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_mount_options *opt =
208 ceph_inode_to_client(&ci->vfs_inode)->mount_options;
209 struct ceph_file_info *fi;
211 dout("%s %p %p 0%o (%s)\n", __func__, inode, file,
212 inode->i_mode, isdir ? "dir" : "regular");
213 BUG_ON(inode->i_fop->release != ceph_release);
216 struct ceph_dir_file_info *dfi =
217 kmem_cache_zalloc(ceph_dir_file_cachep, GFP_KERNEL);
221 file->private_data = dfi;
222 fi = &dfi->file_info;
223 dfi->next_offset = 2;
224 dfi->readdir_cache_idx = -1;
226 fi = kmem_cache_zalloc(ceph_file_cachep, GFP_KERNEL);
230 if (opt->flags & CEPH_MOUNT_OPT_NOPAGECACHE)
231 fi->flags |= CEPH_F_SYNC;
233 file->private_data = fi;
236 ceph_get_fmode(ci, fmode, 1);
239 spin_lock_init(&fi->rw_contexts_lock);
240 INIT_LIST_HEAD(&fi->rw_contexts);
241 fi->filp_gen = READ_ONCE(ceph_inode_to_client(inode)->filp_gen);
247 * initialize private struct file data.
248 * if we fail, clean up by dropping fmode reference on the ceph_inode
250 static int ceph_init_file(struct inode *inode, struct file *file, int fmode)
254 switch (inode->i_mode & S_IFMT) {
256 ceph_fscache_use_cookie(inode, file->f_mode & FMODE_WRITE);
259 ret = ceph_init_file_info(inode, file, fmode,
260 S_ISDIR(inode->i_mode));
264 dout("init_file %p %p 0%o (symlink)\n", inode, file,
269 dout("init_file %p %p 0%o (special)\n", inode, file,
272 * we need to drop the open ref now, since we don't
273 * have .release set to ceph_release.
275 BUG_ON(inode->i_fop->release == ceph_release);
277 /* call the proper open fop */
278 ret = inode->i_fop->open(inode, file);
284 * try renew caps after session gets killed.
286 int ceph_renew_caps(struct inode *inode, int fmode)
288 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
289 struct ceph_inode_info *ci = ceph_inode(inode);
290 struct ceph_mds_request *req;
291 int err, flags, wanted;
293 spin_lock(&ci->i_ceph_lock);
294 __ceph_touch_fmode(ci, mdsc, fmode);
295 wanted = __ceph_caps_file_wanted(ci);
296 if (__ceph_is_any_real_caps(ci) &&
297 (!(wanted & CEPH_CAP_ANY_WR) || ci->i_auth_cap)) {
298 int issued = __ceph_caps_issued(ci, NULL);
299 spin_unlock(&ci->i_ceph_lock);
300 dout("renew caps %p want %s issued %s updating mds_wanted\n",
301 inode, ceph_cap_string(wanted), ceph_cap_string(issued));
302 ceph_check_caps(ci, 0, NULL);
305 spin_unlock(&ci->i_ceph_lock);
308 if ((wanted & CEPH_CAP_FILE_RD) && (wanted & CEPH_CAP_FILE_WR))
310 else if (wanted & CEPH_CAP_FILE_RD)
312 else if (wanted & CEPH_CAP_FILE_WR)
315 if (wanted & CEPH_CAP_FILE_LAZYIO)
319 req = prepare_open_request(inode->i_sb, flags, 0);
325 req->r_inode = inode;
329 err = ceph_mdsc_do_request(mdsc, NULL, req);
330 ceph_mdsc_put_request(req);
332 dout("renew caps %p open result=%d\n", inode, err);
333 return err < 0 ? err : 0;
337 * If we already have the requisite capabilities, we can satisfy
338 * the open request locally (no need to request new caps from the
339 * MDS). We do, however, need to inform the MDS (asynchronously)
340 * if our wanted caps set expands.
342 int ceph_open(struct inode *inode, struct file *file)
344 struct ceph_inode_info *ci = ceph_inode(inode);
345 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
346 struct ceph_mds_client *mdsc = fsc->mdsc;
347 struct ceph_mds_request *req;
348 struct ceph_file_info *fi = file->private_data;
350 int flags, fmode, wanted;
353 dout("open file %p is already opened\n", file);
357 /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */
358 flags = file->f_flags & ~(O_CREAT|O_EXCL);
359 if (S_ISDIR(inode->i_mode))
360 flags = O_DIRECTORY; /* mds likes to know */
362 dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode,
363 ceph_vinop(inode), file, flags, file->f_flags);
364 fmode = ceph_flags_to_mode(flags);
365 wanted = ceph_caps_for_mode(fmode);
367 /* snapped files are read-only */
368 if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE))
371 /* trivially open snapdir */
372 if (ceph_snap(inode) == CEPH_SNAPDIR) {
373 return ceph_init_file(inode, file, fmode);
377 * No need to block if we have caps on the auth MDS (for
378 * write) or any MDS (for read). Update wanted set
381 spin_lock(&ci->i_ceph_lock);
382 if (__ceph_is_any_real_caps(ci) &&
383 (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) {
384 int mds_wanted = __ceph_caps_mds_wanted(ci, true);
385 int issued = __ceph_caps_issued(ci, NULL);
387 dout("open %p fmode %d want %s issued %s using existing\n",
388 inode, fmode, ceph_cap_string(wanted),
389 ceph_cap_string(issued));
390 __ceph_touch_fmode(ci, mdsc, fmode);
391 spin_unlock(&ci->i_ceph_lock);
394 if ((issued & wanted) != wanted &&
395 (mds_wanted & wanted) != wanted &&
396 ceph_snap(inode) != CEPH_SNAPDIR)
397 ceph_check_caps(ci, 0, NULL);
399 return ceph_init_file(inode, file, fmode);
400 } else if (ceph_snap(inode) != CEPH_NOSNAP &&
401 (ci->i_snap_caps & wanted) == wanted) {
402 __ceph_touch_fmode(ci, mdsc, fmode);
403 spin_unlock(&ci->i_ceph_lock);
404 return ceph_init_file(inode, file, fmode);
407 spin_unlock(&ci->i_ceph_lock);
409 dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted));
410 req = prepare_open_request(inode->i_sb, flags, 0);
415 req->r_inode = inode;
419 err = ceph_mdsc_do_request(mdsc, NULL, req);
421 err = ceph_init_file(inode, file, req->r_fmode);
422 ceph_mdsc_put_request(req);
423 dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode));
428 /* Clone the layout from a synchronous create, if the dir now has Dc caps */
430 cache_file_layout(struct inode *dst, struct inode *src)
432 struct ceph_inode_info *cdst = ceph_inode(dst);
433 struct ceph_inode_info *csrc = ceph_inode(src);
435 spin_lock(&cdst->i_ceph_lock);
436 if ((__ceph_caps_issued(cdst, NULL) & CEPH_CAP_DIR_CREATE) &&
437 !ceph_file_layout_is_valid(&cdst->i_cached_layout)) {
438 memcpy(&cdst->i_cached_layout, &csrc->i_layout,
439 sizeof(cdst->i_cached_layout));
440 rcu_assign_pointer(cdst->i_cached_layout.pool_ns,
441 ceph_try_get_string(csrc->i_layout.pool_ns));
443 spin_unlock(&cdst->i_ceph_lock);
447 * Try to set up an async create. We need caps, a file layout, and inode number,
448 * and either a lease on the dentry or complete dir info. If any of those
449 * criteria are not satisfied, then return false and the caller can go
452 static int try_prep_async_create(struct inode *dir, struct dentry *dentry,
453 struct ceph_file_layout *lo, u64 *pino)
455 struct ceph_inode_info *ci = ceph_inode(dir);
456 struct ceph_dentry_info *di = ceph_dentry(dentry);
457 int got = 0, want = CEPH_CAP_FILE_EXCL | CEPH_CAP_DIR_CREATE;
460 spin_lock(&ci->i_ceph_lock);
461 /* No auth cap means no chance for Dc caps */
465 /* Any delegated inos? */
466 if (xa_empty(&ci->i_auth_cap->session->s_delegated_inos))
469 if (!ceph_file_layout_is_valid(&ci->i_cached_layout))
472 if ((__ceph_caps_issued(ci, NULL) & want) != want)
475 if (d_in_lookup(dentry)) {
476 if (!__ceph_dir_is_complete(ci))
478 spin_lock(&dentry->d_lock);
479 di->lease_shared_gen = atomic_read(&ci->i_shared_gen);
480 spin_unlock(&dentry->d_lock);
481 } else if (atomic_read(&ci->i_shared_gen) !=
482 READ_ONCE(di->lease_shared_gen)) {
486 ino = ceph_get_deleg_ino(ci->i_auth_cap->session);
491 ceph_take_cap_refs(ci, want, false);
492 memcpy(lo, &ci->i_cached_layout, sizeof(*lo));
493 rcu_assign_pointer(lo->pool_ns,
494 ceph_try_get_string(ci->i_cached_layout.pool_ns));
497 spin_unlock(&ci->i_ceph_lock);
501 static void restore_deleg_ino(struct inode *dir, u64 ino)
503 struct ceph_inode_info *ci = ceph_inode(dir);
504 struct ceph_mds_session *s = NULL;
506 spin_lock(&ci->i_ceph_lock);
508 s = ceph_get_mds_session(ci->i_auth_cap->session);
509 spin_unlock(&ci->i_ceph_lock);
511 int err = ceph_restore_deleg_ino(s, ino);
513 pr_warn("ceph: unable to restore delegated ino 0x%llx to session: %d\n",
515 ceph_put_mds_session(s);
519 static void ceph_async_create_cb(struct ceph_mds_client *mdsc,
520 struct ceph_mds_request *req)
522 int result = req->r_err ? req->r_err :
523 le32_to_cpu(req->r_reply_info.head->result);
525 if (result == -EJUKEBOX)
528 mapping_set_error(req->r_parent->i_mapping, result);
531 struct dentry *dentry = req->r_dentry;
532 struct inode *inode = d_inode(dentry);
535 char *path = ceph_mdsc_build_path(req->r_dentry, &pathlen,
538 ceph_dir_clear_complete(req->r_parent);
539 if (!d_unhashed(dentry))
542 ceph_inode_shutdown(inode);
544 pr_warn("ceph: async create failure path=(%llx)%s result=%d!\n",
545 base, IS_ERR(path) ? "<<bad>>" : path, result);
546 ceph_mdsc_free_path(path, pathlen);
549 if (req->r_target_inode) {
550 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
551 u64 ino = ceph_vino(req->r_target_inode).ino;
553 if (req->r_deleg_ino != ino)
554 pr_warn("%s: inode number mismatch! err=%d deleg_ino=0x%llx target=0x%llx\n",
555 __func__, req->r_err, req->r_deleg_ino, ino);
556 mapping_set_error(req->r_target_inode->i_mapping, result);
558 spin_lock(&ci->i_ceph_lock);
559 if (ci->i_ceph_flags & CEPH_I_ASYNC_CREATE) {
560 ci->i_ceph_flags &= ~CEPH_I_ASYNC_CREATE;
561 wake_up_bit(&ci->i_ceph_flags, CEPH_ASYNC_CREATE_BIT);
563 ceph_kick_flushing_inode_caps(req->r_session, ci);
564 spin_unlock(&ci->i_ceph_lock);
565 } else if (!result) {
566 pr_warn("%s: no req->r_target_inode for 0x%llx\n", __func__,
570 ceph_mdsc_release_dir_caps(req);
573 static int ceph_finish_async_create(struct inode *dir, struct dentry *dentry,
574 struct file *file, umode_t mode,
575 struct ceph_mds_request *req,
576 struct ceph_acl_sec_ctx *as_ctx,
577 struct ceph_file_layout *lo)
581 struct ceph_mds_reply_inode in = { };
582 struct ceph_mds_reply_info_in iinfo = { .in = &in };
583 struct ceph_inode_info *ci = ceph_inode(dir);
585 struct timespec64 now;
586 struct ceph_string *pool_ns;
587 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
588 struct ceph_vino vino = { .ino = req->r_deleg_ino,
589 .snap = CEPH_NOSNAP };
591 ktime_get_real_ts64(&now);
593 inode = ceph_get_inode(dentry->d_sb, vino);
595 return PTR_ERR(inode);
597 iinfo.inline_version = CEPH_INLINE_NONE;
598 iinfo.change_attr = 1;
599 ceph_encode_timespec64(&iinfo.btime, &now);
601 iinfo.xattr_len = ARRAY_SIZE(xattr_buf);
602 iinfo.xattr_data = xattr_buf;
603 memset(iinfo.xattr_data, 0, iinfo.xattr_len);
605 in.ino = cpu_to_le64(vino.ino);
606 in.snapid = cpu_to_le64(CEPH_NOSNAP);
607 in.version = cpu_to_le64(1); // ???
608 in.cap.caps = in.cap.wanted = cpu_to_le32(CEPH_CAP_ALL_FILE);
609 in.cap.cap_id = cpu_to_le64(1);
610 in.cap.realm = cpu_to_le64(ci->i_snap_realm->ino);
611 in.cap.flags = CEPH_CAP_FLAG_AUTH;
612 in.ctime = in.mtime = in.atime = iinfo.btime;
613 in.truncate_seq = cpu_to_le32(1);
614 in.truncate_size = cpu_to_le64(-1ULL);
615 in.xattr_version = cpu_to_le64(1);
616 in.uid = cpu_to_le32(from_kuid(&init_user_ns, current_fsuid()));
617 if (dir->i_mode & S_ISGID) {
618 in.gid = cpu_to_le32(from_kgid(&init_user_ns, dir->i_gid));
620 /* Directories always inherit the setgid bit. */
623 else if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP) &&
624 !in_group_p(dir->i_gid) &&
625 !capable_wrt_inode_uidgid(&init_user_ns, dir, CAP_FSETID))
628 in.gid = cpu_to_le32(from_kgid(&init_user_ns, current_fsgid()));
630 in.mode = cpu_to_le32((u32)mode);
632 in.nlink = cpu_to_le32(1);
633 in.max_size = cpu_to_le64(lo->stripe_unit);
635 ceph_file_layout_to_legacy(lo, &in.layout);
636 /* lo is private, so pool_ns can't change */
637 pool_ns = rcu_dereference_raw(lo->pool_ns);
639 iinfo.pool_ns_len = pool_ns->len;
640 iinfo.pool_ns_data = pool_ns->str;
643 down_read(&mdsc->snap_rwsem);
644 ret = ceph_fill_inode(inode, NULL, &iinfo, NULL, req->r_session,
646 up_read(&mdsc->snap_rwsem);
648 dout("%s failed to fill inode: %d\n", __func__, ret);
649 ceph_dir_clear_complete(dir);
650 if (!d_unhashed(dentry))
652 if (inode->i_state & I_NEW)
653 discard_new_inode(inode);
657 dout("%s d_adding new inode 0x%llx to 0x%llx/%s\n", __func__,
658 vino.ino, ceph_ino(dir), dentry->d_name.name);
659 ceph_dir_clear_ordered(dir);
660 ceph_init_inode_acls(inode, as_ctx);
661 if (inode->i_state & I_NEW) {
663 * If it's not I_NEW, then someone created this before
664 * we got here. Assume the server is aware of it at
665 * that point and don't worry about setting
666 * CEPH_I_ASYNC_CREATE.
668 ceph_inode(inode)->i_ceph_flags = CEPH_I_ASYNC_CREATE;
669 unlock_new_inode(inode);
671 if (d_in_lookup(dentry) || d_really_is_negative(dentry)) {
672 if (!d_unhashed(dentry))
674 dn = d_splice_alias(inode, dentry);
675 WARN_ON_ONCE(dn && dn != dentry);
677 file->f_mode |= FMODE_CREATED;
678 ret = finish_open(file, dentry, ceph_open);
684 * Do a lookup + open with a single request. If we get a non-existent
685 * file or symlink, return 1 so the VFS can retry.
687 int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
688 struct file *file, unsigned flags, umode_t mode)
690 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
691 struct ceph_mds_client *mdsc = fsc->mdsc;
692 struct ceph_mds_request *req;
694 struct ceph_acl_sec_ctx as_ctx = {};
695 bool try_async = ceph_test_mount_opt(fsc, ASYNC_DIROPS);
699 dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n",
701 d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode);
703 if (dentry->d_name.len > NAME_MAX)
704 return -ENAMETOOLONG;
706 if (flags & O_CREAT) {
707 if (ceph_quota_is_max_files_exceeded(dir))
709 err = ceph_pre_init_acls(dir, &mode, &as_ctx);
712 err = ceph_security_init_secctx(dentry, mode, &as_ctx);
715 } else if (!d_in_lookup(dentry)) {
716 /* If it's not being looked up, it's negative */
721 req = prepare_open_request(dir->i_sb, flags, mode);
726 req->r_dentry = dget(dentry);
728 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
729 if (ceph_security_xattr_wanted(dir))
730 mask |= CEPH_CAP_XATTR_SHARED;
731 req->r_args.open.mask = cpu_to_le32(mask);
735 if (flags & O_CREAT) {
736 struct ceph_file_layout lo;
738 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
739 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
740 if (as_ctx.pagelist) {
741 req->r_pagelist = as_ctx.pagelist;
742 as_ctx.pagelist = NULL;
746 try_prep_async_create(dir, dentry, &lo,
747 &req->r_deleg_ino))) {
748 set_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags);
749 req->r_args.open.flags |= cpu_to_le32(CEPH_O_EXCL);
750 req->r_callback = ceph_async_create_cb;
751 err = ceph_mdsc_submit_request(mdsc, dir, req);
753 err = ceph_finish_async_create(dir, dentry,
756 } else if (err == -EJUKEBOX) {
757 restore_deleg_ino(dir, req->r_deleg_ino);
758 ceph_mdsc_put_request(req);
760 ceph_put_string(rcu_dereference_raw(lo.pool_ns));
763 ceph_put_string(rcu_dereference_raw(lo.pool_ns));
768 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
769 err = ceph_mdsc_do_request(mdsc,
770 (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
772 if (err == -ENOENT) {
773 dentry = ceph_handle_snapdir(req, dentry);
774 if (IS_ERR(dentry)) {
775 err = PTR_ERR(dentry);
781 if (!err && (flags & O_CREAT) && !req->r_reply_info.head->is_dentry)
782 err = ceph_handle_notrace_create(dir, dentry);
784 if (d_in_lookup(dentry)) {
785 dn = ceph_finish_lookup(req, dentry, err);
789 /* we were given a hashed negative dentry */
794 if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) {
795 /* make vfs retry on splice, ENOENT, or symlink */
796 dout("atomic_open finish_no_open on dn %p\n", dn);
797 err = finish_no_open(file, dn);
799 dout("atomic_open finish_open on dn %p\n", dn);
800 if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) {
801 struct inode *newino = d_inode(dentry);
803 cache_file_layout(dir, newino);
804 ceph_init_inode_acls(newino, &as_ctx);
805 file->f_mode |= FMODE_CREATED;
807 err = finish_open(file, dentry, ceph_open);
810 ceph_mdsc_put_request(req);
812 ceph_release_acl_sec_ctx(&as_ctx);
813 dout("atomic_open result=%d\n", err);
817 int ceph_release(struct inode *inode, struct file *file)
819 struct ceph_inode_info *ci = ceph_inode(inode);
821 if (S_ISDIR(inode->i_mode)) {
822 struct ceph_dir_file_info *dfi = file->private_data;
823 dout("release inode %p dir file %p\n", inode, file);
824 WARN_ON(!list_empty(&dfi->file_info.rw_contexts));
826 ceph_put_fmode(ci, dfi->file_info.fmode, 1);
828 if (dfi->last_readdir)
829 ceph_mdsc_put_request(dfi->last_readdir);
830 kfree(dfi->last_name);
831 kfree(dfi->dir_info);
832 kmem_cache_free(ceph_dir_file_cachep, dfi);
834 struct ceph_file_info *fi = file->private_data;
835 dout("release inode %p regular file %p\n", inode, file);
836 WARN_ON(!list_empty(&fi->rw_contexts));
838 ceph_fscache_unuse_cookie(inode, file->f_mode & FMODE_WRITE);
839 ceph_put_fmode(ci, fi->fmode, 1);
841 kmem_cache_free(ceph_file_cachep, fi);
844 /* wake up anyone waiting for caps on this inode */
845 wake_up_all(&ci->i_cap_wq);
856 * Completely synchronous read and write methods. Direct from __user
857 * buffer to osd, or directly to user pages (if O_DIRECT).
859 * If the read spans object boundary, just do multiple reads. (That's not
860 * atomic, but good enough for now.)
862 * If we get a short result from the OSD, check against i_size; we need to
863 * only return a short read to the caller if we hit EOF.
865 static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *to,
868 struct file *file = iocb->ki_filp;
869 struct inode *inode = file_inode(file);
870 struct ceph_inode_info *ci = ceph_inode(inode);
871 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
872 struct ceph_osd_client *osdc = &fsc->client->osdc;
874 u64 off = iocb->ki_pos;
875 u64 len = iov_iter_count(to);
876 u64 i_size = i_size_read(inode);
878 dout("sync_read on file %p %llu~%u %s\n", file, off, (unsigned)len,
879 (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
884 * flush any page cache pages in this range. this
885 * will make concurrent normal and sync io slow,
886 * but it will at least behave sensibly when they are
889 ret = filemap_write_and_wait_range(inode->i_mapping,
895 while ((len = iov_iter_count(to)) > 0) {
896 struct ceph_osd_request *req;
904 req = ceph_osdc_new_request(osdc, &ci->i_layout,
905 ci->i_vino, off, &len, 0, 1,
906 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
907 NULL, ci->i_truncate_seq,
908 ci->i_truncate_size, false);
914 more = len < iov_iter_count(to);
916 num_pages = calc_pages_for(off, len);
917 page_off = off & ~PAGE_MASK;
918 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
920 ceph_osdc_put_request(req);
921 ret = PTR_ERR(pages);
925 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_off,
927 ret = ceph_osdc_start_request(osdc, req, false);
929 ret = ceph_osdc_wait_request(osdc, req);
931 ceph_update_read_metrics(&fsc->mdsc->metric,
932 req->r_start_latency,
936 ceph_osdc_put_request(req);
938 i_size = i_size_read(inode);
939 dout("sync_read %llu~%llu got %zd i_size %llu%s\n",
940 off, len, ret, i_size, (more ? " MORE" : ""));
944 if (ret >= 0 && ret < len && (off + ret < i_size)) {
945 int zlen = min(len - ret, i_size - off - ret);
946 int zoff = page_off + ret;
947 dout("sync_read zero gap %llu~%llu\n",
948 off + ret, off + ret + zlen);
949 ceph_zero_page_vector_range(zoff, zlen, pages);
954 left = ret > 0 ? ret : 0;
957 page_off = off & ~PAGE_MASK;
958 len = min_t(size_t, left, PAGE_SIZE - page_off);
959 SetPageUptodate(pages[idx]);
960 copied = copy_page_to_iter(pages[idx++],
969 ceph_release_page_vector(pages, num_pages);
972 if (ret == -EBLOCKLISTED)
973 fsc->blocklisted = true;
977 if (off >= i_size || !more)
981 if (off > iocb->ki_pos) {
983 *retry_op = CHECK_EOF;
984 ret = i_size - iocb->ki_pos;
985 iocb->ki_pos = i_size;
987 ret = off - iocb->ki_pos;
992 dout("sync_read result %zd retry_op %d\n", ret, *retry_op);
996 struct ceph_aio_request {
1002 struct list_head osd_reqs;
1004 atomic_t pending_reqs;
1005 struct timespec64 mtime;
1006 struct ceph_cap_flush *prealloc_cf;
1009 struct ceph_aio_work {
1010 struct work_struct work;
1011 struct ceph_osd_request *req;
1014 static void ceph_aio_retry_work(struct work_struct *work);
1016 static void ceph_aio_complete(struct inode *inode,
1017 struct ceph_aio_request *aio_req)
1019 struct ceph_inode_info *ci = ceph_inode(inode);
1022 if (!atomic_dec_and_test(&aio_req->pending_reqs))
1025 if (aio_req->iocb->ki_flags & IOCB_DIRECT)
1026 inode_dio_end(inode);
1028 ret = aio_req->error;
1030 ret = aio_req->total_len;
1032 dout("ceph_aio_complete %p rc %d\n", inode, ret);
1034 if (ret >= 0 && aio_req->write) {
1037 loff_t endoff = aio_req->iocb->ki_pos + aio_req->total_len;
1038 if (endoff > i_size_read(inode)) {
1039 if (ceph_inode_set_size(inode, endoff))
1040 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1043 spin_lock(&ci->i_ceph_lock);
1044 ci->i_inline_version = CEPH_INLINE_NONE;
1045 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1046 &aio_req->prealloc_cf);
1047 spin_unlock(&ci->i_ceph_lock);
1049 __mark_inode_dirty(inode, dirty);
1053 ceph_put_cap_refs(ci, (aio_req->write ? CEPH_CAP_FILE_WR :
1056 aio_req->iocb->ki_complete(aio_req->iocb, ret);
1058 ceph_free_cap_flush(aio_req->prealloc_cf);
1062 static void ceph_aio_complete_req(struct ceph_osd_request *req)
1064 int rc = req->r_result;
1065 struct inode *inode = req->r_inode;
1066 struct ceph_aio_request *aio_req = req->r_priv;
1067 struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0);
1068 struct ceph_client_metric *metric = &ceph_sb_to_mdsc(inode->i_sb)->metric;
1069 unsigned int len = osd_data->bvec_pos.iter.bi_size;
1071 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_BVECS);
1072 BUG_ON(!osd_data->num_bvecs);
1074 dout("ceph_aio_complete_req %p rc %d bytes %u\n", inode, rc, len);
1076 if (rc == -EOLDSNAPC) {
1077 struct ceph_aio_work *aio_work;
1078 BUG_ON(!aio_req->write);
1080 aio_work = kmalloc(sizeof(*aio_work), GFP_NOFS);
1082 INIT_WORK(&aio_work->work, ceph_aio_retry_work);
1083 aio_work->req = req;
1084 queue_work(ceph_inode_to_client(inode)->inode_wq,
1089 } else if (!aio_req->write) {
1092 if (rc >= 0 && len > rc) {
1094 int zlen = len - rc;
1097 * If read is satisfied by single OSD request,
1098 * it can pass EOF. Otherwise read is within
1101 if (aio_req->num_reqs == 1) {
1102 loff_t i_size = i_size_read(inode);
1103 loff_t endoff = aio_req->iocb->ki_pos + rc;
1104 if (endoff < i_size)
1105 zlen = min_t(size_t, zlen,
1107 aio_req->total_len = rc + zlen;
1110 iov_iter_bvec(&i, READ, osd_data->bvec_pos.bvecs,
1111 osd_data->num_bvecs, len);
1112 iov_iter_advance(&i, rc);
1113 iov_iter_zero(zlen, &i);
1117 /* r_start_latency == 0 means the request was not submitted */
1118 if (req->r_start_latency) {
1120 ceph_update_write_metrics(metric, req->r_start_latency,
1121 req->r_end_latency, len, rc);
1123 ceph_update_read_metrics(metric, req->r_start_latency,
1124 req->r_end_latency, len, rc);
1127 put_bvecs(osd_data->bvec_pos.bvecs, osd_data->num_bvecs,
1128 aio_req->should_dirty);
1129 ceph_osdc_put_request(req);
1132 cmpxchg(&aio_req->error, 0, rc);
1134 ceph_aio_complete(inode, aio_req);
1138 static void ceph_aio_retry_work(struct work_struct *work)
1140 struct ceph_aio_work *aio_work =
1141 container_of(work, struct ceph_aio_work, work);
1142 struct ceph_osd_request *orig_req = aio_work->req;
1143 struct ceph_aio_request *aio_req = orig_req->r_priv;
1144 struct inode *inode = orig_req->r_inode;
1145 struct ceph_inode_info *ci = ceph_inode(inode);
1146 struct ceph_snap_context *snapc;
1147 struct ceph_osd_request *req;
1150 spin_lock(&ci->i_ceph_lock);
1151 if (__ceph_have_pending_cap_snap(ci)) {
1152 struct ceph_cap_snap *capsnap =
1153 list_last_entry(&ci->i_cap_snaps,
1154 struct ceph_cap_snap,
1156 snapc = ceph_get_snap_context(capsnap->context);
1158 BUG_ON(!ci->i_head_snapc);
1159 snapc = ceph_get_snap_context(ci->i_head_snapc);
1161 spin_unlock(&ci->i_ceph_lock);
1163 req = ceph_osdc_alloc_request(orig_req->r_osdc, snapc, 1,
1171 req->r_flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE;
1172 ceph_oloc_copy(&req->r_base_oloc, &orig_req->r_base_oloc);
1173 ceph_oid_copy(&req->r_base_oid, &orig_req->r_base_oid);
1175 req->r_ops[0] = orig_req->r_ops[0];
1177 req->r_mtime = aio_req->mtime;
1178 req->r_data_offset = req->r_ops[0].extent.offset;
1180 ret = ceph_osdc_alloc_messages(req, GFP_NOFS);
1182 ceph_osdc_put_request(req);
1187 ceph_osdc_put_request(orig_req);
1189 req->r_callback = ceph_aio_complete_req;
1190 req->r_inode = inode;
1191 req->r_priv = aio_req;
1193 ret = ceph_osdc_start_request(req->r_osdc, req, false);
1196 req->r_result = ret;
1197 ceph_aio_complete_req(req);
1200 ceph_put_snap_context(snapc);
1205 ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter,
1206 struct ceph_snap_context *snapc,
1207 struct ceph_cap_flush **pcf)
1209 struct file *file = iocb->ki_filp;
1210 struct inode *inode = file_inode(file);
1211 struct ceph_inode_info *ci = ceph_inode(inode);
1212 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1213 struct ceph_client_metric *metric = &fsc->mdsc->metric;
1214 struct ceph_vino vino;
1215 struct ceph_osd_request *req;
1216 struct bio_vec *bvecs;
1217 struct ceph_aio_request *aio_req = NULL;
1221 struct timespec64 mtime = current_time(inode);
1222 size_t count = iov_iter_count(iter);
1223 loff_t pos = iocb->ki_pos;
1224 bool write = iov_iter_rw(iter) == WRITE;
1225 bool should_dirty = !write && iter_is_iovec(iter);
1227 if (write && ceph_snap(file_inode(file)) != CEPH_NOSNAP)
1230 dout("sync_direct_%s on file %p %lld~%u snapc %p seq %lld\n",
1231 (write ? "write" : "read"), file, pos, (unsigned)count,
1232 snapc, snapc ? snapc->seq : 0);
1237 ceph_fscache_invalidate(inode, true);
1239 ret2 = invalidate_inode_pages2_range(inode->i_mapping,
1241 (pos + count - 1) >> PAGE_SHIFT);
1243 dout("invalidate_inode_pages2_range returned %d\n", ret2);
1245 flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE;
1247 flags = CEPH_OSD_FLAG_READ;
1250 while (iov_iter_count(iter) > 0) {
1251 u64 size = iov_iter_count(iter);
1255 size = min_t(u64, size, fsc->mount_options->wsize);
1257 size = min_t(u64, size, fsc->mount_options->rsize);
1259 vino = ceph_vino(inode);
1260 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1261 vino, pos, &size, 0,
1263 write ? CEPH_OSD_OP_WRITE :
1267 ci->i_truncate_size,
1274 len = iter_get_bvecs_alloc(iter, size, &bvecs, &num_pages);
1276 ceph_osdc_put_request(req);
1281 osd_req_op_extent_update(req, 0, len);
1284 * To simplify error handling, allow AIO when IO within i_size
1285 * or IO can be satisfied by single OSD request.
1287 if (pos == iocb->ki_pos && !is_sync_kiocb(iocb) &&
1288 (len == count || pos + count <= i_size_read(inode))) {
1289 aio_req = kzalloc(sizeof(*aio_req), GFP_KERNEL);
1291 aio_req->iocb = iocb;
1292 aio_req->write = write;
1293 aio_req->should_dirty = should_dirty;
1294 INIT_LIST_HEAD(&aio_req->osd_reqs);
1296 aio_req->mtime = mtime;
1297 swap(aio_req->prealloc_cf, *pcf);
1305 * throw out any page cache pages in this range. this
1308 truncate_inode_pages_range(inode->i_mapping, pos,
1309 PAGE_ALIGN(pos + len) - 1);
1311 req->r_mtime = mtime;
1314 osd_req_op_extent_osd_data_bvecs(req, 0, bvecs, num_pages, len);
1317 aio_req->total_len += len;
1318 aio_req->num_reqs++;
1319 atomic_inc(&aio_req->pending_reqs);
1321 req->r_callback = ceph_aio_complete_req;
1322 req->r_inode = inode;
1323 req->r_priv = aio_req;
1324 list_add_tail(&req->r_private_item, &aio_req->osd_reqs);
1330 ret = ceph_osdc_start_request(req->r_osdc, req, false);
1332 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1335 ceph_update_write_metrics(metric, req->r_start_latency,
1336 req->r_end_latency, len, ret);
1338 ceph_update_read_metrics(metric, req->r_start_latency,
1339 req->r_end_latency, len, ret);
1341 size = i_size_read(inode);
1345 if (ret >= 0 && ret < len && pos + ret < size) {
1347 int zlen = min_t(size_t, len - ret,
1350 iov_iter_bvec(&i, READ, bvecs, num_pages, len);
1351 iov_iter_advance(&i, ret);
1352 iov_iter_zero(zlen, &i);
1359 put_bvecs(bvecs, num_pages, should_dirty);
1360 ceph_osdc_put_request(req);
1365 if (!write && pos >= size)
1368 if (write && pos > size) {
1369 if (ceph_inode_set_size(inode, pos))
1370 ceph_check_caps(ceph_inode(inode),
1371 CHECK_CAPS_AUTHONLY,
1377 LIST_HEAD(osd_reqs);
1379 if (aio_req->num_reqs == 0) {
1384 ceph_get_cap_refs(ci, write ? CEPH_CAP_FILE_WR :
1387 list_splice(&aio_req->osd_reqs, &osd_reqs);
1388 inode_dio_begin(inode);
1389 while (!list_empty(&osd_reqs)) {
1390 req = list_first_entry(&osd_reqs,
1391 struct ceph_osd_request,
1393 list_del_init(&req->r_private_item);
1395 ret = ceph_osdc_start_request(req->r_osdc,
1398 req->r_result = ret;
1399 ceph_aio_complete_req(req);
1402 return -EIOCBQUEUED;
1405 if (ret != -EOLDSNAPC && pos > iocb->ki_pos) {
1406 ret = pos - iocb->ki_pos;
1413 * Synchronous write, straight from __user pointer or user pages.
1415 * If write spans object boundary, just do multiple writes. (For a
1416 * correct atomic write, we should e.g. take write locks on all
1417 * objects, rollback on failure, etc.)
1420 ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
1421 struct ceph_snap_context *snapc)
1423 struct file *file = iocb->ki_filp;
1424 struct inode *inode = file_inode(file);
1425 struct ceph_inode_info *ci = ceph_inode(inode);
1426 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1427 struct ceph_vino vino;
1428 struct ceph_osd_request *req;
1429 struct page **pages;
1435 bool check_caps = false;
1436 struct timespec64 mtime = current_time(inode);
1437 size_t count = iov_iter_count(from);
1439 if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
1442 dout("sync_write on file %p %lld~%u snapc %p seq %lld\n",
1443 file, pos, (unsigned)count, snapc, snapc->seq);
1445 ret = filemap_write_and_wait_range(inode->i_mapping,
1446 pos, pos + count - 1);
1450 ceph_fscache_invalidate(inode, false);
1451 ret = invalidate_inode_pages2_range(inode->i_mapping,
1453 (pos + count - 1) >> PAGE_SHIFT);
1455 dout("invalidate_inode_pages2_range returned %d\n", ret);
1457 flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE;
1459 while ((len = iov_iter_count(from)) > 0) {
1463 vino = ceph_vino(inode);
1464 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1465 vino, pos, &len, 0, 1,
1466 CEPH_OSD_OP_WRITE, flags, snapc,
1468 ci->i_truncate_size,
1476 * write from beginning of first page,
1477 * regardless of io alignment
1479 num_pages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1481 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
1482 if (IS_ERR(pages)) {
1483 ret = PTR_ERR(pages);
1488 for (n = 0; n < num_pages; n++) {
1489 size_t plen = min_t(size_t, left, PAGE_SIZE);
1490 ret = copy_page_from_iter(pages[n], 0, plen, from);
1499 ceph_release_page_vector(pages, num_pages);
1503 req->r_inode = inode;
1505 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0,
1508 req->r_mtime = mtime;
1509 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1511 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1513 ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
1514 req->r_end_latency, len, ret);
1516 ceph_osdc_put_request(req);
1518 ceph_set_error_write(ci);
1522 ceph_clear_error_write(ci);
1525 if (pos > i_size_read(inode)) {
1526 check_caps = ceph_inode_set_size(inode, pos);
1528 ceph_check_caps(ceph_inode(inode),
1529 CHECK_CAPS_AUTHONLY,
1535 if (ret != -EOLDSNAPC && written > 0) {
1543 * Wrap generic_file_aio_read with checks for cap bits on the inode.
1544 * Atomically grab references, so that those bits are not released
1545 * back to the MDS mid-read.
1547 * Hmm, the sync read case isn't actually async... should it be?
1549 static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to)
1551 struct file *filp = iocb->ki_filp;
1552 struct ceph_file_info *fi = filp->private_data;
1553 size_t len = iov_iter_count(to);
1554 struct inode *inode = file_inode(filp);
1555 struct ceph_inode_info *ci = ceph_inode(inode);
1556 bool direct_lock = iocb->ki_flags & IOCB_DIRECT;
1558 int want = 0, got = 0;
1559 int retry_op = 0, read = 0;
1562 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
1563 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode);
1565 if (ceph_inode_is_shutdown(inode))
1569 ceph_start_io_direct(inode);
1571 ceph_start_io_read(inode);
1573 if (!(fi->flags & CEPH_F_SYNC) && !direct_lock)
1574 want |= CEPH_CAP_FILE_CACHE;
1575 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1576 want |= CEPH_CAP_FILE_LAZYIO;
1578 ret = ceph_get_caps(filp, CEPH_CAP_FILE_RD, want, -1, &got);
1581 ceph_end_io_direct(inode);
1583 ceph_end_io_read(inode);
1587 if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 ||
1588 (iocb->ki_flags & IOCB_DIRECT) ||
1589 (fi->flags & CEPH_F_SYNC)) {
1591 dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n",
1592 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
1593 ceph_cap_string(got));
1595 if (ci->i_inline_version == CEPH_INLINE_NONE) {
1596 if (!retry_op && (iocb->ki_flags & IOCB_DIRECT)) {
1597 ret = ceph_direct_read_write(iocb, to,
1599 if (ret >= 0 && ret < len)
1600 retry_op = CHECK_EOF;
1602 ret = ceph_sync_read(iocb, to, &retry_op);
1605 retry_op = READ_INLINE;
1608 CEPH_DEFINE_RW_CONTEXT(rw_ctx, got);
1609 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
1610 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
1611 ceph_cap_string(got));
1612 ceph_add_rw_context(fi, &rw_ctx);
1613 ret = generic_file_read_iter(iocb, to);
1614 ceph_del_rw_context(fi, &rw_ctx);
1617 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
1618 inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
1619 ceph_put_cap_refs(ci, got);
1622 ceph_end_io_direct(inode);
1624 ceph_end_io_read(inode);
1626 if (retry_op > HAVE_RETRIED && ret >= 0) {
1628 struct page *page = NULL;
1630 if (retry_op == READ_INLINE) {
1631 page = __page_cache_alloc(GFP_KERNEL);
1636 statret = __ceph_do_getattr(inode, page,
1637 CEPH_STAT_CAP_INLINE_DATA, !!page);
1641 if (statret == -ENODATA) {
1642 BUG_ON(retry_op != READ_INLINE);
1648 i_size = i_size_read(inode);
1649 if (retry_op == READ_INLINE) {
1650 BUG_ON(ret > 0 || read > 0);
1651 if (iocb->ki_pos < i_size &&
1652 iocb->ki_pos < PAGE_SIZE) {
1653 loff_t end = min_t(loff_t, i_size,
1654 iocb->ki_pos + len);
1655 end = min_t(loff_t, end, PAGE_SIZE);
1657 zero_user_segment(page, statret, end);
1658 ret = copy_page_to_iter(page,
1659 iocb->ki_pos & ~PAGE_MASK,
1660 end - iocb->ki_pos, to);
1661 iocb->ki_pos += ret;
1664 if (iocb->ki_pos < i_size && read < len) {
1665 size_t zlen = min_t(size_t, len - read,
1666 i_size - iocb->ki_pos);
1667 ret = iov_iter_zero(zlen, to);
1668 iocb->ki_pos += ret;
1671 __free_pages(page, 0);
1675 /* hit EOF or hole? */
1676 if (retry_op == CHECK_EOF && iocb->ki_pos < i_size &&
1678 dout("sync_read hit hole, ppos %lld < size %lld"
1679 ", reading more\n", iocb->ki_pos, i_size);
1683 retry_op = HAVE_RETRIED;
1695 * Take cap references to avoid releasing caps to MDS mid-write.
1697 * If we are synchronous, and write with an old snap context, the OSD
1698 * may return EOLDSNAPC. In that case, retry the write.. _after_
1699 * dropping our cap refs and allowing the pending snap to logically
1700 * complete _before_ this write occurs.
1702 * If we are near ENOSPC, write synchronously.
1704 static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
1706 struct file *file = iocb->ki_filp;
1707 struct ceph_file_info *fi = file->private_data;
1708 struct inode *inode = file_inode(file);
1709 struct ceph_inode_info *ci = ceph_inode(inode);
1710 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1711 struct ceph_osd_client *osdc = &fsc->client->osdc;
1712 struct ceph_cap_flush *prealloc_cf;
1713 ssize_t count, written = 0;
1714 int err, want = 0, got;
1715 bool direct_lock = false;
1719 loff_t limit = max(i_size_read(inode), fsc->max_file_size);
1721 if (ceph_inode_is_shutdown(inode))
1724 if (ceph_snap(inode) != CEPH_NOSNAP)
1727 prealloc_cf = ceph_alloc_cap_flush();
1731 if ((iocb->ki_flags & (IOCB_DIRECT | IOCB_APPEND)) == IOCB_DIRECT)
1736 ceph_start_io_direct(inode);
1738 ceph_start_io_write(inode);
1740 /* We can write back this queue in page reclaim */
1741 current->backing_dev_info = inode_to_bdi(inode);
1743 if (iocb->ki_flags & IOCB_APPEND) {
1744 err = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
1749 err = generic_write_checks(iocb, from);
1754 if (unlikely(pos >= limit)) {
1758 iov_iter_truncate(from, limit - pos);
1761 count = iov_iter_count(from);
1762 if (ceph_quota_is_max_bytes_exceeded(inode, pos + count)) {
1767 down_read(&osdc->lock);
1768 map_flags = osdc->osdmap->flags;
1769 pool_flags = ceph_pg_pool_flags(osdc->osdmap, ci->i_layout.pool_id);
1770 up_read(&osdc->lock);
1771 if ((map_flags & CEPH_OSDMAP_FULL) ||
1772 (pool_flags & CEPH_POOL_FLAG_FULL)) {
1777 err = file_remove_privs(file);
1781 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1782 err = ceph_uninline_data(file, NULL);
1787 dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
1788 inode, ceph_vinop(inode), pos, count, i_size_read(inode));
1789 if (!(fi->flags & CEPH_F_SYNC) && !direct_lock)
1790 want |= CEPH_CAP_FILE_BUFFER;
1791 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1792 want |= CEPH_CAP_FILE_LAZYIO;
1794 err = ceph_get_caps(file, CEPH_CAP_FILE_WR, want, pos + count, &got);
1798 err = file_update_time(file);
1802 inode_inc_iversion_raw(inode);
1804 dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
1805 inode, ceph_vinop(inode), pos, count, ceph_cap_string(got));
1807 if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 ||
1808 (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC) ||
1809 (ci->i_ceph_flags & CEPH_I_ERROR_WRITE)) {
1810 struct ceph_snap_context *snapc;
1811 struct iov_iter data;
1813 spin_lock(&ci->i_ceph_lock);
1814 if (__ceph_have_pending_cap_snap(ci)) {
1815 struct ceph_cap_snap *capsnap =
1816 list_last_entry(&ci->i_cap_snaps,
1817 struct ceph_cap_snap,
1819 snapc = ceph_get_snap_context(capsnap->context);
1821 BUG_ON(!ci->i_head_snapc);
1822 snapc = ceph_get_snap_context(ci->i_head_snapc);
1824 spin_unlock(&ci->i_ceph_lock);
1826 /* we might need to revert back to that point */
1828 if (iocb->ki_flags & IOCB_DIRECT)
1829 written = ceph_direct_read_write(iocb, &data, snapc,
1832 written = ceph_sync_write(iocb, &data, pos, snapc);
1834 ceph_end_io_direct(inode);
1836 ceph_end_io_write(inode);
1838 iov_iter_advance(from, written);
1839 ceph_put_snap_context(snapc);
1842 * No need to acquire the i_truncate_mutex. Because
1843 * the MDS revokes Fwb caps before sending truncate
1844 * message to us. We can't get Fwb cap while there
1845 * are pending vmtruncate. So write and vmtruncate
1846 * can not run at the same time
1848 written = generic_perform_write(file, from, pos);
1849 if (likely(written >= 0))
1850 iocb->ki_pos = pos + written;
1851 ceph_end_io_write(inode);
1857 spin_lock(&ci->i_ceph_lock);
1858 ci->i_inline_version = CEPH_INLINE_NONE;
1859 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1861 spin_unlock(&ci->i_ceph_lock);
1863 __mark_inode_dirty(inode, dirty);
1864 if (ceph_quota_is_max_bytes_approaching(inode, iocb->ki_pos))
1865 ceph_check_caps(ci, 0, NULL);
1868 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
1869 inode, ceph_vinop(inode), pos, (unsigned)count,
1870 ceph_cap_string(got));
1871 ceph_put_cap_refs(ci, got);
1873 if (written == -EOLDSNAPC) {
1874 dout("aio_write %p %llx.%llx %llu~%u" "got EOLDSNAPC, retrying\n",
1875 inode, ceph_vinop(inode), pos, (unsigned)count);
1880 if ((map_flags & CEPH_OSDMAP_NEARFULL) ||
1881 (pool_flags & CEPH_POOL_FLAG_NEARFULL))
1882 iocb->ki_flags |= IOCB_DSYNC;
1883 written = generic_write_sync(iocb, written);
1888 ceph_put_cap_refs(ci, got);
1891 ceph_end_io_direct(inode);
1893 ceph_end_io_write(inode);
1895 ceph_free_cap_flush(prealloc_cf);
1896 current->backing_dev_info = NULL;
1897 return written ? written : err;
1901 * llseek. be sure to verify file size on SEEK_END.
1903 static loff_t ceph_llseek(struct file *file, loff_t offset, int whence)
1905 struct inode *inode = file->f_mapping->host;
1906 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1912 if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) {
1913 ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
1918 i_size = i_size_read(inode);
1925 * Here we special-case the lseek(fd, 0, SEEK_CUR)
1926 * position-querying operation. Avoid rewriting the "same"
1927 * f_pos value back to the file because a concurrent read(),
1928 * write() or lseek() might have altered it
1934 offset += file->f_pos;
1937 if (offset < 0 || offset >= i_size) {
1943 if (offset < 0 || offset >= i_size) {
1951 ret = vfs_setpos(file, offset, max(i_size, fsc->max_file_size));
1954 inode_unlock(inode);
1958 static inline void ceph_zero_partial_page(
1959 struct inode *inode, loff_t offset, unsigned size)
1962 pgoff_t index = offset >> PAGE_SHIFT;
1964 page = find_lock_page(inode->i_mapping, index);
1966 wait_on_page_writeback(page);
1967 zero_user(page, offset & (PAGE_SIZE - 1), size);
1973 static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset,
1976 loff_t nearly = round_up(offset, PAGE_SIZE);
1977 if (offset < nearly) {
1978 loff_t size = nearly - offset;
1981 ceph_zero_partial_page(inode, offset, size);
1985 if (length >= PAGE_SIZE) {
1986 loff_t size = round_down(length, PAGE_SIZE);
1987 truncate_pagecache_range(inode, offset, offset + size - 1);
1992 ceph_zero_partial_page(inode, offset, length);
1995 static int ceph_zero_partial_object(struct inode *inode,
1996 loff_t offset, loff_t *length)
1998 struct ceph_inode_info *ci = ceph_inode(inode);
1999 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
2000 struct ceph_osd_request *req;
2006 op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE;
2009 op = CEPH_OSD_OP_ZERO;
2012 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
2016 CEPH_OSD_FLAG_WRITE,
2023 req->r_mtime = inode->i_mtime;
2024 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
2026 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
2030 ceph_osdc_put_request(req);
2036 static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length)
2039 struct ceph_inode_info *ci = ceph_inode(inode);
2040 s32 stripe_unit = ci->i_layout.stripe_unit;
2041 s32 stripe_count = ci->i_layout.stripe_count;
2042 s32 object_size = ci->i_layout.object_size;
2043 u64 object_set_size = object_size * stripe_count;
2046 /* round offset up to next period boundary */
2047 nearly = offset + object_set_size - 1;
2049 nearly -= do_div(t, object_set_size);
2051 while (length && offset < nearly) {
2052 loff_t size = length;
2053 ret = ceph_zero_partial_object(inode, offset, &size);
2059 while (length >= object_set_size) {
2061 loff_t pos = offset;
2062 for (i = 0; i < stripe_count; ++i) {
2063 ret = ceph_zero_partial_object(inode, pos, NULL);
2068 offset += object_set_size;
2069 length -= object_set_size;
2072 loff_t size = length;
2073 ret = ceph_zero_partial_object(inode, offset, &size);
2082 static long ceph_fallocate(struct file *file, int mode,
2083 loff_t offset, loff_t length)
2085 struct ceph_file_info *fi = file->private_data;
2086 struct inode *inode = file_inode(file);
2087 struct ceph_inode_info *ci = ceph_inode(inode);
2088 struct ceph_cap_flush *prealloc_cf;
2095 if (mode != (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2098 if (!S_ISREG(inode->i_mode))
2101 prealloc_cf = ceph_alloc_cap_flush();
2107 if (ceph_snap(inode) != CEPH_NOSNAP) {
2112 if (ci->i_inline_version != CEPH_INLINE_NONE) {
2113 ret = ceph_uninline_data(file, NULL);
2118 size = i_size_read(inode);
2120 /* Are we punching a hole beyond EOF? */
2123 if ((offset + length) > size)
2124 length = size - offset;
2126 if (fi->fmode & CEPH_FILE_MODE_LAZY)
2127 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
2129 want = CEPH_CAP_FILE_BUFFER;
2131 ret = ceph_get_caps(file, CEPH_CAP_FILE_WR, want, endoff, &got);
2135 filemap_invalidate_lock(inode->i_mapping);
2136 ceph_fscache_invalidate(inode, false);
2137 ceph_zero_pagecache_range(inode, offset, length);
2138 ret = ceph_zero_objects(inode, offset, length);
2141 spin_lock(&ci->i_ceph_lock);
2142 ci->i_inline_version = CEPH_INLINE_NONE;
2143 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
2145 spin_unlock(&ci->i_ceph_lock);
2147 __mark_inode_dirty(inode, dirty);
2149 filemap_invalidate_unlock(inode->i_mapping);
2151 ceph_put_cap_refs(ci, got);
2153 inode_unlock(inode);
2154 ceph_free_cap_flush(prealloc_cf);
2159 * This function tries to get FILE_WR capabilities for dst_ci and FILE_RD for
2160 * src_ci. Two attempts are made to obtain both caps, and an error is return if
2161 * this fails; zero is returned on success.
2163 static int get_rd_wr_caps(struct file *src_filp, int *src_got,
2164 struct file *dst_filp,
2165 loff_t dst_endoff, int *dst_got)
2168 bool retrying = false;
2171 ret = ceph_get_caps(dst_filp, CEPH_CAP_FILE_WR, CEPH_CAP_FILE_BUFFER,
2172 dst_endoff, dst_got);
2177 * Since we're already holding the FILE_WR capability for the dst file,
2178 * we would risk a deadlock by using ceph_get_caps. Thus, we'll do some
2179 * retry dance instead to try to get both capabilities.
2181 ret = ceph_try_get_caps(file_inode(src_filp),
2182 CEPH_CAP_FILE_RD, CEPH_CAP_FILE_SHARED,
2185 /* Start by dropping dst_ci caps and getting src_ci caps */
2186 ceph_put_cap_refs(ceph_inode(file_inode(dst_filp)), *dst_got);
2189 /* ceph_try_get_caps masks EAGAIN */
2193 ret = ceph_get_caps(src_filp, CEPH_CAP_FILE_RD,
2194 CEPH_CAP_FILE_SHARED, -1, src_got);
2197 /*... drop src_ci caps too, and retry */
2198 ceph_put_cap_refs(ceph_inode(file_inode(src_filp)), *src_got);
2205 static void put_rd_wr_caps(struct ceph_inode_info *src_ci, int src_got,
2206 struct ceph_inode_info *dst_ci, int dst_got)
2208 ceph_put_cap_refs(src_ci, src_got);
2209 ceph_put_cap_refs(dst_ci, dst_got);
2213 * This function does several size-related checks, returning an error if:
2214 * - source file is smaller than off+len
2215 * - destination file size is not OK (inode_newsize_ok())
2216 * - max bytes quotas is exceeded
2218 static int is_file_size_ok(struct inode *src_inode, struct inode *dst_inode,
2219 loff_t src_off, loff_t dst_off, size_t len)
2221 loff_t size, endoff;
2223 size = i_size_read(src_inode);
2225 * Don't copy beyond source file EOF. Instead of simply setting length
2226 * to (size - src_off), just drop to VFS default implementation, as the
2227 * local i_size may be stale due to other clients writing to the source
2230 if (src_off + len > size) {
2231 dout("Copy beyond EOF (%llu + %zu > %llu)\n",
2232 src_off, len, size);
2235 size = i_size_read(dst_inode);
2237 endoff = dst_off + len;
2238 if (inode_newsize_ok(dst_inode, endoff))
2241 if (ceph_quota_is_max_bytes_exceeded(dst_inode, endoff))
2247 static struct ceph_osd_request *
2248 ceph_alloc_copyfrom_request(struct ceph_osd_client *osdc,
2250 struct ceph_object_id *src_oid,
2251 struct ceph_object_locator *src_oloc,
2252 struct ceph_object_id *dst_oid,
2253 struct ceph_object_locator *dst_oloc,
2254 u32 truncate_seq, u64 truncate_size)
2256 struct ceph_osd_request *req;
2258 u32 src_fadvise_flags =
2259 CEPH_OSD_OP_FLAG_FADVISE_SEQUENTIAL |
2260 CEPH_OSD_OP_FLAG_FADVISE_NOCACHE;
2261 u32 dst_fadvise_flags =
2262 CEPH_OSD_OP_FLAG_FADVISE_SEQUENTIAL |
2263 CEPH_OSD_OP_FLAG_FADVISE_DONTNEED;
2265 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_KERNEL);
2267 return ERR_PTR(-ENOMEM);
2269 req->r_flags = CEPH_OSD_FLAG_WRITE;
2271 ceph_oloc_copy(&req->r_t.base_oloc, dst_oloc);
2272 ceph_oid_copy(&req->r_t.base_oid, dst_oid);
2274 ret = osd_req_op_copy_from_init(req, src_snapid, 0,
2280 CEPH_OSD_COPY_FROM_FLAG_TRUNCATE_SEQ);
2284 ret = ceph_osdc_alloc_messages(req, GFP_KERNEL);
2291 ceph_osdc_put_request(req);
2292 return ERR_PTR(ret);
2295 static ssize_t ceph_do_objects_copy(struct ceph_inode_info *src_ci, u64 *src_off,
2296 struct ceph_inode_info *dst_ci, u64 *dst_off,
2297 struct ceph_fs_client *fsc,
2298 size_t len, unsigned int flags)
2300 struct ceph_object_locator src_oloc, dst_oloc;
2301 struct ceph_object_id src_oid, dst_oid;
2302 struct ceph_osd_client *osdc;
2303 struct ceph_osd_request *req;
2305 u64 src_objnum, src_objoff, dst_objnum, dst_objoff;
2306 u32 src_objlen, dst_objlen;
2307 u32 object_size = src_ci->i_layout.object_size;
2310 src_oloc.pool = src_ci->i_layout.pool_id;
2311 src_oloc.pool_ns = ceph_try_get_string(src_ci->i_layout.pool_ns);
2312 dst_oloc.pool = dst_ci->i_layout.pool_id;
2313 dst_oloc.pool_ns = ceph_try_get_string(dst_ci->i_layout.pool_ns);
2314 osdc = &fsc->client->osdc;
2316 while (len >= object_size) {
2317 ceph_calc_file_object_mapping(&src_ci->i_layout, *src_off,
2318 object_size, &src_objnum,
2319 &src_objoff, &src_objlen);
2320 ceph_calc_file_object_mapping(&dst_ci->i_layout, *dst_off,
2321 object_size, &dst_objnum,
2322 &dst_objoff, &dst_objlen);
2323 ceph_oid_init(&src_oid);
2324 ceph_oid_printf(&src_oid, "%llx.%08llx",
2325 src_ci->i_vino.ino, src_objnum);
2326 ceph_oid_init(&dst_oid);
2327 ceph_oid_printf(&dst_oid, "%llx.%08llx",
2328 dst_ci->i_vino.ino, dst_objnum);
2329 /* Do an object remote copy */
2330 req = ceph_alloc_copyfrom_request(osdc, src_ci->i_vino.snap,
2331 &src_oid, &src_oloc,
2332 &dst_oid, &dst_oloc,
2333 dst_ci->i_truncate_seq,
2334 dst_ci->i_truncate_size);
2338 ceph_osdc_start_request(osdc, req, false);
2339 ret = ceph_osdc_wait_request(osdc, req);
2340 ceph_update_copyfrom_metrics(&fsc->mdsc->metric,
2341 req->r_start_latency,
2344 ceph_osdc_put_request(req);
2347 if (ret == -EOPNOTSUPP) {
2348 fsc->have_copy_from2 = false;
2349 pr_notice("OSDs don't support copy-from2; disabling copy offload\n");
2351 dout("ceph_osdc_copy_from returned %d\n", ret);
2357 bytes += object_size;
2358 *src_off += object_size;
2359 *dst_off += object_size;
2363 ceph_oloc_destroy(&src_oloc);
2364 ceph_oloc_destroy(&dst_oloc);
2368 static ssize_t __ceph_copy_file_range(struct file *src_file, loff_t src_off,
2369 struct file *dst_file, loff_t dst_off,
2370 size_t len, unsigned int flags)
2372 struct inode *src_inode = file_inode(src_file);
2373 struct inode *dst_inode = file_inode(dst_file);
2374 struct ceph_inode_info *src_ci = ceph_inode(src_inode);
2375 struct ceph_inode_info *dst_ci = ceph_inode(dst_inode);
2376 struct ceph_cap_flush *prealloc_cf;
2377 struct ceph_fs_client *src_fsc = ceph_inode_to_client(src_inode);
2379 ssize_t ret = -EIO, bytes;
2380 u64 src_objnum, dst_objnum, src_objoff, dst_objoff;
2381 u32 src_objlen, dst_objlen;
2382 int src_got = 0, dst_got = 0, err, dirty;
2384 if (src_inode->i_sb != dst_inode->i_sb) {
2385 struct ceph_fs_client *dst_fsc = ceph_inode_to_client(dst_inode);
2387 if (ceph_fsid_compare(&src_fsc->client->fsid,
2388 &dst_fsc->client->fsid)) {
2389 dout("Copying files across clusters: src: %pU dst: %pU\n",
2390 &src_fsc->client->fsid, &dst_fsc->client->fsid);
2394 if (ceph_snap(dst_inode) != CEPH_NOSNAP)
2398 * Some of the checks below will return -EOPNOTSUPP, which will force a
2399 * fallback to the default VFS copy_file_range implementation. This is
2400 * desirable in several cases (for ex, the 'len' is smaller than the
2401 * size of the objects, or in cases where that would be more
2405 if (ceph_test_mount_opt(src_fsc, NOCOPYFROM))
2408 if (!src_fsc->have_copy_from2)
2412 * Striped file layouts require that we copy partial objects, but the
2413 * OSD copy-from operation only supports full-object copies. Limit
2414 * this to non-striped file layouts for now.
2416 if ((src_ci->i_layout.stripe_unit != dst_ci->i_layout.stripe_unit) ||
2417 (src_ci->i_layout.stripe_count != 1) ||
2418 (dst_ci->i_layout.stripe_count != 1) ||
2419 (src_ci->i_layout.object_size != dst_ci->i_layout.object_size)) {
2420 dout("Invalid src/dst files layout\n");
2424 if (len < src_ci->i_layout.object_size)
2425 return -EOPNOTSUPP; /* no remote copy will be done */
2427 prealloc_cf = ceph_alloc_cap_flush();
2431 /* Start by sync'ing the source and destination files */
2432 ret = file_write_and_wait_range(src_file, src_off, (src_off + len));
2434 dout("failed to write src file (%zd)\n", ret);
2437 ret = file_write_and_wait_range(dst_file, dst_off, (dst_off + len));
2439 dout("failed to write dst file (%zd)\n", ret);
2444 * We need FILE_WR caps for dst_ci and FILE_RD for src_ci as other
2445 * clients may have dirty data in their caches. And OSDs know nothing
2446 * about caps, so they can't safely do the remote object copies.
2448 err = get_rd_wr_caps(src_file, &src_got,
2449 dst_file, (dst_off + len), &dst_got);
2451 dout("get_rd_wr_caps returned %d\n", err);
2456 ret = is_file_size_ok(src_inode, dst_inode, src_off, dst_off, len);
2460 /* Drop dst file cached pages */
2461 ceph_fscache_invalidate(dst_inode, false);
2462 ret = invalidate_inode_pages2_range(dst_inode->i_mapping,
2463 dst_off >> PAGE_SHIFT,
2464 (dst_off + len) >> PAGE_SHIFT);
2466 dout("Failed to invalidate inode pages (%zd)\n", ret);
2469 ceph_calc_file_object_mapping(&src_ci->i_layout, src_off,
2470 src_ci->i_layout.object_size,
2471 &src_objnum, &src_objoff, &src_objlen);
2472 ceph_calc_file_object_mapping(&dst_ci->i_layout, dst_off,
2473 dst_ci->i_layout.object_size,
2474 &dst_objnum, &dst_objoff, &dst_objlen);
2475 /* object-level offsets need to the same */
2476 if (src_objoff != dst_objoff) {
2482 * Do a manual copy if the object offset isn't object aligned.
2483 * 'src_objlen' contains the bytes left until the end of the object,
2484 * starting at the src_off
2487 dout("Initial partial copy of %u bytes\n", src_objlen);
2490 * we need to temporarily drop all caps as we'll be calling
2491 * {read,write}_iter, which will get caps again.
2493 put_rd_wr_caps(src_ci, src_got, dst_ci, dst_got);
2494 ret = do_splice_direct(src_file, &src_off, dst_file,
2495 &dst_off, src_objlen, flags);
2496 /* Abort on short copies or on error */
2497 if (ret < src_objlen) {
2498 dout("Failed partial copy (%zd)\n", ret);
2502 err = get_rd_wr_caps(src_file, &src_got,
2503 dst_file, (dst_off + len), &dst_got);
2506 err = is_file_size_ok(src_inode, dst_inode,
2507 src_off, dst_off, len);
2512 size = i_size_read(dst_inode);
2513 bytes = ceph_do_objects_copy(src_ci, &src_off, dst_ci, &dst_off,
2514 src_fsc, len, flags);
2520 dout("Copied %zu bytes out of %zu\n", bytes, len);
2524 file_update_time(dst_file);
2525 inode_inc_iversion_raw(dst_inode);
2527 if (dst_off > size) {
2528 /* Let the MDS know about dst file size change */
2529 if (ceph_inode_set_size(dst_inode, dst_off) ||
2530 ceph_quota_is_max_bytes_approaching(dst_inode, dst_off))
2531 ceph_check_caps(dst_ci, CHECK_CAPS_AUTHONLY, NULL);
2534 spin_lock(&dst_ci->i_ceph_lock);
2535 dst_ci->i_inline_version = CEPH_INLINE_NONE;
2536 dirty = __ceph_mark_dirty_caps(dst_ci, CEPH_CAP_FILE_WR, &prealloc_cf);
2537 spin_unlock(&dst_ci->i_ceph_lock);
2539 __mark_inode_dirty(dst_inode, dirty);
2542 put_rd_wr_caps(src_ci, src_got, dst_ci, dst_got);
2545 * Do the final manual copy if we still have some bytes left, unless
2546 * there were errors in remote object copies (len >= object_size).
2548 if (len && (len < src_ci->i_layout.object_size)) {
2549 dout("Final partial copy of %zu bytes\n", len);
2550 bytes = do_splice_direct(src_file, &src_off, dst_file,
2551 &dst_off, len, flags);
2555 dout("Failed partial copy (%zd)\n", bytes);
2559 ceph_free_cap_flush(prealloc_cf);
2564 static ssize_t ceph_copy_file_range(struct file *src_file, loff_t src_off,
2565 struct file *dst_file, loff_t dst_off,
2566 size_t len, unsigned int flags)
2570 ret = __ceph_copy_file_range(src_file, src_off, dst_file, dst_off,
2573 if (ret == -EOPNOTSUPP || ret == -EXDEV)
2574 ret = generic_copy_file_range(src_file, src_off, dst_file,
2575 dst_off, len, flags);
2579 const struct file_operations ceph_file_fops = {
2581 .release = ceph_release,
2582 .llseek = ceph_llseek,
2583 .read_iter = ceph_read_iter,
2584 .write_iter = ceph_write_iter,
2586 .fsync = ceph_fsync,
2588 .setlease = simple_nosetlease,
2589 .flock = ceph_flock,
2590 .splice_read = generic_file_splice_read,
2591 .splice_write = iter_file_splice_write,
2592 .unlocked_ioctl = ceph_ioctl,
2593 .compat_ioctl = compat_ptr_ioctl,
2594 .fallocate = ceph_fallocate,
2595 .copy_file_range = ceph_copy_file_range,