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"
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
;
73 switch (inode
->i_mode
& S_IFMT
) {
76 dout("init_file %p %p 0%o (regular)\n", inode
, file
,
78 cf
= kmem_cache_alloc(ceph_file_cachep
, GFP_NOFS
| __GFP_ZERO
);
80 ceph_put_fmode(ceph_inode(inode
), fmode
); /* clean up */
85 file
->private_data
= cf
;
86 BUG_ON(inode
->i_fop
->release
!= ceph_release
);
90 dout("init_file %p %p 0%o (symlink)\n", inode
, file
,
92 ceph_put_fmode(ceph_inode(inode
), fmode
); /* clean up */
96 dout("init_file %p %p 0%o (special)\n", inode
, file
,
99 * we need to drop the open ref now, since we don't
100 * have .release set to ceph_release.
102 ceph_put_fmode(ceph_inode(inode
), fmode
); /* clean up */
103 BUG_ON(inode
->i_fop
->release
== ceph_release
);
105 /* call the proper open fop */
106 ret
= inode
->i_fop
->open(inode
, file
);
112 * If we already have the requisite capabilities, we can satisfy
113 * the open request locally (no need to request new caps from the
114 * MDS). We do, however, need to inform the MDS (asynchronously)
115 * if our wanted caps set expands.
117 int ceph_open(struct inode
*inode
, struct file
*file
)
119 struct ceph_inode_info
*ci
= ceph_inode(inode
);
120 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
121 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
122 struct ceph_mds_request
*req
;
123 struct ceph_file_info
*cf
= file
->private_data
;
124 struct inode
*parent_inode
= NULL
;
126 int flags
, fmode
, wanted
;
129 dout("open file %p is already opened\n", file
);
133 /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */
134 flags
= file
->f_flags
& ~(O_CREAT
|O_EXCL
);
135 if (S_ISDIR(inode
->i_mode
))
136 flags
= O_DIRECTORY
; /* mds likes to know */
138 dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode
,
139 ceph_vinop(inode
), file
, flags
, file
->f_flags
);
140 fmode
= ceph_flags_to_mode(flags
);
141 wanted
= ceph_caps_for_mode(fmode
);
143 /* snapped files are read-only */
144 if (ceph_snap(inode
) != CEPH_NOSNAP
&& (file
->f_mode
& FMODE_WRITE
))
147 /* trivially open snapdir */
148 if (ceph_snap(inode
) == CEPH_SNAPDIR
) {
149 spin_lock(&ci
->i_ceph_lock
);
150 __ceph_get_fmode(ci
, fmode
);
151 spin_unlock(&ci
->i_ceph_lock
);
152 return ceph_init_file(inode
, file
, fmode
);
156 * No need to block if we have caps on the auth MDS (for
157 * write) or any MDS (for read). Update wanted set
160 spin_lock(&ci
->i_ceph_lock
);
161 if (__ceph_is_any_real_caps(ci
) &&
162 (((fmode
& CEPH_FILE_MODE_WR
) == 0) || ci
->i_auth_cap
)) {
163 int mds_wanted
= __ceph_caps_mds_wanted(ci
);
164 int issued
= __ceph_caps_issued(ci
, NULL
);
166 dout("open %p fmode %d want %s issued %s using existing\n",
167 inode
, fmode
, ceph_cap_string(wanted
),
168 ceph_cap_string(issued
));
169 __ceph_get_fmode(ci
, fmode
);
170 spin_unlock(&ci
->i_ceph_lock
);
173 if ((issued
& wanted
) != wanted
&&
174 (mds_wanted
& wanted
) != wanted
&&
175 ceph_snap(inode
) != CEPH_SNAPDIR
)
176 ceph_check_caps(ci
, 0, NULL
);
178 return ceph_init_file(inode
, file
, fmode
);
179 } else if (ceph_snap(inode
) != CEPH_NOSNAP
&&
180 (ci
->i_snap_caps
& wanted
) == wanted
) {
181 __ceph_get_fmode(ci
, fmode
);
182 spin_unlock(&ci
->i_ceph_lock
);
183 return ceph_init_file(inode
, file
, fmode
);
185 spin_unlock(&ci
->i_ceph_lock
);
187 dout("open fmode %d wants %s\n", fmode
, ceph_cap_string(wanted
));
188 req
= prepare_open_request(inode
->i_sb
, flags
, 0);
193 req
->r_inode
= inode
;
196 if (flags
& (O_CREAT
|O_TRUNC
))
197 parent_inode
= ceph_get_dentry_parent_inode(file
->f_dentry
);
198 err
= ceph_mdsc_do_request(mdsc
, parent_inode
, req
);
201 err
= ceph_init_file(inode
, file
, req
->r_fmode
);
202 ceph_mdsc_put_request(req
);
203 dout("open result=%d on %llx.%llx\n", err
, ceph_vinop(inode
));
210 * Do a lookup + open with a single request. If we get a non-existent
211 * file or symlink, return 1 so the VFS can retry.
213 int ceph_atomic_open(struct inode
*dir
, struct dentry
*dentry
,
214 struct file
*file
, unsigned flags
, umode_t mode
,
217 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
218 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
219 struct ceph_mds_request
*req
;
223 dout("atomic_open %p dentry %p '%.*s' %s flags %d mode 0%o\n",
224 dir
, dentry
, dentry
->d_name
.len
, dentry
->d_name
.name
,
225 d_unhashed(dentry
) ? "unhashed" : "hashed", flags
, mode
);
227 if (dentry
->d_name
.len
> NAME_MAX
)
228 return -ENAMETOOLONG
;
230 err
= ceph_init_dentry(dentry
);
235 req
= prepare_open_request(dir
->i_sb
, flags
, mode
);
238 req
->r_dentry
= dget(dentry
);
240 if (flags
& O_CREAT
) {
241 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
242 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
244 req
->r_locked_dir
= dir
; /* caller holds dir->i_mutex */
245 err
= ceph_mdsc_do_request(mdsc
,
246 (flags
& (O_CREAT
|O_TRUNC
)) ? dir
: NULL
,
251 err
= ceph_handle_snapdir(req
, dentry
, err
);
252 if (err
== 0 && (flags
& O_CREAT
) && !req
->r_reply_info
.head
->is_dentry
)
253 err
= ceph_handle_notrace_create(dir
, dentry
);
255 if (d_unhashed(dentry
)) {
256 dn
= ceph_finish_lookup(req
, dentry
, err
);
260 /* we were given a hashed negative dentry */
265 if (dn
|| dentry
->d_inode
== NULL
|| S_ISLNK(dentry
->d_inode
->i_mode
)) {
266 /* make vfs retry on splice, ENOENT, or symlink */
267 dout("atomic_open finish_no_open on dn %p\n", dn
);
268 err
= finish_no_open(file
, dn
);
270 dout("atomic_open finish_open on dn %p\n", dn
);
271 if (req
->r_op
== CEPH_MDS_OP_CREATE
&& req
->r_reply_info
.has_create_ino
) {
272 *opened
|= FILE_CREATED
;
274 err
= finish_open(file
, dentry
, ceph_open
, opened
);
278 ceph_mdsc_put_request(req
);
279 dout("atomic_open result=%d\n", err
);
283 int ceph_release(struct inode
*inode
, struct file
*file
)
285 struct ceph_inode_info
*ci
= ceph_inode(inode
);
286 struct ceph_file_info
*cf
= file
->private_data
;
288 dout("release inode %p file %p\n", inode
, file
);
289 ceph_put_fmode(ci
, cf
->fmode
);
290 if (cf
->last_readdir
)
291 ceph_mdsc_put_request(cf
->last_readdir
);
292 kfree(cf
->last_name
);
295 kmem_cache_free(ceph_file_cachep
, cf
);
297 /* wake up anyone waiting for caps on this inode */
298 wake_up_all(&ci
->i_cap_wq
);
303 * Read a range of bytes striped over one or more objects. Iterate over
304 * objects we stripe over. (That's not atomic, but good enough for now.)
306 * If we get a short result from the OSD, check against i_size; we need to
307 * only return a short read to the caller if we hit EOF.
309 static int striped_read(struct inode
*inode
,
311 struct page
**pages
, int num_pages
,
312 int *checkeof
, bool o_direct
,
313 unsigned long buf_align
)
315 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
316 struct ceph_inode_info
*ci
= ceph_inode(inode
);
317 u64 pos
, this_len
, left
;
318 int io_align
, page_align
;
321 struct page
**page_pos
;
323 bool hit_stripe
, was_short
;
326 * we may need to do multiple reads. not atomic, unfortunately.
331 pages_left
= num_pages
;
333 io_align
= off
& ~PAGE_MASK
;
337 page_align
= (pos
- io_align
+ buf_align
) & ~PAGE_MASK
;
339 page_align
= pos
& ~PAGE_MASK
;
341 ret
= ceph_osdc_readpages(&fsc
->client
->osdc
, ceph_vino(inode
),
342 &ci
->i_layout
, pos
, &this_len
,
345 page_pos
, pages_left
, page_align
);
348 hit_stripe
= this_len
< left
;
349 was_short
= ret
>= 0 && ret
< this_len
;
350 dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos
, left
, read
,
351 ret
, hit_stripe
? " HITSTRIPE" : "", was_short
? " SHORT" : "");
355 if (was_short
&& (pos
+ ret
< inode
->i_size
)) {
356 u64 tmp
= min(this_len
- ret
,
357 inode
->i_size
- pos
- ret
);
358 dout(" zero gap %llu to %llu\n",
359 pos
+ ret
, pos
+ ret
+ tmp
);
360 ceph_zero_page_vector_range(page_align
+ read
+ ret
,
365 didpages
= (page_align
+ ret
) >> PAGE_CACHE_SHIFT
;
369 page_pos
+= didpages
;
370 pages_left
-= didpages
;
372 /* hit stripe and need continue*/
373 if (left
&& hit_stripe
&& pos
< inode
->i_size
)
379 /* did we bounce off eof? */
380 if (pos
+ left
> inode
->i_size
)
384 dout("striped_read returns %d\n", ret
);
389 * Completely synchronous read and write methods. Direct from __user
390 * buffer to osd, or directly to user pages (if O_DIRECT).
392 * If the read spans object boundary, just do multiple reads.
394 static ssize_t
ceph_sync_read(struct file
*file
, char __user
*data
,
395 unsigned len
, loff_t
*poff
, int *checkeof
)
397 struct inode
*inode
= file_inode(file
);
402 dout("sync_read on file %p %llu~%u %s\n", file
, off
, len
,
403 (file
->f_flags
& O_DIRECT
) ? "O_DIRECT" : "");
405 if (file
->f_flags
& O_DIRECT
) {
406 num_pages
= calc_pages_for((unsigned long)data
, len
);
407 pages
= ceph_get_direct_page_vector(data
, num_pages
, true);
409 num_pages
= calc_pages_for(off
, len
);
410 pages
= ceph_alloc_page_vector(num_pages
, GFP_NOFS
);
413 return PTR_ERR(pages
);
416 * flush any page cache pages in this range. this
417 * will make concurrent normal and sync io slow,
418 * but it will at least behave sensibly when they are
421 ret
= filemap_write_and_wait(inode
->i_mapping
);
425 ret
= striped_read(inode
, off
, len
, pages
, num_pages
, checkeof
,
426 file
->f_flags
& O_DIRECT
,
427 (unsigned long)data
& ~PAGE_MASK
);
429 if (ret
>= 0 && (file
->f_flags
& O_DIRECT
) == 0)
430 ret
= ceph_copy_page_vector_to_user(pages
, data
, off
, ret
);
435 if (file
->f_flags
& O_DIRECT
)
436 ceph_put_page_vector(pages
, num_pages
, true);
438 ceph_release_page_vector(pages
, num_pages
);
439 dout("sync_read result %d\n", ret
);
444 * Write commit request unsafe callback, called to tell us when a
445 * request is unsafe (that is, in flight--has been handed to the
446 * messenger to send to its target osd). It is called again when
447 * we've received a response message indicating the request is
448 * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request
449 * is completed early (and unsuccessfully) due to a timeout or
452 * This is used if we requested both an ACK and ONDISK commit reply
455 static void ceph_sync_write_unsafe(struct ceph_osd_request
*req
, bool unsafe
)
457 struct ceph_inode_info
*ci
= ceph_inode(req
->r_inode
);
459 dout("%s %p tid %llu %ssafe\n", __func__
, req
, req
->r_tid
,
462 ceph_get_cap_refs(ci
, CEPH_CAP_FILE_WR
);
463 spin_lock(&ci
->i_unsafe_lock
);
464 list_add_tail(&req
->r_unsafe_item
,
465 &ci
->i_unsafe_writes
);
466 spin_unlock(&ci
->i_unsafe_lock
);
468 spin_lock(&ci
->i_unsafe_lock
);
469 list_del_init(&req
->r_unsafe_item
);
470 spin_unlock(&ci
->i_unsafe_lock
);
471 ceph_put_cap_refs(ci
, CEPH_CAP_FILE_WR
);
476 * Synchronous write, straight from __user pointer or user pages (if
479 * If write spans object boundary, just do multiple writes. (For a
480 * correct atomic write, we should e.g. take write locks on all
481 * objects, rollback on failure, etc.)
483 static ssize_t
ceph_sync_write(struct file
*file
, const char __user
*data
,
484 size_t left
, loff_t pos
, loff_t
*ppos
)
486 struct inode
*inode
= file_inode(file
);
487 struct ceph_inode_info
*ci
= ceph_inode(inode
);
488 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
489 struct ceph_snap_context
*snapc
;
490 struct ceph_vino vino
;
491 struct ceph_osd_request
*req
;
499 int page_align
, io_align
;
500 unsigned long buf_align
;
502 struct timespec mtime
= CURRENT_TIME
;
503 bool own_pages
= false;
505 if (ceph_snap(file_inode(file
)) != CEPH_NOSNAP
)
508 dout("sync_write on file %p %lld~%u %s\n", file
, pos
,
509 (unsigned)left
, (file
->f_flags
& O_DIRECT
) ? "O_DIRECT" : "");
511 ret
= filemap_write_and_wait_range(inode
->i_mapping
, pos
, pos
+ left
);
515 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
516 pos
>> PAGE_CACHE_SHIFT
,
517 (pos
+ left
) >> PAGE_CACHE_SHIFT
);
519 dout("invalidate_inode_pages2_range returned %d\n", ret
);
521 flags
= CEPH_OSD_FLAG_ORDERSNAP
|
522 CEPH_OSD_FLAG_ONDISK
|
524 if ((file
->f_flags
& (O_SYNC
|O_DIRECT
)) == 0)
525 flags
|= CEPH_OSD_FLAG_ACK
;
527 num_ops
++; /* Also include a 'startsync' command. */
530 * we may need to do multiple writes here if we span an object
531 * boundary. this isn't atomic, unfortunately. :(
534 io_align
= pos
& ~PAGE_MASK
;
535 buf_align
= (unsigned long)data
& ~PAGE_MASK
;
538 snapc
= ci
->i_snap_realm
->cached_context
;
539 vino
= ceph_vino(inode
);
540 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
541 vino
, pos
, &len
, num_ops
,
542 CEPH_OSD_OP_WRITE
, flags
, snapc
,
543 ci
->i_truncate_seq
, ci
->i_truncate_size
,
548 /* write from beginning of first page, regardless of io alignment */
549 page_align
= file
->f_flags
& O_DIRECT
? buf_align
: io_align
;
550 num_pages
= calc_pages_for(page_align
, len
);
551 if (file
->f_flags
& O_DIRECT
) {
552 pages
= ceph_get_direct_page_vector(data
, num_pages
, false);
554 ret
= PTR_ERR(pages
);
559 * throw out any page cache pages in this range. this
562 truncate_inode_pages_range(inode
->i_mapping
, pos
,
563 (pos
+len
) | (PAGE_CACHE_SIZE
-1));
565 pages
= ceph_alloc_page_vector(num_pages
, GFP_NOFS
);
567 ret
= PTR_ERR(pages
);
570 ret
= ceph_copy_user_to_page_vector(pages
, data
, pos
, len
);
572 ceph_release_page_vector(pages
, num_pages
);
576 if ((file
->f_flags
& O_SYNC
) == 0) {
577 /* get a second commit callback */
578 req
->r_unsafe_callback
= ceph_sync_write_unsafe
;
579 req
->r_inode
= inode
;
583 osd_req_op_extent_osd_data_pages(req
, 0, pages
, len
, page_align
,
586 /* BUG_ON(vino.snap != CEPH_NOSNAP); */
587 ceph_osdc_build_request(req
, pos
, snapc
, vino
.snap
, &mtime
);
589 ret
= ceph_osdc_start_request(&fsc
->client
->osdc
, req
, false);
591 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
593 if (file
->f_flags
& O_DIRECT
)
594 ceph_put_page_vector(pages
, num_pages
, false);
595 else if (file
->f_flags
& O_SYNC
)
596 ceph_release_page_vector(pages
, num_pages
);
599 ceph_osdc_put_request(req
);
610 if (pos
> i_size_read(inode
))
611 check_caps
= ceph_inode_set_size(inode
, pos
);
613 ceph_check_caps(ceph_inode(inode
), CHECK_CAPS_AUTHONLY
,
615 } else if (ret
!= -EOLDSNAPC
&& written
> 0) {
622 * Wrap generic_file_aio_read with checks for cap bits on the inode.
623 * Atomically grab references, so that those bits are not released
624 * back to the MDS mid-read.
626 * Hmm, the sync read case isn't actually async... should it be?
628 static ssize_t
ceph_aio_read(struct kiocb
*iocb
, const struct iovec
*iov
,
629 unsigned long nr_segs
, loff_t pos
)
631 struct file
*filp
= iocb
->ki_filp
;
632 struct ceph_file_info
*fi
= filp
->private_data
;
633 loff_t
*ppos
= &iocb
->ki_pos
;
634 size_t len
= iov
->iov_len
;
635 struct inode
*inode
= file_inode(filp
);
636 struct ceph_inode_info
*ci
= ceph_inode(inode
);
637 void __user
*base
= iov
->iov_base
;
640 int checkeof
= 0, read
= 0;
642 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
643 inode
, ceph_vinop(inode
), pos
, (unsigned)len
, inode
);
645 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
646 want
= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_LAZYIO
;
648 want
= CEPH_CAP_FILE_CACHE
;
649 ret
= ceph_get_caps(ci
, CEPH_CAP_FILE_RD
, want
, &got
, -1);
652 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
653 inode
, ceph_vinop(inode
), pos
, (unsigned)len
,
654 ceph_cap_string(got
));
656 if ((got
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)) == 0 ||
657 (iocb
->ki_filp
->f_flags
& O_DIRECT
) ||
658 (fi
->flags
& CEPH_F_SYNC
))
659 /* hmm, this isn't really async... */
660 ret
= ceph_sync_read(filp
, base
, len
, ppos
, &checkeof
);
662 ret
= generic_file_aio_read(iocb
, iov
, nr_segs
, pos
);
665 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
666 inode
, ceph_vinop(inode
), ceph_cap_string(got
), (int)ret
);
667 ceph_put_cap_refs(ci
, got
);
669 if (checkeof
&& ret
>= 0) {
670 int statret
= ceph_do_getattr(inode
, CEPH_STAT_CAP_SIZE
);
672 /* hit EOF or hole? */
673 if (statret
== 0 && *ppos
< inode
->i_size
) {
674 dout("aio_read sync_read hit hole, ppos %lld < size %lld, reading more\n", *ppos
, inode
->i_size
);
689 * Take cap references to avoid releasing caps to MDS mid-write.
691 * If we are synchronous, and write with an old snap context, the OSD
692 * may return EOLDSNAPC. In that case, retry the write.. _after_
693 * dropping our cap refs and allowing the pending snap to logically
694 * complete _before_ this write occurs.
696 * If we are near ENOSPC, write synchronously.
698 static ssize_t
ceph_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
699 unsigned long nr_segs
, loff_t pos
)
701 struct file
*file
= iocb
->ki_filp
;
702 struct ceph_file_info
*fi
= file
->private_data
;
703 struct inode
*inode
= file_inode(file
);
704 struct ceph_inode_info
*ci
= ceph_inode(inode
);
705 struct ceph_osd_client
*osdc
=
706 &ceph_sb_to_client(inode
->i_sb
)->client
->osdc
;
707 ssize_t count
, written
= 0;
710 if (ceph_snap(inode
) != CEPH_NOSNAP
)
713 mutex_lock(&inode
->i_mutex
);
715 err
= generic_segment_checks(iov
, &nr_segs
, &count
, VERIFY_READ
);
719 /* We can write back this queue in page reclaim */
720 current
->backing_dev_info
= file
->f_mapping
->backing_dev_info
;
722 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
729 err
= file_remove_suid(file
);
733 err
= file_update_time(file
);
738 if (ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_FULL
)) {
743 dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
744 inode
, ceph_vinop(inode
), pos
, count
, inode
->i_size
);
745 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
746 want
= CEPH_CAP_FILE_BUFFER
| CEPH_CAP_FILE_LAZYIO
;
748 want
= CEPH_CAP_FILE_BUFFER
;
750 err
= ceph_get_caps(ci
, CEPH_CAP_FILE_WR
, want
, &got
, pos
+ count
);
754 dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
755 inode
, ceph_vinop(inode
), pos
, count
, ceph_cap_string(got
));
757 if ((got
& (CEPH_CAP_FILE_BUFFER
|CEPH_CAP_FILE_LAZYIO
)) == 0 ||
758 (iocb
->ki_filp
->f_flags
& O_DIRECT
) ||
759 (fi
->flags
& CEPH_F_SYNC
)) {
760 mutex_unlock(&inode
->i_mutex
);
761 written
= ceph_sync_write(file
, iov
->iov_base
, count
,
763 if (written
== -EOLDSNAPC
) {
764 dout("aio_write %p %llx.%llx %llu~%u"
765 "got EOLDSNAPC, retrying\n",
766 inode
, ceph_vinop(inode
),
767 pos
, (unsigned)iov
->iov_len
);
768 mutex_lock(&inode
->i_mutex
);
773 * No need to acquire the i_truncate_mutex. Because
774 * the MDS revokes Fwb caps before sending truncate
775 * message to us. We can't get Fwb cap while there
776 * are pending vmtruncate. So write and vmtruncate
777 * can not run at the same time
779 written
= generic_file_buffered_write(iocb
, iov
, nr_segs
,
782 mutex_unlock(&inode
->i_mutex
);
787 spin_lock(&ci
->i_ceph_lock
);
788 dirty
= __ceph_mark_dirty_caps(ci
, CEPH_CAP_FILE_WR
);
789 spin_unlock(&ci
->i_ceph_lock
);
791 __mark_inode_dirty(inode
, dirty
);
794 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
795 inode
, ceph_vinop(inode
), pos
, (unsigned)iov
->iov_len
,
796 ceph_cap_string(got
));
797 ceph_put_cap_refs(ci
, got
);
800 ((file
->f_flags
& O_SYNC
) || IS_SYNC(file
->f_mapping
->host
) ||
801 ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_NEARFULL
))) {
802 err
= vfs_fsync_range(file
, pos
, pos
+ written
- 1, 1);
810 mutex_unlock(&inode
->i_mutex
);
812 current
->backing_dev_info
= NULL
;
813 return written
? written
: err
;
817 * llseek. be sure to verify file size on SEEK_END.
819 static loff_t
ceph_llseek(struct file
*file
, loff_t offset
, int whence
)
821 struct inode
*inode
= file
->f_mapping
->host
;
824 mutex_lock(&inode
->i_mutex
);
826 if (whence
== SEEK_END
|| whence
== SEEK_DATA
|| whence
== SEEK_HOLE
) {
827 ret
= ceph_do_getattr(inode
, CEPH_STAT_CAP_SIZE
);
836 offset
+= inode
->i_size
;
840 * Here we special-case the lseek(fd, 0, SEEK_CUR)
841 * position-querying operation. Avoid rewriting the "same"
842 * f_pos value back to the file because a concurrent read(),
843 * write() or lseek() might have altered it
846 offset
= file
->f_pos
;
849 offset
+= file
->f_pos
;
852 if (offset
>= inode
->i_size
) {
858 if (offset
>= inode
->i_size
) {
862 offset
= inode
->i_size
;
866 offset
= vfs_setpos(file
, offset
, inode
->i_sb
->s_maxbytes
);
869 mutex_unlock(&inode
->i_mutex
);
873 static inline void ceph_zero_partial_page(
874 struct inode
*inode
, loff_t offset
, unsigned size
)
877 pgoff_t index
= offset
>> PAGE_CACHE_SHIFT
;
879 page
= find_lock_page(inode
->i_mapping
, index
);
881 wait_on_page_writeback(page
);
882 zero_user(page
, offset
& (PAGE_CACHE_SIZE
- 1), size
);
884 page_cache_release(page
);
888 static void ceph_zero_pagecache_range(struct inode
*inode
, loff_t offset
,
891 loff_t nearly
= round_up(offset
, PAGE_CACHE_SIZE
);
892 if (offset
< nearly
) {
893 loff_t size
= nearly
- offset
;
896 ceph_zero_partial_page(inode
, offset
, size
);
900 if (length
>= PAGE_CACHE_SIZE
) {
901 loff_t size
= round_down(length
, PAGE_CACHE_SIZE
);
902 truncate_pagecache_range(inode
, offset
, offset
+ size
- 1);
907 ceph_zero_partial_page(inode
, offset
, length
);
910 static int ceph_zero_partial_object(struct inode
*inode
,
911 loff_t offset
, loff_t
*length
)
913 struct ceph_inode_info
*ci
= ceph_inode(inode
);
914 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
915 struct ceph_osd_request
*req
;
921 op
= offset
? CEPH_OSD_OP_DELETE
: CEPH_OSD_OP_TRUNCATE
;
924 op
= CEPH_OSD_OP_ZERO
;
927 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
931 CEPH_OSD_FLAG_WRITE
|
932 CEPH_OSD_FLAG_ONDISK
,
939 ceph_osdc_build_request(req
, offset
, NULL
, ceph_vino(inode
).snap
,
942 ret
= ceph_osdc_start_request(&fsc
->client
->osdc
, req
, false);
944 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
948 ceph_osdc_put_request(req
);
954 static int ceph_zero_objects(struct inode
*inode
, loff_t offset
, loff_t length
)
957 struct ceph_inode_info
*ci
= ceph_inode(inode
);
958 s32 stripe_unit
= ceph_file_layout_su(ci
->i_layout
);
959 s32 stripe_count
= ceph_file_layout_stripe_count(ci
->i_layout
);
960 s32 object_size
= ceph_file_layout_object_size(ci
->i_layout
);
961 u64 object_set_size
= object_size
* stripe_count
;
964 /* round offset up to next period boundary */
965 nearly
= offset
+ object_set_size
- 1;
967 nearly
-= do_div(t
, object_set_size
);
969 while (length
&& offset
< nearly
) {
970 loff_t size
= length
;
971 ret
= ceph_zero_partial_object(inode
, offset
, &size
);
977 while (length
>= object_set_size
) {
980 for (i
= 0; i
< stripe_count
; ++i
) {
981 ret
= ceph_zero_partial_object(inode
, pos
, NULL
);
986 offset
+= object_set_size
;
987 length
-= object_set_size
;
990 loff_t size
= length
;
991 ret
= ceph_zero_partial_object(inode
, offset
, &size
);
1000 static long ceph_fallocate(struct file
*file
, int mode
,
1001 loff_t offset
, loff_t length
)
1003 struct ceph_file_info
*fi
= file
->private_data
;
1004 struct inode
*inode
= file
->f_dentry
->d_inode
;
1005 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1006 struct ceph_osd_client
*osdc
=
1007 &ceph_inode_to_client(inode
)->client
->osdc
;
1014 if (!S_ISREG(inode
->i_mode
))
1017 if (IS_SWAPFILE(inode
))
1020 mutex_lock(&inode
->i_mutex
);
1022 if (ceph_snap(inode
) != CEPH_NOSNAP
) {
1027 if (ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_FULL
) &&
1028 !(mode
& FALLOC_FL_PUNCH_HOLE
)) {
1033 size
= i_size_read(inode
);
1034 if (!(mode
& FALLOC_FL_KEEP_SIZE
))
1035 endoff
= offset
+ length
;
1037 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
1038 want
= CEPH_CAP_FILE_BUFFER
| CEPH_CAP_FILE_LAZYIO
;
1040 want
= CEPH_CAP_FILE_BUFFER
;
1042 ret
= ceph_get_caps(ci
, CEPH_CAP_FILE_WR
, want
, &got
, endoff
);
1046 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
1048 ceph_zero_pagecache_range(inode
, offset
, length
);
1049 ret
= ceph_zero_objects(inode
, offset
, length
);
1050 } else if (endoff
> size
) {
1051 truncate_pagecache_range(inode
, size
, -1);
1052 if (ceph_inode_set_size(inode
, endoff
))
1053 ceph_check_caps(ceph_inode(inode
),
1054 CHECK_CAPS_AUTHONLY
, NULL
);
1058 spin_lock(&ci
->i_ceph_lock
);
1059 dirty
= __ceph_mark_dirty_caps(ci
, CEPH_CAP_FILE_WR
);
1060 spin_unlock(&ci
->i_ceph_lock
);
1062 __mark_inode_dirty(inode
, dirty
);
1065 ceph_put_cap_refs(ci
, got
);
1067 mutex_unlock(&inode
->i_mutex
);
1071 const struct file_operations ceph_file_fops
= {
1073 .release
= ceph_release
,
1074 .llseek
= ceph_llseek
,
1075 .read
= do_sync_read
,
1076 .write
= do_sync_write
,
1077 .aio_read
= ceph_aio_read
,
1078 .aio_write
= ceph_aio_write
,
1080 .fsync
= ceph_fsync
,
1082 .flock
= ceph_flock
,
1083 .splice_read
= generic_file_splice_read
,
1084 .splice_write
= generic_file_splice_write
,
1085 .unlocked_ioctl
= ceph_ioctl
,
1086 .compat_ioctl
= ceph_ioctl
,
1087 .fallocate
= ceph_fallocate
,