v9fs: readpage support
Subsystems (2)
RFC 1/1 v9fs: readpage support
v9fs mmap support was originally removed from v9fs at Al Viro's request, but recently there have been requests from folks who want readpage functionality (primarily to enable execution of files mounted via 9P). This patch adds readpage support (but not writepage which contained most of the objectionable code). It passes FSX (and other regressions) so it should be relatively safe, but I'm a little worried that I have to do more to invalidate anything cached by readpage.
---
fs/9p/Makefile | 1 fs/9p/v9fs_vfs.h | 1 fs/9p/vfs_addr.c | 118 +++++++++++++++++++++++++++++++++++++++++++++++++++++ fs/9p/vfs_file.c | 4 ++ fs/9p/vfs_inode.c | 1 5 files changed, 125 insertions(+), 0 deletions(-) create mode 100644 fs/9p/vfs_addr.c
applies-to: 531356e3b004da443c01dac74420f4f28b91679f 4dea4edce8d34325c023c51cb693a690c972e043
fs/9p/Makefile +1 -0
@@ -3,6 +3,7 @@ obj-$(CONFIG_9P_FS) := 9p2000.o 9p2000-objs := \ vfs_super.o \ vfs_inode.o \ + vfs_addr.o \ vfs_file.o \ vfs_dir.o \ vfs_dentry.o \
fs/9p/v9fs_vfs.h +1 -0
@@ -39,6 +39,7 @@*/ extern struct file_system_type v9fs_fs_type; +extern struct address_space_operations v9fs_addr_operations;extern struct file_operations v9fs_file_operations;extern struct file_operations v9fs_dir_operations;extern struct dentry_operations v9fs_dentry_operations;
fs/9p/vfs_addr.c new file +118 -0
118 lines not on this page · raw message
fs/9p/vfs_file.c +4 -0
@@ -306,6 +306,9 @@ v9fs_file_write(struct file *filp, const total += result; }while (count); +if ((inode->i_mapping) && (inode->i_mapping->nrpages)) +invalidate_inode_pages2 (inode->i_mapping); +kfree (buf);return total; }
@@ -317,4 +320,5 @@ struct file_operations v9fs_file_operati .open = v9fs_file_open, .release = v9fs_dir_release, .lock = v9fs_file_lock, + .mmap = generic_file_mmap, };
fs/9p/vfs_inode.c +1 -0
@@ -246,6 +246,7 @@ struct inode *v9fs_get_inode(struct supe inode->i_blocks =0 ; inode->i_rdev =0 ; inode->i_atime = inode->i_mtime = inode->i_ctime =CURRENT_TIME ; + inode->i_mapping->a_ops = &v9fs_addr_operations;switch (mode &S_IFMT ) {case S_IFIFO :
---
0.99.9j
Quoted 7 lines
+/*
+ *v9fs_vfs_readpage- read an entire page in from 9P
+ *
+ * @file: file being read
+ * @page: structure to page
+ *
+ */
this isn't actually a valid kerneldoc comment, they start with /** e.g.
/**
* v9fs_vfs_readpage - read an entire page in from 9P
* @file: file being read
* @page: structure to page
*/
not that I think kerneldoc comments on implementations driver / fs API methods are very useful in general..
fs/9p/vfs_addr.c line 57
Quoted 1 line
fs/9p/vfs_addr.c new file +1
-0+loff_t offset = ((loff_t ) page->index) <<PAGE_CACHE_SHIFT ;
please use the page_offset() helper.
fs/9p/vfs_addr.c line 68
Quoted 6 lines
fs/9p/vfs_addr.c new file +6
-0+struct v9fs_fcall *fcall =NULL ; +int fid = v9f->fid; +int total =0 ; +int result =0 ; + +page_cache_get (page);
->readpage is synchronous, so you shouldn't need to grab a reference.
fs/9p/vfs_addr.c line 106
Quoted 2 lines
fs/9p/vfs_addr.c new file +2
-0+if (PageError (page)) +ClearPageError (page);
this doesn't make a lot of sense.
fs/9p/vfs_addr.c line 109
Quoted 3 lines
fs/9p/vfs_addr.c new file +3
-0+ retval =0 ; + + Error:
labels always at positions zero or one please. I'd call this out instead of erorr aswell as it's used for the successfull completion case aswell.
fs/9p/vfs_file.c v9fs_file_write(struct file *filp, const · line 311
Quoted 11 lines
fs/9p/vfs_file.c +3
-0@@ -306,6 +306,9 @@ v9fs_file_write(struct file *filp, const total += result; }while (count); +if ((inode->i_mapping) && (inode->i_mapping->nrpages)) +invalidate_inode_pages2 (inode->i_mapping); +
how could inode->i_mapping be zero here? Also the parenthes around those values are superflous.
Quoted 10 lines · Christoph Hellwig
fs/9p/vfs_addr.c line 68
Quoted 7 lines
fs/9p/vfs_addr.c new file +6
-0+struct v9fs_fcall *fcall =NULL ; +int fid = v9f->fid; +int total =0 ; +int result =0 ; + +page_cache_get (page);->readpage is synchronous, so you shouldn't need to grab a reference.
Weird things happen if I don't have page_cache_get - perhaps there's something else I need to call if I don't call that?
bash-2.05b# /root/fsx/fsx -N 1000 -W -d testfile
mapped writes DISABLED
1 write 0x1f7e6 thru 0x2250e (0x2d29 bytes)
2 write 0x10f13 thru 0x16b11 (0x5bff bytes)
3 mapread 0x1f48a thru 0x2250e (0x3085 bytes)
Bad page state at free_hot_cold_page (in process 'fsx', page c0000000006175e0) flags:0x0000700000000008 mapping:c0000000073606b0 mapcount:0 count:0
Backtrace: Call Trace: [C0000000072274B0] [C000000000025890] .show_stack+0x50/0x1cc (unreliable) [C000000007227560] [C00000000007D0E4] .bad_page+0x90/0xec [C0000000072275E0] [C00000000007DAA4] .free_hot_cold_page+0x80/0x1ac [C000000007227680] [C00000000007DC1C] .__pagevec_free+0x4c/0x78 [C000000007227710] [C000000000088B3C] .release_pages+0x1fc/0x2c0 [C000000007227840] [C000000000088F64] .__pagevec_lru_add+0x138/0x170 [C0000000072278F0] [C000000000081C48] .__do_page_cache_readahead+0x27c/0x338 [C000000007227A50] [C00000000007981C] .filemap_nopage+0x308/0x460 [C000000007227B20] [C000000000091C30] .__handle_mm_fault+0x51c/0xe48 [C000000007227C20] [C0000000002BD5E8] .do_page_fault+0x524/0x7b4 [C000000007227E30] [C000000000004760] .handle_page_fault+0x20/0x54
Trying to fix it up, but a reboot is needed
I've incorporated the rest of your suggestions and found a bug in the meanwhile. I'll be submitting an amended patch to Andrew sometime this week unless I hear back from you on the above issue.
Thanks for the comments Christoph.
-eric