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>
15 #include "mds_client.h"
18 static __le32 ceph_flags_sys2wire(u32 flags)
22 switch (flags & O_ACCMODE) {
24 wire_flags |= CEPH_O_RDONLY;
27 wire_flags |= CEPH_O_WRONLY;
30 wire_flags |= CEPH_O_RDWR;
36 #define ceph_sys2wire(a) if (flags & a) { wire_flags |= CEPH_##a; flags &= ~a; }
38 ceph_sys2wire(O_CREAT);
39 ceph_sys2wire(O_EXCL);
40 ceph_sys2wire(O_TRUNC);
41 ceph_sys2wire(O_DIRECTORY);
42 ceph_sys2wire(O_NOFOLLOW);
47 dout("unused open flags: %x\n", flags);
49 return cpu_to_le32(wire_flags);
53 * Ceph file operations
55 * Implement basic open/close functionality, and implement
58 * We implement three modes of file I/O:
59 * - buffered uses the generic_file_aio_{read,write} helpers
61 * - synchronous is used when there is multi-client read/write
62 * sharing, avoids the page cache, and synchronously waits for an
65 * - direct io takes the variant of the sync path that references
66 * user pages directly.
68 * fsync() flushes and waits on dirty pages, but just queues metadata
69 * for writeback: since the MDS can recover size and mtime there is no
70 * need to wait for MDS acknowledgement.
74 * How many pages to get in one call to iov_iter_get_pages(). This
75 * determines the size of the on-stack array used as a buffer.
77 #define ITER_GET_BVECS_PAGES 64
79 static ssize_t __iter_get_bvecs(struct iov_iter *iter, size_t maxsize,
80 struct bio_vec *bvecs)
85 if (maxsize > iov_iter_count(iter))
86 maxsize = iov_iter_count(iter);
88 while (size < maxsize) {
89 struct page *pages[ITER_GET_BVECS_PAGES];
94 bytes = iov_iter_get_pages(iter, pages, maxsize - size,
95 ITER_GET_BVECS_PAGES, &start);
99 iov_iter_advance(iter, bytes);
102 for ( ; bytes; idx++, bvec_idx++) {
103 struct bio_vec bv = {
104 .bv_page = pages[idx],
105 .bv_len = min_t(int, bytes, PAGE_SIZE - start),
109 bvecs[bvec_idx] = bv;
119 * iov_iter_get_pages() only considers one iov_iter segment, no matter
120 * what maxsize or maxpages are given. For ITER_BVEC that is a single
123 * Attempt to get up to @maxsize bytes worth of pages from @iter.
124 * Return the number of bytes in the created bio_vec array, or an error.
126 static ssize_t iter_get_bvecs_alloc(struct iov_iter *iter, size_t maxsize,
127 struct bio_vec **bvecs, int *num_bvecs)
130 size_t orig_count = iov_iter_count(iter);
134 iov_iter_truncate(iter, maxsize);
135 npages = iov_iter_npages(iter, INT_MAX);
136 iov_iter_reexpand(iter, orig_count);
139 * __iter_get_bvecs() may populate only part of the array -- zero it
142 bv = kvmalloc_array(npages, sizeof(*bv), GFP_KERNEL | __GFP_ZERO);
146 bytes = __iter_get_bvecs(iter, maxsize, bv);
149 * No pages were pinned -- just free the array.
160 static void put_bvecs(struct bio_vec *bvecs, int num_bvecs, bool should_dirty)
164 for (i = 0; i < num_bvecs; i++) {
165 if (bvecs[i].bv_page) {
167 set_page_dirty_lock(bvecs[i].bv_page);
168 put_page(bvecs[i].bv_page);
175 * Prepare an open request. Preallocate ceph_cap to avoid an
176 * inopportune ENOMEM later.
178 static struct ceph_mds_request *
179 prepare_open_request(struct super_block *sb, int flags, int create_mode)
181 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
182 struct ceph_mds_client *mdsc = fsc->mdsc;
183 struct ceph_mds_request *req;
184 int want_auth = USE_ANY_MDS;
185 int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN;
187 if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC))
188 want_auth = USE_AUTH_MDS;
190 req = ceph_mdsc_create_request(mdsc, op, want_auth);
193 req->r_fmode = ceph_flags_to_mode(flags);
194 req->r_args.open.flags = ceph_flags_sys2wire(flags);
195 req->r_args.open.mode = cpu_to_le32(create_mode);
200 static int ceph_init_file_info(struct inode *inode, struct file *file,
201 int fmode, bool isdir)
203 struct ceph_file_info *fi;
205 dout("%s %p %p 0%o (%s)\n", __func__, inode, file,
206 inode->i_mode, isdir ? "dir" : "regular");
207 BUG_ON(inode->i_fop->release != ceph_release);
210 struct ceph_dir_file_info *dfi =
211 kmem_cache_zalloc(ceph_dir_file_cachep, GFP_KERNEL);
213 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
217 file->private_data = dfi;
218 fi = &dfi->file_info;
219 dfi->next_offset = 2;
220 dfi->readdir_cache_idx = -1;
222 fi = kmem_cache_zalloc(ceph_file_cachep, GFP_KERNEL);
224 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
228 file->private_data = fi;
232 spin_lock_init(&fi->rw_contexts_lock);
233 INIT_LIST_HEAD(&fi->rw_contexts);
239 * initialize private struct file data.
240 * if we fail, clean up by dropping fmode reference on the ceph_inode
242 static int ceph_init_file(struct inode *inode, struct file *file, int fmode)
246 switch (inode->i_mode & S_IFMT) {
248 ceph_fscache_register_inode_cookie(inode);
249 ceph_fscache_file_set_cookie(inode, file);
252 ret = ceph_init_file_info(inode, file, fmode,
253 S_ISDIR(inode->i_mode));
259 dout("init_file %p %p 0%o (symlink)\n", inode, file,
261 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
265 dout("init_file %p %p 0%o (special)\n", inode, file,
268 * we need to drop the open ref now, since we don't
269 * have .release set to ceph_release.
271 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
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)
285 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
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 wanted = __ceph_caps_file_wanted(ci);
292 if (__ceph_is_any_real_caps(ci) &&
293 (!(wanted & CEPH_CAP_ANY_WR) || ci->i_auth_cap)) {
294 int issued = __ceph_caps_issued(ci, NULL);
295 spin_unlock(&ci->i_ceph_lock);
296 dout("renew caps %p want %s issued %s updating mds_wanted\n",
297 inode, ceph_cap_string(wanted), ceph_cap_string(issued));
298 ceph_check_caps(ci, 0, NULL);
301 spin_unlock(&ci->i_ceph_lock);
304 if ((wanted & CEPH_CAP_FILE_RD) && (wanted & CEPH_CAP_FILE_WR))
306 else if (wanted & CEPH_CAP_FILE_RD)
308 else if (wanted & CEPH_CAP_FILE_WR)
311 if (wanted & CEPH_CAP_FILE_LAZYIO)
315 req = prepare_open_request(inode->i_sb, flags, 0);
321 req->r_inode = inode;
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 spin_lock(&ci->i_ceph_lock);
371 __ceph_get_fmode(ci, fmode);
372 spin_unlock(&ci->i_ceph_lock);
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_get_fmode(ci, 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_get_fmode(ci, 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));
430 * Do a lookup + open with a single request. If we get a non-existent
431 * file or symlink, return 1 so the VFS can retry.
433 int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
434 struct file *file, unsigned flags, umode_t mode)
436 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
437 struct ceph_mds_client *mdsc = fsc->mdsc;
438 struct ceph_mds_request *req;
440 struct ceph_acls_info acls = {};
444 dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n",
446 d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode);
448 if (dentry->d_name.len > NAME_MAX)
449 return -ENAMETOOLONG;
451 if (flags & O_CREAT) {
452 if (ceph_quota_is_max_files_exceeded(dir))
454 err = ceph_pre_init_acls(dir, &mode, &acls);
460 req = prepare_open_request(dir->i_sb, flags, mode);
465 req->r_dentry = dget(dentry);
467 if (flags & O_CREAT) {
468 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
469 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
471 req->r_pagelist = acls.pagelist;
472 acls.pagelist = NULL;
476 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
477 if (ceph_security_xattr_wanted(dir))
478 mask |= CEPH_CAP_XATTR_SHARED;
479 req->r_args.open.mask = cpu_to_le32(mask);
482 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
483 err = ceph_mdsc_do_request(mdsc,
484 (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
486 err = ceph_handle_snapdir(req, dentry, err);
490 if ((flags & O_CREAT) && !req->r_reply_info.head->is_dentry)
491 err = ceph_handle_notrace_create(dir, dentry);
493 if (d_in_lookup(dentry)) {
494 dn = ceph_finish_lookup(req, dentry, err);
498 /* we were given a hashed negative dentry */
503 if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) {
504 /* make vfs retry on splice, ENOENT, or symlink */
505 dout("atomic_open finish_no_open on dn %p\n", dn);
506 err = finish_no_open(file, dn);
508 dout("atomic_open finish_open on dn %p\n", dn);
509 if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) {
510 ceph_init_inode_acls(d_inode(dentry), &acls);
511 file->f_mode |= FMODE_CREATED;
513 err = finish_open(file, dentry, ceph_open);
516 if (!req->r_err && req->r_target_inode)
517 ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode);
518 ceph_mdsc_put_request(req);
520 ceph_release_acls_info(&acls);
521 dout("atomic_open result=%d\n", err);
525 int ceph_release(struct inode *inode, struct file *file)
527 struct ceph_inode_info *ci = ceph_inode(inode);
529 if (S_ISDIR(inode->i_mode)) {
530 struct ceph_dir_file_info *dfi = file->private_data;
531 dout("release inode %p dir file %p\n", inode, file);
532 WARN_ON(!list_empty(&dfi->file_info.rw_contexts));
534 ceph_put_fmode(ci, dfi->file_info.fmode);
536 if (dfi->last_readdir)
537 ceph_mdsc_put_request(dfi->last_readdir);
538 kfree(dfi->last_name);
539 kfree(dfi->dir_info);
540 kmem_cache_free(ceph_dir_file_cachep, dfi);
542 struct ceph_file_info *fi = file->private_data;
543 dout("release inode %p regular file %p\n", inode, file);
544 WARN_ON(!list_empty(&fi->rw_contexts));
546 ceph_put_fmode(ci, fi->fmode);
547 kmem_cache_free(ceph_file_cachep, fi);
550 /* wake up anyone waiting for caps on this inode */
551 wake_up_all(&ci->i_cap_wq);
562 * Completely synchronous read and write methods. Direct from __user
563 * buffer to osd, or directly to user pages (if O_DIRECT).
565 * If the read spans object boundary, just do multiple reads. (That's not
566 * atomic, but good enough for now.)
568 * If we get a short result from the OSD, check against i_size; we need to
569 * only return a short read to the caller if we hit EOF.
571 static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *to,
574 struct file *file = iocb->ki_filp;
575 struct inode *inode = file_inode(file);
576 struct ceph_inode_info *ci = ceph_inode(inode);
577 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
578 struct ceph_osd_client *osdc = &fsc->client->osdc;
580 u64 off = iocb->ki_pos;
581 u64 len = iov_iter_count(to);
583 dout("sync_read on file %p %llu~%u %s\n", file, off, (unsigned)len,
584 (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
589 * flush any page cache pages in this range. this
590 * will make concurrent normal and sync io slow,
591 * but it will at least behave sensibly when they are
594 ret = filemap_write_and_wait_range(inode->i_mapping,
600 while ((len = iov_iter_count(to)) > 0) {
601 struct ceph_osd_request *req;
608 req = ceph_osdc_new_request(osdc, &ci->i_layout,
609 ci->i_vino, off, &len, 0, 1,
610 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
611 NULL, ci->i_truncate_seq,
612 ci->i_truncate_size, false);
618 more = len < iov_iter_count(to);
620 if (unlikely(iov_iter_is_pipe(to))) {
621 ret = iov_iter_get_pages_alloc(to, &pages, len,
624 ceph_osdc_put_request(req);
628 num_pages = DIV_ROUND_UP(ret + page_off, PAGE_SIZE);
631 osd_req_op_extent_update(req, 0, len);
635 num_pages = calc_pages_for(off, len);
636 page_off = off & ~PAGE_MASK;
637 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
639 ceph_osdc_put_request(req);
640 ret = PTR_ERR(pages);
645 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_off,
647 ret = ceph_osdc_start_request(osdc, req, false);
649 ret = ceph_osdc_wait_request(osdc, req);
650 ceph_osdc_put_request(req);
652 i_size = i_size_read(inode);
653 dout("sync_read %llu~%llu got %zd i_size %llu%s\n",
654 off, len, ret, i_size, (more ? " MORE" : ""));
658 if (ret >= 0 && ret < len && (off + ret < i_size)) {
659 int zlen = min(len - ret, i_size - off - ret);
660 int zoff = page_off + ret;
661 dout("sync_read zero gap %llu~%llu\n",
662 off + ret, off + ret + zlen);
663 ceph_zero_page_vector_range(zoff, zlen, pages);
667 if (unlikely(iov_iter_is_pipe(to))) {
669 iov_iter_advance(to, ret);
672 iov_iter_advance(to, 0);
674 ceph_put_page_vector(pages, num_pages, false);
677 size_t left = ret > 0 ? ret : 0;
680 page_off = off & ~PAGE_MASK;
681 len = min_t(size_t, left, PAGE_SIZE - page_off);
682 copied = copy_page_to_iter(pages[idx++],
691 ceph_release_page_vector(pages, num_pages);
694 if (ret <= 0 || off >= i_size || !more)
698 if (off > iocb->ki_pos) {
700 iov_iter_count(to) > 0 && off >= i_size_read(inode))
701 *retry_op = CHECK_EOF;
702 ret = off - iocb->ki_pos;
706 dout("sync_read result %zd retry_op %d\n", ret, *retry_op);
710 struct ceph_aio_request {
716 struct list_head osd_reqs;
718 atomic_t pending_reqs;
719 struct timespec64 mtime;
720 struct ceph_cap_flush *prealloc_cf;
723 struct ceph_aio_work {
724 struct work_struct work;
725 struct ceph_osd_request *req;
728 static void ceph_aio_retry_work(struct work_struct *work);
730 static void ceph_aio_complete(struct inode *inode,
731 struct ceph_aio_request *aio_req)
733 struct ceph_inode_info *ci = ceph_inode(inode);
736 if (!atomic_dec_and_test(&aio_req->pending_reqs))
739 ret = aio_req->error;
741 ret = aio_req->total_len;
743 dout("ceph_aio_complete %p rc %d\n", inode, ret);
745 if (ret >= 0 && aio_req->write) {
748 loff_t endoff = aio_req->iocb->ki_pos + aio_req->total_len;
749 if (endoff > i_size_read(inode)) {
750 if (ceph_inode_set_size(inode, endoff))
751 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
754 spin_lock(&ci->i_ceph_lock);
755 ci->i_inline_version = CEPH_INLINE_NONE;
756 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
757 &aio_req->prealloc_cf);
758 spin_unlock(&ci->i_ceph_lock);
760 __mark_inode_dirty(inode, dirty);
764 ceph_put_cap_refs(ci, (aio_req->write ? CEPH_CAP_FILE_WR :
767 aio_req->iocb->ki_complete(aio_req->iocb, ret, 0);
769 ceph_free_cap_flush(aio_req->prealloc_cf);
773 static void ceph_aio_complete_req(struct ceph_osd_request *req)
775 int rc = req->r_result;
776 struct inode *inode = req->r_inode;
777 struct ceph_aio_request *aio_req = req->r_priv;
778 struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0);
780 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_BVECS);
781 BUG_ON(!osd_data->num_bvecs);
783 dout("ceph_aio_complete_req %p rc %d bytes %u\n",
784 inode, rc, osd_data->bvec_pos.iter.bi_size);
786 if (rc == -EOLDSNAPC) {
787 struct ceph_aio_work *aio_work;
788 BUG_ON(!aio_req->write);
790 aio_work = kmalloc(sizeof(*aio_work), GFP_NOFS);
792 INIT_WORK(&aio_work->work, ceph_aio_retry_work);
794 queue_work(ceph_inode_to_client(inode)->inode_wq,
799 } else if (!aio_req->write) {
802 if (rc >= 0 && osd_data->bvec_pos.iter.bi_size > rc) {
804 int zlen = osd_data->bvec_pos.iter.bi_size - rc;
807 * If read is satisfied by single OSD request,
808 * it can pass EOF. Otherwise read is within
811 if (aio_req->num_reqs == 1) {
812 loff_t i_size = i_size_read(inode);
813 loff_t endoff = aio_req->iocb->ki_pos + rc;
815 zlen = min_t(size_t, zlen,
817 aio_req->total_len = rc + zlen;
820 iov_iter_bvec(&i, READ, osd_data->bvec_pos.bvecs,
822 osd_data->bvec_pos.iter.bi_size);
823 iov_iter_advance(&i, rc);
824 iov_iter_zero(zlen, &i);
828 put_bvecs(osd_data->bvec_pos.bvecs, osd_data->num_bvecs,
829 aio_req->should_dirty);
830 ceph_osdc_put_request(req);
833 cmpxchg(&aio_req->error, 0, rc);
835 ceph_aio_complete(inode, aio_req);
839 static void ceph_aio_retry_work(struct work_struct *work)
841 struct ceph_aio_work *aio_work =
842 container_of(work, struct ceph_aio_work, work);
843 struct ceph_osd_request *orig_req = aio_work->req;
844 struct ceph_aio_request *aio_req = orig_req->r_priv;
845 struct inode *inode = orig_req->r_inode;
846 struct ceph_inode_info *ci = ceph_inode(inode);
847 struct ceph_snap_context *snapc;
848 struct ceph_osd_request *req;
851 spin_lock(&ci->i_ceph_lock);
852 if (__ceph_have_pending_cap_snap(ci)) {
853 struct ceph_cap_snap *capsnap =
854 list_last_entry(&ci->i_cap_snaps,
855 struct ceph_cap_snap,
857 snapc = ceph_get_snap_context(capsnap->context);
859 BUG_ON(!ci->i_head_snapc);
860 snapc = ceph_get_snap_context(ci->i_head_snapc);
862 spin_unlock(&ci->i_ceph_lock);
864 req = ceph_osdc_alloc_request(orig_req->r_osdc, snapc, 1,
872 req->r_flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE;
873 ceph_oloc_copy(&req->r_base_oloc, &orig_req->r_base_oloc);
874 ceph_oid_copy(&req->r_base_oid, &orig_req->r_base_oid);
876 req->r_ops[0] = orig_req->r_ops[0];
878 req->r_mtime = aio_req->mtime;
879 req->r_data_offset = req->r_ops[0].extent.offset;
881 ret = ceph_osdc_alloc_messages(req, GFP_NOFS);
883 ceph_osdc_put_request(req);
888 ceph_osdc_put_request(orig_req);
890 req->r_callback = ceph_aio_complete_req;
891 req->r_inode = inode;
892 req->r_priv = aio_req;
894 ret = ceph_osdc_start_request(req->r_osdc, req, false);
898 ceph_aio_complete_req(req);
901 ceph_put_snap_context(snapc);
906 ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter,
907 struct ceph_snap_context *snapc,
908 struct ceph_cap_flush **pcf)
910 struct file *file = iocb->ki_filp;
911 struct inode *inode = file_inode(file);
912 struct ceph_inode_info *ci = ceph_inode(inode);
913 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
914 struct ceph_vino vino;
915 struct ceph_osd_request *req;
916 struct bio_vec *bvecs;
917 struct ceph_aio_request *aio_req = NULL;
921 struct timespec64 mtime = current_time(inode);
922 size_t count = iov_iter_count(iter);
923 loff_t pos = iocb->ki_pos;
924 bool write = iov_iter_rw(iter) == WRITE;
925 bool should_dirty = !write && iter_is_iovec(iter);
927 if (write && ceph_snap(file_inode(file)) != CEPH_NOSNAP)
930 dout("sync_direct_%s on file %p %lld~%u snapc %p seq %lld\n",
931 (write ? "write" : "read"), file, pos, (unsigned)count,
932 snapc, snapc ? snapc->seq : 0);
934 ret = filemap_write_and_wait_range(inode->i_mapping,
935 pos, pos + count - 1);
940 int ret2 = invalidate_inode_pages2_range(inode->i_mapping,
942 (pos + count - 1) >> PAGE_SHIFT);
944 dout("invalidate_inode_pages2_range returned %d\n", ret2);
946 flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE;
948 flags = CEPH_OSD_FLAG_READ;
951 while (iov_iter_count(iter) > 0) {
952 u64 size = iov_iter_count(iter);
956 size = min_t(u64, size, fsc->mount_options->wsize);
958 size = min_t(u64, size, fsc->mount_options->rsize);
960 vino = ceph_vino(inode);
961 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
964 write ? CEPH_OSD_OP_WRITE :
975 len = iter_get_bvecs_alloc(iter, size, &bvecs, &num_pages);
977 ceph_osdc_put_request(req);
982 osd_req_op_extent_update(req, 0, len);
985 * To simplify error handling, allow AIO when IO within i_size
986 * or IO can be satisfied by single OSD request.
988 if (pos == iocb->ki_pos && !is_sync_kiocb(iocb) &&
989 (len == count || pos + count <= i_size_read(inode))) {
990 aio_req = kzalloc(sizeof(*aio_req), GFP_KERNEL);
992 aio_req->iocb = iocb;
993 aio_req->write = write;
994 aio_req->should_dirty = should_dirty;
995 INIT_LIST_HEAD(&aio_req->osd_reqs);
997 aio_req->mtime = mtime;
998 swap(aio_req->prealloc_cf, *pcf);
1006 * throw out any page cache pages in this range. this
1009 truncate_inode_pages_range(inode->i_mapping, pos,
1010 (pos+len) | (PAGE_SIZE - 1));
1012 req->r_mtime = mtime;
1015 osd_req_op_extent_osd_data_bvecs(req, 0, bvecs, num_pages, len);
1018 aio_req->total_len += len;
1019 aio_req->num_reqs++;
1020 atomic_inc(&aio_req->pending_reqs);
1022 req->r_callback = ceph_aio_complete_req;
1023 req->r_inode = inode;
1024 req->r_priv = aio_req;
1025 list_add_tail(&req->r_unsafe_item, &aio_req->osd_reqs);
1031 ret = ceph_osdc_start_request(req->r_osdc, req, false);
1033 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1035 size = i_size_read(inode);
1039 if (ret >= 0 && ret < len && pos + ret < size) {
1041 int zlen = min_t(size_t, len - ret,
1044 iov_iter_bvec(&i, READ, bvecs, num_pages, len);
1045 iov_iter_advance(&i, ret);
1046 iov_iter_zero(zlen, &i);
1053 put_bvecs(bvecs, num_pages, should_dirty);
1054 ceph_osdc_put_request(req);
1059 if (!write && pos >= size)
1062 if (write && pos > size) {
1063 if (ceph_inode_set_size(inode, pos))
1064 ceph_check_caps(ceph_inode(inode),
1065 CHECK_CAPS_AUTHONLY,
1071 LIST_HEAD(osd_reqs);
1073 if (aio_req->num_reqs == 0) {
1078 ceph_get_cap_refs(ci, write ? CEPH_CAP_FILE_WR :
1081 list_splice(&aio_req->osd_reqs, &osd_reqs);
1082 while (!list_empty(&osd_reqs)) {
1083 req = list_first_entry(&osd_reqs,
1084 struct ceph_osd_request,
1086 list_del_init(&req->r_unsafe_item);
1088 ret = ceph_osdc_start_request(req->r_osdc,
1091 req->r_result = ret;
1092 ceph_aio_complete_req(req);
1095 return -EIOCBQUEUED;
1098 if (ret != -EOLDSNAPC && pos > iocb->ki_pos) {
1099 ret = pos - iocb->ki_pos;
1106 * Synchronous write, straight from __user pointer or user pages.
1108 * If write spans object boundary, just do multiple writes. (For a
1109 * correct atomic write, we should e.g. take write locks on all
1110 * objects, rollback on failure, etc.)
1113 ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
1114 struct ceph_snap_context *snapc)
1116 struct file *file = iocb->ki_filp;
1117 struct inode *inode = file_inode(file);
1118 struct ceph_inode_info *ci = ceph_inode(inode);
1119 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1120 struct ceph_vino vino;
1121 struct ceph_osd_request *req;
1122 struct page **pages;
1128 bool check_caps = false;
1129 struct timespec64 mtime = current_time(inode);
1130 size_t count = iov_iter_count(from);
1132 if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
1135 dout("sync_write on file %p %lld~%u snapc %p seq %lld\n",
1136 file, pos, (unsigned)count, snapc, snapc->seq);
1138 ret = filemap_write_and_wait_range(inode->i_mapping,
1139 pos, pos + count - 1);
1143 ret = invalidate_inode_pages2_range(inode->i_mapping,
1145 (pos + count - 1) >> PAGE_SHIFT);
1147 dout("invalidate_inode_pages2_range returned %d\n", ret);
1149 flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE;
1151 while ((len = iov_iter_count(from)) > 0) {
1155 vino = ceph_vino(inode);
1156 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1157 vino, pos, &len, 0, 1,
1158 CEPH_OSD_OP_WRITE, flags, snapc,
1160 ci->i_truncate_size,
1168 * write from beginning of first page,
1169 * regardless of io alignment
1171 num_pages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1173 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
1174 if (IS_ERR(pages)) {
1175 ret = PTR_ERR(pages);
1180 for (n = 0; n < num_pages; n++) {
1181 size_t plen = min_t(size_t, left, PAGE_SIZE);
1182 ret = copy_page_from_iter(pages[n], 0, plen, from);
1191 ceph_release_page_vector(pages, num_pages);
1195 req->r_inode = inode;
1197 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0,
1200 req->r_mtime = mtime;
1201 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1203 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1206 ceph_osdc_put_request(req);
1208 ceph_set_error_write(ci);
1212 ceph_clear_error_write(ci);
1215 if (pos > i_size_read(inode)) {
1216 check_caps = ceph_inode_set_size(inode, pos);
1218 ceph_check_caps(ceph_inode(inode),
1219 CHECK_CAPS_AUTHONLY,
1225 if (ret != -EOLDSNAPC && written > 0) {
1233 * Wrap generic_file_aio_read with checks for cap bits on the inode.
1234 * Atomically grab references, so that those bits are not released
1235 * back to the MDS mid-read.
1237 * Hmm, the sync read case isn't actually async... should it be?
1239 static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to)
1241 struct file *filp = iocb->ki_filp;
1242 struct ceph_file_info *fi = filp->private_data;
1243 size_t len = iov_iter_count(to);
1244 struct inode *inode = file_inode(filp);
1245 struct ceph_inode_info *ci = ceph_inode(inode);
1246 struct page *pinned_page = NULL;
1249 int retry_op = 0, read = 0;
1252 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
1253 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode);
1255 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1256 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1258 want = CEPH_CAP_FILE_CACHE;
1259 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page);
1263 if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 ||
1264 (iocb->ki_flags & IOCB_DIRECT) ||
1265 (fi->flags & CEPH_F_SYNC)) {
1267 dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n",
1268 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
1269 ceph_cap_string(got));
1271 if (ci->i_inline_version == CEPH_INLINE_NONE) {
1272 if (!retry_op && (iocb->ki_flags & IOCB_DIRECT)) {
1273 ret = ceph_direct_read_write(iocb, to,
1275 if (ret >= 0 && ret < len)
1276 retry_op = CHECK_EOF;
1278 ret = ceph_sync_read(iocb, to, &retry_op);
1281 retry_op = READ_INLINE;
1284 CEPH_DEFINE_RW_CONTEXT(rw_ctx, got);
1285 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
1286 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
1287 ceph_cap_string(got));
1288 ceph_add_rw_context(fi, &rw_ctx);
1289 ret = generic_file_read_iter(iocb, to);
1290 ceph_del_rw_context(fi, &rw_ctx);
1292 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
1293 inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
1295 put_page(pinned_page);
1298 ceph_put_cap_refs(ci, got);
1299 if (retry_op > HAVE_RETRIED && ret >= 0) {
1301 struct page *page = NULL;
1303 if (retry_op == READ_INLINE) {
1304 page = __page_cache_alloc(GFP_KERNEL);
1309 statret = __ceph_do_getattr(inode, page,
1310 CEPH_STAT_CAP_INLINE_DATA, !!page);
1314 if (statret == -ENODATA) {
1315 BUG_ON(retry_op != READ_INLINE);
1321 i_size = i_size_read(inode);
1322 if (retry_op == READ_INLINE) {
1323 BUG_ON(ret > 0 || read > 0);
1324 if (iocb->ki_pos < i_size &&
1325 iocb->ki_pos < PAGE_SIZE) {
1326 loff_t end = min_t(loff_t, i_size,
1327 iocb->ki_pos + len);
1328 end = min_t(loff_t, end, PAGE_SIZE);
1330 zero_user_segment(page, statret, end);
1331 ret = copy_page_to_iter(page,
1332 iocb->ki_pos & ~PAGE_MASK,
1333 end - iocb->ki_pos, to);
1334 iocb->ki_pos += ret;
1337 if (iocb->ki_pos < i_size && read < len) {
1338 size_t zlen = min_t(size_t, len - read,
1339 i_size - iocb->ki_pos);
1340 ret = iov_iter_zero(zlen, to);
1341 iocb->ki_pos += ret;
1344 __free_pages(page, 0);
1348 /* hit EOF or hole? */
1349 if (retry_op == CHECK_EOF && iocb->ki_pos < i_size &&
1351 dout("sync_read hit hole, ppos %lld < size %lld"
1352 ", reading more\n", iocb->ki_pos, i_size);
1356 retry_op = HAVE_RETRIED;
1368 * Take cap references to avoid releasing caps to MDS mid-write.
1370 * If we are synchronous, and write with an old snap context, the OSD
1371 * may return EOLDSNAPC. In that case, retry the write.. _after_
1372 * dropping our cap refs and allowing the pending snap to logically
1373 * complete _before_ this write occurs.
1375 * If we are near ENOSPC, write synchronously.
1377 static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
1379 struct file *file = iocb->ki_filp;
1380 struct ceph_file_info *fi = file->private_data;
1381 struct inode *inode = file_inode(file);
1382 struct ceph_inode_info *ci = ceph_inode(inode);
1383 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1384 struct ceph_cap_flush *prealloc_cf;
1385 ssize_t count, written = 0;
1388 loff_t limit = max(i_size_read(inode), fsc->max_file_size);
1390 if (ceph_snap(inode) != CEPH_NOSNAP)
1393 prealloc_cf = ceph_alloc_cap_flush();
1400 /* We can write back this queue in page reclaim */
1401 current->backing_dev_info = inode_to_bdi(inode);
1403 if (iocb->ki_flags & IOCB_APPEND) {
1404 err = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
1409 err = generic_write_checks(iocb, from);
1414 if (unlikely(pos >= limit)) {
1418 iov_iter_truncate(from, limit - pos);
1421 count = iov_iter_count(from);
1422 if (ceph_quota_is_max_bytes_exceeded(inode, pos + count)) {
1427 err = file_remove_privs(file);
1431 err = file_update_time(file);
1435 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1436 err = ceph_uninline_data(file, NULL);
1441 /* FIXME: not complete since it doesn't account for being at quota */
1442 if (ceph_osdmap_flag(&fsc->client->osdc, CEPH_OSDMAP_FULL)) {
1447 dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
1448 inode, ceph_vinop(inode), pos, count, i_size_read(inode));
1449 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1450 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1452 want = CEPH_CAP_FILE_BUFFER;
1454 err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, pos + count,
1459 dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
1460 inode, ceph_vinop(inode), pos, count, ceph_cap_string(got));
1462 if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 ||
1463 (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC) ||
1464 (ci->i_ceph_flags & CEPH_I_ERROR_WRITE)) {
1465 struct ceph_snap_context *snapc;
1466 struct iov_iter data;
1467 inode_unlock(inode);
1469 spin_lock(&ci->i_ceph_lock);
1470 if (__ceph_have_pending_cap_snap(ci)) {
1471 struct ceph_cap_snap *capsnap =
1472 list_last_entry(&ci->i_cap_snaps,
1473 struct ceph_cap_snap,
1475 snapc = ceph_get_snap_context(capsnap->context);
1477 BUG_ON(!ci->i_head_snapc);
1478 snapc = ceph_get_snap_context(ci->i_head_snapc);
1480 spin_unlock(&ci->i_ceph_lock);
1482 /* we might need to revert back to that point */
1484 if (iocb->ki_flags & IOCB_DIRECT)
1485 written = ceph_direct_read_write(iocb, &data, snapc,
1488 written = ceph_sync_write(iocb, &data, pos, snapc);
1490 iov_iter_advance(from, written);
1491 ceph_put_snap_context(snapc);
1494 * No need to acquire the i_truncate_mutex. Because
1495 * the MDS revokes Fwb caps before sending truncate
1496 * message to us. We can't get Fwb cap while there
1497 * are pending vmtruncate. So write and vmtruncate
1498 * can not run at the same time
1500 written = generic_perform_write(file, from, pos);
1501 if (likely(written >= 0))
1502 iocb->ki_pos = pos + written;
1503 inode_unlock(inode);
1509 spin_lock(&ci->i_ceph_lock);
1510 ci->i_inline_version = CEPH_INLINE_NONE;
1511 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1513 spin_unlock(&ci->i_ceph_lock);
1515 __mark_inode_dirty(inode, dirty);
1516 if (ceph_quota_is_max_bytes_approaching(inode, iocb->ki_pos))
1517 ceph_check_caps(ci, CHECK_CAPS_NODELAY, NULL);
1520 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
1521 inode, ceph_vinop(inode), pos, (unsigned)count,
1522 ceph_cap_string(got));
1523 ceph_put_cap_refs(ci, got);
1525 if (written == -EOLDSNAPC) {
1526 dout("aio_write %p %llx.%llx %llu~%u" "got EOLDSNAPC, retrying\n",
1527 inode, ceph_vinop(inode), pos, (unsigned)count);
1532 if (ceph_osdmap_flag(&fsc->client->osdc, CEPH_OSDMAP_NEARFULL))
1533 iocb->ki_flags |= IOCB_DSYNC;
1534 written = generic_write_sync(iocb, written);
1540 inode_unlock(inode);
1542 ceph_free_cap_flush(prealloc_cf);
1543 current->backing_dev_info = NULL;
1544 return written ? written : err;
1548 * llseek. be sure to verify file size on SEEK_END.
1550 static loff_t ceph_llseek(struct file *file, loff_t offset, int whence)
1552 struct inode *inode = file->f_mapping->host;
1553 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1559 if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) {
1560 ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
1565 i_size = i_size_read(inode);
1572 * Here we special-case the lseek(fd, 0, SEEK_CUR)
1573 * position-querying operation. Avoid rewriting the "same"
1574 * f_pos value back to the file because a concurrent read(),
1575 * write() or lseek() might have altered it
1581 offset += file->f_pos;
1584 if (offset < 0 || offset >= i_size) {
1590 if (offset < 0 || offset >= i_size) {
1598 ret = vfs_setpos(file, offset, max(i_size, fsc->max_file_size));
1601 inode_unlock(inode);
1605 static inline void ceph_zero_partial_page(
1606 struct inode *inode, loff_t offset, unsigned size)
1609 pgoff_t index = offset >> PAGE_SHIFT;
1611 page = find_lock_page(inode->i_mapping, index);
1613 wait_on_page_writeback(page);
1614 zero_user(page, offset & (PAGE_SIZE - 1), size);
1620 static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset,
1623 loff_t nearly = round_up(offset, PAGE_SIZE);
1624 if (offset < nearly) {
1625 loff_t size = nearly - offset;
1628 ceph_zero_partial_page(inode, offset, size);
1632 if (length >= PAGE_SIZE) {
1633 loff_t size = round_down(length, PAGE_SIZE);
1634 truncate_pagecache_range(inode, offset, offset + size - 1);
1639 ceph_zero_partial_page(inode, offset, length);
1642 static int ceph_zero_partial_object(struct inode *inode,
1643 loff_t offset, loff_t *length)
1645 struct ceph_inode_info *ci = ceph_inode(inode);
1646 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1647 struct ceph_osd_request *req;
1653 op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE;
1656 op = CEPH_OSD_OP_ZERO;
1659 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1663 CEPH_OSD_FLAG_WRITE,
1670 req->r_mtime = inode->i_mtime;
1671 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1673 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1677 ceph_osdc_put_request(req);
1683 static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length)
1686 struct ceph_inode_info *ci = ceph_inode(inode);
1687 s32 stripe_unit = ci->i_layout.stripe_unit;
1688 s32 stripe_count = ci->i_layout.stripe_count;
1689 s32 object_size = ci->i_layout.object_size;
1690 u64 object_set_size = object_size * stripe_count;
1693 /* round offset up to next period boundary */
1694 nearly = offset + object_set_size - 1;
1696 nearly -= do_div(t, object_set_size);
1698 while (length && offset < nearly) {
1699 loff_t size = length;
1700 ret = ceph_zero_partial_object(inode, offset, &size);
1706 while (length >= object_set_size) {
1708 loff_t pos = offset;
1709 for (i = 0; i < stripe_count; ++i) {
1710 ret = ceph_zero_partial_object(inode, pos, NULL);
1715 offset += object_set_size;
1716 length -= object_set_size;
1719 loff_t size = length;
1720 ret = ceph_zero_partial_object(inode, offset, &size);
1729 static long ceph_fallocate(struct file *file, int mode,
1730 loff_t offset, loff_t length)
1732 struct ceph_file_info *fi = file->private_data;
1733 struct inode *inode = file_inode(file);
1734 struct ceph_inode_info *ci = ceph_inode(inode);
1735 struct ceph_cap_flush *prealloc_cf;
1742 if (mode != (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1745 if (!S_ISREG(inode->i_mode))
1748 prealloc_cf = ceph_alloc_cap_flush();
1754 if (ceph_snap(inode) != CEPH_NOSNAP) {
1759 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1760 ret = ceph_uninline_data(file, NULL);
1765 size = i_size_read(inode);
1767 /* Are we punching a hole beyond EOF? */
1770 if ((offset + length) > size)
1771 length = size - offset;
1773 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1774 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1776 want = CEPH_CAP_FILE_BUFFER;
1778 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, endoff, &got, NULL);
1782 ceph_zero_pagecache_range(inode, offset, length);
1783 ret = ceph_zero_objects(inode, offset, length);
1786 spin_lock(&ci->i_ceph_lock);
1787 ci->i_inline_version = CEPH_INLINE_NONE;
1788 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1790 spin_unlock(&ci->i_ceph_lock);
1792 __mark_inode_dirty(inode, dirty);
1795 ceph_put_cap_refs(ci, got);
1797 inode_unlock(inode);
1798 ceph_free_cap_flush(prealloc_cf);
1803 * This function tries to get FILE_WR capabilities for dst_ci and FILE_RD for
1804 * src_ci. Two attempts are made to obtain both caps, and an error is return if
1805 * this fails; zero is returned on success.
1807 static int get_rd_wr_caps(struct ceph_inode_info *src_ci,
1808 loff_t src_endoff, int *src_got,
1809 struct ceph_inode_info *dst_ci,
1810 loff_t dst_endoff, int *dst_got)
1813 bool retrying = false;
1816 ret = ceph_get_caps(dst_ci, CEPH_CAP_FILE_WR, CEPH_CAP_FILE_BUFFER,
1817 dst_endoff, dst_got, NULL);
1822 * Since we're already holding the FILE_WR capability for the dst file,
1823 * we would risk a deadlock by using ceph_get_caps. Thus, we'll do some
1824 * retry dance instead to try to get both capabilities.
1826 ret = ceph_try_get_caps(src_ci, CEPH_CAP_FILE_RD, CEPH_CAP_FILE_SHARED,
1829 /* Start by dropping dst_ci caps and getting src_ci caps */
1830 ceph_put_cap_refs(dst_ci, *dst_got);
1833 /* ceph_try_get_caps masks EAGAIN */
1837 ret = ceph_get_caps(src_ci, CEPH_CAP_FILE_RD,
1838 CEPH_CAP_FILE_SHARED, src_endoff,
1842 /*... drop src_ci caps too, and retry */
1843 ceph_put_cap_refs(src_ci, *src_got);
1850 static void put_rd_wr_caps(struct ceph_inode_info *src_ci, int src_got,
1851 struct ceph_inode_info *dst_ci, int dst_got)
1853 ceph_put_cap_refs(src_ci, src_got);
1854 ceph_put_cap_refs(dst_ci, dst_got);
1858 * This function does several size-related checks, returning an error if:
1859 * - source file is smaller than off+len
1860 * - destination file size is not OK (inode_newsize_ok())
1861 * - max bytes quotas is exceeded
1863 static int is_file_size_ok(struct inode *src_inode, struct inode *dst_inode,
1864 loff_t src_off, loff_t dst_off, size_t len)
1866 loff_t size, endoff;
1868 size = i_size_read(src_inode);
1870 * Don't copy beyond source file EOF. Instead of simply setting length
1871 * to (size - src_off), just drop to VFS default implementation, as the
1872 * local i_size may be stale due to other clients writing to the source
1875 if (src_off + len > size) {
1876 dout("Copy beyond EOF (%llu + %zu > %llu)\n",
1877 src_off, len, size);
1880 size = i_size_read(dst_inode);
1882 endoff = dst_off + len;
1883 if (inode_newsize_ok(dst_inode, endoff))
1886 if (ceph_quota_is_max_bytes_exceeded(dst_inode, endoff))
1892 static ssize_t ceph_copy_file_range(struct file *src_file, loff_t src_off,
1893 struct file *dst_file, loff_t dst_off,
1894 size_t len, unsigned int flags)
1896 struct inode *src_inode = file_inode(src_file);
1897 struct inode *dst_inode = file_inode(dst_file);
1898 struct ceph_inode_info *src_ci = ceph_inode(src_inode);
1899 struct ceph_inode_info *dst_ci = ceph_inode(dst_inode);
1900 struct ceph_cap_flush *prealloc_cf;
1901 struct ceph_object_locator src_oloc, dst_oloc;
1902 struct ceph_object_id src_oid, dst_oid;
1903 loff_t endoff = 0, size;
1905 u64 src_objnum, dst_objnum, src_objoff, dst_objoff;
1906 u32 src_objlen, dst_objlen, object_size;
1907 int src_got = 0, dst_got = 0, err, dirty;
1908 bool do_final_copy = false;
1910 if (src_inode == dst_inode)
1912 if (ceph_snap(dst_inode) != CEPH_NOSNAP)
1916 * Some of the checks below will return -EOPNOTSUPP, which will force a
1917 * fallback to the default VFS copy_file_range implementation. This is
1918 * desirable in several cases (for ex, the 'len' is smaller than the
1919 * size of the objects, or in cases where that would be more
1923 if (ceph_test_mount_opt(ceph_inode_to_client(src_inode), NOCOPYFROM))
1926 if ((src_ci->i_layout.stripe_unit != dst_ci->i_layout.stripe_unit) ||
1927 (src_ci->i_layout.stripe_count != dst_ci->i_layout.stripe_count) ||
1928 (src_ci->i_layout.object_size != dst_ci->i_layout.object_size))
1931 if (len < src_ci->i_layout.object_size)
1932 return -EOPNOTSUPP; /* no remote copy will be done */
1934 prealloc_cf = ceph_alloc_cap_flush();
1938 /* Start by sync'ing the source and destination files */
1939 ret = file_write_and_wait_range(src_file, src_off, (src_off + len));
1941 dout("failed to write src file (%zd)\n", ret);
1944 ret = file_write_and_wait_range(dst_file, dst_off, (dst_off + len));
1946 dout("failed to write dst file (%zd)\n", ret);
1951 * We need FILE_WR caps for dst_ci and FILE_RD for src_ci as other
1952 * clients may have dirty data in their caches. And OSDs know nothing
1953 * about caps, so they can't safely do the remote object copies.
1955 err = get_rd_wr_caps(src_ci, (src_off + len), &src_got,
1956 dst_ci, (dst_off + len), &dst_got);
1958 dout("get_rd_wr_caps returned %d\n", err);
1963 ret = is_file_size_ok(src_inode, dst_inode, src_off, dst_off, len);
1967 size = i_size_read(dst_inode);
1968 endoff = dst_off + len;
1970 /* Drop dst file cached pages */
1971 ret = invalidate_inode_pages2_range(dst_inode->i_mapping,
1972 dst_off >> PAGE_SHIFT,
1973 endoff >> PAGE_SHIFT);
1975 dout("Failed to invalidate inode pages (%zd)\n", ret);
1978 src_oloc.pool = src_ci->i_layout.pool_id;
1979 src_oloc.pool_ns = ceph_try_get_string(src_ci->i_layout.pool_ns);
1980 dst_oloc.pool = dst_ci->i_layout.pool_id;
1981 dst_oloc.pool_ns = ceph_try_get_string(dst_ci->i_layout.pool_ns);
1983 ceph_calc_file_object_mapping(&src_ci->i_layout, src_off,
1984 src_ci->i_layout.object_size,
1985 &src_objnum, &src_objoff, &src_objlen);
1986 ceph_calc_file_object_mapping(&dst_ci->i_layout, dst_off,
1987 dst_ci->i_layout.object_size,
1988 &dst_objnum, &dst_objoff, &dst_objlen);
1989 /* object-level offsets need to the same */
1990 if (src_objoff != dst_objoff) {
1996 * Do a manual copy if the object offset isn't object aligned.
1997 * 'src_objlen' contains the bytes left until the end of the object,
1998 * starting at the src_off
2002 * we need to temporarily drop all caps as we'll be calling
2003 * {read,write}_iter, which will get caps again.
2005 put_rd_wr_caps(src_ci, src_got, dst_ci, dst_got);
2006 ret = do_splice_direct(src_file, &src_off, dst_file,
2007 &dst_off, src_objlen, flags);
2009 dout("do_splice_direct returned %d\n", err);
2013 err = get_rd_wr_caps(src_ci, (src_off + len),
2015 (dst_off + len), &dst_got);
2018 err = is_file_size_ok(src_inode, dst_inode,
2019 src_off, dst_off, len);
2023 object_size = src_ci->i_layout.object_size;
2024 while (len >= object_size) {
2025 ceph_calc_file_object_mapping(&src_ci->i_layout, src_off,
2026 object_size, &src_objnum,
2027 &src_objoff, &src_objlen);
2028 ceph_calc_file_object_mapping(&dst_ci->i_layout, dst_off,
2029 object_size, &dst_objnum,
2030 &dst_objoff, &dst_objlen);
2031 ceph_oid_init(&src_oid);
2032 ceph_oid_printf(&src_oid, "%llx.%08llx",
2033 src_ci->i_vino.ino, src_objnum);
2034 ceph_oid_init(&dst_oid);
2035 ceph_oid_printf(&dst_oid, "%llx.%08llx",
2036 dst_ci->i_vino.ino, dst_objnum);
2037 /* Do an object remote copy */
2038 err = ceph_osdc_copy_from(
2039 &ceph_inode_to_client(src_inode)->client->osdc,
2040 src_ci->i_vino.snap, 0,
2041 &src_oid, &src_oloc,
2042 CEPH_OSD_OP_FLAG_FADVISE_SEQUENTIAL |
2043 CEPH_OSD_OP_FLAG_FADVISE_NOCACHE,
2044 &dst_oid, &dst_oloc,
2045 CEPH_OSD_OP_FLAG_FADVISE_SEQUENTIAL |
2046 CEPH_OSD_OP_FLAG_FADVISE_DONTNEED, 0);
2048 dout("ceph_osdc_copy_from returned %d\n", err);
2054 src_off += object_size;
2055 dst_off += object_size;
2060 /* We still need one final local copy */
2061 do_final_copy = true;
2063 file_update_time(dst_file);
2064 if (endoff > size) {
2067 /* Let the MDS know about dst file size change */
2068 if (ceph_quota_is_max_bytes_approaching(dst_inode, endoff))
2069 caps_flags |= CHECK_CAPS_NODELAY;
2070 if (ceph_inode_set_size(dst_inode, endoff))
2071 caps_flags |= CHECK_CAPS_AUTHONLY;
2073 ceph_check_caps(dst_ci, caps_flags, NULL);
2076 spin_lock(&dst_ci->i_ceph_lock);
2077 dst_ci->i_inline_version = CEPH_INLINE_NONE;
2078 dirty = __ceph_mark_dirty_caps(dst_ci, CEPH_CAP_FILE_WR, &prealloc_cf);
2079 spin_unlock(&dst_ci->i_ceph_lock);
2081 __mark_inode_dirty(dst_inode, dirty);
2084 put_rd_wr_caps(src_ci, src_got, dst_ci, dst_got);
2086 if (do_final_copy) {
2087 err = do_splice_direct(src_file, &src_off, dst_file,
2088 &dst_off, len, flags);
2090 dout("do_splice_direct returned %d\n", err);
2098 ceph_free_cap_flush(prealloc_cf);
2103 const struct file_operations ceph_file_fops = {
2105 .release = ceph_release,
2106 .llseek = ceph_llseek,
2107 .read_iter = ceph_read_iter,
2108 .write_iter = ceph_write_iter,
2110 .fsync = ceph_fsync,
2112 .flock = ceph_flock,
2113 .splice_read = generic_file_splice_read,
2114 .splice_write = iter_file_splice_write,
2115 .unlocked_ioctl = ceph_ioctl,
2116 .compat_ioctl = ceph_ioctl,
2117 .fallocate = ceph_fallocate,
2118 .copy_file_range = ceph_copy_file_range,