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->meta_err = errseq_sample(&ci->i_meta_err);
237 fi->filp_gen = READ_ONCE(ceph_inode_to_client(inode)->filp_gen);
243 * initialize private struct file data.
244 * if we fail, clean up by dropping fmode reference on the ceph_inode
246 static int ceph_init_file(struct inode *inode, struct file *file, int fmode)
250 switch (inode->i_mode & S_IFMT) {
252 ceph_fscache_register_inode_cookie(inode);
253 ceph_fscache_file_set_cookie(inode, file);
256 ret = ceph_init_file_info(inode, file, fmode,
257 S_ISDIR(inode->i_mode));
261 dout("init_file %p %p 0%o (symlink)\n", inode, file,
266 dout("init_file %p %p 0%o (special)\n", inode, file,
269 * we need to drop the open ref now, since we don't
270 * have .release set to ceph_release.
272 BUG_ON(inode->i_fop->release == ceph_release);
274 /* call the proper open fop */
275 ret = inode->i_fop->open(inode, file);
281 * try renew caps after session gets killed.
283 int ceph_renew_caps(struct inode *inode, int fmode)
285 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
286 struct ceph_inode_info *ci = ceph_inode(inode);
287 struct ceph_mds_request *req;
288 int err, flags, wanted;
290 spin_lock(&ci->i_ceph_lock);
291 __ceph_touch_fmode(ci, mdsc, fmode);
292 wanted = __ceph_caps_file_wanted(ci);
293 if (__ceph_is_any_real_caps(ci) &&
294 (!(wanted & CEPH_CAP_ANY_WR) || ci->i_auth_cap)) {
295 int issued = __ceph_caps_issued(ci, NULL);
296 spin_unlock(&ci->i_ceph_lock);
297 dout("renew caps %p want %s issued %s updating mds_wanted\n",
298 inode, ceph_cap_string(wanted), ceph_cap_string(issued));
299 ceph_check_caps(ci, 0, NULL);
302 spin_unlock(&ci->i_ceph_lock);
305 if ((wanted & CEPH_CAP_FILE_RD) && (wanted & CEPH_CAP_FILE_WR))
307 else if (wanted & CEPH_CAP_FILE_RD)
309 else if (wanted & CEPH_CAP_FILE_WR)
312 if (wanted & CEPH_CAP_FILE_LAZYIO)
316 req = prepare_open_request(inode->i_sb, flags, 0);
322 req->r_inode = inode;
326 err = ceph_mdsc_do_request(mdsc, NULL, req);
327 ceph_mdsc_put_request(req);
329 dout("renew caps %p open result=%d\n", inode, err);
330 return err < 0 ? err : 0;
334 * If we already have the requisite capabilities, we can satisfy
335 * the open request locally (no need to request new caps from the
336 * MDS). We do, however, need to inform the MDS (asynchronously)
337 * if our wanted caps set expands.
339 int ceph_open(struct inode *inode, struct file *file)
341 struct ceph_inode_info *ci = ceph_inode(inode);
342 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
343 struct ceph_mds_client *mdsc = fsc->mdsc;
344 struct ceph_mds_request *req;
345 struct ceph_file_info *fi = file->private_data;
347 int flags, fmode, wanted;
350 dout("open file %p is already opened\n", file);
354 /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */
355 flags = file->f_flags & ~(O_CREAT|O_EXCL);
356 if (S_ISDIR(inode->i_mode))
357 flags = O_DIRECTORY; /* mds likes to know */
359 dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode,
360 ceph_vinop(inode), file, flags, file->f_flags);
361 fmode = ceph_flags_to_mode(flags);
362 wanted = ceph_caps_for_mode(fmode);
364 /* snapped files are read-only */
365 if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE))
368 /* trivially open snapdir */
369 if (ceph_snap(inode) == CEPH_SNAPDIR) {
370 return ceph_init_file(inode, file, fmode);
374 * No need to block if we have caps on the auth MDS (for
375 * write) or any MDS (for read). Update wanted set
378 spin_lock(&ci->i_ceph_lock);
379 if (__ceph_is_any_real_caps(ci) &&
380 (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) {
381 int mds_wanted = __ceph_caps_mds_wanted(ci, true);
382 int issued = __ceph_caps_issued(ci, NULL);
384 dout("open %p fmode %d want %s issued %s using existing\n",
385 inode, fmode, ceph_cap_string(wanted),
386 ceph_cap_string(issued));
387 __ceph_touch_fmode(ci, mdsc, fmode);
388 spin_unlock(&ci->i_ceph_lock);
391 if ((issued & wanted) != wanted &&
392 (mds_wanted & wanted) != wanted &&
393 ceph_snap(inode) != CEPH_SNAPDIR)
394 ceph_check_caps(ci, 0, NULL);
396 return ceph_init_file(inode, file, fmode);
397 } else if (ceph_snap(inode) != CEPH_NOSNAP &&
398 (ci->i_snap_caps & wanted) == wanted) {
399 __ceph_touch_fmode(ci, mdsc, fmode);
400 spin_unlock(&ci->i_ceph_lock);
401 return ceph_init_file(inode, file, fmode);
404 spin_unlock(&ci->i_ceph_lock);
406 dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted));
407 req = prepare_open_request(inode->i_sb, flags, 0);
412 req->r_inode = inode;
416 err = ceph_mdsc_do_request(mdsc, NULL, req);
418 err = ceph_init_file(inode, file, req->r_fmode);
419 ceph_mdsc_put_request(req);
420 dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode));
425 /* Clone the layout from a synchronous create, if the dir now has Dc caps */
427 cache_file_layout(struct inode *dst, struct inode *src)
429 struct ceph_inode_info *cdst = ceph_inode(dst);
430 struct ceph_inode_info *csrc = ceph_inode(src);
432 spin_lock(&cdst->i_ceph_lock);
433 if ((__ceph_caps_issued(cdst, NULL) & CEPH_CAP_DIR_CREATE) &&
434 !ceph_file_layout_is_valid(&cdst->i_cached_layout)) {
435 memcpy(&cdst->i_cached_layout, &csrc->i_layout,
436 sizeof(cdst->i_cached_layout));
437 rcu_assign_pointer(cdst->i_cached_layout.pool_ns,
438 ceph_try_get_string(csrc->i_layout.pool_ns));
440 spin_unlock(&cdst->i_ceph_lock);
444 * Try to set up an async create. We need caps, a file layout, and inode number,
445 * and either a lease on the dentry or complete dir info. If any of those
446 * criteria are not satisfied, then return false and the caller can go
449 static int try_prep_async_create(struct inode *dir, struct dentry *dentry,
450 struct ceph_file_layout *lo, u64 *pino)
452 struct ceph_inode_info *ci = ceph_inode(dir);
453 struct ceph_dentry_info *di = ceph_dentry(dentry);
454 int got = 0, want = CEPH_CAP_FILE_EXCL | CEPH_CAP_DIR_CREATE;
457 spin_lock(&ci->i_ceph_lock);
458 /* No auth cap means no chance for Dc caps */
462 /* Any delegated inos? */
463 if (xa_empty(&ci->i_auth_cap->session->s_delegated_inos))
466 if (!ceph_file_layout_is_valid(&ci->i_cached_layout))
469 if ((__ceph_caps_issued(ci, NULL) & want) != want)
472 if (d_in_lookup(dentry)) {
473 if (!__ceph_dir_is_complete(ci))
475 spin_lock(&dentry->d_lock);
476 di->lease_shared_gen = atomic_read(&ci->i_shared_gen);
477 spin_unlock(&dentry->d_lock);
478 } else if (atomic_read(&ci->i_shared_gen) !=
479 READ_ONCE(di->lease_shared_gen)) {
483 ino = ceph_get_deleg_ino(ci->i_auth_cap->session);
488 ceph_take_cap_refs(ci, want, false);
489 memcpy(lo, &ci->i_cached_layout, sizeof(*lo));
490 rcu_assign_pointer(lo->pool_ns,
491 ceph_try_get_string(ci->i_cached_layout.pool_ns));
494 spin_unlock(&ci->i_ceph_lock);
498 static void restore_deleg_ino(struct inode *dir, u64 ino)
500 struct ceph_inode_info *ci = ceph_inode(dir);
501 struct ceph_mds_session *s = NULL;
503 spin_lock(&ci->i_ceph_lock);
505 s = ceph_get_mds_session(ci->i_auth_cap->session);
506 spin_unlock(&ci->i_ceph_lock);
508 int err = ceph_restore_deleg_ino(s, ino);
510 pr_warn("ceph: unable to restore delegated ino 0x%llx to session: %d\n",
512 ceph_put_mds_session(s);
516 static void ceph_async_create_cb(struct ceph_mds_client *mdsc,
517 struct ceph_mds_request *req)
519 int result = req->r_err ? req->r_err :
520 le32_to_cpu(req->r_reply_info.head->result);
522 if (result == -EJUKEBOX)
525 mapping_set_error(req->r_parent->i_mapping, result);
528 struct dentry *dentry = req->r_dentry;
531 char *path = ceph_mdsc_build_path(req->r_dentry, &pathlen,
534 ceph_dir_clear_complete(req->r_parent);
535 if (!d_unhashed(dentry))
538 /* FIXME: start returning I/O errors on all accesses? */
539 pr_warn("ceph: async create failure path=(%llx)%s result=%d!\n",
540 base, IS_ERR(path) ? "<<bad>>" : path, result);
541 ceph_mdsc_free_path(path, pathlen);
544 if (req->r_target_inode) {
545 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
546 u64 ino = ceph_vino(req->r_target_inode).ino;
548 if (req->r_deleg_ino != ino)
549 pr_warn("%s: inode number mismatch! err=%d deleg_ino=0x%llx target=0x%llx\n",
550 __func__, req->r_err, req->r_deleg_ino, ino);
551 mapping_set_error(req->r_target_inode->i_mapping, result);
553 spin_lock(&ci->i_ceph_lock);
554 if (ci->i_ceph_flags & CEPH_I_ASYNC_CREATE) {
555 ci->i_ceph_flags &= ~CEPH_I_ASYNC_CREATE;
556 wake_up_bit(&ci->i_ceph_flags, CEPH_ASYNC_CREATE_BIT);
558 ceph_kick_flushing_inode_caps(req->r_session, ci);
559 spin_unlock(&ci->i_ceph_lock);
561 pr_warn("%s: no req->r_target_inode for 0x%llx\n", __func__,
565 ceph_mdsc_release_dir_caps(req);
568 static int ceph_finish_async_create(struct inode *dir, struct dentry *dentry,
569 struct file *file, umode_t mode,
570 struct ceph_mds_request *req,
571 struct ceph_acl_sec_ctx *as_ctx,
572 struct ceph_file_layout *lo)
576 struct ceph_mds_reply_inode in = { };
577 struct ceph_mds_reply_info_in iinfo = { .in = &in };
578 struct ceph_inode_info *ci = ceph_inode(dir);
580 struct timespec64 now;
581 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
582 struct ceph_vino vino = { .ino = req->r_deleg_ino,
583 .snap = CEPH_NOSNAP };
585 ktime_get_real_ts64(&now);
587 inode = ceph_get_inode(dentry->d_sb, vino);
589 return PTR_ERR(inode);
591 iinfo.inline_version = CEPH_INLINE_NONE;
592 iinfo.change_attr = 1;
593 ceph_encode_timespec64(&iinfo.btime, &now);
595 iinfo.xattr_len = ARRAY_SIZE(xattr_buf);
596 iinfo.xattr_data = xattr_buf;
597 memset(iinfo.xattr_data, 0, iinfo.xattr_len);
599 in.ino = cpu_to_le64(vino.ino);
600 in.snapid = cpu_to_le64(CEPH_NOSNAP);
601 in.version = cpu_to_le64(1); // ???
602 in.cap.caps = in.cap.wanted = cpu_to_le32(CEPH_CAP_ALL_FILE);
603 in.cap.cap_id = cpu_to_le64(1);
604 in.cap.realm = cpu_to_le64(ci->i_snap_realm->ino);
605 in.cap.flags = CEPH_CAP_FLAG_AUTH;
606 in.ctime = in.mtime = in.atime = iinfo.btime;
607 in.mode = cpu_to_le32((u32)mode);
608 in.truncate_seq = cpu_to_le32(1);
609 in.truncate_size = cpu_to_le64(-1ULL);
610 in.xattr_version = cpu_to_le64(1);
611 in.uid = cpu_to_le32(from_kuid(&init_user_ns, current_fsuid()));
612 in.gid = cpu_to_le32(from_kgid(&init_user_ns, dir->i_mode & S_ISGID ?
613 dir->i_gid : current_fsgid()));
614 in.nlink = cpu_to_le32(1);
615 in.max_size = cpu_to_le64(lo->stripe_unit);
617 ceph_file_layout_to_legacy(lo, &in.layout);
619 down_read(&mdsc->snap_rwsem);
620 ret = ceph_fill_inode(inode, NULL, &iinfo, NULL, req->r_session,
622 up_read(&mdsc->snap_rwsem);
624 dout("%s failed to fill inode: %d\n", __func__, ret);
625 ceph_dir_clear_complete(dir);
626 if (!d_unhashed(dentry))
628 if (inode->i_state & I_NEW)
629 discard_new_inode(inode);
633 dout("%s d_adding new inode 0x%llx to 0x%llx/%s\n", __func__,
634 vino.ino, ceph_ino(dir), dentry->d_name.name);
635 ceph_dir_clear_ordered(dir);
636 ceph_init_inode_acls(inode, as_ctx);
637 if (inode->i_state & I_NEW) {
639 * If it's not I_NEW, then someone created this before
640 * we got here. Assume the server is aware of it at
641 * that point and don't worry about setting
642 * CEPH_I_ASYNC_CREATE.
644 ceph_inode(inode)->i_ceph_flags = CEPH_I_ASYNC_CREATE;
645 unlock_new_inode(inode);
647 if (d_in_lookup(dentry) || d_really_is_negative(dentry)) {
648 if (!d_unhashed(dentry))
650 dn = d_splice_alias(inode, dentry);
651 WARN_ON_ONCE(dn && dn != dentry);
653 file->f_mode |= FMODE_CREATED;
654 ret = finish_open(file, dentry, ceph_open);
660 * Do a lookup + open with a single request. If we get a non-existent
661 * file or symlink, return 1 so the VFS can retry.
663 int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
664 struct file *file, unsigned flags, umode_t mode)
666 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
667 struct ceph_mds_client *mdsc = fsc->mdsc;
668 struct ceph_mds_request *req;
670 struct ceph_acl_sec_ctx as_ctx = {};
671 bool try_async = ceph_test_mount_opt(fsc, ASYNC_DIROPS);
675 dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n",
677 d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode);
679 if (dentry->d_name.len > NAME_MAX)
680 return -ENAMETOOLONG;
682 if (flags & O_CREAT) {
683 if (ceph_quota_is_max_files_exceeded(dir))
685 err = ceph_pre_init_acls(dir, &mode, &as_ctx);
688 err = ceph_security_init_secctx(dentry, mode, &as_ctx);
691 } else if (!d_in_lookup(dentry)) {
692 /* If it's not being looked up, it's negative */
697 req = prepare_open_request(dir->i_sb, flags, mode);
702 req->r_dentry = dget(dentry);
704 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
705 if (ceph_security_xattr_wanted(dir))
706 mask |= CEPH_CAP_XATTR_SHARED;
707 req->r_args.open.mask = cpu_to_le32(mask);
710 if (flags & O_CREAT) {
711 struct ceph_file_layout lo;
713 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
714 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
715 if (as_ctx.pagelist) {
716 req->r_pagelist = as_ctx.pagelist;
717 as_ctx.pagelist = NULL;
721 try_prep_async_create(dir, dentry, &lo,
722 &req->r_deleg_ino))) {
723 set_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags);
724 req->r_args.open.flags |= cpu_to_le32(CEPH_O_EXCL);
725 req->r_callback = ceph_async_create_cb;
726 err = ceph_mdsc_submit_request(mdsc, dir, req);
728 err = ceph_finish_async_create(dir, dentry,
731 } else if (err == -EJUKEBOX) {
732 restore_deleg_ino(dir, req->r_deleg_ino);
733 ceph_mdsc_put_request(req);
741 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
742 err = ceph_mdsc_do_request(mdsc,
743 (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
745 if (err == -ENOENT) {
746 dentry = ceph_handle_snapdir(req, dentry);
747 if (IS_ERR(dentry)) {
748 err = PTR_ERR(dentry);
754 if (!err && (flags & O_CREAT) && !req->r_reply_info.head->is_dentry)
755 err = ceph_handle_notrace_create(dir, dentry);
757 if (d_in_lookup(dentry)) {
758 dn = ceph_finish_lookup(req, dentry, err);
762 /* we were given a hashed negative dentry */
767 if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) {
768 /* make vfs retry on splice, ENOENT, or symlink */
769 dout("atomic_open finish_no_open on dn %p\n", dn);
770 err = finish_no_open(file, dn);
772 dout("atomic_open finish_open on dn %p\n", dn);
773 if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) {
774 struct inode *newino = d_inode(dentry);
776 cache_file_layout(dir, newino);
777 ceph_init_inode_acls(newino, &as_ctx);
778 file->f_mode |= FMODE_CREATED;
780 err = finish_open(file, dentry, ceph_open);
783 ceph_mdsc_put_request(req);
785 ceph_release_acl_sec_ctx(&as_ctx);
786 dout("atomic_open result=%d\n", err);
790 int ceph_release(struct inode *inode, struct file *file)
792 struct ceph_inode_info *ci = ceph_inode(inode);
794 if (S_ISDIR(inode->i_mode)) {
795 struct ceph_dir_file_info *dfi = file->private_data;
796 dout("release inode %p dir file %p\n", inode, file);
797 WARN_ON(!list_empty(&dfi->file_info.rw_contexts));
799 ceph_put_fmode(ci, dfi->file_info.fmode, 1);
801 if (dfi->last_readdir)
802 ceph_mdsc_put_request(dfi->last_readdir);
803 kfree(dfi->last_name);
804 kfree(dfi->dir_info);
805 kmem_cache_free(ceph_dir_file_cachep, dfi);
807 struct ceph_file_info *fi = file->private_data;
808 dout("release inode %p regular file %p\n", inode, file);
809 WARN_ON(!list_empty(&fi->rw_contexts));
811 ceph_put_fmode(ci, fi->fmode, 1);
813 kmem_cache_free(ceph_file_cachep, fi);
816 /* wake up anyone waiting for caps on this inode */
817 wake_up_all(&ci->i_cap_wq);
828 * Completely synchronous read and write methods. Direct from __user
829 * buffer to osd, or directly to user pages (if O_DIRECT).
831 * If the read spans object boundary, just do multiple reads. (That's not
832 * atomic, but good enough for now.)
834 * If we get a short result from the OSD, check against i_size; we need to
835 * only return a short read to the caller if we hit EOF.
837 static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *to,
840 struct file *file = iocb->ki_filp;
841 struct inode *inode = file_inode(file);
842 struct ceph_inode_info *ci = ceph_inode(inode);
843 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
844 struct ceph_osd_client *osdc = &fsc->client->osdc;
846 u64 off = iocb->ki_pos;
847 u64 len = iov_iter_count(to);
849 dout("sync_read on file %p %llu~%u %s\n", file, off, (unsigned)len,
850 (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
855 * flush any page cache pages in this range. this
856 * will make concurrent normal and sync io slow,
857 * but it will at least behave sensibly when they are
860 ret = filemap_write_and_wait_range(inode->i_mapping,
866 while ((len = iov_iter_count(to)) > 0) {
867 struct ceph_osd_request *req;
876 req = ceph_osdc_new_request(osdc, &ci->i_layout,
877 ci->i_vino, off, &len, 0, 1,
878 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
879 NULL, ci->i_truncate_seq,
880 ci->i_truncate_size, false);
886 more = len < iov_iter_count(to);
888 num_pages = calc_pages_for(off, len);
889 page_off = off & ~PAGE_MASK;
890 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
892 ceph_osdc_put_request(req);
893 ret = PTR_ERR(pages);
897 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_off,
899 ret = ceph_osdc_start_request(osdc, req, false);
901 ret = ceph_osdc_wait_request(osdc, req);
903 ceph_update_read_metrics(&fsc->mdsc->metric,
904 req->r_start_latency,
908 ceph_osdc_put_request(req);
910 i_size = i_size_read(inode);
911 dout("sync_read %llu~%llu got %zd i_size %llu%s\n",
912 off, len, ret, i_size, (more ? " MORE" : ""));
916 if (ret >= 0 && ret < len && (off + ret < i_size)) {
917 int zlen = min(len - ret, i_size - off - ret);
918 int zoff = page_off + ret;
919 dout("sync_read zero gap %llu~%llu\n",
920 off + ret, off + ret + zlen);
921 ceph_zero_page_vector_range(zoff, zlen, pages);
926 left = ret > 0 ? ret : 0;
929 page_off = off & ~PAGE_MASK;
930 len = min_t(size_t, left, PAGE_SIZE - page_off);
931 SetPageUptodate(pages[idx]);
932 copied = copy_page_to_iter(pages[idx++],
941 ceph_release_page_vector(pages, num_pages);
944 if (ret == -EBLOCKLISTED)
945 fsc->blocklisted = true;
949 if (off >= i_size || !more)
953 if (off > iocb->ki_pos) {
955 iov_iter_count(to) > 0 && off >= i_size_read(inode))
956 *retry_op = CHECK_EOF;
957 ret = off - iocb->ki_pos;
961 dout("sync_read result %zd retry_op %d\n", ret, *retry_op);
965 struct ceph_aio_request {
971 struct list_head osd_reqs;
973 atomic_t pending_reqs;
974 struct timespec64 mtime;
975 struct ceph_cap_flush *prealloc_cf;
978 struct ceph_aio_work {
979 struct work_struct work;
980 struct ceph_osd_request *req;
983 static void ceph_aio_retry_work(struct work_struct *work);
985 static void ceph_aio_complete(struct inode *inode,
986 struct ceph_aio_request *aio_req)
988 struct ceph_inode_info *ci = ceph_inode(inode);
991 if (!atomic_dec_and_test(&aio_req->pending_reqs))
994 if (aio_req->iocb->ki_flags & IOCB_DIRECT)
995 inode_dio_end(inode);
997 ret = aio_req->error;
999 ret = aio_req->total_len;
1001 dout("ceph_aio_complete %p rc %d\n", inode, ret);
1003 if (ret >= 0 && aio_req->write) {
1006 loff_t endoff = aio_req->iocb->ki_pos + aio_req->total_len;
1007 if (endoff > i_size_read(inode)) {
1008 if (ceph_inode_set_size(inode, endoff))
1009 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1012 spin_lock(&ci->i_ceph_lock);
1013 ci->i_inline_version = CEPH_INLINE_NONE;
1014 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1015 &aio_req->prealloc_cf);
1016 spin_unlock(&ci->i_ceph_lock);
1018 __mark_inode_dirty(inode, dirty);
1022 ceph_put_cap_refs(ci, (aio_req->write ? CEPH_CAP_FILE_WR :
1025 aio_req->iocb->ki_complete(aio_req->iocb, ret, 0);
1027 ceph_free_cap_flush(aio_req->prealloc_cf);
1031 static void ceph_aio_complete_req(struct ceph_osd_request *req)
1033 int rc = req->r_result;
1034 struct inode *inode = req->r_inode;
1035 struct ceph_aio_request *aio_req = req->r_priv;
1036 struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0);
1037 struct ceph_client_metric *metric = &ceph_sb_to_mdsc(inode->i_sb)->metric;
1039 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_BVECS);
1040 BUG_ON(!osd_data->num_bvecs);
1042 dout("ceph_aio_complete_req %p rc %d bytes %u\n",
1043 inode, rc, osd_data->bvec_pos.iter.bi_size);
1045 if (rc == -EOLDSNAPC) {
1046 struct ceph_aio_work *aio_work;
1047 BUG_ON(!aio_req->write);
1049 aio_work = kmalloc(sizeof(*aio_work), GFP_NOFS);
1051 INIT_WORK(&aio_work->work, ceph_aio_retry_work);
1052 aio_work->req = req;
1053 queue_work(ceph_inode_to_client(inode)->inode_wq,
1058 } else if (!aio_req->write) {
1061 if (rc >= 0 && osd_data->bvec_pos.iter.bi_size > rc) {
1063 int zlen = osd_data->bvec_pos.iter.bi_size - rc;
1066 * If read is satisfied by single OSD request,
1067 * it can pass EOF. Otherwise read is within
1070 if (aio_req->num_reqs == 1) {
1071 loff_t i_size = i_size_read(inode);
1072 loff_t endoff = aio_req->iocb->ki_pos + rc;
1073 if (endoff < i_size)
1074 zlen = min_t(size_t, zlen,
1076 aio_req->total_len = rc + zlen;
1079 iov_iter_bvec(&i, READ, osd_data->bvec_pos.bvecs,
1080 osd_data->num_bvecs,
1081 osd_data->bvec_pos.iter.bi_size);
1082 iov_iter_advance(&i, rc);
1083 iov_iter_zero(zlen, &i);
1087 /* r_start_latency == 0 means the request was not submitted */
1088 if (req->r_start_latency) {
1090 ceph_update_write_metrics(metric, req->r_start_latency,
1091 req->r_end_latency, rc);
1093 ceph_update_read_metrics(metric, req->r_start_latency,
1094 req->r_end_latency, rc);
1097 put_bvecs(osd_data->bvec_pos.bvecs, osd_data->num_bvecs,
1098 aio_req->should_dirty);
1099 ceph_osdc_put_request(req);
1102 cmpxchg(&aio_req->error, 0, rc);
1104 ceph_aio_complete(inode, aio_req);
1108 static void ceph_aio_retry_work(struct work_struct *work)
1110 struct ceph_aio_work *aio_work =
1111 container_of(work, struct ceph_aio_work, work);
1112 struct ceph_osd_request *orig_req = aio_work->req;
1113 struct ceph_aio_request *aio_req = orig_req->r_priv;
1114 struct inode *inode = orig_req->r_inode;
1115 struct ceph_inode_info *ci = ceph_inode(inode);
1116 struct ceph_snap_context *snapc;
1117 struct ceph_osd_request *req;
1120 spin_lock(&ci->i_ceph_lock);
1121 if (__ceph_have_pending_cap_snap(ci)) {
1122 struct ceph_cap_snap *capsnap =
1123 list_last_entry(&ci->i_cap_snaps,
1124 struct ceph_cap_snap,
1126 snapc = ceph_get_snap_context(capsnap->context);
1128 BUG_ON(!ci->i_head_snapc);
1129 snapc = ceph_get_snap_context(ci->i_head_snapc);
1131 spin_unlock(&ci->i_ceph_lock);
1133 req = ceph_osdc_alloc_request(orig_req->r_osdc, snapc, 1,
1141 req->r_flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE;
1142 ceph_oloc_copy(&req->r_base_oloc, &orig_req->r_base_oloc);
1143 ceph_oid_copy(&req->r_base_oid, &orig_req->r_base_oid);
1145 req->r_ops[0] = orig_req->r_ops[0];
1147 req->r_mtime = aio_req->mtime;
1148 req->r_data_offset = req->r_ops[0].extent.offset;
1150 ret = ceph_osdc_alloc_messages(req, GFP_NOFS);
1152 ceph_osdc_put_request(req);
1157 ceph_osdc_put_request(orig_req);
1159 req->r_callback = ceph_aio_complete_req;
1160 req->r_inode = inode;
1161 req->r_priv = aio_req;
1163 ret = ceph_osdc_start_request(req->r_osdc, req, false);
1166 req->r_result = ret;
1167 ceph_aio_complete_req(req);
1170 ceph_put_snap_context(snapc);
1175 ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter,
1176 struct ceph_snap_context *snapc,
1177 struct ceph_cap_flush **pcf)
1179 struct file *file = iocb->ki_filp;
1180 struct inode *inode = file_inode(file);
1181 struct ceph_inode_info *ci = ceph_inode(inode);
1182 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1183 struct ceph_client_metric *metric = &fsc->mdsc->metric;
1184 struct ceph_vino vino;
1185 struct ceph_osd_request *req;
1186 struct bio_vec *bvecs;
1187 struct ceph_aio_request *aio_req = NULL;
1191 struct timespec64 mtime = current_time(inode);
1192 size_t count = iov_iter_count(iter);
1193 loff_t pos = iocb->ki_pos;
1194 bool write = iov_iter_rw(iter) == WRITE;
1195 bool should_dirty = !write && iter_is_iovec(iter);
1197 if (write && ceph_snap(file_inode(file)) != CEPH_NOSNAP)
1200 dout("sync_direct_%s on file %p %lld~%u snapc %p seq %lld\n",
1201 (write ? "write" : "read"), file, pos, (unsigned)count,
1202 snapc, snapc ? snapc->seq : 0);
1205 int ret2 = invalidate_inode_pages2_range(inode->i_mapping,
1207 (pos + count - 1) >> PAGE_SHIFT);
1209 dout("invalidate_inode_pages2_range returned %d\n", ret2);
1211 flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE;
1213 flags = CEPH_OSD_FLAG_READ;
1216 while (iov_iter_count(iter) > 0) {
1217 u64 size = iov_iter_count(iter);
1221 size = min_t(u64, size, fsc->mount_options->wsize);
1223 size = min_t(u64, size, fsc->mount_options->rsize);
1225 vino = ceph_vino(inode);
1226 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1227 vino, pos, &size, 0,
1229 write ? CEPH_OSD_OP_WRITE :
1233 ci->i_truncate_size,
1240 len = iter_get_bvecs_alloc(iter, size, &bvecs, &num_pages);
1242 ceph_osdc_put_request(req);
1247 osd_req_op_extent_update(req, 0, len);
1250 * To simplify error handling, allow AIO when IO within i_size
1251 * or IO can be satisfied by single OSD request.
1253 if (pos == iocb->ki_pos && !is_sync_kiocb(iocb) &&
1254 (len == count || pos + count <= i_size_read(inode))) {
1255 aio_req = kzalloc(sizeof(*aio_req), GFP_KERNEL);
1257 aio_req->iocb = iocb;
1258 aio_req->write = write;
1259 aio_req->should_dirty = should_dirty;
1260 INIT_LIST_HEAD(&aio_req->osd_reqs);
1262 aio_req->mtime = mtime;
1263 swap(aio_req->prealloc_cf, *pcf);
1271 * throw out any page cache pages in this range. this
1274 truncate_inode_pages_range(inode->i_mapping, pos,
1275 PAGE_ALIGN(pos + len) - 1);
1277 req->r_mtime = mtime;
1280 osd_req_op_extent_osd_data_bvecs(req, 0, bvecs, num_pages, len);
1283 aio_req->total_len += len;
1284 aio_req->num_reqs++;
1285 atomic_inc(&aio_req->pending_reqs);
1287 req->r_callback = ceph_aio_complete_req;
1288 req->r_inode = inode;
1289 req->r_priv = aio_req;
1290 list_add_tail(&req->r_private_item, &aio_req->osd_reqs);
1296 ret = ceph_osdc_start_request(req->r_osdc, req, false);
1298 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1301 ceph_update_write_metrics(metric, req->r_start_latency,
1302 req->r_end_latency, ret);
1304 ceph_update_read_metrics(metric, req->r_start_latency,
1305 req->r_end_latency, ret);
1307 size = i_size_read(inode);
1311 if (ret >= 0 && ret < len && pos + ret < size) {
1313 int zlen = min_t(size_t, len - ret,
1316 iov_iter_bvec(&i, READ, bvecs, num_pages, len);
1317 iov_iter_advance(&i, ret);
1318 iov_iter_zero(zlen, &i);
1325 put_bvecs(bvecs, num_pages, should_dirty);
1326 ceph_osdc_put_request(req);
1331 if (!write && pos >= size)
1334 if (write && pos > size) {
1335 if (ceph_inode_set_size(inode, pos))
1336 ceph_check_caps(ceph_inode(inode),
1337 CHECK_CAPS_AUTHONLY,
1343 LIST_HEAD(osd_reqs);
1345 if (aio_req->num_reqs == 0) {
1350 ceph_get_cap_refs(ci, write ? CEPH_CAP_FILE_WR :
1353 list_splice(&aio_req->osd_reqs, &osd_reqs);
1354 inode_dio_begin(inode);
1355 while (!list_empty(&osd_reqs)) {
1356 req = list_first_entry(&osd_reqs,
1357 struct ceph_osd_request,
1359 list_del_init(&req->r_private_item);
1361 ret = ceph_osdc_start_request(req->r_osdc,
1364 req->r_result = ret;
1365 ceph_aio_complete_req(req);
1368 return -EIOCBQUEUED;
1371 if (ret != -EOLDSNAPC && pos > iocb->ki_pos) {
1372 ret = pos - iocb->ki_pos;
1379 * Synchronous write, straight from __user pointer or user pages.
1381 * If write spans object boundary, just do multiple writes. (For a
1382 * correct atomic write, we should e.g. take write locks on all
1383 * objects, rollback on failure, etc.)
1386 ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
1387 struct ceph_snap_context *snapc)
1389 struct file *file = iocb->ki_filp;
1390 struct inode *inode = file_inode(file);
1391 struct ceph_inode_info *ci = ceph_inode(inode);
1392 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1393 struct ceph_vino vino;
1394 struct ceph_osd_request *req;
1395 struct page **pages;
1401 bool check_caps = false;
1402 struct timespec64 mtime = current_time(inode);
1403 size_t count = iov_iter_count(from);
1405 if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
1408 dout("sync_write on file %p %lld~%u snapc %p seq %lld\n",
1409 file, pos, (unsigned)count, snapc, snapc->seq);
1411 ret = filemap_write_and_wait_range(inode->i_mapping,
1412 pos, pos + count - 1);
1416 ret = invalidate_inode_pages2_range(inode->i_mapping,
1418 (pos + count - 1) >> PAGE_SHIFT);
1420 dout("invalidate_inode_pages2_range returned %d\n", ret);
1422 flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE;
1424 while ((len = iov_iter_count(from)) > 0) {
1428 vino = ceph_vino(inode);
1429 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1430 vino, pos, &len, 0, 1,
1431 CEPH_OSD_OP_WRITE, flags, snapc,
1433 ci->i_truncate_size,
1441 * write from beginning of first page,
1442 * regardless of io alignment
1444 num_pages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1446 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
1447 if (IS_ERR(pages)) {
1448 ret = PTR_ERR(pages);
1453 for (n = 0; n < num_pages; n++) {
1454 size_t plen = min_t(size_t, left, PAGE_SIZE);
1455 ret = copy_page_from_iter(pages[n], 0, plen, from);
1464 ceph_release_page_vector(pages, num_pages);
1468 req->r_inode = inode;
1470 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0,
1473 req->r_mtime = mtime;
1474 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1476 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1478 ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
1479 req->r_end_latency, ret);
1481 ceph_osdc_put_request(req);
1483 ceph_set_error_write(ci);
1487 ceph_clear_error_write(ci);
1490 if (pos > i_size_read(inode)) {
1491 check_caps = ceph_inode_set_size(inode, pos);
1493 ceph_check_caps(ceph_inode(inode),
1494 CHECK_CAPS_AUTHONLY,
1500 if (ret != -EOLDSNAPC && written > 0) {
1508 * Wrap generic_file_aio_read with checks for cap bits on the inode.
1509 * Atomically grab references, so that those bits are not released
1510 * back to the MDS mid-read.
1512 * Hmm, the sync read case isn't actually async... should it be?
1514 static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to)
1516 struct file *filp = iocb->ki_filp;
1517 struct ceph_file_info *fi = filp->private_data;
1518 size_t len = iov_iter_count(to);
1519 struct inode *inode = file_inode(filp);
1520 struct ceph_inode_info *ci = ceph_inode(inode);
1521 bool direct_lock = iocb->ki_flags & IOCB_DIRECT;
1524 int retry_op = 0, read = 0;
1527 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
1528 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode);
1531 ceph_start_io_direct(inode);
1533 ceph_start_io_read(inode);
1535 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1536 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1538 want = CEPH_CAP_FILE_CACHE;
1539 ret = ceph_get_caps(filp, CEPH_CAP_FILE_RD, want, -1, &got);
1541 if (iocb->ki_flags & IOCB_DIRECT)
1542 ceph_end_io_direct(inode);
1544 ceph_end_io_read(inode);
1548 if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 ||
1549 (iocb->ki_flags & IOCB_DIRECT) ||
1550 (fi->flags & CEPH_F_SYNC)) {
1552 dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n",
1553 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
1554 ceph_cap_string(got));
1556 if (ci->i_inline_version == CEPH_INLINE_NONE) {
1557 if (!retry_op && (iocb->ki_flags & IOCB_DIRECT)) {
1558 ret = ceph_direct_read_write(iocb, to,
1560 if (ret >= 0 && ret < len)
1561 retry_op = CHECK_EOF;
1563 ret = ceph_sync_read(iocb, to, &retry_op);
1566 retry_op = READ_INLINE;
1569 CEPH_DEFINE_RW_CONTEXT(rw_ctx, got);
1570 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
1571 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
1572 ceph_cap_string(got));
1573 ceph_add_rw_context(fi, &rw_ctx);
1574 ret = generic_file_read_iter(iocb, to);
1575 ceph_del_rw_context(fi, &rw_ctx);
1578 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
1579 inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
1580 ceph_put_cap_refs(ci, got);
1583 ceph_end_io_direct(inode);
1585 ceph_end_io_read(inode);
1587 if (retry_op > HAVE_RETRIED && ret >= 0) {
1589 struct page *page = NULL;
1591 if (retry_op == READ_INLINE) {
1592 page = __page_cache_alloc(GFP_KERNEL);
1597 statret = __ceph_do_getattr(inode, page,
1598 CEPH_STAT_CAP_INLINE_DATA, !!page);
1602 if (statret == -ENODATA) {
1603 BUG_ON(retry_op != READ_INLINE);
1609 i_size = i_size_read(inode);
1610 if (retry_op == READ_INLINE) {
1611 BUG_ON(ret > 0 || read > 0);
1612 if (iocb->ki_pos < i_size &&
1613 iocb->ki_pos < PAGE_SIZE) {
1614 loff_t end = min_t(loff_t, i_size,
1615 iocb->ki_pos + len);
1616 end = min_t(loff_t, end, PAGE_SIZE);
1618 zero_user_segment(page, statret, end);
1619 ret = copy_page_to_iter(page,
1620 iocb->ki_pos & ~PAGE_MASK,
1621 end - iocb->ki_pos, to);
1622 iocb->ki_pos += ret;
1625 if (iocb->ki_pos < i_size && read < len) {
1626 size_t zlen = min_t(size_t, len - read,
1627 i_size - iocb->ki_pos);
1628 ret = iov_iter_zero(zlen, to);
1629 iocb->ki_pos += ret;
1632 __free_pages(page, 0);
1636 /* hit EOF or hole? */
1637 if (retry_op == CHECK_EOF && iocb->ki_pos < i_size &&
1639 dout("sync_read hit hole, ppos %lld < size %lld"
1640 ", reading more\n", iocb->ki_pos, i_size);
1644 retry_op = HAVE_RETRIED;
1656 * Take cap references to avoid releasing caps to MDS mid-write.
1658 * If we are synchronous, and write with an old snap context, the OSD
1659 * may return EOLDSNAPC. In that case, retry the write.. _after_
1660 * dropping our cap refs and allowing the pending snap to logically
1661 * complete _before_ this write occurs.
1663 * If we are near ENOSPC, write synchronously.
1665 static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
1667 struct file *file = iocb->ki_filp;
1668 struct ceph_file_info *fi = file->private_data;
1669 struct inode *inode = file_inode(file);
1670 struct ceph_inode_info *ci = ceph_inode(inode);
1671 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1672 struct ceph_osd_client *osdc = &fsc->client->osdc;
1673 struct ceph_cap_flush *prealloc_cf;
1674 ssize_t count, written = 0;
1676 bool direct_lock = false;
1680 loff_t limit = max(i_size_read(inode), fsc->max_file_size);
1682 if (ceph_snap(inode) != CEPH_NOSNAP)
1685 prealloc_cf = ceph_alloc_cap_flush();
1689 if ((iocb->ki_flags & (IOCB_DIRECT | IOCB_APPEND)) == IOCB_DIRECT)
1694 ceph_start_io_direct(inode);
1696 ceph_start_io_write(inode);
1698 /* We can write back this queue in page reclaim */
1699 current->backing_dev_info = inode_to_bdi(inode);
1701 if (iocb->ki_flags & IOCB_APPEND) {
1702 err = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
1707 err = generic_write_checks(iocb, from);
1712 if (unlikely(pos >= limit)) {
1716 iov_iter_truncate(from, limit - pos);
1719 count = iov_iter_count(from);
1720 if (ceph_quota_is_max_bytes_exceeded(inode, pos + count)) {
1725 err = file_remove_privs(file);
1729 err = file_update_time(file);
1733 inode_inc_iversion_raw(inode);
1735 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1736 err = ceph_uninline_data(file, NULL);
1741 down_read(&osdc->lock);
1742 map_flags = osdc->osdmap->flags;
1743 pool_flags = ceph_pg_pool_flags(osdc->osdmap, ci->i_layout.pool_id);
1744 up_read(&osdc->lock);
1745 if ((map_flags & CEPH_OSDMAP_FULL) ||
1746 (pool_flags & CEPH_POOL_FLAG_FULL)) {
1751 dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
1752 inode, ceph_vinop(inode), pos, count, i_size_read(inode));
1753 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1754 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1756 want = CEPH_CAP_FILE_BUFFER;
1758 err = ceph_get_caps(file, CEPH_CAP_FILE_WR, want, pos + count, &got);
1762 dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
1763 inode, ceph_vinop(inode), pos, count, ceph_cap_string(got));
1765 if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 ||
1766 (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC) ||
1767 (ci->i_ceph_flags & CEPH_I_ERROR_WRITE)) {
1768 struct ceph_snap_context *snapc;
1769 struct iov_iter data;
1771 spin_lock(&ci->i_ceph_lock);
1772 if (__ceph_have_pending_cap_snap(ci)) {
1773 struct ceph_cap_snap *capsnap =
1774 list_last_entry(&ci->i_cap_snaps,
1775 struct ceph_cap_snap,
1777 snapc = ceph_get_snap_context(capsnap->context);
1779 BUG_ON(!ci->i_head_snapc);
1780 snapc = ceph_get_snap_context(ci->i_head_snapc);
1782 spin_unlock(&ci->i_ceph_lock);
1784 /* we might need to revert back to that point */
1786 if (iocb->ki_flags & IOCB_DIRECT)
1787 written = ceph_direct_read_write(iocb, &data, snapc,
1790 written = ceph_sync_write(iocb, &data, pos, snapc);
1792 ceph_end_io_direct(inode);
1794 ceph_end_io_write(inode);
1796 iov_iter_advance(from, written);
1797 ceph_put_snap_context(snapc);
1800 * No need to acquire the i_truncate_mutex. Because
1801 * the MDS revokes Fwb caps before sending truncate
1802 * message to us. We can't get Fwb cap while there
1803 * are pending vmtruncate. So write and vmtruncate
1804 * can not run at the same time
1806 written = generic_perform_write(file, from, pos);
1807 if (likely(written >= 0))
1808 iocb->ki_pos = pos + written;
1809 ceph_end_io_write(inode);
1815 spin_lock(&ci->i_ceph_lock);
1816 ci->i_inline_version = CEPH_INLINE_NONE;
1817 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1819 spin_unlock(&ci->i_ceph_lock);
1821 __mark_inode_dirty(inode, dirty);
1822 if (ceph_quota_is_max_bytes_approaching(inode, iocb->ki_pos))
1823 ceph_check_caps(ci, 0, NULL);
1826 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
1827 inode, ceph_vinop(inode), pos, (unsigned)count,
1828 ceph_cap_string(got));
1829 ceph_put_cap_refs(ci, got);
1831 if (written == -EOLDSNAPC) {
1832 dout("aio_write %p %llx.%llx %llu~%u" "got EOLDSNAPC, retrying\n",
1833 inode, ceph_vinop(inode), pos, (unsigned)count);
1838 if ((map_flags & CEPH_OSDMAP_NEARFULL) ||
1839 (pool_flags & CEPH_POOL_FLAG_NEARFULL))
1840 iocb->ki_flags |= IOCB_DSYNC;
1841 written = generic_write_sync(iocb, written);
1847 ceph_end_io_direct(inode);
1849 ceph_end_io_write(inode);
1851 ceph_free_cap_flush(prealloc_cf);
1852 current->backing_dev_info = NULL;
1853 return written ? written : err;
1857 * llseek. be sure to verify file size on SEEK_END.
1859 static loff_t ceph_llseek(struct file *file, loff_t offset, int whence)
1861 struct inode *inode = file->f_mapping->host;
1862 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1868 if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) {
1869 ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
1874 i_size = i_size_read(inode);
1881 * Here we special-case the lseek(fd, 0, SEEK_CUR)
1882 * position-querying operation. Avoid rewriting the "same"
1883 * f_pos value back to the file because a concurrent read(),
1884 * write() or lseek() might have altered it
1890 offset += file->f_pos;
1893 if (offset < 0 || offset >= i_size) {
1899 if (offset < 0 || offset >= i_size) {
1907 ret = vfs_setpos(file, offset, max(i_size, fsc->max_file_size));
1910 inode_unlock(inode);
1914 static inline void ceph_zero_partial_page(
1915 struct inode *inode, loff_t offset, unsigned size)
1918 pgoff_t index = offset >> PAGE_SHIFT;
1920 page = find_lock_page(inode->i_mapping, index);
1922 wait_on_page_writeback(page);
1923 zero_user(page, offset & (PAGE_SIZE - 1), size);
1929 static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset,
1932 loff_t nearly = round_up(offset, PAGE_SIZE);
1933 if (offset < nearly) {
1934 loff_t size = nearly - offset;
1937 ceph_zero_partial_page(inode, offset, size);
1941 if (length >= PAGE_SIZE) {
1942 loff_t size = round_down(length, PAGE_SIZE);
1943 truncate_pagecache_range(inode, offset, offset + size - 1);
1948 ceph_zero_partial_page(inode, offset, length);
1951 static int ceph_zero_partial_object(struct inode *inode,
1952 loff_t offset, loff_t *length)
1954 struct ceph_inode_info *ci = ceph_inode(inode);
1955 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1956 struct ceph_osd_request *req;
1962 op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE;
1965 op = CEPH_OSD_OP_ZERO;
1968 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1972 CEPH_OSD_FLAG_WRITE,
1979 req->r_mtime = inode->i_mtime;
1980 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1982 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1986 ceph_osdc_put_request(req);
1992 static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length)
1995 struct ceph_inode_info *ci = ceph_inode(inode);
1996 s32 stripe_unit = ci->i_layout.stripe_unit;
1997 s32 stripe_count = ci->i_layout.stripe_count;
1998 s32 object_size = ci->i_layout.object_size;
1999 u64 object_set_size = object_size * stripe_count;
2002 /* round offset up to next period boundary */
2003 nearly = offset + object_set_size - 1;
2005 nearly -= do_div(t, object_set_size);
2007 while (length && offset < nearly) {
2008 loff_t size = length;
2009 ret = ceph_zero_partial_object(inode, offset, &size);
2015 while (length >= object_set_size) {
2017 loff_t pos = offset;
2018 for (i = 0; i < stripe_count; ++i) {
2019 ret = ceph_zero_partial_object(inode, pos, NULL);
2024 offset += object_set_size;
2025 length -= object_set_size;
2028 loff_t size = length;
2029 ret = ceph_zero_partial_object(inode, offset, &size);
2038 static long ceph_fallocate(struct file *file, int mode,
2039 loff_t offset, loff_t length)
2041 struct ceph_file_info *fi = file->private_data;
2042 struct inode *inode = file_inode(file);
2043 struct ceph_inode_info *ci = ceph_inode(inode);
2044 struct ceph_cap_flush *prealloc_cf;
2051 if (mode != (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2054 if (!S_ISREG(inode->i_mode))
2057 prealloc_cf = ceph_alloc_cap_flush();
2063 if (ceph_snap(inode) != CEPH_NOSNAP) {
2068 if (ci->i_inline_version != CEPH_INLINE_NONE) {
2069 ret = ceph_uninline_data(file, NULL);
2074 size = i_size_read(inode);
2076 /* Are we punching a hole beyond EOF? */
2079 if ((offset + length) > size)
2080 length = size - offset;
2082 if (fi->fmode & CEPH_FILE_MODE_LAZY)
2083 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
2085 want = CEPH_CAP_FILE_BUFFER;
2087 ret = ceph_get_caps(file, CEPH_CAP_FILE_WR, want, endoff, &got);
2091 ceph_zero_pagecache_range(inode, offset, length);
2092 ret = ceph_zero_objects(inode, offset, length);
2095 spin_lock(&ci->i_ceph_lock);
2096 ci->i_inline_version = CEPH_INLINE_NONE;
2097 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
2099 spin_unlock(&ci->i_ceph_lock);
2101 __mark_inode_dirty(inode, dirty);
2104 ceph_put_cap_refs(ci, got);
2106 inode_unlock(inode);
2107 ceph_free_cap_flush(prealloc_cf);
2112 * This function tries to get FILE_WR capabilities for dst_ci and FILE_RD for
2113 * src_ci. Two attempts are made to obtain both caps, and an error is return if
2114 * this fails; zero is returned on success.
2116 static int get_rd_wr_caps(struct file *src_filp, int *src_got,
2117 struct file *dst_filp,
2118 loff_t dst_endoff, int *dst_got)
2121 bool retrying = false;
2124 ret = ceph_get_caps(dst_filp, CEPH_CAP_FILE_WR, CEPH_CAP_FILE_BUFFER,
2125 dst_endoff, dst_got);
2130 * Since we're already holding the FILE_WR capability for the dst file,
2131 * we would risk a deadlock by using ceph_get_caps. Thus, we'll do some
2132 * retry dance instead to try to get both capabilities.
2134 ret = ceph_try_get_caps(file_inode(src_filp),
2135 CEPH_CAP_FILE_RD, CEPH_CAP_FILE_SHARED,
2138 /* Start by dropping dst_ci caps and getting src_ci caps */
2139 ceph_put_cap_refs(ceph_inode(file_inode(dst_filp)), *dst_got);
2142 /* ceph_try_get_caps masks EAGAIN */
2146 ret = ceph_get_caps(src_filp, CEPH_CAP_FILE_RD,
2147 CEPH_CAP_FILE_SHARED, -1, src_got);
2150 /*... drop src_ci caps too, and retry */
2151 ceph_put_cap_refs(ceph_inode(file_inode(src_filp)), *src_got);
2158 static void put_rd_wr_caps(struct ceph_inode_info *src_ci, int src_got,
2159 struct ceph_inode_info *dst_ci, int dst_got)
2161 ceph_put_cap_refs(src_ci, src_got);
2162 ceph_put_cap_refs(dst_ci, dst_got);
2166 * This function does several size-related checks, returning an error if:
2167 * - source file is smaller than off+len
2168 * - destination file size is not OK (inode_newsize_ok())
2169 * - max bytes quotas is exceeded
2171 static int is_file_size_ok(struct inode *src_inode, struct inode *dst_inode,
2172 loff_t src_off, loff_t dst_off, size_t len)
2174 loff_t size, endoff;
2176 size = i_size_read(src_inode);
2178 * Don't copy beyond source file EOF. Instead of simply setting length
2179 * to (size - src_off), just drop to VFS default implementation, as the
2180 * local i_size may be stale due to other clients writing to the source
2183 if (src_off + len > size) {
2184 dout("Copy beyond EOF (%llu + %zu > %llu)\n",
2185 src_off, len, size);
2188 size = i_size_read(dst_inode);
2190 endoff = dst_off + len;
2191 if (inode_newsize_ok(dst_inode, endoff))
2194 if (ceph_quota_is_max_bytes_exceeded(dst_inode, endoff))
2200 static ssize_t ceph_do_objects_copy(struct ceph_inode_info *src_ci, u64 *src_off,
2201 struct ceph_inode_info *dst_ci, u64 *dst_off,
2202 struct ceph_fs_client *fsc,
2203 size_t len, unsigned int flags)
2205 struct ceph_object_locator src_oloc, dst_oloc;
2206 struct ceph_object_id src_oid, dst_oid;
2208 u64 src_objnum, src_objoff, dst_objnum, dst_objoff;
2209 u32 src_objlen, dst_objlen;
2210 u32 object_size = src_ci->i_layout.object_size;
2213 src_oloc.pool = src_ci->i_layout.pool_id;
2214 src_oloc.pool_ns = ceph_try_get_string(src_ci->i_layout.pool_ns);
2215 dst_oloc.pool = dst_ci->i_layout.pool_id;
2216 dst_oloc.pool_ns = ceph_try_get_string(dst_ci->i_layout.pool_ns);
2218 while (len >= object_size) {
2219 ceph_calc_file_object_mapping(&src_ci->i_layout, *src_off,
2220 object_size, &src_objnum,
2221 &src_objoff, &src_objlen);
2222 ceph_calc_file_object_mapping(&dst_ci->i_layout, *dst_off,
2223 object_size, &dst_objnum,
2224 &dst_objoff, &dst_objlen);
2225 ceph_oid_init(&src_oid);
2226 ceph_oid_printf(&src_oid, "%llx.%08llx",
2227 src_ci->i_vino.ino, src_objnum);
2228 ceph_oid_init(&dst_oid);
2229 ceph_oid_printf(&dst_oid, "%llx.%08llx",
2230 dst_ci->i_vino.ino, dst_objnum);
2231 /* Do an object remote copy */
2232 ret = ceph_osdc_copy_from(&fsc->client->osdc,
2233 src_ci->i_vino.snap, 0,
2234 &src_oid, &src_oloc,
2235 CEPH_OSD_OP_FLAG_FADVISE_SEQUENTIAL |
2236 CEPH_OSD_OP_FLAG_FADVISE_NOCACHE,
2237 &dst_oid, &dst_oloc,
2238 CEPH_OSD_OP_FLAG_FADVISE_SEQUENTIAL |
2239 CEPH_OSD_OP_FLAG_FADVISE_DONTNEED,
2240 dst_ci->i_truncate_seq,
2241 dst_ci->i_truncate_size,
2242 CEPH_OSD_COPY_FROM_FLAG_TRUNCATE_SEQ);
2244 if (ret == -EOPNOTSUPP) {
2245 fsc->have_copy_from2 = false;
2246 pr_notice("OSDs don't support copy-from2; disabling copy offload\n");
2248 dout("ceph_osdc_copy_from returned %d\n", ret);
2254 bytes += object_size;
2255 *src_off += object_size;
2256 *dst_off += object_size;
2260 ceph_oloc_destroy(&src_oloc);
2261 ceph_oloc_destroy(&dst_oloc);
2265 static ssize_t __ceph_copy_file_range(struct file *src_file, loff_t src_off,
2266 struct file *dst_file, loff_t dst_off,
2267 size_t len, unsigned int flags)
2269 struct inode *src_inode = file_inode(src_file);
2270 struct inode *dst_inode = file_inode(dst_file);
2271 struct ceph_inode_info *src_ci = ceph_inode(src_inode);
2272 struct ceph_inode_info *dst_ci = ceph_inode(dst_inode);
2273 struct ceph_cap_flush *prealloc_cf;
2274 struct ceph_fs_client *src_fsc = ceph_inode_to_client(src_inode);
2276 ssize_t ret = -EIO, bytes;
2277 u64 src_objnum, dst_objnum, src_objoff, dst_objoff;
2278 u32 src_objlen, dst_objlen;
2279 int src_got = 0, dst_got = 0, err, dirty;
2281 if (src_inode->i_sb != dst_inode->i_sb) {
2282 struct ceph_fs_client *dst_fsc = ceph_inode_to_client(dst_inode);
2284 if (ceph_fsid_compare(&src_fsc->client->fsid,
2285 &dst_fsc->client->fsid)) {
2286 dout("Copying files across clusters: src: %pU dst: %pU\n",
2287 &src_fsc->client->fsid, &dst_fsc->client->fsid);
2291 if (ceph_snap(dst_inode) != CEPH_NOSNAP)
2295 * Some of the checks below will return -EOPNOTSUPP, which will force a
2296 * fallback to the default VFS copy_file_range implementation. This is
2297 * desirable in several cases (for ex, the 'len' is smaller than the
2298 * size of the objects, or in cases where that would be more
2302 if (ceph_test_mount_opt(src_fsc, NOCOPYFROM))
2305 if (!src_fsc->have_copy_from2)
2309 * Striped file layouts require that we copy partial objects, but the
2310 * OSD copy-from operation only supports full-object copies. Limit
2311 * this to non-striped file layouts for now.
2313 if ((src_ci->i_layout.stripe_unit != dst_ci->i_layout.stripe_unit) ||
2314 (src_ci->i_layout.stripe_count != 1) ||
2315 (dst_ci->i_layout.stripe_count != 1) ||
2316 (src_ci->i_layout.object_size != dst_ci->i_layout.object_size)) {
2317 dout("Invalid src/dst files layout\n");
2321 if (len < src_ci->i_layout.object_size)
2322 return -EOPNOTSUPP; /* no remote copy will be done */
2324 prealloc_cf = ceph_alloc_cap_flush();
2328 /* Start by sync'ing the source and destination files */
2329 ret = file_write_and_wait_range(src_file, src_off, (src_off + len));
2331 dout("failed to write src file (%zd)\n", ret);
2334 ret = file_write_and_wait_range(dst_file, dst_off, (dst_off + len));
2336 dout("failed to write dst file (%zd)\n", ret);
2341 * We need FILE_WR caps for dst_ci and FILE_RD for src_ci as other
2342 * clients may have dirty data in their caches. And OSDs know nothing
2343 * about caps, so they can't safely do the remote object copies.
2345 err = get_rd_wr_caps(src_file, &src_got,
2346 dst_file, (dst_off + len), &dst_got);
2348 dout("get_rd_wr_caps returned %d\n", err);
2353 ret = is_file_size_ok(src_inode, dst_inode, src_off, dst_off, len);
2357 /* Drop dst file cached pages */
2358 ret = invalidate_inode_pages2_range(dst_inode->i_mapping,
2359 dst_off >> PAGE_SHIFT,
2360 (dst_off + len) >> PAGE_SHIFT);
2362 dout("Failed to invalidate inode pages (%zd)\n", ret);
2365 ceph_calc_file_object_mapping(&src_ci->i_layout, src_off,
2366 src_ci->i_layout.object_size,
2367 &src_objnum, &src_objoff, &src_objlen);
2368 ceph_calc_file_object_mapping(&dst_ci->i_layout, dst_off,
2369 dst_ci->i_layout.object_size,
2370 &dst_objnum, &dst_objoff, &dst_objlen);
2371 /* object-level offsets need to the same */
2372 if (src_objoff != dst_objoff) {
2378 * Do a manual copy if the object offset isn't object aligned.
2379 * 'src_objlen' contains the bytes left until the end of the object,
2380 * starting at the src_off
2383 dout("Initial partial copy of %u bytes\n", src_objlen);
2386 * we need to temporarily drop all caps as we'll be calling
2387 * {read,write}_iter, which will get caps again.
2389 put_rd_wr_caps(src_ci, src_got, dst_ci, dst_got);
2390 ret = do_splice_direct(src_file, &src_off, dst_file,
2391 &dst_off, src_objlen, flags);
2392 /* Abort on short copies or on error */
2393 if (ret < src_objlen) {
2394 dout("Failed partial copy (%zd)\n", ret);
2398 err = get_rd_wr_caps(src_file, &src_got,
2399 dst_file, (dst_off + len), &dst_got);
2402 err = is_file_size_ok(src_inode, dst_inode,
2403 src_off, dst_off, len);
2408 size = i_size_read(dst_inode);
2409 bytes = ceph_do_objects_copy(src_ci, &src_off, dst_ci, &dst_off,
2410 src_fsc, len, flags);
2416 dout("Copied %zu bytes out of %zu\n", bytes, len);
2420 file_update_time(dst_file);
2421 inode_inc_iversion_raw(dst_inode);
2423 if (dst_off > size) {
2424 /* Let the MDS know about dst file size change */
2425 if (ceph_inode_set_size(dst_inode, dst_off) ||
2426 ceph_quota_is_max_bytes_approaching(dst_inode, dst_off))
2427 ceph_check_caps(dst_ci, CHECK_CAPS_AUTHONLY, NULL);
2430 spin_lock(&dst_ci->i_ceph_lock);
2431 dst_ci->i_inline_version = CEPH_INLINE_NONE;
2432 dirty = __ceph_mark_dirty_caps(dst_ci, CEPH_CAP_FILE_WR, &prealloc_cf);
2433 spin_unlock(&dst_ci->i_ceph_lock);
2435 __mark_inode_dirty(dst_inode, dirty);
2438 put_rd_wr_caps(src_ci, src_got, dst_ci, dst_got);
2441 * Do the final manual copy if we still have some bytes left, unless
2442 * there were errors in remote object copies (len >= object_size).
2444 if (len && (len < src_ci->i_layout.object_size)) {
2445 dout("Final partial copy of %zu bytes\n", len);
2446 bytes = do_splice_direct(src_file, &src_off, dst_file,
2447 &dst_off, len, flags);
2451 dout("Failed partial copy (%zd)\n", bytes);
2455 ceph_free_cap_flush(prealloc_cf);
2460 static ssize_t ceph_copy_file_range(struct file *src_file, loff_t src_off,
2461 struct file *dst_file, loff_t dst_off,
2462 size_t len, unsigned int flags)
2466 ret = __ceph_copy_file_range(src_file, src_off, dst_file, dst_off,
2469 if (ret == -EOPNOTSUPP || ret == -EXDEV)
2470 ret = generic_copy_file_range(src_file, src_off, dst_file,
2471 dst_off, len, flags);
2475 const struct file_operations ceph_file_fops = {
2477 .release = ceph_release,
2478 .llseek = ceph_llseek,
2479 .read_iter = ceph_read_iter,
2480 .write_iter = ceph_write_iter,
2482 .fsync = ceph_fsync,
2484 .setlease = simple_nosetlease,
2485 .flock = ceph_flock,
2486 .splice_read = generic_file_splice_read,
2487 .splice_write = iter_file_splice_write,
2488 .unlocked_ioctl = ceph_ioctl,
2489 .compat_ioctl = compat_ptr_ioctl,
2490 .fallocate = ceph_fallocate,
2491 .copy_file_range = ceph_copy_file_range,