put the xfs xfile abstraction on a diet v3
Subsystems (9)
Hi all,
this series refactors and simplifies the code in the xfs xfile abstraction, which is a thing layer on a kernel-use shmem file.
To do this is needs a slighly lower level exports from shmem.c, which I combined with improving an assert and documentation there.
Now that this series has a fair amount of reviews we also need to think how we can merge it best. Should the shmem changes be merged through the xfs tree? Should there be a shared branch?
The series is against Linux 6.8-rc2.
Changes since v2:
- include internal.h in workingset.h to avoid a buildbot warning
- update the usage comment for shmem_get_folio
- improve the xfile_load/store documentation
- don't use highmem for xfiles for real
- only call filemap_check_wb_err to check for errors on the mapping
Changes since v1:
- fix reading i_blocks
- provide wrappers for reading i_size and i_blocks
- update the xfile load/store documentation
- improve a commit message
- use shmem_kernel_file_setup
- add a missing folio unlock in the hwpoison path in xfile_get_page
- fix checking for shmem mappings
- improve I/O error handling (Darrick)
- convert to folios (partially from Darrick)
Diffstat:
Documentation/filesystems/xfs/xfs-online-fsck-design.rst | 25 - fs/xfs/scrub/rtsummary.c | 6 fs/xfs/scrub/trace.h | 81 ++- fs/xfs/scrub/xfarray.c | 234 ++++------ fs/xfs/scrub/xfarray.h | 11 fs/xfs/scrub/xfile.c | 345 +++++---------- fs/xfs/scrub/xfile.h | 62 -- include/linux/shmem_fs.h | 6 include/linux/swap.h | 10 mm/filemap.c | 9 mm/internal.h | 4 mm/shmem.c | 37 + mm/workingset.c | 1 13 files changed, 349 insertions(+), 482 deletions(-)
PATCH 1/20 mm: move mapping_set_update out of <linux/swap.h>
mapping_set_update is only used inside mm/. Move mapping_set_update to mm/internal.h and turn it into an inline function instead of a macro.
--- include/linux/swap.h | 10 ---------- mm/filemap.c | 9 +++++++++ mm/internal.h | 4 ++++ mm/workingset.c | 1 + 4 files changed, 14 insertions(+), 10 deletions(-)
include/linux/swap.h +0 -10
@@ -350,16 +350,6 @@ void *workingset_eviction(struct folio *folio, struct mem_cgroup *target_memcg);void workingset_refault (struct folio *folio,void *shadow);void workingset_activation (struct folio *folio); -/* Only track the nodes of mappings with shadow entries */ -void workingset_update_node (struct xa_node *node); -extern struct list_lru shadow_nodes; -#define mapping_set_update (xas, mapping) do { \ - if (!dax_mapping(mapping) && !shmem_mapping(mapping)) { \ - xas_set_update(xas, workingset_update_node); \ - xas_set_lru(xas, &shadow_nodes); \ - } \ -} while (0) -/* linux/mm/page_alloc.c */ extern unsigned long totalreserve_pages;
mm/filemap.c +9 -0
@@ -124,6 +124,15 @@* ->private_lock (zap_pte_range->block_dirty_folio) */ +static void mapping_set_update (struct xa_state *xas, +struct address_space *mapping) +{ +if (dax_mapping (mapping) ||shmem_mapping (mapping)) +return ; +xas_set_update (xas, workingset_update_node); +xas_set_lru (xas, &shadow_nodes); +} +static void page_cache_delete (struct address_space *mapping,struct folio *folio,void *shadow) {
mm/internal.h +4 -0
@@ -1266,4 +1266,8 @@ static inline void shrinker_debugfs_remove(struct dentry *debugfs_entry, }#endif /* CONFIG_SHRINKER_DEBUG */ +/* Only track the nodes of mappings with shadow entries */ +void workingset_update_node (struct xa_node *node); +extern struct list_lru shadow_nodes; +#endif /* __MM_INTERNAL_H */
mm/workingset.c +1 -0
@@ -16,6 +16,7 @@#include <linux/dax.h> #include <linux/fs.h> #include <linux/mm.h> +#include "internal.h" /* * Double CLOCK lists
Signature · 1 line
2.39.2
PATCH 2/20 shmem: move shmem_mapping out of line
In reply to Christoph Hellwigshmem_aops really should not be exported to the world. Move shmem_mapping and export it as internal for the one semi-legitimate modular user in udmabuf.
--- include/linux/shmem_fs.h | 6 +----- mm/shmem.c | 11 ++++++++--- 2 files changed, 9 insertions(+), 8 deletions(-)
include/linux/shmem_fs.h +1 -5
@@ -97,11 +97,7 @@ extern unsigned long shmem_get_unmapped_area(struct file *, unsigned long addr, unsigned long len, unsigned long pgoff, unsigned long flags);extern int shmem_lock (struct file *file,int lock,struct ucounts *ucounts);#ifdef CONFIG_SHMEM -extern const struct address_space_operations shmem_aops; -static inline bool shmem_mapping (struct address_space *mapping) -{ -return mapping->a_ops == &shmem_aops; -} +bool shmem_mapping (struct address_space *mapping);#else static inline bool shmem_mapping (struct address_space *mapping) {
mm/shmem.c +8 -3
@@ -254,7 +254,7 @@ static void shmem_inode_unacct_blocks(struct inode *inode, long pages) }static const struct super_operations shmem_ops; -const struct address_space_operations shmem_aops; +static const struct address_space_operations shmem_aops;static const struct file_operations shmem_file_operations;static const struct inode_operations shmem_inode_operations;static const struct inode_operations shmem_dir_inode_operations;
@@ -263,6 +263,12 @@ static const struct vm_operations_struct shmem_vm_ops;static const struct vm_operations_struct shmem_anon_vm_ops;static struct file_system_type shmem_fs_type; +bool shmem_mapping (struct address_space *mapping) +{ +return mapping->a_ops == &shmem_aops; +} +EXPORT_SYMBOL_GPL (shmem_mapping); +bool vma_is_anon_shmem (struct vm_area_struct *vma) {return vma->vm_ops == &shmem_anon_vm_ops;
@@ -4466,7 +4472,7 @@ static int shmem_error_remove_folio(struct address_space *mapping,return 0 ; } -const struct address_space_operations shmem_aops = { +static const struct address_space_operations shmem_aops = { .writepage = shmem_writepage, .dirty_folio = noop_dirty_folio, #ifdef CONFIG_TMPFS
@@ -4478,7 +4484,6 @@ const struct address_space_operations shmem_aops = {#endif .error_remove_folio = shmem_error_remove_folio, }; -EXPORT_SYMBOL (shmem_aops);static const struct file_operations shmem_file_operations = { .mmap = shmem_mmap,
Signature · 1 line
2.39.2
Hugh Dickins
maintainerRe: [PATCH 02/20] shmem: move shmem_mapping out of line
Quoted 6 lines · Christoph Hellwig
shmem_aopsreally should not be exported to the world. Moveshmem_mappingand export it as internal for the one semi-legitimate modular user in udmabuf.Signed-off-by Christoph Hellwig <[email protected]>Reviewed-by Matthew Wilcox (Oracle) <[email protected]>
I'm okay with this change, but let's be polite and Cc the author of the change which this reverts - v5.11's commit 30e6a51dbb05 ("mm/shmem.c: make shmem_mapping() inline").
Interesting to notice that commit did EXPORT_SYMBOL(shmem_aops), whereas shmem_mapping() had not been exported before; you're now exporting it _GPL(), like other examples in mm/shmem.c: agreed.
Hugh
Quoted 69 lines
--- include/linux/shmem_fs.h | 6 +----- mm/shmem.c | 11 ++++++++--- 2 files changed, 9 insertions(+), 8 deletions(-)include/linux/shmem_fs.h +1
-5@@ -97,11 +97,7 @@ extern unsigned long shmem_get_unmapped_area(struct file *, unsigned long addr, unsigned long len, unsigned long pgoff, unsigned long flags);extern int shmem_lock (struct file *file,int lock,struct ucounts *ucounts);#ifdef CONFIG_SHMEM -extern const struct address_space_operations shmem_aops; -static inline bool shmem_mapping (struct address_space *mapping) -{ -return mapping->a_ops == &shmem_aops; -} +bool shmem_mapping (struct address_space *mapping);#else static inline bool shmem_mapping (struct address_space *mapping) {mm/shmem.c +8
-3@@ -254,7 +254,7 @@ static void shmem_inode_unacct_blocks(struct inode *inode, long pages) }static const struct super_operations shmem_ops; -const struct address_space_operations shmem_aops; +static const struct address_space_operations shmem_aops;static const struct file_operations shmem_file_operations;static const struct inode_operations shmem_inode_operations;static const struct inode_operations shmem_dir_inode_operations;@@ -263,6 +263,12 @@ static const struct vm_operations_struct shmem_vm_ops;static const struct vm_operations_struct shmem_anon_vm_ops;static struct file_system_type shmem_fs_type; +bool shmem_mapping (struct address_space *mapping) +{ +return mapping->a_ops == &shmem_aops; +} +EXPORT_SYMBOL_GPL (shmem_mapping); +bool vma_is_anon_shmem (struct vm_area_struct *vma) {return vma->vm_ops == &shmem_anon_vm_ops;@@ -4466,7 +4472,7 @@ static int shmem_error_remove_folio(struct address_space *mapping,return 0 ; } -const struct address_space_operations shmem_aops = { +static const struct address_space_operations shmem_aops = { .writepage = shmem_writepage, .dirty_folio = noop_dirty_folio, #ifdef CONFIG_TMPFS@@ -4478,7 +4484,6 @@ const struct address_space_operations shmem_aops = {#endif .error_remove_folio = shmem_error_remove_folio, }; -EXPORT_SYMBOL (shmem_aops);static const struct file_operations shmem_file_operations = { .mmap = shmem_mmap,Signature · 1 line
2.39.2
PATCH 3/20 shmem: set a_ops earlier in shmem_symlink
In reply to Christoph HellwigSet the a_ops in shmem_symlink before reading a folio from the mapping to prepare for asserting that shmem_get_folio is only called on shmem mappings.
--- mm/shmem.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
mm/shmem.c +1 -1
@@ -3506,10 +3506,10 @@ static int shmem_symlink(struct mnt_idmap *idmap, struct inode *dir, inode->i_op = &shmem_short_symlink_operations; }else {inode_nohighmem (inode); + inode->i_mapping->a_ops = &shmem_aops; error =shmem_get_folio (inode,0 , &folio,SGP_WRITE );if (error) goto out_remove_offset; - inode->i_mapping->a_ops = &shmem_aops; inode->i_op = &shmem_symlink_inode_operations;memcpy (folio_address (folio), symname, len);folio_mark_uptodate (folio);
Signature · 1 line
2.39.2
PATCH 4/20 shmem: move the shmem_mapping assert into shmem_get_folio_gfp
In reply to Christoph HellwigMove the check that the inode really is a shmemfs one from shmem_read_folio_gfp to shmem_get_folio_gfp given that shmem_get_folio can also be called from outside of shmem.c. Also turn it into a WARN_ON_ONCE and error return instead of BUG_ON to be less severe.
--- mm/shmem.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
mm/shmem.c +3 -1
@@ -1972,6 +1972,9 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,int error;bool alloced; +if (WARN_ON_ONCE (!shmem_mapping (inode->i_mapping))) +return -EINVAL ; +if (index > (MAX_LFS_FILESIZE >>PAGE_SHIFT ))return -EFBIG ; repeat:
@@ -4915,7 +4918,6 @@ struct folio *shmem_read_folio_gfp(struct address_space *mapping,struct folio *folio;int error; -BUG_ON (!shmem_mapping (mapping)); error =shmem_get_folio_gfp (inode, index, &folio,SGP_CACHE , gfp,NULL ,NULL );if (error)
Signature · 1 line
2.39.2
PATCH 5/20 shmem: export shmem_get_folio
In reply to Christoph HellwigExport shmem_get_folio as a slightly lower-level variant of shmem_read_folio_gfp. This will be useful for XFS xfile use cases that want to pass SGP_NOALLOC or get a locked page, which the thin shmem_read_folio_gfp wrapper can't provide.
--- mm/shmem.c | 15 +++++++++++++++ 1 file changed, 15 insertions(+)
mm/shmem.c +15 -0
@@ -2137,12 +2137,27 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,return error; } +/** +* shmem_get_folio - find and get a reference to a shmem folio. +* @inode: inode to search +* @index: the page index. +* @foliop: pointer to the found folio if one was found +* @sgp: SGP_* flags to control behavior +* +* Looks up the page cache entry at @inode & @index. +* +* If this function returns a folio, it is returned with an increased refcount. +* +* Return: The found folio, %NULL if SGP_READ or SGP_NOALLOC was passed in @sgp +* and no folio was found at @index, or an ERR_PTR() otherwise. +*/ int shmem_get_folio (struct inode *inode,pgoff_t index,struct folio **foliop,enum sgp_type sgp) {return shmem_get_folio_gfp (inode, index, foliop, sgp,mapping_gfp_mask (inode->i_mapping),NULL ,NULL ); } +EXPORT_SYMBOL_GPL (shmem_get_folio);/* * This is like autoremove_wake_function, but it removes the wait queue
Signature · 1 line
2.39.2
Hugh Dickins
maintainerRe: [PATCH 05/20] shmem: export shmem_get_folio
Quoted 20 lines · Christoph Hellwig
Export
shmem_get_folioas a slightly lower-level variant ofshmem_read_folio_gfp. This will be useful for XFS xfile use cases that want to passSGP_NOALLOCor get a locked page, which the thinshmem_read_folio_gfpwrapper can't provide.Signed-off-by Christoph Hellwig <[email protected]>Reviewed-by Matthew Wilcox (Oracle) <[email protected]>Reviewed-by Darrick J. Wong <[email protected]>--- mm/shmem.c | 15 +++++++++++++++ 1 file changed, 15 insertions(+)mm/shmem.c +0
-0@@ -2137,12 +2137,27 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,return error; }
mm/shmem.c static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index, · line 2149
Quoted 10 lines
mm/shmem.c +10
-0+/** +* shmem_get_folio - find and get a reference to a shmem folio. +* @inode: inode to search +* @index: the page index. +* @foliop: pointer to the found folio if one was found +* @sgp: SGP_* flags to control behavior +* +* Looks up the page cache entry at @inode & @index. +* +* If this function returns a folio, it is returned with an increased refcount.
If this function returns a folio, it is returned locked, with an increased refcount.
(Important to mention that it's locked, since that differs from shmem_read_folio_gfp().)
Hugh
Quoted 16 lines
mm/shmem.c +5
-0+* +* Return: The found folio, %NULL if SGP_READ or SGP_NOALLOC was passed in @sgp +* and no folio was found at @index, or an ERR_PTR() otherwise. +*/ int shmem_get_folio (struct inode *inode,pgoff_t index,struct folio **foliop,enum sgp_type sgp) {return shmem_get_folio_gfp (inode, index, foliop, sgp,mapping_gfp_mask (inode->i_mapping),NULL ,NULL ); } +EXPORT_SYMBOL_GPL (shmem_get_folio);/* * This is like autoremove_wake_function, but it removes the wait queue Signature · 1 line
2.39.2
Matthew Wilcox
maintainerRe: [PATCH 05/20] shmem: export shmem_get_folio
In reply to Christoph HellwigQuoted 5 lines · Christoph Hellwig
+/**
+ *shmem_get_folio- find and get a reference to a shmem folio.
+ * @inode: inode to search
+ * @index: the page index.
+ * @foliop: pointer to the found folio if one was found
Quoted 8 lines
+ * @sgp: SGP_* flags to control behavior
+ *
+ * Looks up the page cache entry at @inode & @index.
+ *
+ * If this function returns a folio, it is returned with an increased refcount.
+ *
+ * Return: The found folio, %NULL ifSGP_READorSGP_NOALLOCwas passed in @sgp
+ * and no folio was found at @index, or anERR_PTR()otherwise.
I know I gave an R-b on this earlier, but Hugh made me look again, and this comment clearly does not reflect what the function does. Presumably it returns an errno and sets foliop if it returns 0?
Also, should this function be called shmem_lock_folio() to mirror filemap_lock_folio()?
Quoted 8 lines
+ */
int shmem_get_folio(struct inode *inode,pgoff_tindex, struct folio **foliop,mm/shmem.c +1
-0enum sgp_type sgp) {return shmem_get_folio_gfp (inode, index, foliop, sgp,mapping_gfp_mask (inode->i_mapping),NULL ,NULL ); } +EXPORT_SYMBOL_GPL (shmem_get_folio);
Re: [PATCH 05/20] shmem: export shmem_get_folio
Quoted 3 lines · Matthew Wilcox
I know I gave an R-b on this earlier, but Hugh made me look again, and this comment clearly does not reflect what the function does. Presumably it returns an errno and sets foliop if it returns 0?
Almost. With SGP_READ it can set *foliop to NULL and still return 0.
Quoted 2 lines
Also, should this function be called
shmem_lock_folio()to mirrorfilemap_lock_folio()?
shmem_get_folio can also allocate (and sometimes zero) a new folio. Except for the different calling conventions that closest filemap equivalent is __filemap_get_folio. For now I'd like to avoid the bikeshedding on the name and just get the work done.
PATCH 6/20 shmem: export shmem_kernel_file_setup
In reply to Christoph HellwigXFS wants to use this for it's internal in-memory data structures and currently duplicates the functionality. Export shmem_kernel_file_setup to allow XFS to switch over to using the proper kernel API.
--- mm/shmem.c | 1 + 1 file changed, 1 insertion(+)
mm/shmem.c +1 -0
@@ -4856,6 +4856,7 @@ struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned lon {return __shmem_file_setup (shm_mnt, name, size, flags,S_PRIVATE ); } +EXPORT_SYMBOL_GPL (shmem_kernel_file_setup);/** * shmem_file_setup - get an unlinked file living in tmpfs
Signature · 1 line
2.39.2
Re: [PATCH 06/20] shmem: export shmem_kernel_file_setup
Quoted 6 lines · Christoph Hellwig
XFS wants to use this for it's internal in-memory data structures and currently duplicates the functionality. Export
shmem_kernel_file_setupto allow XFS to switch over to using the proper kernel API.Signed-off-by Christoph Hellwig <[email protected]>Reviewed-by Matthew Wilcox (Oracle) <[email protected]>
Looks ok,
--D
Quoted 20 lines
--- mm/shmem.c | 1 + 1 file changed, 1 insertion(+)mm/shmem.c +1
-0@@ -4856,6 +4856,7 @@ struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned lon {return __shmem_file_setup (shm_mnt, name, size, flags,S_PRIVATE ); } +EXPORT_SYMBOL_GPL (shmem_kernel_file_setup);/** * shmem_file_setup - get an unlinked file living in tmpfs Signature · 1 line
2.39.2
PATCH 7/20 shmem: document how to "persist" data when using shmem_*file_setup
In reply to Christoph HellwigAdd a blurb that simply dirtying the folio will persist data for in-kernel shmem files. This is what most of the callers already do.
--- mm/shmem.c | 4 ++++ 1 file changed, 4 insertions(+)
mm/shmem.c +4 -0
@@ -2148,6 +2148,10 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,* * If this function returns a folio, it is returned with an increased refcount. * +* If the caller modifies data in the folio, it must call folio_mark_dirty() +* before unlocking the folio to ensure that the folio is not reclaimed. +* These is no need to reserve space before calling folio_mark_dirty(). +* * Return: The found folio, %NULL if SGP_READ or SGP_NOALLOC was passed in @sgp * and no folio was found at @index, or an ERR_PTR() otherwise. */
Signature · 1 line
2.39.2
Re: [PATCH 07/20] shmem: document how to "persist" data when using shmem_*file_setup
Quoted 5 lines · Christoph Hellwig
Add a blurb that simply dirtying the folio will persist data for in-kernel shmem files. This is what most of the callers already do.
Signed-off-by Christoph Hellwig <[email protected]>Reviewed-by Darrick J. Wong <[email protected]>
I noticed today that drivers/gpu/drm/i915/gem/i915_gem_shmem.c is going to a lot of effort to call write_begin and write_end. I'll take that out as part of the shmem conversion to buffered_write_ops.
Re: [PATCH 07/20] shmem: document how to "persist" data when using shmem_*file_setup
Quoted 4 lines · Matthew Wilcox
Quoted 3 lines · Christoph Hellwig
Add a blurb that simply dirtying the folio will persist data for in-kernel shmem files. This is what most of the callers already do.
Quoted 8 lines
Quoted 2 lines
Signed-off-by Christoph Hellwig <[email protected]>Reviewed-by Darrick J. Wong <[email protected]>Reviewed-by Matthew Wilcox (Oracle) <[email protected]>I noticed today that
drivers/gpu/drm/i915/gem/i915_gem_shmem.cis going to a lot of effort to callwrite_beginandwrite_end. I'll take that out as part of the shmem conversion tobuffered_write_ops.
I was planning to look into that, but I'll happily leave it to. It also calls into ->writepage in a somewhat gross way.
Re: [PATCH 07/20] shmem: document how to "persist" data when using shmem_*file_setup
In reply to Matthew WilcoxQuoted 1 line · Matthew Wilcox
take that out as part of the shmem conversion to
buffered_write_ops.
Oh, and I think shmem is actually a good candidate for iomap_file_buffered_write. I think it just needs to plug shmem_get_folio into iomap_folio_ops ->get_folio and provide an identify mapped iomap like the block device.
PATCH 8/20 xfs: remove xfile_stat
In reply to Christoph Hellwigvfs_getattr is needed to query inode attributes for unknown underlying file systems. But shmemfs is well known for users of shmem_file_setup and shmem_read_mapping_page_gfp that rely on it not needing specific inode revalidation and having a normal mapping. Remove the detour through the getattr method and an extra wrapper, and just read the inode size and i_bytes directly in the scrub tracing code.
--- fs/xfs/scrub/trace.h | 34 ++++++++++------------------------ fs/xfs/scrub/xfile.c | 19 ------------------- fs/xfs/scrub/xfile.h | 7 ------- 3 files changed, 10 insertions(+), 50 deletions(-)
fs/xfs/scrub/trace.h +10 -24
@@ -861,18 +861,11 @@ TRACE_EVENT(xfile_destroy,__field (loff_t, size) ),TP_fast_assign ( - struct xfile_stat statbuf; -int ret; - - ret =xfile_stat (xf, &statbuf); -if (!ret) { - __entry->bytes = statbuf.bytes; - __entry->size = statbuf.size; - }else { - __entry->bytes =-1 ; - __entry->size =-1 ; - } - __entry->ino =file_inode (xf->file)->i_ino; + struct inode *inode = file_inode(xf->file); + + __entry->ino = inode->i_ino; + __entry->bytes = inode->i_blocks <<SECTOR_SHIFT ; + __entry->size =i_size_read (inode); ), TP_printk("xfino 0x%lx mem_bytes 0x%llx isize 0x%llx", __entry->ino,
@@ -891,19 +884,12 @@ DECLARE_EVENT_CLASS(xfile_class, __field(unsigned long long, bytecount) ),TP_fast_assign ( - struct xfile_stat statbuf; -int ret; - - ret =xfile_stat (xf, &statbuf); -if (!ret) { - __entry->bytes_used = statbuf.bytes; - __entry->size = statbuf.size; - }else { - __entry->bytes_used =-1 ; - __entry->size =-1 ; - } - __entry->ino =file_inode (xf->file)->i_ino; + struct inode *inode = file_inode(xf->file); + + __entry->ino = inode->i_ino; + __entry->bytes_used = inode->i_blocks <<SECTOR_SHIFT ; __entry->pos = pos; + __entry->size =i_size_read (inode); __entry->bytecount = bytecount; ), TP_printk("xfino 0x%lx mem_bytes 0x%llx pos 0x%llx bytecount 0x%llx isize 0x%llx",
fs/xfs/scrub/xfile.c +0 -19
@@ -291,25 +291,6 @@ xfile_seek_data(return ret; } -/* Query stat information for an xfile. */ -int -xfile_stat ( -struct xfile *xf, -struct xfile_stat *statbuf) -{ -struct kstat ks; -int error; - - error =vfs_getattr_nosec (&xf->file->f_path, &ks, -STATX_SIZE |STATX_BLOCKS ,AT_STATX_DONT_SYNC ); -if (error) -return error; - - statbuf->size = ks.size; - statbuf->bytes = ks.blocks <<SECTOR_SHIFT ; -return 0 ; -} -/* * Grab the (locked) page for a memory object. The object cannot span a page * boundary. Returns 0 (and a locked page) if successful, -ENOTBLK if we
fs/xfs/scrub/xfile.h +0 -7
@@ -63,13 +63,6 @@ xfile_obj_store(struct xfile *xf, const void *buf, size_t count, loff_t pos)loff_t xfile_seek_data (struct xfile *xf,loff_t pos); -struct xfile_stat { -loff_t size; -unsigned long long bytes; -}; - -int xfile_stat (struct xfile *xf,struct xfile_stat *statbuf); -int xfile_get_page (struct xfile *xf,loff_t offset,unsigned int len,struct xfile_page *xbuf);int xfile_put_page (struct xfile *xf,struct xfile_page *xbuf);
Signature · 1 line
2.39.2
PATCH 9/20 xfs: remove the xfile_pread/pwrite APIs
In reply to Christoph HellwigAll current and pending xfile users use the xfile_obj_load and xfile_obj_store API, so make those the actual implementation.
--- .../xfs/xfs-online-fsck-design.rst | 23 +++----- fs/xfs/scrub/rtsummary.c | 6 +-- fs/xfs/scrub/trace.h | 4 +- fs/xfs/scrub/xfarray.c | 18 +++---- fs/xfs/scrub/xfile.c | 54 +++++++++---------- fs/xfs/scrub/xfile.h | 32 +---------- 6 files changed, 48 insertions(+), 89 deletions(-)
Documentation/filesystems/xfs/xfs-online-fsck-design.rst +8 -15
@@ -1915,19 +1915,13 @@ four of those five higher level data structures. The fifth use case is discussed in the :ref:`realtime summary <rtsummary>` case study. -The most general storage interface supported by the xfile enables the reading -and writing of arbitrary quantities of data at arbitrary offsets in the xfile. -This capability is provided by ``xfile_pread`` and ``xfile_pwrite`` functions, -which behave similarly to their userspace counterparts. XFS is very record-based, which suggests that the ability to load and store complete records is important. -To support these cases, a pair of ``xfile_obj_load`` and ``xfile_obj_store`` -functions are provided to read and persist objects into an xfile. -They are internally the same as pread and pwrite, except that they treat any -error as an out of memory error. -For online repair, squashing error conditions in this manner is an acceptable -behavior because the only reaction is to abort the operation back to userspace. -All five xfile usecases can be serviced by these four functions. +To support these cases, a pair of ``xfile_load`` and ``xfile_store`` +functions are provided to read and persist objects into an xfile that treat any +error as an out of memory error. For online repair, squashing error conditions +in this manner is an acceptable behavior because the only reaction is to abort +the operation back to userspace. However, no discussion of file access idioms is complete without answering the question, "But what about mmap?"
@@ -1939,10 +1933,9 @@ tmpfs can only push a pagecache folio to the swap cache if the folio is neither pinned nor locked, which means the xfile must not pin too many folios. Short term direct access to xfile contents is done by locking the pagecache -folio and mapping it into kernel address space. -Programmatic access (e.g. pread and pwrite) uses this mechanism. -Folio locks are not supposed to be held for long periods of time, so long -term direct access to xfile contents is done by bumping the folio refcount, +folio and mapping it into kernel address space. Object load and store uses this +mechanism. Folio locks are not supposed to be held for long periods of time, so +long term direct access to xfile contents is done by bumping the folio refcount, mapping it into kernel address space, and dropping the folio lock. These long term users *must* be responsive to memory reclaim by hooking into the shrinker infrastructure to know when to release folios.
fs/xfs/scrub/rtsummary.c +3 -3
@@ -118,7 +118,7 @@ xfsum_load(xfs_rtsumoff_t sumoff, union xfs_suminfo_raw *rawinfo) { - return xfile_obj_load(sc->xfile, rawinfo, + return xfile_load(sc->xfile, rawinfo, sizeof(union xfs_suminfo_raw), sumoff << XFS_WORDLOG); }
@@ -129,7 +129,7 @@ xfsum_store(xfs_rtsumoff_t sumoff, const union xfs_suminfo_raw rawinfo) { - return xfile_obj_store(sc->xfile, &rawinfo, + return xfile_store(sc->xfile, &rawinfo, sizeof(union xfs_suminfo_raw), sumoff << XFS_WORDLOG); }
@@ -141,7 +141,7 @@ xfsum_copyout(union xfs_suminfo_raw *rawinfo, unsigned int nr_words) { - return xfile_obj_load(sc->xfile, rawinfo, nr_words << XFS_WORDLOG, + return xfile_load(sc->xfile, rawinfo, nr_words << XFS_WORDLOG, sumoff << XFS_WORDLOG); }
fs/xfs/scrub/trace.h +2 -2
@@ -903,8 +903,8 @@ DECLARE_EVENT_CLASS(xfile_class,DEFINE_EVENT (xfile_class, name, \ TP_PROTO(struct xfile *xf, loff_t pos, unsigned long long bytecount), \ TP_ARGS(xf, pos, bytecount)) -DEFINE_XFILE_EVENT (xfile_pread); -DEFINE_XFILE_EVENT (xfile_pwrite); +DEFINE_XFILE_EVENT (xfile_load); +DEFINE_XFILE_EVENT (xfile_store);DEFINE_XFILE_EVENT (xfile_seek_data);DEFINE_XFILE_EVENT (xfile_get_page);DEFINE_XFILE_EVENT (xfile_put_page);
fs/xfs/scrub/xfarray.c +9 -9
@@ -136,7 +136,7 @@ xfarray_load(if (idx >= array->nr)return -ENODATA ; -return xfile_obj_load (array->xfile, ptr, array->obj_size, +return xfile_load (array->xfile, ptr, array->obj_size,xfarray_pos (array, idx)); }
@@ -152,7 +152,7 @@ xfarray_is_unset(if (array->unset_slots ==0 )return false; - error =xfile_obj_load (array->xfile, temp, array->obj_size, pos); + error =xfile_load (array->xfile, temp, array->obj_size, pos);if (!error &&xfarray_element_is_null (array, temp))return true;
@@ -184,7 +184,7 @@ xfarray_unset(return 0 ;memset (temp,0 , array->obj_size); - error =xfile_obj_store (array->xfile, temp, array->obj_size, pos); + error =xfile_store (array->xfile, temp, array->obj_size, pos);if (error)return error;
@@ -209,7 +209,7 @@ xfarray_store(ASSERT (!xfarray_element_is_null (array, ptr)); - ret =xfile_obj_store (array->xfile, ptr, array->obj_size, + ret =xfile_store (array->xfile, ptr, array->obj_size,xfarray_pos (array, idx));if (ret)return ret;
@@ -245,12 +245,12 @@ xfarray_store_anywhere(for (pos =0 ; pos < endpos && array->unset_slots >0 ; pos += array->obj_size) { - error =xfile_obj_load (array->xfile, temp, array->obj_size, + error =xfile_load (array->xfile, temp, array->obj_size, pos);if (error || !xfarray_element_is_null (array, temp))continue ; - error =xfile_obj_store (array->xfile, ptr, array->obj_size, + error =xfile_store (array->xfile, ptr, array->obj_size, pos);if (error)return error;
@@ -552,7 +552,7 @@ xfarray_isort(trace_xfarray_isort (si, lo, hi);xfarray_sort_bump_loads (si); - error =xfile_obj_load (si->array->xfile, scratch, len, lo_pos); + error =xfile_load (si->array->xfile, scratch, len, lo_pos);if (error)return error;
@@ -560,7 +560,7 @@ xfarray_isort(sort (scratch, hi - lo +1 , si->array->obj_size, si->cmp_fn,NULL );xfarray_sort_bump_stores (si); -return xfile_obj_store (si->array->xfile, scratch, len, lo_pos); +return xfile_store (si->array->xfile, scratch, len, lo_pos); }/* Grab a page for sorting records. */
@@ -858,7 +858,7 @@ xfarray_sort_load_cached(if (xfarray_sort_terminated (si, &error))return error; -return xfile_obj_load (si->array->xfile, ptr, +return xfile_load (si->array->xfile, ptr, si->array->obj_size, idx_pos); }
fs/xfs/scrub/xfile.c +24 -30
129 lines not on this page · raw message
fs/xfs/scrub/xfile.h +2 -30
@@ -29,38 +29,10 @@ struct xfile {int xfile_create (const char *description,loff_t isize,struct xfile **xfilep);void xfile_destroy (struct xfile *xf); -ssize_t xfile_pread (struct xfile *xf,void *buf,size_t count,loff_t pos); -ssize_t xfile_pwrite (struct xfile *xf,const void *buf,size_t count, +int xfile_load (struct xfile *xf,void *buf,size_t count,loff_t pos); +int xfile_store (struct xfile *xf,const void *buf,size_t count,loff_t pos); -/* -* Load an object. Since we're treating this file as "memory", any error or -* short IO is treated as a failure to allocate memory. -*/ -static inline int -xfile_obj_load (struct xfile *xf,void *buf,size_t count,loff_t pos) -{ -ssize_t ret =xfile_pread (xf, buf, count, pos); - -if (ret <0 || ret != count) -return -ENOMEM ; -return 0 ; -} - -/* -* Store an object. Since we're treating this file as "memory", any error or -* short IO is treated as a failure to allocate memory. -*/ -static inline int -xfile_obj_store (struct xfile *xf,const void *buf,size_t count,loff_t pos) -{ -ssize_t ret =xfile_pwrite (xf, buf, count, pos); - -if (ret <0 || ret != count) -return -ENOMEM ; -return 0 ; -} -loff_t xfile_seek_data (struct xfile *xf,loff_t pos); int xfile_get_page(struct xfile *xf, loff_t offset, unsigned int len,
Signature · 1 line
2.39.2
PATCH 10/20 xfs: don't try to handle non-update pages in xfile_obj_load
In reply to Christoph Hellwigshmem_read_mapping_page_gfp always returns an uptodate page or an ERR_PTR. Remove the code that tries to handle a non-uptodate page.
--- fs/xfs/scrub/xfile.c | 20 ++++++++------------ 1 file changed, 8 insertions(+), 12 deletions(-)
fs/xfs/scrub/xfile.c +8 -12
@@ -166,18 +166,14 @@ xfile_load( goto advance; } -if (PageUptodate (page)) { -/* -* xfile pages must never be mapped into userspace, so -* we skip the dcache flush. -*/ - kaddr =kmap_local_page (page); - p = kaddr +offset_in_page (pos); -memcpy (buf, p, len); -kunmap_local (kaddr); - }else { -memset (buf,0 , len); - } +/* +* xfile pages must never be mapped into userspace, so +* we skip the dcache flush. +*/ + kaddr =kmap_local_page (page); + p = kaddr +offset_in_page (pos); +memcpy (buf, p, len); +kunmap_local (kaddr);put_page (page); advance:
Signature · 1 line
2.39.2
PATCH 11/20 xfs: shmem_file_setup can't return NULL
In reply to Christoph Hellwigshmem_file_setup always returns a struct file pointer or an ERR_PTR, so remove the code to check for a NULL return.
--- fs/xfs/scrub/xfile.c | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-)
fs/xfs/scrub/xfile.c +1 -3
@@ -62,15 +62,13 @@ xfile_create( {struct inode *inode;struct xfile *xf; -int error = -ENOMEM ; +int error; xf =kmalloc (sizeof (struct xfile ),XCHK_GFP_FLAGS );if (!xf)return -ENOMEM ; xf->file =shmem_file_setup (description, isize,0 ); -if (!xf->file) - goto out_xfile;if (IS_ERR (xf->file)) { error =PTR_ERR (xf->file); goto out_xfile;
Signature · 1 line
2.39.2
PATCH 12/20 xfs: don't modify file and inode flags for shmem files
In reply to Christoph Hellwigshmem_file_setup is explicitly intended for a file that can be fully read and written by kernel users without restrictions. Don't poke into internals to change random flags in the file or inode.
--- fs/xfs/scrub/xfile.c | 17 +---------------- 1 file changed, 1 insertion(+), 16 deletions(-)
fs/xfs/scrub/xfile.c +1 -16
@@ -68,28 +68,13 @@ xfile_create(if (!xf)return -ENOMEM ; - xf->file =shmem_file_setup (description, isize,0 ); + xf->file =shmem_kernel_file_setup (description, isize,0 );if (IS_ERR (xf->file)) { error =PTR_ERR (xf->file); goto out_xfile; } -/* -* We want a large sparse file that we can pread, pwrite, and seek. -* xfile users are responsible for keeping the xfile hidden away from -* all other callers, so we skip timestamp updates and security checks. -* Make the inode only accessible by root, just in case the xfile ever -* escapes. -*/ - xf->file->f_mode |=FMODE_PREAD |FMODE_PWRITE |FMODE_NOCMTIME | -FMODE_LSEEK ; - xf->file->f_flags |=O_RDWR |O_LARGEFILE |O_NOATIME ; inode =file_inode (xf->file); - inode->i_flags |=S_PRIVATE |S_NOCMTIME |S_NOATIME ; - inode->i_mode &= ~0177 ; - inode->i_uid =GLOBAL_ROOT_UID ; - inode->i_gid =GLOBAL_ROOT_GID ; -lockdep_set_class (&inode->i_rwsem, &xfile_i_mutex_key);trace_xfile_create (xf);
Signature · 1 line
2.39.2
Hugh Dickins
maintainerRe: [PATCH 12/20] xfs: don't modify file and inode flags for shmem files
Quoted 15 lines · Christoph Hellwig
shmem_file_setupis explicitly intended for a file that can be fully read and written by kernel users without restrictions. Don't poke into internals to change random flags in the file or inode.Signed-off-by Christoph Hellwig <[email protected]>Reviewed-by Darrick J. Wong <[email protected]>--- fs/xfs/scrub/xfile.c | 17 +---------------- 1 file changed, 1 insertion(+), 16 deletions(-)fs/xfs/scrub/xfile.c +0
-0
fs/xfs/scrub/xfile.c xfile_create( · line 71
Quoted 5 lines
fs/xfs/scrub/xfile.c +1
-1@@ -68,28 +68,13 @@ xfile_create(if (!xf)return -ENOMEM ; - xf->file =shmem_file_setup (description, isize,0 ); + xf->file =shmem_kernel_file_setup (description, isize,0 );
I think you probably want to say VM_NORESERVE there, instead of 0.
I see from the (deleted) comment below that "We want a large sparse file", and the VM_NORESERVE does become important on large sparse files. It is how normal flles and memfds are set up (whereas SysV SHM says 0 to reserve size at setup). It affects the Committed_AS seen in /proc/meminfo, and how __vm_enough_memory() behaves (depending on /proc/sys/vm/overcommit_memory).
Hmm, I think mm/shmem.c is not prepared for the case of flags 0 there, and then the file growing bigger than isize later on - I suspect that Committed_AS will end up underflowing. shmem.c ought to be defensive against that, sure, but I don't think such a case has come up before.
I see you have two calls to xfile_create(), one with what looks like a known fixed non-0 isize, but one with isize 0 which will grow later.
Quoted 5 lines
fs/xfs/scrub/xfile.c +0
-0if (IS_ERR (xf->file)) { error =PTR_ERR (xf->file); goto out_xfile; }
fs/xfs/scrub/xfile.c xfile_create( · line 86
Quoted 10 lines
fs/xfs/scrub/xfile.c +0
-10-/* -* We want a large sparse file that we can pread, pwrite, and seek. -* xfile users are responsible for keeping the xfile hidden away from -* all other callers, so we skip timestamp updates and security checks. -* Make the inode only accessible by root, just in case the xfile ever -* escapes. -*/ - xf->file->f_mode |=FMODE_PREAD |FMODE_PWRITE |FMODE_NOCMTIME | -FMODE_LSEEK ; - xf->file->f_flags |=O_RDWR |O_LARGEFILE |O_NOATIME ;
I've forgotten offhand how O_LARGEFILE propagates down to where a file gets grown, and I realize that 32-bit is not your primary concern: but please double-check that it works correctly without O_LARGEFILE now (maybe the new folio ops avoid everywhere that largefile would be checked).
Hugh
Quoted 11 lines
fs/xfs/scrub/xfile.c +0
-5inode =file_inode (xf->file); - inode->i_flags |=S_PRIVATE |S_NOCMTIME |S_NOATIME ; - inode->i_mode &= ~0177 ; - inode->i_uid =GLOBAL_ROOT_UID ; - inode->i_gid =GLOBAL_ROOT_GID ; -lockdep_set_class (&inode->i_rwsem, &xfile_i_mutex_key);trace_xfile_create (xf);Signature · 1 line
2.39.2
PATCH 13/20 xfs: don't allow highmem pages in xfile mappings
In reply to Christoph HellwigXFS is generally used on 64-bit, non-highmem platforms and xfile mappings are accessed all the time. Reduce our pain by not allowing any highmem mappings in the xfile page cache and remove all the kmap calls for it.
--- fs/xfs/scrub/xfarray.c | 3 +-- fs/xfs/scrub/xfile.c | 21 +++++++++------------ 2 files changed, 10 insertions(+), 14 deletions(-)
fs/xfs/scrub/xfarray.c +1 -2
@@ -580,7 +580,7 @@ xfarray_sort_get_page(* xfile pages must never be mapped into userspace, so we skip the * dcache flush when mapping the page. */ - si->page_kaddr =kmap_local_page (si->xfpage.page); + si->page_kaddr =page_address (si->xfpage.page);return 0 ; }
@@ -592,7 +592,6 @@ xfarray_sort_put_page(if (!si->page_kaddr)return 0 ; -kunmap_local (si->page_kaddr); si->page_kaddr =NULL ;return xfile_put_page (si->array->xfile, &si->xfpage);
fs/xfs/scrub/xfile.c +9 -12
@@ -77,6 +77,12 @@ xfile_create( inode =file_inode (xf->file);lockdep_set_class (&inode->i_rwsem, &xfile_i_mutex_key); +/* +* We don't want to bother with kmapping data during repair, so don't +* allow highmem pages to back this mapping. +*/ +mapping_set_gfp_mask (inode->i_mapping,GFP_KERNEL ); +trace_xfile_create (xf); *xfilep = xf;
@@ -126,7 +132,6 @@ xfile_load( pflags =memalloc_nofs_save ();while (count >0 ) { -void *p, *kaddr;unsigned int len; len =min_t (ssize_t, count,PAGE_SIZE -offset_in_page (pos));
@@ -153,10 +158,7 @@ xfile_load(* xfile pages must never be mapped into userspace, so * we skip the dcache flush. */ - kaddr =kmap_local_page (page); - p = kaddr +offset_in_page (pos); -memcpy (buf, p, len); -kunmap_local (kaddr); +memcpy (buf,page_address (page) +offset_in_page (pos), len);put_page (page); advance:
@@ -221,14 +223,13 @@ xfile_store(* the dcache flush. If the page is not uptodate, zero it * before writing data. */ - kaddr =kmap_local_page (page); + kaddr =page_address (page);if (!PageUptodate (page)) {memset (kaddr,0 ,PAGE_SIZE );SetPageUptodate (page); } p = kaddr +offset_in_page (pos);memcpy (p, buf, len); -kunmap_local (kaddr); ret = aops->write_end (NULL , mapping, pos, len, len, page, fsdata);
@@ -314,11 +315,7 @@ xfile_get_page(* to the caller and make sure the backing store will hold on to them. */ if (!PageUptodate (page)) { -void *kaddr; - - kaddr =kmap_local_page (page); -memset (kaddr,0 ,PAGE_SIZE ); -kunmap_local (kaddr); +memset (page_address (page),0 ,PAGE_SIZE );SetPageUptodate (page); }
Signature · 1 line
2.39.2
Hugh Dickins
maintainerRe: [PATCH 13/20] xfs: don't allow highmem pages in xfile mappings
Quoted 38 lines · Christoph Hellwig
XFS is generally used on 64-bit, non-highmem platforms and xfile mappings are accessed all the time. Reduce our pain by not allowing any highmem mappings in the xfile page cache and remove all the kmap calls for it.
Signed-off-by Christoph Hellwig <[email protected]>Reviewed-by Darrick J. Wong <[email protected]>--- fs/xfs/scrub/xfarray.c | 3 +-- fs/xfs/scrub/xfile.c | 21 +++++++++------------ 2 files changed, 10 insertions(+), 14 deletions(-)fs/xfs/scrub/xfarray.c +1
-2@@ -580,7 +580,7 @@ xfarray_sort_get_page(* xfile pages must never be mapped into userspace, so we skip the * dcache flush when mapping the page. */ - si->page_kaddr =kmap_local_page (si->xfpage.page); + si->page_kaddr =page_address (si->xfpage.page);return 0 ; }@@ -592,7 +592,6 @@ xfarray_sort_put_page(if (!si->page_kaddr)return 0 ; -kunmap_local (si->page_kaddr); si->page_kaddr =NULL ;return xfile_put_page (si->array->xfile, &si->xfpage);fs/xfs/scrub/xfile.c +0
-0
fs/xfs/scrub/xfile.c xfile_create( · line 84
Quoted 8 lines
fs/xfs/scrub/xfile.c +5
-0@@ -77,6 +77,12 @@ xfile_create( inode =file_inode (xf->file);lockdep_set_class (&inode->i_rwsem, &xfile_i_mutex_key); +/* +* We don't want to bother with kmapping data during repair, so don't +* allow highmem pages to back this mapping. +*/ +mapping_set_gfp_mask (inode->i_mapping,GFP_KERNEL );
I had been going to point you to inode_nohighmem(inode), which is how that is usually done. But I share Matthew's uncertainty about cpusets and the __GFP_HARDWALL inside GFP_USER: you may well be right to stick with GFP_KERNEL, just as you've done it here. I suppose it depends on whether you see this as a system operation or a user-invoked operation.
I did also think that perhaps you could mask off __GFP_FS here, to avoid having to remember those memalloc_nofs_save()s elsewhere; but that's probably a bad idea, I doubt anywhere else does it that way.
So, no change required, but I wanted to think about it.
Hugh
Quoted 55 lines
fs/xfs/scrub/xfile.c +4
-12+trace_xfile_create (xf); *xfilep = xf;@@ -126,7 +132,6 @@ xfile_load( pflags =memalloc_nofs_save ();while (count >0 ) { -void *p, *kaddr;unsigned int len; len =min_t (ssize_t, count,PAGE_SIZE -offset_in_page (pos));@@ -153,10 +158,7 @@ xfile_load(* xfile pages must never be mapped into userspace, so * we skip the dcache flush. */ - kaddr =kmap_local_page (page); - p = kaddr +offset_in_page (pos); -memcpy (buf, p, len); -kunmap_local (kaddr); +memcpy (buf,page_address (page) +offset_in_page (pos), len);put_page (page); advance:@@ -221,14 +223,13 @@ xfile_store(* the dcache flush. If the page is not uptodate, zero it * before writing data. */ - kaddr =kmap_local_page (page); + kaddr =page_address (page);if (!PageUptodate (page)) {memset (kaddr,0 ,PAGE_SIZE );SetPageUptodate (page); } p = kaddr +offset_in_page (pos);memcpy (p, buf, len); -kunmap_local (kaddr); ret = aops->write_end (NULL , mapping, pos, len, len, page, fsdata);@@ -314,11 +315,7 @@ xfile_get_page(* to the caller and make sure the backing store will hold on to them. */ if (!PageUptodate (page)) { -void *kaddr; - - kaddr =kmap_local_page (page); -memset (kaddr,0 ,PAGE_SIZE ); -kunmap_local (kaddr); +memset (page_address (page),0 ,PAGE_SIZE );SetPageUptodate (page); }Signature · 1 line
2.39.2
PATCH 14/20 xfs: use shmem_get_folio in xfile_obj_store
In reply to Christoph HellwigSwitch to using shmem_get_folio and manually dirtying the page instead of abusing aops->write_begin and aops->write_end in xfile_get_page.
This simplifies the code by not doing indirect calls of not actually exported interfaces that don't really fit the use case very well, and happens to get us large folio support for free.
--- fs/xfs/scrub/xfile.c | 73 ++++++++++++++++---------------------------- 1 file changed, 27 insertions(+), 46 deletions(-)
fs/xfs/scrub/xfile.c +27 -46
@@ -183,11 +183,7 @@ xfile_store( loff_t pos) {struct inode *inode =file_inode (xf->file); -struct address_space *mapping = inode->i_mapping; -const struct address_space_operations *aops = mapping->a_ops; -struct page *page =NULL ;unsigned int pflags; -int error =0 ;if (count >MAX_RW_COUNT )return -ENOMEM ;
@@ -196,60 +192,45 @@ xfile_store(trace_xfile_store (xf, pos, count); +/* +* Increase the file size first so that shmem_get_folio(..., SGP_CACHE), +* actually allocates a folio instead of erroring out. +*/ +if (pos + count >i_size_read (inode)) +i_size_write (inode, pos + count); + pflags =memalloc_nofs_save ();while (count >0 ) { -void *fsdata =NULL ; -void *p, *kaddr; +struct folio *folio;unsigned int len; -int ret; +unsigned int offset; - len =min_t (ssize_t, count,PAGE_SIZE -offset_in_page (pos)); - -/* -* We call write_begin directly here to avoid all the freezer -* protection lock-taking that happens in the normal path. -* shmem doesn't support fs freeze, but lockdep doesn't know -* that and will trip over that. -*/ - error = aops->write_begin (NULL , mapping, pos, len, &page, - &fsdata); -if (error) { - error = -ENOMEM ; +if (shmem_get_folio (inode, pos >>PAGE_SHIFT , &folio, +SGP_CACHE ) <0 )break ; - } - -/* -* xfile pages must never be mapped into userspace, so we skip -* the dcache flush. If the page is not uptodate, zero it -* before writing data. -*/ - kaddr =page_address (page); -if (!PageUptodate (page)) { -memset (kaddr,0 ,PAGE_SIZE ); -SetPageUptodate (page); - } - p = kaddr +offset_in_page (pos); -memcpy (p, buf, len); - - ret = aops->write_end (NULL , mapping, pos, len, len, page, - fsdata); -if (ret <0 ) { - error = -ENOMEM ; +if (filemap_check_wb_err (inode->i_mapping,0 )) { +folio_unlock (folio); +folio_put (folio);break ; } -if (ret != len) { - error = -ENOMEM ; -break ; - } + offset =offset_in_folio (folio, pos); + len =min_t (ssize_t, count,folio_size (folio) - offset); +memcpy (folio_address (folio) + offset, buf, len); + +folio_mark_dirty (folio); +folio_unlock (folio); +folio_put (folio); - count -= ret; - pos += ret; - buf += ret; + count -= len; + pos += len; + buf += len; }memalloc_nofs_restore (pflags); -return error; +if (count) +return -ENOMEM ; +return 0 ; }/* Find the next written area in the xfile data for a given offset. */
Signature · 1 line
2.39.2
Hugh Dickins
maintainerRe: [PATCH 14/20] xfs: use shmem_get_folio in xfile_obj_store
Quoted 53 lines · Christoph Hellwig
Switch to using
shmem_get_folioand manually dirtying the page instead of abusingaops->write_beginandaops->write_endinxfile_get_page.This simplifies the code by not doing indirect calls of not actually exported interfaces that don't really fit the use case very well, and happens to get us large folio support for free.
Signed-off-by Christoph Hellwig <[email protected]>Reviewed-by Darrick J. Wong <[email protected]>--- fs/xfs/scrub/xfile.c | 73 ++++++++++++++++---------------------------- 1 file changed, 27 insertions(+), 46 deletions(-)fs/xfs/scrub/xfile.c +9
-11@@ -183,11 +183,7 @@ xfile_store( loff_t pos) {struct inode *inode =file_inode (xf->file); -struct address_space *mapping = inode->i_mapping; -const struct address_space_operations *aops = mapping->a_ops; -struct page *page =NULL ;unsigned int pflags; -int error =0 ;if (count >MAX_RW_COUNT )return -ENOMEM ;@@ -196,60 +192,45 @@ xfile_store(trace_xfile_store (xf, pos, count); +/* +* Increase the file size first so that shmem_get_folio(..., SGP_CACHE), +* actually allocates a folio instead of erroring out. +*/ +if (pos + count >i_size_read (inode)) +i_size_write (inode, pos + count); + pflags =memalloc_nofs_save ();while (count >0 ) { -void *fsdata =NULL ; -void *p, *kaddr; +struct folio *folio;unsigned int len; -int ret; +unsigned int offset; - len =min_t (ssize_t, count,PAGE_SIZE -offset_in_page (pos)); - -/* -* We call write_begin directly here to avoid all the freezer
fs/xfs/scrub/xfile.c xfile_store( · line 209
Quoted 10 lines
fs/xfs/scrub/xfile.c +2
-8-* protection lock-taking that happens in the normal path. -* shmem doesn't support fs freeze, but lockdep doesn't know -* that and will trip over that. -*/ - error = aops->write_begin (NULL , mapping, pos, len, &page, - &fsdata); -if (error) { - error = -ENOMEM ; +if (shmem_get_folio (inode, pos >>PAGE_SHIFT , &folio, +SGP_CACHE ) <0 )o
SGP_CACHE is the safest choice, yes. It will tend to do an unnecessary clear_highpage() which you immediately overwrite with the actual data; but saves calculating exactly what needs to be zeroed above and below the data - not worth bothering with, unless shaving off CPU cycles.
Quoted 11 lines
fs/xfs/scrub/xfile.c +0
-10break ; - } - -/* -* xfile pages must never be mapped into userspace, so we skip -* the dcache flush. If the page is not uptodate, zero it -* before writing data. -*/ - kaddr =page_address (page); -if (!PageUptodate (page)) { -memset (kaddr,0 ,PAGE_SIZE );
fs/xfs/scrub/xfile.c xfile_store( · line 211
Quoted 10 lines
fs/xfs/scrub/xfile.c +1
-9-SetPageUptodate (page); - } - p = kaddr +offset_in_page (pos); -memcpy (p, buf, len); - - ret = aops->write_end (NULL , mapping, pos, len, len, page, - fsdata); -if (ret <0 ) { - error = -ENOMEM ; +if (filemap_check_wb_err (inode->i_mapping,0 )) {
Hah. I was sceptical what that could ever achieve on shmem (the "wb" is misleading); but it's an ingenious suggestion from Matthew, to avoid our current horrid folio+page HWPoison handling. I followed it up a bit, and it does look as if this filemap_check_wb_err() technique should work for that; but it's not something I've tried at all.
And that's all I've got to say on the series (I read on, but certainly did not delve into the folio sorting stuff): looks good, but the VM_NORESERVE question probably needs attention (and that docbook comment to mention "locked").
XFS tree stil seems to me the right home for it all.
Hugh
Quoted 37 lines
fs/xfs/scrub/xfile.c +15
-8+folio_unlock (folio); +folio_put (folio);break ; } -if (ret != len) { - error = -ENOMEM ; -break ; - } + offset =offset_in_folio (folio, pos); + len =min_t (ssize_t, count,folio_size (folio) - offset); +memcpy (folio_address (folio) + offset, buf, len); + +folio_mark_dirty (folio); +folio_unlock (folio); +folio_put (folio); - count -= ret; - pos += ret; - buf += ret; + count -= len; + pos += len; + buf += len; }memalloc_nofs_restore (pflags); -return error; +if (count) +return -ENOMEM ; +return 0 ; }/* Find the next written area in the xfile data for a given offset. */ Signature · 1 line
2.39.2
PATCH 15/20 xfs: use shmem_get_folio in in xfile_load
In reply to Christoph HellwigSwitch to using shmem_get_folio in xfile_load instead of using shmem_read_mapping_page_gfp. This gets us support for large folios and also optimized reading from unallocated space, as shmem_get_folio with SGP_READ won't allocate a page for them just to zero the content.
--- fs/xfs/scrub/xfile.c | 61 +++++++++++++++++++------------------------- 1 file changed, 26 insertions(+), 35 deletions(-)
fs/xfs/scrub/xfile.c +26 -35
@@ -34,13 +34,6 @@* xfiles assume that the caller will handle all required concurrency * management; standard vfs locks (freezer and inode) are not taken. Reads * and writes are satisfied directly from the page cache. -* -* NOTE: The current shmemfs implementation has a quirk that in-kernel reads -* of a hole cause a page to be mapped into the file. If you are going to -* create a sparse xfile, please be careful about reading from uninitialized -* parts of the file. These pages are !Uptodate and will eventually be -* reclaimed if not written, but in the short term this boosts memory -* consumption. */ /*
@@ -118,10 +111,7 @@ xfile_load( loff_t pos) {struct inode *inode =file_inode (xf->file); -struct address_space *mapping = inode->i_mapping; -struct page *page =NULL ;unsigned int pflags; -int error =0 ;if (count >MAX_RW_COUNT )return -ENOMEM ;
@@ -132,43 +122,44 @@ xfile_load( pflags =memalloc_nofs_save ();while (count >0 ) { +struct folio *folio;unsigned int len; +unsigned int offset; - len =min_t (ssize_t, count,PAGE_SIZE -offset_in_page (pos)); - -/* -* In-kernel reads of a shmem file cause it to allocate a page -* if the mapping shows a hole. Therefore, if we hit ENOMEM -* we can continue by zeroing the caller's buffer. -*/ - page =shmem_read_mapping_page_gfp (mapping, pos >>PAGE_SHIFT , - __GFP_NOWARN); -if (IS_ERR (page)) { - error =PTR_ERR (page); -if (error != -ENOMEM ) { - error = -ENOMEM ; +if (shmem_get_folio (inode, pos >>PAGE_SHIFT , &folio, +SGP_READ ) <0 ) +break ; +if (!folio) { +/* +* No data stored at this offset, just zero the output +* buffer until the next page boundary. +*/ + len =min_t (ssize_t, count, +PAGE_SIZE -offset_in_page (pos)); +memset (buf,0 , len); + }else { +if (filemap_check_wb_err (inode->i_mapping,0 )) { +folio_unlock (folio); +folio_put (folio);break ; } -memset (buf,0 , len); - goto advance; - } - -/* -* xfile pages must never be mapped into userspace, so -* we skip the dcache flush. -*/ -memcpy (buf,page_address (page) +offset_in_page (pos), len); -put_page (page); + offset =offset_in_folio (folio, pos); + len =min_t (ssize_t, count,folio_size (folio) - offset); +memcpy (buf,folio_address (folio) + offset, len); -advance: +folio_unlock (folio); +folio_put (folio); + } count -= len; pos += len; buf += len; }memalloc_nofs_restore (pflags); -return error; +if (count) +return -ENOMEM ; +return 0 ; }/*
Signature · 1 line
2.39.2
PATCH 16/20 xfs: add file_{get,put}_folio
In reply to Christoph HellwigFrom: "Darrick J. Wong" <[email protected]>
Add helper similar to file_{get,set}_page, but which deal with folios and don't allocate new folio unless explicitly asked to, which map to shmem_get_folio instead of calling into the aops.
--- fs/xfs/scrub/trace.h | 2 ++ fs/xfs/scrub/xfile.c | 74 ++++++++++++++++++++++++++++++++++++++++++++ fs/xfs/scrub/xfile.h | 7 +++++ 3 files changed, 83 insertions(+)
fs/xfs/scrub/trace.h +2 -0
@@ -908,6 +908,8 @@ DEFINE_XFILE_EVENT(xfile_store);DEFINE_XFILE_EVENT (xfile_seek_data);DEFINE_XFILE_EVENT (xfile_get_page);DEFINE_XFILE_EVENT (xfile_put_page); +DEFINE_XFILE_EVENT (xfile_get_folio); +DEFINE_XFILE_EVENT (xfile_put_folio); TRACE_EVENT(xfarray_create, TP_PROTO(struct xfarray *xfa, unsigned long long required_capacity),
fs/xfs/scrub/xfile.c +74 -0
77 lines not on this page · raw message
fs/xfs/scrub/xfile.h +7 -0
@@ -39,4 +39,11 @@ int xfile_get_page(struct xfile *xf, loff_t offset, unsigned int len, struct xfile_page *xbuf);int xfile_put_page (struct xfile *xf,struct xfile_page *xbuf); +#define XFILE_MAX_FOLIO_SIZE (PAGE_SIZE << MAX_PAGECACHE_ORDER) + +#define XFILE_ALLOC (1 << 0)/* allocate folio if not present */ +struct folio *xfile_get_folio (struct xfile *xf,loff_t offset,size_t len, +unsigned int flags); +void xfile_put_folio (struct xfile *xf,struct folio *folio); +#endif /* __XFS_SCRUB_XFILE_H__ */
Signature · 1 line
2.39.2
PATCH 17/20 xfs: remove xfarray_sortinfo.page_kaddr
In reply to Christoph HellwigNow that xfile pages don't need kmapping, there is no need to cache the kernel virtual address for them.
--- fs/xfs/scrub/xfarray.c | 22 ++++------------------ fs/xfs/scrub/xfarray.h | 1 - 2 files changed, 4 insertions(+), 19 deletions(-)
fs/xfs/scrub/xfarray.c +4 -18
@@ -570,18 +570,7 @@ xfarray_sort_get_page( loff_t pos, uint64_t len) { - int error; - - error =xfile_get_page (si->array->xfile, pos, len, &si->xfpage); -if (error) -return error; - -/* -* xfile pages must never be mapped into userspace, so we skip the -* dcache flush when mapping the page. -*/ - si->page_kaddr =page_address (si->xfpage.page); -return 0 ; +return xfile_get_page (si->array->xfile, pos, len, &si->xfpage); }/* Release a page we grabbed for sorting records. */
@@ -589,11 +578,8 @@ static inline intxfarray_sort_put_page ( struct xfarray_sortinfo *si) { -if (!si->page_kaddr) +if (!xfile_page_cached (&si->xfpage))return 0 ; - - si->page_kaddr =NULL ; -return xfile_put_page (si->array->xfile, &si->xfpage); }
@@ -636,7 +622,7 @@ xfarray_pagesort(return error;xfarray_sort_bump_heapsorts (si); - startp = si->page_kaddr +offset_in_page (lo_pos); + startp =page_address (si->xfpage.page) +offset_in_page (lo_pos);sort (startp, hi - lo +1 , si->array->obj_size, si->cmp_fn,NULL );xfarray_sort_bump_stores (si);
@@ -883,7 +869,7 @@ xfarray_sort_load_cached(return error; } -memcpy (ptr, si->page_kaddr +offset_in_page (idx_pos), +memcpy (ptr,page_address (si->xfpage.page) +offset_in_page (idx_pos), si->array->obj_size);return 0 ; }
fs/xfs/scrub/xfarray.h +0 -1
@@ -107,7 +107,6 @@ struct xfarray_sortinfo {/* Cache a page here for faster access. */ struct xfile_page xfpage; -void *page_kaddr;#ifdef DEBUG /* Performance statistics. */
Signature · 1 line
2.39.2
PATCH 18/20 xfs: fix a comment in xfarray.c
In reply to Christoph Hellwigxfiles are shmem files, not memfds.
--- fs/xfs/scrub/xfarray.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
fs/xfs/scrub/xfarray.c +1 -1
@@ -16,7 +16,7 @@* Large Arrays of Fixed-Size Records * ================================== * -* This memory array uses an xfile (which itself is a memfd "file") to store +* This memory array uses an xfile (which itself is a shmem file) to store * large numbers of fixed-size records in memory that can be paged out. This * puts less stress on the memory reclaim algorithms during an online repair * because we don't have to pin so much memory. However, array access is less
Signature · 1 line
2.39.2
PATCH 19/20 xfs: convert xfarray_pagesort to deal with large folios
In reply to Christoph HellwigFrom: "Darrick J. Wong" <[email protected]>
Convert xfarray_pagesort to handle large folios by introducing a new xfile_get_folio routine that can return a folio of arbitrary size, and using heapsort on the full folio. This also corrects an off-by-one bug in the calculation of len in xfarray_pagesort that was papered over by xfarray_want_pagesort.
--- fs/xfs/scrub/trace.h | 43 ++++++++- fs/xfs/scrub/xfarray.c | 201 +++++++++++++++++++---------------------- fs/xfs/scrub/xfarray.h | 10 +- 3 files changed, 143 insertions(+), 111 deletions(-)
fs/xfs/scrub/trace.h +42 -1
@@ -956,7 +956,7 @@ TRACE_EVENT(xfarray_isort, __entry->hi - __entry->lo) ); -TRACE_EVENT(xfarray_pagesort, +TRACE_EVENT(xfarray_foliosort, TP_PROTO(struct xfarray_sortinfo *si, uint64_t lo, uint64_t hi), TP_ARGS(si, lo, hi), TP_STRUCT__entry(
@@ -1027,6 +1027,47 @@ TRACE_EVENT(xfarray_sort, __entry->bytes) ); +TRACE_EVENT (xfarray_sort_scan, + TP_PROTO(struct xfarray_sortinfo *si, unsigned long long idx), +TP_ARGS (si, idx), +TP_STRUCT__entry ( + __field(unsigned long, ino) + __field(unsigned long long, nr) + __field(size_t, obj_size) + __field(unsigned long long, idx) + __field(unsigned long long, folio_pos) + __field(unsigned long, folio_bytes) + __field(unsigned long long, first_idx) + __field(unsigned long long, last_idx) + ), +TP_fast_assign ( + __entry->nr = si->array->nr; + __entry->obj_size = si->array->obj_size; + __entry->ino = file_inode(si->array->xfile->file)->i_ino; + __entry->idx = idx; + if (si->folio) { + __entry->folio_pos = folio_pos(si->folio); + __entry->folio_bytes = folio_size(si->folio); + __entry->first_idx = si->first_folio_idx; + __entry->last_idx = si->last_folio_idx; + } else { + __entry->folio_pos =0 ; + __entry->folio_bytes =0 ; + __entry->first_idx =0 ; + __entry->last_idx =0 ; + } + ), +TP_printk ("xfino 0x%lx nr %llu objsz %zu idx %llu folio_pos 0x%llx folio_bytes 0x%lx first_idx %llu last_idx %llu" , + __entry->ino, + __entry->nr, + __entry->obj_size, + __entry->idx, + __entry->folio_pos, + __entry->folio_bytes, + __entry->first_idx, + __entry->last_idx) +); + TRACE_EVENT(xfarray_sort_stats, TP_PROTO(struct xfarray_sortinfo *si, int error), TP_ARGS(si, error),
fs/xfs/scrub/xfarray.c +93 -108
291 lines not on this page · raw message
fs/xfs/scrub/xfarray.h +8 -2
@@ -105,8 +105,14 @@ struct xfarray_sortinfo {/* XFARRAY_SORT_* flags; see below. */ unsigned int flags; -/* Cache a page here for faster access. */ -struct xfile_page xfpage; +/* Cache a folio here for faster scanning for pivots */ +struct folio *folio; + +/* First array index in folio that is completely readable */ +xfarray_idx_t first_folio_idx; + +/* Last array index in folio that is completely readable */ +xfarray_idx_t last_folio_idx;#ifdef DEBUG /* Performance statistics. */
Signature · 1 line
2.39.2
PATCH 20/20 xfs: remove xfile_{get,put}_page
In reply to Christoph HellwigFrom: "Darrick J. Wong" <[email protected]>
These functions aren't used anymore, so get rid of them.
--- .../xfs/xfs-online-fsck-design.rst | 2 +- fs/xfs/scrub/trace.h | 2 - fs/xfs/scrub/xfile.c | 104 ------------------ fs/xfs/scrub/xfile.h | 20 ---- 4 files changed, 1 insertion(+), 127 deletions(-)
Documentation/filesystems/xfs/xfs-online-fsck-design.rst +1 -1
@@ -1940,7 +1940,7 @@ mapping it into kernel address space, and dropping the folio lock. These long term users *must* be responsive to memory reclaim by hooking into the shrinker infrastructure to know when to release folios. -The ``xfile_get_page`` and ``xfile_put_page`` functions are provided to +The ``xfile_get_folio`` and ``xfile_put_folio`` functions are provided to retrieve the (locked) folio that backs part of an xfile and to release it. The only code to use these folio lease functions are the xfarray :ref:`sorting<xfarray_sort>` algorithms and the :ref:`in-memory
fs/xfs/scrub/trace.h +0 -2
@@ -906,8 +906,6 @@ DEFINE_EVENT(xfile_class, name, \DEFINE_XFILE_EVENT (xfile_load);DEFINE_XFILE_EVENT (xfile_store);DEFINE_XFILE_EVENT (xfile_seek_data); -DEFINE_XFILE_EVENT (xfile_get_page); -DEFINE_XFILE_EVENT (xfile_put_page);DEFINE_XFILE_EVENT (xfile_get_folio);DEFINE_XFILE_EVENT (xfile_put_folio);
fs/xfs/scrub/xfile.c +0 -104
110 lines not on this page · raw message
fs/xfs/scrub/xfile.h +0 -20
@@ -6,22 +6,6 @@#ifndef __XFS_SCRUB_XFILE_H__#define __XFS_SCRUB_XFILE_H__ -struct xfile_page { -struct page *page; -void *fsdata; -loff_t pos; -}; - -static inline bool xfile_page_cached (const struct xfile_page *xfpage) -{ -return xfpage->page !=NULL ; -} - -static inline pgoff_t xfile_page_index (const struct xfile_page *xfpage) -{ -return xfpage->page->index; -} -struct xfile {struct file *file; };
@@ -35,10 +19,6 @@ int xfile_store(struct xfile *xf, const void *buf, size_t count,loff_t xfile_seek_data (struct xfile *xf,loff_t pos); -int xfile_get_page (struct xfile *xf,loff_t offset,unsigned int len, -struct xfile_page *xbuf); -int xfile_put_page (struct xfile *xf,struct xfile_page *xbuf); -#define XFILE_MAX_FOLIO_SIZE (PAGE_SIZE << MAX_PAGECACHE_ORDER)#define XFILE_ALLOC (1 << 0)/* allocate folio if not present */
Signature · 1 line
2.39.2