CACHEFILES) := cachefiles.o
fs/cachefiles/daemon.c +66 -13

127 lines not on this page · raw message

fs/cachefiles/internal.h +47 -0

78 lines not on this page · raw message

fs/cachefiles/namei.c +14 -2
index ca9f3e4ec4b3..facf2ebe464b 100644
@@ -452,10 +452,9 @@ struct file *cachefiles_create_tmpfile(struct cachefiles_object *object)
 	struct dentry *fan = volume->fanout[(u8)object->cookie->key_hash];
 	struct file *file;
 	struct path path;
-	uint64_t ni_size = object->cookie->object_size;
+	uint64_t ni_size;
 	long ret;
 
-	ni_size = round_up(ni_size, CACHEFILES_DIO_BLOCK_SIZE);
 
 	cachefiles_begin_secure(cache, &saved_cred);
 
@@ -481,6 +480,15 @@ struct file *cachefiles_create_tmpfile(struct cachefiles_object *object)
 		goto out_dput;
 	}
 
+	ret = cachefiles_ondemand_init_object(object);
+	if (ret < 0) {
+		file = ERR_PTR(ret);
+		goto out_unuse;
+	}
+
+	ni_size = object->cookie->object_size;
+	ni_size = round_up(ni_size, CACHEFILES_DIO_BLOCK_SIZE);
+
 	if (ni_size > 0) {
 		trace_cachefiles_trunc(object, d_backing_inode(path.dentry), 0, ni_size,
 				       cachefiles_trunc_expand_tmpfile);
@@ -586,6 +594,10 @@ static bool cachefiles_open_file(struct cachefiles_object *object,
 	}
 	_debug("file -> %pd positive", dentry);
 
+	ret = cachefiles_ondemand_init_object(object);
+	if (ret < 0)
+		goto error_fput;
+
 	ret = cachefiles_check_auxdata(object, file);
 	if (ret < 0)
 		goto check_failed;
fs/cachefiles/ondemand.c new file +371 -0

371 lines not on this page · raw message

include/linux/fscache.h +1 -0
index e25539072463..72585c9729a2 100644
@@ -39,6 +39,7 @@ struct fscache_cookie;
 #define FSCACHE_ADV_SINGLE_CHUNK	0x01 /* The object is a single chunk of data */
 #define FSCACHE_ADV_WRITE_CACHE		0x00 /* Do cache if written to locally */
 #define FSCACHE_ADV_WRITE_NOCACHE	0x02 /* Don't cache if written to locally */
+#define FSCACHE_ADV_WANT_CACHE_SIZE	0x04 /* Retrieve cache size at runtime */
 
 #define FSCACHE_INVAL_DIO_WRITE		0x01 /* Invalidate due to DIO write */
 
include/trace/events/cachefiles.h +2 -0
index 311c14a20e70..93df9391bd7f 100644
@@ -31,6 +31,8 @@ enum cachefiles_obj_ref_trace {
 	cachefiles_obj_see_lookup_failed,
 	cachefiles_obj_see_withdraw_cookie,
 	cachefiles_obj_see_withdrawal,
+	cachefiles_obj_get_ondemand_fd,
+	cachefiles_obj_put_ondemand_fd,
 };
 
 enum fscache_why_object_killed {
include/uapi/linux/cachefiles.h new file +50 -0

50 lines not on this page · raw message

Signature · 1 line
2.27.0
maintainer
#4 lore raw

Re: [PATCH v9 02/21] cachefiles: notify user daemon when looking up cookie

fs/cachefiles/Kconfig config CACHEFILES_ERROR_INJECTION · line 37

Quoted 4 lines · Jeffle Xu
fs/cachefiles/Kconfig +4 -0
+	help
+	  This permits on-demand read mode of cachefiles.  In this mode, when
+	  cache miss, the cachefiles backend instead of netfs, is responsible
+	  for fetching data, e.g. through user daemon.

How about:

help
This permits userspace to enable the cachefiles on-demand read mode. In this mode, when a cache miss occurs, responsibility for fetching the data lies with the cachefiles backend instead of with the netfs and is delegated to userspace.

fs/cachefiles/daemon.c static int cachefiles_daemon_open(struct inode *inode, struct file *file) · line 147

Quoted 5 lines
fs/cachefiles/daemon.c +5 -0
+	/*
+	 * 1) Cache has been marked as dead state, and then 2) flush all
+	 * pending requests in @reqs xarray. The barrier inside set_bit()
+	 * will ensure that above two ops won't be reordered.
+	 */

What set_bit()? What "above two ops"? And that's not how barriers work; they provide a partial ordering relative to another pair of barriered ops.

Also, set_bit() can't be relied upon to imply a barrier - see Documentation/memory-barriers.txt.

fs/cachefiles/daemon.c static __poll_t cachefiles_daemon_poll(struct file *file, · line 346

Quoted 2 lines
fs/cachefiles/daemon.c +2 -0
+	if (IS_ENABLED(CONFIG_CACHEFILES_ONDEMAND) &&
+	    test_bit(CACHEFILES_ONDEMAND_MODE, &cache->flags)) {

It might be worth abstracting this into an inline function in internal.h:

	static inline bool cachefiles_in_ondemand_mode(cache)
	{
		return IS_ENABLED(CONFIG_CACHEFILES_ONDEMAND) &&
			test_bit(CACHEFILES_ONDEMAND_MODE, &cache->flags)
	}
Quoted 1 line

+#ifdef CONFIG_CACHEFILES_ONDEMAND

This looks like it ought to be superfluous, given the preceding test - though I can see why you need it:

fs/cachefiles/internal.h struct cachefiles_cache { · line 116

Quoted 5 lines
fs/cachefiles/internal.h +5 -0
+#ifdef CONFIG_CACHEFILES_ONDEMAND
+	struct xarray			reqs;		/* xarray of pending on-demand requests */
+	struct xarray			ondemand_ids;	/* xarray for ondemand_id allocation */
+	u32				ondemand_id_next;
+#endif

I'm tempted to say that you should just make them non-conditional. It's not like there's likely to be more than one or two cachefiles_cache structs on a system.

David

author
#5 lore raw

Re: [PATCH v9 02/21] cachefiles: notify user daemon when looking up cookie

Hi David,

Thanks for reviewing :)

Quoted 15 lines · David Howells

fs/cachefiles/Kconfig config CACHEFILES_ERROR_INJECTION · line 37

Quoted 4 lines · Jeffle Xu
fs/cachefiles/Kconfig +4 -0
+	help
+	  This permits on-demand read mode of cachefiles.  In this mode, when
+	  cache miss, the cachefiles backend instead of netfs, is responsible
+	  for fetching data, e.g. through user daemon.

How about:

help
This permits userspace to enable the cachefiles on-demand read mode. In this mode, when a cache miss occurs, responsibility for fetching the data lies with the cachefiles backend instead of with the netfs and is delegated to userspace.

Quoted 7 lines

fs/cachefiles/daemon.c static int cachefiles_daemon_open(struct inode *inode, struct file *file) · line 147

Quoted 5 lines
fs/cachefiles/daemon.c +5 -0
+	/*
+	 * 1) Cache has been marked as dead state, and then 2) flush all
+	 * pending requests in @reqs xarray. The barrier inside set_bit()
+	 * will ensure that above two ops won't be reordered.
+	 */

What set_bit()?

"set_bit(CACHEFILES_DEAD, &cache->flags);" in cachefiles_daemon_release()

Quoted 1 line

What "above two ops"?

The two operations I mentioned in the comment:
1) Cache has been marked as dead state, and then
2) flush all pending requests in @reqs xarray.

Quoted 7 lines

And that's not how barriers work; they

provide a partial ordering relative to another pair of barriered ops.

Also, set_bit() can't be relied upon to imply a barrier - see Documentation/memory-barriers.txt.

Yeah, it seems that set_bit() doesn't imply with a memory barrier, though the x86 implementation (arch/x86/boot/bitops.h) indeed implies a barrier, which may misleads me. Thanks for pointing it out. Then maybe a full barrier is needed here before flushing the @reqs xarray.

Quoted 11 lines

fs/cachefiles/daemon.c static __poll_t cachefiles_daemon_poll(struct file *file, · line 346

Quoted 3 lines
fs/cachefiles/daemon.c +2 -0
+	if (IS_ENABLED(CONFIG_CACHEFILES_ONDEMAND) &&
+	    test_bit(CACHEFILES_ONDEMAND_MODE, &cache->flags)) {

It might be worth abstracting this into an inline function in internal.h:

	static inline bool cachefiles_in_ondemand_mode(cache)
	{
		return IS_ENABLED(CONFIG_CACHEFILES_ONDEMAND) &&
			test_bit(CACHEFILES_ONDEMAND_MODE, &cache->flags)
	}

Okay, will be fixed in the next version.

Quoted 5 lines
Quoted 2 lines

+#ifdef CONFIG_CACHEFILES_ONDEMAND

This looks like it ought to be superfluous, given the preceding test - though I can see why you need it:

Sorry I can't see the context. But I guess you are referring to the snippet of cachefiles_daemon_poll()?

```
+	if (IS_ENABLED(CONFIG_CACHEFILES_ONDEMAND) &&
+	    test_bit(CACHEFILES_ONDEMAND_MODE, &cache->flags)) {
+#ifdef CONFIG_CACHEFILES_ONDEMAND
+		if (!xa_empty(&cache->reqs))
+			mask |= EPOLLIN;
```

Yes the implementation here is indeed not elegant enough. As you described below, if @reqs is defined non-conditionally in struct cachefiles_cache, then the superfluous magic here is not needed then.

Quoted 10 lines

fs/cachefiles/internal.h struct cachefiles_cache { · line 116

Quoted 6 lines
fs/cachefiles/internal.h +5 -0
+#ifdef CONFIG_CACHEFILES_ONDEMAND
+	struct xarray			reqs;		/* xarray of pending on-demand requests */
+	struct xarray			ondemand_ids;	/* xarray for ondemand_id allocation */
+	u32				ondemand_id_next;
+#endif

I'm tempted to say that you should just make them non-conditional. It's not like there's likely to be more than one or two cachefiles_cache structs on a system.

Okay, sounds reasonable.

Signature · 2 lines
Thanks,
Jeffle
maintainer
#6 lore raw

EMFILE/ENFILE mitigation needed in erofs?

In reply to Jeffle Xu

fs/cachefiles/ondemand.c line 166

Quoted 1 line · Jeffle Xu
fs/cachefiles/ondemand.c new file +1 -0
+	fd_install(fd, file);

Do you need to mitigate potential EMFILE/ENFILE problems? You're potentially trebling up the number of accounted systemwide fds: one for erofs itself, one anonfd per cache object file to communicate with the daemon and one in the daemon to talk to the server. Cachefiles has a fourth internally, but it's kept off the books - further, cachefiles closes them fairly quickly after a period of nonuse.

David

author
#7 lore raw

Re: EMFILE/ENFILE mitigation needed in erofs?

Quoted 11 lines · David Howells

fs/cachefiles/ondemand.c line 166

Quoted 1 line · Jeffle Xu
fs/cachefiles/ondemand.c new file +1 -0
+	fd_install(fd, file);

Do you need to mitigate potential EMFILE/ENFILE problems? You're potentially trebling up the number of accounted systemwide fds: one for erofs itself, one anonfd per cache object file to communicate with the daemon and one in the daemon to talk to the server. Cachefiles has a fourth internally, but it's kept off the books - further, cachefiles closes them fairly quickly after a period of nonuse.

Hi, thanks for pointing it out.

1. Actually in our using scenarios, one erofs filesystem is formed of several chunk-deduplicated blobs (which are really cached by Cachefiles), while each blob can contain many files of erofs. For example, one container image for node.js will correspond to ~20 blob files in total. Only these blob files are cached by Cachefiles. In densely employed environment, there could be hundreds of containers and thus thousands of backing files on one machine. That is, only tens of thousands of fds/files is needed in this case.

2. Our user daemon will configure rlimit-nofile to a reasonably large (e.g. 1 million) value, so that it won't fail when trying to allocate fds.

https://github.com/dragonflyoss/image-service/blob/master/src/bin/nydusd/main.rs#L152

3. Our user daemon will close the anonymous fd once the corresponding backing file has fully downloaded, to free the fd resources.

4. Even if fd/file allocation fails (in cachefiles_ondemand_get_fd()), the INIT request will fail, and thus the erofs mount will fail then. That is, it won't break the upper erofs in this case.

5. If later we find that the number of fds/files is indeed an issue, then we also plan to make the user daemon close some fds to spare some free resources. And then the Cachefiles kernel module needs to reallocate an anonymous fd for the backing file when cache miss. But it remains to be done later if it's really needed.

Signature · 2 lines
Thanks,
Jeffle
maintainer
#8 lore raw

Re: EMFILE/ENFILE mitigation needed in erofs?

Quoted 2 lines · JeffleXu

2. Our user daemon will configure rlimit-nofile to a reasonably large (e.g. 1 million) value, so that it won't fail when trying to allocate fds.

There's a system-wide limit also; simply increasing the rlimit won't override that.

David

#9 lore raw

Re: EMFILE/ENFILE mitigation needed in erofs?

Hi David,

Quoted 7 lines · David Howells
Quoted 2 lines · JeffleXu

2. Our user daemon will configure rlimit-nofile to a reasonably large (e.g. 1 million) value, so that it won't fail when trying to allocate fds.

There's a system-wide limit also; simply increasing the rlimit won't override that.

Yes, I suggest that we should add some words to document this to system administrators to take care of `/proc/sys/fs/file-max', but I think it's typically not a problem about our on-demand cases.

Since each cookie equals to an erofs device, so not too many erofs devices (much like docker layers) for one erofs images and they are all handled when mounting (which needs privilege permissions.)

And due to this, fscache dir can be easily backed up, restored, and transfered since they are really golden erofs image files.

Thanks,
Gao Xiang

Quoted 2 lines

David

author
#10 lore raw

PATCH v9 3/21 cachefiles: unbind cachefiles gracefully in on-demand mode

In reply to Jeffle Xu

Add a refcount to avoid the deadlock in on-demand read mode. The on-demand read mode will pin the corresponding cachefiles object for each anonymous fd. The cachefiles object is unpinned when the anonymous fd gets closed. When the user daemon exits and the fd of "/dev/cachefiles" device node gets closed, it will wait for all cahcefiles objects gets withdrawn. Then if there's any anonymous fd getting closed after the fd of the device node, the user daemon will hang forever, waiting for all objects getting withdrawn.

To fix this, add a refcount indicating if there's any object pinned by anonymous fds. The cachefiles cache gets unbound and withdrawn when the refcount decreased to 0. It won't change the behaviour of the original mode, in which case the cachefiles cache gets unbound and withdrawn as long as the fd of the device node gets closed. Besides, kref_get() is adequate whilst kref_get_unless_zero() is not needed here, since no more anonymous fd will be created when the .release() callback of the device node fd has already been called.

---
 fs/cachefiles/daemon.c   | 24 +++++++++++++++++++++---
 fs/cachefiles/internal.h |  3 +++
 fs/cachefiles/ondemand.c |  3 +++
 3 files changed, 27 insertions(+), 3 deletions(-)
fs/cachefiles/daemon.c +21 -3
index 69ca22aa6abf..2e946e4eb65a 100644
@@ -111,6 +111,7 @@ static int cachefiles_daemon_open(struct inode *inode, struct file *file)
 	INIT_LIST_HEAD(&cache->volumes);
 	INIT_LIST_HEAD(&cache->object_list);
 	spin_lock_init(&cache->object_list_lock);
+	kref_init(&cache->unbind_pincount);
 #ifdef CONFIG_CACHEFILES_ONDEMAND
 	xa_init_flags(&cache->reqs, XA_FLAGS_ALLOC);
 	xa_init_flags(&cache->ondemand_ids, XA_FLAGS_ALLOC1);
@@ -157,6 +158,25 @@ static void cachefiles_flush_reqs(struct cachefiles_cache *cache)
 }
 #endif
 
+static void cachefiles_release_cache(struct kref *kref)
+{
+	struct cachefiles_cache *cache;
+
+	cache = container_of(kref, struct cachefiles_cache, unbind_pincount);
+	cachefiles_daemon_unbind(cache);
+	kfree(cache);
+}
+
+void cachefiles_put_unbind_pincount(struct cachefiles_cache *cache)
+{
+	kref_put(&cache->unbind_pincount, cachefiles_release_cache);
+}
+
+void cachefiles_get_unbind_pincount(struct cachefiles_cache *cache)
+{
+	kref_get(&cache->unbind_pincount);
+}
+
 /*
  * Release a cache.
  */
@@ -173,14 +193,12 @@ static int cachefiles_daemon_release(struct inode *inode, struct file *file)
 #ifdef CONFIG_CACHEFILES_ONDEMAND
 	cachefiles_flush_reqs(cache);
 #endif
-	cachefiles_daemon_unbind(cache);
-
 	/* clean up the control file interface */
 	cache->cachefilesd = NULL;
 	file->private_data = NULL;
 	cachefiles_open = 0;
 
-	kfree(cache);
+	cachefiles_put_unbind_pincount(cache);
 
 	_leave("");
 	return 0;
fs/cachefiles/internal.h +3 -0
index 8ebe238af20b..9b83d8c82709 100644
@@ -109,6 +109,7 @@ struct cachefiles_cache {
 	char				*rootdirname;	/* name of cache root directory */
 	char				*secctx;	/* LSM security context */
 	char				*tag;		/* cache binding tag */
+	struct kref			unbind_pincount;/* refcount to do daemon unbind */
 #ifdef CONFIG_CACHEFILES_ONDEMAND
 	struct xarray			reqs;		/* xarray of pending on-demand requests */
 	struct xarray			ondemand_ids;	/* xarray for ondemand_id allocation */
@@ -167,6 +168,8 @@ extern int cachefiles_has_space(struct cachefiles_cache *cache,
  * daemon.c
  */
 extern const struct file_operations cachefiles_daemon_fops;
+extern void cachefiles_get_unbind_pincount(struct cachefiles_cache *cache);
+extern void cachefiles_put_unbind_pincount(struct cachefiles_cache *cache);
 
 /*
  * error_inject.c
fs/cachefiles/ondemand.c +3 -0
index 890cd3ecc2f0..eec883640efa 100644
@@ -14,6 +14,7 @@ static int cachefiles_ondemand_fd_release(struct inode *inode,
 	object->ondemand_id = CACHEFILES_ONDEMAND_ID_CLOSED;
 	xa_erase(&cache->ondemand_ids, object_id);
 	cachefiles_put_object(object, cachefiles_obj_put_ondemand_fd);
+	cachefiles_put_unbind_pincount(cache);
 	return 0;
 }
 
@@ -169,6 +170,8 @@ static int cachefiles_ondemand_get_fd(struct cachefiles_req *req)
 	load->fd = fd;
 	req->msg.object_id = object_id;
 	object->ondemand_id = object_id;
+
+	cachefiles_get_unbind_pincount(cache);
 	return 0;
 
 err_put_fd:
Signature · 1 line
2.27.0
maintainer
#11 lore raw

Re: [PATCH v9 03/21] cachefiles: unbind cachefiles gracefully in on-demand mode

fs/cachefiles/internal.h struct cachefiles_cache { · line 112

Quoted 1 line · Jeffle Xu
fs/cachefiles/internal.h +1 -0
+	struct kref			unbind_pincount;/* refcount to do daemon unbind */

Please use refcount_t or atomic_t, especially as this isn't the refcount for the structure.

fs/cachefiles/daemon.c static int cachefiles_daemon_release(struct inode *inode, struct file *file) · line 199

Quoted 6 lines
fs/cachefiles/daemon.c +0 -2
-	cachefiles_daemon_unbind(cache);
-
 	/* clean up the control file interface */
 	cache->cachefilesd = NULL;
 	file->private_data = NULL;
 	cachefiles_open = 0;

Please call cachefiles_daemon_unbind() before the cleanup.

David

author
#12 lore raw

Re: [PATCH v9 03/21] cachefiles: unbind cachefiles gracefully in on-demand mode

Quoted 6 lines · David Howells

fs/cachefiles/internal.h struct cachefiles_cache { · line 112

Quoted 1 line · Jeffle Xu
fs/cachefiles/internal.h +1 -0
+	struct kref			unbind_pincount;/* refcount to do daemon unbind */

Please use refcount_t or atomic_t, especially as this isn't the refcount for the structure.

Okay, will be done in the next version.

Quoted 9 lines

fs/cachefiles/daemon.c static int cachefiles_daemon_release(struct inode *inode, struct file *file) · line 199

Quoted 7 lines
fs/cachefiles/daemon.c +0 -2
-	cachefiles_daemon_unbind(cache);
-
 	/* clean up the control file interface */
 	cache->cachefilesd = NULL;
 	file->private_data = NULL;
 	cachefiles_open = 0;

Please call cachefiles_daemon_unbind() before the cleanup.

Since the cachefiles_struct struct will be freed once the pincount is decreased to 0, "cache->cachefilesd = NULL;" needs to be done before decreasing the pincount. BTW, "cachefiles_open = 0;" indeed should be done only when pincount has been decreased to 0.

Signature · 2 lines
Thanks,
Jeffle
author
#13 lore raw

PATCH v9 4/21 cachefiles: notify user daemon when withdrawing cookie

In reply to Jeffle Xu

Notify the user daemon that cookie is going to be withdrawn, providing a hint that the associated anonymous fd can be closed.

Be noted that this is only a hint. The user daemon may close the associated anonymous fd when receiving the CLOSE request, then it will receive another anonymous fd when the cookie gets looked up. Or it may ignore the CLOSE request, and keep writing data through the anonymous fd. However the next time the cookie gets looked up, the user daemon will still receive another new anonymous fd.

---
 fs/cachefiles/interface.c       |  2 ++
 fs/cachefiles/internal.h        |  5 +++++
 fs/cachefiles/ondemand.c        | 38 +++++++++++++++++++++++++++++++++
 include/uapi/linux/cachefiles.h |  1 +
 4 files changed, 46 insertions(+)
fs/cachefiles/interface.c +2 -0
index ae93cee9d25d..a69073a1d3f0 100644
@@ -362,6 +362,8 @@ static void cachefiles_withdraw_cookie(struct fscache_cookie *cookie)
 		spin_unlock(&cache->object_list_lock);
 	}
 
+	cachefiles_ondemand_clean_object(object);
+
 	if (object->file) {
 		cachefiles_begin_secure(cache, &saved_cred);
 		cachefiles_clean_up_object(object, cache);
fs/cachefiles/internal.h +5 -0
index 9b83d8c82709..15332eae43c0 100644
@@ -286,6 +286,7 @@ extern int cachefiles_ondemand_copen(struct cachefiles_cache *cache,
 				     char *args);
 
 extern int cachefiles_ondemand_init_object(struct cachefiles_object *object);
+extern void cachefiles_ondemand_clean_object(struct cachefiles_object *object);
 
 #else
 static inline ssize_t cachefiles_ondemand_daemon_read(struct cachefiles_cache *cache,
@@ -298,6 +299,10 @@ static inline int cachefiles_ondemand_init_object(struct cachefiles_object *obje
 {
 	return 0;
 }
+
+static inline void cachefiles_ondemand_clean_object(struct cachefiles_object *object)
+{
+}
 #endif
 
 /*
fs/cachefiles/ondemand.c +38 -0
index eec883640efa..7ce383536f27 100644
@@ -229,6 +229,12 @@ ssize_t cachefiles_ondemand_daemon_read(struct cachefiles_cache *cache,
 		goto err_put_fd;
 	}
 
+	/* CLOSE request has no reply */
+	if (msg->opcode == CACHEFILES_OP_CLOSE) {
+		xa_erase(&cache->reqs, id);
+		complete(&req->done);
+	}
+
 	return n;
 
 err_put_fd:
@@ -293,6 +299,13 @@ static int cachefiles_ondemand_send_req(struct cachefiles_object *object,
 			goto out;
 		}
 
+		if (opcode != CACHEFILES_OP_OPEN && object->ondemand_id <= 0) {
+			WARN_ON_ONCE(object->ondemand_id == 0);
+			xas_unlock(&xas);
+			ret = -EIO;
+			goto out;
+		}
+
 		xas.xa_index = 0;
 		xas_find_marked(&xas, UINT_MAX, XA_FREE_MARK);
 		if (xas.xa_node == XAS_RESTART)
@@ -349,6 +362,25 @@ static int cachefiles_ondemand_init_open_req(struct cachefiles_req *req,
 	return 0;
 }
 
+static int cachefiles_ondemand_init_close_req(struct cachefiles_req *req,
+					      void *private)
+{
+	struct cachefiles_object *object = req->object;
+	int object_id = object->ondemand_id;
+
+	/*
+	 * It's possiblie that object id is still 0 if the cookie looking up
+	 * phase failed before OPEN request has ever been sent. Also avoid
+	 * sending CLOSE request for CACHEFILES_ONDEMAND_ID_CLOSED, which means
+	 * anon_fd has already been closed.
+	 */
+	if (object_id <= 0)
+		return -ENOENT;
+
+	req->msg.object_id = object_id;
+	return 0;
+}
+
 int cachefiles_ondemand_init_object(struct cachefiles_object *object)
 {
 	struct fscache_cookie *cookie = object->cookie;
@@ -372,3 +404,9 @@ int cachefiles_ondemand_init_object(struct cachefiles_object *object)
 	return cachefiles_ondemand_send_req(object, CACHEFILES_OP_OPEN,
 			data_len, cachefiles_ondemand_init_open_req, NULL);
 }
+
+void cachefiles_ondemand_clean_object(struct cachefiles_object *object)
+{
+	cachefiles_ondemand_send_req(object, CACHEFILES_OP_CLOSE, 0,
+			cachefiles_ondemand_init_close_req, NULL);
+}
include/uapi/linux/cachefiles.h +1 -0
index 521f2fe4fe9c..37a0071037c8 100644
@@ -12,6 +12,7 @@
 
 enum cachefiles_opcode {
 	CACHEFILES_OP_OPEN,
+	CACHEFILES_OP_CLOSE,
 };
 
 /*
Signature · 1 line
2.27.0
maintainer Acked-by on v9 4/21
#14 lore raw

Re: [PATCH v9 04/21] cachefiles: notify user daemon when withdrawing cookie

fs/cachefiles/ondemand.c static int cachefiles_ondemand_init_open_req(struct cachefiles_req *req, · line 372

Quoted 1 line · Jeffle Xu
fs/cachefiles/ondemand.c +1 -0
+	 * It's possiblie that object id is still 0 if the cookie looking up

possiblie -> possible

Otherwise:

author
#15 lore raw

Re: [PATCH v9 04/21] cachefiles: notify user daemon when withdrawing cookie

Quoted 5 lines · David Howells

fs/cachefiles/ondemand.c static int cachefiles_ondemand_init_open_req(struct cachefiles_req *req, · line 372

Quoted 1 line · Jeffle Xu
fs/cachefiles/ondemand.c +1 -0
+	 * It's possiblie that object id is still 0 if the cookie looking up

possiblie -> possible

Thanks.

Quoted 4 lines

Otherwise:

Signature · 2 lines
Thanks,
Jeffle
author
#16 lore raw

PATCH v9 5/21 cachefiles: implement on-demand read

In reply to Jeffle Xu

Implement the data plane of on-demand read mode.

The early implementation [1] place the entry to cachefiles_ondemand_read() in fscache_read(). However, fscache_read() can only detect if the requested file range is fully cache miss, whilst we need to notify the user daemon as long as there's a hole inside the requested file range.

Thus the entry is now placed in cachefiles_prepare_read(). When working in on-demand read mode, once a hole detected, the read routine will send a READ request to the user daemon. The user daemon needs to fetch the data and write it to the cache file. After sending the READ request, the read routine will hang there, until the READ request is handled by the user daemon. Then it will retry to read from the same file range. If no progress encountered, the read routine will fail then.

A new NETFS_SREQ_ONDEMAND flag is introduced to indicate that on-demand read should be done when a cache miss encountered.

[1] https://lore.kernel.org/all/[email protected]/ #v8

---
 fs/cachefiles/internal.h        |  9 ++++
 fs/cachefiles/io.c              | 26 ++++++++++-
 fs/cachefiles/ondemand.c        | 77 +++++++++++++++++++++++++++++++++
 include/linux/netfs.h           |  2 +
 include/uapi/linux/cachefiles.h | 17 ++++++++
 5 files changed, 129 insertions(+), 2 deletions(-)
fs/cachefiles/internal.h +9 -0
index 15332eae43c0..3025556ff7d4 100644
@@ -288,6 +288,9 @@ extern int cachefiles_ondemand_copen(struct cachefiles_cache *cache,
 extern int cachefiles_ondemand_init_object(struct cachefiles_object *object);
 extern void cachefiles_ondemand_clean_object(struct cachefiles_object *object);
 
+extern int cachefiles_ondemand_read(struct cachefiles_object *object,
+				    loff_t pos, size_t len);
+
 #else
 static inline ssize_t cachefiles_ondemand_daemon_read(struct cachefiles_cache *cache,
 					char __user *_buffer, size_t buflen)
@@ -303,6 +306,12 @@ static inline int cachefiles_ondemand_init_object(struct cachefiles_object *obje
 static inline void cachefiles_ondemand_clean_object(struct cachefiles_object *object)
 {
 }
+
+static inline int cachefiles_ondemand_read(struct cachefiles_object *object,
+					   loff_t pos, size_t len)
+{
+	return -EOPNOTSUPP;
+}
 #endif
 
 /*
fs/cachefiles/io.c +24 -2
index 50a14e8f0aac..ccf77a969653 100644
@@ -95,6 +95,7 @@ static int cachefiles_read(struct netfs_cache_resources *cres,
 	       file, file_inode(file)->i_ino, start_pos, len,
 	       i_size_read(file_inode(file)));
 
+retry:
 	/* If the caller asked us to seek for data before doing the read, then
 	 * we should do that now.  If we find a gap, we fill it with zeros.
 	 */
@@ -119,6 +120,16 @@ static int cachefiles_read(struct netfs_cache_resources *cres,
 			if (read_hole == NETFS_READ_HOLE_FAIL)
 				goto presubmission_error;
 
+			if (read_hole == NETFS_READ_HOLE_ONDEMAND) {
+				ret = cachefiles_ondemand_read(object, off, len);
+				if (ret)
+					goto presubmission_error;
+
+				/* fail the read if no progress achieved */
+				read_hole = NETFS_READ_HOLE_FAIL;
+				goto retry;
+			}
+
 			iov_iter_zero(len, iter);
 			skipped = len;
 			ret = 0;
@@ -403,6 +414,7 @@ static enum netfs_io_source cachefiles_prepare_read(struct netfs_io_subrequest *
 	enum netfs_io_source ret = NETFS_DOWNLOAD_FROM_SERVER;
 	loff_t off, to;
 	ino_t ino = file ? file_inode(file)->i_ino : 0;
+	int rc;
 
 	_enter("%zx @%llx/%llx", subreq->len, subreq->start, i_size);
 
@@ -415,7 +427,8 @@ static enum netfs_io_source cachefiles_prepare_read(struct netfs_io_subrequest *
 	if (test_bit(FSCACHE_COOKIE_NO_DATA_TO_READ, &cookie->flags)) {
 		__set_bit(NETFS_SREQ_COPY_TO_CACHE, &subreq->flags);
 		why = cachefiles_trace_read_no_data;
-		goto out_no_object;
+		if (!test_bit(NETFS_SREQ_ONDEMAND, &subreq->flags))
+			goto out_no_object;
 	}
 
 	/* The object and the file may be being created in the background. */
@@ -432,7 +445,7 @@ static enum netfs_io_source cachefiles_prepare_read(struct netfs_io_subrequest *
 	object = cachefiles_cres_object(cres);
 	cache = object->volume->cache;
 	cachefiles_begin_secure(cache, &saved_cred);
-
+retry:
 	off = cachefiles_inject_read_error();
 	if (off == 0)
 		off = vfs_llseek(file, subreq->start, SEEK_DATA);
@@ -483,6 +496,15 @@ static enum netfs_io_source cachefiles_prepare_read(struct netfs_io_subrequest *
 
 download_and_store:
 	__set_bit(NETFS_SREQ_COPY_TO_CACHE, &subreq->flags);
+	if (test_bit(NETFS_SREQ_ONDEMAND, &subreq->flags)) {
+		rc = cachefiles_ondemand_read(object, subreq->start,
+					      subreq->len);
+		if (!rc) {
+			__clear_bit(NETFS_SREQ_ONDEMAND, &subreq->flags);
+			goto retry;
+		}
+		ret = NETFS_INVALID_READ;
+	}
 out:
 	cachefiles_end_secure(cache, saved_cred);
 out_no_object:
fs/cachefiles/ondemand.c +77 -0

105 lines not on this page · raw message

include/linux/netfs.h +2 -0
index c7bf1eaf51d5..02dbde48bb68 100644
@@ -159,6 +159,7 @@ struct netfs_io_subrequest {
 #define NETFS_SREQ_SHORT_IO		2	/* Set if the I/O was short */
 #define NETFS_SREQ_SEEK_DATA_READ	3	/* Set if ->read() should SEEK_DATA first */
 #define NETFS_SREQ_NO_PROGRESS		4	/* Set if we didn't manage to read any data */
+#define NETFS_SREQ_ONDEMAND		5	/* Set if it's from on-demand read mode */
 };
 
 enum netfs_io_origin {
@@ -222,6 +223,7 @@ enum netfs_read_from_hole {
 	NETFS_READ_HOLE_IGNORE,
 	NETFS_READ_HOLE_CLEAR,
 	NETFS_READ_HOLE_FAIL,
+	NETFS_READ_HOLE_ONDEMAND,
 };
 
 /*
include/uapi/linux/cachefiles.h +17 -0
index 37a0071037c8..028fbf15e02b 100644
@@ -3,6 +3,7 @@
 #define _LINUX_CACHEFILES_H
 
 #include <linux/types.h>
+#include <linux/ioctl.h>
 
 /*
  * Fscache ensures that the maximum length of cookie key is 255. The volume key
@@ -13,6 +14,7 @@
 enum cachefiles_opcode {
 	CACHEFILES_OP_OPEN,
 	CACHEFILES_OP_CLOSE,
+	CACHEFILES_OP_READ,
 };
 
 /*
@@ -48,4 +50,19 @@ struct cachefiles_open {
 	__u8  data[];
 };
 
+/*
+ * @off		indicates the starting offset of the requested file range
+ * @len		indicates the length of the requested file range
+ */
+struct cachefiles_read {
+	__u64 off;
+	__u64 len;
+};
+
+/*
+ * Reply for READ request (Completion for READ)
+ * arg for CACHEFILES_IOC_CREAD ioctl is the @id field of READ request.
+ */
+#define CACHEFILES_IOC_CREAD	_IOW(0x98, 1, int)
+
 #endif
Signature · 1 line
2.27.0
maintainer Acked-by on v9 5/21
#17 lore raw

Re: [PATCH v9 05/21] cachefiles: implement on-demand read

Quoted 2 lines · Jeffle Xu

A new NETFS_SREQ_ONDEMAND flag is introduced to indicate that on-demand read should be done when a cache miss encountered.

That may conflict with changes I'm making - but it's just a matter of flag renumbering.

Quoted 1 line

+#define CACHEFILES_IOC_CREAD _IOW(0x98, 1, int)

I wonder if CACHEFILES_IOC_READ_COMPLETE would be a better name, but apart from that:

author
#18 lore raw

Re: [PATCH v9 05/21] cachefiles: implement on-demand read

Quoted 11 lines · David Howells
Quoted 2 lines · Jeffle Xu

A new NETFS_SREQ_ONDEMAND flag is introduced to indicate that on-demand read should be done when a cache miss encountered.

That may conflict with changes I'm making - but it's just a matter of flag renumbering.

Quoted 1 line

+#define CACHEFILES_IOC_CREAD _IOW(0x98, 1, int)

I wonder if CACHEFILES_IOC_READ_COMPLETE would be a better name,

Okay, it sounds more readable. Thanks.

but apart

Quoted 3 lines

from that:

Signature · 2 lines
Thanks,
Jeffle
author
#19 lore raw

PATCH v9 6/21 cachefiles: enable on-demand read mode

In reply to Jeffle Xu

Enable on-demand read mode by adding an optional parameter to the "bind" command.

On-demand mode will be turned on when this parameter is "ondemand", i.e. "bind ondemand". Otherwise cachefiles will work in the original mode.

---
 fs/cachefiles/daemon.c | 13 ++++++++-----
 fs/cachefiles/io.c     | 11 -----------
 2 files changed, 8 insertions(+), 16 deletions(-)
fs/cachefiles/daemon.c +8 -5
index 2e946e4eb65a..c8bde21ace6a 100644
@@ -758,11 +758,6 @@ static int cachefiles_daemon_bind(struct cachefiles_cache *cache, char *args)
 	    cache->brun_percent  >= 100)
 		return -ERANGE;
 
-	if (*args) {
-		pr_err("'bind' command doesn't take an argument\n");
-		return -EINVAL;
-	}
-
 	if (!cache->rootdirname) {
 		pr_err("No cache directory specified\n");
 		return -EINVAL;
@@ -774,6 +769,14 @@ static int cachefiles_daemon_bind(struct cachefiles_cache *cache, char *args)
 		return -EBUSY;
 	}
 
+	if (IS_ENABLED(CONFIG_CACHEFILES_ONDEMAND) &&
+	    !strcmp(args, "ondemand")) {
+		set_bit(CACHEFILES_ONDEMAND_MODE, &cache->flags);
+	} else if (*args) {
+		pr_err("'bind' command doesn't take an argument\n");
+		return -EINVAL;
+	}
+
 	/* Make sure we have copies of the tag string */
 	if (!cache->tag) {
 		/*
fs/cachefiles/io.c +0 -11
index ccf77a969653..000a28f46e59 100644
@@ -95,7 +95,6 @@ static int cachefiles_read(struct netfs_cache_resources *cres,
 	       file, file_inode(file)->i_ino, start_pos, len,
 	       i_size_read(file_inode(file)));
 
-retry:
 	/* If the caller asked us to seek for data before doing the read, then
 	 * we should do that now.  If we find a gap, we fill it with zeros.
 	 */
@@ -120,16 +119,6 @@ static int cachefiles_read(struct netfs_cache_resources *cres,
 			if (read_hole == NETFS_READ_HOLE_FAIL)
 				goto presubmission_error;
 
-			if (read_hole == NETFS_READ_HOLE_ONDEMAND) {
-				ret = cachefiles_ondemand_read(object, off, len);
-				if (ret)
-					goto presubmission_error;
-
-				/* fail the read if no progress achieved */
-				read_hole = NETFS_READ_HOLE_FAIL;
-				goto retry;
-			}
-
 			iov_iter_zero(len, iter);
 			skipped = len;
 			ret = 0;
Signature · 1 line
2.27.0
maintainer
#20 lore raw

Re: [PATCH v9 06/21] cachefiles: enable on-demand read mode

fs/cachefiles/daemon.c static int cachefiles_daemon_bind(struct cachefiles_cache *cache, char *args) · line 776

Quoted 5 lines · Jeffle Xu
fs/cachefiles/daemon.c +5 -0
+	if (IS_ENABLED(CONFIG_CACHEFILES_ONDEMAND) &&
+	    !strcmp(args, "ondemand")) {
+		set_bit(CACHEFILES_ONDEMAND_MODE, &cache->flags);
+	} else if (*args) {
+		pr_err("'bind' command doesn't take an argument\n");

The error message isn't true if CONFIG_CACHEFILES_ONDEMAND=y. It would be better to say "Invalid argument to the 'bind' command".

Quoted 8 lines
fs/cachefiles/io.c +0 -1
-retry:
 	/* If the caller asked us to seek for data before doing the read, then
 	 * we should do that now.  If we find a gap, we fill it with zeros.
 	 */
@@ -120,16 +119,6 @@ static int cachefiles_read(struct netfs_cache_resources *cres,
 			if (read_hole == NETFS_READ_HOLE_FAIL)
 				goto presubmission_error;
 

fs/cachefiles/io.c static int cachefiles_read(struct netfs_cache_resources *cres, · line 132

Quoted 10 lines
fs/cachefiles/io.c +0 -10
-			if (read_hole == NETFS_READ_HOLE_ONDEMAND) {
-				ret = cachefiles_ondemand_read(object, off, len);
-				if (ret)
-					goto presubmission_error;
-
-				/* fail the read if no progress achieved */
-				read_hole = NETFS_READ_HOLE_FAIL;
-				goto retry;
-			}
-

Unexplained deletion of newly added code.

David

author
#21 lore raw

Re: [PATCH v9 06/21] cachefiles: enable on-demand read mode

Quoted 10 lines · David Howells

fs/cachefiles/daemon.c static int cachefiles_daemon_bind(struct cachefiles_cache *cache, char *args) · line 776

Quoted 5 lines · Jeffle Xu
fs/cachefiles/daemon.c +5 -0
+	if (IS_ENABLED(CONFIG_CACHEFILES_ONDEMAND) &&
+	    !strcmp(args, "ondemand")) {
+		set_bit(CACHEFILES_ONDEMAND_MODE, &cache->flags);
+	} else if (*args) {
+		pr_err("'bind' command doesn't take an argument\n");

The error message isn't true if CONFIG_CACHEFILES_ONDEMAND=y. It would be better to say "Invalid argument to the 'bind' command".

Right. Or users may gets confused then. Will be fixed in the next version.

Quoted 9 lines
Quoted 9 lines
fs/cachefiles/io.c +0 -1
-retry:
 	/* If the caller asked us to seek for data before doing the read, then
 	 * we should do that now.  If we find a gap, we fill it with zeros.
 	 */
@@ -120,16 +119,6 @@ static int cachefiles_read(struct netfs_cache_resources *cres,
 			if (read_hole == NETFS_READ_HOLE_FAIL)
 				goto presubmission_error;
 
Quoted 11 lines

fs/cachefiles/io.c static int cachefiles_read(struct netfs_cache_resources *cres, · line 132

Quoted 10 lines
fs/cachefiles/io.c +0 -10
-			if (read_hole == NETFS_READ_HOLE_ONDEMAND) {
-				ret = cachefiles_ondemand_read(object, off, len);
-				if (ret)
-					goto presubmission_error;
-
-				/* fail the read if no progress achieved */
-				read_hole = NETFS_READ_HOLE_FAIL;
-				goto retry;
-			}
-

Sorry, it's my mistake when doing "git rebase". The previous version (v8) actually calls cachefiles_ondemand_read() in cachefiles_read(). However as explained in the commit message of patch 5 ("cachefiles:
implement on-demand read"), fscache_read() can only detect if the requested file range is fully cache miss, whilst it can't detect if it is partial cache miss, i.e. there's a hole inside the requested file range.

Thus in this patchset (v9), we move the entry of calling cachefiles_ondemand_read() from cachefiles_read() to cachefiles_prepare_read(). The above "deletion of newly added code" is actually reverting the previous change to cachefiles_read(). It was mistakenly merged to this patch when I was doing "git rebase"... Actually it should be merged to patch 5 ("cachefiles: implement on-demand read"), which initially introduce the change to cachefiles_read().

Apologize for the careless mistake...

Signature · 2 lines
Thanks,
Jeffle
author
#22 lore raw

PATCH v9 7/21 cachefiles: add tracepoints for on-demand read mode

In reply to Jeffle Xu

Add tracepoints for on-demand read mode. Currently following tracepoints are added:

OPEN request / COPEN reply
CLOSE request
READ request / CREAD reply
write through anonymous fd
release of anonymous fd

---
 fs/cachefiles/ondemand.c          |   7 ++
 include/trace/events/cachefiles.h | 174 ++++++++++++++++++++++++++++++
 2 files changed, 181 insertions(+)
fs/cachefiles/ondemand.c +7 -0
index 10bdac26ce23..3be65b825037 100644
@@ -30,6 +30,7 @@ static int cachefiles_ondemand_fd_release(struct inode *inode,
 	xa_unlock(&cache->reqs);
 
 	xa_erase(&cache->ondemand_ids, object_id);
+	trace_cachefiles_ondemand_fd_release(object, object_id);
 	cachefiles_put_object(object, cachefiles_obj_put_ondemand_fd);
 	cachefiles_put_unbind_pincount(cache);
 	return 0;
@@ -55,6 +56,7 @@ static ssize_t cachefiles_ondemand_fd_write_iter(struct kiocb *kiocb,
 	if (ret < 0)
 		return ret;
 
+	trace_cachefiles_ondemand_fd_write(object, file_inode(file), pos, len);
 	ret = __cachefiles_write(object, file, pos, iter, NULL, NULL);
 	if (!ret)
 		ret = len;
@@ -93,6 +95,7 @@ static long cachefiles_ondemand_fd_ioctl(struct file *filp, unsigned int ioctl,
 	if (!req)
 		return -EINVAL;
 
+	trace_cachefiles_ondemand_cread(object, id);
 	complete(&req->done);
 	return 0;
 }
@@ -166,6 +169,7 @@ int cachefiles_ondemand_copen(struct cachefiles_cache *cache, char *args)
 		clear_bit(FSCACHE_COOKIE_NO_DATA_TO_READ, &cookie->flags);
 	else
 		set_bit(FSCACHE_COOKIE_NO_DATA_TO_READ, &cookie->flags);
+	trace_cachefiles_ondemand_copen(req->object, id, size);
 
 out:
 	complete(&req->done);
@@ -213,6 +217,7 @@ static int cachefiles_ondemand_get_fd(struct cachefiles_req *req)
 	object->ondemand_id = object_id;
 
 	cachefiles_get_unbind_pincount(cache);
+	trace_cachefiles_ondemand_open(object, &req->msg, load);
 	return 0;
 
 err_put_fd:
@@ -419,6 +424,7 @@ static int cachefiles_ondemand_init_close_req(struct cachefiles_req *req,
 		return -ENOENT;
 
 	req->msg.object_id = object_id;
+	trace_cachefiles_ondemand_close(object, &req->msg);
 	return 0;
 }
 
@@ -445,6 +451,7 @@ static int cachefiles_ondemand_init_read_req(struct cachefiles_req *req,
 	req->msg.object_id = object_id;
 	load->off = read_ctx->off;
 	load->len = read_ctx->len;
+	trace_cachefiles_ondemand_read(object, &req->msg, load);
 	return 0;
 }
 
include/trace/events/cachefiles.h +174 -0

180 lines not on this page · raw message

Signature · 1 line
2.27.0
maintainer Acked-by on v9 7/21
#23 lore raw

Re: [PATCH v9 07/21] cachefiles: add tracepoints for on-demand read mode

Quoted 10 lines · Jeffle Xu

Add tracepoints for on-demand read mode. Currently following tracepoints are added:

OPEN request / COPEN reply
CLOSE request
READ request / CREAD reply
write through anonymous fd
release of anonymous fd

author
#24 lore raw

PATCH v9 8/21 cachefiles: document on-demand read mode

In reply to Jeffle Xu

Document new user interface introduced by on-demand read mode.

---
 .../filesystems/caching/cachefiles.rst        | 170 ++++++++++++++++++
 1 file changed, 170 insertions(+)
Documentation/filesystems/caching/cachefiles.rst +170 -0
index 8bf396b76359..c10a16957141 100644
@@ -28,6 +28,7 @@ Cache on Already Mounted Filesystem
 
  (*) Debugging.
 
+ (*) On-demand Read.
 
 
 Overview
@@ -482,3 +483,172 @@ the control file.  For example::
 	echo $((1|4|8)) >/sys/module/cachefiles/parameters/debug
 
 will turn on all function entry debugging.
+
+
+On-demand Read
+==============
+
+When working in its original mode, cachefiles mainly serves as a local cache
+for a remote networking fs - while in on-demand read mode, cachefiles can boost
+the scenario where on-demand read semantics is needed, e.g. container image
+distribution.
+
+The essential difference between these two modes is that, in original mode,
+when a cache miss occurs, the netfs will fetch the data from the remote server
+and then write it to the cache file.  With on-demand read mode, however,
+fetching the data and writing it into the cache is delegated to a user daemon.
+
+``CONFIG_CACHEFILES_ONDEMAND`` should be enabled to support on-demand read mode.
+
+
+Protocol Communication
+----------------------
+
+The on-demand read mode relies on a simple protocol used for communication
+between kernel and user daemon. The protocol can be modeled as::
+
+	kernel --[request]--> user daemon --[reply]--> kernel
+
+The cachefiles kernel module will send requests to the user daemon when needed.
+The user daemon needs to poll on the devnode ('/dev/cachefiles') to check if
+there's a pending request to be processed.  A POLLIN event will be returned
+when there's a pending request.
+
+The user daemon then reads the devnode to fetch a request and process it
+accordingly.  It is worth noting that each read only gets one request. When
+finished processing the request, the user daemon needs to write the reply to
+the devnode.
+
+Each request starts with a message header of the form::
+
+	struct cachefiles_msg {
+		__u32 msg_id;
+		__u32 opcode;
+		__u32 len;
+		__u32 object_id;
+		__u8  data[];
+	};
+
+	where:
+
+	* ``msg_id`` is a unique ID identifying this request among all pending
+	  requests.
+
+	* ``opcode`` indicates the type of this request.
+
+	* ``object_id`` is a unique ID identifying the cache file operated on.
+
+	* ``data`` indicates the payload of this request.
+
+	* ``len`` indicates the whole length of this request, including the
+	  header and following type-specific payload.
+
+
+Turn on On-demand Mode
+----------------------
+
+An optional parameter is added to the "bind" command::
+
+	bind [ondemand]
+
+When the "bind" command takes without argument, it defaults to the original
+mode.  When the "bind" command is given the "ondemand" argument, i.e.
+"bind ondemand", on-demand read mode will be enabled.
+
+
+The OPEN Request
+----------------
+
+When the netfs opens a cache file for the first time, a request with the
+CACHEFILES_OP_OPEN opcode, a.k.a an OPEN request will be sent to the user
+daemon.  The payload format is of the form::
+
+	struct cachefiles_open {
+		__u32 volume_key_size;
+		__u32 cookie_key_size;
+		__u32 fd;
+		__u32 flags;
+		__u8  data[];
+	};
+
+	where:
+
+	* ``data`` contains the volume_key followed directly by the cookie_key.
+	  The volume key is a NUL-terminated string; the cookie key is binary
+	  data.
+
+	* ``volume_key_size`` indicates the size of the volume key in bytes.
+
+	* ``cookie_key_size`` indicates the size of the cookie key in bytes.
+
+	* ``fd`` indicates an anonymous fd referring to the cache file, through
+	  which the user daemon can perform write/llseek file operations on the
+	  cache file.
+
+
+The user daemon is able to distinguish the requested cache file with the given
+(volume_key, cookie_key) pair. Each cache file has a unique object_id, while it
+may have multiple anonymous fds. The user daemon may duplicate anonymous fds
+from the initial anonymous fd indicated by the @fd field through dup(). Thus
+each object_id can be mapped to multiple anonymous fds, while the usr daemon
+itself needs to maintain the mapping.
+
+With the given anonymous fd, the user daemon can fetch data and write it to the
+cache file in the background, even when kernel has not triggered a cache miss
+yet.
+
+The user daemon should complete the READ request by issuing a "copen" (complete
+open) command on the devnode::
+
+	copen <msg_id>,<cache_size>
+
+	* ``msg_id`` must match the msg_id field of the previous OPEN request.
+
+	* When >= 0, ``cache_size`` indicates the size of the cache file;
+	  when < 0, ``cache_size`` indicates the error code ecountered by the
+	  user daemon.
+
+
+The CLOSE Request
+-----------------
+
+When a cookie withdrawn, a CLOSE request (opcode CACHEFILES_OP_CLOSE) will be
+sent to the user daemon. It will notify the user daemon to close all anonymous
+fds associated with the given object_id.  The CLOSE request has no extea
+payload.
+
+
+The READ Request
+----------------
+
+When on-demand read mode is turned on, and a cache miss encountered, the kernel
+will send a READ request (opcode CACHEFILES_OP_READ) to the user daemon. This
+will tell the user daemon to fetch data of the requested file range. The payload
+is of the form::
+
+	struct cachefiles_read {
+		__u64 off;
+		__u64 len;
+	};
+
+	where:
+
+	* ``off`` indicates the starting offset of the requested file range.
+
+	* ``len`` indicates the length of the requested file range.
+
+
+When receiving a READ request, the user daemon needs to fetch the data of the
+requested file range, and then write it to the cache file identified by
+object_id.
+
+To finish processing the READ request, the user daemon should reply with the
+CACHEFILES_IOC_CREAD ioctl on one of the anonymous fds associated with the given
+object_id in the READ request.  The ioctl is of the form::
+
+	ioctl(fd, CACHEFILES_IOC_CREAD, msg_id);
+
+	* ``fd`` is one of the anonymous fds associated with the given object_id
+	  in the READ request.
+
+	* ``msg_id`` must match the msg_id field of the previous READ request.
Signature · 1 line
2.27.0
maintainer
#25 lore raw

Re: [PATCH v9 08/21] cachefiles: document on-demand read mode

Quoted 1 line · Jeffle Xu

+When working in its original mode, cachefiles mainly

I'd delete "mainly" there.

Review

Quoted 3 lines

serves as a local cache

(quoted diff) +2 -0
+for a remote networking fs - while in on-demand read mode, cachefiles can boost
+the scenario where on-demand read semantics is

is -> are.

Documentation/filesystems/caching/cachefiles.rst the control file. For example:: · line 499

Quoted 4 lines
Documentation/filesystems/caching/cachefiles.rst +4 -0
+The essential difference between these two modes is that, in original mode,
+when a cache miss occurs, the netfs will fetch the data from the remote server
+and then write it to the cache file.  With on-demand read mode, however,
+fetching the data and writing it into the cache is delegated to a user daemon.

The starting sentence seems off. How about:

The essential difference between these two modes is seen when a cache miss
occurs: In the original mode, the netfs will fetch the data from the remote server and then write it to the cache file; in on-demand read mode, fetching data and writing it into the cache is delegated to a user daemon.

Documentation/filesystems/caching/cachefiles.rst the control file. For example::

Quoted 4 lines
Documentation/filesystems/caching/cachefiles.rst +4 -0
+Protocol Communication
+----------------------
+
+The on-demand read mode relies on

relies on -> uses

Quoted 1 line

a simple protocol used

Delete "used".

Documentation/filesystems/caching/cachefiles.rst the control file. For example::

Quoted 7 lines

for communication

Documentation/filesystems/caching/cachefiles.rst +6 -0
+between kernel and user daemon. The protocol can be modeled as::
+
+	kernel --[request]--> user daemon --[reply]--> kernel
+
+The cachefiles kernel module will send requests to the user daemon when needed.
+The user daemon needs to

needs to -> should

Quoted 1 line

poll on

poll on -> poll

Documentation/filesystems/caching/cachefiles.rst the control file. For example::

Quoted 6 lines

the devnode ('/dev/cachefiles') to check if

Documentation/filesystems/caching/cachefiles.rst +5 -0
+there's a pending request to be processed.  A POLLIN event will be returned
+when there's a pending request.
+
+The user daemon then reads the devnode to fetch a request and process it
+accordingly.

Reading the devnode doesn't process the request, so I think something like:

"... and process it accordingly" -> "... that it can then process."

or:

"... and process it accordingly" -> "... to process."

Quoted 1 line

It is worth noting

"It should be noted"

Quoted 1 line

that each read only gets one request. When

... it has ...

Quoted 1 line

+finished processing the request, the user daemon needs to

needs to -> should write

Quoted 19 lines

write the reply to

Documentation/filesystems/caching/cachefiles.rst +18 -0
+the devnode.
+
+Each request starts with a message header of the form::
+
+	struct cachefiles_msg {
+		__u32 msg_id;
+		__u32 opcode;
+		__u32 len;
+		__u32 object_id;
+		__u8  data[];
+	};
+
+	where:
+
+	* ``msg_id`` is a unique ID identifying this request among all pending
+	  requests.
+
+	* ``opcode`` indicates the type of this request.

Documentation/filesystems/caching/cachefiles.rst the control file. For example:: · line 547

Quoted 10 lines
Documentation/filesystems/caching/cachefiles.rst +10 -0
+
+	* ``object_id`` is a unique ID identifying the cache file operated on.
+
+	* ``data`` indicates the payload of this request.
+
+	* ``len`` indicates the whole length of this request, including the
+	  header and following type-specific payload.
+
+
+Turn on On-demand Mode

Turning on

Review

Quoted 3 lines

+----------------------

(quoted diff) +2 -0
+
+An optional parameter is added

is added -> becomes available

Documentation/filesystems/caching/cachefiles.rst the control file. For example::

Quoted 5 lines

to the "bind" command::

Documentation/filesystems/caching/cachefiles.rst +4 -0
+
+	bind [ondemand]
+
+When the "bind" command takes without

takes without -> is given no

Quoted 2 lines

argument, it defaults to the original
+mode. When the "bind" command is given

When it is given

Quoted 26 lines

the "ondemand" argument, i.e.
+"bind ondemand", on-demand read mode will be enabled.

Documentation/filesystems/caching/cachefiles.rst +24 -0
+
+
+The OPEN Request
+----------------
+
+When the netfs opens a cache file for the first time, a request with the
+CACHEFILES_OP_OPEN opcode, a.k.a an OPEN request will be sent to the user
+daemon.  The payload format is of the form::
+
+	struct cachefiles_open {
+		__u32 volume_key_size;
+		__u32 cookie_key_size;
+		__u32 fd;
+		__u32 flags;
+		__u8  data[];
+	};
+
+	where:
+
+	* ``data`` contains the volume_key followed directly by the cookie_key.
+	  The volume key is a NUL-terminated string; the cookie key is binary
+	  data.
+
+	* ``volume_key_size`` indicates the size of the volume key in bytes.

Documentation/filesystems/caching/cachefiles.rst the control file. For example::

Quoted 10 lines
Documentation/filesystems/caching/cachefiles.rst +10 -0
+
+	* ``cookie_key_size`` indicates the size of the cookie key in bytes.
+
+	* ``fd`` indicates an anonymous fd referring to the cache file, through
+	  which the user daemon can perform write/llseek file operations on the
+	  cache file.
+
+
+The user daemon is able to distinguish the requested cache file with the given
+(volume_key, cookie_key) pair.

"The user daemon can use the given (volume_key, cookie_key) pair to distinguish the requested cache file." might sound better.

Documentation/filesystems/caching/cachefiles.rst the control file. For example::

Quoted 11 lines

Each cache file has a unique object_id, while it

Documentation/filesystems/caching/cachefiles.rst +10 -0
+may have multiple anonymous fds. The user daemon may duplicate anonymous fds
+from the initial anonymous fd indicated by the @fd field through dup(). Thus
+each object_id can be mapped to multiple anonymous fds, while the usr daemon
+itself needs to maintain the mapping.
+
+With the given anonymous fd, the user daemon can fetch data and write it to the
+cache file in the background, even when kernel has not triggered a cache miss
+yet.
+
+The user daemon should complete the READ request

READ request -> OPEN request?

Documentation/filesystems/caching/cachefiles.rst the control file. For example::

Quoted 9 lines

by issuing a "copen" (complete

Documentation/filesystems/caching/cachefiles.rst +8 -0
+open) command on the devnode::
+
+	copen <msg_id>,<cache_size>
+
+	* ``msg_id`` must match the msg_id field of the previous OPEN request.
+
+	* When >= 0, ``cache_size`` indicates the size of the cache file;
+	  when < 0, ``cache_size`` indicates the

the -> any

Quoted 1 line

error code ecountered

encountered

Documentation/filesystems/caching/cachefiles.rst the control file. For example::

Quoted 9 lines

by the

Documentation/filesystems/caching/cachefiles.rst +8 -0
+	  user daemon.
+
+
+The CLOSE Request
+-----------------
+
+When a cookie withdrawn, a CLOSE request (opcode CACHEFILES_OP_CLOSE) will be
+sent to the user daemon. It will notify

It will notify -> This tells

Quoted 2 lines

the user daemon to close all anonymous
+fds associated with the given object_id. The CLOSE request has no extea

extra

Documentation/filesystems/caching/cachefiles.rst the control file. For example::

Quoted 7 lines
Documentation/filesystems/caching/cachefiles.rst +7 -0
+payload.
+
+
+The READ Request
+----------------
+
+When on-demand read mode is turned on, and a cache miss encountered,

"When a cache miss is encountered in on-demand read mode,"

Review

Quoted 3 lines

the kernel

(quoted diff) +2 -0
+will send a READ request (opcode CACHEFILES_OP_READ) to the user daemon. This
+will tell

will tell -> tells/asks

Quoted 1 line

the user daemon to fetch data

data -> the contents

Quoted 6 lines

of the requested file range. The payload

Documentation/filesystems/caching/cachefiles.rst +5 -0
+is of the form::
+
+	struct cachefiles_read {
+		__u64 off;
+		__u64 len;

Documentation/filesystems/caching/cachefiles.rst the control file. For example::

Quoted 10 lines
Documentation/filesystems/caching/cachefiles.rst +10 -0
+	};
+
+	where:
+
+	* ``off`` indicates the starting offset of the requested file range.
+
+	* ``len`` indicates the length of the requested file range.
+
+
+When receiving

receiving -> it receives

Quoted 1 line

a READ request, the user daemon needs to

needs to -> should

Quoted 1 line

fetch the

requested

Quoted 2 lines

data of the
+requested file range,

"of the requested file range," -> "" (including the comma, I think)

Quoted 1 line

and then

"then" -> ""

Documentation/filesystems/caching/cachefiles.rst the control file. For example::

Quoted 4 lines

write it to the cache file identified by

Documentation/filesystems/caching/cachefiles.rst +3 -0
+object_id.
+
+To finish

When it has finished

Quoted 1 line

processing the READ request, the user daemon should reply with

with -> by using

Review

Quoted 3 lines

the

(quoted diff) +2 -0
+CACHEFILES_IOC_CREAD ioctl on one of the anonymous fds associated with the given
+object_id

given object_id -> object_id given

Documentation/filesystems/caching/cachefiles.rst the control file. For example:: · line 652

Quoted 6 lines

in the READ request. The ioctl is of the form::

Documentation/filesystems/caching/cachefiles.rst +5 -0
+
+	ioctl(fd, CACHEFILES_IOC_CREAD, msg_id);
+
+	* ``fd`` is one of the anonymous fds associated with the given object_id
+	  in the READ request.

the given object_id in the READ request -> object_id

Documentation/filesystems/caching/cachefiles.rst the control file. For example:: · line 654

Quoted 2 lines
Documentation/filesystems/caching/cachefiles.rst +2 -0
+
+	* ``msg_id`` must match the msg_id field of the previous READ request.

By "previous READ request" is this referring to something different to "the READ request" you mentioned against the fd parameter?

David

author
#26 lore raw

Re: [PATCH v9 08/21] cachefiles: document on-demand read mode

Hi David, thanks for polishing the documents. It's a detailed and meticulous review again. Really thanks for your time :) I will fix all these in the next version.

Quoted 9 lines · David Howells

Documentation/filesystems/caching/cachefiles.rst the control file. For example:: · line 499

Quoted 4 lines · Jeffle Xu
Documentation/filesystems/caching/cachefiles.rst +4 -0
+The essential difference between these two modes is that, in original mode,
+when a cache miss occurs, the netfs will fetch the data from the remote server
+and then write it to the cache file.  With on-demand read mode, however,
+fetching the data and writing it into the cache is delegated to a user daemon.

The starting sentence seems off. How about:

Quoted 4 lines

The essential difference between these two modes is seen when a cache miss
occurs: In the original mode, the netfs will fetch the data from the remote server and then write it to the cache file; in on-demand read mode, fetching data and writing it into the cache is delegated to a user daemon.

Okay, it sounds better.

Quoted 9 lines

Documentation/filesystems/caching/cachefiles.rst the control file. For example::

Quoted 6 lines

the devnode ('/dev/cachefiles') to check if

Documentation/filesystems/caching/cachefiles.rst +5 -0
+there's a pending request to be processed.  A POLLIN event will be returned
+when there's a pending request.
+
+The user daemon then reads the devnode to fetch a request and process it
+accordingly.

Reading the devnode doesn't process the request, so I think something like:

Quoted 5 lines

"... and process it accordingly" -> "... that it can then process."

or:

"... and process it accordingly" -> "... to process."

Yeah the original statement is indeed misleading.

Quoted 5 lines
Quoted 5 lines

Each cache file has a unique object_id, while it

Documentation/filesystems/caching/cachefiles.rst +4 -0
+may have multiple anonymous fds. The user daemon may duplicate anonymous fds
+from the initial anonymous fd indicated by the @fd field through dup(). Thus
+each object_id can be mapped to multiple anonymous fds, while the usr daemon
+itself needs to maintain the mapping.
Quoted 8 lines

Documentation/filesystems/caching/cachefiles.rst the control file. For example::

Quoted 6 lines
Documentation/filesystems/caching/cachefiles.rst +6 -0
+
+With the given anonymous fd, the user daemon can fetch data and write it to the
+cache file in the background, even when kernel has not triggered a cache miss
+yet.
+
+The user daemon should complete the READ request

READ request -> OPEN request?

Good catch. Will be fixed.

Quoted 9 lines

Documentation/filesystems/caching/cachefiles.rst the control file. For example:: · line 652

Quoted 6 lines

in the READ request. The ioctl is of the form::

Documentation/filesystems/caching/cachefiles.rst +5 -0
+
+	ioctl(fd, CACHEFILES_IOC_CREAD, msg_id);
+
+	* ``fd`` is one of the anonymous fds associated with the given object_id
+	  in the READ request.

the given object_id in the READ request -> object_id

Quoted 5 lines

Documentation/filesystems/caching/cachefiles.rst the control file. For example:: · line 654

Quoted 2 lines
Documentation/filesystems/caching/cachefiles.rst +2 -0
+
+	* ``msg_id`` must match the msg_id field of the previous READ request.

By "previous READ request" is this referring to something different to "the READ request" you mentioned against the fd parameter?

Actually it is referring to the same thing (the same READ request). I will change the statement simply to:

``msg_id`` must match the msg_id field of the READ request.

Signature · 2 lines
Thanks,
Jeffle
author
#27 lore raw

PATCH v9 9/21 erofs: make erofs_map_blocks() generally available

In reply to Jeffle Xu

... so that it can be used in the following introduced fscache mode.

---
 fs/erofs/data.c     | 4 ++--
 fs/erofs/internal.h | 2 ++
 2 files changed, 4 insertions(+), 2 deletions(-)
fs/erofs/data.c +2 -2
index 780db1e5f4b7..bc22642358ec 100644
@@ -110,8 +110,8 @@ static int erofs_map_blocks_flatmode(struct inode *inode,
 	return 0;
 }
 
-static int erofs_map_blocks(struct inode *inode,
-			    struct erofs_map_blocks *map, int flags)
+int erofs_map_blocks(struct inode *inode,
+		     struct erofs_map_blocks *map, int flags)
 {
 	struct super_block *sb = inode->i_sb;
 	struct erofs_inode *vi = EROFS_I(inode);
fs/erofs/internal.h +2 -0
index 5298c4ee277d..fe9564e5091e 100644
@@ -486,6 +486,8 @@ void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb,
 int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *dev);
 int erofs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 		 u64 start, u64 len);
+int erofs_map_blocks(struct inode *inode,
+		     struct erofs_map_blocks *map, int flags);
 
 /* inode.c */
 static inline unsigned long erofs_inode_hash(erofs_nid_t nid)
Signature · 1 line
2.27.0
author
#28 lore raw

PATCH v9 10/21 erofs: add fscache mode check helper

In reply to Jeffle Xu

Until then erofs is exactly blockdev based filesystem.

A new fscache-based mode is going to be introduced for erofs to support scenarios where on-demand read semantics is needed, e.g. container image distribution. In this case, erofs could be mounted from data blobs through fscache.

Add a helper checking which mode erofs works in, and twist the code in prep for the following fscache mode.

---
 fs/erofs/internal.h |  5 +++++
 fs/erofs/super.c    | 44 +++++++++++++++++++++++++++++---------------
 2 files changed, 34 insertions(+), 15 deletions(-)
fs/erofs/internal.h +5 -0
index fe9564e5091e..05a97533b1e9 100644
@@ -161,6 +161,11 @@ struct erofs_sb_info {
 #define set_opt(opt, option)	((opt)->mount_opt |= EROFS_MOUNT_##option)
 #define test_opt(opt, option)	((opt)->mount_opt & EROFS_MOUNT_##option)
 
+static inline bool erofs_is_fscache_mode(struct super_block *sb)
+{
+	return IS_ENABLED(CONFIG_EROFS_FS_ONDEMAND) && !sb->s_bdev;
+}
+
 enum {
 	EROFS_ZIP_CACHE_DISABLED,
 	EROFS_ZIP_CACHE_READAHEAD,
fs/erofs/super.c +29 -15

72 lines not on this page · raw message

Signature · 1 line
2.27.0
Reviewed-by on v9 10/21
#29 lore raw

Re: [PATCH v9 10/21] erofs: add fscache mode check helper

Quoted 9 lines · Jeffle Xu

Until then erofs is exactly blockdev based filesystem.

A new fscache-based mode is going to be introduced for erofs to support scenarios where on-demand read semantics is needed, e.g. container image distribution. In this case, erofs could be mounted from data blobs through fscache.

Add a helper checking which mode erofs works in, and twist the code in prep for the following fscache mode.

in preparation for the upcoming fscache mode.

Quoted 2 lines

Thanks,
Gao Xiang

Quoted 102 lines
---
 fs/erofs/internal.h |  5 +++++
 fs/erofs/super.c    | 44 +++++++++++++++++++++++++++++---------------
 2 files changed, 34 insertions(+), 15 deletions(-)
fs/erofs/internal.h +5 -0
index fe9564e5091e..05a97533b1e9 100644
@@ -161,6 +161,11 @@ struct erofs_sb_info {
 #define set_opt(opt, option)	((opt)->mount_opt |= EROFS_MOUNT_##option)
 #define test_opt(opt, option)	((opt)->mount_opt & EROFS_MOUNT_##option)
 
+static inline bool erofs_is_fscache_mode(struct super_block *sb)
+{
+	return IS_ENABLED(CONFIG_EROFS_FS_ONDEMAND) && !sb->s_bdev;
+}
+
 enum {
 	EROFS_ZIP_CACHE_DISABLED,
 	EROFS_ZIP_CACHE_READAHEAD,
fs/erofs/super.c +29 -15
index 0c4b41130c2f..724d5ff0d78c 100644
@@ -259,15 +259,19 @@ static int erofs_init_devices(struct super_block *sb,
 		}
 		dis = ptr + erofs_blkoff(pos);
 
-		bdev = blkdev_get_by_path(dif->path,
-					  FMODE_READ | FMODE_EXCL,
-					  sb->s_type);
-		if (IS_ERR(bdev)) {
-			err = PTR_ERR(bdev);
-			break;
+		if (!erofs_is_fscache_mode(sb)) {
+			bdev = blkdev_get_by_path(dif->path,
+						  FMODE_READ | FMODE_EXCL,
+						  sb->s_type);
+			if (IS_ERR(bdev)) {
+				err = PTR_ERR(bdev);
+				break;
+			}
+			dif->bdev = bdev;
+			dif->dax_dev = fs_dax_get_by_bdev(bdev,
+							  &dif->dax_part_off);
 		}
-		dif->bdev = bdev;
-		dif->dax_dev = fs_dax_get_by_bdev(bdev, &dif->dax_part_off);
+
 		dif->blocks = le32_to_cpu(dis->blocks);
 		dif->mapped_blkaddr = le32_to_cpu(dis->mapped_blkaddr);
 		sbi->total_blocks += dif->blocks;
@@ -586,21 +590,28 @@ static int erofs_fc_fill_super(struct super_block *sb, struct fs_context *fc)
 
 	sb->s_magic = EROFS_SUPER_MAGIC;
 
-	if (!sb_set_blocksize(sb, EROFS_BLKSIZ)) {
-		erofs_err(sb, "failed to set erofs blksize");
-		return -EINVAL;
-	}
-
 	sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
 	if (!sbi)
 		return -ENOMEM;
 
 	sb->s_fs_info = sbi;
 	sbi->opt = ctx->opt;
-	sbi->dax_dev = fs_dax_get_by_bdev(sb->s_bdev, &sbi->dax_part_off);
 	sbi->devs = ctx->devs;
 	ctx->devs = NULL;
 
+	if (erofs_is_fscache_mode(sb)) {
+		sb->s_blocksize = EROFS_BLKSIZ;
+		sb->s_blocksize_bits = LOG_BLOCK_SIZE;
+	} else {
+		if (!sb_set_blocksize(sb, EROFS_BLKSIZ)) {
+			erofs_err(sb, "failed to set erofs blksize");
+			return -EINVAL;
+		}
+
+		sbi->dax_dev = fs_dax_get_by_bdev(sb->s_bdev,
+						  &sbi->dax_part_off);
+	}
+
 	err = erofs_read_superblock(sb);
 	if (err)
 		return err;
@@ -857,7 +868,10 @@ static int erofs_statfs(struct dentry *dentry, struct kstatfs *buf)
 {
 	struct super_block *sb = dentry->d_sb;
 	struct erofs_sb_info *sbi = EROFS_SB(sb);
-	u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
+	u64 id = 0;
+
+	if (!erofs_is_fscache_mode(sb))
+		id = huge_encode_dev(sb->s_bdev->bd_dev);
 
 	buf->f_type = sb->s_magic;
 	buf->f_bsize = EROFS_BLKSIZ;
Signature · 1 line
2.27.0
author
#30 lore raw

PATCH v9 11/21 erofs: register fscache volume

In reply to Jeffle Xu

A new fscache based mode is going to be introduced for erofs, in which case on-demand read semantics is implemented through fscache.

As the first step, register fscache volume for each erofs filesystem. That means, data blobs can not be shared among erofs filesystems. In the following iteration, we are going to introduce the domain semantics, in which case several erofs filesystems can belong to one domain, and data blobs can be shared among these erofs filesystems of one domain.

---
 fs/erofs/Kconfig    | 10 ++++++++++
 fs/erofs/Makefile   |  1 +
 fs/erofs/fscache.c  | 37 +++++++++++++++++++++++++++++++++++++
 fs/erofs/internal.h | 16 ++++++++++++++++
 fs/erofs/super.c    |  5 +++++
 5 files changed, 69 insertions(+)
 create mode 100644 fs/erofs/fscache.c
fs/erofs/Kconfig +10 -0
index f57255ab88ed..85490370e0ca 100644
@@ -98,3 +98,13 @@ config EROFS_FS_ZIP_LZMA
 	  systems will be readable without selecting this option.
 
 	  If unsure, say N.
+
+config EROFS_FS_ONDEMAND
+	bool "EROFS fscache-based on-demand read support"
+	depends on CACHEFILES_ONDEMAND && (EROFS_FS=m && FSCACHE || EROFS_FS=y && FSCACHE=y)
+	default n
+	help
+	  This permits EROFS to use fscache-backed data blobs with on-demand
+	  read support.
+
+	  If unsure, say N.
fs/erofs/Makefile +1 -0
index 8a3317e38e5a..99bbc597a3e9 100644
@@ -5,3 +5,4 @@ erofs-objs := super.o inode.o data.o namei.o dir.o utils.o pcpubuf.o sysfs.o
 erofs-$(CONFIG_EROFS_FS_XATTR) += xattr.o
 erofs-$(CONFIG_EROFS_FS_ZIP) += decompressor.o zmap.o zdata.o
 erofs-$(CONFIG_EROFS_FS_ZIP_LZMA) += decompressor_lzma.o
+erofs-$(CONFIG_EROFS_FS_ONDEMAND) += fscache.o
fs/erofs/fscache.c new file +37 -0
new file mode 100644
index 000000000000..7a6d0239ebb1
@@ -0,0 +1,37 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright (C) 2022, Alibaba Cloud
+ */
+#include <linux/fscache.h>
+#include "internal.h"
+
+int erofs_fscache_register_fs(struct super_block *sb)
+{
+	struct erofs_sb_info *sbi = EROFS_SB(sb);
+	struct fscache_volume *volume;
+	char *name;
+	int ret = 0;
+
+	name = kasprintf(GFP_KERNEL, "erofs,%s", sbi->opt.fsid);
+	if (!name)
+		return -ENOMEM;
+
+	volume = fscache_acquire_volume(name, NULL, NULL, 0);
+	if (IS_ERR_OR_NULL(volume)) {
+		erofs_err(sb, "failed to register volume for %s", name);
+		ret = volume ? PTR_ERR(volume) : -EOPNOTSUPP;
+		volume = NULL;
+	}
+
+	sbi->volume = volume;
+	kfree(name);
+	return ret;
+}
+
+void erofs_fscache_unregister_fs(struct super_block *sb)
+{
+	struct erofs_sb_info *sbi = EROFS_SB(sb);
+
+	fscache_relinquish_volume(sbi->volume, NULL, false);
+	sbi->volume = NULL;
+}
fs/erofs/internal.h +16 -0
index 05a97533b1e9..e4f6a13f161f 100644
@@ -74,6 +74,7 @@ struct erofs_mount_opts {
 	unsigned int max_sync_decompress_pages;
 #endif
 	unsigned int mount_opt;
+	char *fsid;
 };
 
 struct erofs_dev_context {
@@ -146,6 +147,9 @@ struct erofs_sb_info {
 	/* sysfs support */
 	struct kobject s_kobj;		/* /sys/fs/erofs/<devname> */
 	struct completion s_kobj_unregister;
+
+	/* fscache support */
+	struct fscache_volume *volume;
 };
 
 #define EROFS_SB(sb) ((struct erofs_sb_info *)(sb)->s_fs_info)
@@ -618,6 +622,18 @@ static inline int z_erofs_load_lzma_config(struct super_block *sb,
 }
 #endif	/* !CONFIG_EROFS_FS_ZIP */
 
+/* fscache.c */
+#ifdef CONFIG_EROFS_FS_ONDEMAND
+int erofs_fscache_register_fs(struct super_block *sb);
+void erofs_fscache_unregister_fs(struct super_block *sb);
+#else
+static inline int erofs_fscache_register_fs(struct super_block *sb)
+{
+	return 0;
+}
+static inline void erofs_fscache_unregister_fs(struct super_block *sb) {}
+#endif
+
 #define EFSCORRUPTED    EUCLEAN         /* Filesystem is corrupted */
 
 #endif	/* __EROFS_INTERNAL_H */
fs/erofs/super.c +5 -0
index 724d5ff0d78c..fd8daa447237 100644
@@ -602,6 +602,10 @@ static int erofs_fc_fill_super(struct super_block *sb, struct fs_context *fc)
 	if (erofs_is_fscache_mode(sb)) {
 		sb->s_blocksize = EROFS_BLKSIZ;
 		sb->s_blocksize_bits = LOG_BLOCK_SIZE;
+
+		err = erofs_fscache_register_fs(sb);
+		if (err)
+			return err;
 	} else {
 		if (!sb_set_blocksize(sb, EROFS_BLKSIZ)) {
 			erofs_err(sb, "failed to set erofs blksize");
@@ -768,6 +772,7 @@ static void erofs_kill_sb(struct super_block *sb)
 
 	erofs_free_dev_context(sbi->devs);
 	fs_put_dax(sbi->dax_dev);
+	erofs_fscache_unregister_fs(sb);
 	kfree(sbi);
 	sb->s_fs_info = NULL;
 }
Signature · 1 line
2.27.0
author
#31 lore raw

PATCH v9 12/21 erofs: add fscache context helper functions

In reply to Jeffle Xu

Introduce a context structure for managing data blobs, and helper functions for initializing and cleaning up this context structure.

---
 fs/erofs/fscache.c  | 46 +++++++++++++++++++++++++++++++++++++++++++++
 fs/erofs/internal.h | 19 +++++++++++++++++++
 2 files changed, 65 insertions(+)
fs/erofs/fscache.c +46 -0
index 7a6d0239ebb1..67a3c4935245 100644
@@ -5,6 +5,52 @@
 #include <linux/fscache.h>
 #include "internal.h"
 
+/*
+ * Create an fscache context for data blob.
+ * Return: 0 on success and allocated fscache context is assigned to @fscache,
+ *	   negative error number on failure.
+ */
+int erofs_fscache_register_cookie(struct super_block *sb,
+				  struct erofs_fscache **fscache, char *name)
+{
+	struct fscache_volume *volume = EROFS_SB(sb)->volume;
+	struct erofs_fscache *ctx;
+	struct fscache_cookie *cookie;
+
+	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
+	if (!ctx)
+		return -ENOMEM;
+
+	cookie = fscache_acquire_cookie(volume, FSCACHE_ADV_WANT_CACHE_SIZE,
+					name, strlen(name), NULL, 0, 0);
+	if (!cookie) {
+		erofs_err(sb, "failed to get cookie for %s", name);
+		kfree(name);
+		return -EINVAL;
+	}
+
+	fscache_use_cookie(cookie, false);
+	ctx->cookie = cookie;
+
+	*fscache = ctx;
+	return 0;
+}
+
+void erofs_fscache_unregister_cookie(struct erofs_fscache **fscache)
+{
+	struct erofs_fscache *ctx = *fscache;
+
+	if (!ctx)
+		return;
+
+	fscache_unuse_cookie(ctx->cookie, NULL, NULL);
+	fscache_relinquish_cookie(ctx->cookie, false);
+	ctx->cookie = NULL;
+
+	kfree(ctx);
+	*fscache = NULL;
+}
+
 int erofs_fscache_register_fs(struct super_block *sb)
 {
 	struct erofs_sb_info *sbi = EROFS_SB(sb);
fs/erofs/internal.h +19 -0
index e4f6a13f161f..b1f19f058503 100644
@@ -97,6 +97,10 @@ struct erofs_sb_lz4_info {
 	u16 max_pclusterblks;
 };
 
+struct erofs_fscache {
+	struct fscache_cookie *cookie;
+};
+
 struct erofs_sb_info {
 	struct erofs_mount_opts opt;	/* options */
 #ifdef CONFIG_EROFS_FS_ZIP
@@ -626,12 +630,27 @@ static inline int z_erofs_load_lzma_config(struct super_block *sb,
 #ifdef CONFIG_EROFS_FS_ONDEMAND
 int erofs_fscache_register_fs(struct super_block *sb);
 void erofs_fscache_unregister_fs(struct super_block *sb);
+
+int erofs_fscache_register_cookie(struct super_block *sb,
+				  struct erofs_fscache **fscache, char *name);
+void erofs_fscache_unregister_cookie(struct erofs_fscache **fscache);
 #else
 static inline int erofs_fscache_register_fs(struct super_block *sb)
 {
 	return 0;
 }
 static inline void erofs_fscache_unregister_fs(struct super_block *sb) {}
+
+static inline int erofs_fscache_register_cookie(struct super_block *sb,
+						struct erofs_fscache **fscache,
+						char *name)
+{
+	return -EOPNOTSUPP;
+}
+
+static inline void erofs_fscache_unregister_cookie(struct erofs_fscache **fscache)
+{
+}
 #endif
 
 #define EFSCORRUPTED    EUCLEAN         /* Filesystem is corrupted */
Signature · 1 line
2.27.0
author
#32 lore raw

PATCH v9 13/21 erofs: add anonymous inode caching metadata for data blobs

In reply to Jeffle Xu

Introduce one anonymous inode for data blobs so that erofs can cache metadata directly within such anonymous inode.

---
 fs/erofs/fscache.c  | 39 ++++++++++++++++++++++++++++++++++++---
 fs/erofs/internal.h |  6 ++++--
 2 files changed, 40 insertions(+), 5 deletions(-)
fs/erofs/fscache.c +36 -3
index 67a3c4935245..1c88614203d2 100644
@@ -5,17 +5,22 @@
 #include <linux/fscache.h>
 #include "internal.h"
 
+static const struct address_space_operations erofs_fscache_meta_aops = {
+};
+
 /*
  * Create an fscache context for data blob.
  * Return: 0 on success and allocated fscache context is assigned to @fscache,
  *	   negative error number on failure.
  */
 int erofs_fscache_register_cookie(struct super_block *sb,
-				  struct erofs_fscache **fscache, char *name)
+				  struct erofs_fscache **fscache,
+				  char *name, bool need_inode)
 {
 	struct fscache_volume *volume = EROFS_SB(sb)->volume;
 	struct erofs_fscache *ctx;
 	struct fscache_cookie *cookie;
+	int ret;
 
 	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
 	if (!ctx)
@@ -25,15 +30,40 @@ int erofs_fscache_register_cookie(struct super_block *sb,
 					name, strlen(name), NULL, 0, 0);
 	if (!cookie) {
 		erofs_err(sb, "failed to get cookie for %s", name);
-		kfree(name);
-		return -EINVAL;
+		ret = -EINVAL;
+		goto err;
 	}
 
 	fscache_use_cookie(cookie, false);
 	ctx->cookie = cookie;
 
+	if (need_inode) {
+		struct inode *const inode = new_inode(sb);
+
+		if (!inode) {
+			erofs_err(sb, "failed to get anon inode for %s", name);
+			ret = -ENOMEM;
+			goto err_cookie;
+		}
+
+		set_nlink(inode, 1);
+		inode->i_size = OFFSET_MAX;
+		inode->i_mapping->a_ops = &erofs_fscache_meta_aops;
+		mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
+
+		ctx->inode = inode;
+	}
+
 	*fscache = ctx;
 	return 0;
+
+err_cookie:
+	fscache_unuse_cookie(ctx->cookie, NULL, NULL);
+	fscache_relinquish_cookie(ctx->cookie, false);
+	ctx->cookie = NULL;
+err:
+	kfree(ctx);
+	return ret;
 }
 
 void erofs_fscache_unregister_cookie(struct erofs_fscache **fscache)
@@ -47,6 +77,9 @@ void erofs_fscache_unregister_cookie(struct erofs_fscache **fscache)
 	fscache_relinquish_cookie(ctx->cookie, false);
 	ctx->cookie = NULL;
 
+	iput(ctx->inode);
+	ctx->inode = NULL;
+
 	kfree(ctx);
 	*fscache = NULL;
 }
fs/erofs/internal.h +4 -2
index b1f19f058503..5867cb63fd74 100644
@@ -99,6 +99,7 @@ struct erofs_sb_lz4_info {
 
 struct erofs_fscache {
 	struct fscache_cookie *cookie;
+	struct inode *inode;
 };
 
 struct erofs_sb_info {
@@ -632,7 +633,8 @@ int erofs_fscache_register_fs(struct super_block *sb);
 void erofs_fscache_unregister_fs(struct super_block *sb);
 
 int erofs_fscache_register_cookie(struct super_block *sb,
-				  struct erofs_fscache **fscache, char *name);
+				  struct erofs_fscache **fscache,
+				  char *name, bool need_inode);
 void erofs_fscache_unregister_cookie(struct erofs_fscache **fscache);
 #else
 static inline int erofs_fscache_register_fs(struct super_block *sb)
@@ -643,7 +645,7 @@ static inline void erofs_fscache_unregister_fs(struct super_block *sb) {}
 
 static inline int erofs_fscache_register_cookie(struct super_block *sb,
 						struct erofs_fscache **fscache,
-						char *name)
+						char *name, bool need_inode)
 {
 	return -EOPNOTSUPP;
 }
Signature · 1 line
2.27.0
author
#33 lore raw

PATCH v9 14/21 erofs: add erofs_fscache_read_folios() helper

In reply to Jeffle Xu

Add erofs_fscache_read_folios() helper reading from fscache. It supports on-demand read semantics. That is, it will make the backend prepare for the data when cache miss. Once data ready, it will read from the cache.

This helper can then be used to implement .readpage()/.readahead() of on-demand read semantics.

---
 fs/erofs/fscache.c | 53 ++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 53 insertions(+)
fs/erofs/fscache.c +53 -0
index 1c88614203d2..066f68c062e2 100644
@@ -5,6 +5,59 @@
 #include <linux/fscache.h>
 #include "internal.h"
 
+/*
+ * Read data from fscache and fill the read data into page cache described by
+ * @start/len, which shall be both aligned with PAGE_SIZE. @pstart describes
+ * the start physical address in the cache file.
+ */
+static int erofs_fscache_read_folios(struct fscache_cookie *cookie,
+				     struct address_space *mapping,
+				     loff_t start, size_t len,
+				     loff_t pstart)
+{
+	enum netfs_io_source source;
+	struct netfs_io_subrequest subreq;
+	struct netfs_io_request rreq;
+	struct netfs_cache_resources *cres = &rreq.cache_resources;
+	struct iov_iter iter;
+	size_t done = 0;
+	int ret;
+
+	memset(&rreq, 0, sizeof(rreq));
+	memset(&subreq, 0, sizeof(subreq));
+	subreq.rreq = &rreq;
+
+	ret = fscache_begin_read_operation(cres, cookie);
+	if (ret)
+		return ret;
+
+	while (done < len) {
+		subreq.start = pstart + done;
+		subreq.len = len - done;
+		subreq.flags = 1 << NETFS_SREQ_ONDEMAND;
+
+		source = cres->ops->prepare_read(&subreq, LLONG_MAX);
+		if (WARN_ON(subreq.len == 0))
+			source = NETFS_INVALID_READ;
+		if (source != NETFS_READ_FROM_CACHE) {
+			ret = -EIO;
+			goto out;
+		}
+
+		iov_iter_xarray(&iter, READ, &mapping->i_pages,
+				start + done, subreq.len);
+		ret = fscache_read(cres, subreq.start, &iter,
+				   NETFS_READ_HOLE_FAIL, NULL, NULL);
+		if (ret)
+			goto out;
+
+		done += subreq.len;
+	}
+out:
+	fscache_end_operation(cres);
+	return ret;
+}
+
 static const struct address_space_operations erofs_fscache_meta_aops = {
 };
 
Signature · 1 line
2.27.0
author
#34 lore raw

PATCH v9 15/21 erofs: register fscache context for primary data blob

In reply to Jeffle Xu

Registers fscache context for primary data blob. Also move the initialization of s_op and related fields forward, since anonymous inode will be allocated under the super block when registering the fscache context.

Something worth mentioning about the cleanup routine.

1. The fscache context will instantiate anonymous inodes under the super block. Release these anonymous inodes when .put_super() is called, or we'll get "VFS: Busy inodes after unmount." warning.

2. The fscache context is initialized prior to the root inode. If .kill_sb() is called when mount failed, .put_super() won't be called when root inode has not been initialized yet. Thus .kill_sb() shall also contain the cleanup routine.

---
 fs/erofs/internal.h |  1 +
 fs/erofs/super.c    | 15 +++++++++++----
 2 files changed, 12 insertions(+), 4 deletions(-)
fs/erofs/internal.h +1 -0
index 5867cb63fd74..386658416159 100644
@@ -155,6 +155,7 @@ struct erofs_sb_info {
 
 	/* fscache support */
 	struct fscache_volume *volume;
+	struct erofs_fscache *s_fscache;
 };
 
 #define EROFS_SB(sb) ((struct erofs_sb_info *)(sb)->s_fs_info)
fs/erofs/super.c +11 -4
index fd8daa447237..61dc900295f9 100644
@@ -589,6 +589,9 @@ static int erofs_fc_fill_super(struct super_block *sb, struct fs_context *fc)
 	int err;
 
 	sb->s_magic = EROFS_SUPER_MAGIC;
+	sb->s_flags |= SB_RDONLY | SB_NOATIME;
+	sb->s_maxbytes = MAX_LFS_FILESIZE;
+	sb->s_op = &erofs_sops;
 
 	sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
 	if (!sbi)
@@ -606,6 +609,11 @@ static int erofs_fc_fill_super(struct super_block *sb, struct fs_context *fc)
 		err = erofs_fscache_register_fs(sb);
 		if (err)
 			return err;
+
+		err = erofs_fscache_register_cookie(sb, &sbi->s_fscache,
+						    sbi->opt.fsid, true);
+		if (err)
+			return err;
 	} else {
 		if (!sb_set_blocksize(sb, EROFS_BLKSIZ)) {
 			erofs_err(sb, "failed to set erofs blksize");
@@ -628,11 +636,8 @@ static int erofs_fc_fill_super(struct super_block *sb, struct fs_context *fc)
 			clear_opt(&sbi->opt, DAX_ALWAYS);
 		}
 	}
-	sb->s_flags |= SB_RDONLY | SB_NOATIME;
-	sb->s_maxbytes = MAX_LFS_FILESIZE;
-	sb->s_time_gran = 1;
 
-	sb->s_op = &erofs_sops;
+	sb->s_time_gran = 1;
 	sb->s_xattr = erofs_xattr_handlers;
 
 	if (test_opt(&sbi->opt, POSIX_ACL))
@@ -772,6 +777,7 @@ static void erofs_kill_sb(struct super_block *sb)
 
 	erofs_free_dev_context(sbi->devs);
 	fs_put_dax(sbi->dax_dev);
+	erofs_fscache_unregister_cookie(&sbi->s_fscache);
 	erofs_fscache_unregister_fs(sb);
 	kfree(sbi);
 	sb->s_fs_info = NULL;
@@ -790,6 +796,7 @@ static void erofs_put_super(struct super_block *sb)
 	iput(sbi->managed_cache);
 	sbi->managed_cache = NULL;
 #endif
+	erofs_fscache_unregister_cookie(&sbi->s_fscache);
 }
 
 static struct file_system_type erofs_fs_type = {
Signature · 1 line
2.27.0
author
#35 lore raw

PATCH v9 16/21 erofs: register fscache context for extra data blobs

In reply to Jeffle Xu

Similar to the multi device mode, erofs could be mounted from one primary data blob (mandatory) and multiple extra data blobs (optional).

Register fscache context for each extra data blob.

---
 fs/erofs/data.c     | 3 +++
 fs/erofs/internal.h | 2 ++
 fs/erofs/super.c    | 8 +++++++-
 3 files changed, 12 insertions(+), 1 deletion(-)
fs/erofs/data.c +3 -0
index bc22642358ec..14b64d960541 100644
@@ -199,6 +199,7 @@ int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map)
 	map->m_bdev = sb->s_bdev;
 	map->m_daxdev = EROFS_SB(sb)->dax_dev;
 	map->m_dax_part_off = EROFS_SB(sb)->dax_part_off;
+	map->m_fscache = EROFS_SB(sb)->s_fscache;
 
 	if (map->m_deviceid) {
 		down_read(&devs->rwsem);
@@ -210,6 +211,7 @@ int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map)
 		map->m_bdev = dif->bdev;
 		map->m_daxdev = dif->dax_dev;
 		map->m_dax_part_off = dif->dax_part_off;
+		map->m_fscache = dif->fscache;
 		up_read(&devs->rwsem);
 	} else if (devs->extra_devices) {
 		down_read(&devs->rwsem);
@@ -227,6 +229,7 @@ int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map)
 				map->m_bdev = dif->bdev;
 				map->m_daxdev = dif->dax_dev;
 				map->m_dax_part_off = dif->dax_part_off;
+				map->m_fscache = dif->fscache;
 				break;
 			}
 		}
fs/erofs/internal.h +2 -0
index 386658416159..fa488af8dfcf 100644
@@ -49,6 +49,7 @@ typedef u32 erofs_blk_t;
 
 struct erofs_device_info {
 	char *path;
+	struct erofs_fscache *fscache;
 	struct block_device *bdev;
 	struct dax_device *dax_dev;
 	u64 dax_part_off;
@@ -482,6 +483,7 @@ static inline int z_erofs_map_blocks_iter(struct inode *inode,
 #endif	/* !CONFIG_EROFS_FS_ZIP */
 
 struct erofs_map_dev {
+	struct erofs_fscache *m_fscache;
 	struct block_device *m_bdev;
 	struct dax_device *m_daxdev;
 	u64 m_dax_part_off;
fs/erofs/super.c +7 -1
index 61dc900295f9..c6755bcae4a6 100644
@@ -259,7 +259,12 @@ static int erofs_init_devices(struct super_block *sb,
 		}
 		dis = ptr + erofs_blkoff(pos);
 
-		if (!erofs_is_fscache_mode(sb)) {
+		if (erofs_is_fscache_mode(sb)) {
+			err = erofs_fscache_register_cookie(sb, &dif->fscache,
+							    dif->path, false);
+			if (err)
+				break;
+		} else {
 			bdev = blkdev_get_by_path(dif->path,
 						  FMODE_READ | FMODE_EXCL,
 						  sb->s_type);
@@ -710,6 +715,7 @@ static int erofs_release_device_info(int id, void *ptr, void *data)
 	fs_put_dax(dif->dax_dev);
 	if (dif->bdev)
 		blkdev_put(dif->bdev, FMODE_READ | FMODE_EXCL);
+	erofs_fscache_unregister_cookie(&dif->fscache);
 	kfree(dif->path);
 	kfree(dif);
 	return 0;
Signature · 1 line
2.27.0
Reviewed-by on v9 16/21
#36 lore raw

Re: [PATCH v9 16/21] erofs: register fscache context for extra data blobs

Quoted 6 lines · Jeffle Xu

Similar to the multi device mode, erofs could be mounted from one primary data blob (mandatory) and multiple extra data blobs (optional).

Register fscache context for each extra data blob.

Thanks,
Gao Xiang

Quoted 82 lines
---
 fs/erofs/data.c     | 3 +++
 fs/erofs/internal.h | 2 ++
 fs/erofs/super.c    | 8 +++++++-
 3 files changed, 12 insertions(+), 1 deletion(-)
fs/erofs/data.c +3 -0
index bc22642358ec..14b64d960541 100644
@@ -199,6 +199,7 @@ int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map)
 	map->m_bdev = sb->s_bdev;
 	map->m_daxdev = EROFS_SB(sb)->dax_dev;
 	map->m_dax_part_off = EROFS_SB(sb)->dax_part_off;
+	map->m_fscache = EROFS_SB(sb)->s_fscache;
 
 	if (map->m_deviceid) {
 		down_read(&devs->rwsem);
@@ -210,6 +211,7 @@ int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map)
 		map->m_bdev = dif->bdev;
 		map->m_daxdev = dif->dax_dev;
 		map->m_dax_part_off = dif->dax_part_off;
+		map->m_fscache = dif->fscache;
 		up_read(&devs->rwsem);
 	} else if (devs->extra_devices) {
 		down_read(&devs->rwsem);
@@ -227,6 +229,7 @@ int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map)
 				map->m_bdev = dif->bdev;
 				map->m_daxdev = dif->dax_dev;
 				map->m_dax_part_off = dif->dax_part_off;
+				map->m_fscache = dif->fscache;
 				break;
 			}
 		}
fs/erofs/internal.h +2 -0
index 386658416159..fa488af8dfcf 100644
@@ -49,6 +49,7 @@ typedef u32 erofs_blk_t;
 
 struct erofs_device_info {
 	char *path;
+	struct erofs_fscache *fscache;
 	struct block_device *bdev;
 	struct dax_device *dax_dev;
 	u64 dax_part_off;
@@ -482,6 +483,7 @@ static inline int z_erofs_map_blocks_iter(struct inode *inode,
 #endif	/* !CONFIG_EROFS_FS_ZIP */
 
 struct erofs_map_dev {
+	struct erofs_fscache *m_fscache;
 	struct block_device *m_bdev;
 	struct dax_device *m_daxdev;
 	u64 m_dax_part_off;
fs/erofs/super.c +7 -1
index 61dc900295f9..c6755bcae4a6 100644
@@ -259,7 +259,12 @@ static int erofs_init_devices(struct super_block *sb,
 		}
 		dis = ptr + erofs_blkoff(pos);
 
-		if (!erofs_is_fscache_mode(sb)) {
+		if (erofs_is_fscache_mode(sb)) {
+			err = erofs_fscache_register_cookie(sb, &dif->fscache,
+							    dif->path, false);
+			if (err)
+				break;
+		} else {
 			bdev = blkdev_get_by_path(dif->path,
 						  FMODE_READ | FMODE_EXCL,
 						  sb->s_type);
@@ -710,6 +715,7 @@ static int erofs_release_device_info(int id, void *ptr, void *data)
 	fs_put_dax(dif->dax_dev);
 	if (dif->bdev)
 		blkdev_put(dif->bdev, FMODE_READ | FMODE_EXCL);
+	erofs_fscache_unregister_cookie(&dif->fscache);
 	kfree(dif->path);
 	kfree(dif);
 	return 0;
Signature · 1 line
2.27.0
author
#37 lore raw

PATCH v9 17/21 erofs: implement fscache-based metadata read

In reply to Jeffle Xu

Implement the data plane of reading metadata from primary data blob over fscache.

---
 fs/erofs/data.c    | 19 +++++++++++++++----
 fs/erofs/fscache.c | 27 +++++++++++++++++++++++++++
 2 files changed, 42 insertions(+), 4 deletions(-)
fs/erofs/data.c +15 -4
index 14b64d960541..bb9c1fd48c19 100644
@@ -6,6 +6,7 @@
  */
 #include "internal.h"
 #include <linux/prefetch.h>
+#include <linux/sched/mm.h>
 #include <linux/dax.h>
 #include <trace/events/erofs.h>
 
@@ -35,14 +36,20 @@ void *erofs_bread(struct erofs_buf *buf, struct inode *inode,
 	erofs_off_t offset = blknr_to_addr(blkaddr);
 	pgoff_t index = offset >> PAGE_SHIFT;
 	struct page *page = buf->page;
+	struct folio *folio;
+	unsigned int nofs_flag;
 
 	if (!page || page->index != index) {
 		erofs_put_metabuf(buf);
-		page = read_cache_page_gfp(mapping, index,
-				mapping_gfp_constraint(mapping, ~__GFP_FS));
-		if (IS_ERR(page))
-			return page;
+
+		nofs_flag = memalloc_nofs_save();
+		folio = read_cache_folio(mapping, index, NULL, NULL);
+		memalloc_nofs_restore(nofs_flag);
+		if (IS_ERR(folio))
+			return folio;
+
 		/* should already be PageUptodate, no need to lock page */
+		page = folio_file_page(folio, index);
 		buf->page = page;
 	}
 	if (buf->kmap_type == EROFS_NO_KMAP) {
@@ -63,6 +70,10 @@ void *erofs_bread(struct erofs_buf *buf, struct inode *inode,
 void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb,
 			 erofs_blk_t blkaddr, enum erofs_kmap_type type)
 {
+	if (erofs_is_fscache_mode(sb))
+		return erofs_bread(buf, EROFS_SB(sb)->s_fscache->inode,
+				   blkaddr, type);
+
 	return erofs_bread(buf, sb->s_bdev->bd_inode, blkaddr, type);
 }
 
fs/erofs/fscache.c +27 -0
index 066f68c062e2..3f00eb34ac35 100644
@@ -58,7 +58,34 @@ static int erofs_fscache_read_folios(struct fscache_cookie *cookie,
 	return ret;
 }
 
+static int erofs_fscache_meta_readpage(struct file *data, struct page *page)
+{
+	int ret;
+	struct folio *folio = page_folio(page);
+	struct super_block *sb = folio_mapping(folio)->host->i_sb;
+	struct erofs_map_dev mdev = {
+		.m_deviceid = 0,
+		.m_pa = folio_pos(folio),
+	};
+
+	ret = erofs_map_dev(sb, &mdev);
+	if (ret)
+		goto out;
+
+	ret = erofs_fscache_read_folios(mdev.m_fscache->cookie,
+			folio_mapping(folio), folio_pos(folio),
+			folio_size(folio), mdev.m_pa);
+	if (ret)
+		goto out;
+
+	folio_mark_uptodate(folio);
+out:
+	folio_unlock(folio);
+	return ret;
+}
+
 static const struct address_space_operations erofs_fscache_meta_aops = {
+	.readpage = erofs_fscache_meta_readpage,
 };
 
 /*
Signature · 1 line
2.27.0
#38 lore raw

Re: [PATCH v9 17/21] erofs: implement fscache-based metadata read

Quoted 76 lines · Jeffle Xu

Implement the data plane of reading metadata from primary data blob over fscache.

---
 fs/erofs/data.c    | 19 +++++++++++++++----
 fs/erofs/fscache.c | 27 +++++++++++++++++++++++++++
 2 files changed, 42 insertions(+), 4 deletions(-)
fs/erofs/data.c +15 -4
index 14b64d960541..bb9c1fd48c19 100644
@@ -6,6 +6,7 @@
  */
 #include "internal.h"
 #include <linux/prefetch.h>
+#include <linux/sched/mm.h>
 #include <linux/dax.h>
 #include <trace/events/erofs.h>
 
@@ -35,14 +36,20 @@ void *erofs_bread(struct erofs_buf *buf, struct inode *inode,
 	erofs_off_t offset = blknr_to_addr(blkaddr);
 	pgoff_t index = offset >> PAGE_SHIFT;
 	struct page *page = buf->page;
+	struct folio *folio;
+	unsigned int nofs_flag;
 
 	if (!page || page->index != index) {
 		erofs_put_metabuf(buf);
-		page = read_cache_page_gfp(mapping, index,
-				mapping_gfp_constraint(mapping, ~__GFP_FS));
-		if (IS_ERR(page))
-			return page;
+
+		nofs_flag = memalloc_nofs_save();
+		folio = read_cache_folio(mapping, index, NULL, NULL);
+		memalloc_nofs_restore(nofs_flag);
+		if (IS_ERR(folio))
+			return folio;
+
 		/* should already be PageUptodate, no need to lock page */
+		page = folio_file_page(folio, index);
 		buf->page = page;
 	}
 	if (buf->kmap_type == EROFS_NO_KMAP) {
@@ -63,6 +70,10 @@ void *erofs_bread(struct erofs_buf *buf, struct inode *inode,
 void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb,
 			 erofs_blk_t blkaddr, enum erofs_kmap_type type)
 {
+	if (erofs_is_fscache_mode(sb))
+		return erofs_bread(buf, EROFS_SB(sb)->s_fscache->inode,
+				   blkaddr, type);
+
 	return erofs_bread(buf, sb->s_bdev->bd_inode, blkaddr, type);
 }
 
fs/erofs/fscache.c +11 -0
index 066f68c062e2..3f00eb34ac35 100644
@@ -58,7 +58,34 @@ static int erofs_fscache_read_folios(struct fscache_cookie *cookie,
 	return ret;
 }
 
+static int erofs_fscache_meta_readpage(struct file *data, struct page *page)
+{
+	int ret;
+	struct folio *folio = page_folio(page);
+	struct super_block *sb = folio_mapping(folio)->host->i_sb;
+	struct erofs_map_dev mdev = {
+		.m_deviceid = 0,
+		.m_pa = folio_pos(folio),
+	};
+
+	ret = erofs_map_dev(sb, &mdev);

fs/erofs/fscache.c static int erofs_fscache_read_folios(struct fscache_cookie *cookie, · line 81

Quoted 10 lines
fs/erofs/fscache.c +10 -0
+	if (ret)
+		goto out;
+
+	ret = erofs_fscache_read_folios(mdev.m_fscache->cookie,
+			folio_mapping(folio), folio_pos(folio),
+			folio_size(folio), mdev.m_pa);
+	if (ret)
+		goto out;
+
+	folio_mark_uptodate(folio);
	if (!ret)
		folio_mark_uptodate(folio);

Thanks,
Gao Xiang

Quoted 12 lines
fs/erofs/fscache.c +6 -0
+out:
+	folio_unlock(folio);
+	return ret;
+}
+
 static const struct address_space_operations erofs_fscache_meta_aops = {
+	.readpage = erofs_fscache_meta_readpage,
 };
 
 /*
Signature · 1 line
2.27.0
author
#39 lore raw

PATCH v9 18/21 erofs: implement fscache-based data read for non-inline layout

In reply to Jeffle Xu

Implement the data plane of reading data from data blobs over fscache for non-inline layout.

---
 fs/erofs/fscache.c  | 51 +++++++++++++++++++++++++++++++++++++++++++++
 fs/erofs/inode.c    |  4 ++++
 fs/erofs/internal.h |  2 ++
 3 files changed, 57 insertions(+)
fs/erofs/fscache.c +51 -0
index 3f00eb34ac35..b799b0fe1b67 100644
@@ -84,10 +84,61 @@ static int erofs_fscache_meta_readpage(struct file *data, struct page *page)
 	return ret;
 }
 
+static int erofs_fscache_readpage(struct file *file, struct page *page)
+{
+	struct folio *folio = page_folio(page);
+	struct inode *inode = folio_mapping(folio)->host;
+	struct super_block *sb = inode->i_sb;
+	struct erofs_map_blocks map;
+	struct erofs_map_dev mdev;
+	erofs_off_t pos;
+	loff_t pstart;
+	int ret = 0;
+
+	DBG_BUGON(folio_size(folio) != EROFS_BLKSIZ);
+
+	pos = folio_pos(folio);
+	map.m_la = pos;
+
+	ret = erofs_map_blocks(inode, &map, EROFS_GET_BLOCKS_RAW);
+	if (ret)
+		goto out_unlock;
+
+	if (!(map.m_flags & EROFS_MAP_MAPPED)) {
+		folio_zero_range(folio, 0, folio_size(folio));
+		goto out_uptodate;
+	}
+
+	mdev = (struct erofs_map_dev) {
+		.m_deviceid = map.m_deviceid,
+		.m_pa = map.m_pa,
+	};
+
+	ret = erofs_map_dev(sb, &mdev);
+	if (ret)
+		goto out_unlock;
+
+	pstart = mdev.m_pa + (pos - map.m_la);
+	ret = erofs_fscache_read_folios(mdev.m_fscache->cookie,
+			folio_mapping(folio), folio_pos(folio),
+			folio_size(folio), pstart);
+
+out_uptodate:
+	if (!ret)
+		folio_mark_uptodate(folio);
+out_unlock:
+	folio_unlock(folio);
+	return ret;
+}
+
 static const struct address_space_operations erofs_fscache_meta_aops = {
 	.readpage = erofs_fscache_meta_readpage,
 };
 
+const struct address_space_operations erofs_fscache_access_aops = {
+	.readpage = erofs_fscache_readpage,
+};
+
 /*
  * Create an fscache context for data blob.
  * Return: 0 on success and allocated fscache context is assigned to @fscache,
fs/erofs/inode.c +4 -0
index e8b37ba5e9ad..8d3f56c6469b 100644
@@ -297,6 +297,10 @@ static int erofs_fill_inode(struct inode *inode, int isdir)
 		goto out_unlock;
 	}
 	inode->i_mapping->a_ops = &erofs_raw_access_aops;
+#ifdef CONFIG_EROFS_FS_ONDEMAND
+	if (erofs_is_fscache_mode(inode->i_sb))
+		inode->i_mapping->a_ops = &erofs_fscache_access_aops;
+#endif
 
 out_unlock:
 	erofs_put_metabuf(&buf);
fs/erofs/internal.h +2 -0
index fa488af8dfcf..c8f6ac910976 100644
@@ -639,6 +639,8 @@ int erofs_fscache_register_cookie(struct super_block *sb,
 				  struct erofs_fscache **fscache,
 				  char *name, bool need_inode);
 void erofs_fscache_unregister_cookie(struct erofs_fscache **fscache);
+
+extern const struct address_space_operations erofs_fscache_access_aops;
 #else
 static inline int erofs_fscache_register_fs(struct super_block *sb)
 {
Signature · 1 line
2.27.0
#40 lore raw

Re: [PATCH v9 18/21] erofs: implement fscache-based data read for non-inline layout

Quoted 19 lines · Jeffle Xu

Implement the data plane of reading data from data blobs over fscache for non-inline layout.

---
 fs/erofs/fscache.c  | 51 +++++++++++++++++++++++++++++++++++++++++++++
 fs/erofs/inode.c    |  4 ++++
 fs/erofs/internal.h |  2 ++
 3 files changed, 57 insertions(+)
fs/erofs/fscache.c +0 -0
index 3f00eb34ac35..b799b0fe1b67 100644
@@ -84,10 +84,61 @@ static int erofs_fscache_meta_readpage(struct file *data, struct page *page)
 	return ret;
 }
 

fs/erofs/fscache.c static int erofs_fscache_meta_readpage(struct file *data, struct page *page) · line 96

Quoted 10 lines
fs/erofs/fscache.c +10 -0
+static int erofs_fscache_readpage(struct file *file, struct page *page)
+{
+	struct folio *folio = page_folio(page);
+	struct inode *inode = folio_mapping(folio)->host;
+	struct super_block *sb = inode->i_sb;
+	struct erofs_map_blocks map;
+	struct erofs_map_dev mdev;
+	erofs_off_t pos;
+	loff_t pstart;
+	int ret = 0;

Redundant assignment now?

Quoted 78 lines
fs/erofs/fscache.c +41 -0
+
+	DBG_BUGON(folio_size(folio) != EROFS_BLKSIZ);
+
+	pos = folio_pos(folio);
+	map.m_la = pos;
+
+	ret = erofs_map_blocks(inode, &map, EROFS_GET_BLOCKS_RAW);
+	if (ret)
+		goto out_unlock;
+
+	if (!(map.m_flags & EROFS_MAP_MAPPED)) {
+		folio_zero_range(folio, 0, folio_size(folio));
+		goto out_uptodate;
+	}
+
+	mdev = (struct erofs_map_dev) {
+		.m_deviceid = map.m_deviceid,
+		.m_pa = map.m_pa,
+	};
+
+	ret = erofs_map_dev(sb, &mdev);
+	if (ret)
+		goto out_unlock;
+
+	pstart = mdev.m_pa + (pos - map.m_la);
+	ret = erofs_fscache_read_folios(mdev.m_fscache->cookie,
+			folio_mapping(folio), folio_pos(folio),
+			folio_size(folio), pstart);
+
+out_uptodate:
+	if (!ret)
+		folio_mark_uptodate(folio);
+out_unlock:
+	folio_unlock(folio);
+	return ret;
+}
+
 static const struct address_space_operations erofs_fscache_meta_aops = {
 	.readpage = erofs_fscache_meta_readpage,
 };
 
+const struct address_space_operations erofs_fscache_access_aops = {
+	.readpage = erofs_fscache_readpage,
+};
+
 /*
  * Create an fscache context for data blob.
  * Return: 0 on success and allocated fscache context is assigned to @fscache,
fs/erofs/inode.c +4 -0
index e8b37ba5e9ad..8d3f56c6469b 100644
@@ -297,6 +297,10 @@ static int erofs_fill_inode(struct inode *inode, int isdir)
 		goto out_unlock;
 	}
 	inode->i_mapping->a_ops = &erofs_raw_access_aops;
+#ifdef CONFIG_EROFS_FS_ONDEMAND
+	if (erofs_is_fscache_mode(inode->i_sb))
+		inode->i_mapping->a_ops = &erofs_fscache_access_aops;
+#endif
 
 out_unlock:
 	erofs_put_metabuf(&buf);
fs/erofs/internal.h +2 -0
index fa488af8dfcf..c8f6ac910976 100644
@@ -639,6 +639,8 @@ int erofs_fscache_register_cookie(struct super_block *sb,
 				  struct erofs_fscache **fscache,
 				  char *name, bool need_inode);
 void erofs_fscache_unregister_cookie(struct erofs_fscache **fscache);
+
+extern const struct address_space_operations erofs_fscache_access_aops;
 #else
 static inline int erofs_fscache_register_fs(struct super_block *sb)
 {
Signature · 1 line
2.27.0
author
#41 lore raw

PATCH v9 19/21 erofs: implement fscache-based data read for inline layout

In reply to Jeffle Xu

Implement the data plane of reading data from data blobs over fscache for inline layout.

For the heading non-inline part, the data plane for non-inline layout is reused, while only the tail packing part needs special handling.

---
 fs/erofs/fscache.c | 32 ++++++++++++++++++++++++++++++++
 1 file changed, 32 insertions(+)
fs/erofs/fscache.c +32 -0
index b799b0fe1b67..08849c15500f 100644
@@ -84,6 +84,33 @@ static int erofs_fscache_meta_readpage(struct file *data, struct page *page)
 	return ret;
 }
 
+static int erofs_fscache_readpage_inline(struct folio *folio,
+					 struct erofs_map_blocks *map)
+{
+	struct super_block *sb = folio_mapping(folio)->host->i_sb;
+	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
+	erofs_blk_t blknr;
+	size_t offset, len;
+	void *src, *dst;
+
+	/* For tail packing layout, the offset may be non-zero. */
+	offset = erofs_blkoff(map->m_pa);
+	blknr = erofs_blknr(map->m_pa);
+	len = map->m_llen;
+
+	src = erofs_read_metabuf(&buf, sb, blknr, EROFS_KMAP);
+	if (IS_ERR(src))
+		return PTR_ERR(src);
+
+	dst = kmap_local_folio(folio, 0);
+	memcpy(dst, src + offset, len);
+	memset(dst + len, 0, PAGE_SIZE - len);
+	kunmap_local(dst);
+
+	erofs_put_metabuf(&buf);
+	return 0;
+}
+
 static int erofs_fscache_readpage(struct file *file, struct page *page)
 {
 	struct folio *folio = page_folio(page);
@@ -109,6 +136,11 @@ static int erofs_fscache_readpage(struct file *file, struct page *page)
 		goto out_uptodate;
 	}
 
+	if (map.m_flags & EROFS_MAP_META) {
+		ret = erofs_fscache_readpage_inline(folio, &map);
+		goto out_uptodate;
+	}
+
 	mdev = (struct erofs_map_dev) {
 		.m_deviceid = map.m_deviceid,
 		.m_pa = map.m_pa,
Signature · 1 line
2.27.0
Reviewed-by on v9 19/21
#42 lore raw

Re: [PATCH v9 19/21] erofs: implement fscache-based data read for inline layout

Quoted 7 lines · Jeffle Xu

Implement the data plane of reading data from data blobs over fscache for inline layout.

For the heading non-inline part, the data plane for non-inline layout is reused, while only the tail packing part needs special handling.

Thanks,
Gao Xiang

Quoted 56 lines
---
 fs/erofs/fscache.c | 32 ++++++++++++++++++++++++++++++++
 1 file changed, 32 insertions(+)
fs/erofs/fscache.c +32 -0
index b799b0fe1b67..08849c15500f 100644
@@ -84,6 +84,33 @@ static int erofs_fscache_meta_readpage(struct file *data, struct page *page)
 	return ret;
 }
 
+static int erofs_fscache_readpage_inline(struct folio *folio,
+					 struct erofs_map_blocks *map)
+{
+	struct super_block *sb = folio_mapping(folio)->host->i_sb;
+	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
+	erofs_blk_t blknr;
+	size_t offset, len;
+	void *src, *dst;
+
+	/* For tail packing layout, the offset may be non-zero. */
+	offset = erofs_blkoff(map->m_pa);
+	blknr = erofs_blknr(map->m_pa);
+	len = map->m_llen;
+
+	src = erofs_read_metabuf(&buf, sb, blknr, EROFS_KMAP);
+	if (IS_ERR(src))
+		return PTR_ERR(src);
+
+	dst = kmap_local_folio(folio, 0);
+	memcpy(dst, src + offset, len);
+	memset(dst + len, 0, PAGE_SIZE - len);
+	kunmap_local(dst);
+
+	erofs_put_metabuf(&buf);
+	return 0;
+}
+
 static int erofs_fscache_readpage(struct file *file, struct page *page)
 {
 	struct folio *folio = page_folio(page);
@@ -109,6 +136,11 @@ static int erofs_fscache_readpage(struct file *file, struct page *page)
 		goto out_uptodate;
 	}
 
+	if (map.m_flags & EROFS_MAP_META) {
+		ret = erofs_fscache_readpage_inline(folio, &map);
+		goto out_uptodate;
+	}
+
 	mdev = (struct erofs_map_dev) {
 		.m_deviceid = map.m_deviceid,
 		.m_pa = map.m_pa,
Signature · 1 line
2.27.0
author
#43 lore raw

PATCH v9 20/21 erofs: implement fscache-based data readahead

In reply to Jeffle Xu

Implement fscache-based data readahead. Also registers an individual bdi for each erofs instance to enable readahead.

---
 fs/erofs/fscache.c | 86 ++++++++++++++++++++++++++++++++++++++++++++++
 fs/erofs/super.c   |  4 +++
 2 files changed, 90 insertions(+)
fs/erofs/fscache.c +86 -0
index 08849c15500f..eaa50692ddba 100644
@@ -163,12 +163,98 @@ static int erofs_fscache_readpage(struct file *file, struct page *page)
 	return ret;
 }
 
+static void erofs_fscache_unlock_folios(struct readahead_control *rac,
+					size_t len)
+{
+	while (len) {
+		struct folio *folio = readahead_folio(rac);
+
+		len -= folio_size(folio);
+		folio_mark_uptodate(folio);
+		folio_unlock(folio);
+	}
+}
+
+static void erofs_fscache_readahead(struct readahead_control *rac)
+{
+	struct inode *inode = rac->mapping->host;
+	struct super_block *sb = inode->i_sb;
+	size_t len, count, done = 0;
+	erofs_off_t pos;
+	loff_t start, offset;
+	int ret;
+
+	if (!readahead_count(rac))
+		return;
+
+	start = readahead_pos(rac);
+	len = readahead_length(rac);
+
+	do {
+		struct erofs_map_blocks map;
+		struct erofs_map_dev mdev;
+
+		pos = start + done;
+		map.m_la = pos;
+
+		ret = erofs_map_blocks(inode, &map, EROFS_GET_BLOCKS_RAW);
+		if (ret)
+			return;
+
+		offset = start + done;
+		count = min_t(size_t, map.m_llen - (pos - map.m_la),
+			      len - done);
+
+		if (!(map.m_flags & EROFS_MAP_MAPPED)) {
+			struct iov_iter iter;
+
+			iov_iter_xarray(&iter, READ, &rac->mapping->i_pages,
+					offset, count);
+			iov_iter_zero(count, &iter);
+
+			erofs_fscache_unlock_folios(rac, count);
+			ret = count;
+			continue;
+		}
+
+		if (map.m_flags & EROFS_MAP_META) {
+			struct folio *folio = readahead_folio(rac);
+
+			ret = erofs_fscache_readpage_inline(folio, &map);
+			if (!ret) {
+				folio_mark_uptodate(folio);
+				ret = folio_size(folio);
+			}
+
+			folio_unlock(folio);
+			continue;
+		}
+
+		mdev = (struct erofs_map_dev) {
+			.m_deviceid = map.m_deviceid,
+			.m_pa = map.m_pa,
+		};
+		ret = erofs_map_dev(sb, &mdev);
+		if (ret)
+			return;
+
+		ret = erofs_fscache_read_folios(mdev.m_fscache->cookie,
+				rac->mapping, offset, count,
+				mdev.m_pa + (pos - map.m_la));
+		if (!ret) {
+			erofs_fscache_unlock_folios(rac, count);
+			ret = count;
+		}
+	} while (ret > 0 && ((done += ret) < len));
+}
+
 static const struct address_space_operations erofs_fscache_meta_aops = {
 	.readpage = erofs_fscache_meta_readpage,
 };
 
 const struct address_space_operations erofs_fscache_access_aops = {
 	.readpage = erofs_fscache_readpage,
+	.readahead = erofs_fscache_readahead,
 };
 
 /*
fs/erofs/super.c +4 -0
index c6755bcae4a6..f68ba929100d 100644
@@ -619,6 +619,10 @@ static int erofs_fc_fill_super(struct super_block *sb, struct fs_context *fc)
 						    sbi->opt.fsid, true);
 		if (err)
 			return err;
+
+		err = super_setup_bdi(sb);
+		if (err)
+			return err;
 	} else {
 		if (!sb_set_blocksize(sb, EROFS_BLKSIZ)) {
 			erofs_err(sb, "failed to set erofs blksize");
Signature · 1 line
2.27.0
#44 lore raw

Re: [PATCH v9 20/21] erofs: implement fscache-based data readahead

Quoted 89 lines · Jeffle Xu

Implement fscache-based data readahead. Also registers an individual bdi for each erofs instance to enable readahead.

---
 fs/erofs/fscache.c | 86 ++++++++++++++++++++++++++++++++++++++++++++++
 fs/erofs/super.c   |  4 +++
 2 files changed, 90 insertions(+)
fs/erofs/fscache.c +72 -0
index 08849c15500f..eaa50692ddba 100644
@@ -163,12 +163,98 @@ static int erofs_fscache_readpage(struct file *file, struct page *page)
 	return ret;
 }
 
+static void erofs_fscache_unlock_folios(struct readahead_control *rac,
+					size_t len)
+{
+	while (len) {
+		struct folio *folio = readahead_folio(rac);
+
+		len -= folio_size(folio);
+		folio_mark_uptodate(folio);
+		folio_unlock(folio);
+	}
+}
+
+static void erofs_fscache_readahead(struct readahead_control *rac)
+{
+	struct inode *inode = rac->mapping->host;
+	struct super_block *sb = inode->i_sb;
+	size_t len, count, done = 0;
+	erofs_off_t pos;
+	loff_t start, offset;
+	int ret;
+
+	if (!readahead_count(rac))
+		return;
+
+	start = readahead_pos(rac);
+	len = readahead_length(rac);
+
+	do {
+		struct erofs_map_blocks map;
+		struct erofs_map_dev mdev;
+
+		pos = start + done;
+		map.m_la = pos;
+
+		ret = erofs_map_blocks(inode, &map, EROFS_GET_BLOCKS_RAW);
+		if (ret)
+			return;
+
+		offset = start + done;
+		count = min_t(size_t, map.m_llen - (pos - map.m_la),
+			      len - done);
+
+		if (!(map.m_flags & EROFS_MAP_MAPPED)) {
+			struct iov_iter iter;
+
+			iov_iter_xarray(&iter, READ, &rac->mapping->i_pages,
+					offset, count);
+			iov_iter_zero(count, &iter);
+
+			erofs_fscache_unlock_folios(rac, count);
+			ret = count;
+			continue;
+		}
+
+		if (map.m_flags & EROFS_MAP_META) {
+			struct folio *folio = readahead_folio(rac);
+
+			ret = erofs_fscache_readpage_inline(folio, &map);
+			if (!ret) {
+				folio_mark_uptodate(folio);
+				ret = folio_size(folio);
+			}
+
+			folio_unlock(folio);
+			continue;
+		}
+
+		mdev = (struct erofs_map_dev) {
+			.m_deviceid = map.m_deviceid,
+			.m_pa = map.m_pa,
+		};
+		ret = erofs_map_dev(sb, &mdev);

fs/erofs/fscache.c static int erofs_fscache_readpage(struct file *file, struct page *page) · line 247

Quoted 10 lines
fs/erofs/fscache.c +10 -0
+		if (ret)
+			return;
+
+		ret = erofs_fscache_read_folios(mdev.m_fscache->cookie,
+				rac->mapping, offset, count,
+				mdev.m_pa + (pos - map.m_la));
+		if (!ret) {
+			erofs_fscache_unlock_folios(rac, count);
+			ret = count;
+		}

I think this really needs a comment why we don't need to unlock folios for the error cases.

Thanks,
Gao Xiang

Quoted 3 lines
fs/erofs/fscache.c +3 -0
+	} while (ret > 0 && ((done += ret) < len));
+}
+
author
#45 lore raw

PATCH v9 21/21 erofs: add 'fsid' mount option

In reply to Jeffle Xu

Introduce 'fsid' mount option to enable on-demand read sementics, in which case, erofs will be mounted from data blobs. Users could specify the name of primary data blob by this mount option.

---
 fs/erofs/super.c | 31 ++++++++++++++++++++++++++++++-
 fs/erofs/sysfs.c |  4 ++--
 2 files changed, 32 insertions(+), 3 deletions(-)
fs/erofs/super.c +30 -1
index f68ba929100d..4a623630e1c4 100644
@@ -371,6 +371,8 @@ static int erofs_read_superblock(struct super_block *sb)
 
 	if (erofs_sb_has_ztailpacking(sbi))
 		erofs_info(sb, "EXPERIMENTAL compressed inline data feature in use. Use at your own risk!");
+	if (erofs_is_fscache_mode(sb))
+		erofs_info(sb, "EXPERIMENTAL fscache-based on-demand read feature in use. Use at your own risk!");
 out:
 	erofs_put_metabuf(&buf);
 	return ret;
@@ -399,6 +401,7 @@ enum {
 	Opt_dax,
 	Opt_dax_enum,
 	Opt_device,
+	Opt_fsid,
 	Opt_err
 };
 
@@ -423,6 +426,7 @@ static const struct fs_parameter_spec erofs_fs_parameters[] = {
 	fsparam_flag("dax",             Opt_dax),
 	fsparam_enum("dax",		Opt_dax_enum, erofs_dax_param_enums),
 	fsparam_string("device",	Opt_device),
+	fsparam_string("fsid",		Opt_fsid),
 	{}
 };
 
@@ -518,6 +522,16 @@ static int erofs_fc_parse_param(struct fs_context *fc,
 		}
 		++ctx->devs->extra_devices;
 		break;
+	case Opt_fsid:
+#ifdef CONFIG_EROFS_FS_ONDEMAND
+		kfree(ctx->opt.fsid);
+		ctx->opt.fsid = kstrdup(param->string, GFP_KERNEL);
+		if (!ctx->opt.fsid)
+			return -ENOMEM;
+#else
+		errorfc(fc, "fsid option not supported");
+#endif
+		break;
 	default:
 		return -ENOPARAM;
 	}
@@ -604,6 +618,7 @@ static int erofs_fc_fill_super(struct super_block *sb, struct fs_context *fc)
 
 	sb->s_fs_info = sbi;
 	sbi->opt = ctx->opt;
+	ctx->opt.fsid = NULL;
 	sbi->devs = ctx->devs;
 	ctx->devs = NULL;
 
@@ -690,6 +705,11 @@ static int erofs_fc_fill_super(struct super_block *sb, struct fs_context *fc)
 
 static int erofs_fc_get_tree(struct fs_context *fc)
 {
+	struct erofs_fs_context *ctx = fc->fs_private;
+
+	if (IS_ENABLED(CONFIG_EROFS_FS_ONDEMAND) && ctx->opt.fsid)
+		return get_tree_nodev(fc, erofs_fc_fill_super);
+
 	return get_tree_bdev(fc, erofs_fc_fill_super);
 }
 
@@ -739,6 +759,7 @@ static void erofs_fc_free(struct fs_context *fc)
 	struct erofs_fs_context *ctx = fc->fs_private;
 
 	erofs_free_dev_context(ctx->devs);
+	kfree(ctx->opt.fsid);
 	kfree(ctx);
 }
 
@@ -779,7 +800,10 @@ static void erofs_kill_sb(struct super_block *sb)
 
 	WARN_ON(sb->s_magic != EROFS_SUPER_MAGIC);
 
-	kill_block_super(sb);
+	if (erofs_is_fscache_mode(sb))
+		generic_shutdown_super(sb);
+	else
+		kill_block_super(sb);
 
 	sbi = EROFS_SB(sb);
 	if (!sbi)
@@ -789,6 +813,7 @@ static void erofs_kill_sb(struct super_block *sb)
 	fs_put_dax(sbi->dax_dev);
 	erofs_fscache_unregister_cookie(&sbi->s_fscache);
 	erofs_fscache_unregister_fs(sb);
+	kfree(sbi->opt.fsid);
 	kfree(sbi);
 	sb->s_fs_info = NULL;
 }
@@ -938,6 +963,10 @@ static int erofs_show_options(struct seq_file *seq, struct dentry *root)
 		seq_puts(seq, ",dax=always");
 	if (test_opt(opt, DAX_NEVER))
 		seq_puts(seq, ",dax=never");
+#ifdef CONFIG_EROFS_FS_ONDEMAND
+	if (opt->fsid)
+		seq_printf(seq, ",fsid=%s", opt->fsid);
+#endif
 	return 0;
 }
 
fs/erofs/sysfs.c +2 -2
index f3babf1e6608..c1383e508bbe 100644
@@ -205,8 +205,8 @@ int erofs_register_sysfs(struct super_block *sb)
 
 	sbi->s_kobj.kset = &erofs_root;
 	init_completion(&sbi->s_kobj_unregister);
-	err = kobject_init_and_add(&sbi->s_kobj, &erofs_sb_ktype, NULL,
-				   "%s", sb->s_id);
+	err = kobject_init_and_add(&sbi->s_kobj, &erofs_sb_ktype, NULL, "%s",
+			erofs_is_fscache_mode(sb) ? sbi->opt.fsid : sb->s_id);
 	if (err)
 		goto put_sb_kobj;
 	return 0;
Signature · 1 line
2.27.0
Reviewed-by on v9 21/21
#46 lore raw

Re: [PATCH v9 21/21] erofs: add 'fsid' mount option

Quoted 5 lines · Jeffle Xu

Introduce 'fsid' mount option to enable on-demand read sementics, in which case, erofs will be mounted from data blobs. Users could specify the name of primary data blob by this mount option.

Thanks,
Gao Xiang

Quoted 127 lines
---
 fs/erofs/super.c | 31 ++++++++++++++++++++++++++++++-
 fs/erofs/sysfs.c |  4 ++--
 2 files changed, 32 insertions(+), 3 deletions(-)
fs/erofs/super.c +30 -1
index f68ba929100d..4a623630e1c4 100644
@@ -371,6 +371,8 @@ static int erofs_read_superblock(struct super_block *sb)
 
 	if (erofs_sb_has_ztailpacking(sbi))
 		erofs_info(sb, "EXPERIMENTAL compressed inline data feature in use. Use at your own risk!");
+	if (erofs_is_fscache_mode(sb))
+		erofs_info(sb, "EXPERIMENTAL fscache-based on-demand read feature in use. Use at your own risk!");
 out:
 	erofs_put_metabuf(&buf);
 	return ret;
@@ -399,6 +401,7 @@ enum {
 	Opt_dax,
 	Opt_dax_enum,
 	Opt_device,
+	Opt_fsid,
 	Opt_err
 };
 
@@ -423,6 +426,7 @@ static const struct fs_parameter_spec erofs_fs_parameters[] = {
 	fsparam_flag("dax",             Opt_dax),
 	fsparam_enum("dax",		Opt_dax_enum, erofs_dax_param_enums),
 	fsparam_string("device",	Opt_device),
+	fsparam_string("fsid",		Opt_fsid),
 	{}
 };
 
@@ -518,6 +522,16 @@ static int erofs_fc_parse_param(struct fs_context *fc,
 		}
 		++ctx->devs->extra_devices;
 		break;
+	case Opt_fsid:
+#ifdef CONFIG_EROFS_FS_ONDEMAND
+		kfree(ctx->opt.fsid);
+		ctx->opt.fsid = kstrdup(param->string, GFP_KERNEL);
+		if (!ctx->opt.fsid)
+			return -ENOMEM;
+#else
+		errorfc(fc, "fsid option not supported");
+#endif
+		break;
 	default:
 		return -ENOPARAM;
 	}
@@ -604,6 +618,7 @@ static int erofs_fc_fill_super(struct super_block *sb, struct fs_context *fc)
 
 	sb->s_fs_info = sbi;
 	sbi->opt = ctx->opt;
+	ctx->opt.fsid = NULL;
 	sbi->devs = ctx->devs;
 	ctx->devs = NULL;
 
@@ -690,6 +705,11 @@ static int erofs_fc_fill_super(struct super_block *sb, struct fs_context *fc)
 
 static int erofs_fc_get_tree(struct fs_context *fc)
 {
+	struct erofs_fs_context *ctx = fc->fs_private;
+
+	if (IS_ENABLED(CONFIG_EROFS_FS_ONDEMAND) && ctx->opt.fsid)
+		return get_tree_nodev(fc, erofs_fc_fill_super);
+
 	return get_tree_bdev(fc, erofs_fc_fill_super);
 }
 
@@ -739,6 +759,7 @@ static void erofs_fc_free(struct fs_context *fc)
 	struct erofs_fs_context *ctx = fc->fs_private;
 
 	erofs_free_dev_context(ctx->devs);
+	kfree(ctx->opt.fsid);
 	kfree(ctx);
 }
 
@@ -779,7 +800,10 @@ static void erofs_kill_sb(struct super_block *sb)
 
 	WARN_ON(sb->s_magic != EROFS_SUPER_MAGIC);
 
-	kill_block_super(sb);
+	if (erofs_is_fscache_mode(sb))
+		generic_shutdown_super(sb);
+	else
+		kill_block_super(sb);
 
 	sbi = EROFS_SB(sb);
 	if (!sbi)
@@ -789,6 +813,7 @@ static void erofs_kill_sb(struct super_block *sb)
 	fs_put_dax(sbi->dax_dev);
 	erofs_fscache_unregister_cookie(&sbi->s_fscache);
 	erofs_fscache_unregister_fs(sb);
+	kfree(sbi->opt.fsid);
 	kfree(sbi);
 	sb->s_fs_info = NULL;
 }
@@ -938,6 +963,10 @@ static int erofs_show_options(struct seq_file *seq, struct dentry *root)
 		seq_puts(seq, ",dax=always");
 	if (test_opt(opt, DAX_NEVER))
 		seq_puts(seq, ",dax=never");
+#ifdef CONFIG_EROFS_FS_ONDEMAND
+	if (opt->fsid)
+		seq_printf(seq, ",fsid=%s", opt->fsid);
+#endif
 	return 0;
 }
 
fs/erofs/sysfs.c +2 -2
index f3babf1e6608..c1383e508bbe 100644
@@ -205,8 +205,8 @@ int erofs_register_sysfs(struct super_block *sb)
 
 	sbi->s_kobj.kset = &erofs_root;
 	init_completion(&sbi->s_kobj_unregister);
-	err = kobject_init_and_add(&sbi->s_kobj, &erofs_sb_ktype, NULL,
-				   "%s", sb->s_id);
+	err = kobject_init_and_add(&sbi->s_kobj, &erofs_sb_ktype, NULL, "%s",
+			erofs_is_fscache_mode(sb) ? sbi->opt.fsid : sb->s_id);
 	if (err)
 		goto put_sb_kobj;
 	return 0;
Signature · 1 line
2.27.0
Tested-by on the v9 series
#47 lore raw

Re: [PATCH v9 00/21] fscache, erofs: fscache-based on-demand read semantics

In reply to Jeffle Xu
Quoted 236 lines · Jeffle Xu

changes since v8:
- rebase to 5.18-rc2
- cachefiles: use object_id rather than anon_fd to uniquely identify a cachefile object to avoid potential issues when the user moves the anonymous fd around, e.g. through dup() (refer to commit message and cachefiles_ondemand_get_fd() of patch 2 for more details) (David Howells)
- cachefiles: add @unbind_pincount refcount to avoid the potential deadlock (refer to commit message of patch3 for more details)
- cachefiles: move the calling site of cachefiles_ondemand_read() from cachefiles_read() to cacehfiles_prep_read() (refer to commit message of patch 5 for more details)
- cachefiles: add tracepoints (patch 7) (David Howells)
- cachefiles: update documentation (patch 8) (David Howells)
- erofs: update Reviewed-by tag from Gao Xiang
- erofs: move the logic of initializing bdev/dax_dev in fscache mode out from patch 15/20. Instead move it into patch 9, so that patch 20 can focus on the mount option handling
- erofs: update the subject line and commit message of patch 12 (Gao Xiang)
- erofs: remove and fold erofs_fscache_get_folio() helper (patch 16) (Gao Xiang)
- erofs: change kmap() to kamp_loacl_folio(), and comment cleanup (patch
18) (Gao Xiang)
- update "advantage of fscache-based on-demand read" section of the cover letter
- we've finished a preliminary end-to-end on-demand download daemon in order to test the fscache on-demand kernel code as a real end-to-end workload for container use cases. The test user guide is added in the cover letter.
- Thanks Zichen Tian for testing

Kernel Patchset
---------------
Git tree:

https://github.com/lostjeffle/linux.git jingbo/dev-erofs-fscache-v9

Gitweb:

https://github.com/lostjeffle/linux/commits/jingbo/dev-erofs-fscache-v9

User Guide for E2E Container Use Case
-------------------------------------
User guide:

https://github.com/dragonflyoss/image-service/blob/fscache/docs/nydus-fscache.md

Video:

     https://youtu.be/F4IF2_DENXo

User Daemon for Quick Test
--------------------------
Git tree:

https://github.com/lostjeffle/demand-read-cachefilesd.git main

Gitweb:

https://github.com/lostjeffle/demand-read-cachefilesd

RFC: https://lore.kernel.org/all/[email protected]/t/
v1: https://lore.kernel.org/lkml/[email protected]/T/
v2: https://lore.kernel.org/all/[email protected]/t/
v3: https://lore.kernel.org/lkml/[email protected]/T/
v4: https://lore.kernel.org/lkml/[email protected]/T/#t
v5: https://lore.kernel.org/lkml/[email protected]/T/
v6: https://lore.kernel.org/lkml/[email protected]/T/
v7: https://lore.kernel.org/lkml/[email protected]/T/
v8: https://lore.kernel.org/all/[email protected]/T/

[Background]
============
Nydus [1] is an image distribution service especially optimized for distribution over network. Nydus is an excellent container image acceleration solution, since it only pulls data from remote when needed, a.k.a. on-demand reading and it also supports chunk-based deduplication, compression, etc.

erofs (Enhanced Read-Only File System) is a filesystem designed for read-only scenarios. (Documentation/filesystem/erofs.rst)

Over the past months we've been focusing on supporting Nydus image service with in-kernel erofs format[2]. In that case, each container image will be organized in one bootstrap (metadata) and (optional) multiple data blobs in erofs format. Massive container images will be stored on one machine.

To accelerate the container startup (fetching container images from remote and then start the container), we do hope that the bootstrap & blob files could support on-demand read. That is, erofs can be mounted and accessed even when the bootstrap/data blob files have not been fully downloaded. Then it'll have native performance after data is available locally.

That means we have to manage the cache state of the bootstrap/data blob files (if cache hit, read directly from the local cache; if cache miss, fetch the data somehow). It would be painful and may be dumb for erofs to implement the cache management itself. Thus we prefer fscache/cachefiles to do the cache management instead.

The fscache on-demand read feature aims to be implemented in a generic way so that it can benefit other use cases and/or filesystems if it's implemented in the fscache subsystem.

[1] https://nydus.dev
[2] https://sched.co/pcdL

[Overall Design]
================
Please refer to patch 7 ("cachefiles: document on-demand read mode") for more details.

When working in the original mode, cachefiles mainly serves as a local cache for remote networking fs, while in on-demand read mode, cachefiles can work in the scenario where on-demand read semantics is needed, e.g. container image distribution.

The essential difference between these two modes is that, in original mode, when cache miss, netfs itself will fetch data from remote, and then write the fetched data into cache file. While in on-demand read mode, a user daemon is responsible for fetching data and then feeds to the kernel fscache side.

The on-demand read mode relies on a simple protocol used for communication between kernel and user daemon.

The proposed implementation relies on the anonymous fd mechanism to avoid the dependence on the format of cache file. When a fscache cachefile is opened for the first time, an anon_fd associated with the cache file is sent to the user daemon. With the given anon_fd, user daemon could fetch and write data into the cache file in the background, even when kernel has not triggered the cache miss. Besides, the write() syscall to the anon_fd will finally call cachefiles kernel module, which will write data to cache file in the latest format of cache file.

1. cache miss
When cache miss, cachefiles kernel module will notify user daemon with the anon_fd, along with the requested file range. When notified, user daemon needs to fetch data of the requested file range, and then write the fetched data into cache file with the given anonymous fd. When finished processing the request, user daemon needs to notify the kernel.

After notifying the user daemon, the kernel read routine will hang there, until the request is handled by user daemon. When it's awaken by the notification from user daemon, i.e. the corresponding hole has been filled by the user daemon, it will retry to read from the same file range.

2. cache hit
Once data is already ready in cache file, netfs will read from cache file directly.

[Advantage of fscache-based on-demand read]
========================================
1. Asynchronous prefetch
In current mechanism, fscache is responsible for cache state management, while the data plane (fetching data from local/remote on cache miss) is done on the user daemon side even without any file system request driven. In addition, if cached data has already been available locally, fscache will use it instead of trapping to user space anymore.

Therefore, different from event-driven approaches, the fscache on-demand user daemon could also fetch data (from remote) asynchronously in the background just like most multi-threaded HTTP downloaders.

2. Flexible request amplification
Since the data plane can be independently controlled by the user daemon, the user daemon can also fetch more data from remote than that the file system actually requests for small I/O sizes. Then, fetched data in bulk will be available at once and fscache won't be trapped into the user daemon again.

3. Support massive blobs
This mechanism can naturally support a large amount of backing files, and thus can benefit the densely employed scenarios. In our use cases, one container image can be formed of one bootstrap (required) and multiple chunk-deduplicated data blobs (optional).

For example, one container image for node.js will correspond to ~20 files in total. In densely employed environment, there could be hundreds of containers and thus thousands of backing files on one machine.

Jeffle Xu (21):
cachefiles: extract write routine
cachefiles: notify user daemon when looking up cookie
cachefiles: unbind cachefiles gracefully in on-demand mode
cachefiles: notify user daemon when withdrawing cookie
cachefiles: implement on-demand read
cachefiles: enable on-demand read mode
cachefiles: add tracepoints for on-demand read mode
cachefiles: document on-demand read mode
erofs: make erofs_map_blocks() generally available
erofs: add fscache mode check helper
erofs: register fscache volume
erofs: add fscache context helper functions
erofs: add anonymous inode caching metadata for data blobs
erofs: add erofs_fscache_read_folios() helper
erofs: register fscache context for primary data blob
erofs: register fscache context for extra data blobs
erofs: implement fscache-based metadata read
erofs: implement fscache-based data read for non-inline layout
erofs: implement fscache-based data read for inline layout
erofs: implement fscache-based data readahead
erofs: add 'fsid' mount option

  .../filesystems/caching/cachefiles.rst        | 170 ++++++
  fs/cachefiles/Kconfig                         |  11 +
  fs/cachefiles/Makefile                        |   1 +
  fs/cachefiles/daemon.c                        | 116 +++-
  fs/cachefiles/interface.c                     |   2 +
  fs/cachefiles/internal.h                      |  74 +++
  fs/cachefiles/io.c                            |  76 ++-
  fs/cachefiles/namei.c                         |  16 +-
  fs/cachefiles/ondemand.c                      | 496 ++++++++++++++++++
  fs/erofs/Kconfig                              |  10 +
  fs/erofs/Makefile                             |   1 +
  fs/erofs/data.c                               |  26 +-
  fs/erofs/fscache.c                            | 365 +++++++++++++
  fs/erofs/inode.c                              |   4 +
  fs/erofs/internal.h                           |  49 ++
  fs/erofs/super.c                              | 105 +++-
  fs/erofs/sysfs.c                              |   4 +-
  include/linux/fscache.h                       |   1 +
  include/linux/netfs.h                         |   2 +
  include/trace/events/cachefiles.h             | 176 +++++++
  include/uapi/linux/cachefiles.h               |  68 +++
  21 files changed, 1694 insertions(+), 79 deletions(-)
Quoted 4 lines

create mode 100644 fs/cachefiles/ondemand.c create mode 100644 fs/erofs/fscache.c create mode 100644 include/uapi/linux/cachefiles.h

Hi Jeffle & Xiang,

Thanks for coming up with such an innovative solution. We interested in this and want to deploy it in our system. So we have performed the tests by user guide and did some error injection tests using User Daemon Demo offered by Jeffle. Hope it can be an upstream feature.

Thanks,
Jia