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/namei.h>
8 #include <linux/writeback.h>
11 #include "mds_client.h"
14 * Ceph file operations
16 * Implement basic open/close functionality, and implement
19 * We implement three modes of file I/O:
20 * - buffered uses the generic_file_aio_{read,write} helpers
22 * - synchronous is used when there is multi-client read/write
23 * sharing, avoids the page cache, and synchronously waits for an
26 * - direct io takes the variant of the sync path that references
27 * user pages directly.
29 * fsync() flushes and waits on dirty pages, but just queues metadata
30 * for writeback: since the MDS can recover size and mtime there is no
31 * need to wait for MDS acknowledgement.
36 * Prepare an open request. Preallocate ceph_cap to avoid an
37 * inopportune ENOMEM later.
39 static struct ceph_mds_request
*
40 prepare_open_request(struct super_block
*sb
, int flags
, int create_mode
)
42 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
43 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
44 struct ceph_mds_request
*req
;
45 int want_auth
= USE_ANY_MDS
;
46 int op
= (flags
& O_CREAT
) ? CEPH_MDS_OP_CREATE
: CEPH_MDS_OP_OPEN
;
48 if (flags
& (O_WRONLY
|O_RDWR
|O_CREAT
|O_TRUNC
))
49 want_auth
= USE_AUTH_MDS
;
51 req
= ceph_mdsc_create_request(mdsc
, op
, want_auth
);
54 req
->r_fmode
= ceph_flags_to_mode(flags
);
55 req
->r_args
.open
.flags
= cpu_to_le32(flags
);
56 req
->r_args
.open
.mode
= cpu_to_le32(create_mode
);
57 req
->r_args
.open
.preferred
= cpu_to_le32(-1);
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 the filp already has private_data, that means the file was
111 * already opened by intent during lookup, and we do nothing.
113 * If we already have the requisite capabilities, we can satisfy
114 * the open request locally (no need to request new caps from the
115 * MDS). We do, however, need to inform the MDS (asynchronously)
116 * if our wanted caps set expands.
118 int ceph_open(struct inode
*inode
, struct file
*file
)
120 struct ceph_inode_info
*ci
= ceph_inode(inode
);
121 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
122 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
123 struct ceph_mds_request
*req
;
124 struct ceph_file_info
*cf
= file
->private_data
;
125 struct inode
*parent_inode
= file
->f_dentry
->d_parent
->d_inode
;
127 int flags
, fmode
, wanted
;
130 dout("open file %p is already opened\n", file
);
134 /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */
135 flags
= file
->f_flags
& ~(O_CREAT
|O_EXCL
);
136 if (S_ISDIR(inode
->i_mode
))
137 flags
= O_DIRECTORY
; /* mds likes to know */
139 dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode
,
140 ceph_vinop(inode
), file
, flags
, file
->f_flags
);
141 fmode
= ceph_flags_to_mode(flags
);
142 wanted
= ceph_caps_for_mode(fmode
);
144 /* snapped files are read-only */
145 if (ceph_snap(inode
) != CEPH_NOSNAP
&& (file
->f_mode
& FMODE_WRITE
))
148 /* trivially open snapdir */
149 if (ceph_snap(inode
) == CEPH_SNAPDIR
) {
150 spin_lock(&inode
->i_lock
);
151 __ceph_get_fmode(ci
, fmode
);
152 spin_unlock(&inode
->i_lock
);
153 return ceph_init_file(inode
, file
, fmode
);
157 * No need to block if we have caps on the auth MDS (for
158 * write) or any MDS (for read). Update wanted set
161 spin_lock(&inode
->i_lock
);
162 if (__ceph_is_any_real_caps(ci
) &&
163 (((fmode
& CEPH_FILE_MODE_WR
) == 0) || ci
->i_auth_cap
)) {
164 int mds_wanted
= __ceph_caps_mds_wanted(ci
);
165 int issued
= __ceph_caps_issued(ci
, NULL
);
167 dout("open %p fmode %d want %s issued %s using existing\n",
168 inode
, fmode
, ceph_cap_string(wanted
),
169 ceph_cap_string(issued
));
170 __ceph_get_fmode(ci
, fmode
);
171 spin_unlock(&inode
->i_lock
);
174 if ((issued
& wanted
) != wanted
&&
175 (mds_wanted
& wanted
) != wanted
&&
176 ceph_snap(inode
) != CEPH_SNAPDIR
)
177 ceph_check_caps(ci
, 0, NULL
);
179 return ceph_init_file(inode
, file
, fmode
);
180 } else if (ceph_snap(inode
) != CEPH_NOSNAP
&&
181 (ci
->i_snap_caps
& wanted
) == wanted
) {
182 __ceph_get_fmode(ci
, fmode
);
183 spin_unlock(&inode
->i_lock
);
184 return ceph_init_file(inode
, file
, fmode
);
186 spin_unlock(&inode
->i_lock
);
188 dout("open fmode %d wants %s\n", fmode
, ceph_cap_string(wanted
));
189 req
= prepare_open_request(inode
->i_sb
, flags
, 0);
194 req
->r_inode
= inode
;
197 err
= ceph_mdsc_do_request(mdsc
, (flags
& (O_CREAT
|O_TRUNC
)) ?
198 parent_inode
: NULL
, req
);
200 err
= ceph_init_file(inode
, file
, req
->r_fmode
);
201 ceph_mdsc_put_request(req
);
202 dout("open result=%d on %llx.%llx\n", err
, ceph_vinop(inode
));
209 * Do a lookup + open with a single request.
211 * If this succeeds, but some subsequent check in the vfs
212 * may_open() fails, the struct *file gets cleaned up (i.e.
213 * ceph_release gets called). So fear not!
217 * path_lookup_open -> LOOKUP_OPEN
218 * path_lookup_create -> LOOKUP_OPEN|LOOKUP_CREATE
220 struct dentry
*ceph_lookup_open(struct inode
*dir
, struct dentry
*dentry
,
221 struct nameidata
*nd
, int mode
,
224 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
225 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
227 struct ceph_mds_request
*req
;
230 int flags
= nd
->intent
.open
.flags
- 1; /* silly vfs! */
232 dout("ceph_lookup_open dentry %p '%.*s' flags %d mode 0%o\n",
233 dentry
, dentry
->d_name
.len
, dentry
->d_name
.name
, flags
, mode
);
236 req
= prepare_open_request(dir
->i_sb
, flags
, mode
);
238 return ERR_CAST(req
);
239 req
->r_dentry
= dget(dentry
);
241 if (flags
& O_CREAT
) {
242 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
243 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
245 req
->r_locked_dir
= dir
; /* caller holds dir->i_mutex */
246 err
= ceph_mdsc_do_request(mdsc
,
247 (flags
& (O_CREAT
|O_TRUNC
)) ? dir
: NULL
,
249 err
= ceph_handle_snapdir(req
, dentry
, err
);
252 if ((flags
& O_CREAT
) && !req
->r_reply_info
.head
->is_dentry
)
253 err
= ceph_handle_notrace_create(dir
, dentry
);
256 file
= lookup_instantiate_filp(nd
, req
->r_dentry
, ceph_open
);
260 ret
= ceph_finish_lookup(req
, dentry
, err
);
261 ceph_mdsc_put_request(req
);
262 dout("ceph_lookup_open result=%p\n", ret
);
266 int ceph_release(struct inode
*inode
, struct file
*file
)
268 struct ceph_inode_info
*ci
= ceph_inode(inode
);
269 struct ceph_file_info
*cf
= file
->private_data
;
271 dout("release inode %p file %p\n", inode
, file
);
272 ceph_put_fmode(ci
, cf
->fmode
);
273 if (cf
->last_readdir
)
274 ceph_mdsc_put_request(cf
->last_readdir
);
275 kfree(cf
->last_name
);
278 kmem_cache_free(ceph_file_cachep
, cf
);
280 /* wake up anyone waiting for caps on this inode */
281 wake_up_all(&ci
->i_cap_wq
);
286 * Read a range of bytes striped over one or more objects. Iterate over
287 * objects we stripe over. (That's not atomic, but good enough for now.)
289 * If we get a short result from the OSD, check against i_size; we need to
290 * only return a short read to the caller if we hit EOF.
292 static int striped_read(struct inode
*inode
,
294 struct page
**pages
, int num_pages
,
295 int *checkeof
, bool o_direct
,
296 unsigned long buf_align
)
298 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
299 struct ceph_inode_info
*ci
= ceph_inode(inode
);
301 int io_align
, page_align
;
302 int left
, pages_left
;
304 struct page
**page_pos
;
306 bool hit_stripe
, was_short
;
309 * we may need to do multiple reads. not atomic, unfortunately.
314 pages_left
= num_pages
;
316 io_align
= off
& ~PAGE_MASK
;
320 page_align
= (pos
- io_align
+ buf_align
) & ~PAGE_MASK
;
322 page_align
= pos
& ~PAGE_MASK
;
324 ret
= ceph_osdc_readpages(&fsc
->client
->osdc
, ceph_vino(inode
),
325 &ci
->i_layout
, pos
, &this_len
,
328 page_pos
, pages_left
, page_align
);
331 hit_stripe
= this_len
< left
;
332 was_short
= ret
>= 0 && ret
< this_len
;
333 dout("striped_read %llu~%u (read %u) got %d%s%s\n", pos
, left
, read
,
334 ret
, hit_stripe
? " HITSTRIPE" : "", was_short
? " SHORT" : "");
337 int didpages
= (page_align
+ ret
) >> PAGE_CACHE_SHIFT
;
339 if (read
< pos
- off
) {
340 dout(" zero gap %llu to %llu\n", off
+ read
, pos
);
341 ceph_zero_page_vector_range(page_align
+ read
,
342 pos
- off
- read
, pages
);
347 page_pos
+= didpages
;
348 pages_left
-= didpages
;
351 if (left
&& hit_stripe
)
356 /* did we bounce off eof? */
357 if (pos
+ left
> inode
->i_size
)
360 /* zero trailing bytes (inside i_size) */
361 if (left
> 0 && pos
< inode
->i_size
) {
362 if (pos
+ left
> inode
->i_size
)
363 left
= inode
->i_size
- pos
;
365 dout("zero tail %d\n", left
);
366 ceph_zero_page_vector_range(page_align
+ read
, left
,
374 dout("striped_read returns %d\n", ret
);
379 * Completely synchronous read and write methods. Direct from __user
380 * buffer to osd, or directly to user pages (if O_DIRECT).
382 * If the read spans object boundary, just do multiple reads.
384 static ssize_t
ceph_sync_read(struct file
*file
, char __user
*data
,
385 unsigned len
, loff_t
*poff
, int *checkeof
)
387 struct inode
*inode
= file
->f_dentry
->d_inode
;
392 dout("sync_read on file %p %llu~%u %s\n", file
, off
, len
,
393 (file
->f_flags
& O_DIRECT
) ? "O_DIRECT" : "");
395 if (file
->f_flags
& O_DIRECT
) {
396 num_pages
= calc_pages_for((unsigned long)data
, len
);
397 pages
= ceph_get_direct_page_vector(data
, num_pages
, true);
399 num_pages
= calc_pages_for(off
, len
);
400 pages
= ceph_alloc_page_vector(num_pages
, GFP_NOFS
);
403 return PTR_ERR(pages
);
406 * flush any page cache pages in this range. this
407 * will make concurrent normal and sync io slow,
408 * but it will at least behave sensibly when they are
411 ret
= filemap_write_and_wait(inode
->i_mapping
);
415 ret
= striped_read(inode
, off
, len
, pages
, num_pages
, checkeof
,
416 file
->f_flags
& O_DIRECT
,
417 (unsigned long)data
& ~PAGE_MASK
);
419 if (ret
>= 0 && (file
->f_flags
& O_DIRECT
) == 0)
420 ret
= ceph_copy_page_vector_to_user(pages
, data
, off
, ret
);
425 if (file
->f_flags
& O_DIRECT
)
426 ceph_put_page_vector(pages
, num_pages
, true);
428 ceph_release_page_vector(pages
, num_pages
);
429 dout("sync_read result %d\n", ret
);
434 * Write commit callback, called if we requested both an ACK and
435 * ONDISK commit reply from the OSD.
437 static void sync_write_commit(struct ceph_osd_request
*req
,
438 struct ceph_msg
*msg
)
440 struct ceph_inode_info
*ci
= ceph_inode(req
->r_inode
);
442 dout("sync_write_commit %p tid %llu\n", req
, req
->r_tid
);
443 spin_lock(&ci
->i_unsafe_lock
);
444 list_del_init(&req
->r_unsafe_item
);
445 spin_unlock(&ci
->i_unsafe_lock
);
446 ceph_put_cap_refs(ci
, CEPH_CAP_FILE_WR
);
450 * Synchronous write, straight from __user pointer or user pages (if
453 * If write spans object boundary, just do multiple writes. (For a
454 * correct atomic write, we should e.g. take write locks on all
455 * objects, rollback on failure, etc.)
457 static ssize_t
ceph_sync_write(struct file
*file
, const char __user
*data
,
458 size_t left
, loff_t
*offset
)
460 struct inode
*inode
= file
->f_dentry
->d_inode
;
461 struct ceph_inode_info
*ci
= ceph_inode(inode
);
462 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
463 struct ceph_osd_request
*req
;
466 long long unsigned pos
;
472 int page_align
, io_align
;
473 unsigned long buf_align
;
475 struct timespec mtime
= CURRENT_TIME
;
477 if (ceph_snap(file
->f_dentry
->d_inode
) != CEPH_NOSNAP
)
480 dout("sync_write on file %p %lld~%u %s\n", file
, *offset
,
481 (unsigned)left
, (file
->f_flags
& O_DIRECT
) ? "O_DIRECT" : "");
483 if (file
->f_flags
& O_APPEND
)
484 pos
= i_size_read(inode
);
488 ret
= filemap_write_and_wait_range(inode
->i_mapping
, pos
, pos
+ left
);
492 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
493 pos
>> PAGE_CACHE_SHIFT
,
494 (pos
+ left
) >> PAGE_CACHE_SHIFT
);
496 dout("invalidate_inode_pages2_range returned %d\n", ret
);
498 flags
= CEPH_OSD_FLAG_ORDERSNAP
|
499 CEPH_OSD_FLAG_ONDISK
|
501 if ((file
->f_flags
& (O_SYNC
|O_DIRECT
)) == 0)
502 flags
|= CEPH_OSD_FLAG_ACK
;
507 * we may need to do multiple writes here if we span an object
508 * boundary. this isn't atomic, unfortunately. :(
511 io_align
= pos
& ~PAGE_MASK
;
512 buf_align
= (unsigned long)data
& ~PAGE_MASK
;
514 if (file
->f_flags
& O_DIRECT
) {
515 /* write from beginning of first page, regardless of
517 page_align
= (pos
- io_align
+ buf_align
) & ~PAGE_MASK
;
518 num_pages
= calc_pages_for((unsigned long)data
, len
);
520 page_align
= pos
& ~PAGE_MASK
;
521 num_pages
= calc_pages_for(pos
, len
);
523 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
524 ceph_vino(inode
), pos
, &len
,
525 CEPH_OSD_OP_WRITE
, flags
,
526 ci
->i_snap_realm
->cached_context
,
528 ci
->i_truncate_seq
, ci
->i_truncate_size
,
529 &mtime
, false, 2, page_align
);
533 if (file
->f_flags
& O_DIRECT
) {
534 pages
= ceph_get_direct_page_vector(data
, num_pages
, false);
536 ret
= PTR_ERR(pages
);
541 * throw out any page cache pages in this range. this
544 truncate_inode_pages_range(inode
->i_mapping
, pos
,
545 (pos
+len
) | (PAGE_CACHE_SIZE
-1));
547 pages
= ceph_alloc_page_vector(num_pages
, GFP_NOFS
);
549 ret
= PTR_ERR(pages
);
552 ret
= ceph_copy_user_to_page_vector(pages
, data
, pos
, len
);
554 ceph_release_page_vector(pages
, num_pages
);
558 if ((file
->f_flags
& O_SYNC
) == 0) {
559 /* get a second commit callback */
560 req
->r_safe_callback
= sync_write_commit
;
561 req
->r_own_pages
= 1;
564 req
->r_pages
= pages
;
565 req
->r_num_pages
= num_pages
;
566 req
->r_inode
= inode
;
568 ret
= ceph_osdc_start_request(&fsc
->client
->osdc
, req
, false);
570 if (req
->r_safe_callback
) {
572 * Add to inode unsafe list only after we
573 * start_request so that a tid has been assigned.
575 spin_lock(&ci
->i_unsafe_lock
);
576 list_add_tail(&req
->r_unsafe_item
,
577 &ci
->i_unsafe_writes
);
578 spin_unlock(&ci
->i_unsafe_lock
);
579 ceph_get_cap_refs(ci
, CEPH_CAP_FILE_WR
);
582 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
583 if (ret
< 0 && req
->r_safe_callback
) {
584 spin_lock(&ci
->i_unsafe_lock
);
585 list_del_init(&req
->r_unsafe_item
);
586 spin_unlock(&ci
->i_unsafe_lock
);
587 ceph_put_cap_refs(ci
, CEPH_CAP_FILE_WR
);
591 if (file
->f_flags
& O_DIRECT
)
592 ceph_put_page_vector(pages
, num_pages
, false);
593 else if (file
->f_flags
& O_SYNC
)
594 ceph_release_page_vector(pages
, num_pages
);
597 ceph_osdc_put_request(req
);
608 if (pos
> i_size_read(inode
))
609 check_caps
= ceph_inode_set_size(inode
, pos
);
611 ceph_check_caps(ceph_inode(inode
), CHECK_CAPS_AUTHONLY
,
618 * Wrap generic_file_aio_read with checks for cap bits on the inode.
619 * Atomically grab references, so that those bits are not released
620 * back to the MDS mid-read.
622 * Hmm, the sync read case isn't actually async... should it be?
624 static ssize_t
ceph_aio_read(struct kiocb
*iocb
, const struct iovec
*iov
,
625 unsigned long nr_segs
, loff_t pos
)
627 struct file
*filp
= iocb
->ki_filp
;
628 struct ceph_file_info
*fi
= filp
->private_data
;
629 loff_t
*ppos
= &iocb
->ki_pos
;
630 size_t len
= iov
->iov_len
;
631 struct inode
*inode
= filp
->f_dentry
->d_inode
;
632 struct ceph_inode_info
*ci
= ceph_inode(inode
);
633 void __user
*base
= iov
->iov_base
;
636 int checkeof
= 0, read
= 0;
638 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
639 inode
, ceph_vinop(inode
), pos
, (unsigned)len
, inode
);
641 __ceph_do_pending_vmtruncate(inode
);
642 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
643 want
= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_LAZYIO
;
645 want
= CEPH_CAP_FILE_CACHE
;
646 ret
= ceph_get_caps(ci
, CEPH_CAP_FILE_RD
, want
, &got
, -1);
649 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
650 inode
, ceph_vinop(inode
), pos
, (unsigned)len
,
651 ceph_cap_string(got
));
653 if ((got
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)) == 0 ||
654 (iocb
->ki_filp
->f_flags
& O_DIRECT
) ||
655 (inode
->i_sb
->s_flags
& MS_SYNCHRONOUS
) ||
656 (fi
->flags
& CEPH_F_SYNC
))
657 /* hmm, this isn't really async... */
658 ret
= ceph_sync_read(filp
, base
, len
, ppos
, &checkeof
);
660 ret
= generic_file_aio_read(iocb
, iov
, nr_segs
, pos
);
663 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
664 inode
, ceph_vinop(inode
), ceph_cap_string(got
), (int)ret
);
665 ceph_put_cap_refs(ci
, got
);
667 if (checkeof
&& ret
>= 0) {
668 int statret
= ceph_do_getattr(inode
, CEPH_STAT_CAP_SIZE
);
670 /* hit EOF or hole? */
671 if (statret
== 0 && *ppos
< inode
->i_size
) {
672 dout("aio_read sync_read hit hole, ppos %lld < size %lld, reading more\n", *ppos
, inode
->i_size
);
687 * Take cap references to avoid releasing caps to MDS mid-write.
689 * If we are synchronous, and write with an old snap context, the OSD
690 * may return EOLDSNAPC. In that case, retry the write.. _after_
691 * dropping our cap refs and allowing the pending snap to logically
692 * complete _before_ this write occurs.
694 * If we are near ENOSPC, write synchronously.
696 static ssize_t
ceph_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
697 unsigned long nr_segs
, loff_t pos
)
699 struct file
*file
= iocb
->ki_filp
;
700 struct ceph_file_info
*fi
= file
->private_data
;
701 struct inode
*inode
= file
->f_dentry
->d_inode
;
702 struct ceph_inode_info
*ci
= ceph_inode(inode
);
703 struct ceph_osd_client
*osdc
=
704 &ceph_sb_to_client(inode
->i_sb
)->client
->osdc
;
705 loff_t endoff
= pos
+ iov
->iov_len
;
709 if (ceph_snap(inode
) != CEPH_NOSNAP
)
713 if (ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_FULL
))
715 __ceph_do_pending_vmtruncate(inode
);
716 dout("aio_write %p %llx.%llx %llu~%u getting caps. i_size %llu\n",
717 inode
, ceph_vinop(inode
), pos
, (unsigned)iov
->iov_len
,
719 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
720 want
= CEPH_CAP_FILE_BUFFER
| CEPH_CAP_FILE_LAZYIO
;
722 want
= CEPH_CAP_FILE_BUFFER
;
723 ret
= ceph_get_caps(ci
, CEPH_CAP_FILE_WR
, want
, &got
, endoff
);
727 dout("aio_write %p %llx.%llx %llu~%u got cap refs on %s\n",
728 inode
, ceph_vinop(inode
), pos
, (unsigned)iov
->iov_len
,
729 ceph_cap_string(got
));
731 if ((got
& (CEPH_CAP_FILE_BUFFER
|CEPH_CAP_FILE_LAZYIO
)) == 0 ||
732 (iocb
->ki_filp
->f_flags
& O_DIRECT
) ||
733 (inode
->i_sb
->s_flags
& MS_SYNCHRONOUS
) ||
734 (fi
->flags
& CEPH_F_SYNC
)) {
735 ret
= ceph_sync_write(file
, iov
->iov_base
, iov
->iov_len
,
739 * buffered write; drop Fw early to avoid slow
740 * revocation if we get stuck on balance_dirty_pages
744 spin_lock(&inode
->i_lock
);
745 dirty
= __ceph_mark_dirty_caps(ci
, CEPH_CAP_FILE_WR
);
746 spin_unlock(&inode
->i_lock
);
747 ceph_put_cap_refs(ci
, got
);
749 ret
= generic_file_aio_write(iocb
, iov
, nr_segs
, pos
);
750 if ((ret
>= 0 || ret
== -EIOCBQUEUED
) &&
751 ((file
->f_flags
& O_SYNC
) || IS_SYNC(file
->f_mapping
->host
)
752 || ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_NEARFULL
))) {
753 err
= vfs_fsync_range(file
, pos
, pos
+ ret
- 1, 1);
759 __mark_inode_dirty(inode
, dirty
);
765 spin_lock(&inode
->i_lock
);
766 dirty
= __ceph_mark_dirty_caps(ci
, CEPH_CAP_FILE_WR
);
767 spin_unlock(&inode
->i_lock
);
769 __mark_inode_dirty(inode
, dirty
);
773 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
774 inode
, ceph_vinop(inode
), pos
, (unsigned)iov
->iov_len
,
775 ceph_cap_string(got
));
776 ceph_put_cap_refs(ci
, got
);
779 if (ret
== -EOLDSNAPC
) {
780 dout("aio_write %p %llx.%llx %llu~%u got EOLDSNAPC, retrying\n",
781 inode
, ceph_vinop(inode
), pos
, (unsigned)iov
->iov_len
);
789 * llseek. be sure to verify file size on SEEK_END.
791 static loff_t
ceph_llseek(struct file
*file
, loff_t offset
, int origin
)
793 struct inode
*inode
= file
->f_mapping
->host
;
796 mutex_lock(&inode
->i_mutex
);
797 __ceph_do_pending_vmtruncate(inode
);
800 ret
= ceph_do_getattr(inode
, CEPH_STAT_CAP_SIZE
);
805 offset
+= inode
->i_size
;
809 * Here we special-case the lseek(fd, 0, SEEK_CUR)
810 * position-querying operation. Avoid rewriting the "same"
811 * f_pos value back to the file because a concurrent read(),
812 * write() or lseek() might have altered it
815 offset
= file
->f_pos
;
818 offset
+= file
->f_pos
;
822 if (offset
< 0 || offset
> inode
->i_sb
->s_maxbytes
) {
827 /* Special lock needed here? */
828 if (offset
!= file
->f_pos
) {
829 file
->f_pos
= offset
;
834 mutex_unlock(&inode
->i_mutex
);
838 const struct file_operations ceph_file_fops
= {
840 .release
= ceph_release
,
841 .llseek
= ceph_llseek
,
842 .read
= do_sync_read
,
843 .write
= do_sync_write
,
844 .aio_read
= ceph_aio_read
,
845 .aio_write
= ceph_aio_write
,
850 .splice_read
= generic_file_splice_read
,
851 .splice_write
= generic_file_splice_write
,
852 .unlocked_ioctl
= ceph_ioctl
,
853 .compat_ioctl
= ceph_ioctl
,