xref: /openbmc/linux/fs/ceph/file.c (revision 260ea95c)
1 #include <linux/ceph/ceph_debug.h>
2 
3 #include <linux/module.h>
4 #include <linux/sched.h>
5 #include <linux/slab.h>
6 #include <linux/file.h>
7 #include <linux/mount.h>
8 #include <linux/namei.h>
9 #include <linux/writeback.h>
10 #include <linux/falloc.h>
11 
12 #include "super.h"
13 #include "mds_client.h"
14 #include "cache.h"
15 
16 static __le32 ceph_flags_sys2wire(u32 flags)
17 {
18 	u32 wire_flags = 0;
19 
20 	switch (flags & O_ACCMODE) {
21 	case O_RDONLY:
22 		wire_flags |= CEPH_O_RDONLY;
23 		break;
24 	case O_WRONLY:
25 		wire_flags |= CEPH_O_WRONLY;
26 		break;
27 	case O_RDWR:
28 		wire_flags |= CEPH_O_RDWR;
29 		break;
30 	}
31 
32 #define ceph_sys2wire(a) if (flags & a) { wire_flags |= CEPH_##a; flags &= ~a; }
33 
34 	ceph_sys2wire(O_CREAT);
35 	ceph_sys2wire(O_EXCL);
36 	ceph_sys2wire(O_TRUNC);
37 	ceph_sys2wire(O_DIRECTORY);
38 	ceph_sys2wire(O_NOFOLLOW);
39 
40 #undef ceph_sys2wire
41 
42 	if (flags)
43 		dout("unused open flags: %x", flags);
44 
45 	return cpu_to_le32(wire_flags);
46 }
47 
48 /*
49  * Ceph file operations
50  *
51  * Implement basic open/close functionality, and implement
52  * read/write.
53  *
54  * We implement three modes of file I/O:
55  *  - buffered uses the generic_file_aio_{read,write} helpers
56  *
57  *  - synchronous is used when there is multi-client read/write
58  *    sharing, avoids the page cache, and synchronously waits for an
59  *    ack from the OSD.
60  *
61  *  - direct io takes the variant of the sync path that references
62  *    user pages directly.
63  *
64  * fsync() flushes and waits on dirty pages, but just queues metadata
65  * for writeback: since the MDS can recover size and mtime there is no
66  * need to wait for MDS acknowledgement.
67  */
68 
69 /*
70  * Calculate the length sum of direct io vectors that can
71  * be combined into one page vector.
72  */
73 static size_t dio_get_pagev_size(const struct iov_iter *it)
74 {
75     const struct iovec *iov = it->iov;
76     const struct iovec *iovend = iov + it->nr_segs;
77     size_t size;
78 
79     size = iov->iov_len - it->iov_offset;
80     /*
81      * An iov can be page vectored when both the current tail
82      * and the next base are page aligned.
83      */
84     while (PAGE_ALIGNED((iov->iov_base + iov->iov_len)) &&
85            (++iov < iovend && PAGE_ALIGNED((iov->iov_base)))) {
86         size += iov->iov_len;
87     }
88     dout("dio_get_pagevlen len = %zu\n", size);
89     return size;
90 }
91 
92 /*
93  * Allocate a page vector based on (@it, @nbytes).
94  * The return value is the tuple describing a page vector,
95  * that is (@pages, @page_align, @num_pages).
96  */
97 static struct page **
98 dio_get_pages_alloc(const struct iov_iter *it, size_t nbytes,
99 		    size_t *page_align, int *num_pages)
100 {
101 	struct iov_iter tmp_it = *it;
102 	size_t align;
103 	struct page **pages;
104 	int ret = 0, idx, npages;
105 
106 	align = (unsigned long)(it->iov->iov_base + it->iov_offset) &
107 		(PAGE_SIZE - 1);
108 	npages = calc_pages_for(align, nbytes);
109 	pages = kvmalloc(sizeof(*pages) * npages, GFP_KERNEL);
110 	if (!pages)
111 		return ERR_PTR(-ENOMEM);
112 
113 	for (idx = 0; idx < npages; ) {
114 		size_t start;
115 		ret = iov_iter_get_pages(&tmp_it, pages + idx, nbytes,
116 					 npages - idx, &start);
117 		if (ret < 0)
118 			goto fail;
119 
120 		iov_iter_advance(&tmp_it, ret);
121 		nbytes -= ret;
122 		idx += (ret + start + PAGE_SIZE - 1) / PAGE_SIZE;
123 	}
124 
125 	BUG_ON(nbytes != 0);
126 	*num_pages = npages;
127 	*page_align = align;
128 	dout("dio_get_pages_alloc: got %d pages align %zu\n", npages, align);
129 	return pages;
130 fail:
131 	ceph_put_page_vector(pages, idx, false);
132 	return ERR_PTR(ret);
133 }
134 
135 /*
136  * Prepare an open request.  Preallocate ceph_cap to avoid an
137  * inopportune ENOMEM later.
138  */
139 static struct ceph_mds_request *
140 prepare_open_request(struct super_block *sb, int flags, int create_mode)
141 {
142 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
143 	struct ceph_mds_client *mdsc = fsc->mdsc;
144 	struct ceph_mds_request *req;
145 	int want_auth = USE_ANY_MDS;
146 	int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN;
147 
148 	if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC))
149 		want_auth = USE_AUTH_MDS;
150 
151 	req = ceph_mdsc_create_request(mdsc, op, want_auth);
152 	if (IS_ERR(req))
153 		goto out;
154 	req->r_fmode = ceph_flags_to_mode(flags);
155 	req->r_args.open.flags = ceph_flags_sys2wire(flags);
156 	req->r_args.open.mode = cpu_to_le32(create_mode);
157 out:
158 	return req;
159 }
160 
161 /*
162  * initialize private struct file data.
163  * if we fail, clean up by dropping fmode reference on the ceph_inode
164  */
165 static int ceph_init_file(struct inode *inode, struct file *file, int fmode)
166 {
167 	struct ceph_file_info *cf;
168 	int ret = 0;
169 
170 	switch (inode->i_mode & S_IFMT) {
171 	case S_IFREG:
172 		ceph_fscache_register_inode_cookie(inode);
173 		ceph_fscache_file_set_cookie(inode, file);
174 	case S_IFDIR:
175 		dout("init_file %p %p 0%o (regular)\n", inode, file,
176 		     inode->i_mode);
177 		cf = kmem_cache_zalloc(ceph_file_cachep, GFP_KERNEL);
178 		if (cf == NULL) {
179 			ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
180 			return -ENOMEM;
181 		}
182 		cf->fmode = fmode;
183 		cf->next_offset = 2;
184 		cf->readdir_cache_idx = -1;
185 		file->private_data = cf;
186 		BUG_ON(inode->i_fop->release != ceph_release);
187 		break;
188 
189 	case S_IFLNK:
190 		dout("init_file %p %p 0%o (symlink)\n", inode, file,
191 		     inode->i_mode);
192 		ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
193 		break;
194 
195 	default:
196 		dout("init_file %p %p 0%o (special)\n", inode, file,
197 		     inode->i_mode);
198 		/*
199 		 * we need to drop the open ref now, since we don't
200 		 * have .release set to ceph_release.
201 		 */
202 		ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
203 		BUG_ON(inode->i_fop->release == ceph_release);
204 
205 		/* call the proper open fop */
206 		ret = inode->i_fop->open(inode, file);
207 	}
208 	return ret;
209 }
210 
211 /*
212  * try renew caps after session gets killed.
213  */
214 int ceph_renew_caps(struct inode *inode)
215 {
216 	struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
217 	struct ceph_inode_info *ci = ceph_inode(inode);
218 	struct ceph_mds_request *req;
219 	int err, flags, wanted;
220 
221 	spin_lock(&ci->i_ceph_lock);
222 	wanted = __ceph_caps_file_wanted(ci);
223 	if (__ceph_is_any_real_caps(ci) &&
224 	    (!(wanted & CEPH_CAP_ANY_WR) || ci->i_auth_cap)) {
225 		int issued = __ceph_caps_issued(ci, NULL);
226 		spin_unlock(&ci->i_ceph_lock);
227 		dout("renew caps %p want %s issued %s updating mds_wanted\n",
228 		     inode, ceph_cap_string(wanted), ceph_cap_string(issued));
229 		ceph_check_caps(ci, 0, NULL);
230 		return 0;
231 	}
232 	spin_unlock(&ci->i_ceph_lock);
233 
234 	flags = 0;
235 	if ((wanted & CEPH_CAP_FILE_RD) && (wanted & CEPH_CAP_FILE_WR))
236 		flags = O_RDWR;
237 	else if (wanted & CEPH_CAP_FILE_RD)
238 		flags = O_RDONLY;
239 	else if (wanted & CEPH_CAP_FILE_WR)
240 		flags = O_WRONLY;
241 #ifdef O_LAZY
242 	if (wanted & CEPH_CAP_FILE_LAZYIO)
243 		flags |= O_LAZY;
244 #endif
245 
246 	req = prepare_open_request(inode->i_sb, flags, 0);
247 	if (IS_ERR(req)) {
248 		err = PTR_ERR(req);
249 		goto out;
250 	}
251 
252 	req->r_inode = inode;
253 	ihold(inode);
254 	req->r_num_caps = 1;
255 	req->r_fmode = -1;
256 
257 	err = ceph_mdsc_do_request(mdsc, NULL, req);
258 	ceph_mdsc_put_request(req);
259 out:
260 	dout("renew caps %p open result=%d\n", inode, err);
261 	return err < 0 ? err : 0;
262 }
263 
264 /*
265  * If we already have the requisite capabilities, we can satisfy
266  * the open request locally (no need to request new caps from the
267  * MDS).  We do, however, need to inform the MDS (asynchronously)
268  * if our wanted caps set expands.
269  */
270 int ceph_open(struct inode *inode, struct file *file)
271 {
272 	struct ceph_inode_info *ci = ceph_inode(inode);
273 	struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
274 	struct ceph_mds_client *mdsc = fsc->mdsc;
275 	struct ceph_mds_request *req;
276 	struct ceph_file_info *cf = file->private_data;
277 	int err;
278 	int flags, fmode, wanted;
279 
280 	if (cf) {
281 		dout("open file %p is already opened\n", file);
282 		return 0;
283 	}
284 
285 	/* filter out O_CREAT|O_EXCL; vfs did that already.  yuck. */
286 	flags = file->f_flags & ~(O_CREAT|O_EXCL);
287 	if (S_ISDIR(inode->i_mode))
288 		flags = O_DIRECTORY;  /* mds likes to know */
289 
290 	dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode,
291 	     ceph_vinop(inode), file, flags, file->f_flags);
292 	fmode = ceph_flags_to_mode(flags);
293 	wanted = ceph_caps_for_mode(fmode);
294 
295 	/* snapped files are read-only */
296 	if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE))
297 		return -EROFS;
298 
299 	/* trivially open snapdir */
300 	if (ceph_snap(inode) == CEPH_SNAPDIR) {
301 		spin_lock(&ci->i_ceph_lock);
302 		__ceph_get_fmode(ci, fmode);
303 		spin_unlock(&ci->i_ceph_lock);
304 		return ceph_init_file(inode, file, fmode);
305 	}
306 
307 	/*
308 	 * No need to block if we have caps on the auth MDS (for
309 	 * write) or any MDS (for read).  Update wanted set
310 	 * asynchronously.
311 	 */
312 	spin_lock(&ci->i_ceph_lock);
313 	if (__ceph_is_any_real_caps(ci) &&
314 	    (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) {
315 		int mds_wanted = __ceph_caps_mds_wanted(ci, true);
316 		int issued = __ceph_caps_issued(ci, NULL);
317 
318 		dout("open %p fmode %d want %s issued %s using existing\n",
319 		     inode, fmode, ceph_cap_string(wanted),
320 		     ceph_cap_string(issued));
321 		__ceph_get_fmode(ci, fmode);
322 		spin_unlock(&ci->i_ceph_lock);
323 
324 		/* adjust wanted? */
325 		if ((issued & wanted) != wanted &&
326 		    (mds_wanted & wanted) != wanted &&
327 		    ceph_snap(inode) != CEPH_SNAPDIR)
328 			ceph_check_caps(ci, 0, NULL);
329 
330 		return ceph_init_file(inode, file, fmode);
331 	} else if (ceph_snap(inode) != CEPH_NOSNAP &&
332 		   (ci->i_snap_caps & wanted) == wanted) {
333 		__ceph_get_fmode(ci, fmode);
334 		spin_unlock(&ci->i_ceph_lock);
335 		return ceph_init_file(inode, file, fmode);
336 	}
337 
338 	spin_unlock(&ci->i_ceph_lock);
339 
340 	dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted));
341 	req = prepare_open_request(inode->i_sb, flags, 0);
342 	if (IS_ERR(req)) {
343 		err = PTR_ERR(req);
344 		goto out;
345 	}
346 	req->r_inode = inode;
347 	ihold(inode);
348 
349 	req->r_num_caps = 1;
350 	err = ceph_mdsc_do_request(mdsc, NULL, req);
351 	if (!err)
352 		err = ceph_init_file(inode, file, req->r_fmode);
353 	ceph_mdsc_put_request(req);
354 	dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode));
355 out:
356 	return err;
357 }
358 
359 
360 /*
361  * Do a lookup + open with a single request.  If we get a non-existent
362  * file or symlink, return 1 so the VFS can retry.
363  */
364 int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
365 		     struct file *file, unsigned flags, umode_t mode,
366 		     int *opened)
367 {
368 	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
369 	struct ceph_mds_client *mdsc = fsc->mdsc;
370 	struct ceph_mds_request *req;
371 	struct dentry *dn;
372 	struct ceph_acls_info acls = {};
373        int mask;
374 	int err;
375 
376 	dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n",
377 	     dir, dentry, dentry,
378 	     d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode);
379 
380 	if (dentry->d_name.len > NAME_MAX)
381 		return -ENAMETOOLONG;
382 
383 	if (flags & O_CREAT) {
384 		err = ceph_pre_init_acls(dir, &mode, &acls);
385 		if (err < 0)
386 			return err;
387 	}
388 
389 	/* do the open */
390 	req = prepare_open_request(dir->i_sb, flags, mode);
391 	if (IS_ERR(req)) {
392 		err = PTR_ERR(req);
393 		goto out_acl;
394 	}
395 	req->r_dentry = dget(dentry);
396 	req->r_num_caps = 2;
397 	if (flags & O_CREAT) {
398 		req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
399 		req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
400 		if (acls.pagelist) {
401 			req->r_pagelist = acls.pagelist;
402 			acls.pagelist = NULL;
403 		}
404 	}
405 
406        mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
407        if (ceph_security_xattr_wanted(dir))
408                mask |= CEPH_CAP_XATTR_SHARED;
409        req->r_args.open.mask = cpu_to_le32(mask);
410 
411 	req->r_parent = dir;
412 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
413 	err = ceph_mdsc_do_request(mdsc,
414 				   (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
415 				   req);
416 	err = ceph_handle_snapdir(req, dentry, err);
417 	if (err)
418 		goto out_req;
419 
420 	if ((flags & O_CREAT) && !req->r_reply_info.head->is_dentry)
421 		err = ceph_handle_notrace_create(dir, dentry);
422 
423 	if (d_in_lookup(dentry)) {
424 		dn = ceph_finish_lookup(req, dentry, err);
425 		if (IS_ERR(dn))
426 			err = PTR_ERR(dn);
427 	} else {
428 		/* we were given a hashed negative dentry */
429 		dn = NULL;
430 	}
431 	if (err)
432 		goto out_req;
433 	if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) {
434 		/* make vfs retry on splice, ENOENT, or symlink */
435 		dout("atomic_open finish_no_open on dn %p\n", dn);
436 		err = finish_no_open(file, dn);
437 	} else {
438 		dout("atomic_open finish_open on dn %p\n", dn);
439 		if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) {
440 			ceph_init_inode_acls(d_inode(dentry), &acls);
441 			*opened |= FILE_CREATED;
442 		}
443 		err = finish_open(file, dentry, ceph_open, opened);
444 	}
445 out_req:
446 	if (!req->r_err && req->r_target_inode)
447 		ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode);
448 	ceph_mdsc_put_request(req);
449 out_acl:
450 	ceph_release_acls_info(&acls);
451 	dout("atomic_open result=%d\n", err);
452 	return err;
453 }
454 
455 int ceph_release(struct inode *inode, struct file *file)
456 {
457 	struct ceph_inode_info *ci = ceph_inode(inode);
458 	struct ceph_file_info *cf = file->private_data;
459 
460 	dout("release inode %p file %p\n", inode, file);
461 	ceph_put_fmode(ci, cf->fmode);
462 	if (cf->last_readdir)
463 		ceph_mdsc_put_request(cf->last_readdir);
464 	kfree(cf->last_name);
465 	kfree(cf->dir_info);
466 	kmem_cache_free(ceph_file_cachep, cf);
467 
468 	/* wake up anyone waiting for caps on this inode */
469 	wake_up_all(&ci->i_cap_wq);
470 	return 0;
471 }
472 
473 enum {
474 	HAVE_RETRIED = 1,
475 	CHECK_EOF =    2,
476 	READ_INLINE =  3,
477 };
478 
479 /*
480  * Read a range of bytes striped over one or more objects.  Iterate over
481  * objects we stripe over.  (That's not atomic, but good enough for now.)
482  *
483  * If we get a short result from the OSD, check against i_size; we need to
484  * only return a short read to the caller if we hit EOF.
485  */
486 static int striped_read(struct inode *inode,
487 			u64 pos, u64 len,
488 			struct page **pages, int num_pages,
489 			int page_align, int *checkeof)
490 {
491 	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
492 	struct ceph_inode_info *ci = ceph_inode(inode);
493 	u64 this_len;
494 	loff_t i_size;
495 	int page_idx;
496 	int ret, read = 0;
497 	bool hit_stripe, was_short;
498 
499 	/*
500 	 * we may need to do multiple reads.  not atomic, unfortunately.
501 	 */
502 more:
503 	this_len = len;
504 	page_idx = (page_align + read) >> PAGE_SHIFT;
505 	ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode),
506 				  &ci->i_layout, pos, &this_len,
507 				  ci->i_truncate_seq, ci->i_truncate_size,
508 				  pages + page_idx, num_pages - page_idx,
509 				  ((page_align + read) & ~PAGE_MASK));
510 	if (ret == -ENOENT)
511 		ret = 0;
512 	hit_stripe = this_len < len;
513 	was_short = ret >= 0 && ret < this_len;
514 	dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, len, read,
515 	     ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : "");
516 
517 	i_size = i_size_read(inode);
518 	if (ret >= 0) {
519 		if (was_short && (pos + ret < i_size)) {
520 			int zlen = min(this_len - ret, i_size - pos - ret);
521 			int zoff = page_align + read + ret;
522 			dout(" zero gap %llu to %llu\n",
523 			     pos + ret, pos + ret + zlen);
524 			ceph_zero_page_vector_range(zoff, zlen, pages);
525 			ret += zlen;
526 		}
527 
528 		read += ret;
529 		pos += ret;
530 		len -= ret;
531 
532 		/* hit stripe and need continue*/
533 		if (len && hit_stripe && pos < i_size)
534 			goto more;
535 	}
536 
537 	if (read > 0) {
538 		ret = read;
539 		/* did we bounce off eof? */
540 		if (pos + len > i_size)
541 			*checkeof = CHECK_EOF;
542 	}
543 
544 	dout("striped_read returns %d\n", ret);
545 	return ret;
546 }
547 
548 /*
549  * Completely synchronous read and write methods.  Direct from __user
550  * buffer to osd, or directly to user pages (if O_DIRECT).
551  *
552  * If the read spans object boundary, just do multiple reads.
553  */
554 static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *to,
555 			      int *checkeof)
556 {
557 	struct file *file = iocb->ki_filp;
558 	struct inode *inode = file_inode(file);
559 	struct page **pages;
560 	u64 off = iocb->ki_pos;
561 	int num_pages;
562 	ssize_t ret;
563 	size_t len = iov_iter_count(to);
564 
565 	dout("sync_read on file %p %llu~%u %s\n", file, off,
566 	     (unsigned)len,
567 	     (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
568 
569 	if (!len)
570 		return 0;
571 	/*
572 	 * flush any page cache pages in this range.  this
573 	 * will make concurrent normal and sync io slow,
574 	 * but it will at least behave sensibly when they are
575 	 * in sequence.
576 	 */
577 	ret = filemap_write_and_wait_range(inode->i_mapping, off,
578 						off + len);
579 	if (ret < 0)
580 		return ret;
581 
582 	if (unlikely(to->type & ITER_PIPE)) {
583 		size_t page_off;
584 		ret = iov_iter_get_pages_alloc(to, &pages, len,
585 					       &page_off);
586 		if (ret <= 0)
587 			return -ENOMEM;
588 		num_pages = DIV_ROUND_UP(ret + page_off, PAGE_SIZE);
589 
590 		ret = striped_read(inode, off, ret, pages, num_pages,
591 				   page_off, checkeof);
592 		if (ret > 0) {
593 			iov_iter_advance(to, ret);
594 			off += ret;
595 		} else {
596 			iov_iter_advance(to, 0);
597 		}
598 		ceph_put_page_vector(pages, num_pages, false);
599 	} else {
600 		num_pages = calc_pages_for(off, len);
601 		pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
602 		if (IS_ERR(pages))
603 			return PTR_ERR(pages);
604 
605 		ret = striped_read(inode, off, len, pages, num_pages,
606 				   (off & ~PAGE_MASK), checkeof);
607 		if (ret > 0) {
608 			int l, k = 0;
609 			size_t left = ret;
610 
611 			while (left) {
612 				size_t page_off = off & ~PAGE_MASK;
613 				size_t copy = min_t(size_t, left,
614 						    PAGE_SIZE - page_off);
615 				l = copy_page_to_iter(pages[k++], page_off,
616 						      copy, to);
617 				off += l;
618 				left -= l;
619 				if (l < copy)
620 					break;
621 			}
622 		}
623 		ceph_release_page_vector(pages, num_pages);
624 	}
625 
626 	if (off > iocb->ki_pos) {
627 		ret = off - iocb->ki_pos;
628 		iocb->ki_pos = off;
629 	}
630 
631 	dout("sync_read result %zd\n", ret);
632 	return ret;
633 }
634 
635 struct ceph_aio_request {
636 	struct kiocb *iocb;
637 	size_t total_len;
638 	int write;
639 	int error;
640 	struct list_head osd_reqs;
641 	unsigned num_reqs;
642 	atomic_t pending_reqs;
643 	struct timespec mtime;
644 	struct ceph_cap_flush *prealloc_cf;
645 };
646 
647 struct ceph_aio_work {
648 	struct work_struct work;
649 	struct ceph_osd_request *req;
650 };
651 
652 static void ceph_aio_retry_work(struct work_struct *work);
653 
654 static void ceph_aio_complete(struct inode *inode,
655 			      struct ceph_aio_request *aio_req)
656 {
657 	struct ceph_inode_info *ci = ceph_inode(inode);
658 	int ret;
659 
660 	if (!atomic_dec_and_test(&aio_req->pending_reqs))
661 		return;
662 
663 	ret = aio_req->error;
664 	if (!ret)
665 		ret = aio_req->total_len;
666 
667 	dout("ceph_aio_complete %p rc %d\n", inode, ret);
668 
669 	if (ret >= 0 && aio_req->write) {
670 		int dirty;
671 
672 		loff_t endoff = aio_req->iocb->ki_pos + aio_req->total_len;
673 		if (endoff > i_size_read(inode)) {
674 			if (ceph_inode_set_size(inode, endoff))
675 				ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
676 		}
677 
678 		spin_lock(&ci->i_ceph_lock);
679 		ci->i_inline_version = CEPH_INLINE_NONE;
680 		dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
681 					       &aio_req->prealloc_cf);
682 		spin_unlock(&ci->i_ceph_lock);
683 		if (dirty)
684 			__mark_inode_dirty(inode, dirty);
685 
686 	}
687 
688 	ceph_put_cap_refs(ci, (aio_req->write ? CEPH_CAP_FILE_WR :
689 						CEPH_CAP_FILE_RD));
690 
691 	aio_req->iocb->ki_complete(aio_req->iocb, ret, 0);
692 
693 	ceph_free_cap_flush(aio_req->prealloc_cf);
694 	kfree(aio_req);
695 }
696 
697 static void ceph_aio_complete_req(struct ceph_osd_request *req)
698 {
699 	int rc = req->r_result;
700 	struct inode *inode = req->r_inode;
701 	struct ceph_aio_request *aio_req = req->r_priv;
702 	struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0);
703 	int num_pages = calc_pages_for((u64)osd_data->alignment,
704 				       osd_data->length);
705 
706 	dout("ceph_aio_complete_req %p rc %d bytes %llu\n",
707 	     inode, rc, osd_data->length);
708 
709 	if (rc == -EOLDSNAPC) {
710 		struct ceph_aio_work *aio_work;
711 		BUG_ON(!aio_req->write);
712 
713 		aio_work = kmalloc(sizeof(*aio_work), GFP_NOFS);
714 		if (aio_work) {
715 			INIT_WORK(&aio_work->work, ceph_aio_retry_work);
716 			aio_work->req = req;
717 			queue_work(ceph_inode_to_client(inode)->wb_wq,
718 				   &aio_work->work);
719 			return;
720 		}
721 		rc = -ENOMEM;
722 	} else if (!aio_req->write) {
723 		if (rc == -ENOENT)
724 			rc = 0;
725 		if (rc >= 0 && osd_data->length > rc) {
726 			int zoff = osd_data->alignment + rc;
727 			int zlen = osd_data->length - rc;
728 			/*
729 			 * If read is satisfied by single OSD request,
730 			 * it can pass EOF. Otherwise read is within
731 			 * i_size.
732 			 */
733 			if (aio_req->num_reqs == 1) {
734 				loff_t i_size = i_size_read(inode);
735 				loff_t endoff = aio_req->iocb->ki_pos + rc;
736 				if (endoff < i_size)
737 					zlen = min_t(size_t, zlen,
738 						     i_size - endoff);
739 				aio_req->total_len = rc + zlen;
740 			}
741 
742 			if (zlen > 0)
743 				ceph_zero_page_vector_range(zoff, zlen,
744 							    osd_data->pages);
745 		}
746 	}
747 
748 	ceph_put_page_vector(osd_data->pages, num_pages, !aio_req->write);
749 	ceph_osdc_put_request(req);
750 
751 	if (rc < 0)
752 		cmpxchg(&aio_req->error, 0, rc);
753 
754 	ceph_aio_complete(inode, aio_req);
755 	return;
756 }
757 
758 static void ceph_aio_retry_work(struct work_struct *work)
759 {
760 	struct ceph_aio_work *aio_work =
761 		container_of(work, struct ceph_aio_work, work);
762 	struct ceph_osd_request *orig_req = aio_work->req;
763 	struct ceph_aio_request *aio_req = orig_req->r_priv;
764 	struct inode *inode = orig_req->r_inode;
765 	struct ceph_inode_info *ci = ceph_inode(inode);
766 	struct ceph_snap_context *snapc;
767 	struct ceph_osd_request *req;
768 	int ret;
769 
770 	spin_lock(&ci->i_ceph_lock);
771 	if (__ceph_have_pending_cap_snap(ci)) {
772 		struct ceph_cap_snap *capsnap =
773 			list_last_entry(&ci->i_cap_snaps,
774 					struct ceph_cap_snap,
775 					ci_item);
776 		snapc = ceph_get_snap_context(capsnap->context);
777 	} else {
778 		BUG_ON(!ci->i_head_snapc);
779 		snapc = ceph_get_snap_context(ci->i_head_snapc);
780 	}
781 	spin_unlock(&ci->i_ceph_lock);
782 
783 	req = ceph_osdc_alloc_request(orig_req->r_osdc, snapc, 2,
784 			false, GFP_NOFS);
785 	if (!req) {
786 		ret = -ENOMEM;
787 		req = orig_req;
788 		goto out;
789 	}
790 
791 	req->r_flags = CEPH_OSD_FLAG_ORDERSNAP | CEPH_OSD_FLAG_WRITE;
792 	ceph_oloc_copy(&req->r_base_oloc, &orig_req->r_base_oloc);
793 	ceph_oid_copy(&req->r_base_oid, &orig_req->r_base_oid);
794 
795 	ret = ceph_osdc_alloc_messages(req, GFP_NOFS);
796 	if (ret) {
797 		ceph_osdc_put_request(req);
798 		req = orig_req;
799 		goto out;
800 	}
801 
802 	req->r_ops[0] = orig_req->r_ops[0];
803 	osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC, 0);
804 
805 	req->r_mtime = aio_req->mtime;
806 	req->r_data_offset = req->r_ops[0].extent.offset;
807 
808 	ceph_osdc_put_request(orig_req);
809 
810 	req->r_callback = ceph_aio_complete_req;
811 	req->r_inode = inode;
812 	req->r_priv = aio_req;
813 	req->r_abort_on_full = true;
814 
815 	ret = ceph_osdc_start_request(req->r_osdc, req, false);
816 out:
817 	if (ret < 0) {
818 		req->r_result = ret;
819 		ceph_aio_complete_req(req);
820 	}
821 
822 	ceph_put_snap_context(snapc);
823 	kfree(aio_work);
824 }
825 
826 static ssize_t
827 ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter,
828 		       struct ceph_snap_context *snapc,
829 		       struct ceph_cap_flush **pcf)
830 {
831 	struct file *file = iocb->ki_filp;
832 	struct inode *inode = file_inode(file);
833 	struct ceph_inode_info *ci = ceph_inode(inode);
834 	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
835 	struct ceph_vino vino;
836 	struct ceph_osd_request *req;
837 	struct page **pages;
838 	struct ceph_aio_request *aio_req = NULL;
839 	int num_pages = 0;
840 	int flags;
841 	int ret;
842 	struct timespec mtime = current_time(inode);
843 	size_t count = iov_iter_count(iter);
844 	loff_t pos = iocb->ki_pos;
845 	bool write = iov_iter_rw(iter) == WRITE;
846 
847 	if (write && ceph_snap(file_inode(file)) != CEPH_NOSNAP)
848 		return -EROFS;
849 
850 	dout("sync_direct_read_write (%s) on file %p %lld~%u\n",
851 	     (write ? "write" : "read"), file, pos, (unsigned)count);
852 
853 	ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count);
854 	if (ret < 0)
855 		return ret;
856 
857 	if (write) {
858 		int ret2 = invalidate_inode_pages2_range(inode->i_mapping,
859 					pos >> PAGE_SHIFT,
860 					(pos + count) >> PAGE_SHIFT);
861 		if (ret2 < 0)
862 			dout("invalidate_inode_pages2_range returned %d\n", ret2);
863 
864 		flags = CEPH_OSD_FLAG_ORDERSNAP | CEPH_OSD_FLAG_WRITE;
865 	} else {
866 		flags = CEPH_OSD_FLAG_READ;
867 	}
868 
869 	while (iov_iter_count(iter) > 0) {
870 		u64 size = dio_get_pagev_size(iter);
871 		size_t start = 0;
872 		ssize_t len;
873 
874 		vino = ceph_vino(inode);
875 		req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
876 					    vino, pos, &size, 0,
877 					    /*include a 'startsync' command*/
878 					    write ? 2 : 1,
879 					    write ? CEPH_OSD_OP_WRITE :
880 						    CEPH_OSD_OP_READ,
881 					    flags, snapc,
882 					    ci->i_truncate_seq,
883 					    ci->i_truncate_size,
884 					    false);
885 		if (IS_ERR(req)) {
886 			ret = PTR_ERR(req);
887 			break;
888 		}
889 
890 		len = size;
891 		pages = dio_get_pages_alloc(iter, len, &start, &num_pages);
892 		if (IS_ERR(pages)) {
893 			ceph_osdc_put_request(req);
894 			ret = PTR_ERR(pages);
895 			break;
896 		}
897 
898 		/*
899 		 * To simplify error handling, allow AIO when IO within i_size
900 		 * or IO can be satisfied by single OSD request.
901 		 */
902 		if (pos == iocb->ki_pos && !is_sync_kiocb(iocb) &&
903 		    (len == count || pos + count <= i_size_read(inode))) {
904 			aio_req = kzalloc(sizeof(*aio_req), GFP_KERNEL);
905 			if (aio_req) {
906 				aio_req->iocb = iocb;
907 				aio_req->write = write;
908 				INIT_LIST_HEAD(&aio_req->osd_reqs);
909 				if (write) {
910 					aio_req->mtime = mtime;
911 					swap(aio_req->prealloc_cf, *pcf);
912 				}
913 			}
914 			/* ignore error */
915 		}
916 
917 		if (write) {
918 			/*
919 			 * throw out any page cache pages in this range. this
920 			 * may block.
921 			 */
922 			truncate_inode_pages_range(inode->i_mapping, pos,
923 					(pos+len) | (PAGE_SIZE - 1));
924 
925 			osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC, 0);
926 			req->r_mtime = mtime;
927 		}
928 
929 		osd_req_op_extent_osd_data_pages(req, 0, pages, len, start,
930 						 false, false);
931 
932 		if (aio_req) {
933 			aio_req->total_len += len;
934 			aio_req->num_reqs++;
935 			atomic_inc(&aio_req->pending_reqs);
936 
937 			req->r_callback = ceph_aio_complete_req;
938 			req->r_inode = inode;
939 			req->r_priv = aio_req;
940 			list_add_tail(&req->r_unsafe_item, &aio_req->osd_reqs);
941 
942 			pos += len;
943 			iov_iter_advance(iter, len);
944 			continue;
945 		}
946 
947 		ret = ceph_osdc_start_request(req->r_osdc, req, false);
948 		if (!ret)
949 			ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
950 
951 		size = i_size_read(inode);
952 		if (!write) {
953 			if (ret == -ENOENT)
954 				ret = 0;
955 			if (ret >= 0 && ret < len && pos + ret < size) {
956 				int zlen = min_t(size_t, len - ret,
957 						 size - pos - ret);
958 				ceph_zero_page_vector_range(start + ret, zlen,
959 							    pages);
960 				ret += zlen;
961 			}
962 			if (ret >= 0)
963 				len = ret;
964 		}
965 
966 		ceph_put_page_vector(pages, num_pages, !write);
967 
968 		ceph_osdc_put_request(req);
969 		if (ret < 0)
970 			break;
971 
972 		pos += len;
973 		iov_iter_advance(iter, len);
974 
975 		if (!write && pos >= size)
976 			break;
977 
978 		if (write && pos > size) {
979 			if (ceph_inode_set_size(inode, pos))
980 				ceph_check_caps(ceph_inode(inode),
981 						CHECK_CAPS_AUTHONLY,
982 						NULL);
983 		}
984 	}
985 
986 	if (aio_req) {
987 		LIST_HEAD(osd_reqs);
988 
989 		if (aio_req->num_reqs == 0) {
990 			kfree(aio_req);
991 			return ret;
992 		}
993 
994 		ceph_get_cap_refs(ci, write ? CEPH_CAP_FILE_WR :
995 					      CEPH_CAP_FILE_RD);
996 
997 		list_splice(&aio_req->osd_reqs, &osd_reqs);
998 		while (!list_empty(&osd_reqs)) {
999 			req = list_first_entry(&osd_reqs,
1000 					       struct ceph_osd_request,
1001 					       r_unsafe_item);
1002 			list_del_init(&req->r_unsafe_item);
1003 			if (ret >= 0)
1004 				ret = ceph_osdc_start_request(req->r_osdc,
1005 							      req, false);
1006 			if (ret < 0) {
1007 				req->r_result = ret;
1008 				ceph_aio_complete_req(req);
1009 			}
1010 		}
1011 		return -EIOCBQUEUED;
1012 	}
1013 
1014 	if (ret != -EOLDSNAPC && pos > iocb->ki_pos) {
1015 		ret = pos - iocb->ki_pos;
1016 		iocb->ki_pos = pos;
1017 	}
1018 	return ret;
1019 }
1020 
1021 /*
1022  * Synchronous write, straight from __user pointer or user pages.
1023  *
1024  * If write spans object boundary, just do multiple writes.  (For a
1025  * correct atomic write, we should e.g. take write locks on all
1026  * objects, rollback on failure, etc.)
1027  */
1028 static ssize_t
1029 ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
1030 		struct ceph_snap_context *snapc)
1031 {
1032 	struct file *file = iocb->ki_filp;
1033 	struct inode *inode = file_inode(file);
1034 	struct ceph_inode_info *ci = ceph_inode(inode);
1035 	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1036 	struct ceph_vino vino;
1037 	struct ceph_osd_request *req;
1038 	struct page **pages;
1039 	u64 len;
1040 	int num_pages;
1041 	int written = 0;
1042 	int flags;
1043 	int ret;
1044 	bool check_caps = false;
1045 	struct timespec mtime = current_time(inode);
1046 	size_t count = iov_iter_count(from);
1047 
1048 	if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
1049 		return -EROFS;
1050 
1051 	dout("sync_write on file %p %lld~%u\n", file, pos, (unsigned)count);
1052 
1053 	ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count);
1054 	if (ret < 0)
1055 		return ret;
1056 
1057 	ret = invalidate_inode_pages2_range(inode->i_mapping,
1058 					    pos >> PAGE_SHIFT,
1059 					    (pos + count) >> PAGE_SHIFT);
1060 	if (ret < 0)
1061 		dout("invalidate_inode_pages2_range returned %d\n", ret);
1062 
1063 	flags = CEPH_OSD_FLAG_ORDERSNAP | CEPH_OSD_FLAG_WRITE;
1064 
1065 	while ((len = iov_iter_count(from)) > 0) {
1066 		size_t left;
1067 		int n;
1068 
1069 		vino = ceph_vino(inode);
1070 		req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1071 					    vino, pos, &len, 0, 1,
1072 					    CEPH_OSD_OP_WRITE, flags, snapc,
1073 					    ci->i_truncate_seq,
1074 					    ci->i_truncate_size,
1075 					    false);
1076 		if (IS_ERR(req)) {
1077 			ret = PTR_ERR(req);
1078 			break;
1079 		}
1080 
1081 		/*
1082 		 * write from beginning of first page,
1083 		 * regardless of io alignment
1084 		 */
1085 		num_pages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1086 
1087 		pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
1088 		if (IS_ERR(pages)) {
1089 			ret = PTR_ERR(pages);
1090 			goto out;
1091 		}
1092 
1093 		left = len;
1094 		for (n = 0; n < num_pages; n++) {
1095 			size_t plen = min_t(size_t, left, PAGE_SIZE);
1096 			ret = copy_page_from_iter(pages[n], 0, plen, from);
1097 			if (ret != plen) {
1098 				ret = -EFAULT;
1099 				break;
1100 			}
1101 			left -= ret;
1102 		}
1103 
1104 		if (ret < 0) {
1105 			ceph_release_page_vector(pages, num_pages);
1106 			goto out;
1107 		}
1108 
1109 		req->r_inode = inode;
1110 
1111 		osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0,
1112 						false, true);
1113 
1114 		req->r_mtime = mtime;
1115 		ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1116 		if (!ret)
1117 			ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1118 
1119 out:
1120 		ceph_osdc_put_request(req);
1121 		if (ret != 0) {
1122 			ceph_set_error_write(ci);
1123 			break;
1124 		}
1125 
1126 		ceph_clear_error_write(ci);
1127 		pos += len;
1128 		written += len;
1129 		if (pos > i_size_read(inode)) {
1130 			check_caps = ceph_inode_set_size(inode, pos);
1131 			if (check_caps)
1132 				ceph_check_caps(ceph_inode(inode),
1133 						CHECK_CAPS_AUTHONLY,
1134 						NULL);
1135 		}
1136 
1137 	}
1138 
1139 	if (ret != -EOLDSNAPC && written > 0) {
1140 		ret = written;
1141 		iocb->ki_pos = pos;
1142 	}
1143 	return ret;
1144 }
1145 
1146 /*
1147  * Wrap generic_file_aio_read with checks for cap bits on the inode.
1148  * Atomically grab references, so that those bits are not released
1149  * back to the MDS mid-read.
1150  *
1151  * Hmm, the sync read case isn't actually async... should it be?
1152  */
1153 static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to)
1154 {
1155 	struct file *filp = iocb->ki_filp;
1156 	struct ceph_file_info *fi = filp->private_data;
1157 	size_t len = iov_iter_count(to);
1158 	struct inode *inode = file_inode(filp);
1159 	struct ceph_inode_info *ci = ceph_inode(inode);
1160 	struct page *pinned_page = NULL;
1161 	ssize_t ret;
1162 	int want, got = 0;
1163 	int retry_op = 0, read = 0;
1164 
1165 again:
1166 	dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
1167 	     inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode);
1168 
1169 	if (fi->fmode & CEPH_FILE_MODE_LAZY)
1170 		want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1171 	else
1172 		want = CEPH_CAP_FILE_CACHE;
1173 	ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page);
1174 	if (ret < 0)
1175 		return ret;
1176 
1177 	if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 ||
1178 	    (iocb->ki_flags & IOCB_DIRECT) ||
1179 	    (fi->flags & CEPH_F_SYNC)) {
1180 
1181 		dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n",
1182 		     inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
1183 		     ceph_cap_string(got));
1184 
1185 		if (ci->i_inline_version == CEPH_INLINE_NONE) {
1186 			if (!retry_op && (iocb->ki_flags & IOCB_DIRECT)) {
1187 				ret = ceph_direct_read_write(iocb, to,
1188 							     NULL, NULL);
1189 				if (ret >= 0 && ret < len)
1190 					retry_op = CHECK_EOF;
1191 			} else {
1192 				ret = ceph_sync_read(iocb, to, &retry_op);
1193 			}
1194 		} else {
1195 			retry_op = READ_INLINE;
1196 		}
1197 	} else {
1198 		dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
1199 		     inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
1200 		     ceph_cap_string(got));
1201 		current->journal_info = filp;
1202 		ret = generic_file_read_iter(iocb, to);
1203 		current->journal_info = NULL;
1204 	}
1205 	dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
1206 	     inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
1207 	if (pinned_page) {
1208 		put_page(pinned_page);
1209 		pinned_page = NULL;
1210 	}
1211 	ceph_put_cap_refs(ci, got);
1212 	if (retry_op > HAVE_RETRIED && ret >= 0) {
1213 		int statret;
1214 		struct page *page = NULL;
1215 		loff_t i_size;
1216 		if (retry_op == READ_INLINE) {
1217 			page = __page_cache_alloc(GFP_KERNEL);
1218 			if (!page)
1219 				return -ENOMEM;
1220 		}
1221 
1222 		statret = __ceph_do_getattr(inode, page,
1223 					    CEPH_STAT_CAP_INLINE_DATA, !!page);
1224 		if (statret < 0) {
1225 			if (page)
1226 				__free_page(page);
1227 			if (statret == -ENODATA) {
1228 				BUG_ON(retry_op != READ_INLINE);
1229 				goto again;
1230 			}
1231 			return statret;
1232 		}
1233 
1234 		i_size = i_size_read(inode);
1235 		if (retry_op == READ_INLINE) {
1236 			BUG_ON(ret > 0 || read > 0);
1237 			if (iocb->ki_pos < i_size &&
1238 			    iocb->ki_pos < PAGE_SIZE) {
1239 				loff_t end = min_t(loff_t, i_size,
1240 						   iocb->ki_pos + len);
1241 				end = min_t(loff_t, end, PAGE_SIZE);
1242 				if (statret < end)
1243 					zero_user_segment(page, statret, end);
1244 				ret = copy_page_to_iter(page,
1245 						iocb->ki_pos & ~PAGE_MASK,
1246 						end - iocb->ki_pos, to);
1247 				iocb->ki_pos += ret;
1248 				read += ret;
1249 			}
1250 			if (iocb->ki_pos < i_size && read < len) {
1251 				size_t zlen = min_t(size_t, len - read,
1252 						    i_size - iocb->ki_pos);
1253 				ret = iov_iter_zero(zlen, to);
1254 				iocb->ki_pos += ret;
1255 				read += ret;
1256 			}
1257 			__free_pages(page, 0);
1258 			return read;
1259 		}
1260 
1261 		/* hit EOF or hole? */
1262 		if (retry_op == CHECK_EOF && iocb->ki_pos < i_size &&
1263 		    ret < len) {
1264 			dout("sync_read hit hole, ppos %lld < size %lld"
1265 			     ", reading more\n", iocb->ki_pos, i_size);
1266 
1267 			read += ret;
1268 			len -= ret;
1269 			retry_op = HAVE_RETRIED;
1270 			goto again;
1271 		}
1272 	}
1273 
1274 	if (ret >= 0)
1275 		ret += read;
1276 
1277 	return ret;
1278 }
1279 
1280 /*
1281  * Take cap references to avoid releasing caps to MDS mid-write.
1282  *
1283  * If we are synchronous, and write with an old snap context, the OSD
1284  * may return EOLDSNAPC.  In that case, retry the write.. _after_
1285  * dropping our cap refs and allowing the pending snap to logically
1286  * complete _before_ this write occurs.
1287  *
1288  * If we are near ENOSPC, write synchronously.
1289  */
1290 static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
1291 {
1292 	struct file *file = iocb->ki_filp;
1293 	struct ceph_file_info *fi = file->private_data;
1294 	struct inode *inode = file_inode(file);
1295 	struct ceph_inode_info *ci = ceph_inode(inode);
1296 	struct ceph_osd_client *osdc =
1297 		&ceph_sb_to_client(inode->i_sb)->client->osdc;
1298 	struct ceph_cap_flush *prealloc_cf;
1299 	ssize_t count, written = 0;
1300 	int err, want, got;
1301 	loff_t pos;
1302 
1303 	if (ceph_snap(inode) != CEPH_NOSNAP)
1304 		return -EROFS;
1305 
1306 	prealloc_cf = ceph_alloc_cap_flush();
1307 	if (!prealloc_cf)
1308 		return -ENOMEM;
1309 
1310 	inode_lock(inode);
1311 
1312 	/* We can write back this queue in page reclaim */
1313 	current->backing_dev_info = inode_to_bdi(inode);
1314 
1315 	if (iocb->ki_flags & IOCB_APPEND) {
1316 		err = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
1317 		if (err < 0)
1318 			goto out;
1319 	}
1320 
1321 	err = generic_write_checks(iocb, from);
1322 	if (err <= 0)
1323 		goto out;
1324 
1325 	pos = iocb->ki_pos;
1326 	count = iov_iter_count(from);
1327 	err = file_remove_privs(file);
1328 	if (err)
1329 		goto out;
1330 
1331 	err = file_update_time(file);
1332 	if (err)
1333 		goto out;
1334 
1335 	if (ci->i_inline_version != CEPH_INLINE_NONE) {
1336 		err = ceph_uninline_data(file, NULL);
1337 		if (err < 0)
1338 			goto out;
1339 	}
1340 
1341 retry_snap:
1342 	/* FIXME: not complete since it doesn't account for being at quota */
1343 	if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL)) {
1344 		err = -ENOSPC;
1345 		goto out;
1346 	}
1347 
1348 	dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
1349 	     inode, ceph_vinop(inode), pos, count, i_size_read(inode));
1350 	if (fi->fmode & CEPH_FILE_MODE_LAZY)
1351 		want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1352 	else
1353 		want = CEPH_CAP_FILE_BUFFER;
1354 	got = 0;
1355 	err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, pos + count,
1356 			    &got, NULL);
1357 	if (err < 0)
1358 		goto out;
1359 
1360 	dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
1361 	     inode, ceph_vinop(inode), pos, count, ceph_cap_string(got));
1362 
1363 	if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 ||
1364 	    (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC) ||
1365 	    (ci->i_ceph_flags & CEPH_I_ERROR_WRITE)) {
1366 		struct ceph_snap_context *snapc;
1367 		struct iov_iter data;
1368 		inode_unlock(inode);
1369 
1370 		spin_lock(&ci->i_ceph_lock);
1371 		if (__ceph_have_pending_cap_snap(ci)) {
1372 			struct ceph_cap_snap *capsnap =
1373 					list_last_entry(&ci->i_cap_snaps,
1374 							struct ceph_cap_snap,
1375 							ci_item);
1376 			snapc = ceph_get_snap_context(capsnap->context);
1377 		} else {
1378 			BUG_ON(!ci->i_head_snapc);
1379 			snapc = ceph_get_snap_context(ci->i_head_snapc);
1380 		}
1381 		spin_unlock(&ci->i_ceph_lock);
1382 
1383 		/* we might need to revert back to that point */
1384 		data = *from;
1385 		if (iocb->ki_flags & IOCB_DIRECT)
1386 			written = ceph_direct_read_write(iocb, &data, snapc,
1387 							 &prealloc_cf);
1388 		else
1389 			written = ceph_sync_write(iocb, &data, pos, snapc);
1390 		if (written == -EOLDSNAPC) {
1391 			dout("aio_write %p %llx.%llx %llu~%u"
1392 				"got EOLDSNAPC, retrying\n",
1393 				inode, ceph_vinop(inode),
1394 				pos, (unsigned)count);
1395 			inode_lock(inode);
1396 			goto retry_snap;
1397 		}
1398 		if (written > 0)
1399 			iov_iter_advance(from, written);
1400 		ceph_put_snap_context(snapc);
1401 	} else {
1402 		/*
1403 		 * No need to acquire the i_truncate_mutex. Because
1404 		 * the MDS revokes Fwb caps before sending truncate
1405 		 * message to us. We can't get Fwb cap while there
1406 		 * are pending vmtruncate. So write and vmtruncate
1407 		 * can not run at the same time
1408 		 */
1409 		written = generic_perform_write(file, from, pos);
1410 		if (likely(written >= 0))
1411 			iocb->ki_pos = pos + written;
1412 		inode_unlock(inode);
1413 	}
1414 
1415 	if (written >= 0) {
1416 		int dirty;
1417 		spin_lock(&ci->i_ceph_lock);
1418 		ci->i_inline_version = CEPH_INLINE_NONE;
1419 		dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1420 					       &prealloc_cf);
1421 		spin_unlock(&ci->i_ceph_lock);
1422 		if (dirty)
1423 			__mark_inode_dirty(inode, dirty);
1424 	}
1425 
1426 	dout("aio_write %p %llx.%llx %llu~%u  dropping cap refs on %s\n",
1427 	     inode, ceph_vinop(inode), pos, (unsigned)count,
1428 	     ceph_cap_string(got));
1429 	ceph_put_cap_refs(ci, got);
1430 
1431 	if (written >= 0) {
1432 		if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_NEARFULL))
1433 			iocb->ki_flags |= IOCB_DSYNC;
1434 
1435 		written = generic_write_sync(iocb, written);
1436 	}
1437 
1438 	goto out_unlocked;
1439 
1440 out:
1441 	inode_unlock(inode);
1442 out_unlocked:
1443 	ceph_free_cap_flush(prealloc_cf);
1444 	current->backing_dev_info = NULL;
1445 	return written ? written : err;
1446 }
1447 
1448 /*
1449  * llseek.  be sure to verify file size on SEEK_END.
1450  */
1451 static loff_t ceph_llseek(struct file *file, loff_t offset, int whence)
1452 {
1453 	struct inode *inode = file->f_mapping->host;
1454 	loff_t i_size;
1455 	loff_t ret;
1456 
1457 	inode_lock(inode);
1458 
1459 	if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) {
1460 		ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
1461 		if (ret < 0)
1462 			goto out;
1463 	}
1464 
1465 	i_size = i_size_read(inode);
1466 	switch (whence) {
1467 	case SEEK_END:
1468 		offset += i_size;
1469 		break;
1470 	case SEEK_CUR:
1471 		/*
1472 		 * Here we special-case the lseek(fd, 0, SEEK_CUR)
1473 		 * position-querying operation.  Avoid rewriting the "same"
1474 		 * f_pos value back to the file because a concurrent read(),
1475 		 * write() or lseek() might have altered it
1476 		 */
1477 		if (offset == 0) {
1478 			ret = file->f_pos;
1479 			goto out;
1480 		}
1481 		offset += file->f_pos;
1482 		break;
1483 	case SEEK_DATA:
1484 		if (offset >= i_size) {
1485 			ret = -ENXIO;
1486 			goto out;
1487 		}
1488 		break;
1489 	case SEEK_HOLE:
1490 		if (offset >= i_size) {
1491 			ret = -ENXIO;
1492 			goto out;
1493 		}
1494 		offset = i_size;
1495 		break;
1496 	}
1497 
1498 	ret = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1499 
1500 out:
1501 	inode_unlock(inode);
1502 	return ret;
1503 }
1504 
1505 static inline void ceph_zero_partial_page(
1506 	struct inode *inode, loff_t offset, unsigned size)
1507 {
1508 	struct page *page;
1509 	pgoff_t index = offset >> PAGE_SHIFT;
1510 
1511 	page = find_lock_page(inode->i_mapping, index);
1512 	if (page) {
1513 		wait_on_page_writeback(page);
1514 		zero_user(page, offset & (PAGE_SIZE - 1), size);
1515 		unlock_page(page);
1516 		put_page(page);
1517 	}
1518 }
1519 
1520 static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset,
1521 				      loff_t length)
1522 {
1523 	loff_t nearly = round_up(offset, PAGE_SIZE);
1524 	if (offset < nearly) {
1525 		loff_t size = nearly - offset;
1526 		if (length < size)
1527 			size = length;
1528 		ceph_zero_partial_page(inode, offset, size);
1529 		offset += size;
1530 		length -= size;
1531 	}
1532 	if (length >= PAGE_SIZE) {
1533 		loff_t size = round_down(length, PAGE_SIZE);
1534 		truncate_pagecache_range(inode, offset, offset + size - 1);
1535 		offset += size;
1536 		length -= size;
1537 	}
1538 	if (length)
1539 		ceph_zero_partial_page(inode, offset, length);
1540 }
1541 
1542 static int ceph_zero_partial_object(struct inode *inode,
1543 				    loff_t offset, loff_t *length)
1544 {
1545 	struct ceph_inode_info *ci = ceph_inode(inode);
1546 	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1547 	struct ceph_osd_request *req;
1548 	int ret = 0;
1549 	loff_t zero = 0;
1550 	int op;
1551 
1552 	if (!length) {
1553 		op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE;
1554 		length = &zero;
1555 	} else {
1556 		op = CEPH_OSD_OP_ZERO;
1557 	}
1558 
1559 	req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1560 					ceph_vino(inode),
1561 					offset, length,
1562 					0, 1, op,
1563 					CEPH_OSD_FLAG_WRITE,
1564 					NULL, 0, 0, false);
1565 	if (IS_ERR(req)) {
1566 		ret = PTR_ERR(req);
1567 		goto out;
1568 	}
1569 
1570 	req->r_mtime = inode->i_mtime;
1571 	ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1572 	if (!ret) {
1573 		ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1574 		if (ret == -ENOENT)
1575 			ret = 0;
1576 	}
1577 	ceph_osdc_put_request(req);
1578 
1579 out:
1580 	return ret;
1581 }
1582 
1583 static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length)
1584 {
1585 	int ret = 0;
1586 	struct ceph_inode_info *ci = ceph_inode(inode);
1587 	s32 stripe_unit = ci->i_layout.stripe_unit;
1588 	s32 stripe_count = ci->i_layout.stripe_count;
1589 	s32 object_size = ci->i_layout.object_size;
1590 	u64 object_set_size = object_size * stripe_count;
1591 	u64 nearly, t;
1592 
1593 	/* round offset up to next period boundary */
1594 	nearly = offset + object_set_size - 1;
1595 	t = nearly;
1596 	nearly -= do_div(t, object_set_size);
1597 
1598 	while (length && offset < nearly) {
1599 		loff_t size = length;
1600 		ret = ceph_zero_partial_object(inode, offset, &size);
1601 		if (ret < 0)
1602 			return ret;
1603 		offset += size;
1604 		length -= size;
1605 	}
1606 	while (length >= object_set_size) {
1607 		int i;
1608 		loff_t pos = offset;
1609 		for (i = 0; i < stripe_count; ++i) {
1610 			ret = ceph_zero_partial_object(inode, pos, NULL);
1611 			if (ret < 0)
1612 				return ret;
1613 			pos += stripe_unit;
1614 		}
1615 		offset += object_set_size;
1616 		length -= object_set_size;
1617 	}
1618 	while (length) {
1619 		loff_t size = length;
1620 		ret = ceph_zero_partial_object(inode, offset, &size);
1621 		if (ret < 0)
1622 			return ret;
1623 		offset += size;
1624 		length -= size;
1625 	}
1626 	return ret;
1627 }
1628 
1629 static long ceph_fallocate(struct file *file, int mode,
1630 				loff_t offset, loff_t length)
1631 {
1632 	struct ceph_file_info *fi = file->private_data;
1633 	struct inode *inode = file_inode(file);
1634 	struct ceph_inode_info *ci = ceph_inode(inode);
1635 	struct ceph_osd_client *osdc =
1636 		&ceph_inode_to_client(inode)->client->osdc;
1637 	struct ceph_cap_flush *prealloc_cf;
1638 	int want, got = 0;
1639 	int dirty;
1640 	int ret = 0;
1641 	loff_t endoff = 0;
1642 	loff_t size;
1643 
1644 	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1645 		return -EOPNOTSUPP;
1646 
1647 	if (!S_ISREG(inode->i_mode))
1648 		return -EOPNOTSUPP;
1649 
1650 	prealloc_cf = ceph_alloc_cap_flush();
1651 	if (!prealloc_cf)
1652 		return -ENOMEM;
1653 
1654 	inode_lock(inode);
1655 
1656 	if (ceph_snap(inode) != CEPH_NOSNAP) {
1657 		ret = -EROFS;
1658 		goto unlock;
1659 	}
1660 
1661 	if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) &&
1662 	    !(mode & FALLOC_FL_PUNCH_HOLE)) {
1663 		ret = -ENOSPC;
1664 		goto unlock;
1665 	}
1666 
1667 	if (ci->i_inline_version != CEPH_INLINE_NONE) {
1668 		ret = ceph_uninline_data(file, NULL);
1669 		if (ret < 0)
1670 			goto unlock;
1671 	}
1672 
1673 	size = i_size_read(inode);
1674 	if (!(mode & FALLOC_FL_KEEP_SIZE)) {
1675 		endoff = offset + length;
1676 		ret = inode_newsize_ok(inode, endoff);
1677 		if (ret)
1678 			goto unlock;
1679 	}
1680 
1681 	if (fi->fmode & CEPH_FILE_MODE_LAZY)
1682 		want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1683 	else
1684 		want = CEPH_CAP_FILE_BUFFER;
1685 
1686 	ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, endoff, &got, NULL);
1687 	if (ret < 0)
1688 		goto unlock;
1689 
1690 	if (mode & FALLOC_FL_PUNCH_HOLE) {
1691 		if (offset < size)
1692 			ceph_zero_pagecache_range(inode, offset, length);
1693 		ret = ceph_zero_objects(inode, offset, length);
1694 	} else if (endoff > size) {
1695 		truncate_pagecache_range(inode, size, -1);
1696 		if (ceph_inode_set_size(inode, endoff))
1697 			ceph_check_caps(ceph_inode(inode),
1698 				CHECK_CAPS_AUTHONLY, NULL);
1699 	}
1700 
1701 	if (!ret) {
1702 		spin_lock(&ci->i_ceph_lock);
1703 		ci->i_inline_version = CEPH_INLINE_NONE;
1704 		dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1705 					       &prealloc_cf);
1706 		spin_unlock(&ci->i_ceph_lock);
1707 		if (dirty)
1708 			__mark_inode_dirty(inode, dirty);
1709 	}
1710 
1711 	ceph_put_cap_refs(ci, got);
1712 unlock:
1713 	inode_unlock(inode);
1714 	ceph_free_cap_flush(prealloc_cf);
1715 	return ret;
1716 }
1717 
1718 const struct file_operations ceph_file_fops = {
1719 	.open = ceph_open,
1720 	.release = ceph_release,
1721 	.llseek = ceph_llseek,
1722 	.read_iter = ceph_read_iter,
1723 	.write_iter = ceph_write_iter,
1724 	.mmap = ceph_mmap,
1725 	.fsync = ceph_fsync,
1726 	.lock = ceph_lock,
1727 	.flock = ceph_flock,
1728 	.splice_read = generic_file_splice_read,
1729 	.splice_write = iter_file_splice_write,
1730 	.unlocked_ioctl = ceph_ioctl,
1731 	.compat_ioctl	= ceph_ioctl,
1732 	.fallocate	= ceph_fallocate,
1733 };
1734 
1735