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
,
281 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
|| S_ISLNK(dentry
->d_inode
->i_mode
)) {
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 u64 tmp
= min(this_len
- ret
,
396 inode
->i_size
- pos
- ret
);
397 dout(" zero gap %llu to %llu\n",
398 pos
+ ret
, pos
+ ret
+ tmp
);
399 ceph_zero_page_vector_range(page_align
+ read
+ ret
,
404 didpages
= (page_align
+ ret
) >> PAGE_CACHE_SHIFT
;
408 page_pos
+= didpages
;
409 pages_left
-= didpages
;
411 /* hit stripe and need continue*/
412 if (left
&& hit_stripe
&& pos
< inode
->i_size
)
418 /* did we bounce off eof? */
419 if (pos
+ left
> inode
->i_size
)
420 *checkeof
= CHECK_EOF
;
423 dout("striped_read returns %d\n", ret
);
428 * Completely synchronous read and write methods. Direct from __user
429 * buffer to osd, or directly to user pages (if O_DIRECT).
431 * If the read spans object boundary, just do multiple reads.
433 static ssize_t
ceph_sync_read(struct kiocb
*iocb
, struct iov_iter
*i
,
436 struct file
*file
= iocb
->ki_filp
;
437 struct inode
*inode
= file_inode(file
);
439 u64 off
= iocb
->ki_pos
;
441 size_t len
= iov_iter_count(i
);
443 dout("sync_read on file %p %llu~%u %s\n", file
, off
,
445 (file
->f_flags
& O_DIRECT
) ? "O_DIRECT" : "");
450 * flush any page cache pages in this range. this
451 * will make concurrent normal and sync io slow,
452 * but it will at least behave sensibly when they are
455 ret
= filemap_write_and_wait_range(inode
->i_mapping
, off
,
460 if (file
->f_flags
& O_DIRECT
) {
461 while (iov_iter_count(i
)) {
465 n
= iov_iter_get_pages_alloc(i
, &pages
, INT_MAX
, &start
);
469 num_pages
= (n
+ start
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
471 ret
= striped_read(inode
, off
, n
,
472 pages
, num_pages
, checkeof
,
475 ceph_put_page_vector(pages
, num_pages
, true);
480 iov_iter_advance(i
, ret
);
485 num_pages
= calc_pages_for(off
, len
);
486 pages
= ceph_alloc_page_vector(num_pages
, GFP_NOFS
);
488 return PTR_ERR(pages
);
489 ret
= striped_read(inode
, off
, len
, pages
,
490 num_pages
, checkeof
, 0, 0);
496 size_t page_off
= off
& ~PAGE_MASK
;
497 size_t copy
= min_t(size_t,
498 PAGE_SIZE
- page_off
, left
);
499 l
= copy_page_to_iter(pages
[k
++], page_off
,
507 ceph_release_page_vector(pages
, num_pages
);
510 if (off
> iocb
->ki_pos
) {
511 ret
= off
- iocb
->ki_pos
;
515 dout("sync_read result %d\n", ret
);
520 * Write commit request unsafe callback, called to tell us when a
521 * request is unsafe (that is, in flight--has been handed to the
522 * messenger to send to its target osd). It is called again when
523 * we've received a response message indicating the request is
524 * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request
525 * is completed early (and unsuccessfully) due to a timeout or
528 * This is used if we requested both an ACK and ONDISK commit reply
531 static void ceph_sync_write_unsafe(struct ceph_osd_request
*req
, bool unsafe
)
533 struct ceph_inode_info
*ci
= ceph_inode(req
->r_inode
);
535 dout("%s %p tid %llu %ssafe\n", __func__
, req
, req
->r_tid
,
538 ceph_get_cap_refs(ci
, CEPH_CAP_FILE_WR
);
539 spin_lock(&ci
->i_unsafe_lock
);
540 list_add_tail(&req
->r_unsafe_item
,
541 &ci
->i_unsafe_writes
);
542 spin_unlock(&ci
->i_unsafe_lock
);
544 spin_lock(&ci
->i_unsafe_lock
);
545 list_del_init(&req
->r_unsafe_item
);
546 spin_unlock(&ci
->i_unsafe_lock
);
547 ceph_put_cap_refs(ci
, CEPH_CAP_FILE_WR
);
553 * Synchronous write, straight from __user pointer or user pages.
555 * If write spans object boundary, just do multiple writes. (For a
556 * correct atomic write, we should e.g. take write locks on all
557 * objects, rollback on failure, etc.)
560 ceph_sync_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
, loff_t pos
)
562 struct file
*file
= iocb
->ki_filp
;
563 struct inode
*inode
= file_inode(file
);
564 struct ceph_inode_info
*ci
= ceph_inode(inode
);
565 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
566 struct ceph_snap_context
*snapc
;
567 struct ceph_vino vino
;
568 struct ceph_osd_request
*req
;
575 struct timespec mtime
= CURRENT_TIME
;
576 size_t count
= iov_iter_count(from
);
578 if (ceph_snap(file_inode(file
)) != CEPH_NOSNAP
)
581 dout("sync_direct_write on file %p %lld~%u\n", file
, pos
,
584 ret
= filemap_write_and_wait_range(inode
->i_mapping
, pos
, pos
+ count
);
588 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
589 pos
>> PAGE_CACHE_SHIFT
,
590 (pos
+ count
) >> PAGE_CACHE_SHIFT
);
592 dout("invalidate_inode_pages2_range returned %d\n", ret
);
594 flags
= CEPH_OSD_FLAG_ORDERSNAP
|
595 CEPH_OSD_FLAG_ONDISK
|
598 while (iov_iter_count(from
) > 0) {
599 u64 len
= iov_iter_single_seg_count(from
);
603 snapc
= ci
->i_snap_realm
->cached_context
;
604 vino
= ceph_vino(inode
);
605 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
607 2,/*include a 'startsync' command*/
608 CEPH_OSD_OP_WRITE
, flags
, snapc
,
617 osd_req_op_init(req
, 1, CEPH_OSD_OP_STARTSYNC
);
619 n
= iov_iter_get_pages_alloc(from
, &pages
, len
, &start
);
620 if (unlikely(n
< 0)) {
622 ceph_osdc_put_request(req
);
626 num_pages
= (n
+ start
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
628 * throw out any page cache pages in this range. this
631 truncate_inode_pages_range(inode
->i_mapping
, pos
,
632 (pos
+n
) | (PAGE_CACHE_SIZE
-1));
633 osd_req_op_extent_osd_data_pages(req
, 0, pages
, n
, start
,
636 /* BUG_ON(vino.snap != CEPH_NOSNAP); */
637 ceph_osdc_build_request(req
, pos
, snapc
, vino
.snap
, &mtime
);
639 ret
= ceph_osdc_start_request(&fsc
->client
->osdc
, req
, false);
641 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
643 ceph_put_page_vector(pages
, num_pages
, false);
645 ceph_osdc_put_request(req
);
650 iov_iter_advance(from
, n
);
652 if (pos
> i_size_read(inode
)) {
653 check_caps
= ceph_inode_set_size(inode
, pos
);
655 ceph_check_caps(ceph_inode(inode
),
661 if (ret
!= -EOLDSNAPC
&& written
> 0) {
670 * Synchronous write, straight from __user pointer or user pages.
672 * If write spans object boundary, just do multiple writes. (For a
673 * correct atomic write, we should e.g. take write locks on all
674 * objects, rollback on failure, etc.)
677 ceph_sync_write(struct kiocb
*iocb
, struct iov_iter
*from
, loff_t pos
)
679 struct file
*file
= iocb
->ki_filp
;
680 struct inode
*inode
= file_inode(file
);
681 struct ceph_inode_info
*ci
= ceph_inode(inode
);
682 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
683 struct ceph_snap_context
*snapc
;
684 struct ceph_vino vino
;
685 struct ceph_osd_request
*req
;
693 struct timespec mtime
= CURRENT_TIME
;
694 size_t count
= iov_iter_count(from
);
696 if (ceph_snap(file_inode(file
)) != CEPH_NOSNAP
)
699 dout("sync_write on file %p %lld~%u\n", file
, pos
, (unsigned)count
);
701 ret
= filemap_write_and_wait_range(inode
->i_mapping
, pos
, pos
+ count
);
705 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
706 pos
>> PAGE_CACHE_SHIFT
,
707 (pos
+ count
) >> PAGE_CACHE_SHIFT
);
709 dout("invalidate_inode_pages2_range returned %d\n", ret
);
711 flags
= CEPH_OSD_FLAG_ORDERSNAP
|
712 CEPH_OSD_FLAG_ONDISK
|
713 CEPH_OSD_FLAG_WRITE
|
716 while ((len
= iov_iter_count(from
)) > 0) {
720 snapc
= ci
->i_snap_realm
->cached_context
;
721 vino
= ceph_vino(inode
);
722 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
723 vino
, pos
, &len
, 0, 1,
724 CEPH_OSD_OP_WRITE
, flags
, snapc
,
734 * write from beginning of first page,
735 * regardless of io alignment
737 num_pages
= (len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
739 pages
= ceph_alloc_page_vector(num_pages
, GFP_NOFS
);
741 ret
= PTR_ERR(pages
);
746 for (n
= 0; n
< num_pages
; n
++) {
747 size_t plen
= min_t(size_t, left
, PAGE_SIZE
);
748 ret
= copy_page_from_iter(pages
[n
], 0, plen
, from
);
757 ceph_release_page_vector(pages
, num_pages
);
761 /* get a second commit callback */
762 req
->r_unsafe_callback
= ceph_sync_write_unsafe
;
763 req
->r_inode
= inode
;
765 osd_req_op_extent_osd_data_pages(req
, 0, pages
, len
, 0,
768 /* BUG_ON(vino.snap != CEPH_NOSNAP); */
769 ceph_osdc_build_request(req
, pos
, snapc
, vino
.snap
, &mtime
);
771 ret
= ceph_osdc_start_request(&fsc
->client
->osdc
, req
, false);
773 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
776 ceph_osdc_put_request(req
);
781 if (pos
> i_size_read(inode
)) {
782 check_caps
= ceph_inode_set_size(inode
, pos
);
784 ceph_check_caps(ceph_inode(inode
),
792 if (ret
!= -EOLDSNAPC
&& written
> 0) {
800 * Wrap generic_file_aio_read with checks for cap bits on the inode.
801 * Atomically grab references, so that those bits are not released
802 * back to the MDS mid-read.
804 * Hmm, the sync read case isn't actually async... should it be?
806 static ssize_t
ceph_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
808 struct file
*filp
= iocb
->ki_filp
;
809 struct ceph_file_info
*fi
= filp
->private_data
;
810 size_t len
= iocb
->ki_nbytes
;
811 struct inode
*inode
= file_inode(filp
);
812 struct ceph_inode_info
*ci
= ceph_inode(inode
);
813 struct page
*pinned_page
= NULL
;
816 int retry_op
= 0, read
= 0;
819 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
820 inode
, ceph_vinop(inode
), iocb
->ki_pos
, (unsigned)len
, inode
);
822 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
823 want
= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_LAZYIO
;
825 want
= CEPH_CAP_FILE_CACHE
;
826 ret
= ceph_get_caps(ci
, CEPH_CAP_FILE_RD
, want
, -1, &got
, &pinned_page
);
830 if ((got
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)) == 0 ||
831 (iocb
->ki_filp
->f_flags
& O_DIRECT
) ||
832 (fi
->flags
& CEPH_F_SYNC
)) {
834 dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n",
835 inode
, ceph_vinop(inode
), iocb
->ki_pos
, (unsigned)len
,
836 ceph_cap_string(got
));
838 if (ci
->i_inline_version
== CEPH_INLINE_NONE
) {
839 /* hmm, this isn't really async... */
840 ret
= ceph_sync_read(iocb
, to
, &retry_op
);
842 retry_op
= READ_INLINE
;
845 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
846 inode
, ceph_vinop(inode
), iocb
->ki_pos
, (unsigned)len
,
847 ceph_cap_string(got
));
849 ret
= generic_file_read_iter(iocb
, to
);
851 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
852 inode
, ceph_vinop(inode
), ceph_cap_string(got
), (int)ret
);
854 page_cache_release(pinned_page
);
857 ceph_put_cap_refs(ci
, got
);
858 if (retry_op
&& ret
>= 0) {
860 struct page
*page
= NULL
;
862 if (retry_op
== READ_INLINE
) {
863 page
= __page_cache_alloc(GFP_NOFS
);
868 statret
= __ceph_do_getattr(inode
, page
,
869 CEPH_STAT_CAP_INLINE_DATA
, !!page
);
872 if (statret
== -ENODATA
) {
873 BUG_ON(retry_op
!= READ_INLINE
);
879 i_size
= i_size_read(inode
);
880 if (retry_op
== READ_INLINE
) {
881 /* does not support inline data > PAGE_SIZE */
882 if (i_size
> PAGE_CACHE_SIZE
) {
884 } else if (iocb
->ki_pos
< i_size
) {
885 loff_t end
= min_t(loff_t
, i_size
,
888 zero_user_segment(page
, statret
, end
);
889 ret
= copy_page_to_iter(page
,
890 iocb
->ki_pos
& ~PAGE_MASK
,
891 end
- iocb
->ki_pos
, to
);
896 __free_pages(page
, 0);
900 /* hit EOF or hole? */
901 if (retry_op
== CHECK_EOF
&& iocb
->ki_pos
< i_size
&&
903 dout("sync_read hit hole, ppos %lld < size %lld"
904 ", reading more\n", iocb
->ki_pos
,
921 * Take cap references to avoid releasing caps to MDS mid-write.
923 * If we are synchronous, and write with an old snap context, the OSD
924 * may return EOLDSNAPC. In that case, retry the write.. _after_
925 * dropping our cap refs and allowing the pending snap to logically
926 * complete _before_ this write occurs.
928 * If we are near ENOSPC, write synchronously.
930 static ssize_t
ceph_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
932 struct file
*file
= iocb
->ki_filp
;
933 struct ceph_file_info
*fi
= file
->private_data
;
934 struct inode
*inode
= file_inode(file
);
935 struct ceph_inode_info
*ci
= ceph_inode(inode
);
936 struct ceph_osd_client
*osdc
=
937 &ceph_sb_to_client(inode
->i_sb
)->client
->osdc
;
938 ssize_t count
= iov_iter_count(from
), written
= 0;
940 loff_t pos
= iocb
->ki_pos
;
942 if (ceph_snap(inode
) != CEPH_NOSNAP
)
945 mutex_lock(&inode
->i_mutex
);
947 /* We can write back this queue in page reclaim */
948 current
->backing_dev_info
= file
->f_mapping
->backing_dev_info
;
950 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
956 iov_iter_truncate(from
, count
);
958 err
= file_remove_suid(file
);
962 err
= file_update_time(file
);
966 if (ci
->i_inline_version
!= CEPH_INLINE_NONE
) {
967 err
= ceph_uninline_data(file
, NULL
);
973 if (ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_FULL
)) {
978 dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
979 inode
, ceph_vinop(inode
), pos
, count
, inode
->i_size
);
980 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
981 want
= CEPH_CAP_FILE_BUFFER
| CEPH_CAP_FILE_LAZYIO
;
983 want
= CEPH_CAP_FILE_BUFFER
;
985 err
= ceph_get_caps(ci
, CEPH_CAP_FILE_WR
, want
, pos
+ count
,
990 dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
991 inode
, ceph_vinop(inode
), pos
, count
, ceph_cap_string(got
));
993 if ((got
& (CEPH_CAP_FILE_BUFFER
|CEPH_CAP_FILE_LAZYIO
)) == 0 ||
994 (file
->f_flags
& O_DIRECT
) || (fi
->flags
& CEPH_F_SYNC
)) {
995 struct iov_iter data
;
996 mutex_unlock(&inode
->i_mutex
);
997 /* we might need to revert back to that point */
999 if (file
->f_flags
& O_DIRECT
)
1000 written
= ceph_sync_direct_write(iocb
, &data
, pos
);
1002 written
= ceph_sync_write(iocb
, &data
, pos
);
1003 if (written
== -EOLDSNAPC
) {
1004 dout("aio_write %p %llx.%llx %llu~%u"
1005 "got EOLDSNAPC, retrying\n",
1006 inode
, ceph_vinop(inode
),
1007 pos
, (unsigned)count
);
1008 mutex_lock(&inode
->i_mutex
);
1012 iov_iter_advance(from
, written
);
1014 loff_t old_size
= inode
->i_size
;
1016 * No need to acquire the i_truncate_mutex. Because
1017 * the MDS revokes Fwb caps before sending truncate
1018 * message to us. We can't get Fwb cap while there
1019 * are pending vmtruncate. So write and vmtruncate
1020 * can not run at the same time
1022 written
= generic_perform_write(file
, from
, pos
);
1023 if (likely(written
>= 0))
1024 iocb
->ki_pos
= pos
+ written
;
1025 if (inode
->i_size
> old_size
)
1026 ceph_fscache_update_objectsize(inode
);
1027 mutex_unlock(&inode
->i_mutex
);
1032 spin_lock(&ci
->i_ceph_lock
);
1033 ci
->i_inline_version
= CEPH_INLINE_NONE
;
1034 dirty
= __ceph_mark_dirty_caps(ci
, CEPH_CAP_FILE_WR
);
1035 spin_unlock(&ci
->i_ceph_lock
);
1037 __mark_inode_dirty(inode
, dirty
);
1040 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
1041 inode
, ceph_vinop(inode
), pos
, (unsigned)count
,
1042 ceph_cap_string(got
));
1043 ceph_put_cap_refs(ci
, got
);
1046 ((file
->f_flags
& O_SYNC
) || IS_SYNC(file
->f_mapping
->host
) ||
1047 ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_NEARFULL
))) {
1048 err
= vfs_fsync_range(file
, pos
, pos
+ written
- 1, 1);
1056 mutex_unlock(&inode
->i_mutex
);
1058 current
->backing_dev_info
= NULL
;
1059 return written
? written
: err
;
1063 * llseek. be sure to verify file size on SEEK_END.
1065 static loff_t
ceph_llseek(struct file
*file
, loff_t offset
, int whence
)
1067 struct inode
*inode
= file
->f_mapping
->host
;
1070 mutex_lock(&inode
->i_mutex
);
1072 if (whence
== SEEK_END
|| whence
== SEEK_DATA
|| whence
== SEEK_HOLE
) {
1073 ret
= ceph_do_getattr(inode
, CEPH_STAT_CAP_SIZE
, false);
1082 offset
+= inode
->i_size
;
1086 * Here we special-case the lseek(fd, 0, SEEK_CUR)
1087 * position-querying operation. Avoid rewriting the "same"
1088 * f_pos value back to the file because a concurrent read(),
1089 * write() or lseek() might have altered it
1092 offset
= file
->f_pos
;
1095 offset
+= file
->f_pos
;
1098 if (offset
>= inode
->i_size
) {
1104 if (offset
>= inode
->i_size
) {
1108 offset
= inode
->i_size
;
1112 offset
= vfs_setpos(file
, offset
, inode
->i_sb
->s_maxbytes
);
1115 mutex_unlock(&inode
->i_mutex
);
1119 static inline void ceph_zero_partial_page(
1120 struct inode
*inode
, loff_t offset
, unsigned size
)
1123 pgoff_t index
= offset
>> PAGE_CACHE_SHIFT
;
1125 page
= find_lock_page(inode
->i_mapping
, index
);
1127 wait_on_page_writeback(page
);
1128 zero_user(page
, offset
& (PAGE_CACHE_SIZE
- 1), size
);
1130 page_cache_release(page
);
1134 static void ceph_zero_pagecache_range(struct inode
*inode
, loff_t offset
,
1137 loff_t nearly
= round_up(offset
, PAGE_CACHE_SIZE
);
1138 if (offset
< nearly
) {
1139 loff_t size
= nearly
- offset
;
1142 ceph_zero_partial_page(inode
, offset
, size
);
1146 if (length
>= PAGE_CACHE_SIZE
) {
1147 loff_t size
= round_down(length
, PAGE_CACHE_SIZE
);
1148 truncate_pagecache_range(inode
, offset
, offset
+ size
- 1);
1153 ceph_zero_partial_page(inode
, offset
, length
);
1156 static int ceph_zero_partial_object(struct inode
*inode
,
1157 loff_t offset
, loff_t
*length
)
1159 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1160 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
1161 struct ceph_osd_request
*req
;
1167 op
= offset
? CEPH_OSD_OP_DELETE
: CEPH_OSD_OP_TRUNCATE
;
1170 op
= CEPH_OSD_OP_ZERO
;
1173 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
1177 CEPH_OSD_FLAG_WRITE
|
1178 CEPH_OSD_FLAG_ONDISK
,
1185 ceph_osdc_build_request(req
, offset
, NULL
, ceph_vino(inode
).snap
,
1188 ret
= ceph_osdc_start_request(&fsc
->client
->osdc
, req
, false);
1190 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
1194 ceph_osdc_put_request(req
);
1200 static int ceph_zero_objects(struct inode
*inode
, loff_t offset
, loff_t length
)
1203 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1204 s32 stripe_unit
= ceph_file_layout_su(ci
->i_layout
);
1205 s32 stripe_count
= ceph_file_layout_stripe_count(ci
->i_layout
);
1206 s32 object_size
= ceph_file_layout_object_size(ci
->i_layout
);
1207 u64 object_set_size
= object_size
* stripe_count
;
1210 /* round offset up to next period boundary */
1211 nearly
= offset
+ object_set_size
- 1;
1213 nearly
-= do_div(t
, object_set_size
);
1215 while (length
&& offset
< nearly
) {
1216 loff_t size
= length
;
1217 ret
= ceph_zero_partial_object(inode
, offset
, &size
);
1223 while (length
>= object_set_size
) {
1225 loff_t pos
= offset
;
1226 for (i
= 0; i
< stripe_count
; ++i
) {
1227 ret
= ceph_zero_partial_object(inode
, pos
, NULL
);
1232 offset
+= object_set_size
;
1233 length
-= object_set_size
;
1236 loff_t size
= length
;
1237 ret
= ceph_zero_partial_object(inode
, offset
, &size
);
1246 static long ceph_fallocate(struct file
*file
, int mode
,
1247 loff_t offset
, loff_t length
)
1249 struct ceph_file_info
*fi
= file
->private_data
;
1250 struct inode
*inode
= file_inode(file
);
1251 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1252 struct ceph_osd_client
*osdc
=
1253 &ceph_inode_to_client(inode
)->client
->osdc
;
1260 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
1263 if (!S_ISREG(inode
->i_mode
))
1266 mutex_lock(&inode
->i_mutex
);
1268 if (ceph_snap(inode
) != CEPH_NOSNAP
) {
1273 if (ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_FULL
) &&
1274 !(mode
& FALLOC_FL_PUNCH_HOLE
)) {
1279 if (ci
->i_inline_version
!= CEPH_INLINE_NONE
) {
1280 ret
= ceph_uninline_data(file
, NULL
);
1285 size
= i_size_read(inode
);
1286 if (!(mode
& FALLOC_FL_KEEP_SIZE
))
1287 endoff
= offset
+ length
;
1289 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
1290 want
= CEPH_CAP_FILE_BUFFER
| CEPH_CAP_FILE_LAZYIO
;
1292 want
= CEPH_CAP_FILE_BUFFER
;
1294 ret
= ceph_get_caps(ci
, CEPH_CAP_FILE_WR
, want
, endoff
, &got
, NULL
);
1298 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
1300 ceph_zero_pagecache_range(inode
, offset
, length
);
1301 ret
= ceph_zero_objects(inode
, offset
, length
);
1302 } else if (endoff
> size
) {
1303 truncate_pagecache_range(inode
, size
, -1);
1304 if (ceph_inode_set_size(inode
, endoff
))
1305 ceph_check_caps(ceph_inode(inode
),
1306 CHECK_CAPS_AUTHONLY
, NULL
);
1310 spin_lock(&ci
->i_ceph_lock
);
1311 ci
->i_inline_version
= CEPH_INLINE_NONE
;
1312 dirty
= __ceph_mark_dirty_caps(ci
, CEPH_CAP_FILE_WR
);
1313 spin_unlock(&ci
->i_ceph_lock
);
1315 __mark_inode_dirty(inode
, dirty
);
1318 ceph_put_cap_refs(ci
, got
);
1320 mutex_unlock(&inode
->i_mutex
);
1324 const struct file_operations ceph_file_fops
= {
1326 .release
= ceph_release
,
1327 .llseek
= ceph_llseek
,
1328 .read
= new_sync_read
,
1329 .write
= new_sync_write
,
1330 .read_iter
= ceph_read_iter
,
1331 .write_iter
= ceph_write_iter
,
1333 .fsync
= ceph_fsync
,
1335 .flock
= ceph_flock
,
1336 .splice_read
= generic_file_splice_read
,
1337 .splice_write
= iter_file_splice_write
,
1338 .unlocked_ioctl
= ceph_ioctl
,
1339 .compat_ioctl
= ceph_ioctl
,
1340 .fallocate
= ceph_fallocate
,