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)
434 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
435 struct ceph_mds_client *mdsc = fsc->mdsc;
436 struct ceph_mds_request *req;
438 struct ceph_acls_info acls = {};
442 dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n",
444 d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode);
446 if (dentry->d_name.len > NAME_MAX)
447 return -ENAMETOOLONG;
449 if (flags & O_CREAT) {
450 if (ceph_quota_is_max_files_exceeded(dir))
452 err = ceph_pre_init_acls(dir, &mode, &acls);
458 req = prepare_open_request(dir->i_sb, flags, mode);
463 req->r_dentry = dget(dentry);
465 if (flags & O_CREAT) {
466 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
467 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
469 req->r_pagelist = acls.pagelist;
470 acls.pagelist = NULL;
474 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
475 if (ceph_security_xattr_wanted(dir))
476 mask |= CEPH_CAP_XATTR_SHARED;
477 req->r_args.open.mask = cpu_to_le32(mask);
480 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
481 err = ceph_mdsc_do_request(mdsc,
482 (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
484 err = ceph_handle_snapdir(req, dentry, err);
488 if ((flags & O_CREAT) && !req->r_reply_info.head->is_dentry)
489 err = ceph_handle_notrace_create(dir, dentry);
491 if (d_in_lookup(dentry)) {
492 dn = ceph_finish_lookup(req, dentry, err);
496 /* we were given a hashed negative dentry */
501 if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) {
502 /* make vfs retry on splice, ENOENT, or symlink */
503 dout("atomic_open finish_no_open on dn %p\n", dn);
504 err = finish_no_open(file, dn);
506 dout("atomic_open finish_open on dn %p\n", dn);
507 if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) {
508 ceph_init_inode_acls(d_inode(dentry), &acls);
509 file->f_mode |= FMODE_CREATED;
511 err = finish_open(file, dentry, ceph_open);
514 if (!req->r_err && req->r_target_inode)
515 ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode);
516 ceph_mdsc_put_request(req);
518 ceph_release_acls_info(&acls);
519 dout("atomic_open result=%d\n", err);
523 int ceph_release(struct inode *inode, struct file *file)
525 struct ceph_inode_info *ci = ceph_inode(inode);
527 if (S_ISDIR(inode->i_mode)) {
528 struct ceph_dir_file_info *dfi = file->private_data;
529 dout("release inode %p dir file %p\n", inode, file);
530 WARN_ON(!list_empty(&dfi->file_info.rw_contexts));
532 ceph_put_fmode(ci, dfi->file_info.fmode);
534 if (dfi->last_readdir)
535 ceph_mdsc_put_request(dfi->last_readdir);
536 kfree(dfi->last_name);
537 kfree(dfi->dir_info);
538 kmem_cache_free(ceph_dir_file_cachep, dfi);
540 struct ceph_file_info *fi = file->private_data;
541 dout("release inode %p regular file %p\n", inode, file);
542 WARN_ON(!list_empty(&fi->rw_contexts));
544 ceph_put_fmode(ci, fi->fmode);
545 kmem_cache_free(ceph_file_cachep, fi);
548 /* wake up anyone waiting for caps on this inode */
549 wake_up_all(&ci->i_cap_wq);
560 * Read a range of bytes striped over one or more objects. Iterate over
561 * objects we stripe over. (That's not atomic, but good enough for now.)
563 * If we get a short result from the OSD, check against i_size; we need to
564 * only return a short read to the caller if we hit EOF.
566 static int striped_read(struct inode *inode,
568 struct page **pages, int num_pages,
569 int page_align, int *checkeof)
571 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
572 struct ceph_inode_info *ci = ceph_inode(inode);
577 bool hit_stripe, was_short;
580 * we may need to do multiple reads. not atomic, unfortunately.
584 page_idx = (page_align + read) >> PAGE_SHIFT;
585 ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode),
586 &ci->i_layout, pos, &this_len,
587 ci->i_truncate_seq, ci->i_truncate_size,
588 pages + page_idx, num_pages - page_idx,
589 ((page_align + read) & ~PAGE_MASK));
592 hit_stripe = this_len < len;
593 was_short = ret >= 0 && ret < this_len;
594 dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, len, read,
595 ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : "");
597 i_size = i_size_read(inode);
599 if (was_short && (pos + ret < i_size)) {
600 int zlen = min(this_len - ret, i_size - pos - ret);
601 int zoff = page_align + read + ret;
602 dout(" zero gap %llu to %llu\n",
603 pos + ret, pos + ret + zlen);
604 ceph_zero_page_vector_range(zoff, zlen, pages);
612 /* hit stripe and need continue*/
613 if (len && hit_stripe && pos < i_size)
619 /* did we bounce off eof? */
620 if (pos + len > i_size)
621 *checkeof = CHECK_EOF;
624 dout("striped_read returns %d\n", ret);
629 * Completely synchronous read and write methods. Direct from __user
630 * buffer to osd, or directly to user pages (if O_DIRECT).
632 * If the read spans object boundary, just do multiple reads.
634 static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *to,
637 struct file *file = iocb->ki_filp;
638 struct inode *inode = file_inode(file);
640 u64 off = iocb->ki_pos;
643 size_t len = iov_iter_count(to);
645 dout("sync_read on file %p %llu~%u %s\n", file, off, (unsigned)len,
646 (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
651 * flush any page cache pages in this range. this
652 * will make concurrent normal and sync io slow,
653 * but it will at least behave sensibly when they are
656 ret = filemap_write_and_wait_range(inode->i_mapping, off,
661 if (unlikely(iov_iter_is_pipe(to))) {
663 ret = iov_iter_get_pages_alloc(to, &pages, len,
667 num_pages = DIV_ROUND_UP(ret + page_off, PAGE_SIZE);
669 ret = striped_read(inode, off, ret, pages, num_pages,
672 iov_iter_advance(to, ret);
675 iov_iter_advance(to, 0);
677 ceph_put_page_vector(pages, num_pages, false);
679 num_pages = calc_pages_for(off, len);
680 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
682 return PTR_ERR(pages);
684 ret = striped_read(inode, off, len, pages, num_pages,
685 (off & ~PAGE_MASK), checkeof);
691 size_t page_off = off & ~PAGE_MASK;
692 size_t copy = min_t(size_t, left,
693 PAGE_SIZE - page_off);
694 l = copy_page_to_iter(pages[k++], page_off,
702 ceph_release_page_vector(pages, num_pages);
705 if (off > iocb->ki_pos) {
706 ret = off - iocb->ki_pos;
710 dout("sync_read result %zd\n", ret);
714 struct ceph_aio_request {
720 struct list_head osd_reqs;
722 atomic_t pending_reqs;
723 struct timespec64 mtime;
724 struct ceph_cap_flush *prealloc_cf;
727 struct ceph_aio_work {
728 struct work_struct work;
729 struct ceph_osd_request *req;
732 static void ceph_aio_retry_work(struct work_struct *work);
734 static void ceph_aio_complete(struct inode *inode,
735 struct ceph_aio_request *aio_req)
737 struct ceph_inode_info *ci = ceph_inode(inode);
740 if (!atomic_dec_and_test(&aio_req->pending_reqs))
743 ret = aio_req->error;
745 ret = aio_req->total_len;
747 dout("ceph_aio_complete %p rc %d\n", inode, ret);
749 if (ret >= 0 && aio_req->write) {
752 loff_t endoff = aio_req->iocb->ki_pos + aio_req->total_len;
753 if (endoff > i_size_read(inode)) {
754 if (ceph_inode_set_size(inode, endoff))
755 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
758 spin_lock(&ci->i_ceph_lock);
759 ci->i_inline_version = CEPH_INLINE_NONE;
760 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
761 &aio_req->prealloc_cf);
762 spin_unlock(&ci->i_ceph_lock);
764 __mark_inode_dirty(inode, dirty);
768 ceph_put_cap_refs(ci, (aio_req->write ? CEPH_CAP_FILE_WR :
771 aio_req->iocb->ki_complete(aio_req->iocb, ret, 0);
773 ceph_free_cap_flush(aio_req->prealloc_cf);
777 static void ceph_aio_complete_req(struct ceph_osd_request *req)
779 int rc = req->r_result;
780 struct inode *inode = req->r_inode;
781 struct ceph_aio_request *aio_req = req->r_priv;
782 struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0);
784 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_BVECS);
785 BUG_ON(!osd_data->num_bvecs);
787 dout("ceph_aio_complete_req %p rc %d bytes %u\n",
788 inode, rc, osd_data->bvec_pos.iter.bi_size);
790 if (rc == -EOLDSNAPC) {
791 struct ceph_aio_work *aio_work;
792 BUG_ON(!aio_req->write);
794 aio_work = kmalloc(sizeof(*aio_work), GFP_NOFS);
796 INIT_WORK(&aio_work->work, ceph_aio_retry_work);
798 queue_work(ceph_inode_to_client(inode)->wb_wq,
803 } else if (!aio_req->write) {
806 if (rc >= 0 && osd_data->bvec_pos.iter.bi_size > rc) {
808 int zlen = osd_data->bvec_pos.iter.bi_size - rc;
811 * If read is satisfied by single OSD request,
812 * it can pass EOF. Otherwise read is within
815 if (aio_req->num_reqs == 1) {
816 loff_t i_size = i_size_read(inode);
817 loff_t endoff = aio_req->iocb->ki_pos + rc;
819 zlen = min_t(size_t, zlen,
821 aio_req->total_len = rc + zlen;
824 iov_iter_bvec(&i, READ, osd_data->bvec_pos.bvecs,
826 osd_data->bvec_pos.iter.bi_size);
827 iov_iter_advance(&i, rc);
828 iov_iter_zero(zlen, &i);
832 put_bvecs(osd_data->bvec_pos.bvecs, osd_data->num_bvecs,
833 aio_req->should_dirty);
834 ceph_osdc_put_request(req);
837 cmpxchg(&aio_req->error, 0, rc);
839 ceph_aio_complete(inode, aio_req);
843 static void ceph_aio_retry_work(struct work_struct *work)
845 struct ceph_aio_work *aio_work =
846 container_of(work, struct ceph_aio_work, work);
847 struct ceph_osd_request *orig_req = aio_work->req;
848 struct ceph_aio_request *aio_req = orig_req->r_priv;
849 struct inode *inode = orig_req->r_inode;
850 struct ceph_inode_info *ci = ceph_inode(inode);
851 struct ceph_snap_context *snapc;
852 struct ceph_osd_request *req;
855 spin_lock(&ci->i_ceph_lock);
856 if (__ceph_have_pending_cap_snap(ci)) {
857 struct ceph_cap_snap *capsnap =
858 list_last_entry(&ci->i_cap_snaps,
859 struct ceph_cap_snap,
861 snapc = ceph_get_snap_context(capsnap->context);
863 BUG_ON(!ci->i_head_snapc);
864 snapc = ceph_get_snap_context(ci->i_head_snapc);
866 spin_unlock(&ci->i_ceph_lock);
868 req = ceph_osdc_alloc_request(orig_req->r_osdc, snapc, 2,
876 req->r_flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE;
877 ceph_oloc_copy(&req->r_base_oloc, &orig_req->r_base_oloc);
878 ceph_oid_copy(&req->r_base_oid, &orig_req->r_base_oid);
880 ret = ceph_osdc_alloc_messages(req, GFP_NOFS);
882 ceph_osdc_put_request(req);
887 req->r_ops[0] = orig_req->r_ops[0];
889 req->r_mtime = aio_req->mtime;
890 req->r_data_offset = req->r_ops[0].extent.offset;
892 ceph_osdc_put_request(orig_req);
894 req->r_callback = ceph_aio_complete_req;
895 req->r_inode = inode;
896 req->r_priv = aio_req;
898 ret = ceph_osdc_start_request(req->r_osdc, req, false);
902 ceph_aio_complete_req(req);
905 ceph_put_snap_context(snapc);
910 ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter,
911 struct ceph_snap_context *snapc,
912 struct ceph_cap_flush **pcf)
914 struct file *file = iocb->ki_filp;
915 struct inode *inode = file_inode(file);
916 struct ceph_inode_info *ci = ceph_inode(inode);
917 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
918 struct ceph_vino vino;
919 struct ceph_osd_request *req;
920 struct bio_vec *bvecs;
921 struct ceph_aio_request *aio_req = NULL;
925 struct timespec64 mtime = current_time(inode);
926 size_t count = iov_iter_count(iter);
927 loff_t pos = iocb->ki_pos;
928 bool write = iov_iter_rw(iter) == WRITE;
929 bool should_dirty = !write && iter_is_iovec(iter);
931 if (write && ceph_snap(file_inode(file)) != CEPH_NOSNAP)
934 dout("sync_direct_%s on file %p %lld~%u snapc %p seq %lld\n",
935 (write ? "write" : "read"), file, pos, (unsigned)count,
938 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count);
943 int ret2 = invalidate_inode_pages2_range(inode->i_mapping,
945 (pos + count) >> PAGE_SHIFT);
947 dout("invalidate_inode_pages2_range returned %d\n", ret2);
949 flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE;
951 flags = CEPH_OSD_FLAG_READ;
954 while (iov_iter_count(iter) > 0) {
955 u64 size = iov_iter_count(iter);
959 size = min_t(u64, size, fsc->mount_options->wsize);
961 size = min_t(u64, size, fsc->mount_options->rsize);
963 vino = ceph_vino(inode);
964 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
967 write ? CEPH_OSD_OP_WRITE :
978 len = iter_get_bvecs_alloc(iter, size, &bvecs, &num_pages);
980 ceph_osdc_put_request(req);
985 osd_req_op_extent_update(req, 0, len);
988 * To simplify error handling, allow AIO when IO within i_size
989 * or IO can be satisfied by single OSD request.
991 if (pos == iocb->ki_pos && !is_sync_kiocb(iocb) &&
992 (len == count || pos + count <= i_size_read(inode))) {
993 aio_req = kzalloc(sizeof(*aio_req), GFP_KERNEL);
995 aio_req->iocb = iocb;
996 aio_req->write = write;
997 aio_req->should_dirty = should_dirty;
998 INIT_LIST_HEAD(&aio_req->osd_reqs);
1000 aio_req->mtime = mtime;
1001 swap(aio_req->prealloc_cf, *pcf);
1009 * throw out any page cache pages in this range. this
1012 truncate_inode_pages_range(inode->i_mapping, pos,
1013 (pos+len) | (PAGE_SIZE - 1));
1015 req->r_mtime = mtime;
1018 osd_req_op_extent_osd_data_bvecs(req, 0, bvecs, num_pages, len);
1021 aio_req->total_len += len;
1022 aio_req->num_reqs++;
1023 atomic_inc(&aio_req->pending_reqs);
1025 req->r_callback = ceph_aio_complete_req;
1026 req->r_inode = inode;
1027 req->r_priv = aio_req;
1028 list_add_tail(&req->r_unsafe_item, &aio_req->osd_reqs);
1034 ret = ceph_osdc_start_request(req->r_osdc, req, false);
1036 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1038 size = i_size_read(inode);
1042 if (ret >= 0 && ret < len && pos + ret < size) {
1044 int zlen = min_t(size_t, len - ret,
1047 iov_iter_bvec(&i, READ, bvecs, num_pages, len);
1048 iov_iter_advance(&i, ret);
1049 iov_iter_zero(zlen, &i);
1056 put_bvecs(bvecs, num_pages, should_dirty);
1057 ceph_osdc_put_request(req);
1062 if (!write && pos >= size)
1065 if (write && pos > size) {
1066 if (ceph_inode_set_size(inode, pos))
1067 ceph_check_caps(ceph_inode(inode),
1068 CHECK_CAPS_AUTHONLY,
1074 LIST_HEAD(osd_reqs);
1076 if (aio_req->num_reqs == 0) {
1081 ceph_get_cap_refs(ci, write ? CEPH_CAP_FILE_WR :
1084 list_splice(&aio_req->osd_reqs, &osd_reqs);
1085 while (!list_empty(&osd_reqs)) {
1086 req = list_first_entry(&osd_reqs,
1087 struct ceph_osd_request,
1089 list_del_init(&req->r_unsafe_item);
1091 ret = ceph_osdc_start_request(req->r_osdc,
1094 req->r_result = ret;
1095 ceph_aio_complete_req(req);
1098 return -EIOCBQUEUED;
1101 if (ret != -EOLDSNAPC && pos > iocb->ki_pos) {
1102 ret = pos - iocb->ki_pos;
1109 * Synchronous write, straight from __user pointer or user pages.
1111 * If write spans object boundary, just do multiple writes. (For a
1112 * correct atomic write, we should e.g. take write locks on all
1113 * objects, rollback on failure, etc.)
1116 ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
1117 struct ceph_snap_context *snapc)
1119 struct file *file = iocb->ki_filp;
1120 struct inode *inode = file_inode(file);
1121 struct ceph_inode_info *ci = ceph_inode(inode);
1122 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1123 struct ceph_vino vino;
1124 struct ceph_osd_request *req;
1125 struct page **pages;
1131 bool check_caps = false;
1132 struct timespec64 mtime = current_time(inode);
1133 size_t count = iov_iter_count(from);
1135 if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
1138 dout("sync_write on file %p %lld~%u snapc %p seq %lld\n",
1139 file, pos, (unsigned)count, snapc, snapc->seq);
1141 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count);
1145 ret = invalidate_inode_pages2_range(inode->i_mapping,
1147 (pos + count) >> PAGE_SHIFT);
1149 dout("invalidate_inode_pages2_range returned %d\n", ret);
1151 flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE;
1153 while ((len = iov_iter_count(from)) > 0) {
1157 vino = ceph_vino(inode);
1158 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1159 vino, pos, &len, 0, 1,
1160 CEPH_OSD_OP_WRITE, flags, snapc,
1162 ci->i_truncate_size,
1170 * write from beginning of first page,
1171 * regardless of io alignment
1173 num_pages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1175 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
1176 if (IS_ERR(pages)) {
1177 ret = PTR_ERR(pages);
1182 for (n = 0; n < num_pages; n++) {
1183 size_t plen = min_t(size_t, left, PAGE_SIZE);
1184 ret = copy_page_from_iter(pages[n], 0, plen, from);
1193 ceph_release_page_vector(pages, num_pages);
1197 req->r_inode = inode;
1199 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0,
1202 req->r_mtime = mtime;
1203 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1205 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1208 ceph_osdc_put_request(req);
1210 ceph_set_error_write(ci);
1214 ceph_clear_error_write(ci);
1217 if (pos > i_size_read(inode)) {
1218 check_caps = ceph_inode_set_size(inode, pos);
1220 ceph_check_caps(ceph_inode(inode),
1221 CHECK_CAPS_AUTHONLY,
1227 if (ret != -EOLDSNAPC && written > 0) {
1235 * Wrap generic_file_aio_read with checks for cap bits on the inode.
1236 * Atomically grab references, so that those bits are not released
1237 * back to the MDS mid-read.
1239 * Hmm, the sync read case isn't actually async... should it be?
1241 static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to)
1243 struct file *filp = iocb->ki_filp;
1244 struct ceph_file_info *fi = filp->private_data;
1245 size_t len = iov_iter_count(to);
1246 struct inode *inode = file_inode(filp);
1247 struct ceph_inode_info *ci = ceph_inode(inode);
1248 struct page *pinned_page = NULL;
1251 int retry_op = 0, read = 0;
1254 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
1255 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode);
1257 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1258 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1260 want = CEPH_CAP_FILE_CACHE;
1261 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page);
1265 if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 ||
1266 (iocb->ki_flags & IOCB_DIRECT) ||
1267 (fi->flags & CEPH_F_SYNC)) {
1269 dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n",
1270 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
1271 ceph_cap_string(got));
1273 if (ci->i_inline_version == CEPH_INLINE_NONE) {
1274 if (!retry_op && (iocb->ki_flags & IOCB_DIRECT)) {
1275 ret = ceph_direct_read_write(iocb, to,
1277 if (ret >= 0 && ret < len)
1278 retry_op = CHECK_EOF;
1280 ret = ceph_sync_read(iocb, to, &retry_op);
1283 retry_op = READ_INLINE;
1286 CEPH_DEFINE_RW_CONTEXT(rw_ctx, got);
1287 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
1288 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
1289 ceph_cap_string(got));
1290 ceph_add_rw_context(fi, &rw_ctx);
1291 ret = generic_file_read_iter(iocb, to);
1292 ceph_del_rw_context(fi, &rw_ctx);
1294 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
1295 inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
1297 put_page(pinned_page);
1300 ceph_put_cap_refs(ci, got);
1301 if (retry_op > HAVE_RETRIED && ret >= 0) {
1303 struct page *page = NULL;
1305 if (retry_op == READ_INLINE) {
1306 page = __page_cache_alloc(GFP_KERNEL);
1311 statret = __ceph_do_getattr(inode, page,
1312 CEPH_STAT_CAP_INLINE_DATA, !!page);
1316 if (statret == -ENODATA) {
1317 BUG_ON(retry_op != READ_INLINE);
1323 i_size = i_size_read(inode);
1324 if (retry_op == READ_INLINE) {
1325 BUG_ON(ret > 0 || read > 0);
1326 if (iocb->ki_pos < i_size &&
1327 iocb->ki_pos < PAGE_SIZE) {
1328 loff_t end = min_t(loff_t, i_size,
1329 iocb->ki_pos + len);
1330 end = min_t(loff_t, end, PAGE_SIZE);
1332 zero_user_segment(page, statret, end);
1333 ret = copy_page_to_iter(page,
1334 iocb->ki_pos & ~PAGE_MASK,
1335 end - iocb->ki_pos, to);
1336 iocb->ki_pos += ret;
1339 if (iocb->ki_pos < i_size && read < len) {
1340 size_t zlen = min_t(size_t, len - read,
1341 i_size - iocb->ki_pos);
1342 ret = iov_iter_zero(zlen, to);
1343 iocb->ki_pos += ret;
1346 __free_pages(page, 0);
1350 /* hit EOF or hole? */
1351 if (retry_op == CHECK_EOF && iocb->ki_pos < i_size &&
1353 dout("sync_read hit hole, ppos %lld < size %lld"
1354 ", reading more\n", iocb->ki_pos, i_size);
1358 retry_op = HAVE_RETRIED;
1370 * Take cap references to avoid releasing caps to MDS mid-write.
1372 * If we are synchronous, and write with an old snap context, the OSD
1373 * may return EOLDSNAPC. In that case, retry the write.. _after_
1374 * dropping our cap refs and allowing the pending snap to logically
1375 * complete _before_ this write occurs.
1377 * If we are near ENOSPC, write synchronously.
1379 static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
1381 struct file *file = iocb->ki_filp;
1382 struct ceph_file_info *fi = file->private_data;
1383 struct inode *inode = file_inode(file);
1384 struct ceph_inode_info *ci = ceph_inode(inode);
1385 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1386 struct ceph_cap_flush *prealloc_cf;
1387 ssize_t count, written = 0;
1390 loff_t limit = max(i_size_read(inode), fsc->max_file_size);
1392 if (ceph_snap(inode) != CEPH_NOSNAP)
1395 prealloc_cf = ceph_alloc_cap_flush();
1402 /* We can write back this queue in page reclaim */
1403 current->backing_dev_info = inode_to_bdi(inode);
1405 if (iocb->ki_flags & IOCB_APPEND) {
1406 err = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
1411 err = generic_write_checks(iocb, from);
1416 if (unlikely(pos >= limit)) {
1420 iov_iter_truncate(from, limit - pos);
1423 count = iov_iter_count(from);
1424 if (ceph_quota_is_max_bytes_exceeded(inode, pos + count)) {
1429 err = file_remove_privs(file);
1433 err = file_update_time(file);
1437 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1438 err = ceph_uninline_data(file, NULL);
1443 /* FIXME: not complete since it doesn't account for being at quota */
1444 if (ceph_osdmap_flag(&fsc->client->osdc, CEPH_OSDMAP_FULL)) {
1449 dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
1450 inode, ceph_vinop(inode), pos, count, i_size_read(inode));
1451 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1452 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1454 want = CEPH_CAP_FILE_BUFFER;
1456 err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, pos + count,
1461 dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
1462 inode, ceph_vinop(inode), pos, count, ceph_cap_string(got));
1464 if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 ||
1465 (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC) ||
1466 (ci->i_ceph_flags & CEPH_I_ERROR_WRITE)) {
1467 struct ceph_snap_context *snapc;
1468 struct iov_iter data;
1469 inode_unlock(inode);
1471 spin_lock(&ci->i_ceph_lock);
1472 if (__ceph_have_pending_cap_snap(ci)) {
1473 struct ceph_cap_snap *capsnap =
1474 list_last_entry(&ci->i_cap_snaps,
1475 struct ceph_cap_snap,
1477 snapc = ceph_get_snap_context(capsnap->context);
1479 BUG_ON(!ci->i_head_snapc);
1480 snapc = ceph_get_snap_context(ci->i_head_snapc);
1482 spin_unlock(&ci->i_ceph_lock);
1484 /* we might need to revert back to that point */
1486 if (iocb->ki_flags & IOCB_DIRECT)
1487 written = ceph_direct_read_write(iocb, &data, snapc,
1490 written = ceph_sync_write(iocb, &data, pos, snapc);
1492 iov_iter_advance(from, written);
1493 ceph_put_snap_context(snapc);
1496 * No need to acquire the i_truncate_mutex. Because
1497 * the MDS revokes Fwb caps before sending truncate
1498 * message to us. We can't get Fwb cap while there
1499 * are pending vmtruncate. So write and vmtruncate
1500 * can not run at the same time
1502 written = generic_perform_write(file, from, pos);
1503 if (likely(written >= 0))
1504 iocb->ki_pos = pos + written;
1505 inode_unlock(inode);
1511 spin_lock(&ci->i_ceph_lock);
1512 ci->i_inline_version = CEPH_INLINE_NONE;
1513 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1515 spin_unlock(&ci->i_ceph_lock);
1517 __mark_inode_dirty(inode, dirty);
1518 if (ceph_quota_is_max_bytes_approaching(inode, iocb->ki_pos))
1519 ceph_check_caps(ci, CHECK_CAPS_NODELAY, NULL);
1522 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
1523 inode, ceph_vinop(inode), pos, (unsigned)count,
1524 ceph_cap_string(got));
1525 ceph_put_cap_refs(ci, got);
1527 if (written == -EOLDSNAPC) {
1528 dout("aio_write %p %llx.%llx %llu~%u" "got EOLDSNAPC, retrying\n",
1529 inode, ceph_vinop(inode), pos, (unsigned)count);
1534 if (ceph_osdmap_flag(&fsc->client->osdc, CEPH_OSDMAP_NEARFULL))
1535 iocb->ki_flags |= IOCB_DSYNC;
1536 written = generic_write_sync(iocb, written);
1542 inode_unlock(inode);
1544 ceph_free_cap_flush(prealloc_cf);
1545 current->backing_dev_info = NULL;
1546 return written ? written : err;
1550 * llseek. be sure to verify file size on SEEK_END.
1552 static loff_t ceph_llseek(struct file *file, loff_t offset, int whence)
1554 struct inode *inode = file->f_mapping->host;
1555 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1561 if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) {
1562 ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
1567 i_size = i_size_read(inode);
1574 * Here we special-case the lseek(fd, 0, SEEK_CUR)
1575 * position-querying operation. Avoid rewriting the "same"
1576 * f_pos value back to the file because a concurrent read(),
1577 * write() or lseek() might have altered it
1583 offset += file->f_pos;
1586 if (offset < 0 || offset >= i_size) {
1592 if (offset < 0 || offset >= i_size) {
1600 ret = vfs_setpos(file, offset, max(i_size, fsc->max_file_size));
1603 inode_unlock(inode);
1607 static inline void ceph_zero_partial_page(
1608 struct inode *inode, loff_t offset, unsigned size)
1611 pgoff_t index = offset >> PAGE_SHIFT;
1613 page = find_lock_page(inode->i_mapping, index);
1615 wait_on_page_writeback(page);
1616 zero_user(page, offset & (PAGE_SIZE - 1), size);
1622 static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset,
1625 loff_t nearly = round_up(offset, PAGE_SIZE);
1626 if (offset < nearly) {
1627 loff_t size = nearly - offset;
1630 ceph_zero_partial_page(inode, offset, size);
1634 if (length >= PAGE_SIZE) {
1635 loff_t size = round_down(length, PAGE_SIZE);
1636 truncate_pagecache_range(inode, offset, offset + size - 1);
1641 ceph_zero_partial_page(inode, offset, length);
1644 static int ceph_zero_partial_object(struct inode *inode,
1645 loff_t offset, loff_t *length)
1647 struct ceph_inode_info *ci = ceph_inode(inode);
1648 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1649 struct ceph_osd_request *req;
1655 op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE;
1658 op = CEPH_OSD_OP_ZERO;
1661 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1665 CEPH_OSD_FLAG_WRITE,
1672 req->r_mtime = inode->i_mtime;
1673 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1675 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1679 ceph_osdc_put_request(req);
1685 static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length)
1688 struct ceph_inode_info *ci = ceph_inode(inode);
1689 s32 stripe_unit = ci->i_layout.stripe_unit;
1690 s32 stripe_count = ci->i_layout.stripe_count;
1691 s32 object_size = ci->i_layout.object_size;
1692 u64 object_set_size = object_size * stripe_count;
1695 /* round offset up to next period boundary */
1696 nearly = offset + object_set_size - 1;
1698 nearly -= do_div(t, object_set_size);
1700 while (length && offset < nearly) {
1701 loff_t size = length;
1702 ret = ceph_zero_partial_object(inode, offset, &size);
1708 while (length >= object_set_size) {
1710 loff_t pos = offset;
1711 for (i = 0; i < stripe_count; ++i) {
1712 ret = ceph_zero_partial_object(inode, pos, NULL);
1717 offset += object_set_size;
1718 length -= object_set_size;
1721 loff_t size = length;
1722 ret = ceph_zero_partial_object(inode, offset, &size);
1731 static long ceph_fallocate(struct file *file, int mode,
1732 loff_t offset, loff_t length)
1734 struct ceph_file_info *fi = file->private_data;
1735 struct inode *inode = file_inode(file);
1736 struct ceph_inode_info *ci = ceph_inode(inode);
1737 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1738 struct ceph_cap_flush *prealloc_cf;
1745 if ((offset + length) > max(i_size_read(inode), fsc->max_file_size))
1748 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1751 if (!S_ISREG(inode->i_mode))
1754 prealloc_cf = ceph_alloc_cap_flush();
1760 if (ceph_snap(inode) != CEPH_NOSNAP) {
1765 if (!(mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)) &&
1766 ceph_quota_is_max_bytes_exceeded(inode, offset + length)) {
1771 if (ceph_osdmap_flag(&fsc->client->osdc, CEPH_OSDMAP_FULL) &&
1772 !(mode & FALLOC_FL_PUNCH_HOLE)) {
1777 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1778 ret = ceph_uninline_data(file, NULL);
1783 size = i_size_read(inode);
1784 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
1785 endoff = offset + length;
1786 ret = inode_newsize_ok(inode, endoff);
1791 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1792 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1794 want = CEPH_CAP_FILE_BUFFER;
1796 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, endoff, &got, NULL);
1800 if (mode & FALLOC_FL_PUNCH_HOLE) {
1802 ceph_zero_pagecache_range(inode, offset, length);
1803 ret = ceph_zero_objects(inode, offset, length);
1804 } else if (endoff > size) {
1805 truncate_pagecache_range(inode, size, -1);
1806 if (ceph_inode_set_size(inode, endoff))
1807 ceph_check_caps(ceph_inode(inode),
1808 CHECK_CAPS_AUTHONLY, NULL);
1812 spin_lock(&ci->i_ceph_lock);
1813 ci->i_inline_version = CEPH_INLINE_NONE;
1814 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1816 spin_unlock(&ci->i_ceph_lock);
1818 __mark_inode_dirty(inode, dirty);
1819 if ((endoff > size) &&
1820 ceph_quota_is_max_bytes_approaching(inode, endoff))
1821 ceph_check_caps(ci, CHECK_CAPS_NODELAY, NULL);
1824 ceph_put_cap_refs(ci, got);
1826 inode_unlock(inode);
1827 ceph_free_cap_flush(prealloc_cf);
1831 const struct file_operations ceph_file_fops = {
1833 .release = ceph_release,
1834 .llseek = ceph_llseek,
1835 .read_iter = ceph_read_iter,
1836 .write_iter = ceph_write_iter,
1838 .fsync = ceph_fsync,
1840 .flock = ceph_flock,
1841 .splice_read = generic_file_splice_read,
1842 .splice_write = iter_file_splice_write,
1843 .unlocked_ioctl = ceph_ioctl,
1844 .compat_ioctl = ceph_ioctl,
1845 .fallocate = ceph_fallocate,