1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
3 #include <linux/ceph/striper.h>
5 #include <linux/module.h>
6 #include <linux/sched.h>
7 #include <linux/slab.h>
8 #include <linux/file.h>
9 #include <linux/mount.h>
10 #include <linux/namei.h>
11 #include <linux/writeback.h>
12 #include <linux/falloc.h>
13 #include <linux/iversion.h>
16 #include "mds_client.h"
20 static __le32 ceph_flags_sys2wire(u32 flags)
24 switch (flags & O_ACCMODE) {
26 wire_flags |= CEPH_O_RDONLY;
29 wire_flags |= CEPH_O_WRONLY;
32 wire_flags |= CEPH_O_RDWR;
38 #define ceph_sys2wire(a) if (flags & a) { wire_flags |= CEPH_##a; flags &= ~a; }
40 ceph_sys2wire(O_CREAT);
41 ceph_sys2wire(O_EXCL);
42 ceph_sys2wire(O_TRUNC);
43 ceph_sys2wire(O_DIRECTORY);
44 ceph_sys2wire(O_NOFOLLOW);
49 dout("unused open flags: %x\n", flags);
51 return cpu_to_le32(wire_flags);
55 * Ceph file operations
57 * Implement basic open/close functionality, and implement
60 * We implement three modes of file I/O:
61 * - buffered uses the generic_file_aio_{read,write} helpers
63 * - synchronous is used when there is multi-client read/write
64 * sharing, avoids the page cache, and synchronously waits for an
67 * - direct io takes the variant of the sync path that references
68 * user pages directly.
70 * fsync() flushes and waits on dirty pages, but just queues metadata
71 * for writeback: since the MDS can recover size and mtime there is no
72 * need to wait for MDS acknowledgement.
76 * How many pages to get in one call to iov_iter_get_pages(). This
77 * determines the size of the on-stack array used as a buffer.
79 #define ITER_GET_BVECS_PAGES 64
81 static ssize_t __iter_get_bvecs(struct iov_iter *iter, size_t maxsize,
82 struct bio_vec *bvecs)
87 if (maxsize > iov_iter_count(iter))
88 maxsize = iov_iter_count(iter);
90 while (size < maxsize) {
91 struct page *pages[ITER_GET_BVECS_PAGES];
96 bytes = iov_iter_get_pages(iter, pages, maxsize - size,
97 ITER_GET_BVECS_PAGES, &start);
101 iov_iter_advance(iter, bytes);
104 for ( ; bytes; idx++, bvec_idx++) {
105 struct bio_vec bv = {
106 .bv_page = pages[idx],
107 .bv_len = min_t(int, bytes, PAGE_SIZE - start),
111 bvecs[bvec_idx] = bv;
121 * iov_iter_get_pages() only considers one iov_iter segment, no matter
122 * what maxsize or maxpages are given. For ITER_BVEC that is a single
125 * Attempt to get up to @maxsize bytes worth of pages from @iter.
126 * Return the number of bytes in the created bio_vec array, or an error.
128 static ssize_t iter_get_bvecs_alloc(struct iov_iter *iter, size_t maxsize,
129 struct bio_vec **bvecs, int *num_bvecs)
132 size_t orig_count = iov_iter_count(iter);
136 iov_iter_truncate(iter, maxsize);
137 npages = iov_iter_npages(iter, INT_MAX);
138 iov_iter_reexpand(iter, orig_count);
141 * __iter_get_bvecs() may populate only part of the array -- zero it
144 bv = kvmalloc_array(npages, sizeof(*bv), GFP_KERNEL | __GFP_ZERO);
148 bytes = __iter_get_bvecs(iter, maxsize, bv);
151 * No pages were pinned -- just free the array.
162 static void put_bvecs(struct bio_vec *bvecs, int num_bvecs, bool should_dirty)
166 for (i = 0; i < num_bvecs; i++) {
167 if (bvecs[i].bv_page) {
169 set_page_dirty_lock(bvecs[i].bv_page);
170 put_page(bvecs[i].bv_page);
177 * Prepare an open request. Preallocate ceph_cap to avoid an
178 * inopportune ENOMEM later.
180 static struct ceph_mds_request *
181 prepare_open_request(struct super_block *sb, int flags, int create_mode)
183 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
184 struct ceph_mds_client *mdsc = fsc->mdsc;
185 struct ceph_mds_request *req;
186 int want_auth = USE_ANY_MDS;
187 int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN;
189 if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC))
190 want_auth = USE_AUTH_MDS;
192 req = ceph_mdsc_create_request(mdsc, op, want_auth);
195 req->r_fmode = ceph_flags_to_mode(flags);
196 req->r_args.open.flags = ceph_flags_sys2wire(flags);
197 req->r_args.open.mode = cpu_to_le32(create_mode);
202 static int ceph_init_file_info(struct inode *inode, struct file *file,
203 int fmode, bool isdir)
205 struct ceph_inode_info *ci = ceph_inode(inode);
206 struct ceph_file_info *fi;
208 dout("%s %p %p 0%o (%s)\n", __func__, inode, file,
209 inode->i_mode, isdir ? "dir" : "regular");
210 BUG_ON(inode->i_fop->release != ceph_release);
213 struct ceph_dir_file_info *dfi =
214 kmem_cache_zalloc(ceph_dir_file_cachep, GFP_KERNEL);
216 ceph_put_fmode(ci, fmode); /* clean up */
220 file->private_data = dfi;
221 fi = &dfi->file_info;
222 dfi->next_offset = 2;
223 dfi->readdir_cache_idx = -1;
225 fi = kmem_cache_zalloc(ceph_file_cachep, GFP_KERNEL);
227 ceph_put_fmode(ci, fmode); /* clean up */
231 file->private_data = fi;
235 spin_lock_init(&fi->rw_contexts_lock);
236 INIT_LIST_HEAD(&fi->rw_contexts);
237 fi->filp_gen = READ_ONCE(ceph_inode_to_client(inode)->filp_gen);
243 * initialize private struct file data.
244 * if we fail, clean up by dropping fmode reference on the ceph_inode
246 static int ceph_init_file(struct inode *inode, struct file *file, int fmode)
250 switch (inode->i_mode & S_IFMT) {
252 ceph_fscache_register_inode_cookie(inode);
253 ceph_fscache_file_set_cookie(inode, file);
256 ret = ceph_init_file_info(inode, file, fmode,
257 S_ISDIR(inode->i_mode));
263 dout("init_file %p %p 0%o (symlink)\n", inode, file,
265 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
269 dout("init_file %p %p 0%o (special)\n", inode, file,
272 * we need to drop the open ref now, since we don't
273 * have .release set to ceph_release.
275 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
276 BUG_ON(inode->i_fop->release == ceph_release);
278 /* call the proper open fop */
279 ret = inode->i_fop->open(inode, file);
285 * try renew caps after session gets killed.
287 int ceph_renew_caps(struct inode *inode)
289 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
290 struct ceph_inode_info *ci = ceph_inode(inode);
291 struct ceph_mds_request *req;
292 int err, flags, wanted;
294 spin_lock(&ci->i_ceph_lock);
295 wanted = __ceph_caps_file_wanted(ci);
296 if (__ceph_is_any_real_caps(ci) &&
297 (!(wanted & CEPH_CAP_ANY_WR) || ci->i_auth_cap)) {
298 int issued = __ceph_caps_issued(ci, NULL);
299 spin_unlock(&ci->i_ceph_lock);
300 dout("renew caps %p want %s issued %s updating mds_wanted\n",
301 inode, ceph_cap_string(wanted), ceph_cap_string(issued));
302 ceph_check_caps(ci, 0, NULL);
305 spin_unlock(&ci->i_ceph_lock);
308 if ((wanted & CEPH_CAP_FILE_RD) && (wanted & CEPH_CAP_FILE_WR))
310 else if (wanted & CEPH_CAP_FILE_RD)
312 else if (wanted & CEPH_CAP_FILE_WR)
315 if (wanted & CEPH_CAP_FILE_LAZYIO)
319 req = prepare_open_request(inode->i_sb, flags, 0);
325 req->r_inode = inode;
330 err = ceph_mdsc_do_request(mdsc, NULL, req);
331 ceph_mdsc_put_request(req);
333 dout("renew caps %p open result=%d\n", inode, err);
334 return err < 0 ? err : 0;
338 * If we already have the requisite capabilities, we can satisfy
339 * the open request locally (no need to request new caps from the
340 * MDS). We do, however, need to inform the MDS (asynchronously)
341 * if our wanted caps set expands.
343 int ceph_open(struct inode *inode, struct file *file)
345 struct ceph_inode_info *ci = ceph_inode(inode);
346 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
347 struct ceph_mds_client *mdsc = fsc->mdsc;
348 struct ceph_mds_request *req;
349 struct ceph_file_info *fi = file->private_data;
351 int flags, fmode, wanted;
354 dout("open file %p is already opened\n", file);
358 /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */
359 flags = file->f_flags & ~(O_CREAT|O_EXCL);
360 if (S_ISDIR(inode->i_mode))
361 flags = O_DIRECTORY; /* mds likes to know */
363 dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode,
364 ceph_vinop(inode), file, flags, file->f_flags);
365 fmode = ceph_flags_to_mode(flags);
366 wanted = ceph_caps_for_mode(fmode);
368 /* snapped files are read-only */
369 if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE))
372 /* trivially open snapdir */
373 if (ceph_snap(inode) == CEPH_SNAPDIR) {
374 spin_lock(&ci->i_ceph_lock);
375 __ceph_get_fmode(ci, fmode);
376 spin_unlock(&ci->i_ceph_lock);
377 return ceph_init_file(inode, file, fmode);
381 * No need to block if we have caps on the auth MDS (for
382 * write) or any MDS (for read). Update wanted set
385 spin_lock(&ci->i_ceph_lock);
386 if (__ceph_is_any_real_caps(ci) &&
387 (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) {
388 int mds_wanted = __ceph_caps_mds_wanted(ci, true);
389 int issued = __ceph_caps_issued(ci, NULL);
391 dout("open %p fmode %d want %s issued %s using existing\n",
392 inode, fmode, ceph_cap_string(wanted),
393 ceph_cap_string(issued));
394 __ceph_get_fmode(ci, fmode);
395 spin_unlock(&ci->i_ceph_lock);
398 if ((issued & wanted) != wanted &&
399 (mds_wanted & wanted) != wanted &&
400 ceph_snap(inode) != CEPH_SNAPDIR)
401 ceph_check_caps(ci, 0, NULL);
403 return ceph_init_file(inode, file, fmode);
404 } else if (ceph_snap(inode) != CEPH_NOSNAP &&
405 (ci->i_snap_caps & wanted) == wanted) {
406 __ceph_get_fmode(ci, fmode);
407 spin_unlock(&ci->i_ceph_lock);
408 return ceph_init_file(inode, file, fmode);
411 spin_unlock(&ci->i_ceph_lock);
413 dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted));
414 req = prepare_open_request(inode->i_sb, flags, 0);
419 req->r_inode = inode;
423 err = ceph_mdsc_do_request(mdsc, NULL, req);
425 err = ceph_init_file(inode, file, req->r_fmode);
426 ceph_mdsc_put_request(req);
427 dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode));
434 * Do a lookup + open with a single request. If we get a non-existent
435 * file or symlink, return 1 so the VFS can retry.
437 int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
438 struct file *file, unsigned flags, umode_t mode)
440 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
441 struct ceph_mds_client *mdsc = fsc->mdsc;
442 struct ceph_mds_request *req;
444 struct ceph_acl_sec_ctx as_ctx = {};
448 dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n",
450 d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode);
452 if (dentry->d_name.len > NAME_MAX)
453 return -ENAMETOOLONG;
456 * Do not truncate the file, since atomic_open is called before the
457 * permission check. The caller will do the truncation afterward.
461 if (flags & O_CREAT) {
462 if (ceph_quota_is_max_files_exceeded(dir))
464 err = ceph_pre_init_acls(dir, &mode, &as_ctx);
467 err = ceph_security_init_secctx(dentry, mode, &as_ctx);
470 } else if (!d_in_lookup(dentry)) {
471 /* If it's not being looked up, it's negative */
476 req = prepare_open_request(dir->i_sb, flags, mode);
481 req->r_dentry = dget(dentry);
483 if (flags & O_CREAT) {
484 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
485 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
486 if (as_ctx.pagelist) {
487 req->r_pagelist = as_ctx.pagelist;
488 as_ctx.pagelist = NULL;
492 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
493 if (ceph_security_xattr_wanted(dir))
494 mask |= CEPH_CAP_XATTR_SHARED;
495 req->r_args.open.mask = cpu_to_le32(mask);
498 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
499 err = ceph_mdsc_do_request(mdsc, (flags & O_CREAT) ? dir : NULL, req);
500 err = ceph_handle_snapdir(req, dentry, err);
504 if ((flags & O_CREAT) && !req->r_reply_info.head->is_dentry)
505 err = ceph_handle_notrace_create(dir, dentry);
507 if (d_in_lookup(dentry)) {
508 dn = ceph_finish_lookup(req, dentry, err);
512 /* we were given a hashed negative dentry */
517 if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) {
518 /* make vfs retry on splice, ENOENT, or symlink */
519 dout("atomic_open finish_no_open on dn %p\n", dn);
520 err = finish_no_open(file, dn);
522 dout("atomic_open finish_open on dn %p\n", dn);
523 if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) {
524 ceph_init_inode_acls(d_inode(dentry), &as_ctx);
525 file->f_mode |= FMODE_CREATED;
527 err = finish_open(file, dentry, ceph_open);
530 if (!req->r_err && req->r_target_inode)
531 ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode);
532 ceph_mdsc_put_request(req);
534 ceph_release_acl_sec_ctx(&as_ctx);
535 dout("atomic_open result=%d\n", err);
539 int ceph_release(struct inode *inode, struct file *file)
541 struct ceph_inode_info *ci = ceph_inode(inode);
543 if (S_ISDIR(inode->i_mode)) {
544 struct ceph_dir_file_info *dfi = file->private_data;
545 dout("release inode %p dir file %p\n", inode, file);
546 WARN_ON(!list_empty(&dfi->file_info.rw_contexts));
548 ceph_put_fmode(ci, dfi->file_info.fmode);
550 if (dfi->last_readdir)
551 ceph_mdsc_put_request(dfi->last_readdir);
552 kfree(dfi->last_name);
553 kfree(dfi->dir_info);
554 kmem_cache_free(ceph_dir_file_cachep, dfi);
556 struct ceph_file_info *fi = file->private_data;
557 dout("release inode %p regular file %p\n", inode, file);
558 WARN_ON(!list_empty(&fi->rw_contexts));
560 ceph_put_fmode(ci, fi->fmode);
561 kmem_cache_free(ceph_file_cachep, fi);
564 /* wake up anyone waiting for caps on this inode */
565 wake_up_all(&ci->i_cap_wq);
576 * Completely synchronous read and write methods. Direct from __user
577 * buffer to osd, or directly to user pages (if O_DIRECT).
579 * If the read spans object boundary, just do multiple reads. (That's not
580 * atomic, but good enough for now.)
582 * If we get a short result from the OSD, check against i_size; we need to
583 * only return a short read to the caller if we hit EOF.
585 static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *to,
588 struct file *file = iocb->ki_filp;
589 struct inode *inode = file_inode(file);
590 struct ceph_inode_info *ci = ceph_inode(inode);
591 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
592 struct ceph_osd_client *osdc = &fsc->client->osdc;
594 u64 off = iocb->ki_pos;
595 u64 len = iov_iter_count(to);
597 dout("sync_read on file %p %llu~%u %s\n", file, off, (unsigned)len,
598 (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
603 * flush any page cache pages in this range. this
604 * will make concurrent normal and sync io slow,
605 * but it will at least behave sensibly when they are
608 ret = filemap_write_and_wait_range(inode->i_mapping,
614 while ((len = iov_iter_count(to)) > 0) {
615 struct ceph_osd_request *req;
622 req = ceph_osdc_new_request(osdc, &ci->i_layout,
623 ci->i_vino, off, &len, 0, 1,
624 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
625 NULL, ci->i_truncate_seq,
626 ci->i_truncate_size, false);
632 more = len < iov_iter_count(to);
634 if (unlikely(iov_iter_is_pipe(to))) {
635 ret = iov_iter_get_pages_alloc(to, &pages, len,
638 ceph_osdc_put_request(req);
642 num_pages = DIV_ROUND_UP(ret + page_off, PAGE_SIZE);
645 osd_req_op_extent_update(req, 0, len);
649 num_pages = calc_pages_for(off, len);
650 page_off = off & ~PAGE_MASK;
651 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
653 ceph_osdc_put_request(req);
654 ret = PTR_ERR(pages);
659 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_off,
661 ret = ceph_osdc_start_request(osdc, req, false);
663 ret = ceph_osdc_wait_request(osdc, req);
664 ceph_osdc_put_request(req);
666 i_size = i_size_read(inode);
667 dout("sync_read %llu~%llu got %zd i_size %llu%s\n",
668 off, len, ret, i_size, (more ? " MORE" : ""));
672 if (ret >= 0 && ret < len && (off + ret < i_size)) {
673 int zlen = min(len - ret, i_size - off - ret);
674 int zoff = page_off + ret;
675 dout("sync_read zero gap %llu~%llu\n",
676 off + ret, off + ret + zlen);
677 ceph_zero_page_vector_range(zoff, zlen, pages);
681 if (unlikely(iov_iter_is_pipe(to))) {
683 iov_iter_advance(to, ret);
686 iov_iter_advance(to, 0);
688 ceph_put_page_vector(pages, num_pages, false);
691 size_t left = ret > 0 ? ret : 0;
694 page_off = off & ~PAGE_MASK;
695 len = min_t(size_t, left, PAGE_SIZE - page_off);
696 copied = copy_page_to_iter(pages[idx++],
705 ceph_release_page_vector(pages, num_pages);
709 if (ret == -EBLACKLISTED)
710 fsc->blacklisted = true;
714 if (off >= i_size || !more)
718 if (off > iocb->ki_pos) {
720 iov_iter_count(to) > 0 && off >= i_size_read(inode))
721 *retry_op = CHECK_EOF;
722 ret = off - iocb->ki_pos;
726 dout("sync_read result %zd retry_op %d\n", ret, *retry_op);
730 struct ceph_aio_request {
736 struct list_head osd_reqs;
738 atomic_t pending_reqs;
739 struct timespec64 mtime;
740 struct ceph_cap_flush *prealloc_cf;
743 struct ceph_aio_work {
744 struct work_struct work;
745 struct ceph_osd_request *req;
748 static void ceph_aio_retry_work(struct work_struct *work);
750 static void ceph_aio_complete(struct inode *inode,
751 struct ceph_aio_request *aio_req)
753 struct ceph_inode_info *ci = ceph_inode(inode);
756 if (!atomic_dec_and_test(&aio_req->pending_reqs))
759 if (aio_req->iocb->ki_flags & IOCB_DIRECT)
760 inode_dio_end(inode);
762 ret = aio_req->error;
764 ret = aio_req->total_len;
766 dout("ceph_aio_complete %p rc %d\n", inode, ret);
768 if (ret >= 0 && aio_req->write) {
771 loff_t endoff = aio_req->iocb->ki_pos + aio_req->total_len;
772 if (endoff > i_size_read(inode)) {
773 if (ceph_inode_set_size(inode, endoff))
774 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
777 spin_lock(&ci->i_ceph_lock);
778 ci->i_inline_version = CEPH_INLINE_NONE;
779 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
780 &aio_req->prealloc_cf);
781 spin_unlock(&ci->i_ceph_lock);
783 __mark_inode_dirty(inode, dirty);
787 ceph_put_cap_refs(ci, (aio_req->write ? CEPH_CAP_FILE_WR :
790 aio_req->iocb->ki_complete(aio_req->iocb, ret, 0);
792 ceph_free_cap_flush(aio_req->prealloc_cf);
796 static void ceph_aio_complete_req(struct ceph_osd_request *req)
798 int rc = req->r_result;
799 struct inode *inode = req->r_inode;
800 struct ceph_aio_request *aio_req = req->r_priv;
801 struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0);
803 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_BVECS);
804 BUG_ON(!osd_data->num_bvecs);
806 dout("ceph_aio_complete_req %p rc %d bytes %u\n",
807 inode, rc, osd_data->bvec_pos.iter.bi_size);
809 if (rc == -EOLDSNAPC) {
810 struct ceph_aio_work *aio_work;
811 BUG_ON(!aio_req->write);
813 aio_work = kmalloc(sizeof(*aio_work), GFP_NOFS);
815 INIT_WORK(&aio_work->work, ceph_aio_retry_work);
817 queue_work(ceph_inode_to_client(inode)->inode_wq,
822 } else if (!aio_req->write) {
825 if (rc >= 0 && osd_data->bvec_pos.iter.bi_size > rc) {
827 int zlen = osd_data->bvec_pos.iter.bi_size - rc;
830 * If read is satisfied by single OSD request,
831 * it can pass EOF. Otherwise read is within
834 if (aio_req->num_reqs == 1) {
835 loff_t i_size = i_size_read(inode);
836 loff_t endoff = aio_req->iocb->ki_pos + rc;
838 zlen = min_t(size_t, zlen,
840 aio_req->total_len = rc + zlen;
843 iov_iter_bvec(&i, READ, osd_data->bvec_pos.bvecs,
845 osd_data->bvec_pos.iter.bi_size);
846 iov_iter_advance(&i, rc);
847 iov_iter_zero(zlen, &i);
851 put_bvecs(osd_data->bvec_pos.bvecs, osd_data->num_bvecs,
852 aio_req->should_dirty);
853 ceph_osdc_put_request(req);
856 cmpxchg(&aio_req->error, 0, rc);
858 ceph_aio_complete(inode, aio_req);
862 static void ceph_aio_retry_work(struct work_struct *work)
864 struct ceph_aio_work *aio_work =
865 container_of(work, struct ceph_aio_work, work);
866 struct ceph_osd_request *orig_req = aio_work->req;
867 struct ceph_aio_request *aio_req = orig_req->r_priv;
868 struct inode *inode = orig_req->r_inode;
869 struct ceph_inode_info *ci = ceph_inode(inode);
870 struct ceph_snap_context *snapc;
871 struct ceph_osd_request *req;
874 spin_lock(&ci->i_ceph_lock);
875 if (__ceph_have_pending_cap_snap(ci)) {
876 struct ceph_cap_snap *capsnap =
877 list_last_entry(&ci->i_cap_snaps,
878 struct ceph_cap_snap,
880 snapc = ceph_get_snap_context(capsnap->context);
882 BUG_ON(!ci->i_head_snapc);
883 snapc = ceph_get_snap_context(ci->i_head_snapc);
885 spin_unlock(&ci->i_ceph_lock);
887 req = ceph_osdc_alloc_request(orig_req->r_osdc, snapc, 1,
895 req->r_flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE;
896 ceph_oloc_copy(&req->r_base_oloc, &orig_req->r_base_oloc);
897 ceph_oid_copy(&req->r_base_oid, &orig_req->r_base_oid);
899 req->r_ops[0] = orig_req->r_ops[0];
901 req->r_mtime = aio_req->mtime;
902 req->r_data_offset = req->r_ops[0].extent.offset;
904 ret = ceph_osdc_alloc_messages(req, GFP_NOFS);
906 ceph_osdc_put_request(req);
911 ceph_osdc_put_request(orig_req);
913 req->r_callback = ceph_aio_complete_req;
914 req->r_inode = inode;
915 req->r_priv = aio_req;
917 ret = ceph_osdc_start_request(req->r_osdc, req, false);
921 ceph_aio_complete_req(req);
924 ceph_put_snap_context(snapc);
929 ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter,
930 struct ceph_snap_context *snapc,
931 struct ceph_cap_flush **pcf)
933 struct file *file = iocb->ki_filp;
934 struct inode *inode = file_inode(file);
935 struct ceph_inode_info *ci = ceph_inode(inode);
936 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
937 struct ceph_vino vino;
938 struct ceph_osd_request *req;
939 struct bio_vec *bvecs;
940 struct ceph_aio_request *aio_req = NULL;
944 struct timespec64 mtime = current_time(inode);
945 size_t count = iov_iter_count(iter);
946 loff_t pos = iocb->ki_pos;
947 bool write = iov_iter_rw(iter) == WRITE;
948 bool should_dirty = !write && iter_is_iovec(iter);
950 if (write && ceph_snap(file_inode(file)) != CEPH_NOSNAP)
953 dout("sync_direct_%s on file %p %lld~%u snapc %p seq %lld\n",
954 (write ? "write" : "read"), file, pos, (unsigned)count,
955 snapc, snapc ? snapc->seq : 0);
958 int ret2 = invalidate_inode_pages2_range(inode->i_mapping,
960 (pos + count - 1) >> PAGE_SHIFT);
962 dout("invalidate_inode_pages2_range returned %d\n", ret2);
964 flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE;
966 flags = CEPH_OSD_FLAG_READ;
969 while (iov_iter_count(iter) > 0) {
970 u64 size = iov_iter_count(iter);
974 size = min_t(u64, size, fsc->mount_options->wsize);
976 size = min_t(u64, size, fsc->mount_options->rsize);
978 vino = ceph_vino(inode);
979 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
982 write ? CEPH_OSD_OP_WRITE :
993 len = iter_get_bvecs_alloc(iter, size, &bvecs, &num_pages);
995 ceph_osdc_put_request(req);
1000 osd_req_op_extent_update(req, 0, len);
1003 * To simplify error handling, allow AIO when IO within i_size
1004 * or IO can be satisfied by single OSD request.
1006 if (pos == iocb->ki_pos && !is_sync_kiocb(iocb) &&
1007 (len == count || pos + count <= i_size_read(inode))) {
1008 aio_req = kzalloc(sizeof(*aio_req), GFP_KERNEL);
1010 aio_req->iocb = iocb;
1011 aio_req->write = write;
1012 aio_req->should_dirty = should_dirty;
1013 INIT_LIST_HEAD(&aio_req->osd_reqs);
1015 aio_req->mtime = mtime;
1016 swap(aio_req->prealloc_cf, *pcf);
1024 * throw out any page cache pages in this range. this
1027 truncate_inode_pages_range(inode->i_mapping, pos,
1028 PAGE_ALIGN(pos + len) - 1);
1030 req->r_mtime = mtime;
1033 osd_req_op_extent_osd_data_bvecs(req, 0, bvecs, num_pages, len);
1036 aio_req->total_len += len;
1037 aio_req->num_reqs++;
1038 atomic_inc(&aio_req->pending_reqs);
1040 req->r_callback = ceph_aio_complete_req;
1041 req->r_inode = inode;
1042 req->r_priv = aio_req;
1043 list_add_tail(&req->r_private_item, &aio_req->osd_reqs);
1049 ret = ceph_osdc_start_request(req->r_osdc, req, false);
1051 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1053 size = i_size_read(inode);
1057 if (ret >= 0 && ret < len && pos + ret < size) {
1059 int zlen = min_t(size_t, len - ret,
1062 iov_iter_bvec(&i, READ, bvecs, num_pages, len);
1063 iov_iter_advance(&i, ret);
1064 iov_iter_zero(zlen, &i);
1071 put_bvecs(bvecs, num_pages, should_dirty);
1072 ceph_osdc_put_request(req);
1077 if (!write && pos >= size)
1080 if (write && pos > size) {
1081 if (ceph_inode_set_size(inode, pos))
1082 ceph_check_caps(ceph_inode(inode),
1083 CHECK_CAPS_AUTHONLY,
1089 LIST_HEAD(osd_reqs);
1091 if (aio_req->num_reqs == 0) {
1096 ceph_get_cap_refs(ci, write ? CEPH_CAP_FILE_WR :
1099 list_splice(&aio_req->osd_reqs, &osd_reqs);
1100 inode_dio_begin(inode);
1101 while (!list_empty(&osd_reqs)) {
1102 req = list_first_entry(&osd_reqs,
1103 struct ceph_osd_request,
1105 list_del_init(&req->r_private_item);
1107 ret = ceph_osdc_start_request(req->r_osdc,
1110 req->r_result = ret;
1111 ceph_aio_complete_req(req);
1114 return -EIOCBQUEUED;
1117 if (ret != -EOLDSNAPC && pos > iocb->ki_pos) {
1118 ret = pos - iocb->ki_pos;
1125 * Synchronous write, straight from __user pointer or user pages.
1127 * If write spans object boundary, just do multiple writes. (For a
1128 * correct atomic write, we should e.g. take write locks on all
1129 * objects, rollback on failure, etc.)
1132 ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
1133 struct ceph_snap_context *snapc)
1135 struct file *file = iocb->ki_filp;
1136 struct inode *inode = file_inode(file);
1137 struct ceph_inode_info *ci = ceph_inode(inode);
1138 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1139 struct ceph_vino vino;
1140 struct ceph_osd_request *req;
1141 struct page **pages;
1147 bool check_caps = false;
1148 struct timespec64 mtime = current_time(inode);
1149 size_t count = iov_iter_count(from);
1151 if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
1154 dout("sync_write on file %p %lld~%u snapc %p seq %lld\n",
1155 file, pos, (unsigned)count, snapc, snapc->seq);
1157 ret = filemap_write_and_wait_range(inode->i_mapping,
1158 pos, pos + count - 1);
1162 ret = invalidate_inode_pages2_range(inode->i_mapping,
1164 (pos + count - 1) >> PAGE_SHIFT);
1166 dout("invalidate_inode_pages2_range returned %d\n", ret);
1168 flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE;
1170 while ((len = iov_iter_count(from)) > 0) {
1174 vino = ceph_vino(inode);
1175 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1176 vino, pos, &len, 0, 1,
1177 CEPH_OSD_OP_WRITE, flags, snapc,
1179 ci->i_truncate_size,
1187 * write from beginning of first page,
1188 * regardless of io alignment
1190 num_pages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1192 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
1193 if (IS_ERR(pages)) {
1194 ret = PTR_ERR(pages);
1199 for (n = 0; n < num_pages; n++) {
1200 size_t plen = min_t(size_t, left, PAGE_SIZE);
1201 ret = copy_page_from_iter(pages[n], 0, plen, from);
1210 ceph_release_page_vector(pages, num_pages);
1214 req->r_inode = inode;
1216 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0,
1219 req->r_mtime = mtime;
1220 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1222 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1225 ceph_osdc_put_request(req);
1227 ceph_set_error_write(ci);
1231 ceph_clear_error_write(ci);
1234 if (pos > i_size_read(inode)) {
1235 check_caps = ceph_inode_set_size(inode, pos);
1237 ceph_check_caps(ceph_inode(inode),
1238 CHECK_CAPS_AUTHONLY,
1244 if (ret != -EOLDSNAPC && written > 0) {
1252 * Wrap generic_file_aio_read with checks for cap bits on the inode.
1253 * Atomically grab references, so that those bits are not released
1254 * back to the MDS mid-read.
1256 * Hmm, the sync read case isn't actually async... should it be?
1258 static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to)
1260 struct file *filp = iocb->ki_filp;
1261 struct ceph_file_info *fi = filp->private_data;
1262 size_t len = iov_iter_count(to);
1263 struct inode *inode = file_inode(filp);
1264 struct ceph_inode_info *ci = ceph_inode(inode);
1265 struct page *pinned_page = NULL;
1266 bool direct_lock = iocb->ki_flags & IOCB_DIRECT;
1269 int retry_op = 0, read = 0;
1272 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
1273 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode);
1276 ceph_start_io_direct(inode);
1278 ceph_start_io_read(inode);
1280 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1281 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1283 want = CEPH_CAP_FILE_CACHE;
1284 ret = ceph_get_caps(filp, CEPH_CAP_FILE_RD, want, -1,
1285 &got, &pinned_page);
1287 if (iocb->ki_flags & IOCB_DIRECT)
1288 ceph_end_io_direct(inode);
1290 ceph_end_io_read(inode);
1294 if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 ||
1295 (iocb->ki_flags & IOCB_DIRECT) ||
1296 (fi->flags & CEPH_F_SYNC)) {
1298 dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n",
1299 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
1300 ceph_cap_string(got));
1302 if (ci->i_inline_version == CEPH_INLINE_NONE) {
1303 if (!retry_op && (iocb->ki_flags & IOCB_DIRECT)) {
1304 ret = ceph_direct_read_write(iocb, to,
1306 if (ret >= 0 && ret < len)
1307 retry_op = CHECK_EOF;
1309 ret = ceph_sync_read(iocb, to, &retry_op);
1312 retry_op = READ_INLINE;
1315 CEPH_DEFINE_RW_CONTEXT(rw_ctx, got);
1316 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
1317 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
1318 ceph_cap_string(got));
1319 ceph_add_rw_context(fi, &rw_ctx);
1320 ret = generic_file_read_iter(iocb, to);
1321 ceph_del_rw_context(fi, &rw_ctx);
1324 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
1325 inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
1327 put_page(pinned_page);
1330 ceph_put_cap_refs(ci, got);
1333 ceph_end_io_direct(inode);
1335 ceph_end_io_read(inode);
1337 if (retry_op > HAVE_RETRIED && ret >= 0) {
1339 struct page *page = NULL;
1341 if (retry_op == READ_INLINE) {
1342 page = __page_cache_alloc(GFP_KERNEL);
1347 statret = __ceph_do_getattr(inode, page,
1348 CEPH_STAT_CAP_INLINE_DATA, !!page);
1352 if (statret == -ENODATA) {
1353 BUG_ON(retry_op != READ_INLINE);
1359 i_size = i_size_read(inode);
1360 if (retry_op == READ_INLINE) {
1361 BUG_ON(ret > 0 || read > 0);
1362 if (iocb->ki_pos < i_size &&
1363 iocb->ki_pos < PAGE_SIZE) {
1364 loff_t end = min_t(loff_t, i_size,
1365 iocb->ki_pos + len);
1366 end = min_t(loff_t, end, PAGE_SIZE);
1368 zero_user_segment(page, statret, end);
1369 ret = copy_page_to_iter(page,
1370 iocb->ki_pos & ~PAGE_MASK,
1371 end - iocb->ki_pos, to);
1372 iocb->ki_pos += ret;
1375 if (iocb->ki_pos < i_size && read < len) {
1376 size_t zlen = min_t(size_t, len - read,
1377 i_size - iocb->ki_pos);
1378 ret = iov_iter_zero(zlen, to);
1379 iocb->ki_pos += ret;
1382 __free_pages(page, 0);
1386 /* hit EOF or hole? */
1387 if (retry_op == CHECK_EOF && iocb->ki_pos < i_size &&
1389 dout("sync_read hit hole, ppos %lld < size %lld"
1390 ", reading more\n", iocb->ki_pos, i_size);
1394 retry_op = HAVE_RETRIED;
1406 * Take cap references to avoid releasing caps to MDS mid-write.
1408 * If we are synchronous, and write with an old snap context, the OSD
1409 * may return EOLDSNAPC. In that case, retry the write.. _after_
1410 * dropping our cap refs and allowing the pending snap to logically
1411 * complete _before_ this write occurs.
1413 * If we are near ENOSPC, write synchronously.
1415 static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
1417 struct file *file = iocb->ki_filp;
1418 struct ceph_file_info *fi = file->private_data;
1419 struct inode *inode = file_inode(file);
1420 struct ceph_inode_info *ci = ceph_inode(inode);
1421 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1422 struct ceph_osd_client *osdc = &fsc->client->osdc;
1423 struct ceph_cap_flush *prealloc_cf;
1424 ssize_t count, written = 0;
1426 bool direct_lock = false;
1430 loff_t limit = max(i_size_read(inode), fsc->max_file_size);
1432 if (ceph_snap(inode) != CEPH_NOSNAP)
1435 prealloc_cf = ceph_alloc_cap_flush();
1439 if ((iocb->ki_flags & (IOCB_DIRECT | IOCB_APPEND)) == IOCB_DIRECT)
1444 ceph_start_io_direct(inode);
1446 ceph_start_io_write(inode);
1448 /* We can write back this queue in page reclaim */
1449 current->backing_dev_info = inode_to_bdi(inode);
1451 if (iocb->ki_flags & IOCB_APPEND) {
1452 err = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
1457 err = generic_write_checks(iocb, from);
1462 if (unlikely(pos >= limit)) {
1466 iov_iter_truncate(from, limit - pos);
1469 count = iov_iter_count(from);
1470 if (ceph_quota_is_max_bytes_exceeded(inode, pos + count)) {
1475 down_read(&osdc->lock);
1476 map_flags = osdc->osdmap->flags;
1477 pool_flags = ceph_pg_pool_flags(osdc->osdmap, ci->i_layout.pool_id);
1478 up_read(&osdc->lock);
1479 if ((map_flags & CEPH_OSDMAP_FULL) ||
1480 (pool_flags & CEPH_POOL_FLAG_FULL)) {
1485 err = file_remove_privs(file);
1489 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1490 err = ceph_uninline_data(file, NULL);
1495 dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
1496 inode, ceph_vinop(inode), pos, count, i_size_read(inode));
1497 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1498 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1500 want = CEPH_CAP_FILE_BUFFER;
1502 err = ceph_get_caps(file, CEPH_CAP_FILE_WR, want, pos + count,
1507 err = file_update_time(file);
1511 inode_inc_iversion_raw(inode);
1513 dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
1514 inode, ceph_vinop(inode), pos, count, ceph_cap_string(got));
1516 if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 ||
1517 (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC) ||
1518 (ci->i_ceph_flags & CEPH_I_ERROR_WRITE)) {
1519 struct ceph_snap_context *snapc;
1520 struct iov_iter data;
1522 spin_lock(&ci->i_ceph_lock);
1523 if (__ceph_have_pending_cap_snap(ci)) {
1524 struct ceph_cap_snap *capsnap =
1525 list_last_entry(&ci->i_cap_snaps,
1526 struct ceph_cap_snap,
1528 snapc = ceph_get_snap_context(capsnap->context);
1530 BUG_ON(!ci->i_head_snapc);
1531 snapc = ceph_get_snap_context(ci->i_head_snapc);
1533 spin_unlock(&ci->i_ceph_lock);
1535 /* we might need to revert back to that point */
1537 if (iocb->ki_flags & IOCB_DIRECT)
1538 written = ceph_direct_read_write(iocb, &data, snapc,
1541 written = ceph_sync_write(iocb, &data, pos, snapc);
1543 ceph_end_io_direct(inode);
1545 ceph_end_io_write(inode);
1547 iov_iter_advance(from, written);
1548 ceph_put_snap_context(snapc);
1551 * No need to acquire the i_truncate_mutex. Because
1552 * the MDS revokes Fwb caps before sending truncate
1553 * message to us. We can't get Fwb cap while there
1554 * are pending vmtruncate. So write and vmtruncate
1555 * can not run at the same time
1557 written = generic_perform_write(file, from, pos);
1558 if (likely(written >= 0))
1559 iocb->ki_pos = pos + written;
1560 ceph_end_io_write(inode);
1566 spin_lock(&ci->i_ceph_lock);
1567 ci->i_inline_version = CEPH_INLINE_NONE;
1568 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1570 spin_unlock(&ci->i_ceph_lock);
1572 __mark_inode_dirty(inode, dirty);
1573 if (ceph_quota_is_max_bytes_approaching(inode, iocb->ki_pos))
1574 ceph_check_caps(ci, CHECK_CAPS_NODELAY, NULL);
1577 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
1578 inode, ceph_vinop(inode), pos, (unsigned)count,
1579 ceph_cap_string(got));
1580 ceph_put_cap_refs(ci, got);
1582 if (written == -EOLDSNAPC) {
1583 dout("aio_write %p %llx.%llx %llu~%u" "got EOLDSNAPC, retrying\n",
1584 inode, ceph_vinop(inode), pos, (unsigned)count);
1589 if ((map_flags & CEPH_OSDMAP_NEARFULL) ||
1590 (pool_flags & CEPH_POOL_FLAG_NEARFULL))
1591 iocb->ki_flags |= IOCB_DSYNC;
1592 written = generic_write_sync(iocb, written);
1597 ceph_put_cap_refs(ci, got);
1600 ceph_end_io_direct(inode);
1602 ceph_end_io_write(inode);
1604 ceph_free_cap_flush(prealloc_cf);
1605 current->backing_dev_info = NULL;
1606 return written ? written : err;
1610 * llseek. be sure to verify file size on SEEK_END.
1612 static loff_t ceph_llseek(struct file *file, loff_t offset, int whence)
1614 struct inode *inode = file->f_mapping->host;
1615 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1621 if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) {
1622 ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
1627 i_size = i_size_read(inode);
1634 * Here we special-case the lseek(fd, 0, SEEK_CUR)
1635 * position-querying operation. Avoid rewriting the "same"
1636 * f_pos value back to the file because a concurrent read(),
1637 * write() or lseek() might have altered it
1643 offset += file->f_pos;
1646 if (offset < 0 || offset >= i_size) {
1652 if (offset < 0 || offset >= i_size) {
1660 ret = vfs_setpos(file, offset, max(i_size, fsc->max_file_size));
1663 inode_unlock(inode);
1667 static inline void ceph_zero_partial_page(
1668 struct inode *inode, loff_t offset, unsigned size)
1671 pgoff_t index = offset >> PAGE_SHIFT;
1673 page = find_lock_page(inode->i_mapping, index);
1675 wait_on_page_writeback(page);
1676 zero_user(page, offset & (PAGE_SIZE - 1), size);
1682 static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset,
1685 loff_t nearly = round_up(offset, PAGE_SIZE);
1686 if (offset < nearly) {
1687 loff_t size = nearly - offset;
1690 ceph_zero_partial_page(inode, offset, size);
1694 if (length >= PAGE_SIZE) {
1695 loff_t size = round_down(length, PAGE_SIZE);
1696 truncate_pagecache_range(inode, offset, offset + size - 1);
1701 ceph_zero_partial_page(inode, offset, length);
1704 static int ceph_zero_partial_object(struct inode *inode,
1705 loff_t offset, loff_t *length)
1707 struct ceph_inode_info *ci = ceph_inode(inode);
1708 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1709 struct ceph_osd_request *req;
1715 op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE;
1718 op = CEPH_OSD_OP_ZERO;
1721 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1725 CEPH_OSD_FLAG_WRITE,
1732 req->r_mtime = inode->i_mtime;
1733 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1735 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1739 ceph_osdc_put_request(req);
1745 static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length)
1748 struct ceph_inode_info *ci = ceph_inode(inode);
1749 s32 stripe_unit = ci->i_layout.stripe_unit;
1750 s32 stripe_count = ci->i_layout.stripe_count;
1751 s32 object_size = ci->i_layout.object_size;
1752 u64 object_set_size = object_size * stripe_count;
1755 /* round offset up to next period boundary */
1756 nearly = offset + object_set_size - 1;
1758 nearly -= do_div(t, object_set_size);
1760 while (length && offset < nearly) {
1761 loff_t size = length;
1762 ret = ceph_zero_partial_object(inode, offset, &size);
1768 while (length >= object_set_size) {
1770 loff_t pos = offset;
1771 for (i = 0; i < stripe_count; ++i) {
1772 ret = ceph_zero_partial_object(inode, pos, NULL);
1777 offset += object_set_size;
1778 length -= object_set_size;
1781 loff_t size = length;
1782 ret = ceph_zero_partial_object(inode, offset, &size);
1791 static long ceph_fallocate(struct file *file, int mode,
1792 loff_t offset, loff_t length)
1794 struct ceph_file_info *fi = file->private_data;
1795 struct inode *inode = file_inode(file);
1796 struct ceph_inode_info *ci = ceph_inode(inode);
1797 struct ceph_cap_flush *prealloc_cf;
1804 if (mode != (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1807 if (!S_ISREG(inode->i_mode))
1810 prealloc_cf = ceph_alloc_cap_flush();
1816 if (ceph_snap(inode) != CEPH_NOSNAP) {
1821 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1822 ret = ceph_uninline_data(file, NULL);
1827 size = i_size_read(inode);
1829 /* Are we punching a hole beyond EOF? */
1832 if ((offset + length) > size)
1833 length = size - offset;
1835 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1836 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1838 want = CEPH_CAP_FILE_BUFFER;
1840 ret = ceph_get_caps(file, CEPH_CAP_FILE_WR, want, endoff, &got, NULL);
1844 ceph_zero_pagecache_range(inode, offset, length);
1845 ret = ceph_zero_objects(inode, offset, length);
1848 spin_lock(&ci->i_ceph_lock);
1849 ci->i_inline_version = CEPH_INLINE_NONE;
1850 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1852 spin_unlock(&ci->i_ceph_lock);
1854 __mark_inode_dirty(inode, dirty);
1857 ceph_put_cap_refs(ci, got);
1859 inode_unlock(inode);
1860 ceph_free_cap_flush(prealloc_cf);
1865 * This function tries to get FILE_WR capabilities for dst_ci and FILE_RD for
1866 * src_ci. Two attempts are made to obtain both caps, and an error is return if
1867 * this fails; zero is returned on success.
1869 static int get_rd_wr_caps(struct file *src_filp, int *src_got,
1870 struct file *dst_filp,
1871 loff_t dst_endoff, int *dst_got)
1874 bool retrying = false;
1877 ret = ceph_get_caps(dst_filp, CEPH_CAP_FILE_WR, CEPH_CAP_FILE_BUFFER,
1878 dst_endoff, dst_got, NULL);
1883 * Since we're already holding the FILE_WR capability for the dst file,
1884 * we would risk a deadlock by using ceph_get_caps. Thus, we'll do some
1885 * retry dance instead to try to get both capabilities.
1887 ret = ceph_try_get_caps(file_inode(src_filp),
1888 CEPH_CAP_FILE_RD, CEPH_CAP_FILE_SHARED,
1891 /* Start by dropping dst_ci caps and getting src_ci caps */
1892 ceph_put_cap_refs(ceph_inode(file_inode(dst_filp)), *dst_got);
1895 /* ceph_try_get_caps masks EAGAIN */
1899 ret = ceph_get_caps(src_filp, CEPH_CAP_FILE_RD,
1900 CEPH_CAP_FILE_SHARED, -1, src_got, NULL);
1903 /*... drop src_ci caps too, and retry */
1904 ceph_put_cap_refs(ceph_inode(file_inode(src_filp)), *src_got);
1911 static void put_rd_wr_caps(struct ceph_inode_info *src_ci, int src_got,
1912 struct ceph_inode_info *dst_ci, int dst_got)
1914 ceph_put_cap_refs(src_ci, src_got);
1915 ceph_put_cap_refs(dst_ci, dst_got);
1919 * This function does several size-related checks, returning an error if:
1920 * - source file is smaller than off+len
1921 * - destination file size is not OK (inode_newsize_ok())
1922 * - max bytes quotas is exceeded
1924 static int is_file_size_ok(struct inode *src_inode, struct inode *dst_inode,
1925 loff_t src_off, loff_t dst_off, size_t len)
1927 loff_t size, endoff;
1929 size = i_size_read(src_inode);
1931 * Don't copy beyond source file EOF. Instead of simply setting length
1932 * to (size - src_off), just drop to VFS default implementation, as the
1933 * local i_size may be stale due to other clients writing to the source
1936 if (src_off + len > size) {
1937 dout("Copy beyond EOF (%llu + %zu > %llu)\n",
1938 src_off, len, size);
1941 size = i_size_read(dst_inode);
1943 endoff = dst_off + len;
1944 if (inode_newsize_ok(dst_inode, endoff))
1947 if (ceph_quota_is_max_bytes_exceeded(dst_inode, endoff))
1953 static ssize_t __ceph_copy_file_range(struct file *src_file, loff_t src_off,
1954 struct file *dst_file, loff_t dst_off,
1955 size_t len, unsigned int flags)
1957 struct inode *src_inode = file_inode(src_file);
1958 struct inode *dst_inode = file_inode(dst_file);
1959 struct ceph_inode_info *src_ci = ceph_inode(src_inode);
1960 struct ceph_inode_info *dst_ci = ceph_inode(dst_inode);
1961 struct ceph_cap_flush *prealloc_cf;
1962 struct ceph_fs_client *src_fsc = ceph_inode_to_client(src_inode);
1963 struct ceph_object_locator src_oloc, dst_oloc;
1964 struct ceph_object_id src_oid, dst_oid;
1965 loff_t endoff = 0, size;
1967 u64 src_objnum, dst_objnum, src_objoff, dst_objoff;
1968 u32 src_objlen, dst_objlen, object_size;
1969 int src_got = 0, dst_got = 0, err, dirty;
1970 bool do_final_copy = false;
1972 if (src_inode->i_sb != dst_inode->i_sb) {
1973 struct ceph_fs_client *dst_fsc = ceph_inode_to_client(dst_inode);
1975 if (ceph_fsid_compare(&src_fsc->client->fsid,
1976 &dst_fsc->client->fsid)) {
1977 dout("Copying files across clusters: src: %pU dst: %pU\n",
1978 &src_fsc->client->fsid, &dst_fsc->client->fsid);
1982 if (ceph_snap(dst_inode) != CEPH_NOSNAP)
1986 * Some of the checks below will return -EOPNOTSUPP, which will force a
1987 * fallback to the default VFS copy_file_range implementation. This is
1988 * desirable in several cases (for ex, the 'len' is smaller than the
1989 * size of the objects, or in cases where that would be more
1993 if (ceph_test_mount_opt(src_fsc, NOCOPYFROM))
1997 * Striped file layouts require that we copy partial objects, but the
1998 * OSD copy-from operation only supports full-object copies. Limit
1999 * this to non-striped file layouts for now.
2001 if ((src_ci->i_layout.stripe_unit != dst_ci->i_layout.stripe_unit) ||
2002 (src_ci->i_layout.stripe_count != 1) ||
2003 (dst_ci->i_layout.stripe_count != 1) ||
2004 (src_ci->i_layout.object_size != dst_ci->i_layout.object_size)) {
2005 dout("Invalid src/dst files layout\n");
2009 if (len < src_ci->i_layout.object_size)
2010 return -EOPNOTSUPP; /* no remote copy will be done */
2012 prealloc_cf = ceph_alloc_cap_flush();
2016 /* Start by sync'ing the source and destination files */
2017 ret = file_write_and_wait_range(src_file, src_off, (src_off + len));
2019 dout("failed to write src file (%zd)\n", ret);
2022 ret = file_write_and_wait_range(dst_file, dst_off, (dst_off + len));
2024 dout("failed to write dst file (%zd)\n", ret);
2029 * We need FILE_WR caps for dst_ci and FILE_RD for src_ci as other
2030 * clients may have dirty data in their caches. And OSDs know nothing
2031 * about caps, so they can't safely do the remote object copies.
2033 err = get_rd_wr_caps(src_file, &src_got,
2034 dst_file, (dst_off + len), &dst_got);
2036 dout("get_rd_wr_caps returned %d\n", err);
2041 ret = is_file_size_ok(src_inode, dst_inode, src_off, dst_off, len);
2045 size = i_size_read(dst_inode);
2046 endoff = dst_off + len;
2048 /* Drop dst file cached pages */
2049 ret = invalidate_inode_pages2_range(dst_inode->i_mapping,
2050 dst_off >> PAGE_SHIFT,
2051 endoff >> PAGE_SHIFT);
2053 dout("Failed to invalidate inode pages (%zd)\n", ret);
2056 src_oloc.pool = src_ci->i_layout.pool_id;
2057 src_oloc.pool_ns = ceph_try_get_string(src_ci->i_layout.pool_ns);
2058 dst_oloc.pool = dst_ci->i_layout.pool_id;
2059 dst_oloc.pool_ns = ceph_try_get_string(dst_ci->i_layout.pool_ns);
2061 ceph_calc_file_object_mapping(&src_ci->i_layout, src_off,
2062 src_ci->i_layout.object_size,
2063 &src_objnum, &src_objoff, &src_objlen);
2064 ceph_calc_file_object_mapping(&dst_ci->i_layout, dst_off,
2065 dst_ci->i_layout.object_size,
2066 &dst_objnum, &dst_objoff, &dst_objlen);
2067 /* object-level offsets need to the same */
2068 if (src_objoff != dst_objoff) {
2074 * Do a manual copy if the object offset isn't object aligned.
2075 * 'src_objlen' contains the bytes left until the end of the object,
2076 * starting at the src_off
2080 * we need to temporarily drop all caps as we'll be calling
2081 * {read,write}_iter, which will get caps again.
2083 put_rd_wr_caps(src_ci, src_got, dst_ci, dst_got);
2084 ret = do_splice_direct(src_file, &src_off, dst_file,
2085 &dst_off, src_objlen, flags);
2087 dout("do_splice_direct returned %d\n", err);
2091 err = get_rd_wr_caps(src_file, &src_got,
2092 dst_file, (dst_off + len), &dst_got);
2095 err = is_file_size_ok(src_inode, dst_inode,
2096 src_off, dst_off, len);
2100 object_size = src_ci->i_layout.object_size;
2101 while (len >= object_size) {
2102 ceph_calc_file_object_mapping(&src_ci->i_layout, src_off,
2103 object_size, &src_objnum,
2104 &src_objoff, &src_objlen);
2105 ceph_calc_file_object_mapping(&dst_ci->i_layout, dst_off,
2106 object_size, &dst_objnum,
2107 &dst_objoff, &dst_objlen);
2108 ceph_oid_init(&src_oid);
2109 ceph_oid_printf(&src_oid, "%llx.%08llx",
2110 src_ci->i_vino.ino, src_objnum);
2111 ceph_oid_init(&dst_oid);
2112 ceph_oid_printf(&dst_oid, "%llx.%08llx",
2113 dst_ci->i_vino.ino, dst_objnum);
2114 /* Do an object remote copy */
2115 err = ceph_osdc_copy_from(
2116 &src_fsc->client->osdc,
2117 src_ci->i_vino.snap, 0,
2118 &src_oid, &src_oloc,
2119 CEPH_OSD_OP_FLAG_FADVISE_SEQUENTIAL |
2120 CEPH_OSD_OP_FLAG_FADVISE_NOCACHE,
2121 &dst_oid, &dst_oloc,
2122 CEPH_OSD_OP_FLAG_FADVISE_SEQUENTIAL |
2123 CEPH_OSD_OP_FLAG_FADVISE_DONTNEED, 0);
2125 dout("ceph_osdc_copy_from returned %d\n", err);
2131 src_off += object_size;
2132 dst_off += object_size;
2137 /* We still need one final local copy */
2138 do_final_copy = true;
2140 file_update_time(dst_file);
2141 inode_inc_iversion_raw(dst_inode);
2143 if (endoff > size) {
2146 /* Let the MDS know about dst file size change */
2147 if (ceph_quota_is_max_bytes_approaching(dst_inode, endoff))
2148 caps_flags |= CHECK_CAPS_NODELAY;
2149 if (ceph_inode_set_size(dst_inode, endoff))
2150 caps_flags |= CHECK_CAPS_AUTHONLY;
2152 ceph_check_caps(dst_ci, caps_flags, NULL);
2155 spin_lock(&dst_ci->i_ceph_lock);
2156 dst_ci->i_inline_version = CEPH_INLINE_NONE;
2157 dirty = __ceph_mark_dirty_caps(dst_ci, CEPH_CAP_FILE_WR, &prealloc_cf);
2158 spin_unlock(&dst_ci->i_ceph_lock);
2160 __mark_inode_dirty(dst_inode, dirty);
2163 put_rd_wr_caps(src_ci, src_got, dst_ci, dst_got);
2165 if (do_final_copy) {
2166 err = do_splice_direct(src_file, &src_off, dst_file,
2167 &dst_off, len, flags);
2169 dout("do_splice_direct returned %d\n", err);
2177 ceph_free_cap_flush(prealloc_cf);
2182 static ssize_t ceph_copy_file_range(struct file *src_file, loff_t src_off,
2183 struct file *dst_file, loff_t dst_off,
2184 size_t len, unsigned int flags)
2188 ret = __ceph_copy_file_range(src_file, src_off, dst_file, dst_off,
2191 if (ret == -EOPNOTSUPP || ret == -EXDEV)
2192 ret = generic_copy_file_range(src_file, src_off, dst_file,
2193 dst_off, len, flags);
2197 const struct file_operations ceph_file_fops = {
2199 .release = ceph_release,
2200 .llseek = ceph_llseek,
2201 .read_iter = ceph_read_iter,
2202 .write_iter = ceph_write_iter,
2204 .fsync = ceph_fsync,
2206 .setlease = simple_nosetlease,
2207 .flock = ceph_flock,
2208 .splice_read = generic_file_splice_read,
2209 .splice_write = iter_file_splice_write,
2210 .unlocked_ioctl = ceph_ioctl,
2211 .compat_ioctl = compat_ptr_ioctl,
2212 .fallocate = ceph_fallocate,
2213 .copy_file_range = ceph_copy_file_range,