1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/module.h>
4 #include <linux/sched.h>
5 #include <linux/slab.h>
6 #include <linux/file.h>
7 #include <linux/mount.h>
8 #include <linux/namei.h>
9 #include <linux/writeback.h>
10 #include <linux/aio.h>
11 #include <linux/falloc.h>
14 #include "mds_client.h"
18 * Ceph file operations
20 * Implement basic open/close functionality, and implement
23 * We implement three modes of file I/O:
24 * - buffered uses the generic_file_aio_{read,write} helpers
26 * - synchronous is used when there is multi-client read/write
27 * sharing, avoids the page cache, and synchronously waits for an
30 * - direct io takes the variant of the sync path that references
31 * user pages directly.
33 * fsync() flushes and waits on dirty pages, but just queues metadata
34 * for writeback: since the MDS can recover size and mtime there is no
35 * need to wait for MDS acknowledgement.
40 * Prepare an open request. Preallocate ceph_cap to avoid an
41 * inopportune ENOMEM later.
43 static struct ceph_mds_request
*
44 prepare_open_request(struct super_block
*sb
, int flags
, int create_mode
)
46 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
47 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
48 struct ceph_mds_request
*req
;
49 int want_auth
= USE_ANY_MDS
;
50 int op
= (flags
& O_CREAT
) ? CEPH_MDS_OP_CREATE
: CEPH_MDS_OP_OPEN
;
52 if (flags
& (O_WRONLY
|O_RDWR
|O_CREAT
|O_TRUNC
))
53 want_auth
= USE_AUTH_MDS
;
55 req
= ceph_mdsc_create_request(mdsc
, op
, want_auth
);
58 req
->r_fmode
= ceph_flags_to_mode(flags
);
59 req
->r_args
.open
.flags
= cpu_to_le32(flags
);
60 req
->r_args
.open
.mode
= cpu_to_le32(create_mode
);
66 * initialize private struct file data.
67 * if we fail, clean up by dropping fmode reference on the ceph_inode
69 static int ceph_init_file(struct inode
*inode
, struct file
*file
, int fmode
)
71 struct ceph_file_info
*cf
;
73 struct ceph_inode_info
*ci
= ceph_inode(inode
);
74 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
75 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
77 switch (inode
->i_mode
& S_IFMT
) {
79 /* First file open request creates the cookie, we want to keep
80 * this cookie around for the filetime of the inode as not to
81 * have to worry about fscache register / revoke / operation
84 * Also, if we know the operation is going to invalidate data
85 * (non readonly) just nuke the cache right away.
87 ceph_fscache_register_inode_cookie(mdsc
->fsc
, ci
);
88 if ((fmode
& CEPH_FILE_MODE_WR
))
89 ceph_fscache_invalidate(inode
);
91 dout("init_file %p %p 0%o (regular)\n", inode
, file
,
93 cf
= kmem_cache_alloc(ceph_file_cachep
, GFP_NOFS
| __GFP_ZERO
);
95 ceph_put_fmode(ceph_inode(inode
), fmode
); /* clean up */
100 file
->private_data
= cf
;
101 BUG_ON(inode
->i_fop
->release
!= ceph_release
);
105 dout("init_file %p %p 0%o (symlink)\n", inode
, file
,
107 ceph_put_fmode(ceph_inode(inode
), fmode
); /* clean up */
111 dout("init_file %p %p 0%o (special)\n", inode
, file
,
114 * we need to drop the open ref now, since we don't
115 * have .release set to ceph_release.
117 ceph_put_fmode(ceph_inode(inode
), fmode
); /* clean up */
118 BUG_ON(inode
->i_fop
->release
== ceph_release
);
120 /* call the proper open fop */
121 ret
= inode
->i_fop
->open(inode
, file
);
127 * If we already have the requisite capabilities, we can satisfy
128 * the open request locally (no need to request new caps from the
129 * MDS). We do, however, need to inform the MDS (asynchronously)
130 * if our wanted caps set expands.
132 int ceph_open(struct inode
*inode
, struct file
*file
)
134 struct ceph_inode_info
*ci
= ceph_inode(inode
);
135 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
136 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
137 struct ceph_mds_request
*req
;
138 struct ceph_file_info
*cf
= file
->private_data
;
139 struct inode
*parent_inode
= NULL
;
141 int flags
, fmode
, wanted
;
144 dout("open file %p is already opened\n", file
);
148 /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */
149 flags
= file
->f_flags
& ~(O_CREAT
|O_EXCL
);
150 if (S_ISDIR(inode
->i_mode
))
151 flags
= O_DIRECTORY
; /* mds likes to know */
153 dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode
,
154 ceph_vinop(inode
), file
, flags
, file
->f_flags
);
155 fmode
= ceph_flags_to_mode(flags
);
156 wanted
= ceph_caps_for_mode(fmode
);
158 /* snapped files are read-only */
159 if (ceph_snap(inode
) != CEPH_NOSNAP
&& (file
->f_mode
& FMODE_WRITE
))
162 /* trivially open snapdir */
163 if (ceph_snap(inode
) == CEPH_SNAPDIR
) {
164 spin_lock(&ci
->i_ceph_lock
);
165 __ceph_get_fmode(ci
, fmode
);
166 spin_unlock(&ci
->i_ceph_lock
);
167 return ceph_init_file(inode
, file
, fmode
);
171 * No need to block if we have caps on the auth MDS (for
172 * write) or any MDS (for read). Update wanted set
175 spin_lock(&ci
->i_ceph_lock
);
176 if (__ceph_is_any_real_caps(ci
) &&
177 (((fmode
& CEPH_FILE_MODE_WR
) == 0) || ci
->i_auth_cap
)) {
178 int mds_wanted
= __ceph_caps_mds_wanted(ci
);
179 int issued
= __ceph_caps_issued(ci
, NULL
);
181 dout("open %p fmode %d want %s issued %s using existing\n",
182 inode
, fmode
, ceph_cap_string(wanted
),
183 ceph_cap_string(issued
));
184 __ceph_get_fmode(ci
, fmode
);
185 spin_unlock(&ci
->i_ceph_lock
);
188 if ((issued
& wanted
) != wanted
&&
189 (mds_wanted
& wanted
) != wanted
&&
190 ceph_snap(inode
) != CEPH_SNAPDIR
)
191 ceph_check_caps(ci
, 0, NULL
);
193 return ceph_init_file(inode
, file
, fmode
);
194 } else if (ceph_snap(inode
) != CEPH_NOSNAP
&&
195 (ci
->i_snap_caps
& wanted
) == wanted
) {
196 __ceph_get_fmode(ci
, fmode
);
197 spin_unlock(&ci
->i_ceph_lock
);
198 return ceph_init_file(inode
, file
, fmode
);
201 spin_unlock(&ci
->i_ceph_lock
);
203 dout("open fmode %d wants %s\n", fmode
, ceph_cap_string(wanted
));
204 req
= prepare_open_request(inode
->i_sb
, flags
, 0);
209 req
->r_inode
= inode
;
214 parent_inode
= ceph_get_dentry_parent_inode(file
->f_dentry
);
215 err
= ceph_mdsc_do_request(mdsc
, parent_inode
, req
);
218 err
= ceph_init_file(inode
, file
, req
->r_fmode
);
219 ceph_mdsc_put_request(req
);
220 dout("open result=%d on %llx.%llx\n", err
, ceph_vinop(inode
));
227 * Do a lookup + open with a single request. If we get a non-existent
228 * file or symlink, return 1 so the VFS can retry.
230 int ceph_atomic_open(struct inode
*dir
, struct dentry
*dentry
,
231 struct file
*file
, unsigned flags
, umode_t mode
,
234 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
235 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
236 struct ceph_mds_request
*req
;
240 dout("atomic_open %p dentry %p '%.*s' %s flags %d mode 0%o\n",
241 dir
, dentry
, dentry
->d_name
.len
, dentry
->d_name
.name
,
242 d_unhashed(dentry
) ? "unhashed" : "hashed", flags
, mode
);
244 if (dentry
->d_name
.len
> NAME_MAX
)
245 return -ENAMETOOLONG
;
247 err
= ceph_init_dentry(dentry
);
252 req
= prepare_open_request(dir
->i_sb
, flags
, mode
);
255 req
->r_dentry
= dget(dentry
);
257 if (flags
& O_CREAT
) {
258 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
259 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
261 req
->r_locked_dir
= dir
; /* caller holds dir->i_mutex */
262 err
= ceph_mdsc_do_request(mdsc
,
263 (flags
& (O_CREAT
|O_TRUNC
)) ? dir
: NULL
,
268 err
= ceph_handle_snapdir(req
, dentry
, err
);
269 if (err
== 0 && (flags
& O_CREAT
) && !req
->r_reply_info
.head
->is_dentry
)
270 err
= ceph_handle_notrace_create(dir
, dentry
);
272 if (d_unhashed(dentry
)) {
273 dn
= ceph_finish_lookup(req
, dentry
, err
);
277 /* we were given a hashed negative dentry */
282 if (dn
|| dentry
->d_inode
== NULL
|| S_ISLNK(dentry
->d_inode
->i_mode
)) {
283 /* make vfs retry on splice, ENOENT, or symlink */
284 dout("atomic_open finish_no_open on dn %p\n", dn
);
285 err
= finish_no_open(file
, dn
);
287 dout("atomic_open finish_open on dn %p\n", dn
);
288 if (req
->r_op
== CEPH_MDS_OP_CREATE
&& req
->r_reply_info
.has_create_ino
) {
289 ceph_init_acl(dentry
, dentry
->d_inode
, dir
);
290 *opened
|= FILE_CREATED
;
292 err
= finish_open(file
, dentry
, ceph_open
, opened
);
295 if (!req
->r_err
&& req
->r_target_inode
)
296 ceph_put_fmode(ceph_inode(req
->r_target_inode
), req
->r_fmode
);
297 ceph_mdsc_put_request(req
);
298 dout("atomic_open result=%d\n", err
);
302 int ceph_release(struct inode
*inode
, struct file
*file
)
304 struct ceph_inode_info
*ci
= ceph_inode(inode
);
305 struct ceph_file_info
*cf
= file
->private_data
;
307 dout("release inode %p file %p\n", inode
, file
);
308 ceph_put_fmode(ci
, cf
->fmode
);
309 if (cf
->last_readdir
)
310 ceph_mdsc_put_request(cf
->last_readdir
);
311 kfree(cf
->last_name
);
314 kmem_cache_free(ceph_file_cachep
, cf
);
316 /* wake up anyone waiting for caps on this inode */
317 wake_up_all(&ci
->i_cap_wq
);
322 * Read a range of bytes striped over one or more objects. Iterate over
323 * objects we stripe over. (That's not atomic, but good enough for now.)
325 * If we get a short result from the OSD, check against i_size; we need to
326 * only return a short read to the caller if we hit EOF.
328 static int striped_read(struct inode
*inode
,
330 struct page
**pages
, int num_pages
,
331 int *checkeof
, bool o_direct
,
332 unsigned long buf_align
)
334 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
335 struct ceph_inode_info
*ci
= ceph_inode(inode
);
336 u64 pos
, this_len
, left
;
337 int io_align
, page_align
;
340 struct page
**page_pos
;
342 bool hit_stripe
, was_short
;
345 * we may need to do multiple reads. not atomic, unfortunately.
350 pages_left
= num_pages
;
352 io_align
= off
& ~PAGE_MASK
;
356 page_align
= (pos
- io_align
+ buf_align
) & ~PAGE_MASK
;
358 page_align
= pos
& ~PAGE_MASK
;
360 ret
= ceph_osdc_readpages(&fsc
->client
->osdc
, ceph_vino(inode
),
361 &ci
->i_layout
, pos
, &this_len
,
364 page_pos
, pages_left
, page_align
);
367 hit_stripe
= this_len
< left
;
368 was_short
= ret
>= 0 && ret
< this_len
;
369 dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos
, left
, read
,
370 ret
, hit_stripe
? " HITSTRIPE" : "", was_short
? " SHORT" : "");
374 if (was_short
&& (pos
+ ret
< inode
->i_size
)) {
375 u64 tmp
= min(this_len
- ret
,
376 inode
->i_size
- pos
- ret
);
377 dout(" zero gap %llu to %llu\n",
378 pos
+ ret
, pos
+ ret
+ tmp
);
379 ceph_zero_page_vector_range(page_align
+ read
+ ret
,
384 didpages
= (page_align
+ ret
) >> PAGE_CACHE_SHIFT
;
388 page_pos
+= didpages
;
389 pages_left
-= didpages
;
391 /* hit stripe and need continue*/
392 if (left
&& hit_stripe
&& pos
< inode
->i_size
)
398 /* did we bounce off eof? */
399 if (pos
+ left
> inode
->i_size
)
403 dout("striped_read returns %d\n", ret
);
408 * Completely synchronous read and write methods. Direct from __user
409 * buffer to osd, or directly to user pages (if O_DIRECT).
411 * If the read spans object boundary, just do multiple reads.
413 static ssize_t
ceph_sync_read(struct kiocb
*iocb
, struct iov_iter
*i
,
416 struct file
*file
= iocb
->ki_filp
;
417 struct inode
*inode
= file_inode(file
);
419 u64 off
= iocb
->ki_pos
;
421 size_t len
= i
->count
;
423 dout("sync_read on file %p %llu~%u %s\n", file
, off
,
425 (file
->f_flags
& O_DIRECT
) ? "O_DIRECT" : "");
427 * flush any page cache pages in this range. this
428 * will make concurrent normal and sync io slow,
429 * but it will at least behave sensibly when they are
432 ret
= filemap_write_and_wait_range(inode
->i_mapping
, off
,
437 if (file
->f_flags
& O_DIRECT
) {
438 while (iov_iter_count(i
)) {
439 void __user
*data
= i
->iov
[0].iov_base
+ i
->iov_offset
;
440 size_t len
= i
->iov
[0].iov_len
- i
->iov_offset
;
442 num_pages
= calc_pages_for((unsigned long)data
, len
);
443 pages
= ceph_get_direct_page_vector(data
,
446 return PTR_ERR(pages
);
448 ret
= striped_read(inode
, off
, len
,
449 pages
, num_pages
, checkeof
,
450 1, (unsigned long)data
& ~PAGE_MASK
);
451 ceph_put_page_vector(pages
, num_pages
, true);
456 iov_iter_advance(i
, ret
);
461 num_pages
= calc_pages_for(off
, len
);
462 pages
= ceph_alloc_page_vector(num_pages
, GFP_NOFS
);
464 return PTR_ERR(pages
);
465 ret
= striped_read(inode
, off
, len
, pages
,
466 num_pages
, checkeof
, 0, 0);
469 size_t left
= len
= ret
;
472 void __user
*data
= i
->iov
[0].iov_base
474 l
= min(i
->iov
[0].iov_len
- i
->iov_offset
,
477 ret
= ceph_copy_page_vector_to_user(&pages
[k
],
481 iov_iter_advance(i
, ret
);
484 k
= calc_pages_for(iocb
->ki_pos
,
486 BUG_ON(k
>= num_pages
&& left
);
491 ceph_release_page_vector(pages
, num_pages
);
494 if (off
> iocb
->ki_pos
) {
495 ret
= off
- iocb
->ki_pos
;
499 dout("sync_read result %d\n", ret
);
504 * Write commit request unsafe callback, called to tell us when a
505 * request is unsafe (that is, in flight--has been handed to the
506 * messenger to send to its target osd). It is called again when
507 * we've received a response message indicating the request is
508 * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request
509 * is completed early (and unsuccessfully) due to a timeout or
512 * This is used if we requested both an ACK and ONDISK commit reply
515 static void ceph_sync_write_unsafe(struct ceph_osd_request
*req
, bool unsafe
)
517 struct ceph_inode_info
*ci
= ceph_inode(req
->r_inode
);
519 dout("%s %p tid %llu %ssafe\n", __func__
, req
, req
->r_tid
,
522 ceph_get_cap_refs(ci
, CEPH_CAP_FILE_WR
);
523 spin_lock(&ci
->i_unsafe_lock
);
524 list_add_tail(&req
->r_unsafe_item
,
525 &ci
->i_unsafe_writes
);
526 spin_unlock(&ci
->i_unsafe_lock
);
528 spin_lock(&ci
->i_unsafe_lock
);
529 list_del_init(&req
->r_unsafe_item
);
530 spin_unlock(&ci
->i_unsafe_lock
);
531 ceph_put_cap_refs(ci
, CEPH_CAP_FILE_WR
);
537 * Synchronous write, straight from __user pointer or user pages.
539 * If write spans object boundary, just do multiple writes. (For a
540 * correct atomic write, we should e.g. take write locks on all
541 * objects, rollback on failure, etc.)
544 ceph_sync_direct_write(struct kiocb
*iocb
, const struct iovec
*iov
,
545 unsigned long nr_segs
, size_t count
)
547 struct file
*file
= iocb
->ki_filp
;
548 struct inode
*inode
= file_inode(file
);
549 struct ceph_inode_info
*ci
= ceph_inode(inode
);
550 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
551 struct ceph_snap_context
*snapc
;
552 struct ceph_vino vino
;
553 struct ceph_osd_request
*req
;
561 struct timespec mtime
= CURRENT_TIME
;
562 loff_t pos
= iocb
->ki_pos
;
565 if (ceph_snap(file_inode(file
)) != CEPH_NOSNAP
)
568 dout("sync_direct_write on file %p %lld~%u\n", file
, pos
,
571 ret
= filemap_write_and_wait_range(inode
->i_mapping
, pos
, pos
+ count
);
575 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
576 pos
>> PAGE_CACHE_SHIFT
,
577 (pos
+ count
) >> PAGE_CACHE_SHIFT
);
579 dout("invalidate_inode_pages2_range returned %d\n", ret
);
581 flags
= CEPH_OSD_FLAG_ORDERSNAP
|
582 CEPH_OSD_FLAG_ONDISK
|
585 iov_iter_init(&i
, iov
, nr_segs
, count
, 0);
587 while (iov_iter_count(&i
) > 0) {
588 void __user
*data
= i
.iov
->iov_base
+ i
.iov_offset
;
589 u64 len
= i
.iov
->iov_len
- i
.iov_offset
;
591 page_align
= (unsigned long)data
& ~PAGE_MASK
;
593 snapc
= ci
->i_snap_realm
->cached_context
;
594 vino
= ceph_vino(inode
);
595 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
597 2,/*include a 'startsync' command*/
598 CEPH_OSD_OP_WRITE
, flags
, snapc
,
607 num_pages
= calc_pages_for(page_align
, len
);
608 pages
= ceph_get_direct_page_vector(data
, num_pages
, false);
610 ret
= PTR_ERR(pages
);
615 * throw out any page cache pages in this range. this
618 truncate_inode_pages_range(inode
->i_mapping
, pos
,
619 (pos
+len
) | (PAGE_CACHE_SIZE
-1));
620 osd_req_op_extent_osd_data_pages(req
, 0, pages
, len
, page_align
,
623 /* BUG_ON(vino.snap != CEPH_NOSNAP); */
624 ceph_osdc_build_request(req
, pos
, snapc
, vino
.snap
, &mtime
);
626 ret
= ceph_osdc_start_request(&fsc
->client
->osdc
, req
, false);
628 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
630 ceph_put_page_vector(pages
, num_pages
, false);
633 ceph_osdc_put_request(req
);
637 iov_iter_advance(&i
, (size_t)len
);
639 if (pos
> i_size_read(inode
)) {
640 check_caps
= ceph_inode_set_size(inode
, pos
);
642 ceph_check_caps(ceph_inode(inode
),
650 if (ret
!= -EOLDSNAPC
&& written
> 0) {
659 * Synchronous write, straight from __user pointer or user pages.
661 * If write spans object boundary, just do multiple writes. (For a
662 * correct atomic write, we should e.g. take write locks on all
663 * objects, rollback on failure, etc.)
665 static ssize_t
ceph_sync_write(struct kiocb
*iocb
, const struct iovec
*iov
,
666 unsigned long nr_segs
, size_t count
)
668 struct file
*file
= iocb
->ki_filp
;
669 struct inode
*inode
= file_inode(file
);
670 struct ceph_inode_info
*ci
= ceph_inode(inode
);
671 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
672 struct ceph_snap_context
*snapc
;
673 struct ceph_vino vino
;
674 struct ceph_osd_request
*req
;
682 struct timespec mtime
= CURRENT_TIME
;
683 loff_t pos
= iocb
->ki_pos
;
686 if (ceph_snap(file_inode(file
)) != CEPH_NOSNAP
)
689 dout("sync_write on file %p %lld~%u\n", file
, pos
, (unsigned)count
);
691 ret
= filemap_write_and_wait_range(inode
->i_mapping
, pos
, pos
+ count
);
695 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
696 pos
>> PAGE_CACHE_SHIFT
,
697 (pos
+ count
) >> PAGE_CACHE_SHIFT
);
699 dout("invalidate_inode_pages2_range returned %d\n", ret
);
701 flags
= CEPH_OSD_FLAG_ORDERSNAP
|
702 CEPH_OSD_FLAG_ONDISK
|
703 CEPH_OSD_FLAG_WRITE
|
706 iov_iter_init(&i
, iov
, nr_segs
, count
, 0);
708 while ((len
= iov_iter_count(&i
)) > 0) {
712 snapc
= ci
->i_snap_realm
->cached_context
;
713 vino
= ceph_vino(inode
);
714 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
716 CEPH_OSD_OP_WRITE
, flags
, snapc
,
726 * write from beginning of first page,
727 * regardless of io alignment
729 num_pages
= (len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
731 pages
= ceph_alloc_page_vector(num_pages
, GFP_NOFS
);
733 ret
= PTR_ERR(pages
);
738 for (n
= 0; n
< num_pages
; n
++) {
739 size_t plen
= min_t(size_t, left
, PAGE_SIZE
);
740 ret
= iov_iter_copy_from_user(pages
[n
], &i
, 0, plen
);
746 iov_iter_advance(&i
, ret
);
750 ceph_release_page_vector(pages
, num_pages
);
754 /* get a second commit callback */
755 req
->r_unsafe_callback
= ceph_sync_write_unsafe
;
756 req
->r_inode
= inode
;
758 osd_req_op_extent_osd_data_pages(req
, 0, pages
, len
, 0,
761 /* BUG_ON(vino.snap != CEPH_NOSNAP); */
762 ceph_osdc_build_request(req
, pos
, snapc
, vino
.snap
, &mtime
);
764 ret
= ceph_osdc_start_request(&fsc
->client
->osdc
, req
, false);
766 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
769 ceph_osdc_put_request(req
);
774 if (pos
> i_size_read(inode
)) {
775 check_caps
= ceph_inode_set_size(inode
, pos
);
777 ceph_check_caps(ceph_inode(inode
),
785 if (ret
!= -EOLDSNAPC
&& written
> 0) {
793 * Wrap generic_file_aio_read with checks for cap bits on the inode.
794 * Atomically grab references, so that those bits are not released
795 * back to the MDS mid-read.
797 * Hmm, the sync read case isn't actually async... should it be?
799 static ssize_t
ceph_aio_read(struct kiocb
*iocb
, const struct iovec
*iov
,
800 unsigned long nr_segs
, loff_t pos
)
802 struct file
*filp
= iocb
->ki_filp
;
803 struct ceph_file_info
*fi
= filp
->private_data
;
804 size_t len
= iocb
->ki_nbytes
;
805 struct inode
*inode
= file_inode(filp
);
806 struct ceph_inode_info
*ci
= ceph_inode(inode
);
809 int checkeof
= 0, read
= 0;
812 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
813 inode
, ceph_vinop(inode
), iocb
->ki_pos
, (unsigned)len
, inode
);
815 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
816 want
= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_LAZYIO
;
818 want
= CEPH_CAP_FILE_CACHE
;
819 ret
= ceph_get_caps(ci
, CEPH_CAP_FILE_RD
, want
, &got
, -1);
823 if ((got
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)) == 0 ||
824 (iocb
->ki_filp
->f_flags
& O_DIRECT
) ||
825 (fi
->flags
& CEPH_F_SYNC
)) {
828 dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n",
829 inode
, ceph_vinop(inode
), iocb
->ki_pos
, (unsigned)len
,
830 ceph_cap_string(got
));
833 ret
= generic_segment_checks(iov
, &nr_segs
,
839 iov_iter_init(&i
, iov
, nr_segs
, len
, read
);
841 /* hmm, this isn't really async... */
842 ret
= ceph_sync_read(iocb
, &i
, &checkeof
);
845 * We can't modify the content of iov,
846 * so we only read from beginning.
850 len
= iocb
->ki_nbytes
;
853 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
854 inode
, ceph_vinop(inode
), pos
, (unsigned)len
,
855 ceph_cap_string(got
));
857 ret
= generic_file_aio_read(iocb
, iov
, nr_segs
, pos
);
860 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
861 inode
, ceph_vinop(inode
), ceph_cap_string(got
), (int)ret
);
862 ceph_put_cap_refs(ci
, got
);
864 if (checkeof
&& ret
>= 0) {
865 int statret
= ceph_do_getattr(inode
,
868 /* hit EOF or hole? */
869 if (statret
== 0 && iocb
->ki_pos
< inode
->i_size
&&
871 dout("sync_read hit hole, ppos %lld < size %lld"
872 ", reading more\n", iocb
->ki_pos
,
889 * Take cap references to avoid releasing caps to MDS mid-write.
891 * If we are synchronous, and write with an old snap context, the OSD
892 * may return EOLDSNAPC. In that case, retry the write.. _after_
893 * dropping our cap refs and allowing the pending snap to logically
894 * complete _before_ this write occurs.
896 * If we are near ENOSPC, write synchronously.
898 static ssize_t
ceph_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
899 unsigned long nr_segs
, loff_t pos
)
901 struct file
*file
= iocb
->ki_filp
;
902 struct ceph_file_info
*fi
= file
->private_data
;
903 struct inode
*inode
= file_inode(file
);
904 struct ceph_inode_info
*ci
= ceph_inode(inode
);
905 struct ceph_osd_client
*osdc
=
906 &ceph_sb_to_client(inode
->i_sb
)->client
->osdc
;
907 ssize_t count
, written
= 0;
910 if (ceph_snap(inode
) != CEPH_NOSNAP
)
913 mutex_lock(&inode
->i_mutex
);
915 err
= generic_segment_checks(iov
, &nr_segs
, &count
, VERIFY_READ
);
919 /* We can write back this queue in page reclaim */
920 current
->backing_dev_info
= file
->f_mapping
->backing_dev_info
;
922 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
929 err
= file_remove_suid(file
);
933 err
= file_update_time(file
);
938 if (ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_FULL
)) {
943 dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
944 inode
, ceph_vinop(inode
), pos
, count
, inode
->i_size
);
945 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
946 want
= CEPH_CAP_FILE_BUFFER
| CEPH_CAP_FILE_LAZYIO
;
948 want
= CEPH_CAP_FILE_BUFFER
;
950 err
= ceph_get_caps(ci
, CEPH_CAP_FILE_WR
, want
, &got
, pos
+ count
);
954 dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
955 inode
, ceph_vinop(inode
), pos
, count
, ceph_cap_string(got
));
957 if ((got
& (CEPH_CAP_FILE_BUFFER
|CEPH_CAP_FILE_LAZYIO
)) == 0 ||
958 (file
->f_flags
& O_DIRECT
) || (fi
->flags
& CEPH_F_SYNC
)) {
959 mutex_unlock(&inode
->i_mutex
);
960 if (file
->f_flags
& O_DIRECT
)
961 written
= ceph_sync_direct_write(iocb
, iov
,
964 written
= ceph_sync_write(iocb
, iov
, nr_segs
, count
);
965 if (written
== -EOLDSNAPC
) {
966 dout("aio_write %p %llx.%llx %llu~%u"
967 "got EOLDSNAPC, retrying\n",
968 inode
, ceph_vinop(inode
),
969 pos
, (unsigned)iov
->iov_len
);
970 mutex_lock(&inode
->i_mutex
);
974 loff_t old_size
= inode
->i_size
;
975 struct iov_iter from
;
977 * No need to acquire the i_truncate_mutex. Because
978 * the MDS revokes Fwb caps before sending truncate
979 * message to us. We can't get Fwb cap while there
980 * are pending vmtruncate. So write and vmtruncate
981 * can not run at the same time
983 iov_iter_init(&from
, iov
, nr_segs
, count
, 0);
984 written
= generic_perform_write(file
, &from
, pos
);
985 if (likely(written
>= 0))
986 iocb
->ki_pos
= pos
+ written
;
987 if (inode
->i_size
> old_size
)
988 ceph_fscache_update_objectsize(inode
);
989 mutex_unlock(&inode
->i_mutex
);
994 spin_lock(&ci
->i_ceph_lock
);
995 dirty
= __ceph_mark_dirty_caps(ci
, CEPH_CAP_FILE_WR
);
996 spin_unlock(&ci
->i_ceph_lock
);
998 __mark_inode_dirty(inode
, dirty
);
1001 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
1002 inode
, ceph_vinop(inode
), pos
, (unsigned)iov
->iov_len
,
1003 ceph_cap_string(got
));
1004 ceph_put_cap_refs(ci
, got
);
1007 ((file
->f_flags
& O_SYNC
) || IS_SYNC(file
->f_mapping
->host
) ||
1008 ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_NEARFULL
))) {
1009 err
= vfs_fsync_range(file
, pos
, pos
+ written
- 1, 1);
1017 mutex_unlock(&inode
->i_mutex
);
1019 current
->backing_dev_info
= NULL
;
1020 return written
? written
: err
;
1024 * llseek. be sure to verify file size on SEEK_END.
1026 static loff_t
ceph_llseek(struct file
*file
, loff_t offset
, int whence
)
1028 struct inode
*inode
= file
->f_mapping
->host
;
1031 mutex_lock(&inode
->i_mutex
);
1033 if (whence
== SEEK_END
|| whence
== SEEK_DATA
|| whence
== SEEK_HOLE
) {
1034 ret
= ceph_do_getattr(inode
, CEPH_STAT_CAP_SIZE
);
1043 offset
+= inode
->i_size
;
1047 * Here we special-case the lseek(fd, 0, SEEK_CUR)
1048 * position-querying operation. Avoid rewriting the "same"
1049 * f_pos value back to the file because a concurrent read(),
1050 * write() or lseek() might have altered it
1053 offset
= file
->f_pos
;
1056 offset
+= file
->f_pos
;
1059 if (offset
>= inode
->i_size
) {
1065 if (offset
>= inode
->i_size
) {
1069 offset
= inode
->i_size
;
1073 offset
= vfs_setpos(file
, offset
, inode
->i_sb
->s_maxbytes
);
1076 mutex_unlock(&inode
->i_mutex
);
1080 static inline void ceph_zero_partial_page(
1081 struct inode
*inode
, loff_t offset
, unsigned size
)
1084 pgoff_t index
= offset
>> PAGE_CACHE_SHIFT
;
1086 page
= find_lock_page(inode
->i_mapping
, index
);
1088 wait_on_page_writeback(page
);
1089 zero_user(page
, offset
& (PAGE_CACHE_SIZE
- 1), size
);
1091 page_cache_release(page
);
1095 static void ceph_zero_pagecache_range(struct inode
*inode
, loff_t offset
,
1098 loff_t nearly
= round_up(offset
, PAGE_CACHE_SIZE
);
1099 if (offset
< nearly
) {
1100 loff_t size
= nearly
- offset
;
1103 ceph_zero_partial_page(inode
, offset
, size
);
1107 if (length
>= PAGE_CACHE_SIZE
) {
1108 loff_t size
= round_down(length
, PAGE_CACHE_SIZE
);
1109 truncate_pagecache_range(inode
, offset
, offset
+ size
- 1);
1114 ceph_zero_partial_page(inode
, offset
, length
);
1117 static int ceph_zero_partial_object(struct inode
*inode
,
1118 loff_t offset
, loff_t
*length
)
1120 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1121 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
1122 struct ceph_osd_request
*req
;
1128 op
= offset
? CEPH_OSD_OP_DELETE
: CEPH_OSD_OP_TRUNCATE
;
1131 op
= CEPH_OSD_OP_ZERO
;
1134 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
1138 CEPH_OSD_FLAG_WRITE
|
1139 CEPH_OSD_FLAG_ONDISK
,
1146 ceph_osdc_build_request(req
, offset
, NULL
, ceph_vino(inode
).snap
,
1149 ret
= ceph_osdc_start_request(&fsc
->client
->osdc
, req
, false);
1151 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
1155 ceph_osdc_put_request(req
);
1161 static int ceph_zero_objects(struct inode
*inode
, loff_t offset
, loff_t length
)
1164 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1165 s32 stripe_unit
= ceph_file_layout_su(ci
->i_layout
);
1166 s32 stripe_count
= ceph_file_layout_stripe_count(ci
->i_layout
);
1167 s32 object_size
= ceph_file_layout_object_size(ci
->i_layout
);
1168 u64 object_set_size
= object_size
* stripe_count
;
1171 /* round offset up to next period boundary */
1172 nearly
= offset
+ object_set_size
- 1;
1174 nearly
-= do_div(t
, object_set_size
);
1176 while (length
&& offset
< nearly
) {
1177 loff_t size
= length
;
1178 ret
= ceph_zero_partial_object(inode
, offset
, &size
);
1184 while (length
>= object_set_size
) {
1186 loff_t pos
= offset
;
1187 for (i
= 0; i
< stripe_count
; ++i
) {
1188 ret
= ceph_zero_partial_object(inode
, pos
, NULL
);
1193 offset
+= object_set_size
;
1194 length
-= object_set_size
;
1197 loff_t size
= length
;
1198 ret
= ceph_zero_partial_object(inode
, offset
, &size
);
1207 static long ceph_fallocate(struct file
*file
, int mode
,
1208 loff_t offset
, loff_t length
)
1210 struct ceph_file_info
*fi
= file
->private_data
;
1211 struct inode
*inode
= file_inode(file
);
1212 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1213 struct ceph_osd_client
*osdc
=
1214 &ceph_inode_to_client(inode
)->client
->osdc
;
1221 if (!S_ISREG(inode
->i_mode
))
1224 mutex_lock(&inode
->i_mutex
);
1226 if (ceph_snap(inode
) != CEPH_NOSNAP
) {
1231 if (ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_FULL
) &&
1232 !(mode
& FALLOC_FL_PUNCH_HOLE
)) {
1237 size
= i_size_read(inode
);
1238 if (!(mode
& FALLOC_FL_KEEP_SIZE
))
1239 endoff
= offset
+ length
;
1241 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
1242 want
= CEPH_CAP_FILE_BUFFER
| CEPH_CAP_FILE_LAZYIO
;
1244 want
= CEPH_CAP_FILE_BUFFER
;
1246 ret
= ceph_get_caps(ci
, CEPH_CAP_FILE_WR
, want
, &got
, endoff
);
1250 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
1252 ceph_zero_pagecache_range(inode
, offset
, length
);
1253 ret
= ceph_zero_objects(inode
, offset
, length
);
1254 } else if (endoff
> size
) {
1255 truncate_pagecache_range(inode
, size
, -1);
1256 if (ceph_inode_set_size(inode
, endoff
))
1257 ceph_check_caps(ceph_inode(inode
),
1258 CHECK_CAPS_AUTHONLY
, NULL
);
1262 spin_lock(&ci
->i_ceph_lock
);
1263 dirty
= __ceph_mark_dirty_caps(ci
, CEPH_CAP_FILE_WR
);
1264 spin_unlock(&ci
->i_ceph_lock
);
1266 __mark_inode_dirty(inode
, dirty
);
1269 ceph_put_cap_refs(ci
, got
);
1271 mutex_unlock(&inode
->i_mutex
);
1275 const struct file_operations ceph_file_fops
= {
1277 .release
= ceph_release
,
1278 .llseek
= ceph_llseek
,
1279 .read
= do_sync_read
,
1280 .write
= do_sync_write
,
1281 .aio_read
= ceph_aio_read
,
1282 .aio_write
= ceph_aio_write
,
1284 .fsync
= ceph_fsync
,
1286 .flock
= ceph_flock
,
1287 .splice_read
= generic_file_splice_read
,
1288 .splice_write
= generic_file_splice_write
,
1289 .unlocked_ioctl
= ceph_ioctl
,
1290 .compat_ioctl
= ceph_ioctl
,
1291 .fallocate
= ceph_fallocate
,