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/aio.h>
11 #include <linux/falloc.h>
14 #include "mds_client.h"
18 * Ceph file operations
20 * Implement basic open/close functionality, and implement
23 * We implement three modes of file I/O:
24 * - buffered uses the generic_file_aio_{read,write} helpers
26 * - synchronous is used when there is multi-client read/write
27 * sharing, avoids the page cache, and synchronously waits for an
30 * - direct io takes the variant of the sync path that references
31 * user pages directly.
33 * fsync() flushes and waits on dirty pages, but just queues metadata
34 * for writeback: since the MDS can recover size and mtime there is no
35 * need to wait for MDS acknowledgement.
40 * Prepare an open request. Preallocate ceph_cap to avoid an
41 * inopportune ENOMEM later.
43 static struct ceph_mds_request
*
44 prepare_open_request(struct super_block
*sb
, int flags
, int create_mode
)
46 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
47 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
48 struct ceph_mds_request
*req
;
49 int want_auth
= USE_ANY_MDS
;
50 int op
= (flags
& O_CREAT
) ? CEPH_MDS_OP_CREATE
: CEPH_MDS_OP_OPEN
;
52 if (flags
& (O_WRONLY
|O_RDWR
|O_CREAT
|O_TRUNC
))
53 want_auth
= USE_AUTH_MDS
;
55 req
= ceph_mdsc_create_request(mdsc
, op
, want_auth
);
58 req
->r_fmode
= ceph_flags_to_mode(flags
);
59 req
->r_args
.open
.flags
= cpu_to_le32(flags
);
60 req
->r_args
.open
.mode
= cpu_to_le32(create_mode
);
66 * initialize private struct file data.
67 * if we fail, clean up by dropping fmode reference on the ceph_inode
69 static int ceph_init_file(struct inode
*inode
, struct file
*file
, int fmode
)
71 struct ceph_file_info
*cf
;
73 struct ceph_inode_info
*ci
= ceph_inode(inode
);
74 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
75 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
77 switch (inode
->i_mode
& S_IFMT
) {
79 /* First file open request creates the cookie, we want to keep
80 * this cookie around for the filetime of the inode as not to
81 * have to worry about fscache register / revoke / operation
84 * Also, if we know the operation is going to invalidate data
85 * (non readonly) just nuke the cache right away.
87 ceph_fscache_register_inode_cookie(mdsc
->fsc
, ci
);
88 if ((fmode
& CEPH_FILE_MODE_WR
))
89 ceph_fscache_invalidate(inode
);
91 dout("init_file %p %p 0%o (regular)\n", inode
, file
,
93 cf
= kmem_cache_alloc(ceph_file_cachep
, GFP_NOFS
| __GFP_ZERO
);
95 ceph_put_fmode(ceph_inode(inode
), fmode
); /* clean up */
100 file
->private_data
= cf
;
101 BUG_ON(inode
->i_fop
->release
!= ceph_release
);
105 dout("init_file %p %p 0%o (symlink)\n", inode
, file
,
107 ceph_put_fmode(ceph_inode(inode
), fmode
); /* clean up */
111 dout("init_file %p %p 0%o (special)\n", inode
, file
,
114 * we need to drop the open ref now, since we don't
115 * have .release set to ceph_release.
117 ceph_put_fmode(ceph_inode(inode
), fmode
); /* clean up */
118 BUG_ON(inode
->i_fop
->release
== ceph_release
);
120 /* call the proper open fop */
121 ret
= inode
->i_fop
->open(inode
, file
);
127 * If we already have the requisite capabilities, we can satisfy
128 * the open request locally (no need to request new caps from the
129 * MDS). We do, however, need to inform the MDS (asynchronously)
130 * if our wanted caps set expands.
132 int ceph_open(struct inode
*inode
, struct file
*file
)
134 struct ceph_inode_info
*ci
= ceph_inode(inode
);
135 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
136 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
137 struct ceph_mds_request
*req
;
138 struct ceph_file_info
*cf
= file
->private_data
;
139 struct inode
*parent_inode
= NULL
;
141 int flags
, fmode
, wanted
;
144 dout("open file %p is already opened\n", file
);
148 /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */
149 flags
= file
->f_flags
& ~(O_CREAT
|O_EXCL
);
150 if (S_ISDIR(inode
->i_mode
))
151 flags
= O_DIRECTORY
; /* mds likes to know */
153 dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode
,
154 ceph_vinop(inode
), file
, flags
, file
->f_flags
);
155 fmode
= ceph_flags_to_mode(flags
);
156 wanted
= ceph_caps_for_mode(fmode
);
158 /* snapped files are read-only */
159 if (ceph_snap(inode
) != CEPH_NOSNAP
&& (file
->f_mode
& FMODE_WRITE
))
162 /* trivially open snapdir */
163 if (ceph_snap(inode
) == CEPH_SNAPDIR
) {
164 spin_lock(&ci
->i_ceph_lock
);
165 __ceph_get_fmode(ci
, fmode
);
166 spin_unlock(&ci
->i_ceph_lock
);
167 return ceph_init_file(inode
, file
, fmode
);
171 * No need to block if we have caps on the auth MDS (for
172 * write) or any MDS (for read). Update wanted set
175 spin_lock(&ci
->i_ceph_lock
);
176 if (__ceph_is_any_real_caps(ci
) &&
177 (((fmode
& CEPH_FILE_MODE_WR
) == 0) || ci
->i_auth_cap
)) {
178 int mds_wanted
= __ceph_caps_mds_wanted(ci
);
179 int issued
= __ceph_caps_issued(ci
, NULL
);
181 dout("open %p fmode %d want %s issued %s using existing\n",
182 inode
, fmode
, ceph_cap_string(wanted
),
183 ceph_cap_string(issued
));
184 __ceph_get_fmode(ci
, fmode
);
185 spin_unlock(&ci
->i_ceph_lock
);
188 if ((issued
& wanted
) != wanted
&&
189 (mds_wanted
& wanted
) != wanted
&&
190 ceph_snap(inode
) != CEPH_SNAPDIR
)
191 ceph_check_caps(ci
, 0, NULL
);
193 return ceph_init_file(inode
, file
, fmode
);
194 } else if (ceph_snap(inode
) != CEPH_NOSNAP
&&
195 (ci
->i_snap_caps
& wanted
) == wanted
) {
196 __ceph_get_fmode(ci
, fmode
);
197 spin_unlock(&ci
->i_ceph_lock
);
198 return ceph_init_file(inode
, file
, fmode
);
201 spin_unlock(&ci
->i_ceph_lock
);
203 dout("open fmode %d wants %s\n", fmode
, ceph_cap_string(wanted
));
204 req
= prepare_open_request(inode
->i_sb
, flags
, 0);
209 req
->r_inode
= inode
;
214 parent_inode
= ceph_get_dentry_parent_inode(file
->f_path
.dentry
);
215 err
= ceph_mdsc_do_request(mdsc
, parent_inode
, req
);
218 err
= ceph_init_file(inode
, file
, req
->r_fmode
);
219 ceph_mdsc_put_request(req
);
220 dout("open result=%d on %llx.%llx\n", err
, ceph_vinop(inode
));
227 * Do a lookup + open with a single request. If we get a non-existent
228 * file or symlink, return 1 so the VFS can retry.
230 int ceph_atomic_open(struct inode
*dir
, struct dentry
*dentry
,
231 struct file
*file
, unsigned flags
, umode_t mode
,
234 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
235 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
236 struct ceph_mds_request
*req
;
238 struct ceph_acls_info acls
= {};
241 dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n",
243 d_unhashed(dentry
) ? "unhashed" : "hashed", flags
, mode
);
245 if (dentry
->d_name
.len
> NAME_MAX
)
246 return -ENAMETOOLONG
;
248 err
= ceph_init_dentry(dentry
);
252 if (flags
& O_CREAT
) {
253 err
= ceph_pre_init_acls(dir
, &mode
, &acls
);
259 req
= prepare_open_request(dir
->i_sb
, flags
, mode
);
264 req
->r_dentry
= dget(dentry
);
266 if (flags
& O_CREAT
) {
267 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
268 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
270 req
->r_pagelist
= acls
.pagelist
;
271 acls
.pagelist
= NULL
;
274 req
->r_locked_dir
= dir
; /* caller holds dir->i_mutex */
275 err
= ceph_mdsc_do_request(mdsc
,
276 (flags
& (O_CREAT
|O_TRUNC
)) ? dir
: NULL
,
278 err
= ceph_handle_snapdir(req
, dentry
, err
);
282 if (err
== 0 && (flags
& O_CREAT
) && !req
->r_reply_info
.head
->is_dentry
)
283 err
= ceph_handle_notrace_create(dir
, dentry
);
285 if (d_unhashed(dentry
)) {
286 dn
= ceph_finish_lookup(req
, dentry
, err
);
290 /* we were given a hashed negative dentry */
295 if (dn
|| dentry
->d_inode
== NULL
|| d_is_symlink(dentry
)) {
296 /* make vfs retry on splice, ENOENT, or symlink */
297 dout("atomic_open finish_no_open on dn %p\n", dn
);
298 err
= finish_no_open(file
, dn
);
300 dout("atomic_open finish_open on dn %p\n", dn
);
301 if (req
->r_op
== CEPH_MDS_OP_CREATE
&& req
->r_reply_info
.has_create_ino
) {
302 ceph_init_inode_acls(dentry
->d_inode
, &acls
);
303 *opened
|= FILE_CREATED
;
305 err
= finish_open(file
, dentry
, ceph_open
, opened
);
308 if (!req
->r_err
&& req
->r_target_inode
)
309 ceph_put_fmode(ceph_inode(req
->r_target_inode
), req
->r_fmode
);
310 ceph_mdsc_put_request(req
);
312 ceph_release_acls_info(&acls
);
313 dout("atomic_open result=%d\n", err
);
317 int ceph_release(struct inode
*inode
, struct file
*file
)
319 struct ceph_inode_info
*ci
= ceph_inode(inode
);
320 struct ceph_file_info
*cf
= file
->private_data
;
322 dout("release inode %p file %p\n", inode
, file
);
323 ceph_put_fmode(ci
, cf
->fmode
);
324 if (cf
->last_readdir
)
325 ceph_mdsc_put_request(cf
->last_readdir
);
326 kfree(cf
->last_name
);
329 kmem_cache_free(ceph_file_cachep
, cf
);
331 /* wake up anyone waiting for caps on this inode */
332 wake_up_all(&ci
->i_cap_wq
);
342 * Read a range of bytes striped over one or more objects. Iterate over
343 * objects we stripe over. (That's not atomic, but good enough for now.)
345 * If we get a short result from the OSD, check against i_size; we need to
346 * only return a short read to the caller if we hit EOF.
348 static int striped_read(struct inode
*inode
,
350 struct page
**pages
, int num_pages
,
351 int *checkeof
, bool o_direct
,
352 unsigned long buf_align
)
354 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
355 struct ceph_inode_info
*ci
= ceph_inode(inode
);
356 u64 pos
, this_len
, left
;
357 int io_align
, page_align
;
360 struct page
**page_pos
;
362 bool hit_stripe
, was_short
;
365 * we may need to do multiple reads. not atomic, unfortunately.
370 pages_left
= num_pages
;
372 io_align
= off
& ~PAGE_MASK
;
376 page_align
= (pos
- io_align
+ buf_align
) & ~PAGE_MASK
;
378 page_align
= pos
& ~PAGE_MASK
;
380 ret
= ceph_osdc_readpages(&fsc
->client
->osdc
, ceph_vino(inode
),
381 &ci
->i_layout
, pos
, &this_len
,
384 page_pos
, pages_left
, page_align
);
387 hit_stripe
= this_len
< left
;
388 was_short
= ret
>= 0 && ret
< this_len
;
389 dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos
, left
, read
,
390 ret
, hit_stripe
? " HITSTRIPE" : "", was_short
? " SHORT" : "");
394 if (was_short
&& (pos
+ ret
< inode
->i_size
)) {
395 int zlen
= min(this_len
- ret
,
396 inode
->i_size
- pos
- ret
);
397 int zoff
= (o_direct
? buf_align
: io_align
) +
399 dout(" zero gap %llu to %llu\n",
400 pos
+ ret
, pos
+ ret
+ zlen
);
401 ceph_zero_page_vector_range(zoff
, zlen
, pages
);
405 didpages
= (page_align
+ ret
) >> PAGE_CACHE_SHIFT
;
409 page_pos
+= didpages
;
410 pages_left
-= didpages
;
412 /* hit stripe and need continue*/
413 if (left
&& hit_stripe
&& pos
< inode
->i_size
)
419 /* did we bounce off eof? */
420 if (pos
+ left
> inode
->i_size
)
421 *checkeof
= CHECK_EOF
;
424 dout("striped_read returns %d\n", ret
);
429 * Completely synchronous read and write methods. Direct from __user
430 * buffer to osd, or directly to user pages (if O_DIRECT).
432 * If the read spans object boundary, just do multiple reads.
434 static ssize_t
ceph_sync_read(struct kiocb
*iocb
, struct iov_iter
*i
,
437 struct file
*file
= iocb
->ki_filp
;
438 struct inode
*inode
= file_inode(file
);
440 u64 off
= iocb
->ki_pos
;
442 size_t len
= iov_iter_count(i
);
444 dout("sync_read on file %p %llu~%u %s\n", file
, off
,
446 (file
->f_flags
& O_DIRECT
) ? "O_DIRECT" : "");
451 * flush any page cache pages in this range. this
452 * will make concurrent normal and sync io slow,
453 * but it will at least behave sensibly when they are
456 ret
= filemap_write_and_wait_range(inode
->i_mapping
, off
,
461 if (file
->f_flags
& O_DIRECT
) {
462 while (iov_iter_count(i
)) {
466 n
= iov_iter_get_pages_alloc(i
, &pages
, INT_MAX
, &start
);
470 num_pages
= (n
+ start
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
472 ret
= striped_read(inode
, off
, n
,
473 pages
, num_pages
, checkeof
,
476 ceph_put_page_vector(pages
, num_pages
, true);
481 iov_iter_advance(i
, ret
);
486 num_pages
= calc_pages_for(off
, len
);
487 pages
= ceph_alloc_page_vector(num_pages
, GFP_NOFS
);
489 return PTR_ERR(pages
);
490 ret
= striped_read(inode
, off
, len
, pages
,
491 num_pages
, checkeof
, 0, 0);
497 size_t page_off
= off
& ~PAGE_MASK
;
498 size_t copy
= min_t(size_t,
499 PAGE_SIZE
- page_off
, left
);
500 l
= copy_page_to_iter(pages
[k
++], page_off
,
508 ceph_release_page_vector(pages
, num_pages
);
511 if (off
> iocb
->ki_pos
) {
512 ret
= off
- iocb
->ki_pos
;
516 dout("sync_read result %d\n", ret
);
521 * Write commit request unsafe callback, called to tell us when a
522 * request is unsafe (that is, in flight--has been handed to the
523 * messenger to send to its target osd). It is called again when
524 * we've received a response message indicating the request is
525 * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request
526 * is completed early (and unsuccessfully) due to a timeout or
529 * This is used if we requested both an ACK and ONDISK commit reply
532 static void ceph_sync_write_unsafe(struct ceph_osd_request
*req
, bool unsafe
)
534 struct ceph_inode_info
*ci
= ceph_inode(req
->r_inode
);
536 dout("%s %p tid %llu %ssafe\n", __func__
, req
, req
->r_tid
,
539 ceph_get_cap_refs(ci
, CEPH_CAP_FILE_WR
);
540 spin_lock(&ci
->i_unsafe_lock
);
541 list_add_tail(&req
->r_unsafe_item
,
542 &ci
->i_unsafe_writes
);
543 spin_unlock(&ci
->i_unsafe_lock
);
545 spin_lock(&ci
->i_unsafe_lock
);
546 list_del_init(&req
->r_unsafe_item
);
547 spin_unlock(&ci
->i_unsafe_lock
);
548 ceph_put_cap_refs(ci
, CEPH_CAP_FILE_WR
);
554 * Synchronous write, straight from __user pointer or user pages.
556 * If write spans object boundary, just do multiple writes. (For a
557 * correct atomic write, we should e.g. take write locks on all
558 * objects, rollback on failure, etc.)
561 ceph_sync_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
, loff_t pos
)
563 struct file
*file
= iocb
->ki_filp
;
564 struct inode
*inode
= file_inode(file
);
565 struct ceph_inode_info
*ci
= ceph_inode(inode
);
566 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
567 struct ceph_snap_context
*snapc
;
568 struct ceph_vino vino
;
569 struct ceph_osd_request
*req
;
576 struct timespec mtime
= CURRENT_TIME
;
577 size_t count
= iov_iter_count(from
);
579 if (ceph_snap(file_inode(file
)) != CEPH_NOSNAP
)
582 dout("sync_direct_write on file %p %lld~%u\n", file
, pos
,
585 ret
= filemap_write_and_wait_range(inode
->i_mapping
, pos
, pos
+ count
);
589 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
590 pos
>> PAGE_CACHE_SHIFT
,
591 (pos
+ count
) >> PAGE_CACHE_SHIFT
);
593 dout("invalidate_inode_pages2_range returned %d\n", ret
);
595 flags
= CEPH_OSD_FLAG_ORDERSNAP
|
596 CEPH_OSD_FLAG_ONDISK
|
599 while (iov_iter_count(from
) > 0) {
600 u64 len
= iov_iter_single_seg_count(from
);
604 snapc
= ci
->i_snap_realm
->cached_context
;
605 vino
= ceph_vino(inode
);
606 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
608 2,/*include a 'startsync' command*/
609 CEPH_OSD_OP_WRITE
, flags
, snapc
,
618 osd_req_op_init(req
, 1, CEPH_OSD_OP_STARTSYNC
);
620 n
= iov_iter_get_pages_alloc(from
, &pages
, len
, &start
);
621 if (unlikely(n
< 0)) {
623 ceph_osdc_put_request(req
);
627 num_pages
= (n
+ start
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
629 * throw out any page cache pages in this range. this
632 truncate_inode_pages_range(inode
->i_mapping
, pos
,
633 (pos
+n
) | (PAGE_CACHE_SIZE
-1));
634 osd_req_op_extent_osd_data_pages(req
, 0, pages
, n
, start
,
637 /* BUG_ON(vino.snap != CEPH_NOSNAP); */
638 ceph_osdc_build_request(req
, pos
, snapc
, vino
.snap
, &mtime
);
640 ret
= ceph_osdc_start_request(&fsc
->client
->osdc
, req
, false);
642 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
644 ceph_put_page_vector(pages
, num_pages
, false);
646 ceph_osdc_put_request(req
);
651 iov_iter_advance(from
, n
);
653 if (pos
> i_size_read(inode
)) {
654 check_caps
= ceph_inode_set_size(inode
, pos
);
656 ceph_check_caps(ceph_inode(inode
),
662 if (ret
!= -EOLDSNAPC
&& written
> 0) {
671 * Synchronous write, straight from __user pointer or user pages.
673 * If write spans object boundary, just do multiple writes. (For a
674 * correct atomic write, we should e.g. take write locks on all
675 * objects, rollback on failure, etc.)
678 ceph_sync_write(struct kiocb
*iocb
, struct iov_iter
*from
, loff_t pos
)
680 struct file
*file
= iocb
->ki_filp
;
681 struct inode
*inode
= file_inode(file
);
682 struct ceph_inode_info
*ci
= ceph_inode(inode
);
683 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
684 struct ceph_snap_context
*snapc
;
685 struct ceph_vino vino
;
686 struct ceph_osd_request
*req
;
694 struct timespec mtime
= CURRENT_TIME
;
695 size_t count
= iov_iter_count(from
);
697 if (ceph_snap(file_inode(file
)) != CEPH_NOSNAP
)
700 dout("sync_write on file %p %lld~%u\n", file
, pos
, (unsigned)count
);
702 ret
= filemap_write_and_wait_range(inode
->i_mapping
, pos
, pos
+ count
);
706 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
707 pos
>> PAGE_CACHE_SHIFT
,
708 (pos
+ count
) >> PAGE_CACHE_SHIFT
);
710 dout("invalidate_inode_pages2_range returned %d\n", ret
);
712 flags
= CEPH_OSD_FLAG_ORDERSNAP
|
713 CEPH_OSD_FLAG_ONDISK
|
714 CEPH_OSD_FLAG_WRITE
|
717 while ((len
= iov_iter_count(from
)) > 0) {
721 snapc
= ci
->i_snap_realm
->cached_context
;
722 vino
= ceph_vino(inode
);
723 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
724 vino
, pos
, &len
, 0, 1,
725 CEPH_OSD_OP_WRITE
, flags
, snapc
,
735 * write from beginning of first page,
736 * regardless of io alignment
738 num_pages
= (len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
740 pages
= ceph_alloc_page_vector(num_pages
, GFP_NOFS
);
742 ret
= PTR_ERR(pages
);
747 for (n
= 0; n
< num_pages
; n
++) {
748 size_t plen
= min_t(size_t, left
, PAGE_SIZE
);
749 ret
= copy_page_from_iter(pages
[n
], 0, plen
, from
);
758 ceph_release_page_vector(pages
, num_pages
);
762 /* get a second commit callback */
763 req
->r_unsafe_callback
= ceph_sync_write_unsafe
;
764 req
->r_inode
= inode
;
766 osd_req_op_extent_osd_data_pages(req
, 0, pages
, len
, 0,
769 /* BUG_ON(vino.snap != CEPH_NOSNAP); */
770 ceph_osdc_build_request(req
, pos
, snapc
, vino
.snap
, &mtime
);
772 ret
= ceph_osdc_start_request(&fsc
->client
->osdc
, req
, false);
774 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
777 ceph_osdc_put_request(req
);
782 if (pos
> i_size_read(inode
)) {
783 check_caps
= ceph_inode_set_size(inode
, pos
);
785 ceph_check_caps(ceph_inode(inode
),
793 if (ret
!= -EOLDSNAPC
&& written
> 0) {
801 * Wrap generic_file_aio_read with checks for cap bits on the inode.
802 * Atomically grab references, so that those bits are not released
803 * back to the MDS mid-read.
805 * Hmm, the sync read case isn't actually async... should it be?
807 static ssize_t
ceph_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
809 struct file
*filp
= iocb
->ki_filp
;
810 struct ceph_file_info
*fi
= filp
->private_data
;
811 size_t len
= iocb
->ki_nbytes
;
812 struct inode
*inode
= file_inode(filp
);
813 struct ceph_inode_info
*ci
= ceph_inode(inode
);
814 struct page
*pinned_page
= NULL
;
817 int retry_op
= 0, read
= 0;
820 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
821 inode
, ceph_vinop(inode
), iocb
->ki_pos
, (unsigned)len
, inode
);
823 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
824 want
= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_LAZYIO
;
826 want
= CEPH_CAP_FILE_CACHE
;
827 ret
= ceph_get_caps(ci
, CEPH_CAP_FILE_RD
, want
, -1, &got
, &pinned_page
);
831 if ((got
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)) == 0 ||
832 (iocb
->ki_filp
->f_flags
& O_DIRECT
) ||
833 (fi
->flags
& CEPH_F_SYNC
)) {
835 dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n",
836 inode
, ceph_vinop(inode
), iocb
->ki_pos
, (unsigned)len
,
837 ceph_cap_string(got
));
839 if (ci
->i_inline_version
== CEPH_INLINE_NONE
) {
840 /* hmm, this isn't really async... */
841 ret
= ceph_sync_read(iocb
, to
, &retry_op
);
843 retry_op
= READ_INLINE
;
846 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
847 inode
, ceph_vinop(inode
), iocb
->ki_pos
, (unsigned)len
,
848 ceph_cap_string(got
));
850 ret
= generic_file_read_iter(iocb
, to
);
852 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
853 inode
, ceph_vinop(inode
), ceph_cap_string(got
), (int)ret
);
855 page_cache_release(pinned_page
);
858 ceph_put_cap_refs(ci
, got
);
859 if (retry_op
&& ret
>= 0) {
861 struct page
*page
= NULL
;
863 if (retry_op
== READ_INLINE
) {
864 page
= __page_cache_alloc(GFP_NOFS
);
869 statret
= __ceph_do_getattr(inode
, page
,
870 CEPH_STAT_CAP_INLINE_DATA
, !!page
);
873 if (statret
== -ENODATA
) {
874 BUG_ON(retry_op
!= READ_INLINE
);
880 i_size
= i_size_read(inode
);
881 if (retry_op
== READ_INLINE
) {
882 BUG_ON(ret
> 0 || read
> 0);
883 if (iocb
->ki_pos
< i_size
&&
884 iocb
->ki_pos
< PAGE_CACHE_SIZE
) {
885 loff_t end
= min_t(loff_t
, i_size
,
887 end
= min_t(loff_t
, end
, PAGE_CACHE_SIZE
);
889 zero_user_segment(page
, statret
, end
);
890 ret
= copy_page_to_iter(page
,
891 iocb
->ki_pos
& ~PAGE_MASK
,
892 end
- iocb
->ki_pos
, to
);
896 if (iocb
->ki_pos
< i_size
&& read
< len
) {
897 size_t zlen
= min_t(size_t, len
- read
,
898 i_size
- iocb
->ki_pos
);
899 ret
= iov_iter_zero(zlen
, to
);
903 __free_pages(page
, 0);
907 /* hit EOF or hole? */
908 if (retry_op
== CHECK_EOF
&& iocb
->ki_pos
< i_size
&&
910 dout("sync_read hit hole, ppos %lld < size %lld"
911 ", reading more\n", iocb
->ki_pos
,
928 * Take cap references to avoid releasing caps to MDS mid-write.
930 * If we are synchronous, and write with an old snap context, the OSD
931 * may return EOLDSNAPC. In that case, retry the write.. _after_
932 * dropping our cap refs and allowing the pending snap to logically
933 * complete _before_ this write occurs.
935 * If we are near ENOSPC, write synchronously.
937 static ssize_t
ceph_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
939 struct file
*file
= iocb
->ki_filp
;
940 struct ceph_file_info
*fi
= file
->private_data
;
941 struct inode
*inode
= file_inode(file
);
942 struct ceph_inode_info
*ci
= ceph_inode(inode
);
943 struct ceph_osd_client
*osdc
=
944 &ceph_sb_to_client(inode
->i_sb
)->client
->osdc
;
945 ssize_t count
= iov_iter_count(from
), written
= 0;
947 loff_t pos
= iocb
->ki_pos
;
949 if (ceph_snap(inode
) != CEPH_NOSNAP
)
952 mutex_lock(&inode
->i_mutex
);
954 /* We can write back this queue in page reclaim */
955 current
->backing_dev_info
= inode_to_bdi(inode
);
957 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
963 iov_iter_truncate(from
, count
);
965 err
= file_remove_suid(file
);
969 err
= file_update_time(file
);
973 if (ci
->i_inline_version
!= CEPH_INLINE_NONE
) {
974 err
= ceph_uninline_data(file
, NULL
);
980 if (ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_FULL
)) {
985 dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
986 inode
, ceph_vinop(inode
), pos
, count
, inode
->i_size
);
987 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
988 want
= CEPH_CAP_FILE_BUFFER
| CEPH_CAP_FILE_LAZYIO
;
990 want
= CEPH_CAP_FILE_BUFFER
;
992 err
= ceph_get_caps(ci
, CEPH_CAP_FILE_WR
, want
, pos
+ count
,
997 dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
998 inode
, ceph_vinop(inode
), pos
, count
, ceph_cap_string(got
));
1000 if ((got
& (CEPH_CAP_FILE_BUFFER
|CEPH_CAP_FILE_LAZYIO
)) == 0 ||
1001 (file
->f_flags
& O_DIRECT
) || (fi
->flags
& CEPH_F_SYNC
)) {
1002 struct iov_iter data
;
1003 mutex_unlock(&inode
->i_mutex
);
1004 /* we might need to revert back to that point */
1006 if (file
->f_flags
& O_DIRECT
)
1007 written
= ceph_sync_direct_write(iocb
, &data
, pos
);
1009 written
= ceph_sync_write(iocb
, &data
, pos
);
1010 if (written
== -EOLDSNAPC
) {
1011 dout("aio_write %p %llx.%llx %llu~%u"
1012 "got EOLDSNAPC, retrying\n",
1013 inode
, ceph_vinop(inode
),
1014 pos
, (unsigned)count
);
1015 mutex_lock(&inode
->i_mutex
);
1019 iov_iter_advance(from
, written
);
1021 loff_t old_size
= inode
->i_size
;
1023 * No need to acquire the i_truncate_mutex. Because
1024 * the MDS revokes Fwb caps before sending truncate
1025 * message to us. We can't get Fwb cap while there
1026 * are pending vmtruncate. So write and vmtruncate
1027 * can not run at the same time
1029 written
= generic_perform_write(file
, from
, pos
);
1030 if (likely(written
>= 0))
1031 iocb
->ki_pos
= pos
+ written
;
1032 if (inode
->i_size
> old_size
)
1033 ceph_fscache_update_objectsize(inode
);
1034 mutex_unlock(&inode
->i_mutex
);
1039 spin_lock(&ci
->i_ceph_lock
);
1040 ci
->i_inline_version
= CEPH_INLINE_NONE
;
1041 dirty
= __ceph_mark_dirty_caps(ci
, CEPH_CAP_FILE_WR
);
1042 spin_unlock(&ci
->i_ceph_lock
);
1044 __mark_inode_dirty(inode
, dirty
);
1047 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
1048 inode
, ceph_vinop(inode
), pos
, (unsigned)count
,
1049 ceph_cap_string(got
));
1050 ceph_put_cap_refs(ci
, got
);
1053 ((file
->f_flags
& O_SYNC
) || IS_SYNC(file
->f_mapping
->host
) ||
1054 ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_NEARFULL
))) {
1055 err
= vfs_fsync_range(file
, pos
, pos
+ written
- 1, 1);
1063 mutex_unlock(&inode
->i_mutex
);
1065 current
->backing_dev_info
= NULL
;
1066 return written
? written
: err
;
1070 * llseek. be sure to verify file size on SEEK_END.
1072 static loff_t
ceph_llseek(struct file
*file
, loff_t offset
, int whence
)
1074 struct inode
*inode
= file
->f_mapping
->host
;
1077 mutex_lock(&inode
->i_mutex
);
1079 if (whence
== SEEK_END
|| whence
== SEEK_DATA
|| whence
== SEEK_HOLE
) {
1080 ret
= ceph_do_getattr(inode
, CEPH_STAT_CAP_SIZE
, false);
1089 offset
+= inode
->i_size
;
1093 * Here we special-case the lseek(fd, 0, SEEK_CUR)
1094 * position-querying operation. Avoid rewriting the "same"
1095 * f_pos value back to the file because a concurrent read(),
1096 * write() or lseek() might have altered it
1099 offset
= file
->f_pos
;
1102 offset
+= file
->f_pos
;
1105 if (offset
>= inode
->i_size
) {
1111 if (offset
>= inode
->i_size
) {
1115 offset
= inode
->i_size
;
1119 offset
= vfs_setpos(file
, offset
, inode
->i_sb
->s_maxbytes
);
1122 mutex_unlock(&inode
->i_mutex
);
1126 static inline void ceph_zero_partial_page(
1127 struct inode
*inode
, loff_t offset
, unsigned size
)
1130 pgoff_t index
= offset
>> PAGE_CACHE_SHIFT
;
1132 page
= find_lock_page(inode
->i_mapping
, index
);
1134 wait_on_page_writeback(page
);
1135 zero_user(page
, offset
& (PAGE_CACHE_SIZE
- 1), size
);
1137 page_cache_release(page
);
1141 static void ceph_zero_pagecache_range(struct inode
*inode
, loff_t offset
,
1144 loff_t nearly
= round_up(offset
, PAGE_CACHE_SIZE
);
1145 if (offset
< nearly
) {
1146 loff_t size
= nearly
- offset
;
1149 ceph_zero_partial_page(inode
, offset
, size
);
1153 if (length
>= PAGE_CACHE_SIZE
) {
1154 loff_t size
= round_down(length
, PAGE_CACHE_SIZE
);
1155 truncate_pagecache_range(inode
, offset
, offset
+ size
- 1);
1160 ceph_zero_partial_page(inode
, offset
, length
);
1163 static int ceph_zero_partial_object(struct inode
*inode
,
1164 loff_t offset
, loff_t
*length
)
1166 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1167 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
1168 struct ceph_osd_request
*req
;
1174 op
= offset
? CEPH_OSD_OP_DELETE
: CEPH_OSD_OP_TRUNCATE
;
1177 op
= CEPH_OSD_OP_ZERO
;
1180 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
1184 CEPH_OSD_FLAG_WRITE
|
1185 CEPH_OSD_FLAG_ONDISK
,
1192 ceph_osdc_build_request(req
, offset
, NULL
, ceph_vino(inode
).snap
,
1195 ret
= ceph_osdc_start_request(&fsc
->client
->osdc
, req
, false);
1197 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
1201 ceph_osdc_put_request(req
);
1207 static int ceph_zero_objects(struct inode
*inode
, loff_t offset
, loff_t length
)
1210 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1211 s32 stripe_unit
= ceph_file_layout_su(ci
->i_layout
);
1212 s32 stripe_count
= ceph_file_layout_stripe_count(ci
->i_layout
);
1213 s32 object_size
= ceph_file_layout_object_size(ci
->i_layout
);
1214 u64 object_set_size
= object_size
* stripe_count
;
1217 /* round offset up to next period boundary */
1218 nearly
= offset
+ object_set_size
- 1;
1220 nearly
-= do_div(t
, object_set_size
);
1222 while (length
&& offset
< nearly
) {
1223 loff_t size
= length
;
1224 ret
= ceph_zero_partial_object(inode
, offset
, &size
);
1230 while (length
>= object_set_size
) {
1232 loff_t pos
= offset
;
1233 for (i
= 0; i
< stripe_count
; ++i
) {
1234 ret
= ceph_zero_partial_object(inode
, pos
, NULL
);
1239 offset
+= object_set_size
;
1240 length
-= object_set_size
;
1243 loff_t size
= length
;
1244 ret
= ceph_zero_partial_object(inode
, offset
, &size
);
1253 static long ceph_fallocate(struct file
*file
, int mode
,
1254 loff_t offset
, loff_t length
)
1256 struct ceph_file_info
*fi
= file
->private_data
;
1257 struct inode
*inode
= file_inode(file
);
1258 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1259 struct ceph_osd_client
*osdc
=
1260 &ceph_inode_to_client(inode
)->client
->osdc
;
1267 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
1270 if (!S_ISREG(inode
->i_mode
))
1273 mutex_lock(&inode
->i_mutex
);
1275 if (ceph_snap(inode
) != CEPH_NOSNAP
) {
1280 if (ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_FULL
) &&
1281 !(mode
& FALLOC_FL_PUNCH_HOLE
)) {
1286 if (ci
->i_inline_version
!= CEPH_INLINE_NONE
) {
1287 ret
= ceph_uninline_data(file
, NULL
);
1292 size
= i_size_read(inode
);
1293 if (!(mode
& FALLOC_FL_KEEP_SIZE
))
1294 endoff
= offset
+ length
;
1296 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
1297 want
= CEPH_CAP_FILE_BUFFER
| CEPH_CAP_FILE_LAZYIO
;
1299 want
= CEPH_CAP_FILE_BUFFER
;
1301 ret
= ceph_get_caps(ci
, CEPH_CAP_FILE_WR
, want
, endoff
, &got
, NULL
);
1305 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
1307 ceph_zero_pagecache_range(inode
, offset
, length
);
1308 ret
= ceph_zero_objects(inode
, offset
, length
);
1309 } else if (endoff
> size
) {
1310 truncate_pagecache_range(inode
, size
, -1);
1311 if (ceph_inode_set_size(inode
, endoff
))
1312 ceph_check_caps(ceph_inode(inode
),
1313 CHECK_CAPS_AUTHONLY
, NULL
);
1317 spin_lock(&ci
->i_ceph_lock
);
1318 ci
->i_inline_version
= CEPH_INLINE_NONE
;
1319 dirty
= __ceph_mark_dirty_caps(ci
, CEPH_CAP_FILE_WR
);
1320 spin_unlock(&ci
->i_ceph_lock
);
1322 __mark_inode_dirty(inode
, dirty
);
1325 ceph_put_cap_refs(ci
, got
);
1327 mutex_unlock(&inode
->i_mutex
);
1331 const struct file_operations ceph_file_fops
= {
1333 .release
= ceph_release
,
1334 .llseek
= ceph_llseek
,
1335 .read
= new_sync_read
,
1336 .write
= new_sync_write
,
1337 .read_iter
= ceph_read_iter
,
1338 .write_iter
= ceph_write_iter
,
1340 .fsync
= ceph_fsync
,
1342 .flock
= ceph_flock
,
1343 .splice_read
= generic_file_splice_read
,
1344 .splice_write
= iter_file_splice_write
,
1345 .unlocked_ioctl
= ceph_ioctl
,
1346 .compat_ioctl
= ceph_ioctl
,
1347 .fallocate
= ceph_fallocate
,