1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
3 #include <linux/ceph/striper.h>
5 #include <linux/module.h>
6 #include <linux/sched.h>
7 #include <linux/slab.h>
8 #include <linux/file.h>
9 #include <linux/mount.h>
10 #include <linux/namei.h>
11 #include <linux/writeback.h>
12 #include <linux/falloc.h>
13 #include <linux/iversion.h>
14 #include <linux/ktime.h>
17 #include "mds_client.h"
22 static __le32
ceph_flags_sys2wire(u32 flags
)
26 switch (flags
& O_ACCMODE
) {
28 wire_flags
|= CEPH_O_RDONLY
;
31 wire_flags
|= CEPH_O_WRONLY
;
34 wire_flags
|= CEPH_O_RDWR
;
40 #define ceph_sys2wire(a) if (flags & a) { wire_flags |= CEPH_##a; flags &= ~a; }
42 ceph_sys2wire(O_CREAT
);
43 ceph_sys2wire(O_EXCL
);
44 ceph_sys2wire(O_TRUNC
);
45 ceph_sys2wire(O_DIRECTORY
);
46 ceph_sys2wire(O_NOFOLLOW
);
51 dout("unused open flags: %x\n", flags
);
53 return cpu_to_le32(wire_flags
);
57 * Ceph file operations
59 * Implement basic open/close functionality, and implement
62 * We implement three modes of file I/O:
63 * - buffered uses the generic_file_aio_{read,write} helpers
65 * - synchronous is used when there is multi-client read/write
66 * sharing, avoids the page cache, and synchronously waits for an
69 * - direct io takes the variant of the sync path that references
70 * user pages directly.
72 * fsync() flushes and waits on dirty pages, but just queues metadata
73 * for writeback: since the MDS can recover size and mtime there is no
74 * need to wait for MDS acknowledgement.
78 * How many pages to get in one call to iov_iter_get_pages(). This
79 * determines the size of the on-stack array used as a buffer.
81 #define ITER_GET_BVECS_PAGES 64
83 static ssize_t
__iter_get_bvecs(struct iov_iter
*iter
, size_t maxsize
,
84 struct bio_vec
*bvecs
)
89 if (maxsize
> iov_iter_count(iter
))
90 maxsize
= iov_iter_count(iter
);
92 while (size
< maxsize
) {
93 struct page
*pages
[ITER_GET_BVECS_PAGES
];
98 bytes
= iov_iter_get_pages(iter
, pages
, maxsize
- size
,
99 ITER_GET_BVECS_PAGES
, &start
);
101 return size
?: bytes
;
103 iov_iter_advance(iter
, bytes
);
106 for ( ; bytes
; idx
++, bvec_idx
++) {
107 struct bio_vec bv
= {
108 .bv_page
= pages
[idx
],
109 .bv_len
= min_t(int, bytes
, PAGE_SIZE
- start
),
113 bvecs
[bvec_idx
] = bv
;
123 * iov_iter_get_pages() only considers one iov_iter segment, no matter
124 * what maxsize or maxpages are given. For ITER_BVEC that is a single
127 * Attempt to get up to @maxsize bytes worth of pages from @iter.
128 * Return the number of bytes in the created bio_vec array, or an error.
130 static ssize_t
iter_get_bvecs_alloc(struct iov_iter
*iter
, size_t maxsize
,
131 struct bio_vec
**bvecs
, int *num_bvecs
)
134 size_t orig_count
= iov_iter_count(iter
);
138 iov_iter_truncate(iter
, maxsize
);
139 npages
= iov_iter_npages(iter
, INT_MAX
);
140 iov_iter_reexpand(iter
, orig_count
);
143 * __iter_get_bvecs() may populate only part of the array -- zero it
146 bv
= kvmalloc_array(npages
, sizeof(*bv
), GFP_KERNEL
| __GFP_ZERO
);
150 bytes
= __iter_get_bvecs(iter
, maxsize
, bv
);
153 * No pages were pinned -- just free the array.
164 static void put_bvecs(struct bio_vec
*bvecs
, int num_bvecs
, bool should_dirty
)
168 for (i
= 0; i
< num_bvecs
; i
++) {
169 if (bvecs
[i
].bv_page
) {
171 set_page_dirty_lock(bvecs
[i
].bv_page
);
172 put_page(bvecs
[i
].bv_page
);
179 * Prepare an open request. Preallocate ceph_cap to avoid an
180 * inopportune ENOMEM later.
182 static struct ceph_mds_request
*
183 prepare_open_request(struct super_block
*sb
, int flags
, int create_mode
)
185 struct ceph_mds_client
*mdsc
= ceph_sb_to_mdsc(sb
);
186 struct ceph_mds_request
*req
;
187 int want_auth
= USE_ANY_MDS
;
188 int op
= (flags
& O_CREAT
) ? CEPH_MDS_OP_CREATE
: CEPH_MDS_OP_OPEN
;
190 if (flags
& (O_WRONLY
|O_RDWR
|O_CREAT
|O_TRUNC
))
191 want_auth
= USE_AUTH_MDS
;
193 req
= ceph_mdsc_create_request(mdsc
, op
, want_auth
);
196 req
->r_fmode
= ceph_flags_to_mode(flags
);
197 req
->r_args
.open
.flags
= ceph_flags_sys2wire(flags
);
198 req
->r_args
.open
.mode
= cpu_to_le32(create_mode
);
203 static int ceph_init_file_info(struct inode
*inode
, struct file
*file
,
204 int fmode
, bool isdir
)
206 struct ceph_inode_info
*ci
= ceph_inode(inode
);
207 struct ceph_file_info
*fi
;
209 dout("%s %p %p 0%o (%s)\n", __func__
, inode
, file
,
210 inode
->i_mode
, isdir
? "dir" : "regular");
211 BUG_ON(inode
->i_fop
->release
!= ceph_release
);
214 struct ceph_dir_file_info
*dfi
=
215 kmem_cache_zalloc(ceph_dir_file_cachep
, GFP_KERNEL
);
219 file
->private_data
= dfi
;
220 fi
= &dfi
->file_info
;
221 dfi
->next_offset
= 2;
222 dfi
->readdir_cache_idx
= -1;
224 fi
= kmem_cache_zalloc(ceph_file_cachep
, GFP_KERNEL
);
228 file
->private_data
= fi
;
231 ceph_get_fmode(ci
, fmode
, 1);
234 spin_lock_init(&fi
->rw_contexts_lock
);
235 INIT_LIST_HEAD(&fi
->rw_contexts
);
236 fi
->filp_gen
= READ_ONCE(ceph_inode_to_client(inode
)->filp_gen
);
242 * initialize private struct file data.
243 * if we fail, clean up by dropping fmode reference on the ceph_inode
245 static int ceph_init_file(struct inode
*inode
, struct file
*file
, int fmode
)
249 switch (inode
->i_mode
& S_IFMT
) {
251 ceph_fscache_register_inode_cookie(inode
);
252 ceph_fscache_file_set_cookie(inode
, file
);
255 ret
= ceph_init_file_info(inode
, file
, fmode
,
256 S_ISDIR(inode
->i_mode
));
260 dout("init_file %p %p 0%o (symlink)\n", inode
, file
,
265 dout("init_file %p %p 0%o (special)\n", inode
, file
,
268 * we need to drop the open ref now, since we don't
269 * have .release set to ceph_release.
271 BUG_ON(inode
->i_fop
->release
== ceph_release
);
273 /* call the proper open fop */
274 ret
= inode
->i_fop
->open(inode
, file
);
280 * try renew caps after session gets killed.
282 int ceph_renew_caps(struct inode
*inode
, int fmode
)
284 struct ceph_mds_client
*mdsc
= ceph_sb_to_mdsc(inode
->i_sb
);
285 struct ceph_inode_info
*ci
= ceph_inode(inode
);
286 struct ceph_mds_request
*req
;
287 int err
, flags
, wanted
;
289 spin_lock(&ci
->i_ceph_lock
);
290 __ceph_touch_fmode(ci
, mdsc
, fmode
);
291 wanted
= __ceph_caps_file_wanted(ci
);
292 if (__ceph_is_any_real_caps(ci
) &&
293 (!(wanted
& CEPH_CAP_ANY_WR
) || ci
->i_auth_cap
)) {
294 int issued
= __ceph_caps_issued(ci
, NULL
);
295 spin_unlock(&ci
->i_ceph_lock
);
296 dout("renew caps %p want %s issued %s updating mds_wanted\n",
297 inode
, ceph_cap_string(wanted
), ceph_cap_string(issued
));
298 ceph_check_caps(ci
, 0, NULL
);
301 spin_unlock(&ci
->i_ceph_lock
);
304 if ((wanted
& CEPH_CAP_FILE_RD
) && (wanted
& CEPH_CAP_FILE_WR
))
306 else if (wanted
& CEPH_CAP_FILE_RD
)
308 else if (wanted
& CEPH_CAP_FILE_WR
)
311 if (wanted
& CEPH_CAP_FILE_LAZYIO
)
315 req
= prepare_open_request(inode
->i_sb
, flags
, 0);
321 req
->r_inode
= inode
;
325 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
326 ceph_mdsc_put_request(req
);
328 dout("renew caps %p open result=%d\n", inode
, err
);
329 return err
< 0 ? err
: 0;
333 * If we already have the requisite capabilities, we can satisfy
334 * the open request locally (no need to request new caps from the
335 * MDS). We do, however, need to inform the MDS (asynchronously)
336 * if our wanted caps set expands.
338 int ceph_open(struct inode
*inode
, struct file
*file
)
340 struct ceph_inode_info
*ci
= ceph_inode(inode
);
341 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
342 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
343 struct ceph_mds_request
*req
;
344 struct ceph_file_info
*fi
= file
->private_data
;
346 int flags
, fmode
, wanted
;
349 dout("open file %p is already opened\n", file
);
353 /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */
354 flags
= file
->f_flags
& ~(O_CREAT
|O_EXCL
);
355 if (S_ISDIR(inode
->i_mode
))
356 flags
= O_DIRECTORY
; /* mds likes to know */
358 dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode
,
359 ceph_vinop(inode
), file
, flags
, file
->f_flags
);
360 fmode
= ceph_flags_to_mode(flags
);
361 wanted
= ceph_caps_for_mode(fmode
);
363 /* snapped files are read-only */
364 if (ceph_snap(inode
) != CEPH_NOSNAP
&& (file
->f_mode
& FMODE_WRITE
))
367 /* trivially open snapdir */
368 if (ceph_snap(inode
) == CEPH_SNAPDIR
) {
369 return ceph_init_file(inode
, file
, fmode
);
373 * No need to block if we have caps on the auth MDS (for
374 * write) or any MDS (for read). Update wanted set
377 spin_lock(&ci
->i_ceph_lock
);
378 if (__ceph_is_any_real_caps(ci
) &&
379 (((fmode
& CEPH_FILE_MODE_WR
) == 0) || ci
->i_auth_cap
)) {
380 int mds_wanted
= __ceph_caps_mds_wanted(ci
, true);
381 int issued
= __ceph_caps_issued(ci
, NULL
);
383 dout("open %p fmode %d want %s issued %s using existing\n",
384 inode
, fmode
, ceph_cap_string(wanted
),
385 ceph_cap_string(issued
));
386 __ceph_touch_fmode(ci
, mdsc
, fmode
);
387 spin_unlock(&ci
->i_ceph_lock
);
390 if ((issued
& wanted
) != wanted
&&
391 (mds_wanted
& wanted
) != wanted
&&
392 ceph_snap(inode
) != CEPH_SNAPDIR
)
393 ceph_check_caps(ci
, 0, NULL
);
395 return ceph_init_file(inode
, file
, fmode
);
396 } else if (ceph_snap(inode
) != CEPH_NOSNAP
&&
397 (ci
->i_snap_caps
& wanted
) == wanted
) {
398 __ceph_touch_fmode(ci
, mdsc
, fmode
);
399 spin_unlock(&ci
->i_ceph_lock
);
400 return ceph_init_file(inode
, file
, fmode
);
403 spin_unlock(&ci
->i_ceph_lock
);
405 dout("open fmode %d wants %s\n", fmode
, ceph_cap_string(wanted
));
406 req
= prepare_open_request(inode
->i_sb
, flags
, 0);
411 req
->r_inode
= inode
;
415 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
417 err
= ceph_init_file(inode
, file
, req
->r_fmode
);
418 ceph_mdsc_put_request(req
);
419 dout("open result=%d on %llx.%llx\n", err
, ceph_vinop(inode
));
424 /* Clone the layout from a synchronous create, if the dir now has Dc caps */
426 cache_file_layout(struct inode
*dst
, struct inode
*src
)
428 struct ceph_inode_info
*cdst
= ceph_inode(dst
);
429 struct ceph_inode_info
*csrc
= ceph_inode(src
);
431 spin_lock(&cdst
->i_ceph_lock
);
432 if ((__ceph_caps_issued(cdst
, NULL
) & CEPH_CAP_DIR_CREATE
) &&
433 !ceph_file_layout_is_valid(&cdst
->i_cached_layout
)) {
434 memcpy(&cdst
->i_cached_layout
, &csrc
->i_layout
,
435 sizeof(cdst
->i_cached_layout
));
436 rcu_assign_pointer(cdst
->i_cached_layout
.pool_ns
,
437 ceph_try_get_string(csrc
->i_layout
.pool_ns
));
439 spin_unlock(&cdst
->i_ceph_lock
);
443 * Try to set up an async create. We need caps, a file layout, and inode number,
444 * and either a lease on the dentry or complete dir info. If any of those
445 * criteria are not satisfied, then return false and the caller can go
448 static int try_prep_async_create(struct inode
*dir
, struct dentry
*dentry
,
449 struct ceph_file_layout
*lo
, u64
*pino
)
451 struct ceph_inode_info
*ci
= ceph_inode(dir
);
452 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
453 int got
= 0, want
= CEPH_CAP_FILE_EXCL
| CEPH_CAP_DIR_CREATE
;
456 spin_lock(&ci
->i_ceph_lock
);
457 /* No auth cap means no chance for Dc caps */
461 /* Any delegated inos? */
462 if (xa_empty(&ci
->i_auth_cap
->session
->s_delegated_inos
))
465 if (!ceph_file_layout_is_valid(&ci
->i_cached_layout
))
468 if ((__ceph_caps_issued(ci
, NULL
) & want
) != want
)
471 if (d_in_lookup(dentry
)) {
472 if (!__ceph_dir_is_complete(ci
))
474 spin_lock(&dentry
->d_lock
);
475 di
->lease_shared_gen
= atomic_read(&ci
->i_shared_gen
);
476 spin_unlock(&dentry
->d_lock
);
477 } else if (atomic_read(&ci
->i_shared_gen
) !=
478 READ_ONCE(di
->lease_shared_gen
)) {
482 ino
= ceph_get_deleg_ino(ci
->i_auth_cap
->session
);
487 ceph_take_cap_refs(ci
, want
, false);
488 memcpy(lo
, &ci
->i_cached_layout
, sizeof(*lo
));
489 rcu_assign_pointer(lo
->pool_ns
,
490 ceph_try_get_string(ci
->i_cached_layout
.pool_ns
));
493 spin_unlock(&ci
->i_ceph_lock
);
497 static void restore_deleg_ino(struct inode
*dir
, u64 ino
)
499 struct ceph_inode_info
*ci
= ceph_inode(dir
);
500 struct ceph_mds_session
*s
= NULL
;
502 spin_lock(&ci
->i_ceph_lock
);
504 s
= ceph_get_mds_session(ci
->i_auth_cap
->session
);
505 spin_unlock(&ci
->i_ceph_lock
);
507 int err
= ceph_restore_deleg_ino(s
, ino
);
509 pr_warn("ceph: unable to restore delegated ino 0x%llx to session: %d\n",
511 ceph_put_mds_session(s
);
515 static void ceph_async_create_cb(struct ceph_mds_client
*mdsc
,
516 struct ceph_mds_request
*req
)
518 int result
= req
->r_err
? req
->r_err
:
519 le32_to_cpu(req
->r_reply_info
.head
->result
);
521 if (result
== -EJUKEBOX
)
524 mapping_set_error(req
->r_parent
->i_mapping
, result
);
527 struct dentry
*dentry
= req
->r_dentry
;
530 char *path
= ceph_mdsc_build_path(req
->r_dentry
, &pathlen
,
533 ceph_dir_clear_complete(req
->r_parent
);
534 if (!d_unhashed(dentry
))
537 /* FIXME: start returning I/O errors on all accesses? */
538 pr_warn("ceph: async create failure path=(%llx)%s result=%d!\n",
539 base
, IS_ERR(path
) ? "<<bad>>" : path
, result
);
540 ceph_mdsc_free_path(path
, pathlen
);
543 if (req
->r_target_inode
) {
544 struct ceph_inode_info
*ci
= ceph_inode(req
->r_target_inode
);
545 u64 ino
= ceph_vino(req
->r_target_inode
).ino
;
547 if (req
->r_deleg_ino
!= ino
)
548 pr_warn("%s: inode number mismatch! err=%d deleg_ino=0x%llx target=0x%llx\n",
549 __func__
, req
->r_err
, req
->r_deleg_ino
, ino
);
550 mapping_set_error(req
->r_target_inode
->i_mapping
, result
);
552 spin_lock(&ci
->i_ceph_lock
);
553 if (ci
->i_ceph_flags
& CEPH_I_ASYNC_CREATE
) {
554 ci
->i_ceph_flags
&= ~CEPH_I_ASYNC_CREATE
;
555 wake_up_bit(&ci
->i_ceph_flags
, CEPH_ASYNC_CREATE_BIT
);
557 ceph_kick_flushing_inode_caps(req
->r_session
, ci
);
558 spin_unlock(&ci
->i_ceph_lock
);
560 pr_warn("%s: no req->r_target_inode for 0x%llx\n", __func__
,
564 ceph_mdsc_release_dir_caps(req
);
567 static int ceph_finish_async_create(struct inode
*dir
, struct dentry
*dentry
,
568 struct file
*file
, umode_t mode
,
569 struct ceph_mds_request
*req
,
570 struct ceph_acl_sec_ctx
*as_ctx
,
571 struct ceph_file_layout
*lo
)
575 struct ceph_mds_reply_inode in
= { };
576 struct ceph_mds_reply_info_in iinfo
= { .in
= &in
};
577 struct ceph_inode_info
*ci
= ceph_inode(dir
);
579 struct timespec64 now
;
580 struct ceph_mds_client
*mdsc
= ceph_sb_to_mdsc(dir
->i_sb
);
581 struct ceph_vino vino
= { .ino
= req
->r_deleg_ino
,
582 .snap
= CEPH_NOSNAP
};
584 ktime_get_real_ts64(&now
);
586 inode
= ceph_get_inode(dentry
->d_sb
, vino
);
588 return PTR_ERR(inode
);
590 iinfo
.inline_version
= CEPH_INLINE_NONE
;
591 iinfo
.change_attr
= 1;
592 ceph_encode_timespec64(&iinfo
.btime
, &now
);
594 iinfo
.xattr_len
= ARRAY_SIZE(xattr_buf
);
595 iinfo
.xattr_data
= xattr_buf
;
596 memset(iinfo
.xattr_data
, 0, iinfo
.xattr_len
);
598 in
.ino
= cpu_to_le64(vino
.ino
);
599 in
.snapid
= cpu_to_le64(CEPH_NOSNAP
);
600 in
.version
= cpu_to_le64(1); // ???
601 in
.cap
.caps
= in
.cap
.wanted
= cpu_to_le32(CEPH_CAP_ALL_FILE
);
602 in
.cap
.cap_id
= cpu_to_le64(1);
603 in
.cap
.realm
= cpu_to_le64(ci
->i_snap_realm
->ino
);
604 in
.cap
.flags
= CEPH_CAP_FLAG_AUTH
;
605 in
.ctime
= in
.mtime
= in
.atime
= iinfo
.btime
;
606 in
.truncate_seq
= cpu_to_le32(1);
607 in
.truncate_size
= cpu_to_le64(-1ULL);
608 in
.xattr_version
= cpu_to_le64(1);
609 in
.uid
= cpu_to_le32(from_kuid(&init_user_ns
, current_fsuid()));
610 if (dir
->i_mode
& S_ISGID
) {
611 in
.gid
= cpu_to_le32(from_kgid(&init_user_ns
, dir
->i_gid
));
613 /* Directories always inherit the setgid bit. */
616 else if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
) &&
617 !in_group_p(dir
->i_gid
) &&
618 !capable_wrt_inode_uidgid(&init_user_ns
, dir
, CAP_FSETID
))
621 in
.gid
= cpu_to_le32(from_kgid(&init_user_ns
, current_fsgid()));
623 in
.mode
= cpu_to_le32((u32
)mode
);
625 in
.nlink
= cpu_to_le32(1);
626 in
.max_size
= cpu_to_le64(lo
->stripe_unit
);
628 ceph_file_layout_to_legacy(lo
, &in
.layout
);
630 down_read(&mdsc
->snap_rwsem
);
631 ret
= ceph_fill_inode(inode
, NULL
, &iinfo
, NULL
, req
->r_session
,
633 up_read(&mdsc
->snap_rwsem
);
635 dout("%s failed to fill inode: %d\n", __func__
, ret
);
636 ceph_dir_clear_complete(dir
);
637 if (!d_unhashed(dentry
))
639 if (inode
->i_state
& I_NEW
)
640 discard_new_inode(inode
);
644 dout("%s d_adding new inode 0x%llx to 0x%llx/%s\n", __func__
,
645 vino
.ino
, ceph_ino(dir
), dentry
->d_name
.name
);
646 ceph_dir_clear_ordered(dir
);
647 ceph_init_inode_acls(inode
, as_ctx
);
648 if (inode
->i_state
& I_NEW
) {
650 * If it's not I_NEW, then someone created this before
651 * we got here. Assume the server is aware of it at
652 * that point and don't worry about setting
653 * CEPH_I_ASYNC_CREATE.
655 ceph_inode(inode
)->i_ceph_flags
= CEPH_I_ASYNC_CREATE
;
656 unlock_new_inode(inode
);
658 if (d_in_lookup(dentry
) || d_really_is_negative(dentry
)) {
659 if (!d_unhashed(dentry
))
661 dn
= d_splice_alias(inode
, dentry
);
662 WARN_ON_ONCE(dn
&& dn
!= dentry
);
664 file
->f_mode
|= FMODE_CREATED
;
665 ret
= finish_open(file
, dentry
, ceph_open
);
671 * Do a lookup + open with a single request. If we get a non-existent
672 * file or symlink, return 1 so the VFS can retry.
674 int ceph_atomic_open(struct inode
*dir
, struct dentry
*dentry
,
675 struct file
*file
, unsigned flags
, umode_t mode
)
677 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
678 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
679 struct ceph_mds_request
*req
;
681 struct ceph_acl_sec_ctx as_ctx
= {};
682 bool try_async
= ceph_test_mount_opt(fsc
, ASYNC_DIROPS
);
686 dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n",
688 d_unhashed(dentry
) ? "unhashed" : "hashed", flags
, mode
);
690 if (dentry
->d_name
.len
> NAME_MAX
)
691 return -ENAMETOOLONG
;
693 if (flags
& O_CREAT
) {
694 if (ceph_quota_is_max_files_exceeded(dir
))
696 err
= ceph_pre_init_acls(dir
, &mode
, &as_ctx
);
699 err
= ceph_security_init_secctx(dentry
, mode
, &as_ctx
);
702 } else if (!d_in_lookup(dentry
)) {
703 /* If it's not being looked up, it's negative */
708 req
= prepare_open_request(dir
->i_sb
, flags
, mode
);
713 req
->r_dentry
= dget(dentry
);
715 mask
= CEPH_STAT_CAP_INODE
| CEPH_CAP_AUTH_SHARED
;
716 if (ceph_security_xattr_wanted(dir
))
717 mask
|= CEPH_CAP_XATTR_SHARED
;
718 req
->r_args
.open
.mask
= cpu_to_le32(mask
);
722 if (flags
& O_CREAT
) {
723 struct ceph_file_layout lo
;
725 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
| CEPH_CAP_AUTH_EXCL
;
726 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
727 if (as_ctx
.pagelist
) {
728 req
->r_pagelist
= as_ctx
.pagelist
;
729 as_ctx
.pagelist
= NULL
;
733 try_prep_async_create(dir
, dentry
, &lo
,
734 &req
->r_deleg_ino
))) {
735 set_bit(CEPH_MDS_R_ASYNC
, &req
->r_req_flags
);
736 req
->r_args
.open
.flags
|= cpu_to_le32(CEPH_O_EXCL
);
737 req
->r_callback
= ceph_async_create_cb
;
738 err
= ceph_mdsc_submit_request(mdsc
, dir
, req
);
740 err
= ceph_finish_async_create(dir
, dentry
,
743 } else if (err
== -EJUKEBOX
) {
744 restore_deleg_ino(dir
, req
->r_deleg_ino
);
745 ceph_mdsc_put_request(req
);
753 set_bit(CEPH_MDS_R_PARENT_LOCKED
, &req
->r_req_flags
);
754 err
= ceph_mdsc_do_request(mdsc
,
755 (flags
& (O_CREAT
|O_TRUNC
)) ? dir
: NULL
,
757 if (err
== -ENOENT
) {
758 dentry
= ceph_handle_snapdir(req
, dentry
);
759 if (IS_ERR(dentry
)) {
760 err
= PTR_ERR(dentry
);
766 if (!err
&& (flags
& O_CREAT
) && !req
->r_reply_info
.head
->is_dentry
)
767 err
= ceph_handle_notrace_create(dir
, dentry
);
769 if (d_in_lookup(dentry
)) {
770 dn
= ceph_finish_lookup(req
, dentry
, err
);
774 /* we were given a hashed negative dentry */
779 if (dn
|| d_really_is_negative(dentry
) || d_is_symlink(dentry
)) {
780 /* make vfs retry on splice, ENOENT, or symlink */
781 dout("atomic_open finish_no_open on dn %p\n", dn
);
782 err
= finish_no_open(file
, dn
);
784 dout("atomic_open finish_open on dn %p\n", dn
);
785 if (req
->r_op
== CEPH_MDS_OP_CREATE
&& req
->r_reply_info
.has_create_ino
) {
786 struct inode
*newino
= d_inode(dentry
);
788 cache_file_layout(dir
, newino
);
789 ceph_init_inode_acls(newino
, &as_ctx
);
790 file
->f_mode
|= FMODE_CREATED
;
792 err
= finish_open(file
, dentry
, ceph_open
);
795 ceph_mdsc_put_request(req
);
797 ceph_release_acl_sec_ctx(&as_ctx
);
798 dout("atomic_open result=%d\n", err
);
802 int ceph_release(struct inode
*inode
, struct file
*file
)
804 struct ceph_inode_info
*ci
= ceph_inode(inode
);
806 if (S_ISDIR(inode
->i_mode
)) {
807 struct ceph_dir_file_info
*dfi
= file
->private_data
;
808 dout("release inode %p dir file %p\n", inode
, file
);
809 WARN_ON(!list_empty(&dfi
->file_info
.rw_contexts
));
811 ceph_put_fmode(ci
, dfi
->file_info
.fmode
, 1);
813 if (dfi
->last_readdir
)
814 ceph_mdsc_put_request(dfi
->last_readdir
);
815 kfree(dfi
->last_name
);
816 kfree(dfi
->dir_info
);
817 kmem_cache_free(ceph_dir_file_cachep
, dfi
);
819 struct ceph_file_info
*fi
= file
->private_data
;
820 dout("release inode %p regular file %p\n", inode
, file
);
821 WARN_ON(!list_empty(&fi
->rw_contexts
));
823 ceph_put_fmode(ci
, fi
->fmode
, 1);
825 kmem_cache_free(ceph_file_cachep
, fi
);
828 /* wake up anyone waiting for caps on this inode */
829 wake_up_all(&ci
->i_cap_wq
);
840 * Completely synchronous read and write methods. Direct from __user
841 * buffer to osd, or directly to user pages (if O_DIRECT).
843 * If the read spans object boundary, just do multiple reads. (That's not
844 * atomic, but good enough for now.)
846 * If we get a short result from the OSD, check against i_size; we need to
847 * only return a short read to the caller if we hit EOF.
849 static ssize_t
ceph_sync_read(struct kiocb
*iocb
, struct iov_iter
*to
,
852 struct file
*file
= iocb
->ki_filp
;
853 struct inode
*inode
= file_inode(file
);
854 struct ceph_inode_info
*ci
= ceph_inode(inode
);
855 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
856 struct ceph_osd_client
*osdc
= &fsc
->client
->osdc
;
858 u64 off
= iocb
->ki_pos
;
859 u64 len
= iov_iter_count(to
);
861 dout("sync_read on file %p %llu~%u %s\n", file
, off
, (unsigned)len
,
862 (file
->f_flags
& O_DIRECT
) ? "O_DIRECT" : "");
867 * flush any page cache pages in this range. this
868 * will make concurrent normal and sync io slow,
869 * but it will at least behave sensibly when they are
872 ret
= filemap_write_and_wait_range(inode
->i_mapping
,
878 while ((len
= iov_iter_count(to
)) > 0) {
879 struct ceph_osd_request
*req
;
888 req
= ceph_osdc_new_request(osdc
, &ci
->i_layout
,
889 ci
->i_vino
, off
, &len
, 0, 1,
890 CEPH_OSD_OP_READ
, CEPH_OSD_FLAG_READ
,
891 NULL
, ci
->i_truncate_seq
,
892 ci
->i_truncate_size
, false);
898 more
= len
< iov_iter_count(to
);
900 num_pages
= calc_pages_for(off
, len
);
901 page_off
= off
& ~PAGE_MASK
;
902 pages
= ceph_alloc_page_vector(num_pages
, GFP_KERNEL
);
904 ceph_osdc_put_request(req
);
905 ret
= PTR_ERR(pages
);
909 osd_req_op_extent_osd_data_pages(req
, 0, pages
, len
, page_off
,
911 ret
= ceph_osdc_start_request(osdc
, req
, false);
913 ret
= ceph_osdc_wait_request(osdc
, req
);
915 ceph_update_read_metrics(&fsc
->mdsc
->metric
,
916 req
->r_start_latency
,
920 ceph_osdc_put_request(req
);
922 i_size
= i_size_read(inode
);
923 dout("sync_read %llu~%llu got %zd i_size %llu%s\n",
924 off
, len
, ret
, i_size
, (more
? " MORE" : ""));
928 if (ret
>= 0 && ret
< len
&& (off
+ ret
< i_size
)) {
929 int zlen
= min(len
- ret
, i_size
- off
- ret
);
930 int zoff
= page_off
+ ret
;
931 dout("sync_read zero gap %llu~%llu\n",
932 off
+ ret
, off
+ ret
+ zlen
);
933 ceph_zero_page_vector_range(zoff
, zlen
, pages
);
938 left
= ret
> 0 ? ret
: 0;
941 page_off
= off
& ~PAGE_MASK
;
942 len
= min_t(size_t, left
, PAGE_SIZE
- page_off
);
943 SetPageUptodate(pages
[idx
]);
944 copied
= copy_page_to_iter(pages
[idx
++],
953 ceph_release_page_vector(pages
, num_pages
);
956 if (ret
== -EBLOCKLISTED
)
957 fsc
->blocklisted
= true;
961 if (off
>= i_size
|| !more
)
965 if (off
> iocb
->ki_pos
) {
967 iov_iter_count(to
) > 0 && off
>= i_size_read(inode
))
968 *retry_op
= CHECK_EOF
;
969 ret
= off
- iocb
->ki_pos
;
973 dout("sync_read result %zd retry_op %d\n", ret
, *retry_op
);
977 struct ceph_aio_request
{
983 struct list_head osd_reqs
;
985 atomic_t pending_reqs
;
986 struct timespec64 mtime
;
987 struct ceph_cap_flush
*prealloc_cf
;
990 struct ceph_aio_work
{
991 struct work_struct work
;
992 struct ceph_osd_request
*req
;
995 static void ceph_aio_retry_work(struct work_struct
*work
);
997 static void ceph_aio_complete(struct inode
*inode
,
998 struct ceph_aio_request
*aio_req
)
1000 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1003 if (!atomic_dec_and_test(&aio_req
->pending_reqs
))
1006 if (aio_req
->iocb
->ki_flags
& IOCB_DIRECT
)
1007 inode_dio_end(inode
);
1009 ret
= aio_req
->error
;
1011 ret
= aio_req
->total_len
;
1013 dout("ceph_aio_complete %p rc %d\n", inode
, ret
);
1015 if (ret
>= 0 && aio_req
->write
) {
1018 loff_t endoff
= aio_req
->iocb
->ki_pos
+ aio_req
->total_len
;
1019 if (endoff
> i_size_read(inode
)) {
1020 if (ceph_inode_set_size(inode
, endoff
))
1021 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
1024 spin_lock(&ci
->i_ceph_lock
);
1025 ci
->i_inline_version
= CEPH_INLINE_NONE
;
1026 dirty
= __ceph_mark_dirty_caps(ci
, CEPH_CAP_FILE_WR
,
1027 &aio_req
->prealloc_cf
);
1028 spin_unlock(&ci
->i_ceph_lock
);
1030 __mark_inode_dirty(inode
, dirty
);
1034 ceph_put_cap_refs(ci
, (aio_req
->write
? CEPH_CAP_FILE_WR
:
1037 aio_req
->iocb
->ki_complete(aio_req
->iocb
, ret
, 0);
1039 ceph_free_cap_flush(aio_req
->prealloc_cf
);
1043 static void ceph_aio_complete_req(struct ceph_osd_request
*req
)
1045 int rc
= req
->r_result
;
1046 struct inode
*inode
= req
->r_inode
;
1047 struct ceph_aio_request
*aio_req
= req
->r_priv
;
1048 struct ceph_osd_data
*osd_data
= osd_req_op_extent_osd_data(req
, 0);
1049 struct ceph_client_metric
*metric
= &ceph_sb_to_mdsc(inode
->i_sb
)->metric
;
1050 unsigned int len
= osd_data
->bvec_pos
.iter
.bi_size
;
1052 BUG_ON(osd_data
->type
!= CEPH_OSD_DATA_TYPE_BVECS
);
1053 BUG_ON(!osd_data
->num_bvecs
);
1055 dout("ceph_aio_complete_req %p rc %d bytes %u\n", inode
, rc
, len
);
1057 if (rc
== -EOLDSNAPC
) {
1058 struct ceph_aio_work
*aio_work
;
1059 BUG_ON(!aio_req
->write
);
1061 aio_work
= kmalloc(sizeof(*aio_work
), GFP_NOFS
);
1063 INIT_WORK(&aio_work
->work
, ceph_aio_retry_work
);
1064 aio_work
->req
= req
;
1065 queue_work(ceph_inode_to_client(inode
)->inode_wq
,
1070 } else if (!aio_req
->write
) {
1073 if (rc
>= 0 && len
> rc
) {
1075 int zlen
= len
- rc
;
1078 * If read is satisfied by single OSD request,
1079 * it can pass EOF. Otherwise read is within
1082 if (aio_req
->num_reqs
== 1) {
1083 loff_t i_size
= i_size_read(inode
);
1084 loff_t endoff
= aio_req
->iocb
->ki_pos
+ rc
;
1085 if (endoff
< i_size
)
1086 zlen
= min_t(size_t, zlen
,
1088 aio_req
->total_len
= rc
+ zlen
;
1091 iov_iter_bvec(&i
, READ
, osd_data
->bvec_pos
.bvecs
,
1092 osd_data
->num_bvecs
, len
);
1093 iov_iter_advance(&i
, rc
);
1094 iov_iter_zero(zlen
, &i
);
1098 /* r_start_latency == 0 means the request was not submitted */
1099 if (req
->r_start_latency
) {
1101 ceph_update_write_metrics(metric
, req
->r_start_latency
,
1102 req
->r_end_latency
, len
, rc
);
1104 ceph_update_read_metrics(metric
, req
->r_start_latency
,
1105 req
->r_end_latency
, len
, rc
);
1108 put_bvecs(osd_data
->bvec_pos
.bvecs
, osd_data
->num_bvecs
,
1109 aio_req
->should_dirty
);
1110 ceph_osdc_put_request(req
);
1113 cmpxchg(&aio_req
->error
, 0, rc
);
1115 ceph_aio_complete(inode
, aio_req
);
1119 static void ceph_aio_retry_work(struct work_struct
*work
)
1121 struct ceph_aio_work
*aio_work
=
1122 container_of(work
, struct ceph_aio_work
, work
);
1123 struct ceph_osd_request
*orig_req
= aio_work
->req
;
1124 struct ceph_aio_request
*aio_req
= orig_req
->r_priv
;
1125 struct inode
*inode
= orig_req
->r_inode
;
1126 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1127 struct ceph_snap_context
*snapc
;
1128 struct ceph_osd_request
*req
;
1131 spin_lock(&ci
->i_ceph_lock
);
1132 if (__ceph_have_pending_cap_snap(ci
)) {
1133 struct ceph_cap_snap
*capsnap
=
1134 list_last_entry(&ci
->i_cap_snaps
,
1135 struct ceph_cap_snap
,
1137 snapc
= ceph_get_snap_context(capsnap
->context
);
1139 BUG_ON(!ci
->i_head_snapc
);
1140 snapc
= ceph_get_snap_context(ci
->i_head_snapc
);
1142 spin_unlock(&ci
->i_ceph_lock
);
1144 req
= ceph_osdc_alloc_request(orig_req
->r_osdc
, snapc
, 1,
1152 req
->r_flags
= /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE
;
1153 ceph_oloc_copy(&req
->r_base_oloc
, &orig_req
->r_base_oloc
);
1154 ceph_oid_copy(&req
->r_base_oid
, &orig_req
->r_base_oid
);
1156 req
->r_ops
[0] = orig_req
->r_ops
[0];
1158 req
->r_mtime
= aio_req
->mtime
;
1159 req
->r_data_offset
= req
->r_ops
[0].extent
.offset
;
1161 ret
= ceph_osdc_alloc_messages(req
, GFP_NOFS
);
1163 ceph_osdc_put_request(req
);
1168 ceph_osdc_put_request(orig_req
);
1170 req
->r_callback
= ceph_aio_complete_req
;
1171 req
->r_inode
= inode
;
1172 req
->r_priv
= aio_req
;
1174 ret
= ceph_osdc_start_request(req
->r_osdc
, req
, false);
1177 req
->r_result
= ret
;
1178 ceph_aio_complete_req(req
);
1181 ceph_put_snap_context(snapc
);
1186 ceph_direct_read_write(struct kiocb
*iocb
, struct iov_iter
*iter
,
1187 struct ceph_snap_context
*snapc
,
1188 struct ceph_cap_flush
**pcf
)
1190 struct file
*file
= iocb
->ki_filp
;
1191 struct inode
*inode
= file_inode(file
);
1192 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1193 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
1194 struct ceph_client_metric
*metric
= &fsc
->mdsc
->metric
;
1195 struct ceph_vino vino
;
1196 struct ceph_osd_request
*req
;
1197 struct bio_vec
*bvecs
;
1198 struct ceph_aio_request
*aio_req
= NULL
;
1202 struct timespec64 mtime
= current_time(inode
);
1203 size_t count
= iov_iter_count(iter
);
1204 loff_t pos
= iocb
->ki_pos
;
1205 bool write
= iov_iter_rw(iter
) == WRITE
;
1206 bool should_dirty
= !write
&& iter_is_iovec(iter
);
1208 if (write
&& ceph_snap(file_inode(file
)) != CEPH_NOSNAP
)
1211 dout("sync_direct_%s on file %p %lld~%u snapc %p seq %lld\n",
1212 (write
? "write" : "read"), file
, pos
, (unsigned)count
,
1213 snapc
, snapc
? snapc
->seq
: 0);
1216 int ret2
= invalidate_inode_pages2_range(inode
->i_mapping
,
1218 (pos
+ count
- 1) >> PAGE_SHIFT
);
1220 dout("invalidate_inode_pages2_range returned %d\n", ret2
);
1222 flags
= /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE
;
1224 flags
= CEPH_OSD_FLAG_READ
;
1227 while (iov_iter_count(iter
) > 0) {
1228 u64 size
= iov_iter_count(iter
);
1232 size
= min_t(u64
, size
, fsc
->mount_options
->wsize
);
1234 size
= min_t(u64
, size
, fsc
->mount_options
->rsize
);
1236 vino
= ceph_vino(inode
);
1237 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
1238 vino
, pos
, &size
, 0,
1240 write
? CEPH_OSD_OP_WRITE
:
1244 ci
->i_truncate_size
,
1251 len
= iter_get_bvecs_alloc(iter
, size
, &bvecs
, &num_pages
);
1253 ceph_osdc_put_request(req
);
1258 osd_req_op_extent_update(req
, 0, len
);
1261 * To simplify error handling, allow AIO when IO within i_size
1262 * or IO can be satisfied by single OSD request.
1264 if (pos
== iocb
->ki_pos
&& !is_sync_kiocb(iocb
) &&
1265 (len
== count
|| pos
+ count
<= i_size_read(inode
))) {
1266 aio_req
= kzalloc(sizeof(*aio_req
), GFP_KERNEL
);
1268 aio_req
->iocb
= iocb
;
1269 aio_req
->write
= write
;
1270 aio_req
->should_dirty
= should_dirty
;
1271 INIT_LIST_HEAD(&aio_req
->osd_reqs
);
1273 aio_req
->mtime
= mtime
;
1274 swap(aio_req
->prealloc_cf
, *pcf
);
1282 * throw out any page cache pages in this range. this
1285 truncate_inode_pages_range(inode
->i_mapping
, pos
,
1286 PAGE_ALIGN(pos
+ len
) - 1);
1288 req
->r_mtime
= mtime
;
1291 osd_req_op_extent_osd_data_bvecs(req
, 0, bvecs
, num_pages
, len
);
1294 aio_req
->total_len
+= len
;
1295 aio_req
->num_reqs
++;
1296 atomic_inc(&aio_req
->pending_reqs
);
1298 req
->r_callback
= ceph_aio_complete_req
;
1299 req
->r_inode
= inode
;
1300 req
->r_priv
= aio_req
;
1301 list_add_tail(&req
->r_private_item
, &aio_req
->osd_reqs
);
1307 ret
= ceph_osdc_start_request(req
->r_osdc
, req
, false);
1309 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
1312 ceph_update_write_metrics(metric
, req
->r_start_latency
,
1313 req
->r_end_latency
, len
, ret
);
1315 ceph_update_read_metrics(metric
, req
->r_start_latency
,
1316 req
->r_end_latency
, len
, ret
);
1318 size
= i_size_read(inode
);
1322 if (ret
>= 0 && ret
< len
&& pos
+ ret
< size
) {
1324 int zlen
= min_t(size_t, len
- ret
,
1327 iov_iter_bvec(&i
, READ
, bvecs
, num_pages
, len
);
1328 iov_iter_advance(&i
, ret
);
1329 iov_iter_zero(zlen
, &i
);
1336 put_bvecs(bvecs
, num_pages
, should_dirty
);
1337 ceph_osdc_put_request(req
);
1342 if (!write
&& pos
>= size
)
1345 if (write
&& pos
> size
) {
1346 if (ceph_inode_set_size(inode
, pos
))
1347 ceph_check_caps(ceph_inode(inode
),
1348 CHECK_CAPS_AUTHONLY
,
1354 LIST_HEAD(osd_reqs
);
1356 if (aio_req
->num_reqs
== 0) {
1361 ceph_get_cap_refs(ci
, write
? CEPH_CAP_FILE_WR
:
1364 list_splice(&aio_req
->osd_reqs
, &osd_reqs
);
1365 inode_dio_begin(inode
);
1366 while (!list_empty(&osd_reqs
)) {
1367 req
= list_first_entry(&osd_reqs
,
1368 struct ceph_osd_request
,
1370 list_del_init(&req
->r_private_item
);
1372 ret
= ceph_osdc_start_request(req
->r_osdc
,
1375 req
->r_result
= ret
;
1376 ceph_aio_complete_req(req
);
1379 return -EIOCBQUEUED
;
1382 if (ret
!= -EOLDSNAPC
&& pos
> iocb
->ki_pos
) {
1383 ret
= pos
- iocb
->ki_pos
;
1390 * Synchronous write, straight from __user pointer or user pages.
1392 * If write spans object boundary, just do multiple writes. (For a
1393 * correct atomic write, we should e.g. take write locks on all
1394 * objects, rollback on failure, etc.)
1397 ceph_sync_write(struct kiocb
*iocb
, struct iov_iter
*from
, loff_t pos
,
1398 struct ceph_snap_context
*snapc
)
1400 struct file
*file
= iocb
->ki_filp
;
1401 struct inode
*inode
= file_inode(file
);
1402 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1403 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
1404 struct ceph_vino vino
;
1405 struct ceph_osd_request
*req
;
1406 struct page
**pages
;
1412 bool check_caps
= false;
1413 struct timespec64 mtime
= current_time(inode
);
1414 size_t count
= iov_iter_count(from
);
1416 if (ceph_snap(file_inode(file
)) != CEPH_NOSNAP
)
1419 dout("sync_write on file %p %lld~%u snapc %p seq %lld\n",
1420 file
, pos
, (unsigned)count
, snapc
, snapc
->seq
);
1422 ret
= filemap_write_and_wait_range(inode
->i_mapping
,
1423 pos
, pos
+ count
- 1);
1427 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
1429 (pos
+ count
- 1) >> PAGE_SHIFT
);
1431 dout("invalidate_inode_pages2_range returned %d\n", ret
);
1433 flags
= /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE
;
1435 while ((len
= iov_iter_count(from
)) > 0) {
1439 vino
= ceph_vino(inode
);
1440 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
1441 vino
, pos
, &len
, 0, 1,
1442 CEPH_OSD_OP_WRITE
, flags
, snapc
,
1444 ci
->i_truncate_size
,
1452 * write from beginning of first page,
1453 * regardless of io alignment
1455 num_pages
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1457 pages
= ceph_alloc_page_vector(num_pages
, GFP_KERNEL
);
1458 if (IS_ERR(pages
)) {
1459 ret
= PTR_ERR(pages
);
1464 for (n
= 0; n
< num_pages
; n
++) {
1465 size_t plen
= min_t(size_t, left
, PAGE_SIZE
);
1466 ret
= copy_page_from_iter(pages
[n
], 0, plen
, from
);
1475 ceph_release_page_vector(pages
, num_pages
);
1479 req
->r_inode
= inode
;
1481 osd_req_op_extent_osd_data_pages(req
, 0, pages
, len
, 0,
1484 req
->r_mtime
= mtime
;
1485 ret
= ceph_osdc_start_request(&fsc
->client
->osdc
, req
, false);
1487 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
1489 ceph_update_write_metrics(&fsc
->mdsc
->metric
, req
->r_start_latency
,
1490 req
->r_end_latency
, len
, ret
);
1492 ceph_osdc_put_request(req
);
1494 ceph_set_error_write(ci
);
1498 ceph_clear_error_write(ci
);
1501 if (pos
> i_size_read(inode
)) {
1502 check_caps
= ceph_inode_set_size(inode
, pos
);
1504 ceph_check_caps(ceph_inode(inode
),
1505 CHECK_CAPS_AUTHONLY
,
1511 if (ret
!= -EOLDSNAPC
&& written
> 0) {
1519 * Wrap generic_file_aio_read with checks for cap bits on the inode.
1520 * Atomically grab references, so that those bits are not released
1521 * back to the MDS mid-read.
1523 * Hmm, the sync read case isn't actually async... should it be?
1525 static ssize_t
ceph_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1527 struct file
*filp
= iocb
->ki_filp
;
1528 struct ceph_file_info
*fi
= filp
->private_data
;
1529 size_t len
= iov_iter_count(to
);
1530 struct inode
*inode
= file_inode(filp
);
1531 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1532 bool direct_lock
= iocb
->ki_flags
& IOCB_DIRECT
;
1535 int retry_op
= 0, read
= 0;
1538 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
1539 inode
, ceph_vinop(inode
), iocb
->ki_pos
, (unsigned)len
, inode
);
1542 ceph_start_io_direct(inode
);
1544 ceph_start_io_read(inode
);
1546 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
1547 want
= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_LAZYIO
;
1549 want
= CEPH_CAP_FILE_CACHE
;
1550 ret
= ceph_get_caps(filp
, CEPH_CAP_FILE_RD
, want
, -1, &got
);
1552 if (iocb
->ki_flags
& IOCB_DIRECT
)
1553 ceph_end_io_direct(inode
);
1555 ceph_end_io_read(inode
);
1559 if ((got
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)) == 0 ||
1560 (iocb
->ki_flags
& IOCB_DIRECT
) ||
1561 (fi
->flags
& CEPH_F_SYNC
)) {
1563 dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n",
1564 inode
, ceph_vinop(inode
), iocb
->ki_pos
, (unsigned)len
,
1565 ceph_cap_string(got
));
1567 if (ci
->i_inline_version
== CEPH_INLINE_NONE
) {
1568 if (!retry_op
&& (iocb
->ki_flags
& IOCB_DIRECT
)) {
1569 ret
= ceph_direct_read_write(iocb
, to
,
1571 if (ret
>= 0 && ret
< len
)
1572 retry_op
= CHECK_EOF
;
1574 ret
= ceph_sync_read(iocb
, to
, &retry_op
);
1577 retry_op
= READ_INLINE
;
1580 CEPH_DEFINE_RW_CONTEXT(rw_ctx
, got
);
1581 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
1582 inode
, ceph_vinop(inode
), iocb
->ki_pos
, (unsigned)len
,
1583 ceph_cap_string(got
));
1584 ceph_add_rw_context(fi
, &rw_ctx
);
1585 ret
= generic_file_read_iter(iocb
, to
);
1586 ceph_del_rw_context(fi
, &rw_ctx
);
1589 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
1590 inode
, ceph_vinop(inode
), ceph_cap_string(got
), (int)ret
);
1591 ceph_put_cap_refs(ci
, got
);
1594 ceph_end_io_direct(inode
);
1596 ceph_end_io_read(inode
);
1598 if (retry_op
> HAVE_RETRIED
&& ret
>= 0) {
1600 struct page
*page
= NULL
;
1602 if (retry_op
== READ_INLINE
) {
1603 page
= __page_cache_alloc(GFP_KERNEL
);
1608 statret
= __ceph_do_getattr(inode
, page
,
1609 CEPH_STAT_CAP_INLINE_DATA
, !!page
);
1613 if (statret
== -ENODATA
) {
1614 BUG_ON(retry_op
!= READ_INLINE
);
1620 i_size
= i_size_read(inode
);
1621 if (retry_op
== READ_INLINE
) {
1622 BUG_ON(ret
> 0 || read
> 0);
1623 if (iocb
->ki_pos
< i_size
&&
1624 iocb
->ki_pos
< PAGE_SIZE
) {
1625 loff_t end
= min_t(loff_t
, i_size
,
1626 iocb
->ki_pos
+ len
);
1627 end
= min_t(loff_t
, end
, PAGE_SIZE
);
1629 zero_user_segment(page
, statret
, end
);
1630 ret
= copy_page_to_iter(page
,
1631 iocb
->ki_pos
& ~PAGE_MASK
,
1632 end
- iocb
->ki_pos
, to
);
1633 iocb
->ki_pos
+= ret
;
1636 if (iocb
->ki_pos
< i_size
&& read
< len
) {
1637 size_t zlen
= min_t(size_t, len
- read
,
1638 i_size
- iocb
->ki_pos
);
1639 ret
= iov_iter_zero(zlen
, to
);
1640 iocb
->ki_pos
+= ret
;
1643 __free_pages(page
, 0);
1647 /* hit EOF or hole? */
1648 if (retry_op
== CHECK_EOF
&& iocb
->ki_pos
< i_size
&&
1650 dout("sync_read hit hole, ppos %lld < size %lld"
1651 ", reading more\n", iocb
->ki_pos
, i_size
);
1655 retry_op
= HAVE_RETRIED
;
1667 * Take cap references to avoid releasing caps to MDS mid-write.
1669 * If we are synchronous, and write with an old snap context, the OSD
1670 * may return EOLDSNAPC. In that case, retry the write.. _after_
1671 * dropping our cap refs and allowing the pending snap to logically
1672 * complete _before_ this write occurs.
1674 * If we are near ENOSPC, write synchronously.
1676 static ssize_t
ceph_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1678 struct file
*file
= iocb
->ki_filp
;
1679 struct ceph_file_info
*fi
= file
->private_data
;
1680 struct inode
*inode
= file_inode(file
);
1681 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1682 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
1683 struct ceph_osd_client
*osdc
= &fsc
->client
->osdc
;
1684 struct ceph_cap_flush
*prealloc_cf
;
1685 ssize_t count
, written
= 0;
1687 bool direct_lock
= false;
1691 loff_t limit
= max(i_size_read(inode
), fsc
->max_file_size
);
1693 if (ceph_snap(inode
) != CEPH_NOSNAP
)
1696 prealloc_cf
= ceph_alloc_cap_flush();
1700 if ((iocb
->ki_flags
& (IOCB_DIRECT
| IOCB_APPEND
)) == IOCB_DIRECT
)
1705 ceph_start_io_direct(inode
);
1707 ceph_start_io_write(inode
);
1709 /* We can write back this queue in page reclaim */
1710 current
->backing_dev_info
= inode_to_bdi(inode
);
1712 if (iocb
->ki_flags
& IOCB_APPEND
) {
1713 err
= ceph_do_getattr(inode
, CEPH_STAT_CAP_SIZE
, false);
1718 err
= generic_write_checks(iocb
, from
);
1723 if (unlikely(pos
>= limit
)) {
1727 iov_iter_truncate(from
, limit
- pos
);
1730 count
= iov_iter_count(from
);
1731 if (ceph_quota_is_max_bytes_exceeded(inode
, pos
+ count
)) {
1736 down_read(&osdc
->lock
);
1737 map_flags
= osdc
->osdmap
->flags
;
1738 pool_flags
= ceph_pg_pool_flags(osdc
->osdmap
, ci
->i_layout
.pool_id
);
1739 up_read(&osdc
->lock
);
1740 if ((map_flags
& CEPH_OSDMAP_FULL
) ||
1741 (pool_flags
& CEPH_POOL_FLAG_FULL
)) {
1746 err
= file_remove_privs(file
);
1750 if (ci
->i_inline_version
!= CEPH_INLINE_NONE
) {
1751 err
= ceph_uninline_data(file
, NULL
);
1756 dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
1757 inode
, ceph_vinop(inode
), pos
, count
, i_size_read(inode
));
1758 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
1759 want
= CEPH_CAP_FILE_BUFFER
| CEPH_CAP_FILE_LAZYIO
;
1761 want
= CEPH_CAP_FILE_BUFFER
;
1763 err
= ceph_get_caps(file
, CEPH_CAP_FILE_WR
, want
, pos
+ count
, &got
);
1767 err
= file_update_time(file
);
1771 inode_inc_iversion_raw(inode
);
1773 dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
1774 inode
, ceph_vinop(inode
), pos
, count
, ceph_cap_string(got
));
1776 if ((got
& (CEPH_CAP_FILE_BUFFER
|CEPH_CAP_FILE_LAZYIO
)) == 0 ||
1777 (iocb
->ki_flags
& IOCB_DIRECT
) || (fi
->flags
& CEPH_F_SYNC
) ||
1778 (ci
->i_ceph_flags
& CEPH_I_ERROR_WRITE
)) {
1779 struct ceph_snap_context
*snapc
;
1780 struct iov_iter data
;
1782 spin_lock(&ci
->i_ceph_lock
);
1783 if (__ceph_have_pending_cap_snap(ci
)) {
1784 struct ceph_cap_snap
*capsnap
=
1785 list_last_entry(&ci
->i_cap_snaps
,
1786 struct ceph_cap_snap
,
1788 snapc
= ceph_get_snap_context(capsnap
->context
);
1790 BUG_ON(!ci
->i_head_snapc
);
1791 snapc
= ceph_get_snap_context(ci
->i_head_snapc
);
1793 spin_unlock(&ci
->i_ceph_lock
);
1795 /* we might need to revert back to that point */
1797 if (iocb
->ki_flags
& IOCB_DIRECT
)
1798 written
= ceph_direct_read_write(iocb
, &data
, snapc
,
1801 written
= ceph_sync_write(iocb
, &data
, pos
, snapc
);
1803 ceph_end_io_direct(inode
);
1805 ceph_end_io_write(inode
);
1807 iov_iter_advance(from
, written
);
1808 ceph_put_snap_context(snapc
);
1811 * No need to acquire the i_truncate_mutex. Because
1812 * the MDS revokes Fwb caps before sending truncate
1813 * message to us. We can't get Fwb cap while there
1814 * are pending vmtruncate. So write and vmtruncate
1815 * can not run at the same time
1817 written
= generic_perform_write(file
, from
, pos
);
1818 if (likely(written
>= 0))
1819 iocb
->ki_pos
= pos
+ written
;
1820 ceph_end_io_write(inode
);
1826 spin_lock(&ci
->i_ceph_lock
);
1827 ci
->i_inline_version
= CEPH_INLINE_NONE
;
1828 dirty
= __ceph_mark_dirty_caps(ci
, CEPH_CAP_FILE_WR
,
1830 spin_unlock(&ci
->i_ceph_lock
);
1832 __mark_inode_dirty(inode
, dirty
);
1833 if (ceph_quota_is_max_bytes_approaching(inode
, iocb
->ki_pos
))
1834 ceph_check_caps(ci
, 0, NULL
);
1837 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
1838 inode
, ceph_vinop(inode
), pos
, (unsigned)count
,
1839 ceph_cap_string(got
));
1840 ceph_put_cap_refs(ci
, got
);
1842 if (written
== -EOLDSNAPC
) {
1843 dout("aio_write %p %llx.%llx %llu~%u" "got EOLDSNAPC, retrying\n",
1844 inode
, ceph_vinop(inode
), pos
, (unsigned)count
);
1849 if ((map_flags
& CEPH_OSDMAP_NEARFULL
) ||
1850 (pool_flags
& CEPH_POOL_FLAG_NEARFULL
))
1851 iocb
->ki_flags
|= IOCB_DSYNC
;
1852 written
= generic_write_sync(iocb
, written
);
1857 ceph_put_cap_refs(ci
, got
);
1860 ceph_end_io_direct(inode
);
1862 ceph_end_io_write(inode
);
1864 ceph_free_cap_flush(prealloc_cf
);
1865 current
->backing_dev_info
= NULL
;
1866 return written
? written
: err
;
1870 * llseek. be sure to verify file size on SEEK_END.
1872 static loff_t
ceph_llseek(struct file
*file
, loff_t offset
, int whence
)
1874 struct inode
*inode
= file
->f_mapping
->host
;
1875 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
1881 if (whence
== SEEK_END
|| whence
== SEEK_DATA
|| whence
== SEEK_HOLE
) {
1882 ret
= ceph_do_getattr(inode
, CEPH_STAT_CAP_SIZE
, false);
1887 i_size
= i_size_read(inode
);
1894 * Here we special-case the lseek(fd, 0, SEEK_CUR)
1895 * position-querying operation. Avoid rewriting the "same"
1896 * f_pos value back to the file because a concurrent read(),
1897 * write() or lseek() might have altered it
1903 offset
+= file
->f_pos
;
1906 if (offset
< 0 || offset
>= i_size
) {
1912 if (offset
< 0 || offset
>= i_size
) {
1920 ret
= vfs_setpos(file
, offset
, max(i_size
, fsc
->max_file_size
));
1923 inode_unlock(inode
);
1927 static inline void ceph_zero_partial_page(
1928 struct inode
*inode
, loff_t offset
, unsigned size
)
1931 pgoff_t index
= offset
>> PAGE_SHIFT
;
1933 page
= find_lock_page(inode
->i_mapping
, index
);
1935 wait_on_page_writeback(page
);
1936 zero_user(page
, offset
& (PAGE_SIZE
- 1), size
);
1942 static void ceph_zero_pagecache_range(struct inode
*inode
, loff_t offset
,
1945 loff_t nearly
= round_up(offset
, PAGE_SIZE
);
1946 if (offset
< nearly
) {
1947 loff_t size
= nearly
- offset
;
1950 ceph_zero_partial_page(inode
, offset
, size
);
1954 if (length
>= PAGE_SIZE
) {
1955 loff_t size
= round_down(length
, PAGE_SIZE
);
1956 truncate_pagecache_range(inode
, offset
, offset
+ size
- 1);
1961 ceph_zero_partial_page(inode
, offset
, length
);
1964 static int ceph_zero_partial_object(struct inode
*inode
,
1965 loff_t offset
, loff_t
*length
)
1967 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1968 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
1969 struct ceph_osd_request
*req
;
1975 op
= offset
? CEPH_OSD_OP_DELETE
: CEPH_OSD_OP_TRUNCATE
;
1978 op
= CEPH_OSD_OP_ZERO
;
1981 req
= ceph_osdc_new_request(&fsc
->client
->osdc
, &ci
->i_layout
,
1985 CEPH_OSD_FLAG_WRITE
,
1992 req
->r_mtime
= inode
->i_mtime
;
1993 ret
= ceph_osdc_start_request(&fsc
->client
->osdc
, req
, false);
1995 ret
= ceph_osdc_wait_request(&fsc
->client
->osdc
, req
);
1999 ceph_osdc_put_request(req
);
2005 static int ceph_zero_objects(struct inode
*inode
, loff_t offset
, loff_t length
)
2008 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2009 s32 stripe_unit
= ci
->i_layout
.stripe_unit
;
2010 s32 stripe_count
= ci
->i_layout
.stripe_count
;
2011 s32 object_size
= ci
->i_layout
.object_size
;
2012 u64 object_set_size
= object_size
* stripe_count
;
2015 /* round offset up to next period boundary */
2016 nearly
= offset
+ object_set_size
- 1;
2018 nearly
-= do_div(t
, object_set_size
);
2020 while (length
&& offset
< nearly
) {
2021 loff_t size
= length
;
2022 ret
= ceph_zero_partial_object(inode
, offset
, &size
);
2028 while (length
>= object_set_size
) {
2030 loff_t pos
= offset
;
2031 for (i
= 0; i
< stripe_count
; ++i
) {
2032 ret
= ceph_zero_partial_object(inode
, pos
, NULL
);
2037 offset
+= object_set_size
;
2038 length
-= object_set_size
;
2041 loff_t size
= length
;
2042 ret
= ceph_zero_partial_object(inode
, offset
, &size
);
2051 static long ceph_fallocate(struct file
*file
, int mode
,
2052 loff_t offset
, loff_t length
)
2054 struct ceph_file_info
*fi
= file
->private_data
;
2055 struct inode
*inode
= file_inode(file
);
2056 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2057 struct ceph_cap_flush
*prealloc_cf
;
2064 if (mode
!= (FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2067 if (!S_ISREG(inode
->i_mode
))
2070 prealloc_cf
= ceph_alloc_cap_flush();
2076 if (ceph_snap(inode
) != CEPH_NOSNAP
) {
2081 if (ci
->i_inline_version
!= CEPH_INLINE_NONE
) {
2082 ret
= ceph_uninline_data(file
, NULL
);
2087 size
= i_size_read(inode
);
2089 /* Are we punching a hole beyond EOF? */
2092 if ((offset
+ length
) > size
)
2093 length
= size
- offset
;
2095 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
2096 want
= CEPH_CAP_FILE_BUFFER
| CEPH_CAP_FILE_LAZYIO
;
2098 want
= CEPH_CAP_FILE_BUFFER
;
2100 ret
= ceph_get_caps(file
, CEPH_CAP_FILE_WR
, want
, endoff
, &got
);
2104 filemap_invalidate_lock(inode
->i_mapping
);
2105 ceph_zero_pagecache_range(inode
, offset
, length
);
2106 ret
= ceph_zero_objects(inode
, offset
, length
);
2109 spin_lock(&ci
->i_ceph_lock
);
2110 ci
->i_inline_version
= CEPH_INLINE_NONE
;
2111 dirty
= __ceph_mark_dirty_caps(ci
, CEPH_CAP_FILE_WR
,
2113 spin_unlock(&ci
->i_ceph_lock
);
2115 __mark_inode_dirty(inode
, dirty
);
2117 filemap_invalidate_unlock(inode
->i_mapping
);
2119 ceph_put_cap_refs(ci
, got
);
2121 inode_unlock(inode
);
2122 ceph_free_cap_flush(prealloc_cf
);
2127 * This function tries to get FILE_WR capabilities for dst_ci and FILE_RD for
2128 * src_ci. Two attempts are made to obtain both caps, and an error is return if
2129 * this fails; zero is returned on success.
2131 static int get_rd_wr_caps(struct file
*src_filp
, int *src_got
,
2132 struct file
*dst_filp
,
2133 loff_t dst_endoff
, int *dst_got
)
2136 bool retrying
= false;
2139 ret
= ceph_get_caps(dst_filp
, CEPH_CAP_FILE_WR
, CEPH_CAP_FILE_BUFFER
,
2140 dst_endoff
, dst_got
);
2145 * Since we're already holding the FILE_WR capability for the dst file,
2146 * we would risk a deadlock by using ceph_get_caps. Thus, we'll do some
2147 * retry dance instead to try to get both capabilities.
2149 ret
= ceph_try_get_caps(file_inode(src_filp
),
2150 CEPH_CAP_FILE_RD
, CEPH_CAP_FILE_SHARED
,
2153 /* Start by dropping dst_ci caps and getting src_ci caps */
2154 ceph_put_cap_refs(ceph_inode(file_inode(dst_filp
)), *dst_got
);
2157 /* ceph_try_get_caps masks EAGAIN */
2161 ret
= ceph_get_caps(src_filp
, CEPH_CAP_FILE_RD
,
2162 CEPH_CAP_FILE_SHARED
, -1, src_got
);
2165 /*... drop src_ci caps too, and retry */
2166 ceph_put_cap_refs(ceph_inode(file_inode(src_filp
)), *src_got
);
2173 static void put_rd_wr_caps(struct ceph_inode_info
*src_ci
, int src_got
,
2174 struct ceph_inode_info
*dst_ci
, int dst_got
)
2176 ceph_put_cap_refs(src_ci
, src_got
);
2177 ceph_put_cap_refs(dst_ci
, dst_got
);
2181 * This function does several size-related checks, returning an error if:
2182 * - source file is smaller than off+len
2183 * - destination file size is not OK (inode_newsize_ok())
2184 * - max bytes quotas is exceeded
2186 static int is_file_size_ok(struct inode
*src_inode
, struct inode
*dst_inode
,
2187 loff_t src_off
, loff_t dst_off
, size_t len
)
2189 loff_t size
, endoff
;
2191 size
= i_size_read(src_inode
);
2193 * Don't copy beyond source file EOF. Instead of simply setting length
2194 * to (size - src_off), just drop to VFS default implementation, as the
2195 * local i_size may be stale due to other clients writing to the source
2198 if (src_off
+ len
> size
) {
2199 dout("Copy beyond EOF (%llu + %zu > %llu)\n",
2200 src_off
, len
, size
);
2203 size
= i_size_read(dst_inode
);
2205 endoff
= dst_off
+ len
;
2206 if (inode_newsize_ok(dst_inode
, endoff
))
2209 if (ceph_quota_is_max_bytes_exceeded(dst_inode
, endoff
))
2215 static ssize_t
ceph_do_objects_copy(struct ceph_inode_info
*src_ci
, u64
*src_off
,
2216 struct ceph_inode_info
*dst_ci
, u64
*dst_off
,
2217 struct ceph_fs_client
*fsc
,
2218 size_t len
, unsigned int flags
)
2220 struct ceph_object_locator src_oloc
, dst_oloc
;
2221 struct ceph_object_id src_oid
, dst_oid
;
2223 u64 src_objnum
, src_objoff
, dst_objnum
, dst_objoff
;
2224 u32 src_objlen
, dst_objlen
;
2225 u32 object_size
= src_ci
->i_layout
.object_size
;
2228 src_oloc
.pool
= src_ci
->i_layout
.pool_id
;
2229 src_oloc
.pool_ns
= ceph_try_get_string(src_ci
->i_layout
.pool_ns
);
2230 dst_oloc
.pool
= dst_ci
->i_layout
.pool_id
;
2231 dst_oloc
.pool_ns
= ceph_try_get_string(dst_ci
->i_layout
.pool_ns
);
2233 while (len
>= object_size
) {
2234 ceph_calc_file_object_mapping(&src_ci
->i_layout
, *src_off
,
2235 object_size
, &src_objnum
,
2236 &src_objoff
, &src_objlen
);
2237 ceph_calc_file_object_mapping(&dst_ci
->i_layout
, *dst_off
,
2238 object_size
, &dst_objnum
,
2239 &dst_objoff
, &dst_objlen
);
2240 ceph_oid_init(&src_oid
);
2241 ceph_oid_printf(&src_oid
, "%llx.%08llx",
2242 src_ci
->i_vino
.ino
, src_objnum
);
2243 ceph_oid_init(&dst_oid
);
2244 ceph_oid_printf(&dst_oid
, "%llx.%08llx",
2245 dst_ci
->i_vino
.ino
, dst_objnum
);
2246 /* Do an object remote copy */
2247 ret
= ceph_osdc_copy_from(&fsc
->client
->osdc
,
2248 src_ci
->i_vino
.snap
, 0,
2249 &src_oid
, &src_oloc
,
2250 CEPH_OSD_OP_FLAG_FADVISE_SEQUENTIAL
|
2251 CEPH_OSD_OP_FLAG_FADVISE_NOCACHE
,
2252 &dst_oid
, &dst_oloc
,
2253 CEPH_OSD_OP_FLAG_FADVISE_SEQUENTIAL
|
2254 CEPH_OSD_OP_FLAG_FADVISE_DONTNEED
,
2255 dst_ci
->i_truncate_seq
,
2256 dst_ci
->i_truncate_size
,
2257 CEPH_OSD_COPY_FROM_FLAG_TRUNCATE_SEQ
);
2259 if (ret
== -EOPNOTSUPP
) {
2260 fsc
->have_copy_from2
= false;
2261 pr_notice("OSDs don't support copy-from2; disabling copy offload\n");
2263 dout("ceph_osdc_copy_from returned %d\n", ret
);
2269 bytes
+= object_size
;
2270 *src_off
+= object_size
;
2271 *dst_off
+= object_size
;
2275 ceph_oloc_destroy(&src_oloc
);
2276 ceph_oloc_destroy(&dst_oloc
);
2280 static ssize_t
__ceph_copy_file_range(struct file
*src_file
, loff_t src_off
,
2281 struct file
*dst_file
, loff_t dst_off
,
2282 size_t len
, unsigned int flags
)
2284 struct inode
*src_inode
= file_inode(src_file
);
2285 struct inode
*dst_inode
= file_inode(dst_file
);
2286 struct ceph_inode_info
*src_ci
= ceph_inode(src_inode
);
2287 struct ceph_inode_info
*dst_ci
= ceph_inode(dst_inode
);
2288 struct ceph_cap_flush
*prealloc_cf
;
2289 struct ceph_fs_client
*src_fsc
= ceph_inode_to_client(src_inode
);
2291 ssize_t ret
= -EIO
, bytes
;
2292 u64 src_objnum
, dst_objnum
, src_objoff
, dst_objoff
;
2293 u32 src_objlen
, dst_objlen
;
2294 int src_got
= 0, dst_got
= 0, err
, dirty
;
2296 if (src_inode
->i_sb
!= dst_inode
->i_sb
) {
2297 struct ceph_fs_client
*dst_fsc
= ceph_inode_to_client(dst_inode
);
2299 if (ceph_fsid_compare(&src_fsc
->client
->fsid
,
2300 &dst_fsc
->client
->fsid
)) {
2301 dout("Copying files across clusters: src: %pU dst: %pU\n",
2302 &src_fsc
->client
->fsid
, &dst_fsc
->client
->fsid
);
2306 if (ceph_snap(dst_inode
) != CEPH_NOSNAP
)
2310 * Some of the checks below will return -EOPNOTSUPP, which will force a
2311 * fallback to the default VFS copy_file_range implementation. This is
2312 * desirable in several cases (for ex, the 'len' is smaller than the
2313 * size of the objects, or in cases where that would be more
2317 if (ceph_test_mount_opt(src_fsc
, NOCOPYFROM
))
2320 if (!src_fsc
->have_copy_from2
)
2324 * Striped file layouts require that we copy partial objects, but the
2325 * OSD copy-from operation only supports full-object copies. Limit
2326 * this to non-striped file layouts for now.
2328 if ((src_ci
->i_layout
.stripe_unit
!= dst_ci
->i_layout
.stripe_unit
) ||
2329 (src_ci
->i_layout
.stripe_count
!= 1) ||
2330 (dst_ci
->i_layout
.stripe_count
!= 1) ||
2331 (src_ci
->i_layout
.object_size
!= dst_ci
->i_layout
.object_size
)) {
2332 dout("Invalid src/dst files layout\n");
2336 if (len
< src_ci
->i_layout
.object_size
)
2337 return -EOPNOTSUPP
; /* no remote copy will be done */
2339 prealloc_cf
= ceph_alloc_cap_flush();
2343 /* Start by sync'ing the source and destination files */
2344 ret
= file_write_and_wait_range(src_file
, src_off
, (src_off
+ len
));
2346 dout("failed to write src file (%zd)\n", ret
);
2349 ret
= file_write_and_wait_range(dst_file
, dst_off
, (dst_off
+ len
));
2351 dout("failed to write dst file (%zd)\n", ret
);
2356 * We need FILE_WR caps for dst_ci and FILE_RD for src_ci as other
2357 * clients may have dirty data in their caches. And OSDs know nothing
2358 * about caps, so they can't safely do the remote object copies.
2360 err
= get_rd_wr_caps(src_file
, &src_got
,
2361 dst_file
, (dst_off
+ len
), &dst_got
);
2363 dout("get_rd_wr_caps returned %d\n", err
);
2368 ret
= is_file_size_ok(src_inode
, dst_inode
, src_off
, dst_off
, len
);
2372 /* Drop dst file cached pages */
2373 ret
= invalidate_inode_pages2_range(dst_inode
->i_mapping
,
2374 dst_off
>> PAGE_SHIFT
,
2375 (dst_off
+ len
) >> PAGE_SHIFT
);
2377 dout("Failed to invalidate inode pages (%zd)\n", ret
);
2380 ceph_calc_file_object_mapping(&src_ci
->i_layout
, src_off
,
2381 src_ci
->i_layout
.object_size
,
2382 &src_objnum
, &src_objoff
, &src_objlen
);
2383 ceph_calc_file_object_mapping(&dst_ci
->i_layout
, dst_off
,
2384 dst_ci
->i_layout
.object_size
,
2385 &dst_objnum
, &dst_objoff
, &dst_objlen
);
2386 /* object-level offsets need to the same */
2387 if (src_objoff
!= dst_objoff
) {
2393 * Do a manual copy if the object offset isn't object aligned.
2394 * 'src_objlen' contains the bytes left until the end of the object,
2395 * starting at the src_off
2398 dout("Initial partial copy of %u bytes\n", src_objlen
);
2401 * we need to temporarily drop all caps as we'll be calling
2402 * {read,write}_iter, which will get caps again.
2404 put_rd_wr_caps(src_ci
, src_got
, dst_ci
, dst_got
);
2405 ret
= do_splice_direct(src_file
, &src_off
, dst_file
,
2406 &dst_off
, src_objlen
, flags
);
2407 /* Abort on short copies or on error */
2408 if (ret
< src_objlen
) {
2409 dout("Failed partial copy (%zd)\n", ret
);
2413 err
= get_rd_wr_caps(src_file
, &src_got
,
2414 dst_file
, (dst_off
+ len
), &dst_got
);
2417 err
= is_file_size_ok(src_inode
, dst_inode
,
2418 src_off
, dst_off
, len
);
2423 size
= i_size_read(dst_inode
);
2424 bytes
= ceph_do_objects_copy(src_ci
, &src_off
, dst_ci
, &dst_off
,
2425 src_fsc
, len
, flags
);
2431 dout("Copied %zu bytes out of %zu\n", bytes
, len
);
2435 file_update_time(dst_file
);
2436 inode_inc_iversion_raw(dst_inode
);
2438 if (dst_off
> size
) {
2439 /* Let the MDS know about dst file size change */
2440 if (ceph_inode_set_size(dst_inode
, dst_off
) ||
2441 ceph_quota_is_max_bytes_approaching(dst_inode
, dst_off
))
2442 ceph_check_caps(dst_ci
, CHECK_CAPS_AUTHONLY
, NULL
);
2445 spin_lock(&dst_ci
->i_ceph_lock
);
2446 dst_ci
->i_inline_version
= CEPH_INLINE_NONE
;
2447 dirty
= __ceph_mark_dirty_caps(dst_ci
, CEPH_CAP_FILE_WR
, &prealloc_cf
);
2448 spin_unlock(&dst_ci
->i_ceph_lock
);
2450 __mark_inode_dirty(dst_inode
, dirty
);
2453 put_rd_wr_caps(src_ci
, src_got
, dst_ci
, dst_got
);
2456 * Do the final manual copy if we still have some bytes left, unless
2457 * there were errors in remote object copies (len >= object_size).
2459 if (len
&& (len
< src_ci
->i_layout
.object_size
)) {
2460 dout("Final partial copy of %zu bytes\n", len
);
2461 bytes
= do_splice_direct(src_file
, &src_off
, dst_file
,
2462 &dst_off
, len
, flags
);
2466 dout("Failed partial copy (%zd)\n", bytes
);
2470 ceph_free_cap_flush(prealloc_cf
);
2475 static ssize_t
ceph_copy_file_range(struct file
*src_file
, loff_t src_off
,
2476 struct file
*dst_file
, loff_t dst_off
,
2477 size_t len
, unsigned int flags
)
2481 ret
= __ceph_copy_file_range(src_file
, src_off
, dst_file
, dst_off
,
2484 if (ret
== -EOPNOTSUPP
|| ret
== -EXDEV
)
2485 ret
= generic_copy_file_range(src_file
, src_off
, dst_file
,
2486 dst_off
, len
, flags
);
2490 const struct file_operations ceph_file_fops
= {
2492 .release
= ceph_release
,
2493 .llseek
= ceph_llseek
,
2494 .read_iter
= ceph_read_iter
,
2495 .write_iter
= ceph_write_iter
,
2497 .fsync
= ceph_fsync
,
2499 .setlease
= simple_nosetlease
,
2500 .flock
= ceph_flock
,
2501 .splice_read
= generic_file_splice_read
,
2502 .splice_write
= iter_file_splice_write
,
2503 .unlocked_ioctl
= ceph_ioctl
,
2504 .compat_ioctl
= compat_ptr_ioctl
,
2505 .fallocate
= ceph_fallocate
,
2506 .copy_file_range
= ceph_copy_file_range
,