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
);
62 * initialize private struct file data.
63 * if we fail, clean up by dropping fmode reference on the ceph_inode
65 static int ceph_init_file(struct inode
*inode
, struct file
*file
, int fmode
)
67 struct ceph_file_info
*cf
;
70 switch (inode
->i_mode
& S_IFMT
) {
73 dout("init_file %p %p 0%o (regular)\n", inode
, file
,
75 cf
= kmem_cache_alloc(ceph_file_cachep
, GFP_NOFS
| __GFP_ZERO
);
77 ceph_put_fmode(ceph_inode(inode
), fmode
); /* clean up */
82 file
->private_data
= cf
;
83 BUG_ON(inode
->i_fop
->release
!= ceph_release
);
87 dout("init_file %p %p 0%o (symlink)\n", inode
, file
,
89 ceph_put_fmode(ceph_inode(inode
), fmode
); /* clean up */
93 dout("init_file %p %p 0%o (special)\n", inode
, file
,
96 * we need to drop the open ref now, since we don't
97 * have .release set to ceph_release.
99 ceph_put_fmode(ceph_inode(inode
), fmode
); /* clean up */
100 BUG_ON(inode
->i_fop
->release
== ceph_release
);
102 /* call the proper open fop */
103 ret
= inode
->i_fop
->open(inode
, file
);
109 * If the filp already has private_data, that means the file was
110 * already opened by intent during lookup, and we do nothing.
112 * If we already have the requisite capabilities, we can satisfy
113 * the open request locally (no need to request new caps from the
114 * MDS). We do, however, need to inform the MDS (asynchronously)
115 * if our wanted caps set expands.
117 int ceph_open(struct inode
*inode
, struct file
*file
)
119 struct ceph_inode_info
*ci
= ceph_inode(inode
);
120 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
121 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
122 struct ceph_mds_request
*req
;
123 struct ceph_file_info
*cf
= file
->private_data
;
124 struct inode
*parent_inode
= NULL
;
126 int flags
, fmode
, wanted
;
129 dout("open file %p is already opened\n", file
);
133 /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */
134 flags
= file
->f_flags
& ~(O_CREAT
|O_EXCL
);
135 if (S_ISDIR(inode
->i_mode
))
136 flags
= O_DIRECTORY
; /* mds likes to know */
138 dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode
,
139 ceph_vinop(inode
), file
, flags
, file
->f_flags
);
140 fmode
= ceph_flags_to_mode(flags
);
141 wanted
= ceph_caps_for_mode(fmode
);
143 /* snapped files are read-only */
144 if (ceph_snap(inode
) != CEPH_NOSNAP
&& (file
->f_mode
& FMODE_WRITE
))
147 /* trivially open snapdir */
148 if (ceph_snap(inode
) == CEPH_SNAPDIR
) {
149 spin_lock(&ci
->i_ceph_lock
);
150 __ceph_get_fmode(ci
, fmode
);
151 spin_unlock(&ci
->i_ceph_lock
);
152 return ceph_init_file(inode
, file
, fmode
);
156 * No need to block if we have caps on the auth MDS (for
157 * write) or any MDS (for read). Update wanted set
160 spin_lock(&ci
->i_ceph_lock
);
161 if (__ceph_is_any_real_caps(ci
) &&
162 (((fmode
& CEPH_FILE_MODE_WR
) == 0) || ci
->i_auth_cap
)) {
163 int mds_wanted
= __ceph_caps_mds_wanted(ci
);
164 int issued
= __ceph_caps_issued(ci
, NULL
);
166 dout("open %p fmode %d want %s issued %s using existing\n",
167 inode
, fmode
, ceph_cap_string(wanted
),
168 ceph_cap_string(issued
));
169 __ceph_get_fmode(ci
, fmode
);
170 spin_unlock(&ci
->i_ceph_lock
);
173 if ((issued
& wanted
) != wanted
&&
174 (mds_wanted
& wanted
) != wanted
&&
175 ceph_snap(inode
) != CEPH_SNAPDIR
)
176 ceph_check_caps(ci
, 0, NULL
);
178 return ceph_init_file(inode
, file
, fmode
);
179 } else if (ceph_snap(inode
) != CEPH_NOSNAP
&&
180 (ci
->i_snap_caps
& wanted
) == wanted
) {
181 __ceph_get_fmode(ci
, fmode
);
182 spin_unlock(&ci
->i_ceph_lock
);
183 return ceph_init_file(inode
, file
, fmode
);
185 spin_unlock(&ci
->i_ceph_lock
);
187 dout("open fmode %d wants %s\n", fmode
, ceph_cap_string(wanted
));
188 req
= prepare_open_request(inode
->i_sb
, flags
, 0);
193 req
->r_inode
= inode
;
196 if (flags
& (O_CREAT
|O_TRUNC
))
197 parent_inode
= ceph_get_dentry_parent_inode(file
->f_dentry
);
198 err
= ceph_mdsc_do_request(mdsc
, parent_inode
, req
);
201 err
= ceph_init_file(inode
, file
, req
->r_fmode
);
202 ceph_mdsc_put_request(req
);
203 dout("open result=%d on %llx.%llx\n", err
, ceph_vinop(inode
));
210 * Do a lookup + open with a single request.
212 * If this succeeds, but some subsequent check in the vfs
213 * may_open() fails, the struct *file gets cleaned up (i.e.
214 * ceph_release gets called). So fear not!
216 struct file
*ceph_lookup_open(struct inode
*dir
, struct dentry
*dentry
,
217 struct opendata
*od
, unsigned flags
, umode_t mode
)
219 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
220 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
221 struct file
*file
= NULL
;
222 struct ceph_mds_request
*req
;
226 dout("ceph_lookup_open dentry %p '%.*s' flags %d mode 0%o\n",
227 dentry
, dentry
->d_name
.len
, dentry
->d_name
.name
, flags
, mode
);
230 req
= prepare_open_request(dir
->i_sb
, flags
, mode
);
232 return ERR_CAST(req
);
233 req
->r_dentry
= dget(dentry
);
235 if (flags
& O_CREAT
) {
236 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
237 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
239 req
->r_locked_dir
= dir
; /* caller holds dir->i_mutex */
240 err
= ceph_mdsc_do_request(mdsc
,
241 (flags
& (O_CREAT
|O_TRUNC
)) ? dir
: NULL
,
243 err
= ceph_handle_snapdir(req
, dentry
, err
);
246 if ((flags
& O_CREAT
) && !req
->r_reply_info
.head
->is_dentry
)
247 err
= ceph_handle_notrace_create(dir
, dentry
);
250 file
= finish_open(od
, req
->r_dentry
, ceph_open
);
254 ret
= ceph_finish_lookup(req
, dentry
, err
);
255 ceph_mdsc_put_request(req
);
256 dout("ceph_lookup_open result=%p\n", ret
);
259 return ERR_CAST(ret
);
262 return err
? ERR_PTR(err
) : file
;
265 int ceph_release(struct inode
*inode
, struct file
*file
)
267 struct ceph_inode_info
*ci
= ceph_inode(inode
);
268 struct ceph_file_info
*cf
= file
->private_data
;
270 dout("release inode %p file %p\n", inode
, file
);
271 ceph_put_fmode(ci
, cf
->fmode
);
272 if (cf
->last_readdir
)
273 ceph_mdsc_put_request(cf
->last_readdir
);
274 kfree(cf
->last_name
);
277 kmem_cache_free(ceph_file_cachep
, cf
);
279 /* wake up anyone waiting for caps on this inode */
280 wake_up_all(&ci
->i_cap_wq
);
285 * Read a range of bytes striped over one or more objects. Iterate over
286 * objects we stripe over. (That's not atomic, but good enough for now.)
288 * If we get a short result from the OSD, check against i_size; we need to
289 * only return a short read to the caller if we hit EOF.
291 static int striped_read(struct inode
*inode
,
293 struct page
**pages
, int num_pages
,
294 int *checkeof
, bool o_direct
,
295 unsigned long buf_align
)
297 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
298 struct ceph_inode_info
*ci
= ceph_inode(inode
);
300 int io_align
, page_align
;
301 int left
, pages_left
;
303 struct page
**page_pos
;
305 bool hit_stripe
, was_short
;
308 * we may need to do multiple reads. not atomic, unfortunately.
313 pages_left
= num_pages
;
315 io_align
= off
& ~PAGE_MASK
;
319 page_align
= (pos
- io_align
+ buf_align
) & ~PAGE_MASK
;
321 page_align
= pos
& ~PAGE_MASK
;
323 ret
= ceph_osdc_readpages(&fsc
->client
->osdc
, ceph_vino(inode
),
324 &ci
->i_layout
, pos
, &this_len
,
327 page_pos
, pages_left
, page_align
);
330 hit_stripe
= this_len
< left
;
331 was_short
= ret
>= 0 && ret
< this_len
;
332 dout("striped_read %llu~%u (read %u) got %d%s%s\n", pos
, left
, read
,
333 ret
, hit_stripe
? " HITSTRIPE" : "", was_short
? " SHORT" : "");
336 int didpages
= (page_align
+ ret
) >> PAGE_CACHE_SHIFT
;
338 if (read
< pos
- off
) {
339 dout(" zero gap %llu to %llu\n", off
+ read
, pos
);
340 ceph_zero_page_vector_range(page_align
+ read
,
341 pos
- off
- read
, pages
);
346 page_pos
+= didpages
;
347 pages_left
-= didpages
;
350 if (left
&& hit_stripe
)
355 /* did we bounce off eof? */
356 if (pos
+ left
> inode
->i_size
)
359 /* zero trailing bytes (inside i_size) */
360 if (left
> 0 && pos
< inode
->i_size
) {
361 if (pos
+ left
> inode
->i_size
)
362 left
= inode
->i_size
- pos
;
364 dout("zero tail %d\n", left
);
365 ceph_zero_page_vector_range(page_align
+ read
, left
,
373 dout("striped_read returns %d\n", ret
);
378 * Completely synchronous read and write methods. Direct from __user
379 * buffer to osd, or directly to user pages (if O_DIRECT).
381 * If the read spans object boundary, just do multiple reads.
383 static ssize_t
ceph_sync_read(struct file
*file
, char __user
*data
,
384 unsigned len
, loff_t
*poff
, int *checkeof
)
386 struct inode
*inode
= file
->f_dentry
->d_inode
;
391 dout("sync_read on file %p %llu~%u %s\n", file
, off
, len
,
392 (file
->f_flags
& O_DIRECT
) ? "O_DIRECT" : "");
394 if (file
->f_flags
& O_DIRECT
) {
395 num_pages
= calc_pages_for((unsigned long)data
, len
);
396 pages
= ceph_get_direct_page_vector(data
, num_pages
, true);
398 num_pages
= calc_pages_for(off
, len
);
399 pages
= ceph_alloc_page_vector(num_pages
, GFP_NOFS
);
402 return PTR_ERR(pages
);
405 * flush any page cache pages in this range. this
406 * will make concurrent normal and sync io slow,
407 * but it will at least behave sensibly when they are
410 ret
= filemap_write_and_wait(inode
->i_mapping
);
414 ret
= striped_read(inode
, off
, len
, pages
, num_pages
, checkeof
,
415 file
->f_flags
& O_DIRECT
,
416 (unsigned long)data
& ~PAGE_MASK
);
418 if (ret
>= 0 && (file
->f_flags
& O_DIRECT
) == 0)
419 ret
= ceph_copy_page_vector_to_user(pages
, data
, off
, ret
);
424 if (file
->f_flags
& O_DIRECT
)
425 ceph_put_page_vector(pages
, num_pages
, true);
427 ceph_release_page_vector(pages
, num_pages
);
428 dout("sync_read result %d\n", ret
);
433 * Write commit callback, called if we requested both an ACK and
434 * ONDISK commit reply from the OSD.
436 static void sync_write_commit(struct ceph_osd_request
*req
,
437 struct ceph_msg
*msg
)
439 struct ceph_inode_info
*ci
= ceph_inode(req
->r_inode
);
441 dout("sync_write_commit %p tid %llu\n", req
, req
->r_tid
);
442 spin_lock(&ci
->i_unsafe_lock
);
443 list_del_init(&req
->r_unsafe_item
);
444 spin_unlock(&ci
->i_unsafe_lock
);
445 ceph_put_cap_refs(ci
, CEPH_CAP_FILE_WR
);
449 * Synchronous write, straight from __user pointer or user pages (if
452 * If write spans object boundary, just do multiple writes. (For a
453 * correct atomic write, we should e.g. take write locks on all
454 * objects, rollback on failure, etc.)
456 static ssize_t
ceph_sync_write(struct file
*file
, const char __user
*data
,
457 size_t left
, loff_t
*offset
)
459 struct inode
*inode
= file
->f_dentry
->d_inode
;
460 struct ceph_inode_info
*ci
= ceph_inode(inode
);
461 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
462 struct ceph_osd_request
*req
;
465 long long unsigned pos
;
471 int page_align
, io_align
;
472 unsigned long buf_align
;
474 struct timespec mtime
= CURRENT_TIME
;
476 if (ceph_snap(file
->f_dentry
->d_inode
) != CEPH_NOSNAP
)
479 dout("sync_write on file %p %lld~%u %s\n", file
, *offset
,
480 (unsigned)left
, (file
->f_flags
& O_DIRECT
) ? "O_DIRECT" : "");
482 if (file
->f_flags
& O_APPEND
)
483 pos
= i_size_read(inode
);
487 ret
= filemap_write_and_wait_range(inode
->i_mapping
, pos
, pos
+ left
);
491 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
492 pos
>> PAGE_CACHE_SHIFT
,
493 (pos
+ left
) >> PAGE_CACHE_SHIFT
);
495 dout("invalidate_inode_pages2_range returned %d\n", ret
);
497 flags
= CEPH_OSD_FLAG_ORDERSNAP
|
498 CEPH_OSD_FLAG_ONDISK
|
500 if ((file
->f_flags
& (O_SYNC
|O_DIRECT
)) == 0)
501 flags
|= CEPH_OSD_FLAG_ACK
;
506 * we may need to do multiple writes here if we span an object
507 * boundary. this isn't atomic, unfortunately. :(
510 io_align
= pos
& ~PAGE_MASK
;
511 buf_align
= (unsigned long)data
& ~PAGE_MASK
;
513 if (file
->f_flags
& O_DIRECT
) {
514 /* write from beginning of first page, regardless of
516 page_align
= (pos
- io_align
+ buf_align
) & ~PAGE_MASK
;
517 num_pages
= calc_pages_for((unsigned long)data
, len
);
519 page_align
= pos
& ~PAGE_MASK
;
520 num_pages
= calc_pages_for(pos
, len
);
522 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
523 ceph_vino(inode
), pos
, &len
,
524 CEPH_OSD_OP_WRITE
, flags
,
525 ci
->i_snap_realm
->cached_context
,
527 ci
->i_truncate_seq
, ci
->i_truncate_size
,
528 &mtime
, false, 2, page_align
);
532 if (file
->f_flags
& O_DIRECT
) {
533 pages
= ceph_get_direct_page_vector(data
, num_pages
, false);
535 ret
= PTR_ERR(pages
);
540 * throw out any page cache pages in this range. this
543 truncate_inode_pages_range(inode
->i_mapping
, pos
,
544 (pos
+len
) | (PAGE_CACHE_SIZE
-1));
546 pages
= ceph_alloc_page_vector(num_pages
, GFP_NOFS
);
548 ret
= PTR_ERR(pages
);
551 ret
= ceph_copy_user_to_page_vector(pages
, data
, pos
, len
);
553 ceph_release_page_vector(pages
, num_pages
);
557 if ((file
->f_flags
& O_SYNC
) == 0) {
558 /* get a second commit callback */
559 req
->r_safe_callback
= sync_write_commit
;
560 req
->r_own_pages
= 1;
563 req
->r_pages
= pages
;
564 req
->r_num_pages
= num_pages
;
565 req
->r_inode
= inode
;
567 ret
= ceph_osdc_start_request(&fsc
->client
->osdc
, req
, false);
569 if (req
->r_safe_callback
) {
571 * Add to inode unsafe list only after we
572 * start_request so that a tid has been assigned.
574 spin_lock(&ci
->i_unsafe_lock
);
575 list_add_tail(&req
->r_unsafe_item
,
576 &ci
->i_unsafe_writes
);
577 spin_unlock(&ci
->i_unsafe_lock
);
578 ceph_get_cap_refs(ci
, CEPH_CAP_FILE_WR
);
581 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
582 if (ret
< 0 && req
->r_safe_callback
) {
583 spin_lock(&ci
->i_unsafe_lock
);
584 list_del_init(&req
->r_unsafe_item
);
585 spin_unlock(&ci
->i_unsafe_lock
);
586 ceph_put_cap_refs(ci
, CEPH_CAP_FILE_WR
);
590 if (file
->f_flags
& O_DIRECT
)
591 ceph_put_page_vector(pages
, num_pages
, false);
592 else if (file
->f_flags
& O_SYNC
)
593 ceph_release_page_vector(pages
, num_pages
);
596 ceph_osdc_put_request(req
);
607 if (pos
> i_size_read(inode
))
608 check_caps
= ceph_inode_set_size(inode
, pos
);
610 ceph_check_caps(ceph_inode(inode
), CHECK_CAPS_AUTHONLY
,
617 * Wrap generic_file_aio_read with checks for cap bits on the inode.
618 * Atomically grab references, so that those bits are not released
619 * back to the MDS mid-read.
621 * Hmm, the sync read case isn't actually async... should it be?
623 static ssize_t
ceph_aio_read(struct kiocb
*iocb
, const struct iovec
*iov
,
624 unsigned long nr_segs
, loff_t pos
)
626 struct file
*filp
= iocb
->ki_filp
;
627 struct ceph_file_info
*fi
= filp
->private_data
;
628 loff_t
*ppos
= &iocb
->ki_pos
;
629 size_t len
= iov
->iov_len
;
630 struct inode
*inode
= filp
->f_dentry
->d_inode
;
631 struct ceph_inode_info
*ci
= ceph_inode(inode
);
632 void __user
*base
= iov
->iov_base
;
635 int checkeof
= 0, read
= 0;
637 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
638 inode
, ceph_vinop(inode
), pos
, (unsigned)len
, inode
);
640 __ceph_do_pending_vmtruncate(inode
);
641 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
642 want
= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_LAZYIO
;
644 want
= CEPH_CAP_FILE_CACHE
;
645 ret
= ceph_get_caps(ci
, CEPH_CAP_FILE_RD
, want
, &got
, -1);
648 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
649 inode
, ceph_vinop(inode
), pos
, (unsigned)len
,
650 ceph_cap_string(got
));
652 if ((got
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)) == 0 ||
653 (iocb
->ki_filp
->f_flags
& O_DIRECT
) ||
654 (inode
->i_sb
->s_flags
& MS_SYNCHRONOUS
) ||
655 (fi
->flags
& CEPH_F_SYNC
))
656 /* hmm, this isn't really async... */
657 ret
= ceph_sync_read(filp
, base
, len
, ppos
, &checkeof
);
659 ret
= generic_file_aio_read(iocb
, iov
, nr_segs
, pos
);
662 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
663 inode
, ceph_vinop(inode
), ceph_cap_string(got
), (int)ret
);
664 ceph_put_cap_refs(ci
, got
);
666 if (checkeof
&& ret
>= 0) {
667 int statret
= ceph_do_getattr(inode
, CEPH_STAT_CAP_SIZE
);
669 /* hit EOF or hole? */
670 if (statret
== 0 && *ppos
< inode
->i_size
) {
671 dout("aio_read sync_read hit hole, ppos %lld < size %lld, reading more\n", *ppos
, inode
->i_size
);
686 * Take cap references to avoid releasing caps to MDS mid-write.
688 * If we are synchronous, and write with an old snap context, the OSD
689 * may return EOLDSNAPC. In that case, retry the write.. _after_
690 * dropping our cap refs and allowing the pending snap to logically
691 * complete _before_ this write occurs.
693 * If we are near ENOSPC, write synchronously.
695 static ssize_t
ceph_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
696 unsigned long nr_segs
, loff_t pos
)
698 struct file
*file
= iocb
->ki_filp
;
699 struct ceph_file_info
*fi
= file
->private_data
;
700 struct inode
*inode
= file
->f_dentry
->d_inode
;
701 struct ceph_inode_info
*ci
= ceph_inode(inode
);
702 struct ceph_osd_client
*osdc
=
703 &ceph_sb_to_client(inode
->i_sb
)->client
->osdc
;
704 loff_t endoff
= pos
+ iov
->iov_len
;
708 if (ceph_snap(inode
) != CEPH_NOSNAP
)
712 if (ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_FULL
))
714 __ceph_do_pending_vmtruncate(inode
);
715 dout("aio_write %p %llx.%llx %llu~%u getting caps. i_size %llu\n",
716 inode
, ceph_vinop(inode
), pos
, (unsigned)iov
->iov_len
,
718 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
719 want
= CEPH_CAP_FILE_BUFFER
| CEPH_CAP_FILE_LAZYIO
;
721 want
= CEPH_CAP_FILE_BUFFER
;
722 ret
= ceph_get_caps(ci
, CEPH_CAP_FILE_WR
, want
, &got
, endoff
);
726 dout("aio_write %p %llx.%llx %llu~%u got cap refs on %s\n",
727 inode
, ceph_vinop(inode
), pos
, (unsigned)iov
->iov_len
,
728 ceph_cap_string(got
));
730 if ((got
& (CEPH_CAP_FILE_BUFFER
|CEPH_CAP_FILE_LAZYIO
)) == 0 ||
731 (iocb
->ki_filp
->f_flags
& O_DIRECT
) ||
732 (inode
->i_sb
->s_flags
& MS_SYNCHRONOUS
) ||
733 (fi
->flags
& CEPH_F_SYNC
)) {
734 ret
= ceph_sync_write(file
, iov
->iov_base
, iov
->iov_len
,
738 * buffered write; drop Fw early to avoid slow
739 * revocation if we get stuck on balance_dirty_pages
743 spin_lock(&ci
->i_ceph_lock
);
744 dirty
= __ceph_mark_dirty_caps(ci
, CEPH_CAP_FILE_WR
);
745 spin_unlock(&ci
->i_ceph_lock
);
746 ceph_put_cap_refs(ci
, got
);
748 ret
= generic_file_aio_write(iocb
, iov
, nr_segs
, pos
);
749 if ((ret
>= 0 || ret
== -EIOCBQUEUED
) &&
750 ((file
->f_flags
& O_SYNC
) || IS_SYNC(file
->f_mapping
->host
)
751 || ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_NEARFULL
))) {
752 err
= vfs_fsync_range(file
, pos
, pos
+ ret
- 1, 1);
758 __mark_inode_dirty(inode
, dirty
);
764 spin_lock(&ci
->i_ceph_lock
);
765 dirty
= __ceph_mark_dirty_caps(ci
, CEPH_CAP_FILE_WR
);
766 spin_unlock(&ci
->i_ceph_lock
);
768 __mark_inode_dirty(inode
, dirty
);
772 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
773 inode
, ceph_vinop(inode
), pos
, (unsigned)iov
->iov_len
,
774 ceph_cap_string(got
));
775 ceph_put_cap_refs(ci
, got
);
778 if (ret
== -EOLDSNAPC
) {
779 dout("aio_write %p %llx.%llx %llu~%u got EOLDSNAPC, retrying\n",
780 inode
, ceph_vinop(inode
), pos
, (unsigned)iov
->iov_len
);
788 * llseek. be sure to verify file size on SEEK_END.
790 static loff_t
ceph_llseek(struct file
*file
, loff_t offset
, int origin
)
792 struct inode
*inode
= file
->f_mapping
->host
;
795 mutex_lock(&inode
->i_mutex
);
796 __ceph_do_pending_vmtruncate(inode
);
798 if (origin
== SEEK_END
|| origin
== SEEK_DATA
|| origin
== SEEK_HOLE
) {
799 ret
= ceph_do_getattr(inode
, CEPH_STAT_CAP_SIZE
);
808 offset
+= inode
->i_size
;
812 * Here we special-case the lseek(fd, 0, SEEK_CUR)
813 * position-querying operation. Avoid rewriting the "same"
814 * f_pos value back to the file because a concurrent read(),
815 * write() or lseek() might have altered it
818 offset
= file
->f_pos
;
821 offset
+= file
->f_pos
;
824 if (offset
>= inode
->i_size
) {
830 if (offset
>= inode
->i_size
) {
834 offset
= inode
->i_size
;
838 if (offset
< 0 || offset
> inode
->i_sb
->s_maxbytes
) {
843 /* Special lock needed here? */
844 if (offset
!= file
->f_pos
) {
845 file
->f_pos
= offset
;
850 mutex_unlock(&inode
->i_mutex
);
854 const struct file_operations ceph_file_fops
= {
856 .release
= ceph_release
,
857 .llseek
= ceph_llseek
,
858 .read
= do_sync_read
,
859 .write
= do_sync_write
,
860 .aio_read
= ceph_aio_read
,
861 .aio_write
= ceph_aio_write
,
866 .splice_read
= generic_file_splice_read
,
867 .splice_write
= generic_file_splice_write
,
868 .unlocked_ioctl
= ceph_ioctl
,
869 .compat_ioctl
= ceph_ioctl
,