2 * (C) 2001 Clemson University and The University of Chicago
4 * See COPYING in top-level directory.
8 * Linux VFS inode operations.
12 #include "orangefs-kernel.h"
13 #include "orangefs-bufmap.h"
15 static int read_one_page(struct page
*page
)
19 ssize_t bytes_read
= 0;
20 struct inode
*inode
= page
->mapping
->host
;
21 const __u32 blocksize
= PAGE_CACHE_SIZE
; /* inode->i_blksize */
22 const __u32 blockbits
= PAGE_CACHE_SHIFT
; /* inode->i_blkbits */
24 struct bio_vec bv
= {.bv_page
= page
, .bv_len
= PAGE_SIZE
};
26 iov_iter_bvec(&to
, ITER_BVEC
| READ
, &bv
, 1, PAGE_SIZE
);
28 gossip_debug(GOSSIP_INODE_DEBUG
,
29 "orangefs_readpage called with page %p\n",
32 max_block
= ((inode
->i_size
/ blocksize
) + 1);
34 if (page
->index
< max_block
) {
35 loff_t blockptr_offset
= (((loff_t
) page
->index
) << blockbits
);
37 bytes_read
= orangefs_inode_read(inode
,
42 /* this will only zero remaining unread portions of the page data */
43 iov_iter_zero(~0U, &to
);
44 /* takes care of potential aliasing */
45 flush_dcache_page(page
);
50 SetPageUptodate(page
);
55 /* unlock the page after the ->readpage() routine completes */
60 static int orangefs_readpage(struct file
*file
, struct page
*page
)
62 return read_one_page(page
);
65 static int orangefs_readpages(struct file
*file
,
66 struct address_space
*mapping
,
67 struct list_head
*pages
,
73 gossip_debug(GOSSIP_INODE_DEBUG
, "orangefs_readpages called\n");
75 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
78 page
= list_entry(pages
->prev
, struct page
, lru
);
80 if (!add_to_page_cache(page
,
84 ret
= read_one_page(page
);
85 gossip_debug(GOSSIP_INODE_DEBUG
,
86 "failure adding page to cache, read_one_page returned: %d\n",
89 page_cache_release(page
);
92 BUG_ON(!list_empty(pages
));
96 static void orangefs_invalidatepage(struct page
*page
,
100 gossip_debug(GOSSIP_INODE_DEBUG
,
101 "orangefs_invalidatepage called on page %p "
106 ClearPageUptodate(page
);
107 ClearPageMappedToDisk(page
);
112 static int orangefs_releasepage(struct page
*page
, gfp_t foo
)
114 gossip_debug(GOSSIP_INODE_DEBUG
,
115 "orangefs_releasepage called on page %p\n",
121 * Having a direct_IO entry point in the address_space_operations
122 * struct causes the kernel to allows us to use O_DIRECT on
123 * open. Nothing will ever call this thing, but in the future we
124 * will need to be able to use O_DIRECT on open in order to support
125 * AIO. Modeled after NFS, they do this too.
128 * static ssize_t orangefs_direct_IO(int rw,
129 * struct kiocb *iocb,
130 * struct iov_iter *iter,
133 * gossip_debug(GOSSIP_INODE_DEBUG,
134 * "orangefs_direct_IO: %s\n",
135 * iocb->ki_filp->f_path.dentry->d_name.name);
141 struct backing_dev_info orangefs_backing_dev_info
= {
144 .capabilities
= BDI_CAP_NO_ACCT_DIRTY
| BDI_CAP_NO_WRITEBACK
,
147 /** ORANGEFS2 implementation of address space operations */
148 const struct address_space_operations orangefs_address_operations
= {
149 .readpage
= orangefs_readpage
,
150 .readpages
= orangefs_readpages
,
151 .invalidatepage
= orangefs_invalidatepage
,
152 .releasepage
= orangefs_releasepage
,
153 /* .direct_IO = orangefs_direct_IO */
156 static int orangefs_setattr_size(struct inode
*inode
, struct iattr
*iattr
)
158 struct orangefs_inode_s
*orangefs_inode
= ORANGEFS_I(inode
);
159 struct orangefs_kernel_op_s
*new_op
;
160 loff_t orig_size
= i_size_read(inode
);
163 gossip_debug(GOSSIP_INODE_DEBUG
,
164 "%s: %pU: Handle is %pU | fs_id %d | size is %llu\n",
166 get_khandle_from_ino(inode
),
167 &orangefs_inode
->refn
.khandle
,
168 orangefs_inode
->refn
.fs_id
,
171 truncate_setsize(inode
, iattr
->ia_size
);
173 new_op
= op_alloc(ORANGEFS_VFS_OP_TRUNCATE
);
177 new_op
->upcall
.req
.truncate
.refn
= orangefs_inode
->refn
;
178 new_op
->upcall
.req
.truncate
.size
= (__s64
) iattr
->ia_size
;
180 ret
= service_operation(new_op
, __func__
,
181 get_interruptible_flag(inode
));
184 * the truncate has no downcall members to retrieve, but
185 * the status value tells us if it went through ok or not
187 gossip_debug(GOSSIP_INODE_DEBUG
,
188 "orangefs: orangefs_truncate got return value of %d\n",
197 * Only change the c/mtime if we are changing the size or we are
198 * explicitly asked to change it. This handles the semantic difference
199 * between truncate() and ftruncate() as implemented in the VFS.
201 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
202 * special case where we need to update the times despite not having
203 * these flags set. For all other operations the VFS set these flags
204 * explicitly if it wants a timestamp update.
206 if (orig_size
!= i_size_read(inode
) &&
207 !(iattr
->ia_valid
& (ATTR_CTIME
| ATTR_MTIME
))) {
208 iattr
->ia_ctime
= iattr
->ia_mtime
=
209 current_fs_time(inode
->i_sb
);
210 iattr
->ia_valid
|= ATTR_CTIME
| ATTR_MTIME
;
217 * Change attributes of an object referenced by dentry.
219 int orangefs_setattr(struct dentry
*dentry
, struct iattr
*iattr
)
222 struct inode
*inode
= dentry
->d_inode
;
224 gossip_debug(GOSSIP_INODE_DEBUG
,
225 "orangefs_setattr: called on %s\n",
226 dentry
->d_name
.name
);
228 ret
= inode_change_ok(inode
, iattr
);
232 if ((iattr
->ia_valid
& ATTR_SIZE
) &&
233 iattr
->ia_size
!= i_size_read(inode
)) {
234 ret
= orangefs_setattr_size(inode
, iattr
);
239 setattr_copy(inode
, iattr
);
240 mark_inode_dirty(inode
);
242 ret
= orangefs_inode_setattr(inode
, iattr
);
243 gossip_debug(GOSSIP_INODE_DEBUG
,
244 "orangefs_setattr: inode_setattr returned %d\n",
247 if (!ret
&& (iattr
->ia_valid
& ATTR_MODE
))
248 /* change mod on a file that has ACLs */
249 ret
= posix_acl_chmod(inode
, inode
->i_mode
);
252 gossip_debug(GOSSIP_INODE_DEBUG
, "orangefs_setattr: returning %d\n", ret
);
257 * Obtain attributes of an object given a dentry
259 int orangefs_getattr(struct vfsmount
*mnt
,
260 struct dentry
*dentry
,
264 struct inode
*inode
= dentry
->d_inode
;
265 struct orangefs_inode_s
*orangefs_inode
= NULL
;
267 gossip_debug(GOSSIP_INODE_DEBUG
,
268 "orangefs_getattr: called on %s\n",
269 dentry
->d_name
.name
);
272 * Similar to the above comment, a getattr also expects that all
273 * fields/attributes of the inode would be refreshed. So again, we
274 * dont have too much of a choice but refresh all the attributes.
276 ret
= orangefs_inode_getattr(inode
, ORANGEFS_ATTR_SYS_ALL_NOHINT
, 0);
278 generic_fillattr(inode
, kstat
);
279 /* override block size reported to stat */
280 orangefs_inode
= ORANGEFS_I(inode
);
281 kstat
->blksize
= orangefs_inode
->blksize
;
283 /* assume an I/O error and flag inode as bad */
284 gossip_debug(GOSSIP_INODE_DEBUG
,
285 "%s:%s:%d calling make bad inode\n",
289 orangefs_make_bad_inode(inode
);
294 int orangefs_permission(struct inode
*inode
, int mask
)
298 if (mask
& MAY_NOT_BLOCK
)
301 gossip_debug(GOSSIP_INODE_DEBUG
, "%s: refreshing\n", __func__
);
303 /* Make sure the permission (and other common attrs) are up to date. */
304 ret
= orangefs_inode_getattr(inode
,
305 ORANGEFS_ATTR_SYS_ALL_NOHINT_NOSIZE
, 0);
309 return generic_permission(inode
, mask
);
312 /* ORANGEDS2 implementation of VFS inode operations for files */
313 struct inode_operations orangefs_file_inode_operations
= {
314 .get_acl
= orangefs_get_acl
,
315 .set_acl
= orangefs_set_acl
,
316 .setattr
= orangefs_setattr
,
317 .getattr
= orangefs_getattr
,
318 .setxattr
= generic_setxattr
,
319 .getxattr
= generic_getxattr
,
320 .listxattr
= orangefs_listxattr
,
321 .removexattr
= generic_removexattr
,
322 .permission
= orangefs_permission
,
325 static int orangefs_init_iops(struct inode
*inode
)
327 inode
->i_mapping
->a_ops
= &orangefs_address_operations
;
329 switch (inode
->i_mode
& S_IFMT
) {
331 inode
->i_op
= &orangefs_file_inode_operations
;
332 inode
->i_fop
= &orangefs_file_operations
;
333 inode
->i_blkbits
= PAGE_CACHE_SHIFT
;
336 inode
->i_op
= &orangefs_symlink_inode_operations
;
339 inode
->i_op
= &orangefs_dir_inode_operations
;
340 inode
->i_fop
= &orangefs_dir_operations
;
343 gossip_debug(GOSSIP_INODE_DEBUG
,
344 "%s: unsupported mode\n",
353 * Given a ORANGEFS object identifier (fsid, handle), convert it into a ino_t type
354 * that will be used as a hash-index from where the handle will
355 * be searched for in the VFS hash table of inodes.
357 static inline ino_t
orangefs_handle_hash(struct orangefs_object_kref
*ref
)
361 return orangefs_khandle_to_ino(&(ref
->khandle
));
365 * Called to set up an inode from iget5_locked.
367 static int orangefs_set_inode(struct inode
*inode
, void *data
)
369 struct orangefs_object_kref
*ref
= (struct orangefs_object_kref
*) data
;
370 struct orangefs_inode_s
*orangefs_inode
= NULL
;
372 /* Make sure that we have sane parameters */
375 orangefs_inode
= ORANGEFS_I(inode
);
378 orangefs_inode
->refn
.fs_id
= ref
->fs_id
;
379 orangefs_inode
->refn
.khandle
= ref
->khandle
;
384 * Called to determine if handles match.
386 static int orangefs_test_inode(struct inode
*inode
, void *data
)
388 struct orangefs_object_kref
*ref
= (struct orangefs_object_kref
*) data
;
389 struct orangefs_inode_s
*orangefs_inode
= NULL
;
391 orangefs_inode
= ORANGEFS_I(inode
);
392 return (!ORANGEFS_khandle_cmp(&(orangefs_inode
->refn
.khandle
), &(ref
->khandle
))
393 && orangefs_inode
->refn
.fs_id
== ref
->fs_id
);
397 * Front-end to lookup the inode-cache maintained by the VFS using the ORANGEFS
400 * @sb: the file system super block instance.
401 * @ref: The ORANGEFS object for which we are trying to locate an inode structure.
403 struct inode
*orangefs_iget(struct super_block
*sb
, struct orangefs_object_kref
*ref
)
405 struct inode
*inode
= NULL
;
409 hash
= orangefs_handle_hash(ref
);
410 inode
= iget5_locked(sb
, hash
, orangefs_test_inode
, orangefs_set_inode
, ref
);
411 if (!inode
|| !(inode
->i_state
& I_NEW
))
414 error
= orangefs_inode_getattr(inode
,
415 ORANGEFS_ATTR_SYS_ALL_NOHINT_NOSIZE
, 0);
418 return ERR_PTR(error
);
421 inode
->i_ino
= hash
; /* needed for stat etc */
422 orangefs_init_iops(inode
);
423 unlock_new_inode(inode
);
425 gossip_debug(GOSSIP_INODE_DEBUG
,
426 "iget handle %pU, fsid %d hash %ld i_ino %lu\n",
436 * Allocate an inode for a newly created file and insert it into the inode hash.
438 struct inode
*orangefs_new_inode(struct super_block
*sb
, struct inode
*dir
,
439 int mode
, dev_t dev
, struct orangefs_object_kref
*ref
)
441 unsigned long hash
= orangefs_handle_hash(ref
);
445 gossip_debug(GOSSIP_INODE_DEBUG
,
446 "%s:(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n",
453 inode
= new_inode(sb
);
457 orangefs_set_inode(inode
, ref
);
458 inode
->i_ino
= hash
; /* needed for stat etc */
460 error
= orangefs_inode_getattr(inode
,
461 ORANGEFS_ATTR_SYS_ALL_NOHINT_NOSIZE
, 0);
465 orangefs_init_iops(inode
);
467 inode
->i_mode
= mode
;
468 inode
->i_uid
= current_fsuid();
469 inode
->i_gid
= current_fsgid();
470 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
471 inode
->i_size
= PAGE_CACHE_SIZE
;
474 error
= insert_inode_locked4(inode
, hash
, orangefs_test_inode
, ref
);
478 gossip_debug(GOSSIP_INODE_DEBUG
,
479 "Initializing ACL's for inode %pU\n",
480 get_khandle_from_ino(inode
));
481 orangefs_init_acl(inode
, dir
);
486 return ERR_PTR(error
);