tmpfs: mem_cgroup charge fault to vm_mm not current mm
authorAndres Lagar-Cavilla <andreslc@google.com>
Fri, 20 May 2016 00:12:47 +0000 (17:12 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 20 May 2016 02:12:14 +0000 (19:12 -0700)
Although shmem_fault() has been careful to count a major fault to vm_mm,
shmem_getpage_gfp() has been careless in charging a remote access fault
to current->mm owner's memcg instead of to vma->vm_mm owner's memcg:
that is inconsistent with all the mem_cgroup charging on remote access
faults in mm/memory.c.

Fix it by passing fault_mm along with fault_type to
shmem_get_page_gfp(); but in that case, now knowing the right mm, it's
better for it to handle the PGMAJFAULT updates itself.

And let's keep this clutter out of most callers' way: change the common
shmem_getpage() wrapper to hide fault_mm and fault_type as well as gfp.

Signed-off-by: Andres Lagar-Cavilla <andreslc@google.com>
Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Andres Lagar-Cavilla <andreslc@google.com>
Cc: Yang Shi <yang.shi@linaro.org>
Cc: Ning Qu <quning@gmail.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Konstantin Khlebnikov <koct9i@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/shmem.c

index 6d2de2c1bf11c1bd76b05e7b49ddfb52b04082a1..e418a995427d67b4e386c366da4dd427dc27c889 100644 (file)
@@ -121,13 +121,14 @@ static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
                                struct shmem_inode_info *info, pgoff_t index);
 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
-       struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
+               struct page **pagep, enum sgp_type sgp,
+               gfp_t gfp, struct mm_struct *fault_mm, int *fault_type);
 
 static inline int shmem_getpage(struct inode *inode, pgoff_t index,
-       struct page **pagep, enum sgp_type sgp, int *fault_type)
+               struct page **pagep, enum sgp_type sgp)
 {
        return shmem_getpage_gfp(inode, index, pagep, sgp,
-                       mapping_gfp_mask(inode->i_mapping), fault_type);
+               mapping_gfp_mask(inode->i_mapping), NULL, NULL);
 }
 
 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
@@ -527,7 +528,7 @@ static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
 
        if (partial_start) {
                struct page *page = NULL;
-               shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
+               shmem_getpage(inode, start - 1, &page, SGP_READ);
                if (page) {
                        unsigned int top = PAGE_SIZE;
                        if (start > end) {
@@ -542,7 +543,7 @@ static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
        }
        if (partial_end) {
                struct page *page = NULL;
-               shmem_getpage(inode, end, &page, SGP_READ, NULL);
+               shmem_getpage(inode, end, &page, SGP_READ);
                if (page) {
                        zero_user_segment(page, 0, partial_end);
                        set_page_dirty(page);
@@ -1115,14 +1116,19 @@ static int shmem_replace_page(struct page **pagep, gfp_t gfp,
  *
  * If we allocate a new one we do not mark it dirty. That's up to the
  * vm. If we swap it in we mark it dirty since we also free the swap
- * entry since a page cannot live in both the swap and page cache
+ * entry since a page cannot live in both the swap and page cache.
+ *
+ * fault_mm and fault_type are only supplied by shmem_fault:
+ * otherwise they are NULL.
  */
 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
-       struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
+       struct page **pagep, enum sgp_type sgp, gfp_t gfp,
+       struct mm_struct *fault_mm, int *fault_type)
 {
        struct address_space *mapping = inode->i_mapping;
        struct shmem_inode_info *info;
        struct shmem_sb_info *sbinfo;
+       struct mm_struct *charge_mm;
        struct mem_cgroup *memcg;
        struct page *page;
        swp_entry_t swap;
@@ -1168,14 +1174,19 @@ repeat:
         */
        info = SHMEM_I(inode);
        sbinfo = SHMEM_SB(inode->i_sb);
+       charge_mm = fault_mm ? : current->mm;
 
        if (swap.val) {
                /* Look it up and read it in.. */
                page = lookup_swap_cache(swap);
                if (!page) {
-                       /* here we actually do the io */
-                       if (fault_type)
+                       /* Or update major stats only when swapin succeeds?? */
+                       if (fault_type) {
                                *fault_type |= VM_FAULT_MAJOR;
+                               count_vm_event(PGMAJFAULT);
+                               mem_cgroup_count_vm_event(fault_mm, PGMAJFAULT);
+                       }
+                       /* Here we actually start the io */
                        page = shmem_swapin(swap, gfp, info, index);
                        if (!page) {
                                error = -ENOMEM;
@@ -1202,7 +1213,7 @@ repeat:
                                goto failed;
                }
 
-               error = mem_cgroup_try_charge(page, current->mm, gfp, &memcg,
+               error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
                                false);
                if (!error) {
                        error = shmem_add_to_page_cache(page, mapping, index,
@@ -1263,7 +1274,7 @@ repeat:
                if (sgp == SGP_WRITE)
                        __SetPageReferenced(page);
 
-               error = mem_cgroup_try_charge(page, current->mm, gfp, &memcg,
+               error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
                                false);
                if (error)
                        goto decused;
@@ -1352,6 +1363,7 @@ unlock:
 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 {
        struct inode *inode = file_inode(vma->vm_file);
+       gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
        int error;
        int ret = VM_FAULT_LOCKED;
 
@@ -1413,14 +1425,10 @@ static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
                spin_unlock(&inode->i_lock);
        }
 
-       error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
+       error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, SGP_CACHE,
+                                 gfp, vma->vm_mm, &ret);
        if (error)
                return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
-
-       if (ret & VM_FAULT_MAJOR) {
-               count_vm_event(PGMAJFAULT);
-               mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
-       }
        return ret;
 }
 
@@ -1567,7 +1575,7 @@ shmem_write_begin(struct file *file, struct address_space *mapping,
                        return -EPERM;
        }
 
-       return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
+       return shmem_getpage(inode, index, pagep, SGP_WRITE);
 }
 
 static int
@@ -1633,7 +1641,7 @@ static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
                                break;
                }
 
-               error = shmem_getpage(inode, index, &page, sgp, NULL);
+               error = shmem_getpage(inode, index, &page, sgp);
                if (error) {
                        if (error == -EINVAL)
                                error = 0;
@@ -1749,7 +1757,7 @@ static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
        error = 0;
 
        while (spd.nr_pages < nr_pages) {
-               error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
+               error = shmem_getpage(inode, index, &page, SGP_CACHE);
                if (error)
                        break;
                unlock_page(page);
@@ -1771,8 +1779,7 @@ static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
                page = spd.pages[page_nr];
 
                if (!PageUptodate(page) || page->mapping != mapping) {
-                       error = shmem_getpage(inode, index, &page,
-                                                       SGP_CACHE, NULL);
+                       error = shmem_getpage(inode, index, &page, SGP_CACHE);
                        if (error)
                                break;
                        unlock_page(page);
@@ -2215,8 +2222,7 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
                else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
                        error = -ENOMEM;
                else
-                       error = shmem_getpage(inode, index, &page, SGP_FALLOC,
-                                                                       NULL);
+                       error = shmem_getpage(inode, index, &page, SGP_FALLOC);
                if (error) {
                        /* Remove the !PageUptodate pages we added */
                        shmem_undo_range(inode,
@@ -2534,7 +2540,7 @@ static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *s
                inode->i_op = &shmem_short_symlink_operations;
        } else {
                inode_nohighmem(inode);
-               error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
+               error = shmem_getpage(inode, 0, &page, SGP_WRITE);
                if (error) {
                        iput(inode);
                        return error;
@@ -2575,7 +2581,7 @@ static const char *shmem_get_link(struct dentry *dentry,
                        return ERR_PTR(-ECHILD);
                }
        } else {
-               error = shmem_getpage(inode, 0, &page, SGP_READ, NULL);
+               error = shmem_getpage(inode, 0, &page, SGP_READ);
                if (error)
                        return ERR_PTR(error);
                unlock_page(page);
@@ -3479,7 +3485,8 @@ struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
        int error;
 
        BUG_ON(mapping->a_ops != &shmem_aops);
-       error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
+       error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
+                                 gfp, NULL, NULL);
        if (error)
                page = ERR_PTR(error);
        else