[RFC PATCH 4/8] fs/dax: copy data before write

From: Shiyang Ruan
Date: Mon Apr 27 2020 - 04:48:59 EST


Add dax_copy_edges() into each dax actor functions to perform CoW.

Signed-off-by: Shiyang Ruan <ruansy.fnst@xxxxxxxxxxxxxx>
---
fs/dax.c | 37 ++++++++++++++++++++++++++++++++-----
1 file changed, 32 insertions(+), 5 deletions(-)

diff --git a/fs/dax.c b/fs/dax.c
index 8107ed10c851..13a6a1d3c3b3 100644
--- a/fs/dax.c
+++ b/fs/dax.c
@@ -1276,9 +1276,6 @@ dax_iomap_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
return iov_iter_zero(min(length, end - pos), iter);
}

- if (WARN_ON_ONCE(iomap->type != IOMAP_MAPPED))
- return -EIO;
-
/*
* Write can allocate block for an area which has a hole page mapped
* into page tables. We have to tear down these mappings so that data
@@ -1315,6 +1312,12 @@ dax_iomap_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
break;
}

+ if (iomap != srcmap) {
+ ret = dax_copy_edges(pos, length, srcmap, kaddr, false);
+ if (ret)
+ break;
+ }
+
map_len = PFN_PHYS(map_len);
kaddr += offset;
map_len -= offset;
@@ -1426,6 +1429,7 @@ static vm_fault_t dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp,
vm_fault_t ret = 0;
void *entry;
pfn_t pfn;
+ void *kaddr;

trace_dax_pte_fault(inode, vmf, ret);
/*
@@ -1507,19 +1511,27 @@ static vm_fault_t dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp,

switch (iomap.type) {
case IOMAP_MAPPED:
+cow:
if (iomap.flags & IOMAP_F_NEW) {
count_vm_event(PGMAJFAULT);
count_memcg_event_mm(vma->vm_mm, PGMAJFAULT);
major = VM_FAULT_MAJOR;
}
error = dax_iomap_direct_access(&iomap, pos, PAGE_SIZE, &pfn,
- NULL);
+ &kaddr);
if (error < 0)
goto error_finish_iomap;

entry = dax_insert_entry(&xas, mapping, vmf, entry, pfn,
0, write && !sync);

+ if (srcmap.type != IOMAP_HOLE) {
+ error = dax_copy_edges(pos, PAGE_SIZE, &srcmap, kaddr,
+ false);
+ if (error)
+ goto error_finish_iomap;
+ }
+
/*
* If we are doing synchronous page fault and inode needs fsync,
* we can insert PTE into page tables only after that happens.
@@ -1543,6 +1555,9 @@ static vm_fault_t dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp,

goto finish_iomap;
case IOMAP_UNWRITTEN:
+ if (srcmap.type != IOMAP_HOLE)
+ goto cow;
+ /*FALLTHRU*/
case IOMAP_HOLE:
if (!write) {
ret = dax_load_hole(&xas, mapping, &entry, vmf);
@@ -1650,6 +1665,7 @@ static vm_fault_t dax_iomap_pmd_fault(struct vm_fault *vmf, pfn_t *pfnp,
loff_t pos;
int error;
pfn_t pfn;
+ void *kaddr;

/*
* Check whether offset isn't beyond end of file now. Caller is
@@ -1731,14 +1747,22 @@ static vm_fault_t dax_iomap_pmd_fault(struct vm_fault *vmf, pfn_t *pfnp,

switch (iomap.type) {
case IOMAP_MAPPED:
+cow:
error = dax_iomap_direct_access(&iomap, pos, PMD_SIZE, &pfn,
- NULL);
+ &kaddr);
if (error < 0)
goto finish_iomap;

entry = dax_insert_entry(&xas, mapping, vmf, entry, pfn,
DAX_PMD, write && !sync);

+ if (srcmap.type != IOMAP_HOLE) {
+ error = dax_copy_edges(pos, PMD_SIZE, &srcmap, kaddr,
+ true);
+ if (error)
+ goto unlock_entry;
+ }
+
/*
* If we are doing synchronous page fault and inode needs fsync,
* we can insert PMD into page tables only after that happens.
@@ -1757,6 +1781,9 @@ static vm_fault_t dax_iomap_pmd_fault(struct vm_fault *vmf, pfn_t *pfnp,
result = vmf_insert_pfn_pmd(vmf, pfn, write);
break;
case IOMAP_UNWRITTEN:
+ if (srcmap.type != IOMAP_HOLE)
+ goto cow;
+ /*FALLTHRU*/
case IOMAP_HOLE:
if (WARN_ON_ONCE(write))
break;
--
2.26.2