1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
4 #include <linux/module.h>
5 #include <linux/sched.h>
6 #include <linux/slab.h>
7 #include <linux/file.h>
8 #include <linux/mount.h>
9 #include <linux/namei.h>
10 #include <linux/writeback.h>
11 #include <linux/falloc.h>
14 #include "mds_client.h"
17 static __le32 ceph_flags_sys2wire(u32 flags)
21 switch (flags & O_ACCMODE) {
23 wire_flags |= CEPH_O_RDONLY;
26 wire_flags |= CEPH_O_WRONLY;
29 wire_flags |= CEPH_O_RDWR;
35 #define ceph_sys2wire(a) if (flags & a) { wire_flags |= CEPH_##a; flags &= ~a; }
37 ceph_sys2wire(O_CREAT);
38 ceph_sys2wire(O_EXCL);
39 ceph_sys2wire(O_TRUNC);
40 ceph_sys2wire(O_DIRECTORY);
41 ceph_sys2wire(O_NOFOLLOW);
46 dout("unused open flags: %x\n", flags);
48 return cpu_to_le32(wire_flags);
52 * Ceph file operations
54 * Implement basic open/close functionality, and implement
57 * We implement three modes of file I/O:
58 * - buffered uses the generic_file_aio_{read,write} helpers
60 * - synchronous is used when there is multi-client read/write
61 * sharing, avoids the page cache, and synchronously waits for an
64 * - direct io takes the variant of the sync path that references
65 * user pages directly.
67 * fsync() flushes and waits on dirty pages, but just queues metadata
68 * for writeback: since the MDS can recover size and mtime there is no
69 * need to wait for MDS acknowledgement.
73 * How many pages to get in one call to iov_iter_get_pages(). This
74 * determines the size of the on-stack array used as a buffer.
76 #define ITER_GET_BVECS_PAGES 64
78 static ssize_t __iter_get_bvecs(struct iov_iter *iter, size_t maxsize,
79 struct bio_vec *bvecs)
84 if (maxsize > iov_iter_count(iter))
85 maxsize = iov_iter_count(iter);
87 while (size < maxsize) {
88 struct page *pages[ITER_GET_BVECS_PAGES];
93 bytes = iov_iter_get_pages(iter, pages, maxsize - size,
94 ITER_GET_BVECS_PAGES, &start);
98 iov_iter_advance(iter, bytes);
101 for ( ; bytes; idx++, bvec_idx++) {
102 struct bio_vec bv = {
103 .bv_page = pages[idx],
104 .bv_len = min_t(int, bytes, PAGE_SIZE - start),
108 bvecs[bvec_idx] = bv;
118 * iov_iter_get_pages() only considers one iov_iter segment, no matter
119 * what maxsize or maxpages are given. For ITER_BVEC that is a single
122 * Attempt to get up to @maxsize bytes worth of pages from @iter.
123 * Return the number of bytes in the created bio_vec array, or an error.
125 static ssize_t iter_get_bvecs_alloc(struct iov_iter *iter, size_t maxsize,
126 struct bio_vec **bvecs, int *num_bvecs)
129 size_t orig_count = iov_iter_count(iter);
133 iov_iter_truncate(iter, maxsize);
134 npages = iov_iter_npages(iter, INT_MAX);
135 iov_iter_reexpand(iter, orig_count);
138 * __iter_get_bvecs() may populate only part of the array -- zero it
141 bv = kvmalloc_array(npages, sizeof(*bv), GFP_KERNEL | __GFP_ZERO);
145 bytes = __iter_get_bvecs(iter, maxsize, bv);
148 * No pages were pinned -- just free the array.
159 static void put_bvecs(struct bio_vec *bvecs, int num_bvecs, bool should_dirty)
163 for (i = 0; i < num_bvecs; i++) {
164 if (bvecs[i].bv_page) {
166 set_page_dirty_lock(bvecs[i].bv_page);
167 put_page(bvecs[i].bv_page);
174 * Prepare an open request. Preallocate ceph_cap to avoid an
175 * inopportune ENOMEM later.
177 static struct ceph_mds_request *
178 prepare_open_request(struct super_block *sb, int flags, int create_mode)
180 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
181 struct ceph_mds_client *mdsc = fsc->mdsc;
182 struct ceph_mds_request *req;
183 int want_auth = USE_ANY_MDS;
184 int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN;
186 if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC))
187 want_auth = USE_AUTH_MDS;
189 req = ceph_mdsc_create_request(mdsc, op, want_auth);
192 req->r_fmode = ceph_flags_to_mode(flags);
193 req->r_args.open.flags = ceph_flags_sys2wire(flags);
194 req->r_args.open.mode = cpu_to_le32(create_mode);
199 static int ceph_init_file_info(struct inode *inode, struct file *file,
200 int fmode, bool isdir)
202 struct ceph_file_info *fi;
204 dout("%s %p %p 0%o (%s)\n", __func__, inode, file,
205 inode->i_mode, isdir ? "dir" : "regular");
206 BUG_ON(inode->i_fop->release != ceph_release);
209 struct ceph_dir_file_info *dfi =
210 kmem_cache_zalloc(ceph_dir_file_cachep, GFP_KERNEL);
212 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
216 file->private_data = dfi;
217 fi = &dfi->file_info;
218 dfi->next_offset = 2;
219 dfi->readdir_cache_idx = -1;
221 fi = kmem_cache_zalloc(ceph_file_cachep, GFP_KERNEL);
223 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
227 file->private_data = fi;
231 spin_lock_init(&fi->rw_contexts_lock);
232 INIT_LIST_HEAD(&fi->rw_contexts);
238 * initialize private struct file data.
239 * if we fail, clean up by dropping fmode reference on the ceph_inode
241 static int ceph_init_file(struct inode *inode, struct file *file, int fmode)
245 switch (inode->i_mode & S_IFMT) {
247 ceph_fscache_register_inode_cookie(inode);
248 ceph_fscache_file_set_cookie(inode, file);
250 ret = ceph_init_file_info(inode, file, fmode,
251 S_ISDIR(inode->i_mode));
257 dout("init_file %p %p 0%o (symlink)\n", inode, file,
259 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
263 dout("init_file %p %p 0%o (special)\n", inode, file,
266 * we need to drop the open ref now, since we don't
267 * have .release set to ceph_release.
269 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
270 BUG_ON(inode->i_fop->release == ceph_release);
272 /* call the proper open fop */
273 ret = inode->i_fop->open(inode, file);
279 * try renew caps after session gets killed.
281 int ceph_renew_caps(struct inode *inode)
283 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
284 struct ceph_inode_info *ci = ceph_inode(inode);
285 struct ceph_mds_request *req;
286 int err, flags, wanted;
288 spin_lock(&ci->i_ceph_lock);
289 wanted = __ceph_caps_file_wanted(ci);
290 if (__ceph_is_any_real_caps(ci) &&
291 (!(wanted & CEPH_CAP_ANY_WR) || ci->i_auth_cap)) {
292 int issued = __ceph_caps_issued(ci, NULL);
293 spin_unlock(&ci->i_ceph_lock);
294 dout("renew caps %p want %s issued %s updating mds_wanted\n",
295 inode, ceph_cap_string(wanted), ceph_cap_string(issued));
296 ceph_check_caps(ci, 0, NULL);
299 spin_unlock(&ci->i_ceph_lock);
302 if ((wanted & CEPH_CAP_FILE_RD) && (wanted & CEPH_CAP_FILE_WR))
304 else if (wanted & CEPH_CAP_FILE_RD)
306 else if (wanted & CEPH_CAP_FILE_WR)
309 if (wanted & CEPH_CAP_FILE_LAZYIO)
313 req = prepare_open_request(inode->i_sb, flags, 0);
319 req->r_inode = inode;
324 err = ceph_mdsc_do_request(mdsc, NULL, req);
325 ceph_mdsc_put_request(req);
327 dout("renew caps %p open result=%d\n", inode, err);
328 return err < 0 ? err : 0;
332 * If we already have the requisite capabilities, we can satisfy
333 * the open request locally (no need to request new caps from the
334 * MDS). We do, however, need to inform the MDS (asynchronously)
335 * if our wanted caps set expands.
337 int ceph_open(struct inode *inode, struct file *file)
339 struct ceph_inode_info *ci = ceph_inode(inode);
340 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
341 struct ceph_mds_client *mdsc = fsc->mdsc;
342 struct ceph_mds_request *req;
343 struct ceph_file_info *fi = file->private_data;
345 int flags, fmode, wanted;
348 dout("open file %p is already opened\n", file);
352 /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */
353 flags = file->f_flags & ~(O_CREAT|O_EXCL);
354 if (S_ISDIR(inode->i_mode))
355 flags = O_DIRECTORY; /* mds likes to know */
357 dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode,
358 ceph_vinop(inode), file, flags, file->f_flags);
359 fmode = ceph_flags_to_mode(flags);
360 wanted = ceph_caps_for_mode(fmode);
362 /* snapped files are read-only */
363 if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE))
366 /* trivially open snapdir */
367 if (ceph_snap(inode) == CEPH_SNAPDIR) {
368 spin_lock(&ci->i_ceph_lock);
369 __ceph_get_fmode(ci, fmode);
370 spin_unlock(&ci->i_ceph_lock);
371 return ceph_init_file(inode, file, fmode);
375 * No need to block if we have caps on the auth MDS (for
376 * write) or any MDS (for read). Update wanted set
379 spin_lock(&ci->i_ceph_lock);
380 if (__ceph_is_any_real_caps(ci) &&
381 (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) {
382 int mds_wanted = __ceph_caps_mds_wanted(ci, true);
383 int issued = __ceph_caps_issued(ci, NULL);
385 dout("open %p fmode %d want %s issued %s using existing\n",
386 inode, fmode, ceph_cap_string(wanted),
387 ceph_cap_string(issued));
388 __ceph_get_fmode(ci, fmode);
389 spin_unlock(&ci->i_ceph_lock);
392 if ((issued & wanted) != wanted &&
393 (mds_wanted & wanted) != wanted &&
394 ceph_snap(inode) != CEPH_SNAPDIR)
395 ceph_check_caps(ci, 0, NULL);
397 return ceph_init_file(inode, file, fmode);
398 } else if (ceph_snap(inode) != CEPH_NOSNAP &&
399 (ci->i_snap_caps & wanted) == wanted) {
400 __ceph_get_fmode(ci, fmode);
401 spin_unlock(&ci->i_ceph_lock);
402 return ceph_init_file(inode, file, fmode);
405 spin_unlock(&ci->i_ceph_lock);
407 dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted));
408 req = prepare_open_request(inode->i_sb, flags, 0);
413 req->r_inode = inode;
417 err = ceph_mdsc_do_request(mdsc, NULL, req);
419 err = ceph_init_file(inode, file, req->r_fmode);
420 ceph_mdsc_put_request(req);
421 dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode));
428 * Do a lookup + open with a single request. If we get a non-existent
429 * file or symlink, return 1 so the VFS can retry.
431 int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
432 struct file *file, unsigned flags, umode_t mode,
435 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
436 struct ceph_mds_client *mdsc = fsc->mdsc;
437 struct ceph_mds_request *req;
439 struct ceph_acls_info acls = {};
443 dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n",
445 d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode);
447 if (dentry->d_name.len > NAME_MAX)
448 return -ENAMETOOLONG;
450 if (flags & O_CREAT) {
451 if (ceph_quota_is_max_files_exceeded(dir))
453 err = ceph_pre_init_acls(dir, &mode, &acls);
459 req = prepare_open_request(dir->i_sb, flags, mode);
464 req->r_dentry = dget(dentry);
466 if (flags & O_CREAT) {
467 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
468 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
470 req->r_pagelist = acls.pagelist;
471 acls.pagelist = NULL;
475 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
476 if (ceph_security_xattr_wanted(dir))
477 mask |= CEPH_CAP_XATTR_SHARED;
478 req->r_args.open.mask = cpu_to_le32(mask);
481 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
482 err = ceph_mdsc_do_request(mdsc,
483 (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
485 err = ceph_handle_snapdir(req, dentry, err);
489 if ((flags & O_CREAT) && !req->r_reply_info.head->is_dentry)
490 err = ceph_handle_notrace_create(dir, dentry);
492 if (d_in_lookup(dentry)) {
493 dn = ceph_finish_lookup(req, dentry, err);
497 /* we were given a hashed negative dentry */
502 if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) {
503 /* make vfs retry on splice, ENOENT, or symlink */
504 dout("atomic_open finish_no_open on dn %p\n", dn);
505 err = finish_no_open(file, dn);
507 dout("atomic_open finish_open on dn %p\n", dn);
508 if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) {
509 ceph_init_inode_acls(d_inode(dentry), &acls);
510 *opened |= FILE_CREATED;
512 err = finish_open(file, dentry, ceph_open, opened);
515 if (!req->r_err && req->r_target_inode)
516 ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode);
517 ceph_mdsc_put_request(req);
519 ceph_release_acls_info(&acls);
520 dout("atomic_open result=%d\n", err);
524 int ceph_release(struct inode *inode, struct file *file)
526 struct ceph_inode_info *ci = ceph_inode(inode);
528 if (S_ISDIR(inode->i_mode)) {
529 struct ceph_dir_file_info *dfi = file->private_data;
530 dout("release inode %p dir file %p\n", inode, file);
531 WARN_ON(!list_empty(&dfi->file_info.rw_contexts));
533 ceph_put_fmode(ci, dfi->file_info.fmode);
535 if (dfi->last_readdir)
536 ceph_mdsc_put_request(dfi->last_readdir);
537 kfree(dfi->last_name);
538 kfree(dfi->dir_info);
539 kmem_cache_free(ceph_dir_file_cachep, dfi);
541 struct ceph_file_info *fi = file->private_data;
542 dout("release inode %p regular file %p\n", inode, file);
543 WARN_ON(!list_empty(&fi->rw_contexts));
545 ceph_put_fmode(ci, fi->fmode);
546 kmem_cache_free(ceph_file_cachep, fi);
549 /* wake up anyone waiting for caps on this inode */
550 wake_up_all(&ci->i_cap_wq);
561 * Read a range of bytes striped over one or more objects. Iterate over
562 * objects we stripe over. (That's not atomic, but good enough for now.)
564 * If we get a short result from the OSD, check against i_size; we need to
565 * only return a short read to the caller if we hit EOF.
567 static int striped_read(struct inode *inode,
569 struct page **pages, int num_pages,
570 int page_align, int *checkeof)
572 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
573 struct ceph_inode_info *ci = ceph_inode(inode);
578 bool hit_stripe, was_short;
581 * we may need to do multiple reads. not atomic, unfortunately.
585 page_idx = (page_align + read) >> PAGE_SHIFT;
586 ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode),
587 &ci->i_layout, pos, &this_len,
588 ci->i_truncate_seq, ci->i_truncate_size,
589 pages + page_idx, num_pages - page_idx,
590 ((page_align + read) & ~PAGE_MASK));
593 hit_stripe = this_len < len;
594 was_short = ret >= 0 && ret < this_len;
595 dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, len, read,
596 ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : "");
598 i_size = i_size_read(inode);
600 if (was_short && (pos + ret < i_size)) {
601 int zlen = min(this_len - ret, i_size - pos - ret);
602 int zoff = page_align + read + ret;
603 dout(" zero gap %llu to %llu\n",
604 pos + ret, pos + ret + zlen);
605 ceph_zero_page_vector_range(zoff, zlen, pages);
613 /* hit stripe and need continue*/
614 if (len && hit_stripe && pos < i_size)
620 /* did we bounce off eof? */
621 if (pos + len > i_size)
622 *checkeof = CHECK_EOF;
625 dout("striped_read returns %d\n", ret);
630 * Completely synchronous read and write methods. Direct from __user
631 * buffer to osd, or directly to user pages (if O_DIRECT).
633 * If the read spans object boundary, just do multiple reads.
635 static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *to,
638 struct file *file = iocb->ki_filp;
639 struct inode *inode = file_inode(file);
641 u64 off = iocb->ki_pos;
644 size_t len = iov_iter_count(to);
646 dout("sync_read on file %p %llu~%u %s\n", file, off, (unsigned)len,
647 (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
652 * flush any page cache pages in this range. this
653 * will make concurrent normal and sync io slow,
654 * but it will at least behave sensibly when they are
657 ret = filemap_write_and_wait_range(inode->i_mapping, off,
662 if (unlikely(to->type & ITER_PIPE)) {
664 ret = iov_iter_get_pages_alloc(to, &pages, len,
668 num_pages = DIV_ROUND_UP(ret + page_off, PAGE_SIZE);
670 ret = striped_read(inode, off, ret, pages, num_pages,
673 iov_iter_advance(to, ret);
676 iov_iter_advance(to, 0);
678 ceph_put_page_vector(pages, num_pages, false);
680 num_pages = calc_pages_for(off, len);
681 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
683 return PTR_ERR(pages);
685 ret = striped_read(inode, off, len, pages, num_pages,
686 (off & ~PAGE_MASK), checkeof);
692 size_t page_off = off & ~PAGE_MASK;
693 size_t copy = min_t(size_t, left,
694 PAGE_SIZE - page_off);
695 l = copy_page_to_iter(pages[k++], page_off,
703 ceph_release_page_vector(pages, num_pages);
706 if (off > iocb->ki_pos) {
707 ret = off - iocb->ki_pos;
711 dout("sync_read result %zd\n", ret);
715 struct ceph_aio_request {
721 struct list_head osd_reqs;
723 atomic_t pending_reqs;
724 struct timespec mtime;
725 struct ceph_cap_flush *prealloc_cf;
728 struct ceph_aio_work {
729 struct work_struct work;
730 struct ceph_osd_request *req;
733 static void ceph_aio_retry_work(struct work_struct *work);
735 static void ceph_aio_complete(struct inode *inode,
736 struct ceph_aio_request *aio_req)
738 struct ceph_inode_info *ci = ceph_inode(inode);
741 if (!atomic_dec_and_test(&aio_req->pending_reqs))
744 ret = aio_req->error;
746 ret = aio_req->total_len;
748 dout("ceph_aio_complete %p rc %d\n", inode, ret);
750 if (ret >= 0 && aio_req->write) {
753 loff_t endoff = aio_req->iocb->ki_pos + aio_req->total_len;
754 if (endoff > i_size_read(inode)) {
755 if (ceph_inode_set_size(inode, endoff))
756 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
759 spin_lock(&ci->i_ceph_lock);
760 ci->i_inline_version = CEPH_INLINE_NONE;
761 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
762 &aio_req->prealloc_cf);
763 spin_unlock(&ci->i_ceph_lock);
765 __mark_inode_dirty(inode, dirty);
769 ceph_put_cap_refs(ci, (aio_req->write ? CEPH_CAP_FILE_WR :
772 aio_req->iocb->ki_complete(aio_req->iocb, ret, 0);
774 ceph_free_cap_flush(aio_req->prealloc_cf);
778 static void ceph_aio_complete_req(struct ceph_osd_request *req)
780 int rc = req->r_result;
781 struct inode *inode = req->r_inode;
782 struct ceph_aio_request *aio_req = req->r_priv;
783 struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0);
785 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_BVECS);
786 BUG_ON(!osd_data->num_bvecs);
788 dout("ceph_aio_complete_req %p rc %d bytes %u\n",
789 inode, rc, osd_data->bvec_pos.iter.bi_size);
791 if (rc == -EOLDSNAPC) {
792 struct ceph_aio_work *aio_work;
793 BUG_ON(!aio_req->write);
795 aio_work = kmalloc(sizeof(*aio_work), GFP_NOFS);
797 INIT_WORK(&aio_work->work, ceph_aio_retry_work);
799 queue_work(ceph_inode_to_client(inode)->wb_wq,
804 } else if (!aio_req->write) {
807 if (rc >= 0 && osd_data->bvec_pos.iter.bi_size > rc) {
809 int zlen = osd_data->bvec_pos.iter.bi_size - rc;
812 * If read is satisfied by single OSD request,
813 * it can pass EOF. Otherwise read is within
816 if (aio_req->num_reqs == 1) {
817 loff_t i_size = i_size_read(inode);
818 loff_t endoff = aio_req->iocb->ki_pos + rc;
820 zlen = min_t(size_t, zlen,
822 aio_req->total_len = rc + zlen;
825 iov_iter_bvec(&i, ITER_BVEC, osd_data->bvec_pos.bvecs,
827 osd_data->bvec_pos.iter.bi_size);
828 iov_iter_advance(&i, rc);
829 iov_iter_zero(zlen, &i);
833 put_bvecs(osd_data->bvec_pos.bvecs, osd_data->num_bvecs,
834 aio_req->should_dirty);
835 ceph_osdc_put_request(req);
838 cmpxchg(&aio_req->error, 0, rc);
840 ceph_aio_complete(inode, aio_req);
844 static void ceph_aio_retry_work(struct work_struct *work)
846 struct ceph_aio_work *aio_work =
847 container_of(work, struct ceph_aio_work, work);
848 struct ceph_osd_request *orig_req = aio_work->req;
849 struct ceph_aio_request *aio_req = orig_req->r_priv;
850 struct inode *inode = orig_req->r_inode;
851 struct ceph_inode_info *ci = ceph_inode(inode);
852 struct ceph_snap_context *snapc;
853 struct ceph_osd_request *req;
856 spin_lock(&ci->i_ceph_lock);
857 if (__ceph_have_pending_cap_snap(ci)) {
858 struct ceph_cap_snap *capsnap =
859 list_last_entry(&ci->i_cap_snaps,
860 struct ceph_cap_snap,
862 snapc = ceph_get_snap_context(capsnap->context);
864 BUG_ON(!ci->i_head_snapc);
865 snapc = ceph_get_snap_context(ci->i_head_snapc);
867 spin_unlock(&ci->i_ceph_lock);
869 req = ceph_osdc_alloc_request(orig_req->r_osdc, snapc, 2,
877 req->r_flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE;
878 ceph_oloc_copy(&req->r_base_oloc, &orig_req->r_base_oloc);
879 ceph_oid_copy(&req->r_base_oid, &orig_req->r_base_oid);
881 ret = ceph_osdc_alloc_messages(req, GFP_NOFS);
883 ceph_osdc_put_request(req);
888 req->r_ops[0] = orig_req->r_ops[0];
890 req->r_mtime = aio_req->mtime;
891 req->r_data_offset = req->r_ops[0].extent.offset;
893 ceph_osdc_put_request(orig_req);
895 req->r_callback = ceph_aio_complete_req;
896 req->r_inode = inode;
897 req->r_priv = aio_req;
898 req->r_abort_on_full = true;
900 ret = ceph_osdc_start_request(req->r_osdc, req, false);
904 ceph_aio_complete_req(req);
907 ceph_put_snap_context(snapc);
912 ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter,
913 struct ceph_snap_context *snapc,
914 struct ceph_cap_flush **pcf)
916 struct file *file = iocb->ki_filp;
917 struct inode *inode = file_inode(file);
918 struct ceph_inode_info *ci = ceph_inode(inode);
919 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
920 struct ceph_vino vino;
921 struct ceph_osd_request *req;
922 struct bio_vec *bvecs;
923 struct ceph_aio_request *aio_req = NULL;
927 struct timespec mtime = current_time(inode);
928 size_t count = iov_iter_count(iter);
929 loff_t pos = iocb->ki_pos;
930 bool write = iov_iter_rw(iter) == WRITE;
931 bool should_dirty = !write && iter_is_iovec(iter);
933 if (write && ceph_snap(file_inode(file)) != CEPH_NOSNAP)
936 dout("sync_direct_%s on file %p %lld~%u snapc %p seq %lld\n",
937 (write ? "write" : "read"), file, pos, (unsigned)count,
940 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count);
945 int ret2 = invalidate_inode_pages2_range(inode->i_mapping,
947 (pos + count) >> PAGE_SHIFT);
949 dout("invalidate_inode_pages2_range returned %d\n", ret2);
951 flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE;
953 flags = CEPH_OSD_FLAG_READ;
956 while (iov_iter_count(iter) > 0) {
957 u64 size = iov_iter_count(iter);
961 size = min_t(u64, size, fsc->mount_options->wsize);
963 size = min_t(u64, size, fsc->mount_options->rsize);
965 vino = ceph_vino(inode);
966 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
969 write ? CEPH_OSD_OP_WRITE :
980 len = iter_get_bvecs_alloc(iter, size, &bvecs, &num_pages);
982 ceph_osdc_put_request(req);
987 osd_req_op_extent_update(req, 0, len);
990 * To simplify error handling, allow AIO when IO within i_size
991 * or IO can be satisfied by single OSD request.
993 if (pos == iocb->ki_pos && !is_sync_kiocb(iocb) &&
994 (len == count || pos + count <= i_size_read(inode))) {
995 aio_req = kzalloc(sizeof(*aio_req), GFP_KERNEL);
997 aio_req->iocb = iocb;
998 aio_req->write = write;
999 aio_req->should_dirty = should_dirty;
1000 INIT_LIST_HEAD(&aio_req->osd_reqs);
1002 aio_req->mtime = mtime;
1003 swap(aio_req->prealloc_cf, *pcf);
1011 * throw out any page cache pages in this range. this
1014 truncate_inode_pages_range(inode->i_mapping, pos,
1015 (pos+len) | (PAGE_SIZE - 1));
1017 req->r_mtime = mtime;
1020 osd_req_op_extent_osd_data_bvecs(req, 0, bvecs, num_pages, len);
1023 aio_req->total_len += len;
1024 aio_req->num_reqs++;
1025 atomic_inc(&aio_req->pending_reqs);
1027 req->r_callback = ceph_aio_complete_req;
1028 req->r_inode = inode;
1029 req->r_priv = aio_req;
1030 list_add_tail(&req->r_unsafe_item, &aio_req->osd_reqs);
1036 ret = ceph_osdc_start_request(req->r_osdc, req, false);
1038 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1040 size = i_size_read(inode);
1044 if (ret >= 0 && ret < len && pos + ret < size) {
1046 int zlen = min_t(size_t, len - ret,
1049 iov_iter_bvec(&i, ITER_BVEC, bvecs, num_pages,
1051 iov_iter_advance(&i, ret);
1052 iov_iter_zero(zlen, &i);
1059 put_bvecs(bvecs, num_pages, should_dirty);
1060 ceph_osdc_put_request(req);
1065 if (!write && pos >= size)
1068 if (write && pos > size) {
1069 if (ceph_inode_set_size(inode, pos))
1070 ceph_check_caps(ceph_inode(inode),
1071 CHECK_CAPS_AUTHONLY,
1077 LIST_HEAD(osd_reqs);
1079 if (aio_req->num_reqs == 0) {
1084 ceph_get_cap_refs(ci, write ? CEPH_CAP_FILE_WR :
1087 list_splice(&aio_req->osd_reqs, &osd_reqs);
1088 while (!list_empty(&osd_reqs)) {
1089 req = list_first_entry(&osd_reqs,
1090 struct ceph_osd_request,
1092 list_del_init(&req->r_unsafe_item);
1094 ret = ceph_osdc_start_request(req->r_osdc,
1097 req->r_result = ret;
1098 ceph_aio_complete_req(req);
1101 return -EIOCBQUEUED;
1104 if (ret != -EOLDSNAPC && pos > iocb->ki_pos) {
1105 ret = pos - iocb->ki_pos;
1112 * Synchronous write, straight from __user pointer or user pages.
1114 * If write spans object boundary, just do multiple writes. (For a
1115 * correct atomic write, we should e.g. take write locks on all
1116 * objects, rollback on failure, etc.)
1119 ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
1120 struct ceph_snap_context *snapc)
1122 struct file *file = iocb->ki_filp;
1123 struct inode *inode = file_inode(file);
1124 struct ceph_inode_info *ci = ceph_inode(inode);
1125 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1126 struct ceph_vino vino;
1127 struct ceph_osd_request *req;
1128 struct page **pages;
1134 bool check_caps = false;
1135 struct timespec mtime = current_time(inode);
1136 size_t count = iov_iter_count(from);
1138 if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
1141 dout("sync_write on file %p %lld~%u snapc %p seq %lld\n",
1142 file, pos, (unsigned)count, snapc, snapc->seq);
1144 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count);
1148 ret = invalidate_inode_pages2_range(inode->i_mapping,
1150 (pos + count) >> PAGE_SHIFT);
1152 dout("invalidate_inode_pages2_range returned %d\n", ret);
1154 flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE;
1156 while ((len = iov_iter_count(from)) > 0) {
1160 vino = ceph_vino(inode);
1161 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1162 vino, pos, &len, 0, 1,
1163 CEPH_OSD_OP_WRITE, flags, snapc,
1165 ci->i_truncate_size,
1173 * write from beginning of first page,
1174 * regardless of io alignment
1176 num_pages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1178 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
1179 if (IS_ERR(pages)) {
1180 ret = PTR_ERR(pages);
1185 for (n = 0; n < num_pages; n++) {
1186 size_t plen = min_t(size_t, left, PAGE_SIZE);
1187 ret = copy_page_from_iter(pages[n], 0, plen, from);
1196 ceph_release_page_vector(pages, num_pages);
1200 req->r_inode = inode;
1202 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0,
1205 req->r_mtime = mtime;
1206 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1208 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1211 ceph_osdc_put_request(req);
1213 ceph_set_error_write(ci);
1217 ceph_clear_error_write(ci);
1220 if (pos > i_size_read(inode)) {
1221 check_caps = ceph_inode_set_size(inode, pos);
1223 ceph_check_caps(ceph_inode(inode),
1224 CHECK_CAPS_AUTHONLY,
1230 if (ret != -EOLDSNAPC && written > 0) {
1238 * Wrap generic_file_aio_read with checks for cap bits on the inode.
1239 * Atomically grab references, so that those bits are not released
1240 * back to the MDS mid-read.
1242 * Hmm, the sync read case isn't actually async... should it be?
1244 static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to)
1246 struct file *filp = iocb->ki_filp;
1247 struct ceph_file_info *fi = filp->private_data;
1248 size_t len = iov_iter_count(to);
1249 struct inode *inode = file_inode(filp);
1250 struct ceph_inode_info *ci = ceph_inode(inode);
1251 struct page *pinned_page = NULL;
1254 int retry_op = 0, read = 0;
1257 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
1258 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode);
1260 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1261 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1263 want = CEPH_CAP_FILE_CACHE;
1264 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page);
1268 if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 ||
1269 (iocb->ki_flags & IOCB_DIRECT) ||
1270 (fi->flags & CEPH_F_SYNC)) {
1272 dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n",
1273 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
1274 ceph_cap_string(got));
1276 if (ci->i_inline_version == CEPH_INLINE_NONE) {
1277 if (!retry_op && (iocb->ki_flags & IOCB_DIRECT)) {
1278 ret = ceph_direct_read_write(iocb, to,
1280 if (ret >= 0 && ret < len)
1281 retry_op = CHECK_EOF;
1283 ret = ceph_sync_read(iocb, to, &retry_op);
1286 retry_op = READ_INLINE;
1289 CEPH_DEFINE_RW_CONTEXT(rw_ctx, got);
1290 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
1291 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
1292 ceph_cap_string(got));
1293 ceph_add_rw_context(fi, &rw_ctx);
1294 ret = generic_file_read_iter(iocb, to);
1295 ceph_del_rw_context(fi, &rw_ctx);
1297 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
1298 inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
1300 put_page(pinned_page);
1303 ceph_put_cap_refs(ci, got);
1304 if (retry_op > HAVE_RETRIED && ret >= 0) {
1306 struct page *page = NULL;
1308 if (retry_op == READ_INLINE) {
1309 page = __page_cache_alloc(GFP_KERNEL);
1314 statret = __ceph_do_getattr(inode, page,
1315 CEPH_STAT_CAP_INLINE_DATA, !!page);
1319 if (statret == -ENODATA) {
1320 BUG_ON(retry_op != READ_INLINE);
1326 i_size = i_size_read(inode);
1327 if (retry_op == READ_INLINE) {
1328 BUG_ON(ret > 0 || read > 0);
1329 if (iocb->ki_pos < i_size &&
1330 iocb->ki_pos < PAGE_SIZE) {
1331 loff_t end = min_t(loff_t, i_size,
1332 iocb->ki_pos + len);
1333 end = min_t(loff_t, end, PAGE_SIZE);
1335 zero_user_segment(page, statret, end);
1336 ret = copy_page_to_iter(page,
1337 iocb->ki_pos & ~PAGE_MASK,
1338 end - iocb->ki_pos, to);
1339 iocb->ki_pos += ret;
1342 if (iocb->ki_pos < i_size && read < len) {
1343 size_t zlen = min_t(size_t, len - read,
1344 i_size - iocb->ki_pos);
1345 ret = iov_iter_zero(zlen, to);
1346 iocb->ki_pos += ret;
1349 __free_pages(page, 0);
1353 /* hit EOF or hole? */
1354 if (retry_op == CHECK_EOF && iocb->ki_pos < i_size &&
1356 dout("sync_read hit hole, ppos %lld < size %lld"
1357 ", reading more\n", iocb->ki_pos, i_size);
1361 retry_op = HAVE_RETRIED;
1373 * Take cap references to avoid releasing caps to MDS mid-write.
1375 * If we are synchronous, and write with an old snap context, the OSD
1376 * may return EOLDSNAPC. In that case, retry the write.. _after_
1377 * dropping our cap refs and allowing the pending snap to logically
1378 * complete _before_ this write occurs.
1380 * If we are near ENOSPC, write synchronously.
1382 static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
1384 struct file *file = iocb->ki_filp;
1385 struct ceph_file_info *fi = file->private_data;
1386 struct inode *inode = file_inode(file);
1387 struct ceph_inode_info *ci = ceph_inode(inode);
1388 struct ceph_osd_client *osdc =
1389 &ceph_sb_to_client(inode->i_sb)->client->osdc;
1390 struct ceph_cap_flush *prealloc_cf;
1391 ssize_t count, written = 0;
1395 if (ceph_snap(inode) != CEPH_NOSNAP)
1398 prealloc_cf = ceph_alloc_cap_flush();
1405 /* We can write back this queue in page reclaim */
1406 current->backing_dev_info = inode_to_bdi(inode);
1408 if (iocb->ki_flags & IOCB_APPEND) {
1409 err = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
1414 err = generic_write_checks(iocb, from);
1419 count = iov_iter_count(from);
1420 if (ceph_quota_is_max_bytes_exceeded(inode, pos + count)) {
1425 err = file_remove_privs(file);
1429 err = file_update_time(file);
1433 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1434 err = ceph_uninline_data(file, NULL);
1439 /* FIXME: not complete since it doesn't account for being at quota */
1440 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL)) {
1445 dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
1446 inode, ceph_vinop(inode), pos, count, i_size_read(inode));
1447 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1448 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1450 want = CEPH_CAP_FILE_BUFFER;
1452 err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, pos + count,
1457 dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
1458 inode, ceph_vinop(inode), pos, count, ceph_cap_string(got));
1460 if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 ||
1461 (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC) ||
1462 (ci->i_ceph_flags & CEPH_I_ERROR_WRITE)) {
1463 struct ceph_snap_context *snapc;
1464 struct iov_iter data;
1465 inode_unlock(inode);
1467 spin_lock(&ci->i_ceph_lock);
1468 if (__ceph_have_pending_cap_snap(ci)) {
1469 struct ceph_cap_snap *capsnap =
1470 list_last_entry(&ci->i_cap_snaps,
1471 struct ceph_cap_snap,
1473 snapc = ceph_get_snap_context(capsnap->context);
1475 BUG_ON(!ci->i_head_snapc);
1476 snapc = ceph_get_snap_context(ci->i_head_snapc);
1478 spin_unlock(&ci->i_ceph_lock);
1480 /* we might need to revert back to that point */
1482 if (iocb->ki_flags & IOCB_DIRECT)
1483 written = ceph_direct_read_write(iocb, &data, snapc,
1486 written = ceph_sync_write(iocb, &data, pos, snapc);
1488 iov_iter_advance(from, written);
1489 ceph_put_snap_context(snapc);
1492 * No need to acquire the i_truncate_mutex. Because
1493 * the MDS revokes Fwb caps before sending truncate
1494 * message to us. We can't get Fwb cap while there
1495 * are pending vmtruncate. So write and vmtruncate
1496 * can not run at the same time
1498 written = generic_perform_write(file, from, pos);
1499 if (likely(written >= 0))
1500 iocb->ki_pos = pos + written;
1501 inode_unlock(inode);
1507 spin_lock(&ci->i_ceph_lock);
1508 ci->i_inline_version = CEPH_INLINE_NONE;
1509 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1511 spin_unlock(&ci->i_ceph_lock);
1513 __mark_inode_dirty(inode, dirty);
1514 if (ceph_quota_is_max_bytes_approaching(inode, iocb->ki_pos))
1515 ceph_check_caps(ci, CHECK_CAPS_NODELAY, NULL);
1518 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
1519 inode, ceph_vinop(inode), pos, (unsigned)count,
1520 ceph_cap_string(got));
1521 ceph_put_cap_refs(ci, got);
1523 if (written == -EOLDSNAPC) {
1524 dout("aio_write %p %llx.%llx %llu~%u" "got EOLDSNAPC, retrying\n",
1525 inode, ceph_vinop(inode), pos, (unsigned)count);
1530 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_NEARFULL))
1531 iocb->ki_flags |= IOCB_DSYNC;
1532 written = generic_write_sync(iocb, written);
1538 inode_unlock(inode);
1540 ceph_free_cap_flush(prealloc_cf);
1541 current->backing_dev_info = NULL;
1542 return written ? written : err;
1546 * llseek. be sure to verify file size on SEEK_END.
1548 static loff_t ceph_llseek(struct file *file, loff_t offset, int whence)
1550 struct inode *inode = file->f_mapping->host;
1556 if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) {
1557 ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
1562 i_size = i_size_read(inode);
1569 * Here we special-case the lseek(fd, 0, SEEK_CUR)
1570 * position-querying operation. Avoid rewriting the "same"
1571 * f_pos value back to the file because a concurrent read(),
1572 * write() or lseek() might have altered it
1578 offset += file->f_pos;
1581 if (offset < 0 || offset >= i_size) {
1587 if (offset < 0 || offset >= i_size) {
1595 ret = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1598 inode_unlock(inode);
1602 static inline void ceph_zero_partial_page(
1603 struct inode *inode, loff_t offset, unsigned size)
1606 pgoff_t index = offset >> PAGE_SHIFT;
1608 page = find_lock_page(inode->i_mapping, index);
1610 wait_on_page_writeback(page);
1611 zero_user(page, offset & (PAGE_SIZE - 1), size);
1617 static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset,
1620 loff_t nearly = round_up(offset, PAGE_SIZE);
1621 if (offset < nearly) {
1622 loff_t size = nearly - offset;
1625 ceph_zero_partial_page(inode, offset, size);
1629 if (length >= PAGE_SIZE) {
1630 loff_t size = round_down(length, PAGE_SIZE);
1631 truncate_pagecache_range(inode, offset, offset + size - 1);
1636 ceph_zero_partial_page(inode, offset, length);
1639 static int ceph_zero_partial_object(struct inode *inode,
1640 loff_t offset, loff_t *length)
1642 struct ceph_inode_info *ci = ceph_inode(inode);
1643 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1644 struct ceph_osd_request *req;
1650 op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE;
1653 op = CEPH_OSD_OP_ZERO;
1656 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1660 CEPH_OSD_FLAG_WRITE,
1667 req->r_mtime = inode->i_mtime;
1668 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1670 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1674 ceph_osdc_put_request(req);
1680 static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length)
1683 struct ceph_inode_info *ci = ceph_inode(inode);
1684 s32 stripe_unit = ci->i_layout.stripe_unit;
1685 s32 stripe_count = ci->i_layout.stripe_count;
1686 s32 object_size = ci->i_layout.object_size;
1687 u64 object_set_size = object_size * stripe_count;
1690 /* round offset up to next period boundary */
1691 nearly = offset + object_set_size - 1;
1693 nearly -= do_div(t, object_set_size);
1695 while (length && offset < nearly) {
1696 loff_t size = length;
1697 ret = ceph_zero_partial_object(inode, offset, &size);
1703 while (length >= object_set_size) {
1705 loff_t pos = offset;
1706 for (i = 0; i < stripe_count; ++i) {
1707 ret = ceph_zero_partial_object(inode, pos, NULL);
1712 offset += object_set_size;
1713 length -= object_set_size;
1716 loff_t size = length;
1717 ret = ceph_zero_partial_object(inode, offset, &size);
1726 static long ceph_fallocate(struct file *file, int mode,
1727 loff_t offset, loff_t length)
1729 struct ceph_file_info *fi = file->private_data;
1730 struct inode *inode = file_inode(file);
1731 struct ceph_inode_info *ci = ceph_inode(inode);
1732 struct ceph_osd_client *osdc =
1733 &ceph_inode_to_client(inode)->client->osdc;
1734 struct ceph_cap_flush *prealloc_cf;
1741 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1744 if (!S_ISREG(inode->i_mode))
1747 prealloc_cf = ceph_alloc_cap_flush();
1753 if (ceph_snap(inode) != CEPH_NOSNAP) {
1758 if (!(mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)) &&
1759 ceph_quota_is_max_bytes_exceeded(inode, offset + length)) {
1764 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) &&
1765 !(mode & FALLOC_FL_PUNCH_HOLE)) {
1770 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1771 ret = ceph_uninline_data(file, NULL);
1776 size = i_size_read(inode);
1777 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
1778 endoff = offset + length;
1779 ret = inode_newsize_ok(inode, endoff);
1784 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1785 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1787 want = CEPH_CAP_FILE_BUFFER;
1789 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, endoff, &got, NULL);
1793 if (mode & FALLOC_FL_PUNCH_HOLE) {
1795 ceph_zero_pagecache_range(inode, offset, length);
1796 ret = ceph_zero_objects(inode, offset, length);
1797 } else if (endoff > size) {
1798 truncate_pagecache_range(inode, size, -1);
1799 if (ceph_inode_set_size(inode, endoff))
1800 ceph_check_caps(ceph_inode(inode),
1801 CHECK_CAPS_AUTHONLY, NULL);
1805 spin_lock(&ci->i_ceph_lock);
1806 ci->i_inline_version = CEPH_INLINE_NONE;
1807 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1809 spin_unlock(&ci->i_ceph_lock);
1811 __mark_inode_dirty(inode, dirty);
1812 if ((endoff > size) &&
1813 ceph_quota_is_max_bytes_approaching(inode, endoff))
1814 ceph_check_caps(ci, CHECK_CAPS_NODELAY, NULL);
1817 ceph_put_cap_refs(ci, got);
1819 inode_unlock(inode);
1820 ceph_free_cap_flush(prealloc_cf);
1824 const struct file_operations ceph_file_fops = {
1826 .release = ceph_release,
1827 .llseek = ceph_llseek,
1828 .read_iter = ceph_read_iter,
1829 .write_iter = ceph_write_iter,
1831 .fsync = ceph_fsync,
1833 .flock = ceph_flock,
1834 .splice_read = generic_file_splice_read,
1835 .splice_write = iter_file_splice_write,
1836 .unlocked_ioctl = ceph_ioctl,
1837 .compat_ioctl = ceph_ioctl,
1838 .fallocate = ceph_fallocate,