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/falloc.h>
13 #include "mds_client.h"
17 * Ceph file operations
19 * Implement basic open/close functionality, and implement
22 * We implement three modes of file I/O:
23 * - buffered uses the generic_file_aio_{read,write} helpers
25 * - synchronous is used when there is multi-client read/write
26 * sharing, avoids the page cache, and synchronously waits for an
29 * - direct io takes the variant of the sync path that references
30 * user pages directly.
32 * fsync() flushes and waits on dirty pages, but just queues metadata
33 * for writeback: since the MDS can recover size and mtime there is no
34 * need to wait for MDS acknowledgement.
38 * Calculate the length sum of direct io vectors that can
39 * be combined into one page vector.
41 static size_t dio_get_pagev_size(const struct iov_iter
*it
)
43 const struct iovec
*iov
= it
->iov
;
44 const struct iovec
*iovend
= iov
+ it
->nr_segs
;
47 size
= iov
->iov_len
- it
->iov_offset
;
49 * An iov can be page vectored when both the current tail
50 * and the next base are page aligned.
52 while (PAGE_ALIGNED((iov
->iov_base
+ iov
->iov_len
)) &&
53 (++iov
< iovend
&& PAGE_ALIGNED((iov
->iov_base
)))) {
56 dout("dio_get_pagevlen len = %zu\n", size
);
61 * Allocate a page vector based on (@it, @nbytes).
62 * The return value is the tuple describing a page vector,
63 * that is (@pages, @page_align, @num_pages).
66 dio_get_pages_alloc(const struct iov_iter
*it
, size_t nbytes
,
67 size_t *page_align
, int *num_pages
)
69 struct iov_iter tmp_it
= *it
;
72 int ret
= 0, idx
, npages
;
74 align
= (unsigned long)(it
->iov
->iov_base
+ it
->iov_offset
) &
76 npages
= calc_pages_for(align
, nbytes
);
77 pages
= kmalloc(sizeof(*pages
) * npages
, GFP_KERNEL
);
79 pages
= vmalloc(sizeof(*pages
) * npages
);
81 return ERR_PTR(-ENOMEM
);
84 for (idx
= 0; idx
< npages
; ) {
86 ret
= iov_iter_get_pages(&tmp_it
, pages
+ idx
, nbytes
,
87 npages
- idx
, &start
);
91 iov_iter_advance(&tmp_it
, ret
);
93 idx
+= (ret
+ start
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
99 dout("dio_get_pages_alloc: got %d pages align %zu\n", npages
, align
);
102 ceph_put_page_vector(pages
, idx
, false);
107 * Prepare an open request. Preallocate ceph_cap to avoid an
108 * inopportune ENOMEM later.
110 static struct ceph_mds_request
*
111 prepare_open_request(struct super_block
*sb
, int flags
, int create_mode
)
113 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
114 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
115 struct ceph_mds_request
*req
;
116 int want_auth
= USE_ANY_MDS
;
117 int op
= (flags
& O_CREAT
) ? CEPH_MDS_OP_CREATE
: CEPH_MDS_OP_OPEN
;
119 if (flags
& (O_WRONLY
|O_RDWR
|O_CREAT
|O_TRUNC
))
120 want_auth
= USE_AUTH_MDS
;
122 req
= ceph_mdsc_create_request(mdsc
, op
, want_auth
);
125 req
->r_fmode
= ceph_flags_to_mode(flags
);
126 req
->r_args
.open
.flags
= cpu_to_le32(flags
);
127 req
->r_args
.open
.mode
= cpu_to_le32(create_mode
);
133 * initialize private struct file data.
134 * if we fail, clean up by dropping fmode reference on the ceph_inode
136 static int ceph_init_file(struct inode
*inode
, struct file
*file
, int fmode
)
138 struct ceph_file_info
*cf
;
140 struct ceph_inode_info
*ci
= ceph_inode(inode
);
141 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
142 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
144 switch (inode
->i_mode
& S_IFMT
) {
146 /* First file open request creates the cookie, we want to keep
147 * this cookie around for the filetime of the inode as not to
148 * have to worry about fscache register / revoke / operation
151 * Also, if we know the operation is going to invalidate data
152 * (non readonly) just nuke the cache right away.
154 ceph_fscache_register_inode_cookie(mdsc
->fsc
, ci
);
155 if ((fmode
& CEPH_FILE_MODE_WR
))
156 ceph_fscache_invalidate(inode
);
158 dout("init_file %p %p 0%o (regular)\n", inode
, file
,
160 cf
= kmem_cache_zalloc(ceph_file_cachep
, GFP_KERNEL
);
162 ceph_put_fmode(ceph_inode(inode
), fmode
); /* clean up */
167 cf
->readdir_cache_idx
= -1;
168 file
->private_data
= cf
;
169 BUG_ON(inode
->i_fop
->release
!= ceph_release
);
173 dout("init_file %p %p 0%o (symlink)\n", inode
, file
,
175 ceph_put_fmode(ceph_inode(inode
), fmode
); /* clean up */
179 dout("init_file %p %p 0%o (special)\n", inode
, file
,
182 * we need to drop the open ref now, since we don't
183 * have .release set to ceph_release.
185 ceph_put_fmode(ceph_inode(inode
), fmode
); /* clean up */
186 BUG_ON(inode
->i_fop
->release
== ceph_release
);
188 /* call the proper open fop */
189 ret
= inode
->i_fop
->open(inode
, file
);
195 * If we already have the requisite capabilities, we can satisfy
196 * the open request locally (no need to request new caps from the
197 * MDS). We do, however, need to inform the MDS (asynchronously)
198 * if our wanted caps set expands.
200 int ceph_open(struct inode
*inode
, struct file
*file
)
202 struct ceph_inode_info
*ci
= ceph_inode(inode
);
203 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
204 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
205 struct ceph_mds_request
*req
;
206 struct ceph_file_info
*cf
= file
->private_data
;
208 int flags
, fmode
, wanted
;
211 dout("open file %p is already opened\n", file
);
215 /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */
216 flags
= file
->f_flags
& ~(O_CREAT
|O_EXCL
);
217 if (S_ISDIR(inode
->i_mode
))
218 flags
= O_DIRECTORY
; /* mds likes to know */
220 dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode
,
221 ceph_vinop(inode
), file
, flags
, file
->f_flags
);
222 fmode
= ceph_flags_to_mode(flags
);
223 wanted
= ceph_caps_for_mode(fmode
);
225 /* snapped files are read-only */
226 if (ceph_snap(inode
) != CEPH_NOSNAP
&& (file
->f_mode
& FMODE_WRITE
))
229 /* trivially open snapdir */
230 if (ceph_snap(inode
) == CEPH_SNAPDIR
) {
231 spin_lock(&ci
->i_ceph_lock
);
232 __ceph_get_fmode(ci
, fmode
);
233 spin_unlock(&ci
->i_ceph_lock
);
234 return ceph_init_file(inode
, file
, fmode
);
238 * No need to block if we have caps on the auth MDS (for
239 * write) or any MDS (for read). Update wanted set
242 spin_lock(&ci
->i_ceph_lock
);
243 if (__ceph_is_any_real_caps(ci
) &&
244 (((fmode
& CEPH_FILE_MODE_WR
) == 0) || ci
->i_auth_cap
)) {
245 int mds_wanted
= __ceph_caps_mds_wanted(ci
);
246 int issued
= __ceph_caps_issued(ci
, NULL
);
248 dout("open %p fmode %d want %s issued %s using existing\n",
249 inode
, fmode
, ceph_cap_string(wanted
),
250 ceph_cap_string(issued
));
251 __ceph_get_fmode(ci
, fmode
);
252 spin_unlock(&ci
->i_ceph_lock
);
255 if ((issued
& wanted
) != wanted
&&
256 (mds_wanted
& wanted
) != wanted
&&
257 ceph_snap(inode
) != CEPH_SNAPDIR
)
258 ceph_check_caps(ci
, 0, NULL
);
260 return ceph_init_file(inode
, file
, fmode
);
261 } else if (ceph_snap(inode
) != CEPH_NOSNAP
&&
262 (ci
->i_snap_caps
& wanted
) == wanted
) {
263 __ceph_get_fmode(ci
, fmode
);
264 spin_unlock(&ci
->i_ceph_lock
);
265 return ceph_init_file(inode
, file
, fmode
);
268 spin_unlock(&ci
->i_ceph_lock
);
270 dout("open fmode %d wants %s\n", fmode
, ceph_cap_string(wanted
));
271 req
= prepare_open_request(inode
->i_sb
, flags
, 0);
276 req
->r_inode
= inode
;
280 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
282 err
= ceph_init_file(inode
, file
, req
->r_fmode
);
283 ceph_mdsc_put_request(req
);
284 dout("open result=%d on %llx.%llx\n", err
, ceph_vinop(inode
));
291 * Do a lookup + open with a single request. If we get a non-existent
292 * file or symlink, return 1 so the VFS can retry.
294 int ceph_atomic_open(struct inode
*dir
, struct dentry
*dentry
,
295 struct file
*file
, unsigned flags
, umode_t mode
,
298 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
299 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
300 struct ceph_mds_request
*req
;
302 struct ceph_acls_info acls
= {};
306 dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n",
308 d_unhashed(dentry
) ? "unhashed" : "hashed", flags
, mode
);
310 if (dentry
->d_name
.len
> NAME_MAX
)
311 return -ENAMETOOLONG
;
313 err
= ceph_init_dentry(dentry
);
317 if (flags
& O_CREAT
) {
318 err
= ceph_pre_init_acls(dir
, &mode
, &acls
);
324 req
= prepare_open_request(dir
->i_sb
, flags
, mode
);
329 req
->r_dentry
= dget(dentry
);
331 if (flags
& O_CREAT
) {
332 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
333 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
335 req
->r_pagelist
= acls
.pagelist
;
336 acls
.pagelist
= NULL
;
340 mask
= CEPH_STAT_CAP_INODE
| CEPH_CAP_AUTH_SHARED
;
341 if (ceph_security_xattr_wanted(dir
))
342 mask
|= CEPH_CAP_XATTR_SHARED
;
343 req
->r_args
.open
.mask
= cpu_to_le32(mask
);
345 req
->r_locked_dir
= dir
; /* caller holds dir->i_mutex */
346 err
= ceph_mdsc_do_request(mdsc
,
347 (flags
& (O_CREAT
|O_TRUNC
)) ? dir
: NULL
,
349 err
= ceph_handle_snapdir(req
, dentry
, err
);
353 if ((flags
& O_CREAT
) && !req
->r_reply_info
.head
->is_dentry
)
354 err
= ceph_handle_notrace_create(dir
, dentry
);
356 if (d_unhashed(dentry
)) {
357 dn
= ceph_finish_lookup(req
, dentry
, err
);
361 /* we were given a hashed negative dentry */
366 if (dn
|| d_really_is_negative(dentry
) || d_is_symlink(dentry
)) {
367 /* make vfs retry on splice, ENOENT, or symlink */
368 dout("atomic_open finish_no_open on dn %p\n", dn
);
369 err
= finish_no_open(file
, dn
);
371 dout("atomic_open finish_open on dn %p\n", dn
);
372 if (req
->r_op
== CEPH_MDS_OP_CREATE
&& req
->r_reply_info
.has_create_ino
) {
373 ceph_init_inode_acls(d_inode(dentry
), &acls
);
374 *opened
|= FILE_CREATED
;
376 err
= finish_open(file
, dentry
, ceph_open
, opened
);
379 if (!req
->r_err
&& req
->r_target_inode
)
380 ceph_put_fmode(ceph_inode(req
->r_target_inode
), req
->r_fmode
);
381 ceph_mdsc_put_request(req
);
383 ceph_release_acls_info(&acls
);
384 dout("atomic_open result=%d\n", err
);
388 int ceph_release(struct inode
*inode
, struct file
*file
)
390 struct ceph_inode_info
*ci
= ceph_inode(inode
);
391 struct ceph_file_info
*cf
= file
->private_data
;
393 dout("release inode %p file %p\n", inode
, file
);
394 ceph_put_fmode(ci
, cf
->fmode
);
395 if (cf
->last_readdir
)
396 ceph_mdsc_put_request(cf
->last_readdir
);
397 kfree(cf
->last_name
);
399 kmem_cache_free(ceph_file_cachep
, cf
);
401 /* wake up anyone waiting for caps on this inode */
402 wake_up_all(&ci
->i_cap_wq
);
413 * Read a range of bytes striped over one or more objects. Iterate over
414 * objects we stripe over. (That's not atomic, but good enough for now.)
416 * If we get a short result from the OSD, check against i_size; we need to
417 * only return a short read to the caller if we hit EOF.
419 static int striped_read(struct inode
*inode
,
421 struct page
**pages
, int num_pages
,
424 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
425 struct ceph_inode_info
*ci
= ceph_inode(inode
);
426 u64 pos
, this_len
, left
;
428 int page_align
, pages_left
;
430 struct page
**page_pos
;
431 bool hit_stripe
, was_short
;
434 * we may need to do multiple reads. not atomic, unfortunately.
439 pages_left
= num_pages
;
443 page_align
= pos
& ~PAGE_MASK
;
445 ret
= ceph_osdc_readpages(&fsc
->client
->osdc
, ceph_vino(inode
),
446 &ci
->i_layout
, pos
, &this_len
,
449 page_pos
, pages_left
, page_align
);
452 hit_stripe
= this_len
< left
;
453 was_short
= ret
>= 0 && ret
< this_len
;
454 dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos
, left
, read
,
455 ret
, hit_stripe
? " HITSTRIPE" : "", was_short
? " SHORT" : "");
457 i_size
= i_size_read(inode
);
460 if (was_short
&& (pos
+ ret
< i_size
)) {
461 int zlen
= min(this_len
- ret
, i_size
- pos
- ret
);
462 int zoff
= (off
& ~PAGE_MASK
) + read
+ ret
;
463 dout(" zero gap %llu to %llu\n",
464 pos
+ ret
, pos
+ ret
+ zlen
);
465 ceph_zero_page_vector_range(zoff
, zlen
, pages
);
469 didpages
= (page_align
+ ret
) >> PAGE_SHIFT
;
473 page_pos
+= didpages
;
474 pages_left
-= didpages
;
476 /* hit stripe and need continue*/
477 if (left
&& hit_stripe
&& pos
< i_size
)
483 /* did we bounce off eof? */
484 if (pos
+ left
> i_size
)
485 *checkeof
= CHECK_EOF
;
488 dout("striped_read returns %d\n", ret
);
493 * Completely synchronous read and write methods. Direct from __user
494 * buffer to osd, or directly to user pages (if O_DIRECT).
496 * If the read spans object boundary, just do multiple reads.
498 static ssize_t
ceph_sync_read(struct kiocb
*iocb
, struct iov_iter
*i
,
501 struct file
*file
= iocb
->ki_filp
;
502 struct inode
*inode
= file_inode(file
);
504 u64 off
= iocb
->ki_pos
;
506 size_t len
= iov_iter_count(i
);
508 dout("sync_read on file %p %llu~%u %s\n", file
, off
,
510 (file
->f_flags
& O_DIRECT
) ? "O_DIRECT" : "");
515 * flush any page cache pages in this range. this
516 * will make concurrent normal and sync io slow,
517 * but it will at least behave sensibly when they are
520 ret
= filemap_write_and_wait_range(inode
->i_mapping
, off
,
525 num_pages
= calc_pages_for(off
, len
);
526 pages
= ceph_alloc_page_vector(num_pages
, GFP_KERNEL
);
528 return PTR_ERR(pages
);
529 ret
= striped_read(inode
, off
, len
, pages
,
530 num_pages
, checkeof
);
536 size_t page_off
= off
& ~PAGE_MASK
;
537 size_t copy
= min_t(size_t, left
,
538 PAGE_SIZE
- page_off
);
539 l
= copy_page_to_iter(pages
[k
++], page_off
, copy
, i
);
546 ceph_release_page_vector(pages
, num_pages
);
548 if (off
> iocb
->ki_pos
) {
549 ret
= off
- iocb
->ki_pos
;
553 dout("sync_read result %d\n", ret
);
557 struct ceph_aio_request
{
562 struct list_head osd_reqs
;
564 atomic_t pending_reqs
;
565 struct timespec mtime
;
566 struct ceph_cap_flush
*prealloc_cf
;
569 struct ceph_aio_work
{
570 struct work_struct work
;
571 struct ceph_osd_request
*req
;
574 static void ceph_aio_retry_work(struct work_struct
*work
);
576 static void ceph_aio_complete(struct inode
*inode
,
577 struct ceph_aio_request
*aio_req
)
579 struct ceph_inode_info
*ci
= ceph_inode(inode
);
582 if (!atomic_dec_and_test(&aio_req
->pending_reqs
))
585 ret
= aio_req
->error
;
587 ret
= aio_req
->total_len
;
589 dout("ceph_aio_complete %p rc %d\n", inode
, ret
);
591 if (ret
>= 0 && aio_req
->write
) {
594 loff_t endoff
= aio_req
->iocb
->ki_pos
+ aio_req
->total_len
;
595 if (endoff
> i_size_read(inode
)) {
596 if (ceph_inode_set_size(inode
, endoff
))
597 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
600 spin_lock(&ci
->i_ceph_lock
);
601 ci
->i_inline_version
= CEPH_INLINE_NONE
;
602 dirty
= __ceph_mark_dirty_caps(ci
, CEPH_CAP_FILE_WR
,
603 &aio_req
->prealloc_cf
);
604 spin_unlock(&ci
->i_ceph_lock
);
606 __mark_inode_dirty(inode
, dirty
);
610 ceph_put_cap_refs(ci
, (aio_req
->write
? CEPH_CAP_FILE_WR
:
613 aio_req
->iocb
->ki_complete(aio_req
->iocb
, ret
, 0);
615 ceph_free_cap_flush(aio_req
->prealloc_cf
);
619 static void ceph_aio_complete_req(struct ceph_osd_request
*req
,
620 struct ceph_msg
*msg
)
622 int rc
= req
->r_result
;
623 struct inode
*inode
= req
->r_inode
;
624 struct ceph_aio_request
*aio_req
= req
->r_priv
;
625 struct ceph_osd_data
*osd_data
= osd_req_op_extent_osd_data(req
, 0);
626 int num_pages
= calc_pages_for((u64
)osd_data
->alignment
,
629 dout("ceph_aio_complete_req %p rc %d bytes %llu\n",
630 inode
, rc
, osd_data
->length
);
632 if (rc
== -EOLDSNAPC
) {
633 struct ceph_aio_work
*aio_work
;
634 BUG_ON(!aio_req
->write
);
636 aio_work
= kmalloc(sizeof(*aio_work
), GFP_NOFS
);
638 INIT_WORK(&aio_work
->work
, ceph_aio_retry_work
);
640 queue_work(ceph_inode_to_client(inode
)->wb_wq
,
645 } else if (!aio_req
->write
) {
648 if (rc
>= 0 && osd_data
->length
> rc
) {
649 int zoff
= osd_data
->alignment
+ rc
;
650 int zlen
= osd_data
->length
- rc
;
652 * If read is satisfied by single OSD request,
653 * it can pass EOF. Otherwise read is within
656 if (aio_req
->num_reqs
== 1) {
657 loff_t i_size
= i_size_read(inode
);
658 loff_t endoff
= aio_req
->iocb
->ki_pos
+ rc
;
660 zlen
= min_t(size_t, zlen
,
662 aio_req
->total_len
= rc
+ zlen
;
666 ceph_zero_page_vector_range(zoff
, zlen
,
671 ceph_put_page_vector(osd_data
->pages
, num_pages
, false);
672 ceph_osdc_put_request(req
);
675 cmpxchg(&aio_req
->error
, 0, rc
);
677 ceph_aio_complete(inode
, aio_req
);
681 static void ceph_aio_retry_work(struct work_struct
*work
)
683 struct ceph_aio_work
*aio_work
=
684 container_of(work
, struct ceph_aio_work
, work
);
685 struct ceph_osd_request
*orig_req
= aio_work
->req
;
686 struct ceph_aio_request
*aio_req
= orig_req
->r_priv
;
687 struct inode
*inode
= orig_req
->r_inode
;
688 struct ceph_inode_info
*ci
= ceph_inode(inode
);
689 struct ceph_snap_context
*snapc
;
690 struct ceph_osd_request
*req
;
693 spin_lock(&ci
->i_ceph_lock
);
694 if (__ceph_have_pending_cap_snap(ci
)) {
695 struct ceph_cap_snap
*capsnap
=
696 list_last_entry(&ci
->i_cap_snaps
,
697 struct ceph_cap_snap
,
699 snapc
= ceph_get_snap_context(capsnap
->context
);
701 BUG_ON(!ci
->i_head_snapc
);
702 snapc
= ceph_get_snap_context(ci
->i_head_snapc
);
704 spin_unlock(&ci
->i_ceph_lock
);
706 req
= ceph_osdc_alloc_request(orig_req
->r_osdc
, snapc
, 2,
714 req
->r_flags
= CEPH_OSD_FLAG_ORDERSNAP
|
715 CEPH_OSD_FLAG_ONDISK
|
717 ceph_oloc_copy(&req
->r_base_oloc
, &orig_req
->r_base_oloc
);
718 ceph_oid_copy(&req
->r_base_oid
, &orig_req
->r_base_oid
);
720 ret
= ceph_osdc_alloc_messages(req
, GFP_NOFS
);
722 ceph_osdc_put_request(req
);
727 req
->r_ops
[0] = orig_req
->r_ops
[0];
728 osd_req_op_init(req
, 1, CEPH_OSD_OP_STARTSYNC
, 0);
730 req
->r_mtime
= aio_req
->mtime
;
731 req
->r_data_offset
= req
->r_ops
[0].extent
.offset
;
733 ceph_osdc_put_request(orig_req
);
735 req
->r_callback
= ceph_aio_complete_req
;
736 req
->r_inode
= inode
;
737 req
->r_priv
= aio_req
;
739 ret
= ceph_osdc_start_request(req
->r_osdc
, req
, false);
743 ceph_aio_complete_req(req
, NULL
);
746 ceph_put_snap_context(snapc
);
751 * Write commit request unsafe callback, called to tell us when a
752 * request is unsafe (that is, in flight--has been handed to the
753 * messenger to send to its target osd). It is called again when
754 * we've received a response message indicating the request is
755 * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request
756 * is completed early (and unsuccessfully) due to a timeout or
759 * This is used if we requested both an ACK and ONDISK commit reply
762 static void ceph_sync_write_unsafe(struct ceph_osd_request
*req
, bool unsafe
)
764 struct ceph_inode_info
*ci
= ceph_inode(req
->r_inode
);
766 dout("%s %p tid %llu %ssafe\n", __func__
, req
, req
->r_tid
,
769 ceph_get_cap_refs(ci
, CEPH_CAP_FILE_WR
);
770 spin_lock(&ci
->i_unsafe_lock
);
771 list_add_tail(&req
->r_unsafe_item
,
772 &ci
->i_unsafe_writes
);
773 spin_unlock(&ci
->i_unsafe_lock
);
775 spin_lock(&ci
->i_unsafe_lock
);
776 list_del_init(&req
->r_unsafe_item
);
777 spin_unlock(&ci
->i_unsafe_lock
);
778 ceph_put_cap_refs(ci
, CEPH_CAP_FILE_WR
);
784 ceph_direct_read_write(struct kiocb
*iocb
, struct iov_iter
*iter
,
785 struct ceph_snap_context
*snapc
,
786 struct ceph_cap_flush
**pcf
)
788 struct file
*file
= iocb
->ki_filp
;
789 struct inode
*inode
= file_inode(file
);
790 struct ceph_inode_info
*ci
= ceph_inode(inode
);
791 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
792 struct ceph_vino vino
;
793 struct ceph_osd_request
*req
;
795 struct ceph_aio_request
*aio_req
= NULL
;
799 struct timespec mtime
= current_fs_time(inode
->i_sb
);
800 size_t count
= iov_iter_count(iter
);
801 loff_t pos
= iocb
->ki_pos
;
802 bool write
= iov_iter_rw(iter
) == WRITE
;
804 if (write
&& ceph_snap(file_inode(file
)) != CEPH_NOSNAP
)
807 dout("sync_direct_read_write (%s) on file %p %lld~%u\n",
808 (write
? "write" : "read"), file
, pos
, (unsigned)count
);
810 ret
= filemap_write_and_wait_range(inode
->i_mapping
, pos
, pos
+ count
);
815 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
817 (pos
+ count
) >> PAGE_SHIFT
);
819 dout("invalidate_inode_pages2_range returned %d\n", ret
);
821 flags
= CEPH_OSD_FLAG_ORDERSNAP
|
822 CEPH_OSD_FLAG_ONDISK
|
825 flags
= CEPH_OSD_FLAG_READ
;
828 while (iov_iter_count(iter
) > 0) {
829 u64 size
= dio_get_pagev_size(iter
);
833 vino
= ceph_vino(inode
);
834 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
836 /*include a 'startsync' command*/
838 write
? CEPH_OSD_OP_WRITE
:
850 pages
= dio_get_pages_alloc(iter
, len
, &start
, &num_pages
);
852 ceph_osdc_put_request(req
);
853 ret
= PTR_ERR(pages
);
858 * To simplify error handling, allow AIO when IO within i_size
859 * or IO can be satisfied by single OSD request.
861 if (pos
== iocb
->ki_pos
&& !is_sync_kiocb(iocb
) &&
862 (len
== count
|| pos
+ count
<= i_size_read(inode
))) {
863 aio_req
= kzalloc(sizeof(*aio_req
), GFP_KERNEL
);
865 aio_req
->iocb
= iocb
;
866 aio_req
->write
= write
;
867 INIT_LIST_HEAD(&aio_req
->osd_reqs
);
869 aio_req
->mtime
= mtime
;
870 swap(aio_req
->prealloc_cf
, *pcf
);
878 * throw out any page cache pages in this range. this
881 truncate_inode_pages_range(inode
->i_mapping
, pos
,
882 (pos
+len
) | (PAGE_SIZE
- 1));
884 osd_req_op_init(req
, 1, CEPH_OSD_OP_STARTSYNC
, 0);
885 req
->r_mtime
= mtime
;
888 osd_req_op_extent_osd_data_pages(req
, 0, pages
, len
, start
,
892 aio_req
->total_len
+= len
;
894 atomic_inc(&aio_req
->pending_reqs
);
896 req
->r_callback
= ceph_aio_complete_req
;
897 req
->r_inode
= inode
;
898 req
->r_priv
= aio_req
;
899 list_add_tail(&req
->r_unsafe_item
, &aio_req
->osd_reqs
);
902 iov_iter_advance(iter
, len
);
906 ret
= ceph_osdc_start_request(req
->r_osdc
, req
, false);
908 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
910 size
= i_size_read(inode
);
914 if (ret
>= 0 && ret
< len
&& pos
+ ret
< size
) {
915 int zlen
= min_t(size_t, len
- ret
,
917 ceph_zero_page_vector_range(start
+ ret
, zlen
,
925 ceph_put_page_vector(pages
, num_pages
, false);
927 ceph_osdc_put_request(req
);
932 iov_iter_advance(iter
, len
);
934 if (!write
&& pos
>= size
)
937 if (write
&& pos
> size
) {
938 if (ceph_inode_set_size(inode
, pos
))
939 ceph_check_caps(ceph_inode(inode
),
946 if (aio_req
->num_reqs
== 0) {
951 ceph_get_cap_refs(ci
, write
? CEPH_CAP_FILE_WR
:
954 while (!list_empty(&aio_req
->osd_reqs
)) {
955 req
= list_first_entry(&aio_req
->osd_reqs
,
956 struct ceph_osd_request
,
958 list_del_init(&req
->r_unsafe_item
);
960 ret
= ceph_osdc_start_request(req
->r_osdc
,
964 ceph_aio_complete_req(req
, NULL
);
970 if (ret
!= -EOLDSNAPC
&& pos
> iocb
->ki_pos
) {
971 ret
= pos
- iocb
->ki_pos
;
978 * Synchronous write, straight from __user pointer or user pages.
980 * If write spans object boundary, just do multiple writes. (For a
981 * correct atomic write, we should e.g. take write locks on all
982 * objects, rollback on failure, etc.)
985 ceph_sync_write(struct kiocb
*iocb
, struct iov_iter
*from
, loff_t pos
,
986 struct ceph_snap_context
*snapc
)
988 struct file
*file
= iocb
->ki_filp
;
989 struct inode
*inode
= file_inode(file
);
990 struct ceph_inode_info
*ci
= ceph_inode(inode
);
991 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
992 struct ceph_vino vino
;
993 struct ceph_osd_request
*req
;
1001 struct timespec mtime
= current_fs_time(inode
->i_sb
);
1002 size_t count
= iov_iter_count(from
);
1004 if (ceph_snap(file_inode(file
)) != CEPH_NOSNAP
)
1007 dout("sync_write on file %p %lld~%u\n", file
, pos
, (unsigned)count
);
1009 ret
= filemap_write_and_wait_range(inode
->i_mapping
, pos
, pos
+ count
);
1013 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
1015 (pos
+ count
) >> PAGE_SHIFT
);
1017 dout("invalidate_inode_pages2_range returned %d\n", ret
);
1019 flags
= CEPH_OSD_FLAG_ORDERSNAP
|
1020 CEPH_OSD_FLAG_ONDISK
|
1021 CEPH_OSD_FLAG_WRITE
|
1024 while ((len
= iov_iter_count(from
)) > 0) {
1028 vino
= ceph_vino(inode
);
1029 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
1030 vino
, pos
, &len
, 0, 1,
1031 CEPH_OSD_OP_WRITE
, flags
, snapc
,
1033 ci
->i_truncate_size
,
1041 * write from beginning of first page,
1042 * regardless of io alignment
1044 num_pages
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1046 pages
= ceph_alloc_page_vector(num_pages
, GFP_KERNEL
);
1047 if (IS_ERR(pages
)) {
1048 ret
= PTR_ERR(pages
);
1053 for (n
= 0; n
< num_pages
; n
++) {
1054 size_t plen
= min_t(size_t, left
, PAGE_SIZE
);
1055 ret
= copy_page_from_iter(pages
[n
], 0, plen
, from
);
1064 ceph_release_page_vector(pages
, num_pages
);
1068 /* get a second commit callback */
1069 req
->r_unsafe_callback
= ceph_sync_write_unsafe
;
1070 req
->r_inode
= inode
;
1072 osd_req_op_extent_osd_data_pages(req
, 0, pages
, len
, 0,
1075 req
->r_mtime
= mtime
;
1076 ret
= ceph_osdc_start_request(&fsc
->client
->osdc
, req
, false);
1078 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
1081 ceph_osdc_put_request(req
);
1086 if (pos
> i_size_read(inode
)) {
1087 check_caps
= ceph_inode_set_size(inode
, pos
);
1089 ceph_check_caps(ceph_inode(inode
),
1090 CHECK_CAPS_AUTHONLY
,
1097 if (ret
!= -EOLDSNAPC
&& written
> 0) {
1105 * Wrap generic_file_aio_read with checks for cap bits on the inode.
1106 * Atomically grab references, so that those bits are not released
1107 * back to the MDS mid-read.
1109 * Hmm, the sync read case isn't actually async... should it be?
1111 static ssize_t
ceph_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1113 struct file
*filp
= iocb
->ki_filp
;
1114 struct ceph_file_info
*fi
= filp
->private_data
;
1115 size_t len
= iov_iter_count(to
);
1116 struct inode
*inode
= file_inode(filp
);
1117 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1118 struct page
*pinned_page
= NULL
;
1121 int retry_op
= 0, read
= 0;
1124 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
1125 inode
, ceph_vinop(inode
), iocb
->ki_pos
, (unsigned)len
, inode
);
1127 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
1128 want
= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_LAZYIO
;
1130 want
= CEPH_CAP_FILE_CACHE
;
1131 ret
= ceph_get_caps(ci
, CEPH_CAP_FILE_RD
, want
, -1, &got
, &pinned_page
);
1135 if ((got
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)) == 0 ||
1136 (iocb
->ki_flags
& IOCB_DIRECT
) ||
1137 (fi
->flags
& CEPH_F_SYNC
)) {
1139 dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n",
1140 inode
, ceph_vinop(inode
), iocb
->ki_pos
, (unsigned)len
,
1141 ceph_cap_string(got
));
1143 if (ci
->i_inline_version
== CEPH_INLINE_NONE
) {
1144 if (!retry_op
&& (iocb
->ki_flags
& IOCB_DIRECT
)) {
1145 ret
= ceph_direct_read_write(iocb
, to
,
1147 if (ret
>= 0 && ret
< len
)
1148 retry_op
= CHECK_EOF
;
1150 ret
= ceph_sync_read(iocb
, to
, &retry_op
);
1153 retry_op
= READ_INLINE
;
1156 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
1157 inode
, ceph_vinop(inode
), iocb
->ki_pos
, (unsigned)len
,
1158 ceph_cap_string(got
));
1160 ret
= generic_file_read_iter(iocb
, to
);
1162 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
1163 inode
, ceph_vinop(inode
), ceph_cap_string(got
), (int)ret
);
1165 put_page(pinned_page
);
1168 ceph_put_cap_refs(ci
, got
);
1169 if (retry_op
> HAVE_RETRIED
&& ret
>= 0) {
1171 struct page
*page
= NULL
;
1173 if (retry_op
== READ_INLINE
) {
1174 page
= __page_cache_alloc(GFP_KERNEL
);
1179 statret
= __ceph_do_getattr(inode
, page
,
1180 CEPH_STAT_CAP_INLINE_DATA
, !!page
);
1183 if (statret
== -ENODATA
) {
1184 BUG_ON(retry_op
!= READ_INLINE
);
1190 i_size
= i_size_read(inode
);
1191 if (retry_op
== READ_INLINE
) {
1192 BUG_ON(ret
> 0 || read
> 0);
1193 if (iocb
->ki_pos
< i_size
&&
1194 iocb
->ki_pos
< PAGE_SIZE
) {
1195 loff_t end
= min_t(loff_t
, i_size
,
1196 iocb
->ki_pos
+ len
);
1197 end
= min_t(loff_t
, end
, PAGE_SIZE
);
1199 zero_user_segment(page
, statret
, end
);
1200 ret
= copy_page_to_iter(page
,
1201 iocb
->ki_pos
& ~PAGE_MASK
,
1202 end
- iocb
->ki_pos
, to
);
1203 iocb
->ki_pos
+= ret
;
1206 if (iocb
->ki_pos
< i_size
&& read
< len
) {
1207 size_t zlen
= min_t(size_t, len
- read
,
1208 i_size
- iocb
->ki_pos
);
1209 ret
= iov_iter_zero(zlen
, to
);
1210 iocb
->ki_pos
+= ret
;
1213 __free_pages(page
, 0);
1217 /* hit EOF or hole? */
1218 if (retry_op
== CHECK_EOF
&& iocb
->ki_pos
< i_size
&&
1220 dout("sync_read hit hole, ppos %lld < size %lld"
1221 ", reading more\n", iocb
->ki_pos
, i_size
);
1225 retry_op
= HAVE_RETRIED
;
1237 * Take cap references to avoid releasing caps to MDS mid-write.
1239 * If we are synchronous, and write with an old snap context, the OSD
1240 * may return EOLDSNAPC. In that case, retry the write.. _after_
1241 * dropping our cap refs and allowing the pending snap to logically
1242 * complete _before_ this write occurs.
1244 * If we are near ENOSPC, write synchronously.
1246 static ssize_t
ceph_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1248 struct file
*file
= iocb
->ki_filp
;
1249 struct ceph_file_info
*fi
= file
->private_data
;
1250 struct inode
*inode
= file_inode(file
);
1251 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1252 struct ceph_osd_client
*osdc
=
1253 &ceph_sb_to_client(inode
->i_sb
)->client
->osdc
;
1254 struct ceph_cap_flush
*prealloc_cf
;
1255 ssize_t count
, written
= 0;
1259 if (ceph_snap(inode
) != CEPH_NOSNAP
)
1262 prealloc_cf
= ceph_alloc_cap_flush();
1268 /* We can write back this queue in page reclaim */
1269 current
->backing_dev_info
= inode_to_bdi(inode
);
1271 if (iocb
->ki_flags
& IOCB_APPEND
) {
1272 err
= ceph_do_getattr(inode
, CEPH_STAT_CAP_SIZE
, false);
1277 err
= generic_write_checks(iocb
, from
);
1282 count
= iov_iter_count(from
);
1283 err
= file_remove_privs(file
);
1287 err
= file_update_time(file
);
1291 if (ci
->i_inline_version
!= CEPH_INLINE_NONE
) {
1292 err
= ceph_uninline_data(file
, NULL
);
1298 if (ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_FULL
)) {
1303 dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
1304 inode
, ceph_vinop(inode
), pos
, count
, i_size_read(inode
));
1305 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
1306 want
= CEPH_CAP_FILE_BUFFER
| CEPH_CAP_FILE_LAZYIO
;
1308 want
= CEPH_CAP_FILE_BUFFER
;
1310 err
= ceph_get_caps(ci
, CEPH_CAP_FILE_WR
, want
, pos
+ count
,
1315 dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
1316 inode
, ceph_vinop(inode
), pos
, count
, ceph_cap_string(got
));
1318 if ((got
& (CEPH_CAP_FILE_BUFFER
|CEPH_CAP_FILE_LAZYIO
)) == 0 ||
1319 (iocb
->ki_flags
& IOCB_DIRECT
) || (fi
->flags
& CEPH_F_SYNC
)) {
1320 struct ceph_snap_context
*snapc
;
1321 struct iov_iter data
;
1322 inode_unlock(inode
);
1324 spin_lock(&ci
->i_ceph_lock
);
1325 if (__ceph_have_pending_cap_snap(ci
)) {
1326 struct ceph_cap_snap
*capsnap
=
1327 list_last_entry(&ci
->i_cap_snaps
,
1328 struct ceph_cap_snap
,
1330 snapc
= ceph_get_snap_context(capsnap
->context
);
1332 BUG_ON(!ci
->i_head_snapc
);
1333 snapc
= ceph_get_snap_context(ci
->i_head_snapc
);
1335 spin_unlock(&ci
->i_ceph_lock
);
1337 /* we might need to revert back to that point */
1339 if (iocb
->ki_flags
& IOCB_DIRECT
)
1340 written
= ceph_direct_read_write(iocb
, &data
, snapc
,
1343 written
= ceph_sync_write(iocb
, &data
, pos
, snapc
);
1344 if (written
== -EOLDSNAPC
) {
1345 dout("aio_write %p %llx.%llx %llu~%u"
1346 "got EOLDSNAPC, retrying\n",
1347 inode
, ceph_vinop(inode
),
1348 pos
, (unsigned)count
);
1353 iov_iter_advance(from
, written
);
1354 ceph_put_snap_context(snapc
);
1356 loff_t old_size
= i_size_read(inode
);
1358 * No need to acquire the i_truncate_mutex. Because
1359 * the MDS revokes Fwb caps before sending truncate
1360 * message to us. We can't get Fwb cap while there
1361 * are pending vmtruncate. So write and vmtruncate
1362 * can not run at the same time
1364 written
= generic_perform_write(file
, from
, pos
);
1365 if (likely(written
>= 0))
1366 iocb
->ki_pos
= pos
+ written
;
1367 if (i_size_read(inode
) > old_size
)
1368 ceph_fscache_update_objectsize(inode
);
1369 inode_unlock(inode
);
1374 spin_lock(&ci
->i_ceph_lock
);
1375 ci
->i_inline_version
= CEPH_INLINE_NONE
;
1376 dirty
= __ceph_mark_dirty_caps(ci
, CEPH_CAP_FILE_WR
,
1378 spin_unlock(&ci
->i_ceph_lock
);
1380 __mark_inode_dirty(inode
, dirty
);
1383 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
1384 inode
, ceph_vinop(inode
), pos
, (unsigned)count
,
1385 ceph_cap_string(got
));
1386 ceph_put_cap_refs(ci
, got
);
1389 ((file
->f_flags
& O_SYNC
) || IS_SYNC(file
->f_mapping
->host
) ||
1390 ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_NEARFULL
))) {
1391 err
= vfs_fsync_range(file
, pos
, pos
+ written
- 1, 1);
1399 inode_unlock(inode
);
1401 ceph_free_cap_flush(prealloc_cf
);
1402 current
->backing_dev_info
= NULL
;
1403 return written
? written
: err
;
1407 * llseek. be sure to verify file size on SEEK_END.
1409 static loff_t
ceph_llseek(struct file
*file
, loff_t offset
, int whence
)
1411 struct inode
*inode
= file
->f_mapping
->host
;
1417 if (whence
== SEEK_END
|| whence
== SEEK_DATA
|| whence
== SEEK_HOLE
) {
1418 ret
= ceph_do_getattr(inode
, CEPH_STAT_CAP_SIZE
, false);
1425 i_size
= i_size_read(inode
);
1432 * Here we special-case the lseek(fd, 0, SEEK_CUR)
1433 * position-querying operation. Avoid rewriting the "same"
1434 * f_pos value back to the file because a concurrent read(),
1435 * write() or lseek() might have altered it
1438 offset
= file
->f_pos
;
1441 offset
+= file
->f_pos
;
1444 if (offset
>= i_size
) {
1450 if (offset
>= i_size
) {
1458 offset
= vfs_setpos(file
, offset
, inode
->i_sb
->s_maxbytes
);
1461 inode_unlock(inode
);
1465 static inline void ceph_zero_partial_page(
1466 struct inode
*inode
, loff_t offset
, unsigned size
)
1469 pgoff_t index
= offset
>> PAGE_SHIFT
;
1471 page
= find_lock_page(inode
->i_mapping
, index
);
1473 wait_on_page_writeback(page
);
1474 zero_user(page
, offset
& (PAGE_SIZE
- 1), size
);
1480 static void ceph_zero_pagecache_range(struct inode
*inode
, loff_t offset
,
1483 loff_t nearly
= round_up(offset
, PAGE_SIZE
);
1484 if (offset
< nearly
) {
1485 loff_t size
= nearly
- offset
;
1488 ceph_zero_partial_page(inode
, offset
, size
);
1492 if (length
>= PAGE_SIZE
) {
1493 loff_t size
= round_down(length
, PAGE_SIZE
);
1494 truncate_pagecache_range(inode
, offset
, offset
+ size
- 1);
1499 ceph_zero_partial_page(inode
, offset
, length
);
1502 static int ceph_zero_partial_object(struct inode
*inode
,
1503 loff_t offset
, loff_t
*length
)
1505 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1506 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
1507 struct ceph_osd_request
*req
;
1513 op
= offset
? CEPH_OSD_OP_DELETE
: CEPH_OSD_OP_TRUNCATE
;
1516 op
= CEPH_OSD_OP_ZERO
;
1519 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
1523 CEPH_OSD_FLAG_WRITE
|
1524 CEPH_OSD_FLAG_ONDISK
,
1531 req
->r_mtime
= inode
->i_mtime
;
1532 ret
= ceph_osdc_start_request(&fsc
->client
->osdc
, req
, false);
1534 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
1538 ceph_osdc_put_request(req
);
1544 static int ceph_zero_objects(struct inode
*inode
, loff_t offset
, loff_t length
)
1547 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1548 s32 stripe_unit
= ceph_file_layout_su(ci
->i_layout
);
1549 s32 stripe_count
= ceph_file_layout_stripe_count(ci
->i_layout
);
1550 s32 object_size
= ceph_file_layout_object_size(ci
->i_layout
);
1551 u64 object_set_size
= object_size
* stripe_count
;
1554 /* round offset up to next period boundary */
1555 nearly
= offset
+ object_set_size
- 1;
1557 nearly
-= do_div(t
, object_set_size
);
1559 while (length
&& offset
< nearly
) {
1560 loff_t size
= length
;
1561 ret
= ceph_zero_partial_object(inode
, offset
, &size
);
1567 while (length
>= object_set_size
) {
1569 loff_t pos
= offset
;
1570 for (i
= 0; i
< stripe_count
; ++i
) {
1571 ret
= ceph_zero_partial_object(inode
, pos
, NULL
);
1576 offset
+= object_set_size
;
1577 length
-= object_set_size
;
1580 loff_t size
= length
;
1581 ret
= ceph_zero_partial_object(inode
, offset
, &size
);
1590 static long ceph_fallocate(struct file
*file
, int mode
,
1591 loff_t offset
, loff_t length
)
1593 struct ceph_file_info
*fi
= file
->private_data
;
1594 struct inode
*inode
= file_inode(file
);
1595 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1596 struct ceph_osd_client
*osdc
=
1597 &ceph_inode_to_client(inode
)->client
->osdc
;
1598 struct ceph_cap_flush
*prealloc_cf
;
1605 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
1608 if (!S_ISREG(inode
->i_mode
))
1611 prealloc_cf
= ceph_alloc_cap_flush();
1617 if (ceph_snap(inode
) != CEPH_NOSNAP
) {
1622 if (ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_FULL
) &&
1623 !(mode
& FALLOC_FL_PUNCH_HOLE
)) {
1628 if (ci
->i_inline_version
!= CEPH_INLINE_NONE
) {
1629 ret
= ceph_uninline_data(file
, NULL
);
1634 size
= i_size_read(inode
);
1635 if (!(mode
& FALLOC_FL_KEEP_SIZE
))
1636 endoff
= offset
+ length
;
1638 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
1639 want
= CEPH_CAP_FILE_BUFFER
| CEPH_CAP_FILE_LAZYIO
;
1641 want
= CEPH_CAP_FILE_BUFFER
;
1643 ret
= ceph_get_caps(ci
, CEPH_CAP_FILE_WR
, want
, endoff
, &got
, NULL
);
1647 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
1649 ceph_zero_pagecache_range(inode
, offset
, length
);
1650 ret
= ceph_zero_objects(inode
, offset
, length
);
1651 } else if (endoff
> size
) {
1652 truncate_pagecache_range(inode
, size
, -1);
1653 if (ceph_inode_set_size(inode
, endoff
))
1654 ceph_check_caps(ceph_inode(inode
),
1655 CHECK_CAPS_AUTHONLY
, NULL
);
1659 spin_lock(&ci
->i_ceph_lock
);
1660 ci
->i_inline_version
= CEPH_INLINE_NONE
;
1661 dirty
= __ceph_mark_dirty_caps(ci
, CEPH_CAP_FILE_WR
,
1663 spin_unlock(&ci
->i_ceph_lock
);
1665 __mark_inode_dirty(inode
, dirty
);
1668 ceph_put_cap_refs(ci
, got
);
1670 inode_unlock(inode
);
1671 ceph_free_cap_flush(prealloc_cf
);
1675 const struct file_operations ceph_file_fops
= {
1677 .release
= ceph_release
,
1678 .llseek
= ceph_llseek
,
1679 .read_iter
= ceph_read_iter
,
1680 .write_iter
= ceph_write_iter
,
1682 .fsync
= ceph_fsync
,
1684 .flock
= ceph_flock
,
1685 .splice_read
= generic_file_splice_read
,
1686 .splice_write
= iter_file_splice_write
,
1687 .unlocked_ioctl
= ceph_ioctl
,
1688 .compat_ioctl
= ceph_ioctl
,
1689 .fallocate
= ceph_fallocate
,