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.
39 * Prepare an open request. Preallocate ceph_cap to avoid an
40 * inopportune ENOMEM later.
42 static struct ceph_mds_request
*
43 prepare_open_request(struct super_block
*sb
, int flags
, int create_mode
)
45 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
46 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
47 struct ceph_mds_request
*req
;
48 int want_auth
= USE_ANY_MDS
;
49 int op
= (flags
& O_CREAT
) ? CEPH_MDS_OP_CREATE
: CEPH_MDS_OP_OPEN
;
51 if (flags
& (O_WRONLY
|O_RDWR
|O_CREAT
|O_TRUNC
))
52 want_auth
= USE_AUTH_MDS
;
54 req
= ceph_mdsc_create_request(mdsc
, op
, want_auth
);
57 req
->r_fmode
= ceph_flags_to_mode(flags
);
58 req
->r_args
.open
.flags
= cpu_to_le32(flags
);
59 req
->r_args
.open
.mode
= cpu_to_le32(create_mode
);
65 * initialize private struct file data.
66 * if we fail, clean up by dropping fmode reference on the ceph_inode
68 static int ceph_init_file(struct inode
*inode
, struct file
*file
, int fmode
)
70 struct ceph_file_info
*cf
;
72 struct ceph_inode_info
*ci
= ceph_inode(inode
);
73 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
74 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
76 switch (inode
->i_mode
& S_IFMT
) {
78 /* First file open request creates the cookie, we want to keep
79 * this cookie around for the filetime of the inode as not to
80 * have to worry about fscache register / revoke / operation
83 * Also, if we know the operation is going to invalidate data
84 * (non readonly) just nuke the cache right away.
86 ceph_fscache_register_inode_cookie(mdsc
->fsc
, ci
);
87 if ((fmode
& CEPH_FILE_MODE_WR
))
88 ceph_fscache_invalidate(inode
);
90 dout("init_file %p %p 0%o (regular)\n", inode
, file
,
92 cf
= kmem_cache_alloc(ceph_file_cachep
, GFP_NOFS
| __GFP_ZERO
);
94 ceph_put_fmode(ceph_inode(inode
), fmode
); /* clean up */
99 file
->private_data
= cf
;
100 BUG_ON(inode
->i_fop
->release
!= ceph_release
);
104 dout("init_file %p %p 0%o (symlink)\n", inode
, file
,
106 ceph_put_fmode(ceph_inode(inode
), fmode
); /* clean up */
110 dout("init_file %p %p 0%o (special)\n", inode
, file
,
113 * we need to drop the open ref now, since we don't
114 * have .release set to ceph_release.
116 ceph_put_fmode(ceph_inode(inode
), fmode
); /* clean up */
117 BUG_ON(inode
->i_fop
->release
== ceph_release
);
119 /* call the proper open fop */
120 ret
= inode
->i_fop
->open(inode
, file
);
126 * If we already have the requisite capabilities, we can satisfy
127 * the open request locally (no need to request new caps from the
128 * MDS). We do, however, need to inform the MDS (asynchronously)
129 * if our wanted caps set expands.
131 int ceph_open(struct inode
*inode
, struct file
*file
)
133 struct ceph_inode_info
*ci
= ceph_inode(inode
);
134 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
135 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
136 struct ceph_mds_request
*req
;
137 struct ceph_file_info
*cf
= file
->private_data
;
138 struct inode
*parent_inode
= NULL
;
140 int flags
, fmode
, wanted
;
143 dout("open file %p is already opened\n", file
);
147 /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */
148 flags
= file
->f_flags
& ~(O_CREAT
|O_EXCL
);
149 if (S_ISDIR(inode
->i_mode
))
150 flags
= O_DIRECTORY
; /* mds likes to know */
152 dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode
,
153 ceph_vinop(inode
), file
, flags
, file
->f_flags
);
154 fmode
= ceph_flags_to_mode(flags
);
155 wanted
= ceph_caps_for_mode(fmode
);
157 /* snapped files are read-only */
158 if (ceph_snap(inode
) != CEPH_NOSNAP
&& (file
->f_mode
& FMODE_WRITE
))
161 /* trivially open snapdir */
162 if (ceph_snap(inode
) == CEPH_SNAPDIR
) {
163 spin_lock(&ci
->i_ceph_lock
);
164 __ceph_get_fmode(ci
, fmode
);
165 spin_unlock(&ci
->i_ceph_lock
);
166 return ceph_init_file(inode
, file
, fmode
);
170 * No need to block if we have caps on the auth MDS (for
171 * write) or any MDS (for read). Update wanted set
174 spin_lock(&ci
->i_ceph_lock
);
175 if (__ceph_is_any_real_caps(ci
) &&
176 (((fmode
& CEPH_FILE_MODE_WR
) == 0) || ci
->i_auth_cap
)) {
177 int mds_wanted
= __ceph_caps_mds_wanted(ci
);
178 int issued
= __ceph_caps_issued(ci
, NULL
);
180 dout("open %p fmode %d want %s issued %s using existing\n",
181 inode
, fmode
, ceph_cap_string(wanted
),
182 ceph_cap_string(issued
));
183 __ceph_get_fmode(ci
, fmode
);
184 spin_unlock(&ci
->i_ceph_lock
);
187 if ((issued
& wanted
) != wanted
&&
188 (mds_wanted
& wanted
) != wanted
&&
189 ceph_snap(inode
) != CEPH_SNAPDIR
)
190 ceph_check_caps(ci
, 0, NULL
);
192 return ceph_init_file(inode
, file
, fmode
);
193 } else if (ceph_snap(inode
) != CEPH_NOSNAP
&&
194 (ci
->i_snap_caps
& wanted
) == wanted
) {
195 __ceph_get_fmode(ci
, fmode
);
196 spin_unlock(&ci
->i_ceph_lock
);
197 return ceph_init_file(inode
, file
, fmode
);
200 spin_unlock(&ci
->i_ceph_lock
);
202 dout("open fmode %d wants %s\n", fmode
, ceph_cap_string(wanted
));
203 req
= prepare_open_request(inode
->i_sb
, flags
, 0);
208 req
->r_inode
= inode
;
213 parent_inode
= ceph_get_dentry_parent_inode(file
->f_path
.dentry
);
214 err
= ceph_mdsc_do_request(mdsc
, parent_inode
, req
);
217 err
= ceph_init_file(inode
, file
, req
->r_fmode
);
218 ceph_mdsc_put_request(req
);
219 dout("open result=%d on %llx.%llx\n", err
, ceph_vinop(inode
));
226 * Do a lookup + open with a single request. If we get a non-existent
227 * file or symlink, return 1 so the VFS can retry.
229 int ceph_atomic_open(struct inode
*dir
, struct dentry
*dentry
,
230 struct file
*file
, unsigned flags
, umode_t mode
,
233 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
234 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
235 struct ceph_mds_request
*req
;
237 struct ceph_acls_info acls
= {};
240 dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n",
242 d_unhashed(dentry
) ? "unhashed" : "hashed", flags
, mode
);
244 if (dentry
->d_name
.len
> NAME_MAX
)
245 return -ENAMETOOLONG
;
247 err
= ceph_init_dentry(dentry
);
251 if (flags
& O_CREAT
) {
252 err
= ceph_pre_init_acls(dir
, &mode
, &acls
);
258 req
= prepare_open_request(dir
->i_sb
, flags
, mode
);
263 req
->r_dentry
= dget(dentry
);
265 if (flags
& O_CREAT
) {
266 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
267 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
269 req
->r_pagelist
= acls
.pagelist
;
270 acls
.pagelist
= NULL
;
273 req
->r_locked_dir
= dir
; /* caller holds dir->i_mutex */
274 err
= ceph_mdsc_do_request(mdsc
,
275 (flags
& (O_CREAT
|O_TRUNC
)) ? dir
: NULL
,
280 err
= ceph_handle_snapdir(req
, dentry
, err
);
281 if (err
== 0 && (flags
& O_CREAT
) && !req
->r_reply_info
.head
->is_dentry
)
282 err
= ceph_handle_notrace_create(dir
, dentry
);
284 if (d_unhashed(dentry
)) {
285 dn
= ceph_finish_lookup(req
, dentry
, err
);
289 /* we were given a hashed negative dentry */
294 if (dn
|| dentry
->d_inode
== NULL
|| S_ISLNK(dentry
->d_inode
->i_mode
)) {
295 /* make vfs retry on splice, ENOENT, or symlink */
296 dout("atomic_open finish_no_open on dn %p\n", dn
);
297 err
= finish_no_open(file
, dn
);
299 dout("atomic_open finish_open on dn %p\n", dn
);
300 if (req
->r_op
== CEPH_MDS_OP_CREATE
&& req
->r_reply_info
.has_create_ino
) {
301 ceph_init_inode_acls(dentry
->d_inode
, &acls
);
302 *opened
|= FILE_CREATED
;
304 err
= finish_open(file
, dentry
, ceph_open
, opened
);
307 if (!req
->r_err
&& req
->r_target_inode
)
308 ceph_put_fmode(ceph_inode(req
->r_target_inode
), req
->r_fmode
);
309 ceph_mdsc_put_request(req
);
311 ceph_release_acls_info(&acls
);
312 dout("atomic_open result=%d\n", err
);
316 int ceph_release(struct inode
*inode
, struct file
*file
)
318 struct ceph_inode_info
*ci
= ceph_inode(inode
);
319 struct ceph_file_info
*cf
= file
->private_data
;
321 dout("release inode %p file %p\n", inode
, file
);
322 ceph_put_fmode(ci
, cf
->fmode
);
323 if (cf
->last_readdir
)
324 ceph_mdsc_put_request(cf
->last_readdir
);
325 kfree(cf
->last_name
);
328 kmem_cache_free(ceph_file_cachep
, cf
);
330 /* wake up anyone waiting for caps on this inode */
331 wake_up_all(&ci
->i_cap_wq
);
341 * Read a range of bytes striped over one or more objects. Iterate over
342 * objects we stripe over. (That's not atomic, but good enough for now.)
344 * If we get a short result from the OSD, check against i_size; we need to
345 * only return a short read to the caller if we hit EOF.
347 static int striped_read(struct inode
*inode
,
349 struct page
**pages
, int num_pages
,
350 int *checkeof
, bool o_direct
,
351 unsigned long buf_align
)
353 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
354 struct ceph_inode_info
*ci
= ceph_inode(inode
);
355 u64 pos
, this_len
, left
;
356 int io_align
, page_align
;
359 struct page
**page_pos
;
361 bool hit_stripe
, was_short
;
364 * we may need to do multiple reads. not atomic, unfortunately.
369 pages_left
= num_pages
;
371 io_align
= off
& ~PAGE_MASK
;
375 page_align
= (pos
- io_align
+ buf_align
) & ~PAGE_MASK
;
377 page_align
= pos
& ~PAGE_MASK
;
379 ret
= ceph_osdc_readpages(&fsc
->client
->osdc
, ceph_vino(inode
),
380 &ci
->i_layout
, pos
, &this_len
,
383 page_pos
, pages_left
, page_align
);
386 hit_stripe
= this_len
< left
;
387 was_short
= ret
>= 0 && ret
< this_len
;
388 dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos
, left
, read
,
389 ret
, hit_stripe
? " HITSTRIPE" : "", was_short
? " SHORT" : "");
393 if (was_short
&& (pos
+ ret
< inode
->i_size
)) {
394 u64 tmp
= min(this_len
- ret
,
395 inode
->i_size
- pos
- ret
);
396 dout(" zero gap %llu to %llu\n",
397 pos
+ ret
, pos
+ ret
+ tmp
);
398 ceph_zero_page_vector_range(page_align
+ read
+ ret
,
403 didpages
= (page_align
+ ret
) >> PAGE_CACHE_SHIFT
;
407 page_pos
+= didpages
;
408 pages_left
-= didpages
;
410 /* hit stripe and need continue*/
411 if (left
&& hit_stripe
&& pos
< inode
->i_size
)
417 /* did we bounce off eof? */
418 if (pos
+ left
> inode
->i_size
)
419 *checkeof
= CHECK_EOF
;
422 dout("striped_read returns %d\n", ret
);
427 * Completely synchronous read and write methods. Direct from __user
428 * buffer to osd, or directly to user pages (if O_DIRECT).
430 * If the read spans object boundary, just do multiple reads.
432 static ssize_t
ceph_sync_read(struct kiocb
*iocb
, struct iov_iter
*i
,
435 struct file
*file
= iocb
->ki_filp
;
436 struct inode
*inode
= file_inode(file
);
438 u64 off
= iocb
->ki_pos
;
440 size_t len
= iov_iter_count(i
);
442 dout("sync_read on file %p %llu~%u %s\n", file
, off
,
444 (file
->f_flags
& O_DIRECT
) ? "O_DIRECT" : "");
449 * flush any page cache pages in this range. this
450 * will make concurrent normal and sync io slow,
451 * but it will at least behave sensibly when they are
454 ret
= filemap_write_and_wait_range(inode
->i_mapping
, off
,
459 if (file
->f_flags
& O_DIRECT
) {
460 while (iov_iter_count(i
)) {
464 n
= iov_iter_get_pages_alloc(i
, &pages
, INT_MAX
, &start
);
468 num_pages
= (n
+ start
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
470 ret
= striped_read(inode
, off
, n
,
471 pages
, num_pages
, checkeof
,
474 ceph_put_page_vector(pages
, num_pages
, true);
479 iov_iter_advance(i
, ret
);
484 num_pages
= calc_pages_for(off
, len
);
485 pages
= ceph_alloc_page_vector(num_pages
, GFP_NOFS
);
487 return PTR_ERR(pages
);
488 ret
= striped_read(inode
, off
, len
, pages
,
489 num_pages
, checkeof
, 0, 0);
495 size_t page_off
= off
& ~PAGE_MASK
;
496 size_t copy
= min_t(size_t,
497 PAGE_SIZE
- page_off
, left
);
498 l
= copy_page_to_iter(pages
[k
++], page_off
,
506 ceph_release_page_vector(pages
, num_pages
);
509 if (off
> iocb
->ki_pos
) {
510 ret
= off
- iocb
->ki_pos
;
514 dout("sync_read result %d\n", ret
);
519 * Write commit request unsafe callback, called to tell us when a
520 * request is unsafe (that is, in flight--has been handed to the
521 * messenger to send to its target osd). It is called again when
522 * we've received a response message indicating the request is
523 * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request
524 * is completed early (and unsuccessfully) due to a timeout or
527 * This is used if we requested both an ACK and ONDISK commit reply
530 static void ceph_sync_write_unsafe(struct ceph_osd_request
*req
, bool unsafe
)
532 struct ceph_inode_info
*ci
= ceph_inode(req
->r_inode
);
534 dout("%s %p tid %llu %ssafe\n", __func__
, req
, req
->r_tid
,
537 ceph_get_cap_refs(ci
, CEPH_CAP_FILE_WR
);
538 spin_lock(&ci
->i_unsafe_lock
);
539 list_add_tail(&req
->r_unsafe_item
,
540 &ci
->i_unsafe_writes
);
541 spin_unlock(&ci
->i_unsafe_lock
);
543 spin_lock(&ci
->i_unsafe_lock
);
544 list_del_init(&req
->r_unsafe_item
);
545 spin_unlock(&ci
->i_unsafe_lock
);
546 ceph_put_cap_refs(ci
, CEPH_CAP_FILE_WR
);
552 * Synchronous write, straight from __user pointer or user pages.
554 * If write spans object boundary, just do multiple writes. (For a
555 * correct atomic write, we should e.g. take write locks on all
556 * objects, rollback on failure, etc.)
559 ceph_sync_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
, loff_t pos
)
561 struct file
*file
= iocb
->ki_filp
;
562 struct inode
*inode
= file_inode(file
);
563 struct ceph_inode_info
*ci
= ceph_inode(inode
);
564 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
565 struct ceph_snap_context
*snapc
;
566 struct ceph_vino vino
;
567 struct ceph_osd_request
*req
;
574 struct timespec mtime
= CURRENT_TIME
;
575 size_t count
= iov_iter_count(from
);
577 if (ceph_snap(file_inode(file
)) != CEPH_NOSNAP
)
580 dout("sync_direct_write on file %p %lld~%u\n", file
, pos
,
583 ret
= filemap_write_and_wait_range(inode
->i_mapping
, pos
, pos
+ count
);
587 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
588 pos
>> PAGE_CACHE_SHIFT
,
589 (pos
+ count
) >> PAGE_CACHE_SHIFT
);
591 dout("invalidate_inode_pages2_range returned %d\n", ret
);
593 flags
= CEPH_OSD_FLAG_ORDERSNAP
|
594 CEPH_OSD_FLAG_ONDISK
|
597 while (iov_iter_count(from
) > 0) {
598 u64 len
= iov_iter_single_seg_count(from
);
602 snapc
= ci
->i_snap_realm
->cached_context
;
603 vino
= ceph_vino(inode
);
604 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
606 2,/*include a 'startsync' command*/
607 CEPH_OSD_OP_WRITE
, flags
, snapc
,
616 osd_req_op_init(req
, 1, CEPH_OSD_OP_STARTSYNC
);
618 n
= iov_iter_get_pages_alloc(from
, &pages
, len
, &start
);
619 if (unlikely(n
< 0)) {
621 ceph_osdc_put_request(req
);
625 num_pages
= (n
+ start
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
627 * throw out any page cache pages in this range. this
630 truncate_inode_pages_range(inode
->i_mapping
, pos
,
631 (pos
+n
) | (PAGE_CACHE_SIZE
-1));
632 osd_req_op_extent_osd_data_pages(req
, 0, pages
, n
, start
,
635 /* BUG_ON(vino.snap != CEPH_NOSNAP); */
636 ceph_osdc_build_request(req
, pos
, snapc
, vino
.snap
, &mtime
);
638 ret
= ceph_osdc_start_request(&fsc
->client
->osdc
, req
, false);
640 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
642 ceph_put_page_vector(pages
, num_pages
, false);
644 ceph_osdc_put_request(req
);
649 iov_iter_advance(from
, n
);
651 if (pos
> i_size_read(inode
)) {
652 check_caps
= ceph_inode_set_size(inode
, pos
);
654 ceph_check_caps(ceph_inode(inode
),
660 if (ret
!= -EOLDSNAPC
&& written
> 0) {
669 * Synchronous write, straight from __user pointer or user pages.
671 * If write spans object boundary, just do multiple writes. (For a
672 * correct atomic write, we should e.g. take write locks on all
673 * objects, rollback on failure, etc.)
676 ceph_sync_write(struct kiocb
*iocb
, struct iov_iter
*from
, loff_t pos
)
678 struct file
*file
= iocb
->ki_filp
;
679 struct inode
*inode
= file_inode(file
);
680 struct ceph_inode_info
*ci
= ceph_inode(inode
);
681 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
682 struct ceph_snap_context
*snapc
;
683 struct ceph_vino vino
;
684 struct ceph_osd_request
*req
;
692 struct timespec mtime
= CURRENT_TIME
;
693 size_t count
= iov_iter_count(from
);
695 if (ceph_snap(file_inode(file
)) != CEPH_NOSNAP
)
698 dout("sync_write on file %p %lld~%u\n", file
, pos
, (unsigned)count
);
700 ret
= filemap_write_and_wait_range(inode
->i_mapping
, pos
, pos
+ count
);
704 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
705 pos
>> PAGE_CACHE_SHIFT
,
706 (pos
+ count
) >> PAGE_CACHE_SHIFT
);
708 dout("invalidate_inode_pages2_range returned %d\n", ret
);
710 flags
= CEPH_OSD_FLAG_ORDERSNAP
|
711 CEPH_OSD_FLAG_ONDISK
|
712 CEPH_OSD_FLAG_WRITE
|
715 while ((len
= iov_iter_count(from
)) > 0) {
719 snapc
= ci
->i_snap_realm
->cached_context
;
720 vino
= ceph_vino(inode
);
721 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
722 vino
, pos
, &len
, 0, 1,
723 CEPH_OSD_OP_WRITE
, flags
, snapc
,
733 * write from beginning of first page,
734 * regardless of io alignment
736 num_pages
= (len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
738 pages
= ceph_alloc_page_vector(num_pages
, GFP_NOFS
);
740 ret
= PTR_ERR(pages
);
745 for (n
= 0; n
< num_pages
; n
++) {
746 size_t plen
= min_t(size_t, left
, PAGE_SIZE
);
747 ret
= copy_page_from_iter(pages
[n
], 0, plen
, from
);
756 ceph_release_page_vector(pages
, num_pages
);
760 /* get a second commit callback */
761 req
->r_unsafe_callback
= ceph_sync_write_unsafe
;
762 req
->r_inode
= inode
;
764 osd_req_op_extent_osd_data_pages(req
, 0, pages
, len
, 0,
767 /* BUG_ON(vino.snap != CEPH_NOSNAP); */
768 ceph_osdc_build_request(req
, pos
, snapc
, vino
.snap
, &mtime
);
770 ret
= ceph_osdc_start_request(&fsc
->client
->osdc
, req
, false);
772 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
775 ceph_osdc_put_request(req
);
780 if (pos
> i_size_read(inode
)) {
781 check_caps
= ceph_inode_set_size(inode
, pos
);
783 ceph_check_caps(ceph_inode(inode
),
791 if (ret
!= -EOLDSNAPC
&& written
> 0) {
799 * Wrap generic_file_aio_read with checks for cap bits on the inode.
800 * Atomically grab references, so that those bits are not released
801 * back to the MDS mid-read.
803 * Hmm, the sync read case isn't actually async... should it be?
805 static ssize_t
ceph_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
807 struct file
*filp
= iocb
->ki_filp
;
808 struct ceph_file_info
*fi
= filp
->private_data
;
809 size_t len
= iov_iter_count(to
);
810 struct inode
*inode
= file_inode(filp
);
811 struct ceph_inode_info
*ci
= ceph_inode(inode
);
812 struct page
*pinned_page
= NULL
;
815 int retry_op
= 0, read
= 0;
818 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
819 inode
, ceph_vinop(inode
), iocb
->ki_pos
, (unsigned)len
, inode
);
821 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
822 want
= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_LAZYIO
;
824 want
= CEPH_CAP_FILE_CACHE
;
825 ret
= ceph_get_caps(ci
, CEPH_CAP_FILE_RD
, want
, -1, &got
, &pinned_page
);
829 if ((got
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)) == 0 ||
830 (iocb
->ki_filp
->f_flags
& O_DIRECT
) ||
831 (fi
->flags
& CEPH_F_SYNC
)) {
833 dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n",
834 inode
, ceph_vinop(inode
), iocb
->ki_pos
, (unsigned)len
,
835 ceph_cap_string(got
));
837 if (ci
->i_inline_version
== CEPH_INLINE_NONE
) {
838 /* hmm, this isn't really async... */
839 ret
= ceph_sync_read(iocb
, to
, &retry_op
);
841 retry_op
= READ_INLINE
;
844 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
845 inode
, ceph_vinop(inode
), iocb
->ki_pos
, (unsigned)len
,
846 ceph_cap_string(got
));
848 ret
= generic_file_read_iter(iocb
, to
);
850 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
851 inode
, ceph_vinop(inode
), ceph_cap_string(got
), (int)ret
);
853 page_cache_release(pinned_page
);
856 ceph_put_cap_refs(ci
, got
);
857 if (retry_op
&& ret
>= 0) {
859 struct page
*page
= NULL
;
861 if (retry_op
== READ_INLINE
) {
862 page
= __page_cache_alloc(GFP_NOFS
);
867 statret
= __ceph_do_getattr(inode
, page
,
868 CEPH_STAT_CAP_INLINE_DATA
, !!page
);
871 if (statret
== -ENODATA
) {
872 BUG_ON(retry_op
!= READ_INLINE
);
878 i_size
= i_size_read(inode
);
879 if (retry_op
== READ_INLINE
) {
880 /* does not support inline data > PAGE_SIZE */
881 if (i_size
> PAGE_CACHE_SIZE
) {
883 } else if (iocb
->ki_pos
< i_size
) {
884 loff_t end
= min_t(loff_t
, i_size
,
887 zero_user_segment(page
, statret
, end
);
888 ret
= copy_page_to_iter(page
,
889 iocb
->ki_pos
& ~PAGE_MASK
,
890 end
- iocb
->ki_pos
, to
);
895 __free_pages(page
, 0);
899 /* hit EOF or hole? */
900 if (retry_op
== CHECK_EOF
&& iocb
->ki_pos
< i_size
&&
902 dout("sync_read hit hole, ppos %lld < size %lld"
903 ", reading more\n", iocb
->ki_pos
,
920 * Take cap references to avoid releasing caps to MDS mid-write.
922 * If we are synchronous, and write with an old snap context, the OSD
923 * may return EOLDSNAPC. In that case, retry the write.. _after_
924 * dropping our cap refs and allowing the pending snap to logically
925 * complete _before_ this write occurs.
927 * If we are near ENOSPC, write synchronously.
929 static ssize_t
ceph_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
931 struct file
*file
= iocb
->ki_filp
;
932 struct ceph_file_info
*fi
= file
->private_data
;
933 struct inode
*inode
= file_inode(file
);
934 struct ceph_inode_info
*ci
= ceph_inode(inode
);
935 struct ceph_osd_client
*osdc
=
936 &ceph_sb_to_client(inode
->i_sb
)->client
->osdc
;
937 ssize_t count
= iov_iter_count(from
), written
= 0;
939 loff_t pos
= iocb
->ki_pos
;
941 if (ceph_snap(inode
) != CEPH_NOSNAP
)
944 mutex_lock(&inode
->i_mutex
);
946 /* We can write back this queue in page reclaim */
947 current
->backing_dev_info
= inode_to_bdi(inode
);
949 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
955 iov_iter_truncate(from
, count
);
957 err
= file_remove_suid(file
);
961 err
= file_update_time(file
);
965 if (ci
->i_inline_version
!= CEPH_INLINE_NONE
) {
966 err
= ceph_uninline_data(file
, NULL
);
972 if (ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_FULL
)) {
977 dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
978 inode
, ceph_vinop(inode
), pos
, count
, inode
->i_size
);
979 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
980 want
= CEPH_CAP_FILE_BUFFER
| CEPH_CAP_FILE_LAZYIO
;
982 want
= CEPH_CAP_FILE_BUFFER
;
984 err
= ceph_get_caps(ci
, CEPH_CAP_FILE_WR
, want
, pos
+ count
,
989 dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
990 inode
, ceph_vinop(inode
), pos
, count
, ceph_cap_string(got
));
992 if ((got
& (CEPH_CAP_FILE_BUFFER
|CEPH_CAP_FILE_LAZYIO
)) == 0 ||
993 (file
->f_flags
& O_DIRECT
) || (fi
->flags
& CEPH_F_SYNC
)) {
994 struct iov_iter data
;
995 mutex_unlock(&inode
->i_mutex
);
996 /* we might need to revert back to that point */
998 if (file
->f_flags
& O_DIRECT
)
999 written
= ceph_sync_direct_write(iocb
, &data
, pos
);
1001 written
= ceph_sync_write(iocb
, &data
, pos
);
1002 if (written
== -EOLDSNAPC
) {
1003 dout("aio_write %p %llx.%llx %llu~%u"
1004 "got EOLDSNAPC, retrying\n",
1005 inode
, ceph_vinop(inode
),
1006 pos
, (unsigned)count
);
1007 mutex_lock(&inode
->i_mutex
);
1011 iov_iter_advance(from
, written
);
1013 loff_t old_size
= inode
->i_size
;
1015 * No need to acquire the i_truncate_mutex. Because
1016 * the MDS revokes Fwb caps before sending truncate
1017 * message to us. We can't get Fwb cap while there
1018 * are pending vmtruncate. So write and vmtruncate
1019 * can not run at the same time
1021 written
= generic_perform_write(file
, from
, pos
);
1022 if (likely(written
>= 0))
1023 iocb
->ki_pos
= pos
+ written
;
1024 if (inode
->i_size
> old_size
)
1025 ceph_fscache_update_objectsize(inode
);
1026 mutex_unlock(&inode
->i_mutex
);
1031 spin_lock(&ci
->i_ceph_lock
);
1032 ci
->i_inline_version
= CEPH_INLINE_NONE
;
1033 dirty
= __ceph_mark_dirty_caps(ci
, CEPH_CAP_FILE_WR
);
1034 spin_unlock(&ci
->i_ceph_lock
);
1036 __mark_inode_dirty(inode
, dirty
);
1039 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
1040 inode
, ceph_vinop(inode
), pos
, (unsigned)count
,
1041 ceph_cap_string(got
));
1042 ceph_put_cap_refs(ci
, got
);
1045 ((file
->f_flags
& O_SYNC
) || IS_SYNC(file
->f_mapping
->host
) ||
1046 ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_NEARFULL
))) {
1047 err
= vfs_fsync_range(file
, pos
, pos
+ written
- 1, 1);
1055 mutex_unlock(&inode
->i_mutex
);
1057 current
->backing_dev_info
= NULL
;
1058 return written
? written
: err
;
1062 * llseek. be sure to verify file size on SEEK_END.
1064 static loff_t
ceph_llseek(struct file
*file
, loff_t offset
, int whence
)
1066 struct inode
*inode
= file
->f_mapping
->host
;
1069 mutex_lock(&inode
->i_mutex
);
1071 if (whence
== SEEK_END
|| whence
== SEEK_DATA
|| whence
== SEEK_HOLE
) {
1072 ret
= ceph_do_getattr(inode
, CEPH_STAT_CAP_SIZE
, false);
1081 offset
+= inode
->i_size
;
1085 * Here we special-case the lseek(fd, 0, SEEK_CUR)
1086 * position-querying operation. Avoid rewriting the "same"
1087 * f_pos value back to the file because a concurrent read(),
1088 * write() or lseek() might have altered it
1091 offset
= file
->f_pos
;
1094 offset
+= file
->f_pos
;
1097 if (offset
>= inode
->i_size
) {
1103 if (offset
>= inode
->i_size
) {
1107 offset
= inode
->i_size
;
1111 offset
= vfs_setpos(file
, offset
, inode
->i_sb
->s_maxbytes
);
1114 mutex_unlock(&inode
->i_mutex
);
1118 static inline void ceph_zero_partial_page(
1119 struct inode
*inode
, loff_t offset
, unsigned size
)
1122 pgoff_t index
= offset
>> PAGE_CACHE_SHIFT
;
1124 page
= find_lock_page(inode
->i_mapping
, index
);
1126 wait_on_page_writeback(page
);
1127 zero_user(page
, offset
& (PAGE_CACHE_SIZE
- 1), size
);
1129 page_cache_release(page
);
1133 static void ceph_zero_pagecache_range(struct inode
*inode
, loff_t offset
,
1136 loff_t nearly
= round_up(offset
, PAGE_CACHE_SIZE
);
1137 if (offset
< nearly
) {
1138 loff_t size
= nearly
- offset
;
1141 ceph_zero_partial_page(inode
, offset
, size
);
1145 if (length
>= PAGE_CACHE_SIZE
) {
1146 loff_t size
= round_down(length
, PAGE_CACHE_SIZE
);
1147 truncate_pagecache_range(inode
, offset
, offset
+ size
- 1);
1152 ceph_zero_partial_page(inode
, offset
, length
);
1155 static int ceph_zero_partial_object(struct inode
*inode
,
1156 loff_t offset
, loff_t
*length
)
1158 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1159 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
1160 struct ceph_osd_request
*req
;
1166 op
= offset
? CEPH_OSD_OP_DELETE
: CEPH_OSD_OP_TRUNCATE
;
1169 op
= CEPH_OSD_OP_ZERO
;
1172 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
1176 CEPH_OSD_FLAG_WRITE
|
1177 CEPH_OSD_FLAG_ONDISK
,
1184 ceph_osdc_build_request(req
, offset
, NULL
, ceph_vino(inode
).snap
,
1187 ret
= ceph_osdc_start_request(&fsc
->client
->osdc
, req
, false);
1189 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
1193 ceph_osdc_put_request(req
);
1199 static int ceph_zero_objects(struct inode
*inode
, loff_t offset
, loff_t length
)
1202 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1203 s32 stripe_unit
= ceph_file_layout_su(ci
->i_layout
);
1204 s32 stripe_count
= ceph_file_layout_stripe_count(ci
->i_layout
);
1205 s32 object_size
= ceph_file_layout_object_size(ci
->i_layout
);
1206 u64 object_set_size
= object_size
* stripe_count
;
1209 /* round offset up to next period boundary */
1210 nearly
= offset
+ object_set_size
- 1;
1212 nearly
-= do_div(t
, object_set_size
);
1214 while (length
&& offset
< nearly
) {
1215 loff_t size
= length
;
1216 ret
= ceph_zero_partial_object(inode
, offset
, &size
);
1222 while (length
>= object_set_size
) {
1224 loff_t pos
= offset
;
1225 for (i
= 0; i
< stripe_count
; ++i
) {
1226 ret
= ceph_zero_partial_object(inode
, pos
, NULL
);
1231 offset
+= object_set_size
;
1232 length
-= object_set_size
;
1235 loff_t size
= length
;
1236 ret
= ceph_zero_partial_object(inode
, offset
, &size
);
1245 static long ceph_fallocate(struct file
*file
, int mode
,
1246 loff_t offset
, loff_t length
)
1248 struct ceph_file_info
*fi
= file
->private_data
;
1249 struct inode
*inode
= file_inode(file
);
1250 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1251 struct ceph_osd_client
*osdc
=
1252 &ceph_inode_to_client(inode
)->client
->osdc
;
1259 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
1262 if (!S_ISREG(inode
->i_mode
))
1265 mutex_lock(&inode
->i_mutex
);
1267 if (ceph_snap(inode
) != CEPH_NOSNAP
) {
1272 if (ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_FULL
) &&
1273 !(mode
& FALLOC_FL_PUNCH_HOLE
)) {
1278 if (ci
->i_inline_version
!= CEPH_INLINE_NONE
) {
1279 ret
= ceph_uninline_data(file
, NULL
);
1284 size
= i_size_read(inode
);
1285 if (!(mode
& FALLOC_FL_KEEP_SIZE
))
1286 endoff
= offset
+ length
;
1288 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
1289 want
= CEPH_CAP_FILE_BUFFER
| CEPH_CAP_FILE_LAZYIO
;
1291 want
= CEPH_CAP_FILE_BUFFER
;
1293 ret
= ceph_get_caps(ci
, CEPH_CAP_FILE_WR
, want
, endoff
, &got
, NULL
);
1297 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
1299 ceph_zero_pagecache_range(inode
, offset
, length
);
1300 ret
= ceph_zero_objects(inode
, offset
, length
);
1301 } else if (endoff
> size
) {
1302 truncate_pagecache_range(inode
, size
, -1);
1303 if (ceph_inode_set_size(inode
, endoff
))
1304 ceph_check_caps(ceph_inode(inode
),
1305 CHECK_CAPS_AUTHONLY
, NULL
);
1309 spin_lock(&ci
->i_ceph_lock
);
1310 ci
->i_inline_version
= CEPH_INLINE_NONE
;
1311 dirty
= __ceph_mark_dirty_caps(ci
, CEPH_CAP_FILE_WR
);
1312 spin_unlock(&ci
->i_ceph_lock
);
1314 __mark_inode_dirty(inode
, dirty
);
1317 ceph_put_cap_refs(ci
, got
);
1319 mutex_unlock(&inode
->i_mutex
);
1323 const struct file_operations ceph_file_fops
= {
1325 .release
= ceph_release
,
1326 .llseek
= ceph_llseek
,
1327 .read
= new_sync_read
,
1328 .write
= new_sync_write
,
1329 .read_iter
= ceph_read_iter
,
1330 .write_iter
= ceph_write_iter
,
1332 .fsync
= ceph_fsync
,
1334 .flock
= ceph_flock
,
1335 .splice_read
= generic_file_splice_read
,
1336 .splice_write
= iter_file_splice_write
,
1337 .unlocked_ioctl
= ceph_ioctl
,
1338 .compat_ioctl
= ceph_ioctl
,
1339 .fallocate
= ceph_fallocate
,