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
;
141 switch (inode
->i_mode
& S_IFMT
) {
143 ceph_fscache_register_inode_cookie(inode
);
144 ceph_fscache_file_set_cookie(inode
, file
);
146 dout("init_file %p %p 0%o (regular)\n", inode
, file
,
148 cf
= kmem_cache_zalloc(ceph_file_cachep
, GFP_KERNEL
);
150 ceph_put_fmode(ceph_inode(inode
), fmode
); /* clean up */
155 cf
->readdir_cache_idx
= -1;
156 file
->private_data
= cf
;
157 BUG_ON(inode
->i_fop
->release
!= ceph_release
);
161 dout("init_file %p %p 0%o (symlink)\n", inode
, file
,
163 ceph_put_fmode(ceph_inode(inode
), fmode
); /* clean up */
167 dout("init_file %p %p 0%o (special)\n", inode
, file
,
170 * we need to drop the open ref now, since we don't
171 * have .release set to ceph_release.
173 ceph_put_fmode(ceph_inode(inode
), fmode
); /* clean up */
174 BUG_ON(inode
->i_fop
->release
== ceph_release
);
176 /* call the proper open fop */
177 ret
= inode
->i_fop
->open(inode
, file
);
183 * try renew caps after session gets killed.
185 int ceph_renew_caps(struct inode
*inode
)
187 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(inode
->i_sb
)->mdsc
;
188 struct ceph_inode_info
*ci
= ceph_inode(inode
);
189 struct ceph_mds_request
*req
;
190 int err
, flags
, wanted
;
192 spin_lock(&ci
->i_ceph_lock
);
193 wanted
= __ceph_caps_file_wanted(ci
);
194 if (__ceph_is_any_real_caps(ci
) &&
195 (!(wanted
& CEPH_CAP_ANY_WR
) || ci
->i_auth_cap
)) {
196 int issued
= __ceph_caps_issued(ci
, NULL
);
197 spin_unlock(&ci
->i_ceph_lock
);
198 dout("renew caps %p want %s issued %s updating mds_wanted\n",
199 inode
, ceph_cap_string(wanted
), ceph_cap_string(issued
));
200 ceph_check_caps(ci
, 0, NULL
);
203 spin_unlock(&ci
->i_ceph_lock
);
206 if ((wanted
& CEPH_CAP_FILE_RD
) && (wanted
& CEPH_CAP_FILE_WR
))
208 else if (wanted
& CEPH_CAP_FILE_RD
)
210 else if (wanted
& CEPH_CAP_FILE_WR
)
213 if (wanted
& CEPH_CAP_FILE_LAZYIO
)
217 req
= prepare_open_request(inode
->i_sb
, flags
, 0);
223 req
->r_inode
= inode
;
228 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
229 ceph_mdsc_put_request(req
);
231 dout("renew caps %p open result=%d\n", inode
, err
);
232 return err
< 0 ? err
: 0;
236 * If we already have the requisite capabilities, we can satisfy
237 * the open request locally (no need to request new caps from the
238 * MDS). We do, however, need to inform the MDS (asynchronously)
239 * if our wanted caps set expands.
241 int ceph_open(struct inode
*inode
, struct file
*file
)
243 struct ceph_inode_info
*ci
= ceph_inode(inode
);
244 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
245 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
246 struct ceph_mds_request
*req
;
247 struct ceph_file_info
*cf
= file
->private_data
;
249 int flags
, fmode
, wanted
;
252 dout("open file %p is already opened\n", file
);
256 /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */
257 flags
= file
->f_flags
& ~(O_CREAT
|O_EXCL
);
258 if (S_ISDIR(inode
->i_mode
))
259 flags
= O_DIRECTORY
; /* mds likes to know */
261 dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode
,
262 ceph_vinop(inode
), file
, flags
, file
->f_flags
);
263 fmode
= ceph_flags_to_mode(flags
);
264 wanted
= ceph_caps_for_mode(fmode
);
266 /* snapped files are read-only */
267 if (ceph_snap(inode
) != CEPH_NOSNAP
&& (file
->f_mode
& FMODE_WRITE
))
270 /* trivially open snapdir */
271 if (ceph_snap(inode
) == CEPH_SNAPDIR
) {
272 spin_lock(&ci
->i_ceph_lock
);
273 __ceph_get_fmode(ci
, fmode
);
274 spin_unlock(&ci
->i_ceph_lock
);
275 return ceph_init_file(inode
, file
, fmode
);
279 * No need to block if we have caps on the auth MDS (for
280 * write) or any MDS (for read). Update wanted set
283 spin_lock(&ci
->i_ceph_lock
);
284 if (__ceph_is_any_real_caps(ci
) &&
285 (((fmode
& CEPH_FILE_MODE_WR
) == 0) || ci
->i_auth_cap
)) {
286 int mds_wanted
= __ceph_caps_mds_wanted(ci
, true);
287 int issued
= __ceph_caps_issued(ci
, NULL
);
289 dout("open %p fmode %d want %s issued %s using existing\n",
290 inode
, fmode
, ceph_cap_string(wanted
),
291 ceph_cap_string(issued
));
292 __ceph_get_fmode(ci
, fmode
);
293 spin_unlock(&ci
->i_ceph_lock
);
296 if ((issued
& wanted
) != wanted
&&
297 (mds_wanted
& wanted
) != wanted
&&
298 ceph_snap(inode
) != CEPH_SNAPDIR
)
299 ceph_check_caps(ci
, 0, NULL
);
301 return ceph_init_file(inode
, file
, fmode
);
302 } else if (ceph_snap(inode
) != CEPH_NOSNAP
&&
303 (ci
->i_snap_caps
& wanted
) == wanted
) {
304 __ceph_get_fmode(ci
, fmode
);
305 spin_unlock(&ci
->i_ceph_lock
);
306 return ceph_init_file(inode
, file
, fmode
);
309 spin_unlock(&ci
->i_ceph_lock
);
311 dout("open fmode %d wants %s\n", fmode
, ceph_cap_string(wanted
));
312 req
= prepare_open_request(inode
->i_sb
, flags
, 0);
317 req
->r_inode
= inode
;
321 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
323 err
= ceph_init_file(inode
, file
, req
->r_fmode
);
324 ceph_mdsc_put_request(req
);
325 dout("open result=%d on %llx.%llx\n", err
, ceph_vinop(inode
));
332 * Do a lookup + open with a single request. If we get a non-existent
333 * file or symlink, return 1 so the VFS can retry.
335 int ceph_atomic_open(struct inode
*dir
, struct dentry
*dentry
,
336 struct file
*file
, unsigned flags
, umode_t mode
,
339 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
340 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
341 struct ceph_mds_request
*req
;
343 struct ceph_acls_info acls
= {};
347 dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n",
349 d_unhashed(dentry
) ? "unhashed" : "hashed", flags
, mode
);
351 if (dentry
->d_name
.len
> NAME_MAX
)
352 return -ENAMETOOLONG
;
354 if (flags
& O_CREAT
) {
355 err
= ceph_pre_init_acls(dir
, &mode
, &acls
);
361 req
= prepare_open_request(dir
->i_sb
, flags
, mode
);
366 req
->r_dentry
= dget(dentry
);
368 if (flags
& O_CREAT
) {
369 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
370 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
372 req
->r_pagelist
= acls
.pagelist
;
373 acls
.pagelist
= NULL
;
377 mask
= CEPH_STAT_CAP_INODE
| CEPH_CAP_AUTH_SHARED
;
378 if (ceph_security_xattr_wanted(dir
))
379 mask
|= CEPH_CAP_XATTR_SHARED
;
380 req
->r_args
.open
.mask
= cpu_to_le32(mask
);
383 set_bit(CEPH_MDS_R_PARENT_LOCKED
, &req
->r_req_flags
);
384 err
= ceph_mdsc_do_request(mdsc
,
385 (flags
& (O_CREAT
|O_TRUNC
)) ? dir
: NULL
,
387 err
= ceph_handle_snapdir(req
, dentry
, err
);
391 if ((flags
& O_CREAT
) && !req
->r_reply_info
.head
->is_dentry
)
392 err
= ceph_handle_notrace_create(dir
, dentry
);
394 if (d_in_lookup(dentry
)) {
395 dn
= ceph_finish_lookup(req
, dentry
, err
);
399 /* we were given a hashed negative dentry */
404 if (dn
|| d_really_is_negative(dentry
) || d_is_symlink(dentry
)) {
405 /* make vfs retry on splice, ENOENT, or symlink */
406 dout("atomic_open finish_no_open on dn %p\n", dn
);
407 err
= finish_no_open(file
, dn
);
409 dout("atomic_open finish_open on dn %p\n", dn
);
410 if (req
->r_op
== CEPH_MDS_OP_CREATE
&& req
->r_reply_info
.has_create_ino
) {
411 ceph_init_inode_acls(d_inode(dentry
), &acls
);
412 *opened
|= FILE_CREATED
;
414 err
= finish_open(file
, dentry
, ceph_open
, opened
);
417 if (!req
->r_err
&& req
->r_target_inode
)
418 ceph_put_fmode(ceph_inode(req
->r_target_inode
), req
->r_fmode
);
419 ceph_mdsc_put_request(req
);
421 ceph_release_acls_info(&acls
);
422 dout("atomic_open result=%d\n", err
);
426 int ceph_release(struct inode
*inode
, struct file
*file
)
428 struct ceph_inode_info
*ci
= ceph_inode(inode
);
429 struct ceph_file_info
*cf
= file
->private_data
;
431 dout("release inode %p file %p\n", inode
, file
);
432 ceph_put_fmode(ci
, cf
->fmode
);
433 if (cf
->last_readdir
)
434 ceph_mdsc_put_request(cf
->last_readdir
);
435 kfree(cf
->last_name
);
437 kmem_cache_free(ceph_file_cachep
, cf
);
439 /* wake up anyone waiting for caps on this inode */
440 wake_up_all(&ci
->i_cap_wq
);
451 * Read a range of bytes striped over one or more objects. Iterate over
452 * objects we stripe over. (That's not atomic, but good enough for now.)
454 * If we get a short result from the OSD, check against i_size; we need to
455 * only return a short read to the caller if we hit EOF.
457 static int striped_read(struct inode
*inode
,
459 struct page
**pages
, int num_pages
,
460 int page_align
, int *checkeof
)
462 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
463 struct ceph_inode_info
*ci
= ceph_inode(inode
);
468 bool hit_stripe
, was_short
;
471 * we may need to do multiple reads. not atomic, unfortunately.
475 page_idx
= (page_align
+ read
) >> PAGE_SHIFT
;
476 ret
= ceph_osdc_readpages(&fsc
->client
->osdc
, ceph_vino(inode
),
477 &ci
->i_layout
, pos
, &this_len
,
478 ci
->i_truncate_seq
, ci
->i_truncate_size
,
479 pages
+ page_idx
, num_pages
- page_idx
,
480 ((page_align
+ read
) & ~PAGE_MASK
));
483 hit_stripe
= this_len
< len
;
484 was_short
= ret
>= 0 && ret
< this_len
;
485 dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos
, len
, read
,
486 ret
, hit_stripe
? " HITSTRIPE" : "", was_short
? " SHORT" : "");
488 i_size
= i_size_read(inode
);
490 if (was_short
&& (pos
+ ret
< i_size
)) {
491 int zlen
= min(this_len
- ret
, i_size
- pos
- ret
);
492 int zoff
= page_align
+ read
+ ret
;
493 dout(" zero gap %llu to %llu\n",
494 pos
+ ret
, pos
+ ret
+ zlen
);
495 ceph_zero_page_vector_range(zoff
, zlen
, pages
);
503 /* hit stripe and need continue*/
504 if (len
&& hit_stripe
&& pos
< i_size
)
510 /* did we bounce off eof? */
511 if (pos
+ len
> i_size
)
512 *checkeof
= CHECK_EOF
;
515 dout("striped_read returns %d\n", ret
);
520 * Completely synchronous read and write methods. Direct from __user
521 * buffer to osd, or directly to user pages (if O_DIRECT).
523 * If the read spans object boundary, just do multiple reads.
525 static ssize_t
ceph_sync_read(struct kiocb
*iocb
, struct iov_iter
*to
,
528 struct file
*file
= iocb
->ki_filp
;
529 struct inode
*inode
= file_inode(file
);
531 u64 off
= iocb
->ki_pos
;
534 size_t len
= iov_iter_count(to
);
536 dout("sync_read on file %p %llu~%u %s\n", file
, off
,
538 (file
->f_flags
& O_DIRECT
) ? "O_DIRECT" : "");
543 * flush any page cache pages in this range. this
544 * will make concurrent normal and sync io slow,
545 * but it will at least behave sensibly when they are
548 ret
= filemap_write_and_wait_range(inode
->i_mapping
, off
,
553 if (unlikely(to
->type
& ITER_PIPE
)) {
555 ret
= iov_iter_get_pages_alloc(to
, &pages
, len
,
559 num_pages
= DIV_ROUND_UP(ret
+ page_off
, PAGE_SIZE
);
561 ret
= striped_read(inode
, off
, ret
, pages
, num_pages
,
564 iov_iter_advance(to
, ret
);
567 iov_iter_advance(to
, 0);
569 ceph_put_page_vector(pages
, num_pages
, false);
571 num_pages
= calc_pages_for(off
, len
);
572 pages
= ceph_alloc_page_vector(num_pages
, GFP_KERNEL
);
574 return PTR_ERR(pages
);
576 ret
= striped_read(inode
, off
, len
, pages
, num_pages
,
577 (off
& ~PAGE_MASK
), checkeof
);
583 size_t page_off
= off
& ~PAGE_MASK
;
584 size_t copy
= min_t(size_t, left
,
585 PAGE_SIZE
- page_off
);
586 l
= copy_page_to_iter(pages
[k
++], page_off
,
594 ceph_release_page_vector(pages
, num_pages
);
597 if (off
> iocb
->ki_pos
) {
598 ret
= off
- iocb
->ki_pos
;
602 dout("sync_read result %zd\n", ret
);
606 struct ceph_aio_request
{
611 struct list_head osd_reqs
;
613 atomic_t pending_reqs
;
614 struct timespec mtime
;
615 struct ceph_cap_flush
*prealloc_cf
;
618 struct ceph_aio_work
{
619 struct work_struct work
;
620 struct ceph_osd_request
*req
;
623 static void ceph_aio_retry_work(struct work_struct
*work
);
625 static void ceph_aio_complete(struct inode
*inode
,
626 struct ceph_aio_request
*aio_req
)
628 struct ceph_inode_info
*ci
= ceph_inode(inode
);
631 if (!atomic_dec_and_test(&aio_req
->pending_reqs
))
634 ret
= aio_req
->error
;
636 ret
= aio_req
->total_len
;
638 dout("ceph_aio_complete %p rc %d\n", inode
, ret
);
640 if (ret
>= 0 && aio_req
->write
) {
643 loff_t endoff
= aio_req
->iocb
->ki_pos
+ aio_req
->total_len
;
644 if (endoff
> i_size_read(inode
)) {
645 if (ceph_inode_set_size(inode
, endoff
))
646 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
649 spin_lock(&ci
->i_ceph_lock
);
650 ci
->i_inline_version
= CEPH_INLINE_NONE
;
651 dirty
= __ceph_mark_dirty_caps(ci
, CEPH_CAP_FILE_WR
,
652 &aio_req
->prealloc_cf
);
653 spin_unlock(&ci
->i_ceph_lock
);
655 __mark_inode_dirty(inode
, dirty
);
659 ceph_put_cap_refs(ci
, (aio_req
->write
? CEPH_CAP_FILE_WR
:
662 aio_req
->iocb
->ki_complete(aio_req
->iocb
, ret
, 0);
664 ceph_free_cap_flush(aio_req
->prealloc_cf
);
668 static void ceph_aio_complete_req(struct ceph_osd_request
*req
)
670 int rc
= req
->r_result
;
671 struct inode
*inode
= req
->r_inode
;
672 struct ceph_aio_request
*aio_req
= req
->r_priv
;
673 struct ceph_osd_data
*osd_data
= osd_req_op_extent_osd_data(req
, 0);
674 int num_pages
= calc_pages_for((u64
)osd_data
->alignment
,
677 dout("ceph_aio_complete_req %p rc %d bytes %llu\n",
678 inode
, rc
, osd_data
->length
);
680 if (rc
== -EOLDSNAPC
) {
681 struct ceph_aio_work
*aio_work
;
682 BUG_ON(!aio_req
->write
);
684 aio_work
= kmalloc(sizeof(*aio_work
), GFP_NOFS
);
686 INIT_WORK(&aio_work
->work
, ceph_aio_retry_work
);
688 queue_work(ceph_inode_to_client(inode
)->wb_wq
,
693 } else if (!aio_req
->write
) {
696 if (rc
>= 0 && osd_data
->length
> rc
) {
697 int zoff
= osd_data
->alignment
+ rc
;
698 int zlen
= osd_data
->length
- rc
;
700 * If read is satisfied by single OSD request,
701 * it can pass EOF. Otherwise read is within
704 if (aio_req
->num_reqs
== 1) {
705 loff_t i_size
= i_size_read(inode
);
706 loff_t endoff
= aio_req
->iocb
->ki_pos
+ rc
;
708 zlen
= min_t(size_t, zlen
,
710 aio_req
->total_len
= rc
+ zlen
;
714 ceph_zero_page_vector_range(zoff
, zlen
,
719 ceph_put_page_vector(osd_data
->pages
, num_pages
, !aio_req
->write
);
720 ceph_osdc_put_request(req
);
723 cmpxchg(&aio_req
->error
, 0, rc
);
725 ceph_aio_complete(inode
, aio_req
);
729 static void ceph_aio_retry_work(struct work_struct
*work
)
731 struct ceph_aio_work
*aio_work
=
732 container_of(work
, struct ceph_aio_work
, work
);
733 struct ceph_osd_request
*orig_req
= aio_work
->req
;
734 struct ceph_aio_request
*aio_req
= orig_req
->r_priv
;
735 struct inode
*inode
= orig_req
->r_inode
;
736 struct ceph_inode_info
*ci
= ceph_inode(inode
);
737 struct ceph_snap_context
*snapc
;
738 struct ceph_osd_request
*req
;
741 spin_lock(&ci
->i_ceph_lock
);
742 if (__ceph_have_pending_cap_snap(ci
)) {
743 struct ceph_cap_snap
*capsnap
=
744 list_last_entry(&ci
->i_cap_snaps
,
745 struct ceph_cap_snap
,
747 snapc
= ceph_get_snap_context(capsnap
->context
);
749 BUG_ON(!ci
->i_head_snapc
);
750 snapc
= ceph_get_snap_context(ci
->i_head_snapc
);
752 spin_unlock(&ci
->i_ceph_lock
);
754 req
= ceph_osdc_alloc_request(orig_req
->r_osdc
, snapc
, 2,
762 req
->r_flags
= CEPH_OSD_FLAG_ORDERSNAP
| CEPH_OSD_FLAG_WRITE
;
763 ceph_oloc_copy(&req
->r_base_oloc
, &orig_req
->r_base_oloc
);
764 ceph_oid_copy(&req
->r_base_oid
, &orig_req
->r_base_oid
);
766 ret
= ceph_osdc_alloc_messages(req
, GFP_NOFS
);
768 ceph_osdc_put_request(req
);
773 req
->r_ops
[0] = orig_req
->r_ops
[0];
774 osd_req_op_init(req
, 1, CEPH_OSD_OP_STARTSYNC
, 0);
776 req
->r_mtime
= aio_req
->mtime
;
777 req
->r_data_offset
= req
->r_ops
[0].extent
.offset
;
779 ceph_osdc_put_request(orig_req
);
781 req
->r_callback
= ceph_aio_complete_req
;
782 req
->r_inode
= inode
;
783 req
->r_priv
= aio_req
;
784 req
->r_abort_on_full
= true;
786 ret
= ceph_osdc_start_request(req
->r_osdc
, req
, false);
790 ceph_aio_complete_req(req
);
793 ceph_put_snap_context(snapc
);
798 ceph_direct_read_write(struct kiocb
*iocb
, struct iov_iter
*iter
,
799 struct ceph_snap_context
*snapc
,
800 struct ceph_cap_flush
**pcf
)
802 struct file
*file
= iocb
->ki_filp
;
803 struct inode
*inode
= file_inode(file
);
804 struct ceph_inode_info
*ci
= ceph_inode(inode
);
805 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
806 struct ceph_vino vino
;
807 struct ceph_osd_request
*req
;
809 struct ceph_aio_request
*aio_req
= NULL
;
813 struct timespec mtime
= current_time(inode
);
814 size_t count
= iov_iter_count(iter
);
815 loff_t pos
= iocb
->ki_pos
;
816 bool write
= iov_iter_rw(iter
) == WRITE
;
818 if (write
&& ceph_snap(file_inode(file
)) != CEPH_NOSNAP
)
821 dout("sync_direct_read_write (%s) on file %p %lld~%u\n",
822 (write
? "write" : "read"), file
, pos
, (unsigned)count
);
824 ret
= filemap_write_and_wait_range(inode
->i_mapping
, pos
, pos
+ count
);
829 int ret2
= invalidate_inode_pages2_range(inode
->i_mapping
,
831 (pos
+ count
) >> PAGE_SHIFT
);
833 dout("invalidate_inode_pages2_range returned %d\n", ret2
);
835 flags
= CEPH_OSD_FLAG_ORDERSNAP
| CEPH_OSD_FLAG_WRITE
;
837 flags
= CEPH_OSD_FLAG_READ
;
840 while (iov_iter_count(iter
) > 0) {
841 u64 size
= dio_get_pagev_size(iter
);
845 vino
= ceph_vino(inode
);
846 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
848 /*include a 'startsync' command*/
850 write
? CEPH_OSD_OP_WRITE
:
862 pages
= dio_get_pages_alloc(iter
, len
, &start
, &num_pages
);
864 ceph_osdc_put_request(req
);
865 ret
= PTR_ERR(pages
);
870 * To simplify error handling, allow AIO when IO within i_size
871 * or IO can be satisfied by single OSD request.
873 if (pos
== iocb
->ki_pos
&& !is_sync_kiocb(iocb
) &&
874 (len
== count
|| pos
+ count
<= i_size_read(inode
))) {
875 aio_req
= kzalloc(sizeof(*aio_req
), GFP_KERNEL
);
877 aio_req
->iocb
= iocb
;
878 aio_req
->write
= write
;
879 INIT_LIST_HEAD(&aio_req
->osd_reqs
);
881 aio_req
->mtime
= mtime
;
882 swap(aio_req
->prealloc_cf
, *pcf
);
890 * throw out any page cache pages in this range. this
893 truncate_inode_pages_range(inode
->i_mapping
, pos
,
894 (pos
+len
) | (PAGE_SIZE
- 1));
896 osd_req_op_init(req
, 1, CEPH_OSD_OP_STARTSYNC
, 0);
897 req
->r_mtime
= mtime
;
900 osd_req_op_extent_osd_data_pages(req
, 0, pages
, len
, start
,
904 aio_req
->total_len
+= len
;
906 atomic_inc(&aio_req
->pending_reqs
);
908 req
->r_callback
= ceph_aio_complete_req
;
909 req
->r_inode
= inode
;
910 req
->r_priv
= aio_req
;
911 list_add_tail(&req
->r_unsafe_item
, &aio_req
->osd_reqs
);
914 iov_iter_advance(iter
, len
);
918 ret
= ceph_osdc_start_request(req
->r_osdc
, req
, false);
920 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
922 size
= i_size_read(inode
);
926 if (ret
>= 0 && ret
< len
&& pos
+ ret
< size
) {
927 int zlen
= min_t(size_t, len
- ret
,
929 ceph_zero_page_vector_range(start
+ ret
, zlen
,
937 ceph_put_page_vector(pages
, num_pages
, !write
);
939 ceph_osdc_put_request(req
);
944 iov_iter_advance(iter
, len
);
946 if (!write
&& pos
>= size
)
949 if (write
&& pos
> size
) {
950 if (ceph_inode_set_size(inode
, pos
))
951 ceph_check_caps(ceph_inode(inode
),
960 if (aio_req
->num_reqs
== 0) {
965 ceph_get_cap_refs(ci
, write
? CEPH_CAP_FILE_WR
:
968 list_splice(&aio_req
->osd_reqs
, &osd_reqs
);
969 while (!list_empty(&osd_reqs
)) {
970 req
= list_first_entry(&osd_reqs
,
971 struct ceph_osd_request
,
973 list_del_init(&req
->r_unsafe_item
);
975 ret
= ceph_osdc_start_request(req
->r_osdc
,
979 ceph_aio_complete_req(req
);
985 if (ret
!= -EOLDSNAPC
&& pos
> iocb
->ki_pos
) {
986 ret
= pos
- iocb
->ki_pos
;
993 * Synchronous write, straight from __user pointer or user pages.
995 * If write spans object boundary, just do multiple writes. (For a
996 * correct atomic write, we should e.g. take write locks on all
997 * objects, rollback on failure, etc.)
1000 ceph_sync_write(struct kiocb
*iocb
, struct iov_iter
*from
, loff_t pos
,
1001 struct ceph_snap_context
*snapc
)
1003 struct file
*file
= iocb
->ki_filp
;
1004 struct inode
*inode
= file_inode(file
);
1005 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1006 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
1007 struct ceph_vino vino
;
1008 struct ceph_osd_request
*req
;
1009 struct page
**pages
;
1016 struct timespec mtime
= current_time(inode
);
1017 size_t count
= iov_iter_count(from
);
1019 if (ceph_snap(file_inode(file
)) != CEPH_NOSNAP
)
1022 dout("sync_write on file %p %lld~%u\n", file
, pos
, (unsigned)count
);
1024 ret
= filemap_write_and_wait_range(inode
->i_mapping
, pos
, pos
+ count
);
1028 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
1030 (pos
+ count
) >> PAGE_SHIFT
);
1032 dout("invalidate_inode_pages2_range returned %d\n", ret
);
1034 flags
= CEPH_OSD_FLAG_ORDERSNAP
| CEPH_OSD_FLAG_WRITE
;
1036 while ((len
= iov_iter_count(from
)) > 0) {
1040 vino
= ceph_vino(inode
);
1041 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
1042 vino
, pos
, &len
, 0, 1,
1043 CEPH_OSD_OP_WRITE
, flags
, snapc
,
1045 ci
->i_truncate_size
,
1053 * write from beginning of first page,
1054 * regardless of io alignment
1056 num_pages
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1058 pages
= ceph_alloc_page_vector(num_pages
, GFP_KERNEL
);
1059 if (IS_ERR(pages
)) {
1060 ret
= PTR_ERR(pages
);
1065 for (n
= 0; n
< num_pages
; n
++) {
1066 size_t plen
= min_t(size_t, left
, PAGE_SIZE
);
1067 ret
= copy_page_from_iter(pages
[n
], 0, plen
, from
);
1076 ceph_release_page_vector(pages
, num_pages
);
1080 req
->r_inode
= inode
;
1082 osd_req_op_extent_osd_data_pages(req
, 0, pages
, len
, 0,
1085 req
->r_mtime
= mtime
;
1086 ret
= ceph_osdc_start_request(&fsc
->client
->osdc
, req
, false);
1088 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
1091 ceph_osdc_put_request(req
);
1096 if (pos
> i_size_read(inode
)) {
1097 check_caps
= ceph_inode_set_size(inode
, pos
);
1099 ceph_check_caps(ceph_inode(inode
),
1100 CHECK_CAPS_AUTHONLY
,
1107 if (ret
!= -EOLDSNAPC
&& written
> 0) {
1115 * Wrap generic_file_aio_read with checks for cap bits on the inode.
1116 * Atomically grab references, so that those bits are not released
1117 * back to the MDS mid-read.
1119 * Hmm, the sync read case isn't actually async... should it be?
1121 static ssize_t
ceph_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1123 struct file
*filp
= iocb
->ki_filp
;
1124 struct ceph_file_info
*fi
= filp
->private_data
;
1125 size_t len
= iov_iter_count(to
);
1126 struct inode
*inode
= file_inode(filp
);
1127 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1128 struct page
*pinned_page
= NULL
;
1131 int retry_op
= 0, read
= 0;
1134 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
1135 inode
, ceph_vinop(inode
), iocb
->ki_pos
, (unsigned)len
, inode
);
1137 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
1138 want
= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_LAZYIO
;
1140 want
= CEPH_CAP_FILE_CACHE
;
1141 ret
= ceph_get_caps(ci
, CEPH_CAP_FILE_RD
, want
, -1, &got
, &pinned_page
);
1145 if ((got
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)) == 0 ||
1146 (iocb
->ki_flags
& IOCB_DIRECT
) ||
1147 (fi
->flags
& CEPH_F_SYNC
)) {
1149 dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n",
1150 inode
, ceph_vinop(inode
), iocb
->ki_pos
, (unsigned)len
,
1151 ceph_cap_string(got
));
1153 if (ci
->i_inline_version
== CEPH_INLINE_NONE
) {
1154 if (!retry_op
&& (iocb
->ki_flags
& IOCB_DIRECT
)) {
1155 ret
= ceph_direct_read_write(iocb
, to
,
1157 if (ret
>= 0 && ret
< len
)
1158 retry_op
= CHECK_EOF
;
1160 ret
= ceph_sync_read(iocb
, to
, &retry_op
);
1163 retry_op
= READ_INLINE
;
1166 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
1167 inode
, ceph_vinop(inode
), iocb
->ki_pos
, (unsigned)len
,
1168 ceph_cap_string(got
));
1169 current
->journal_info
= filp
;
1170 ret
= generic_file_read_iter(iocb
, to
);
1171 current
->journal_info
= NULL
;
1173 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
1174 inode
, ceph_vinop(inode
), ceph_cap_string(got
), (int)ret
);
1176 put_page(pinned_page
);
1179 ceph_put_cap_refs(ci
, got
);
1180 if (retry_op
> HAVE_RETRIED
&& ret
>= 0) {
1182 struct page
*page
= NULL
;
1184 if (retry_op
== READ_INLINE
) {
1185 page
= __page_cache_alloc(GFP_KERNEL
);
1190 statret
= __ceph_do_getattr(inode
, page
,
1191 CEPH_STAT_CAP_INLINE_DATA
, !!page
);
1195 if (statret
== -ENODATA
) {
1196 BUG_ON(retry_op
!= READ_INLINE
);
1202 i_size
= i_size_read(inode
);
1203 if (retry_op
== READ_INLINE
) {
1204 BUG_ON(ret
> 0 || read
> 0);
1205 if (iocb
->ki_pos
< i_size
&&
1206 iocb
->ki_pos
< PAGE_SIZE
) {
1207 loff_t end
= min_t(loff_t
, i_size
,
1208 iocb
->ki_pos
+ len
);
1209 end
= min_t(loff_t
, end
, PAGE_SIZE
);
1211 zero_user_segment(page
, statret
, end
);
1212 ret
= copy_page_to_iter(page
,
1213 iocb
->ki_pos
& ~PAGE_MASK
,
1214 end
- iocb
->ki_pos
, to
);
1215 iocb
->ki_pos
+= ret
;
1218 if (iocb
->ki_pos
< i_size
&& read
< len
) {
1219 size_t zlen
= min_t(size_t, len
- read
,
1220 i_size
- iocb
->ki_pos
);
1221 ret
= iov_iter_zero(zlen
, to
);
1222 iocb
->ki_pos
+= ret
;
1225 __free_pages(page
, 0);
1229 /* hit EOF or hole? */
1230 if (retry_op
== CHECK_EOF
&& iocb
->ki_pos
< i_size
&&
1232 dout("sync_read hit hole, ppos %lld < size %lld"
1233 ", reading more\n", iocb
->ki_pos
, i_size
);
1237 retry_op
= HAVE_RETRIED
;
1249 * Take cap references to avoid releasing caps to MDS mid-write.
1251 * If we are synchronous, and write with an old snap context, the OSD
1252 * may return EOLDSNAPC. In that case, retry the write.. _after_
1253 * dropping our cap refs and allowing the pending snap to logically
1254 * complete _before_ this write occurs.
1256 * If we are near ENOSPC, write synchronously.
1258 static ssize_t
ceph_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1260 struct file
*file
= iocb
->ki_filp
;
1261 struct ceph_file_info
*fi
= file
->private_data
;
1262 struct inode
*inode
= file_inode(file
);
1263 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1264 struct ceph_osd_client
*osdc
=
1265 &ceph_sb_to_client(inode
->i_sb
)->client
->osdc
;
1266 struct ceph_cap_flush
*prealloc_cf
;
1267 ssize_t count
, written
= 0;
1271 if (ceph_snap(inode
) != CEPH_NOSNAP
)
1274 prealloc_cf
= ceph_alloc_cap_flush();
1280 /* We can write back this queue in page reclaim */
1281 current
->backing_dev_info
= inode_to_bdi(inode
);
1283 if (iocb
->ki_flags
& IOCB_APPEND
) {
1284 err
= ceph_do_getattr(inode
, CEPH_STAT_CAP_SIZE
, false);
1289 err
= generic_write_checks(iocb
, from
);
1294 count
= iov_iter_count(from
);
1295 err
= file_remove_privs(file
);
1299 err
= file_update_time(file
);
1303 if (ci
->i_inline_version
!= CEPH_INLINE_NONE
) {
1304 err
= ceph_uninline_data(file
, NULL
);
1310 if (ceph_osdmap_flag(osdc
, CEPH_OSDMAP_FULL
)) {
1315 dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
1316 inode
, ceph_vinop(inode
), pos
, count
, i_size_read(inode
));
1317 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
1318 want
= CEPH_CAP_FILE_BUFFER
| CEPH_CAP_FILE_LAZYIO
;
1320 want
= CEPH_CAP_FILE_BUFFER
;
1322 err
= ceph_get_caps(ci
, CEPH_CAP_FILE_WR
, want
, pos
+ count
,
1327 dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
1328 inode
, ceph_vinop(inode
), pos
, count
, ceph_cap_string(got
));
1330 if ((got
& (CEPH_CAP_FILE_BUFFER
|CEPH_CAP_FILE_LAZYIO
)) == 0 ||
1331 (iocb
->ki_flags
& IOCB_DIRECT
) || (fi
->flags
& CEPH_F_SYNC
)) {
1332 struct ceph_snap_context
*snapc
;
1333 struct iov_iter data
;
1334 inode_unlock(inode
);
1336 spin_lock(&ci
->i_ceph_lock
);
1337 if (__ceph_have_pending_cap_snap(ci
)) {
1338 struct ceph_cap_snap
*capsnap
=
1339 list_last_entry(&ci
->i_cap_snaps
,
1340 struct ceph_cap_snap
,
1342 snapc
= ceph_get_snap_context(capsnap
->context
);
1344 BUG_ON(!ci
->i_head_snapc
);
1345 snapc
= ceph_get_snap_context(ci
->i_head_snapc
);
1347 spin_unlock(&ci
->i_ceph_lock
);
1349 /* we might need to revert back to that point */
1351 if (iocb
->ki_flags
& IOCB_DIRECT
)
1352 written
= ceph_direct_read_write(iocb
, &data
, snapc
,
1355 written
= ceph_sync_write(iocb
, &data
, pos
, snapc
);
1356 if (written
== -EOLDSNAPC
) {
1357 dout("aio_write %p %llx.%llx %llu~%u"
1358 "got EOLDSNAPC, retrying\n",
1359 inode
, ceph_vinop(inode
),
1360 pos
, (unsigned)count
);
1365 iov_iter_advance(from
, written
);
1366 ceph_put_snap_context(snapc
);
1369 * No need to acquire the i_truncate_mutex. Because
1370 * the MDS revokes Fwb caps before sending truncate
1371 * message to us. We can't get Fwb cap while there
1372 * are pending vmtruncate. So write and vmtruncate
1373 * can not run at the same time
1375 written
= generic_perform_write(file
, from
, pos
);
1376 if (likely(written
>= 0))
1377 iocb
->ki_pos
= pos
+ written
;
1378 inode_unlock(inode
);
1383 spin_lock(&ci
->i_ceph_lock
);
1384 ci
->i_inline_version
= CEPH_INLINE_NONE
;
1385 dirty
= __ceph_mark_dirty_caps(ci
, CEPH_CAP_FILE_WR
,
1387 spin_unlock(&ci
->i_ceph_lock
);
1389 __mark_inode_dirty(inode
, dirty
);
1392 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
1393 inode
, ceph_vinop(inode
), pos
, (unsigned)count
,
1394 ceph_cap_string(got
));
1395 ceph_put_cap_refs(ci
, got
);
1398 if (ceph_osdmap_flag(osdc
, CEPH_OSDMAP_NEARFULL
))
1399 iocb
->ki_flags
|= IOCB_DSYNC
;
1401 written
= generic_write_sync(iocb
, written
);
1407 inode_unlock(inode
);
1409 ceph_free_cap_flush(prealloc_cf
);
1410 current
->backing_dev_info
= NULL
;
1411 return written
? written
: err
;
1415 * llseek. be sure to verify file size on SEEK_END.
1417 static loff_t
ceph_llseek(struct file
*file
, loff_t offset
, int whence
)
1419 struct inode
*inode
= file
->f_mapping
->host
;
1425 if (whence
== SEEK_END
|| whence
== SEEK_DATA
|| whence
== SEEK_HOLE
) {
1426 ret
= ceph_do_getattr(inode
, CEPH_STAT_CAP_SIZE
, false);
1431 i_size
= i_size_read(inode
);
1438 * Here we special-case the lseek(fd, 0, SEEK_CUR)
1439 * position-querying operation. Avoid rewriting the "same"
1440 * f_pos value back to the file because a concurrent read(),
1441 * write() or lseek() might have altered it
1447 offset
+= file
->f_pos
;
1450 if (offset
>= i_size
) {
1456 if (offset
>= i_size
) {
1464 ret
= vfs_setpos(file
, offset
, inode
->i_sb
->s_maxbytes
);
1467 inode_unlock(inode
);
1471 static inline void ceph_zero_partial_page(
1472 struct inode
*inode
, loff_t offset
, unsigned size
)
1475 pgoff_t index
= offset
>> PAGE_SHIFT
;
1477 page
= find_lock_page(inode
->i_mapping
, index
);
1479 wait_on_page_writeback(page
);
1480 zero_user(page
, offset
& (PAGE_SIZE
- 1), size
);
1486 static void ceph_zero_pagecache_range(struct inode
*inode
, loff_t offset
,
1489 loff_t nearly
= round_up(offset
, PAGE_SIZE
);
1490 if (offset
< nearly
) {
1491 loff_t size
= nearly
- offset
;
1494 ceph_zero_partial_page(inode
, offset
, size
);
1498 if (length
>= PAGE_SIZE
) {
1499 loff_t size
= round_down(length
, PAGE_SIZE
);
1500 truncate_pagecache_range(inode
, offset
, offset
+ size
- 1);
1505 ceph_zero_partial_page(inode
, offset
, length
);
1508 static int ceph_zero_partial_object(struct inode
*inode
,
1509 loff_t offset
, loff_t
*length
)
1511 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1512 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
1513 struct ceph_osd_request
*req
;
1519 op
= offset
? CEPH_OSD_OP_DELETE
: CEPH_OSD_OP_TRUNCATE
;
1522 op
= CEPH_OSD_OP_ZERO
;
1525 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
1529 CEPH_OSD_FLAG_WRITE
,
1536 req
->r_mtime
= inode
->i_mtime
;
1537 ret
= ceph_osdc_start_request(&fsc
->client
->osdc
, req
, false);
1539 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
1543 ceph_osdc_put_request(req
);
1549 static int ceph_zero_objects(struct inode
*inode
, loff_t offset
, loff_t length
)
1552 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1553 s32 stripe_unit
= ci
->i_layout
.stripe_unit
;
1554 s32 stripe_count
= ci
->i_layout
.stripe_count
;
1555 s32 object_size
= ci
->i_layout
.object_size
;
1556 u64 object_set_size
= object_size
* stripe_count
;
1559 /* round offset up to next period boundary */
1560 nearly
= offset
+ object_set_size
- 1;
1562 nearly
-= do_div(t
, object_set_size
);
1564 while (length
&& offset
< nearly
) {
1565 loff_t size
= length
;
1566 ret
= ceph_zero_partial_object(inode
, offset
, &size
);
1572 while (length
>= object_set_size
) {
1574 loff_t pos
= offset
;
1575 for (i
= 0; i
< stripe_count
; ++i
) {
1576 ret
= ceph_zero_partial_object(inode
, pos
, NULL
);
1581 offset
+= object_set_size
;
1582 length
-= object_set_size
;
1585 loff_t size
= length
;
1586 ret
= ceph_zero_partial_object(inode
, offset
, &size
);
1595 static long ceph_fallocate(struct file
*file
, int mode
,
1596 loff_t offset
, loff_t length
)
1598 struct ceph_file_info
*fi
= file
->private_data
;
1599 struct inode
*inode
= file_inode(file
);
1600 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1601 struct ceph_osd_client
*osdc
=
1602 &ceph_inode_to_client(inode
)->client
->osdc
;
1603 struct ceph_cap_flush
*prealloc_cf
;
1610 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
1613 if (!S_ISREG(inode
->i_mode
))
1616 prealloc_cf
= ceph_alloc_cap_flush();
1622 if (ceph_snap(inode
) != CEPH_NOSNAP
) {
1627 if (ceph_osdmap_flag(osdc
, CEPH_OSDMAP_FULL
) &&
1628 !(mode
& FALLOC_FL_PUNCH_HOLE
)) {
1633 if (ci
->i_inline_version
!= CEPH_INLINE_NONE
) {
1634 ret
= ceph_uninline_data(file
, NULL
);
1639 size
= i_size_read(inode
);
1640 if (!(mode
& FALLOC_FL_KEEP_SIZE
))
1641 endoff
= offset
+ length
;
1643 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
1644 want
= CEPH_CAP_FILE_BUFFER
| CEPH_CAP_FILE_LAZYIO
;
1646 want
= CEPH_CAP_FILE_BUFFER
;
1648 ret
= ceph_get_caps(ci
, CEPH_CAP_FILE_WR
, want
, endoff
, &got
, NULL
);
1652 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
1654 ceph_zero_pagecache_range(inode
, offset
, length
);
1655 ret
= ceph_zero_objects(inode
, offset
, length
);
1656 } else if (endoff
> size
) {
1657 truncate_pagecache_range(inode
, size
, -1);
1658 if (ceph_inode_set_size(inode
, endoff
))
1659 ceph_check_caps(ceph_inode(inode
),
1660 CHECK_CAPS_AUTHONLY
, NULL
);
1664 spin_lock(&ci
->i_ceph_lock
);
1665 ci
->i_inline_version
= CEPH_INLINE_NONE
;
1666 dirty
= __ceph_mark_dirty_caps(ci
, CEPH_CAP_FILE_WR
,
1668 spin_unlock(&ci
->i_ceph_lock
);
1670 __mark_inode_dirty(inode
, dirty
);
1673 ceph_put_cap_refs(ci
, got
);
1675 inode_unlock(inode
);
1676 ceph_free_cap_flush(prealloc_cf
);
1680 const struct file_operations ceph_file_fops
= {
1682 .release
= ceph_release
,
1683 .llseek
= ceph_llseek
,
1684 .read_iter
= ceph_read_iter
,
1685 .write_iter
= ceph_write_iter
,
1687 .fsync
= ceph_fsync
,
1689 .flock
= ceph_flock
,
1690 .splice_read
= generic_file_splice_read
,
1691 .splice_write
= iter_file_splice_write
,
1692 .unlocked_ioctl
= ceph_ioctl
,
1693 .compat_ioctl
= ceph_ioctl
,
1694 .fallocate
= ceph_fallocate
,