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>
12 #include "mds_client.h"
15 * Ceph file operations
17 * Implement basic open/close functionality, and implement
20 * We implement three modes of file I/O:
21 * - buffered uses the generic_file_aio_{read,write} helpers
23 * - synchronous is used when there is multi-client read/write
24 * sharing, avoids the page cache, and synchronously waits for an
27 * - direct io takes the variant of the sync path that references
28 * user pages directly.
30 * fsync() flushes and waits on dirty pages, but just queues metadata
31 * for writeback: since the MDS can recover size and mtime there is no
32 * need to wait for MDS acknowledgement.
37 * Prepare an open request. Preallocate ceph_cap to avoid an
38 * inopportune ENOMEM later.
40 static struct ceph_mds_request
*
41 prepare_open_request(struct super_block
*sb
, int flags
, int create_mode
)
43 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
44 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
45 struct ceph_mds_request
*req
;
46 int want_auth
= USE_ANY_MDS
;
47 int op
= (flags
& O_CREAT
) ? CEPH_MDS_OP_CREATE
: CEPH_MDS_OP_OPEN
;
49 if (flags
& (O_WRONLY
|O_RDWR
|O_CREAT
|O_TRUNC
))
50 want_auth
= USE_AUTH_MDS
;
52 req
= ceph_mdsc_create_request(mdsc
, op
, want_auth
);
55 req
->r_fmode
= ceph_flags_to_mode(flags
);
56 req
->r_args
.open
.flags
= cpu_to_le32(flags
);
57 req
->r_args
.open
.mode
= cpu_to_le32(create_mode
);
63 * initialize private struct file data.
64 * if we fail, clean up by dropping fmode reference on the ceph_inode
66 static int ceph_init_file(struct inode
*inode
, struct file
*file
, int fmode
)
68 struct ceph_file_info
*cf
;
71 switch (inode
->i_mode
& S_IFMT
) {
74 dout("init_file %p %p 0%o (regular)\n", inode
, file
,
76 cf
= kmem_cache_alloc(ceph_file_cachep
, GFP_NOFS
| __GFP_ZERO
);
78 ceph_put_fmode(ceph_inode(inode
), fmode
); /* clean up */
83 file
->private_data
= cf
;
84 BUG_ON(inode
->i_fop
->release
!= ceph_release
);
88 dout("init_file %p %p 0%o (symlink)\n", inode
, file
,
90 ceph_put_fmode(ceph_inode(inode
), fmode
); /* clean up */
94 dout("init_file %p %p 0%o (special)\n", inode
, file
,
97 * we need to drop the open ref now, since we don't
98 * have .release set to ceph_release.
100 ceph_put_fmode(ceph_inode(inode
), fmode
); /* clean up */
101 BUG_ON(inode
->i_fop
->release
== ceph_release
);
103 /* call the proper open fop */
104 ret
= inode
->i_fop
->open(inode
, file
);
110 * If we already have the requisite capabilities, we can satisfy
111 * the open request locally (no need to request new caps from the
112 * MDS). We do, however, need to inform the MDS (asynchronously)
113 * if our wanted caps set expands.
115 int ceph_open(struct inode
*inode
, struct file
*file
)
117 struct ceph_inode_info
*ci
= ceph_inode(inode
);
118 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
119 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
120 struct ceph_mds_request
*req
;
121 struct ceph_file_info
*cf
= file
->private_data
;
122 struct inode
*parent_inode
= NULL
;
124 int flags
, fmode
, wanted
;
127 dout("open file %p is already opened\n", file
);
131 /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */
132 flags
= file
->f_flags
& ~(O_CREAT
|O_EXCL
);
133 if (S_ISDIR(inode
->i_mode
))
134 flags
= O_DIRECTORY
; /* mds likes to know */
136 dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode
,
137 ceph_vinop(inode
), file
, flags
, file
->f_flags
);
138 fmode
= ceph_flags_to_mode(flags
);
139 wanted
= ceph_caps_for_mode(fmode
);
141 /* snapped files are read-only */
142 if (ceph_snap(inode
) != CEPH_NOSNAP
&& (file
->f_mode
& FMODE_WRITE
))
145 /* trivially open snapdir */
146 if (ceph_snap(inode
) == CEPH_SNAPDIR
) {
147 spin_lock(&ci
->i_ceph_lock
);
148 __ceph_get_fmode(ci
, fmode
);
149 spin_unlock(&ci
->i_ceph_lock
);
150 return ceph_init_file(inode
, file
, fmode
);
154 * No need to block if we have caps on the auth MDS (for
155 * write) or any MDS (for read). Update wanted set
158 spin_lock(&ci
->i_ceph_lock
);
159 if (__ceph_is_any_real_caps(ci
) &&
160 (((fmode
& CEPH_FILE_MODE_WR
) == 0) || ci
->i_auth_cap
)) {
161 int mds_wanted
= __ceph_caps_mds_wanted(ci
);
162 int issued
= __ceph_caps_issued(ci
, NULL
);
164 dout("open %p fmode %d want %s issued %s using existing\n",
165 inode
, fmode
, ceph_cap_string(wanted
),
166 ceph_cap_string(issued
));
167 __ceph_get_fmode(ci
, fmode
);
168 spin_unlock(&ci
->i_ceph_lock
);
171 if ((issued
& wanted
) != wanted
&&
172 (mds_wanted
& wanted
) != wanted
&&
173 ceph_snap(inode
) != CEPH_SNAPDIR
)
174 ceph_check_caps(ci
, 0, NULL
);
176 return ceph_init_file(inode
, file
, fmode
);
177 } else if (ceph_snap(inode
) != CEPH_NOSNAP
&&
178 (ci
->i_snap_caps
& wanted
) == wanted
) {
179 __ceph_get_fmode(ci
, fmode
);
180 spin_unlock(&ci
->i_ceph_lock
);
181 return ceph_init_file(inode
, file
, fmode
);
183 spin_unlock(&ci
->i_ceph_lock
);
185 dout("open fmode %d wants %s\n", fmode
, ceph_cap_string(wanted
));
186 req
= prepare_open_request(inode
->i_sb
, flags
, 0);
191 req
->r_inode
= inode
;
194 if (flags
& (O_CREAT
|O_TRUNC
))
195 parent_inode
= ceph_get_dentry_parent_inode(file
->f_dentry
);
196 err
= ceph_mdsc_do_request(mdsc
, parent_inode
, req
);
199 err
= ceph_init_file(inode
, file
, req
->r_fmode
);
200 ceph_mdsc_put_request(req
);
201 dout("open result=%d on %llx.%llx\n", err
, ceph_vinop(inode
));
208 * Do a lookup + open with a single request. If we get a non-existent
209 * file or symlink, return 1 so the VFS can retry.
211 int ceph_atomic_open(struct inode
*dir
, struct dentry
*dentry
,
212 struct file
*file
, unsigned flags
, umode_t mode
,
215 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
216 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
217 struct ceph_mds_request
*req
;
221 dout("atomic_open %p dentry %p '%.*s' %s flags %d mode 0%o\n",
222 dir
, dentry
, dentry
->d_name
.len
, dentry
->d_name
.name
,
223 d_unhashed(dentry
) ? "unhashed" : "hashed", flags
, mode
);
225 if (dentry
->d_name
.len
> NAME_MAX
)
226 return -ENAMETOOLONG
;
228 err
= ceph_init_dentry(dentry
);
233 req
= prepare_open_request(dir
->i_sb
, flags
, mode
);
236 req
->r_dentry
= dget(dentry
);
238 if (flags
& O_CREAT
) {
239 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
240 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
242 req
->r_locked_dir
= dir
; /* caller holds dir->i_mutex */
243 err
= ceph_mdsc_do_request(mdsc
,
244 (flags
& (O_CREAT
|O_TRUNC
)) ? dir
: NULL
,
249 err
= ceph_handle_snapdir(req
, dentry
, err
);
250 if (err
== 0 && (flags
& O_CREAT
) && !req
->r_reply_info
.head
->is_dentry
)
251 err
= ceph_handle_notrace_create(dir
, dentry
);
253 if (d_unhashed(dentry
)) {
254 dn
= ceph_finish_lookup(req
, dentry
, err
);
258 /* we were given a hashed negative dentry */
263 if (dn
|| dentry
->d_inode
== NULL
|| S_ISLNK(dentry
->d_inode
->i_mode
)) {
264 /* make vfs retry on splice, ENOENT, or symlink */
265 dout("atomic_open finish_no_open on dn %p\n", dn
);
266 err
= finish_no_open(file
, dn
);
268 dout("atomic_open finish_open on dn %p\n", dn
);
269 if (req
->r_op
== CEPH_MDS_OP_CREATE
&& req
->r_reply_info
.has_create_ino
) {
270 *opened
|= FILE_CREATED
;
272 err
= finish_open(file
, dentry
, ceph_open
, opened
);
276 ceph_mdsc_put_request(req
);
277 dout("atomic_open result=%d\n", err
);
281 int ceph_release(struct inode
*inode
, struct file
*file
)
283 struct ceph_inode_info
*ci
= ceph_inode(inode
);
284 struct ceph_file_info
*cf
= file
->private_data
;
286 dout("release inode %p file %p\n", inode
, file
);
287 ceph_put_fmode(ci
, cf
->fmode
);
288 if (cf
->last_readdir
)
289 ceph_mdsc_put_request(cf
->last_readdir
);
290 kfree(cf
->last_name
);
293 kmem_cache_free(ceph_file_cachep
, cf
);
295 /* wake up anyone waiting for caps on this inode */
296 wake_up_all(&ci
->i_cap_wq
);
301 * Read a range of bytes striped over one or more objects. Iterate over
302 * objects we stripe over. (That's not atomic, but good enough for now.)
304 * If we get a short result from the OSD, check against i_size; we need to
305 * only return a short read to the caller if we hit EOF.
307 static int striped_read(struct inode
*inode
,
309 struct page
**pages
, int num_pages
,
310 int *checkeof
, bool o_direct
,
311 unsigned long buf_align
)
313 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
314 struct ceph_inode_info
*ci
= ceph_inode(inode
);
316 int io_align
, page_align
;
317 int left
, pages_left
;
319 struct page
**page_pos
;
321 bool hit_stripe
, was_short
;
324 * we may need to do multiple reads. not atomic, unfortunately.
329 pages_left
= num_pages
;
331 io_align
= off
& ~PAGE_MASK
;
335 page_align
= (pos
- io_align
+ buf_align
) & ~PAGE_MASK
;
337 page_align
= pos
& ~PAGE_MASK
;
339 ret
= ceph_osdc_readpages(&fsc
->client
->osdc
, ceph_vino(inode
),
340 &ci
->i_layout
, pos
, &this_len
,
343 page_pos
, pages_left
, page_align
);
346 hit_stripe
= this_len
< left
;
347 was_short
= ret
>= 0 && ret
< this_len
;
348 dout("striped_read %llu~%u (read %u) got %d%s%s\n", pos
, left
, read
,
349 ret
, hit_stripe
? " HITSTRIPE" : "", was_short
? " SHORT" : "");
352 int didpages
= (page_align
+ ret
) >> PAGE_CACHE_SHIFT
;
354 if (read
< pos
- off
) {
355 dout(" zero gap %llu to %llu\n", off
+ read
, pos
);
356 ceph_zero_page_vector_range(page_align
+ read
,
357 pos
- off
- read
, pages
);
362 page_pos
+= didpages
;
363 pages_left
-= didpages
;
366 if (left
&& hit_stripe
)
371 /* did we bounce off eof? */
372 if (pos
+ left
> inode
->i_size
)
375 /* zero trailing bytes (inside i_size) */
376 if (left
> 0 && pos
< inode
->i_size
) {
377 if (pos
+ left
> inode
->i_size
)
378 left
= inode
->i_size
- pos
;
380 dout("zero tail %d\n", left
);
381 ceph_zero_page_vector_range(page_align
+ read
, left
,
389 dout("striped_read returns %d\n", ret
);
394 * Completely synchronous read and write methods. Direct from __user
395 * buffer to osd, or directly to user pages (if O_DIRECT).
397 * If the read spans object boundary, just do multiple reads.
399 static ssize_t
ceph_sync_read(struct file
*file
, char __user
*data
,
400 unsigned len
, loff_t
*poff
, int *checkeof
)
402 struct inode
*inode
= file_inode(file
);
407 dout("sync_read on file %p %llu~%u %s\n", file
, off
, len
,
408 (file
->f_flags
& O_DIRECT
) ? "O_DIRECT" : "");
410 if (file
->f_flags
& O_DIRECT
) {
411 num_pages
= calc_pages_for((unsigned long)data
, len
);
412 pages
= ceph_get_direct_page_vector(data
, num_pages
, true);
414 num_pages
= calc_pages_for(off
, len
);
415 pages
= ceph_alloc_page_vector(num_pages
, GFP_NOFS
);
418 return PTR_ERR(pages
);
421 * flush any page cache pages in this range. this
422 * will make concurrent normal and sync io slow,
423 * but it will at least behave sensibly when they are
426 ret
= filemap_write_and_wait(inode
->i_mapping
);
430 ret
= striped_read(inode
, off
, len
, pages
, num_pages
, checkeof
,
431 file
->f_flags
& O_DIRECT
,
432 (unsigned long)data
& ~PAGE_MASK
);
434 if (ret
>= 0 && (file
->f_flags
& O_DIRECT
) == 0)
435 ret
= ceph_copy_page_vector_to_user(pages
, data
, off
, ret
);
440 if (file
->f_flags
& O_DIRECT
)
441 ceph_put_page_vector(pages
, num_pages
, true);
443 ceph_release_page_vector(pages
, num_pages
);
444 dout("sync_read result %d\n", ret
);
449 * Write commit callback, called if we requested both an ACK and
450 * ONDISK commit reply from the OSD.
452 static void sync_write_commit(struct ceph_osd_request
*req
,
453 struct ceph_msg
*msg
)
455 struct ceph_inode_info
*ci
= ceph_inode(req
->r_inode
);
457 dout("sync_write_commit %p tid %llu\n", req
, req
->r_tid
);
458 spin_lock(&ci
->i_unsafe_lock
);
459 list_del_init(&req
->r_unsafe_item
);
460 spin_unlock(&ci
->i_unsafe_lock
);
461 ceph_put_cap_refs(ci
, CEPH_CAP_FILE_WR
);
465 * Synchronous write, straight from __user pointer or user pages (if
468 * If write spans object boundary, just do multiple writes. (For a
469 * correct atomic write, we should e.g. take write locks on all
470 * objects, rollback on failure, etc.)
472 static ssize_t
ceph_sync_write(struct file
*file
, const char __user
*data
,
473 size_t left
, loff_t
*offset
)
475 struct inode
*inode
= file_inode(file
);
476 struct ceph_inode_info
*ci
= ceph_inode(inode
);
477 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
478 struct ceph_osd_request
*req
;
481 long long unsigned pos
;
487 int page_align
, io_align
;
488 unsigned long buf_align
;
490 struct timespec mtime
= CURRENT_TIME
;
492 if (ceph_snap(file_inode(file
)) != CEPH_NOSNAP
)
495 dout("sync_write on file %p %lld~%u %s\n", file
, *offset
,
496 (unsigned)left
, (file
->f_flags
& O_DIRECT
) ? "O_DIRECT" : "");
498 if (file
->f_flags
& O_APPEND
)
499 pos
= i_size_read(inode
);
503 ret
= filemap_write_and_wait_range(inode
->i_mapping
, pos
, pos
+ left
);
507 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
508 pos
>> PAGE_CACHE_SHIFT
,
509 (pos
+ left
) >> PAGE_CACHE_SHIFT
);
511 dout("invalidate_inode_pages2_range returned %d\n", ret
);
513 flags
= CEPH_OSD_FLAG_ORDERSNAP
|
514 CEPH_OSD_FLAG_ONDISK
|
516 if ((file
->f_flags
& (O_SYNC
|O_DIRECT
)) == 0)
517 flags
|= CEPH_OSD_FLAG_ACK
;
522 * we may need to do multiple writes here if we span an object
523 * boundary. this isn't atomic, unfortunately. :(
526 io_align
= pos
& ~PAGE_MASK
;
527 buf_align
= (unsigned long)data
& ~PAGE_MASK
;
529 if (file
->f_flags
& O_DIRECT
) {
530 /* write from beginning of first page, regardless of
532 page_align
= (pos
- io_align
+ buf_align
) & ~PAGE_MASK
;
533 num_pages
= calc_pages_for((unsigned long)data
, len
);
535 page_align
= pos
& ~PAGE_MASK
;
536 num_pages
= calc_pages_for(pos
, len
);
538 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
539 ceph_vino(inode
), pos
, &len
,
540 CEPH_OSD_OP_WRITE
, flags
,
541 ci
->i_snap_realm
->cached_context
,
543 ci
->i_truncate_seq
, ci
->i_truncate_size
,
544 &mtime
, false, page_align
);
548 if (file
->f_flags
& O_DIRECT
) {
549 pages
= ceph_get_direct_page_vector(data
, num_pages
, false);
551 ret
= PTR_ERR(pages
);
556 * throw out any page cache pages in this range. this
559 truncate_inode_pages_range(inode
->i_mapping
, pos
,
560 (pos
+len
) | (PAGE_CACHE_SIZE
-1));
562 pages
= ceph_alloc_page_vector(num_pages
, GFP_NOFS
);
564 ret
= PTR_ERR(pages
);
567 ret
= ceph_copy_user_to_page_vector(pages
, data
, pos
, len
);
569 ceph_release_page_vector(pages
, num_pages
);
573 if ((file
->f_flags
& O_SYNC
) == 0) {
574 /* get a second commit callback */
575 req
->r_safe_callback
= sync_write_commit
;
576 req
->r_own_pages
= 1;
579 req
->r_pages
= pages
;
580 req
->r_num_pages
= num_pages
;
581 req
->r_inode
= inode
;
583 ret
= ceph_osdc_start_request(&fsc
->client
->osdc
, req
, false);
585 if (req
->r_safe_callback
) {
587 * Add to inode unsafe list only after we
588 * start_request so that a tid has been assigned.
590 spin_lock(&ci
->i_unsafe_lock
);
591 list_add_tail(&req
->r_unsafe_item
,
592 &ci
->i_unsafe_writes
);
593 spin_unlock(&ci
->i_unsafe_lock
);
594 ceph_get_cap_refs(ci
, CEPH_CAP_FILE_WR
);
597 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
598 if (ret
< 0 && req
->r_safe_callback
) {
599 spin_lock(&ci
->i_unsafe_lock
);
600 list_del_init(&req
->r_unsafe_item
);
601 spin_unlock(&ci
->i_unsafe_lock
);
602 ceph_put_cap_refs(ci
, CEPH_CAP_FILE_WR
);
606 if (file
->f_flags
& O_DIRECT
)
607 ceph_put_page_vector(pages
, num_pages
, false);
608 else if (file
->f_flags
& O_SYNC
)
609 ceph_release_page_vector(pages
, num_pages
);
612 ceph_osdc_put_request(req
);
623 if (pos
> i_size_read(inode
))
624 check_caps
= ceph_inode_set_size(inode
, pos
);
626 ceph_check_caps(ceph_inode(inode
), CHECK_CAPS_AUTHONLY
,
633 * Wrap generic_file_aio_read with checks for cap bits on the inode.
634 * Atomically grab references, so that those bits are not released
635 * back to the MDS mid-read.
637 * Hmm, the sync read case isn't actually async... should it be?
639 static ssize_t
ceph_aio_read(struct kiocb
*iocb
, const struct iovec
*iov
,
640 unsigned long nr_segs
, loff_t pos
)
642 struct file
*filp
= iocb
->ki_filp
;
643 struct ceph_file_info
*fi
= filp
->private_data
;
644 loff_t
*ppos
= &iocb
->ki_pos
;
645 size_t len
= iov
->iov_len
;
646 struct inode
*inode
= file_inode(filp
);
647 struct ceph_inode_info
*ci
= ceph_inode(inode
);
648 void __user
*base
= iov
->iov_base
;
651 int checkeof
= 0, read
= 0;
653 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
654 inode
, ceph_vinop(inode
), pos
, (unsigned)len
, inode
);
656 __ceph_do_pending_vmtruncate(inode
);
657 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
658 want
= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_LAZYIO
;
660 want
= CEPH_CAP_FILE_CACHE
;
661 ret
= ceph_get_caps(ci
, CEPH_CAP_FILE_RD
, want
, &got
, -1);
664 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
665 inode
, ceph_vinop(inode
), pos
, (unsigned)len
,
666 ceph_cap_string(got
));
668 if ((got
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)) == 0 ||
669 (iocb
->ki_filp
->f_flags
& O_DIRECT
) ||
670 (inode
->i_sb
->s_flags
& MS_SYNCHRONOUS
) ||
671 (fi
->flags
& CEPH_F_SYNC
))
672 /* hmm, this isn't really async... */
673 ret
= ceph_sync_read(filp
, base
, len
, ppos
, &checkeof
);
675 ret
= generic_file_aio_read(iocb
, iov
, nr_segs
, pos
);
678 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
679 inode
, ceph_vinop(inode
), ceph_cap_string(got
), (int)ret
);
680 ceph_put_cap_refs(ci
, got
);
682 if (checkeof
&& ret
>= 0) {
683 int statret
= ceph_do_getattr(inode
, CEPH_STAT_CAP_SIZE
);
685 /* hit EOF or hole? */
686 if (statret
== 0 && *ppos
< inode
->i_size
) {
687 dout("aio_read sync_read hit hole, ppos %lld < size %lld, reading more\n", *ppos
, inode
->i_size
);
702 * Take cap references to avoid releasing caps to MDS mid-write.
704 * If we are synchronous, and write with an old snap context, the OSD
705 * may return EOLDSNAPC. In that case, retry the write.. _after_
706 * dropping our cap refs and allowing the pending snap to logically
707 * complete _before_ this write occurs.
709 * If we are near ENOSPC, write synchronously.
711 static ssize_t
ceph_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
712 unsigned long nr_segs
, loff_t pos
)
714 struct file
*file
= iocb
->ki_filp
;
715 struct ceph_file_info
*fi
= file
->private_data
;
716 struct inode
*inode
= file_inode(file
);
717 struct ceph_inode_info
*ci
= ceph_inode(inode
);
718 struct ceph_osd_client
*osdc
=
719 &ceph_sb_to_client(inode
->i_sb
)->client
->osdc
;
720 loff_t endoff
= pos
+ iov
->iov_len
;
722 int ret
, err
, written
;
724 if (ceph_snap(inode
) != CEPH_NOSNAP
)
729 if (ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_FULL
))
731 __ceph_do_pending_vmtruncate(inode
);
734 * try to do a buffered write. if we don't have sufficient
735 * caps, we'll get -EAGAIN from generic_file_aio_write, or a
736 * short write if we only get caps for some pages.
738 if (!(iocb
->ki_filp
->f_flags
& O_DIRECT
) &&
739 !(inode
->i_sb
->s_flags
& MS_SYNCHRONOUS
) &&
740 !(fi
->flags
& CEPH_F_SYNC
)) {
741 ret
= generic_file_aio_write(iocb
, iov
, nr_segs
, pos
);
745 if ((ret
>= 0 || ret
== -EIOCBQUEUED
) &&
746 ((file
->f_flags
& O_SYNC
) || IS_SYNC(file
->f_mapping
->host
)
747 || ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_NEARFULL
))) {
748 err
= vfs_fsync_range(file
, pos
, pos
+ written
- 1, 1);
752 if ((ret
< 0 && ret
!= -EAGAIN
) || pos
+ written
>= endoff
)
756 dout("aio_write %p %llx.%llx %llu~%u getting caps. i_size %llu\n",
757 inode
, ceph_vinop(inode
), pos
+ written
,
758 (unsigned)iov
->iov_len
- written
, inode
->i_size
);
759 ret
= ceph_get_caps(ci
, CEPH_CAP_FILE_WR
, 0, &got
, endoff
);
763 dout("aio_write %p %llx.%llx %llu~%u got cap refs on %s\n",
764 inode
, ceph_vinop(inode
), pos
+ written
,
765 (unsigned)iov
->iov_len
- written
, ceph_cap_string(got
));
766 ret
= ceph_sync_write(file
, iov
->iov_base
+ written
,
767 iov
->iov_len
- written
, &iocb
->ki_pos
);
770 spin_lock(&ci
->i_ceph_lock
);
771 dirty
= __ceph_mark_dirty_caps(ci
, CEPH_CAP_FILE_WR
);
772 spin_unlock(&ci
->i_ceph_lock
);
774 __mark_inode_dirty(inode
, dirty
);
776 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
777 inode
, ceph_vinop(inode
), pos
+ written
,
778 (unsigned)iov
->iov_len
- written
, ceph_cap_string(got
));
779 ceph_put_cap_refs(ci
, got
);
781 if (ret
== -EOLDSNAPC
) {
782 dout("aio_write %p %llx.%llx %llu~%u got EOLDSNAPC, retrying\n",
783 inode
, ceph_vinop(inode
), pos
, (unsigned)iov
->iov_len
);
791 * llseek. be sure to verify file size on SEEK_END.
793 static loff_t
ceph_llseek(struct file
*file
, loff_t offset
, int whence
)
795 struct inode
*inode
= file
->f_mapping
->host
;
798 mutex_lock(&inode
->i_mutex
);
799 __ceph_do_pending_vmtruncate(inode
);
801 if (whence
== SEEK_END
|| whence
== SEEK_DATA
|| whence
== SEEK_HOLE
) {
802 ret
= ceph_do_getattr(inode
, CEPH_STAT_CAP_SIZE
);
811 offset
+= inode
->i_size
;
815 * Here we special-case the lseek(fd, 0, SEEK_CUR)
816 * position-querying operation. Avoid rewriting the "same"
817 * f_pos value back to the file because a concurrent read(),
818 * write() or lseek() might have altered it
821 offset
= file
->f_pos
;
824 offset
+= file
->f_pos
;
827 if (offset
>= inode
->i_size
) {
833 if (offset
>= inode
->i_size
) {
837 offset
= inode
->i_size
;
841 if (offset
< 0 || offset
> inode
->i_sb
->s_maxbytes
) {
846 /* Special lock needed here? */
847 if (offset
!= file
->f_pos
) {
848 file
->f_pos
= offset
;
853 mutex_unlock(&inode
->i_mutex
);
857 const struct file_operations ceph_file_fops
= {
859 .release
= ceph_release
,
860 .llseek
= ceph_llseek
,
861 .read
= do_sync_read
,
862 .write
= do_sync_write
,
863 .aio_read
= ceph_aio_read
,
864 .aio_write
= ceph_aio_write
,
869 .splice_read
= generic_file_splice_read
,
870 .splice_write
= generic_file_splice_write
,
871 .unlocked_ioctl
= ceph_ioctl
,
872 .compat_ioctl
= ceph_ioctl
,