file.c

   1// SPDX-License-Identifier: GPL-2.0
   2#include <linux/ceph/ceph_debug.h>
   3#include <linux/ceph/striper.h>
   4
   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>
  14#include <linux/ktime.h>
  15#include <linux/splice.h>
  16
  17#include "super.h"
  18#include "mds_client.h"
  19#include "cache.h"
  20#include "io.h"
  21#include "metric.h"
  22
  23static __le32 ceph_flags_sys2wire(struct ceph_mds_client *mdsc, u32 flags)
  24{
  25	struct ceph_client *cl = mdsc->fsc->client;
  26	u32 wire_flags = 0;
  27
  28	switch (flags & O_ACCMODE) {
  29	case O_RDONLY:
  30		wire_flags |= CEPH_O_RDONLY;
  31		break;
  32	case O_WRONLY:
  33		wire_flags |= CEPH_O_WRONLY;
  34		break;
  35	case O_RDWR:
  36		wire_flags |= CEPH_O_RDWR;
  37		break;
  38	}
  39
  40	flags &= ~O_ACCMODE;
  41
  42#define ceph_sys2wire(a) if (flags & a) { wire_flags |= CEPH_##a; flags &= ~a; }
  43
  44	ceph_sys2wire(O_CREAT);
  45	ceph_sys2wire(O_EXCL);
  46	ceph_sys2wire(O_TRUNC);
  47	ceph_sys2wire(O_DIRECTORY);
  48	ceph_sys2wire(O_NOFOLLOW);
  49
  50#undef ceph_sys2wire
  51
  52	if (flags)
  53		doutc(cl, "unused open flags: %x\n", flags);
  54
  55	return cpu_to_le32(wire_flags);
  56}
  57
  58/*
  59 * Ceph file operations
  60 *
  61 * Implement basic open/close functionality, and implement
  62 * read/write.
  63 *
  64 * We implement three modes of file I/O:
  65 *  - buffered uses the generic_file_aio_{read,write} helpers
  66 *
  67 *  - synchronous is used when there is multi-client read/write
  68 *    sharing, avoids the page cache, and synchronously waits for an
  69 *    ack from the OSD.
  70 *
  71 *  - direct io takes the variant of the sync path that references
  72 *    user pages directly.
  73 *
  74 * fsync() flushes and waits on dirty pages, but just queues metadata
  75 * for writeback: since the MDS can recover size and mtime there is no
  76 * need to wait for MDS acknowledgement.
  77 */
  78
  79/*
  80 * How many pages to get in one call to iov_iter_get_pages().  This
  81 * determines the size of the on-stack array used as a buffer.
  82 */
  83#define ITER_GET_BVECS_PAGES	64
  84
  85static ssize_t __iter_get_bvecs(struct iov_iter *iter, size_t maxsize,
  86				struct bio_vec *bvecs)
  87{
  88	size_t size = 0;
  89	int bvec_idx = 0;
  90
  91	if (maxsize > iov_iter_count(iter))
  92		maxsize = iov_iter_count(iter);
  93
  94	while (size < maxsize) {
  95		struct page *pages[ITER_GET_BVECS_PAGES];
  96		ssize_t bytes;
  97		size_t start;
  98		int idx = 0;
  99
 100		bytes = iov_iter_get_pages2(iter, pages, maxsize - size,
 101					   ITER_GET_BVECS_PAGES, &start);
 102		if (bytes < 0)
 103			return size ?: bytes;
 104
 105		size += bytes;
 106
 107		for ( ; bytes; idx++, bvec_idx++) {
 108			int len = min_t(int, bytes, PAGE_SIZE - start);
 109
 110			bvec_set_page(&bvecs[bvec_idx], pages[idx], len, start);
 111			bytes -= len;
 112			start = 0;
 113		}
 114	}
 115
 116	return size;
 117}
 118
 119/*
 120 * iov_iter_get_pages() only considers one iov_iter segment, no matter
 121 * what maxsize or maxpages are given.  For ITER_BVEC that is a single
 122 * page.
 123 *
 124 * Attempt to get up to @maxsize bytes worth of pages from @iter.
 125 * Return the number of bytes in the created bio_vec array, or an error.
 126 */
 127static ssize_t iter_get_bvecs_alloc(struct iov_iter *iter, size_t maxsize,
 128				    struct bio_vec **bvecs, int *num_bvecs)
 129{
 130	struct bio_vec *bv;
 131	size_t orig_count = iov_iter_count(iter);
 132	ssize_t bytes;
 133	int npages;
 134
 135	iov_iter_truncate(iter, maxsize);
 136	npages = iov_iter_npages(iter, INT_MAX);
 137	iov_iter_reexpand(iter, orig_count);
 138
 139	/*
 140	 * __iter_get_bvecs() may populate only part of the array -- zero it
 141	 * out.
 142	 */
 143	bv = kvmalloc_array(npages, sizeof(*bv), GFP_KERNEL | __GFP_ZERO);
 144	if (!bv)
 145		return -ENOMEM;
 146
 147	bytes = __iter_get_bvecs(iter, maxsize, bv);
 148	if (bytes < 0) {
 149		/*
 150		 * No pages were pinned -- just free the array.
 151		 */
 152		kvfree(bv);
 153		return bytes;
 154	}
 155
 156	*bvecs = bv;
 157	*num_bvecs = npages;
 158	return bytes;
 159}
 160
 161static void put_bvecs(struct bio_vec *bvecs, int num_bvecs, bool should_dirty)
 162{
 163	int i;
 164
 165	for (i = 0; i < num_bvecs; i++) {
 166		if (bvecs[i].bv_page) {
 167			if (should_dirty)
 168				set_page_dirty_lock(bvecs[i].bv_page);
 169			put_page(bvecs[i].bv_page);
 170		}
 171	}
 172	kvfree(bvecs);
 173}
 174
 175/*
 176 * Prepare an open request.  Preallocate ceph_cap to avoid an
 177 * inopportune ENOMEM later.
 178 */
 179static struct ceph_mds_request *
 180prepare_open_request(struct super_block *sb, int flags, int create_mode)
 181{
 182	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(sb);
 183	struct ceph_mds_request *req;
 184	int want_auth = USE_ANY_MDS;
 185	int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN;
 186
 187	if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC))
 188		want_auth = USE_AUTH_MDS;
 189
 190	req = ceph_mdsc_create_request(mdsc, op, want_auth);
 191	if (IS_ERR(req))
 192		goto out;
 193	req->r_fmode = ceph_flags_to_mode(flags);
 194	req->r_args.open.flags = ceph_flags_sys2wire(mdsc, flags);
 195	req->r_args.open.mode = cpu_to_le32(create_mode);
 196out:
 197	return req;
 198}
 199
 200static int ceph_init_file_info(struct inode *inode, struct file *file,
 201					int fmode, bool isdir)
 202{
 203	struct ceph_inode_info *ci = ceph_inode(inode);
 204	struct ceph_mount_options *opt =
 205		ceph_inode_to_fs_client(&ci->netfs.inode)->mount_options;
 206	struct ceph_client *cl = ceph_inode_to_client(inode);
 207	struct ceph_file_info *fi;
 208	int ret;
 209
 210	doutc(cl, "%p %llx.%llx %p 0%o (%s)\n", inode, ceph_vinop(inode),
 211	      file, inode->i_mode, isdir ? "dir" : "regular");
 212	BUG_ON(inode->i_fop->release != ceph_release);
 213
 214	if (isdir) {
 215		struct ceph_dir_file_info *dfi =
 216			kmem_cache_zalloc(ceph_dir_file_cachep, GFP_KERNEL);
 217		if (!dfi)
 218			return -ENOMEM;
 219
 220		file->private_data = dfi;
 221		fi = &dfi->file_info;
 222		dfi->next_offset = 2;
 223		dfi->readdir_cache_idx = -1;
 224	} else {
 225		fi = kmem_cache_zalloc(ceph_file_cachep, GFP_KERNEL);
 226		if (!fi)
 227			return -ENOMEM;
 228
 229		if (opt->flags & CEPH_MOUNT_OPT_NOPAGECACHE)
 230			fi->flags |= CEPH_F_SYNC;
 231
 232		file->private_data = fi;
 233	}
 234
 235	ceph_get_fmode(ci, fmode, 1);
 236	fi->fmode = fmode;
 237
 238	spin_lock_init(&fi->rw_contexts_lock);
 239	INIT_LIST_HEAD(&fi->rw_contexts);
 240	fi->filp_gen = READ_ONCE(ceph_inode_to_fs_client(inode)->filp_gen);
 241
 242	if ((file->f_mode & FMODE_WRITE) && ceph_has_inline_data(ci)) {
 243		ret = ceph_uninline_data(file);
 244		if (ret < 0)
 245			goto error;
 246	}
 247
 248	return 0;
 249
 250error:
 251	ceph_fscache_unuse_cookie(inode, file->f_mode & FMODE_WRITE);
 252	ceph_put_fmode(ci, fi->fmode, 1);
 253	kmem_cache_free(ceph_file_cachep, fi);
 254	/* wake up anyone waiting for caps on this inode */
 255	wake_up_all(&ci->i_cap_wq);
 256	return ret;
 257}
 258
 259/*
 260 * initialize private struct file data.
 261 * if we fail, clean up by dropping fmode reference on the ceph_inode
 262 */
 263static int ceph_init_file(struct inode *inode, struct file *file, int fmode)
 264{
 265	struct ceph_client *cl = ceph_inode_to_client(inode);
 266	int ret = 0;
 267
 268	switch (inode->i_mode & S_IFMT) {
 269	case S_IFREG:
 270		ceph_fscache_use_cookie(inode, file->f_mode & FMODE_WRITE);
 271		fallthrough;
 272	case S_IFDIR:
 273		ret = ceph_init_file_info(inode, file, fmode,
 274						S_ISDIR(inode->i_mode));
 275		break;
 276
 277	case S_IFLNK:
 278		doutc(cl, "%p %llx.%llx %p 0%o (symlink)\n", inode,
 279		      ceph_vinop(inode), file, inode->i_mode);
 280		break;
 281
 282	default:
 283		doutc(cl, "%p %llx.%llx %p 0%o (special)\n", inode,
 284		      ceph_vinop(inode), file, inode->i_mode);
 285		/*
 286		 * we need to drop the open ref now, since we don't
 287		 * have .release set to ceph_release.
 288		 */
 289		BUG_ON(inode->i_fop->release == ceph_release);
 290
 291		/* call the proper open fop */
 292		ret = inode->i_fop->open(inode, file);
 293	}
 294	return ret;
 295}
 296
 297/*
 298 * try renew caps after session gets killed.
 299 */
 300int ceph_renew_caps(struct inode *inode, int fmode)
 301{
 302	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
 303	struct ceph_client *cl = mdsc->fsc->client;
 304	struct ceph_inode_info *ci = ceph_inode(inode);
 305	struct ceph_mds_request *req;
 306	int err, flags, wanted;
 307
 308	spin_lock(&ci->i_ceph_lock);
 309	__ceph_touch_fmode(ci, mdsc, fmode);
 310	wanted = __ceph_caps_file_wanted(ci);
 311	if (__ceph_is_any_real_caps(ci) &&
 312	    (!(wanted & CEPH_CAP_ANY_WR) || ci->i_auth_cap)) {
 313		int issued = __ceph_caps_issued(ci, NULL);
 314		spin_unlock(&ci->i_ceph_lock);
 315		doutc(cl, "%p %llx.%llx want %s issued %s updating mds_wanted\n",
 316		      inode, ceph_vinop(inode), ceph_cap_string(wanted),
 317		      ceph_cap_string(issued));
 318		ceph_check_caps(ci, 0);
 319		return 0;
 320	}
 321	spin_unlock(&ci->i_ceph_lock);
 322
 323	flags = 0;
 324	if ((wanted & CEPH_CAP_FILE_RD) && (wanted & CEPH_CAP_FILE_WR))
 325		flags = O_RDWR;
 326	else if (wanted & CEPH_CAP_FILE_RD)
 327		flags = O_RDONLY;
 328	else if (wanted & CEPH_CAP_FILE_WR)
 329		flags = O_WRONLY;
 330#ifdef O_LAZY
 331	if (wanted & CEPH_CAP_FILE_LAZYIO)
 332		flags |= O_LAZY;
 333#endif
 334
 335	req = prepare_open_request(inode->i_sb, flags, 0);
 336	if (IS_ERR(req)) {
 337		err = PTR_ERR(req);
 338		goto out;
 339	}
 340
 341	req->r_inode = inode;
 342	ihold(inode);
 343	req->r_num_caps = 1;
 344
 345	err = ceph_mdsc_do_request(mdsc, NULL, req);
 346	ceph_mdsc_put_request(req);
 347out:
 348	doutc(cl, "%p %llx.%llx open result=%d\n", inode, ceph_vinop(inode),
 349	      err);
 350	return err < 0 ? err : 0;
 351}
 352
 353/*
 354 * If we already have the requisite capabilities, we can satisfy
 355 * the open request locally (no need to request new caps from the
 356 * MDS).  We do, however, need to inform the MDS (asynchronously)
 357 * if our wanted caps set expands.
 358 */
 359int ceph_open(struct inode *inode, struct file *file)
 360{
 361	struct ceph_inode_info *ci = ceph_inode(inode);
 362	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(inode->i_sb);
 363	struct ceph_client *cl = fsc->client;
 364	struct ceph_mds_client *mdsc = fsc->mdsc;
 365	struct ceph_mds_request *req;
 366	struct ceph_file_info *fi = file->private_data;
 367	int err;
 368	int flags, fmode, wanted;
 369	struct dentry *dentry;
 370	char *path;
 371	bool do_sync = false;
 372	int mask = MAY_READ;
 373
 374	if (fi) {
 375		doutc(cl, "file %p is already opened\n", file);
 376		return 0;
 377	}
 378
 379	/* filter out O_CREAT|O_EXCL; vfs did that already.  yuck. */
 380	flags = file->f_flags & ~(O_CREAT|O_EXCL);
 381	if (S_ISDIR(inode->i_mode)) {
 382		flags = O_DIRECTORY;  /* mds likes to know */
 383	} else if (S_ISREG(inode->i_mode)) {
 384		err = fscrypt_file_open(inode, file);
 385		if (err)
 386			return err;
 387	}
 388
 389	doutc(cl, "%p %llx.%llx file %p flags %d (%d)\n", inode,
 390	      ceph_vinop(inode), file, flags, file->f_flags);
 391	fmode = ceph_flags_to_mode(flags);
 392	wanted = ceph_caps_for_mode(fmode);
 393
 394	if (fmode & CEPH_FILE_MODE_WR)
 395		mask |= MAY_WRITE;
 396	dentry = d_find_alias(inode);
 397	if (!dentry) {
 398		do_sync = true;
 399	} else {
 400		struct ceph_path_info path_info = {0};
 401		path = ceph_mdsc_build_path(mdsc, dentry, &path_info, 0);
 402		if (IS_ERR(path)) {
 403			do_sync = true;
 404			err = 0;
 405		} else {
 406			err = ceph_mds_check_access(mdsc, path, mask);
 407		}
 408		ceph_mdsc_free_path_info(&path_info);
 409		dput(dentry);
 410
 411		/* For none EACCES cases will let the MDS do the mds auth check */
 412		if (err == -EACCES) {
 413			return err;
 414		} else if (err < 0) {
 415			do_sync = true;
 416			err = 0;
 417		}
 418	}
 419
 420	/* snapped files are read-only */
 421	if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE))
 422		return -EROFS;
 423
 424	/* trivially open snapdir */
 425	if (ceph_snap(inode) == CEPH_SNAPDIR) {
 426		return ceph_init_file(inode, file, fmode);
 427	}
 428
 429	/*
 430	 * No need to block if we have caps on the auth MDS (for
 431	 * write) or any MDS (for read).  Update wanted set
 432	 * asynchronously.
 433	 */
 434	spin_lock(&ci->i_ceph_lock);
 435	if (!do_sync && __ceph_is_any_real_caps(ci) &&
 436	    (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) {
 437		int mds_wanted = __ceph_caps_mds_wanted(ci, true);
 438		int issued = __ceph_caps_issued(ci, NULL);
 439
 440		doutc(cl, "open %p fmode %d want %s issued %s using existing\n",
 441		      inode, fmode, ceph_cap_string(wanted),
 442		      ceph_cap_string(issued));
 443		__ceph_touch_fmode(ci, mdsc, fmode);
 444		spin_unlock(&ci->i_ceph_lock);
 445
 446		/* adjust wanted? */
 447		if ((issued & wanted) != wanted &&
 448		    (mds_wanted & wanted) != wanted &&
 449		    ceph_snap(inode) != CEPH_SNAPDIR)
 450			ceph_check_caps(ci, 0);
 451
 452		return ceph_init_file(inode, file, fmode);
 453	} else if (!do_sync && ceph_snap(inode) != CEPH_NOSNAP &&
 454		   (ci->i_snap_caps & wanted) == wanted) {
 455		__ceph_touch_fmode(ci, mdsc, fmode);
 456		spin_unlock(&ci->i_ceph_lock);
 457		return ceph_init_file(inode, file, fmode);
 458	}
 459
 460	spin_unlock(&ci->i_ceph_lock);
 461
 462	doutc(cl, "open fmode %d wants %s\n", fmode, ceph_cap_string(wanted));
 463	req = prepare_open_request(inode->i_sb, flags, 0);
 464	if (IS_ERR(req)) {
 465		err = PTR_ERR(req);
 466		goto out;
 467	}
 468	req->r_inode = inode;
 469	ihold(inode);
 470
 471	req->r_num_caps = 1;
 472	err = ceph_mdsc_do_request(mdsc, NULL, req);
 473	if (!err)
 474		err = ceph_init_file(inode, file, req->r_fmode);
 475	ceph_mdsc_put_request(req);
 476	doutc(cl, "open result=%d on %llx.%llx\n", err, ceph_vinop(inode));
 477out:
 478	return err;
 479}
 480
 481/* Clone the layout from a synchronous create, if the dir now has Dc caps */
 482static void
 483cache_file_layout(struct inode *dst, struct inode *src)
 484{
 485	struct ceph_inode_info *cdst = ceph_inode(dst);
 486	struct ceph_inode_info *csrc = ceph_inode(src);
 487
 488	spin_lock(&cdst->i_ceph_lock);
 489	if ((__ceph_caps_issued(cdst, NULL) & CEPH_CAP_DIR_CREATE) &&
 490	    !ceph_file_layout_is_valid(&cdst->i_cached_layout)) {
 491		memcpy(&cdst->i_cached_layout, &csrc->i_layout,
 492			sizeof(cdst->i_cached_layout));
 493		rcu_assign_pointer(cdst->i_cached_layout.pool_ns,
 494				   ceph_try_get_string(csrc->i_layout.pool_ns));
 495	}
 496	spin_unlock(&cdst->i_ceph_lock);
 497}
 498
 499/*
 500 * Try to set up an async create. We need caps, a file layout, and inode number,
 501 * and either a lease on the dentry or complete dir info. If any of those
 502 * criteria are not satisfied, then return false and the caller can go
 503 * synchronous.
 504 */
 505static int try_prep_async_create(struct inode *dir, struct dentry *dentry,
 506				 struct ceph_file_layout *lo, u64 *pino)
 507{
 508	struct ceph_inode_info *ci = ceph_inode(dir);
 509	struct ceph_dentry_info *di = ceph_dentry(dentry);
 510	int got = 0, want = CEPH_CAP_FILE_EXCL | CEPH_CAP_DIR_CREATE;
 511	u64 ino;
 512
 513	spin_lock(&ci->i_ceph_lock);
 514	/* No auth cap means no chance for Dc caps */
 515	if (!ci->i_auth_cap)
 516		goto no_async;
 517
 518	/* Any delegated inos? */
 519	if (xa_empty(&ci->i_auth_cap->session->s_delegated_inos))
 520		goto no_async;
 521
 522	if (!ceph_file_layout_is_valid(&ci->i_cached_layout))
 523		goto no_async;
 524
 525	if ((__ceph_caps_issued(ci, NULL) & want) != want)
 526		goto no_async;
 527
 528	if (d_in_lookup(dentry)) {
 529		if (!__ceph_dir_is_complete(ci))
 530			goto no_async;
 531		spin_lock(&dentry->d_lock);
 532		di->lease_shared_gen = atomic_read(&ci->i_shared_gen);
 533		spin_unlock(&dentry->d_lock);
 534	} else if (atomic_read(&ci->i_shared_gen) !=
 535		   READ_ONCE(di->lease_shared_gen)) {
 536		goto no_async;
 537	}
 538
 539	ino = ceph_get_deleg_ino(ci->i_auth_cap->session);
 540	if (!ino)
 541		goto no_async;
 542
 543	*pino = ino;
 544	ceph_take_cap_refs(ci, want, false);
 545	memcpy(lo, &ci->i_cached_layout, sizeof(*lo));
 546	rcu_assign_pointer(lo->pool_ns,
 547			   ceph_try_get_string(ci->i_cached_layout.pool_ns));
 548	got = want;
 549no_async:
 550	spin_unlock(&ci->i_ceph_lock);
 551	return got;
 552}
 553
 554static void restore_deleg_ino(struct inode *dir, u64 ino)
 555{
 556	struct ceph_client *cl = ceph_inode_to_client(dir);
 557	struct ceph_inode_info *ci = ceph_inode(dir);
 558	struct ceph_mds_session *s = NULL;
 559
 560	spin_lock(&ci->i_ceph_lock);
 561	if (ci->i_auth_cap)
 562		s = ceph_get_mds_session(ci->i_auth_cap->session);
 563	spin_unlock(&ci->i_ceph_lock);
 564	if (s) {
 565		int err = ceph_restore_deleg_ino(s, ino);
 566		if (err)
 567			pr_warn_client(cl,
 568				"unable to restore delegated ino 0x%llx to session: %d\n",
 569				ino, err);
 570		ceph_put_mds_session(s);
 571	}
 572}
 573
 574static void wake_async_create_waiters(struct inode *inode,
 575				      struct ceph_mds_session *session)
 576{
 577	struct ceph_inode_info *ci = ceph_inode(inode);
 578	bool check_cap = false;
 579
 580	spin_lock(&ci->i_ceph_lock);
 581	if (ci->i_ceph_flags & CEPH_I_ASYNC_CREATE) {
 582		clear_and_wake_up_bit(CEPH_ASYNC_CREATE_BIT, &ci->i_ceph_flags);
 583
 584		if (ci->i_ceph_flags & CEPH_I_ASYNC_CHECK_CAPS) {
 585			ci->i_ceph_flags &= ~CEPH_I_ASYNC_CHECK_CAPS;
 586			check_cap = true;
 587		}
 588	}
 589	ceph_kick_flushing_inode_caps(session, ci);
 590	spin_unlock(&ci->i_ceph_lock);
 591
 592	if (check_cap)
 593		ceph_check_caps(ci, CHECK_CAPS_FLUSH);
 594}
 595
 596static void ceph_async_create_cb(struct ceph_mds_client *mdsc,
 597                                 struct ceph_mds_request *req)
 598{
 599	struct ceph_client *cl = mdsc->fsc->client;
 600	struct dentry *dentry = req->r_dentry;
 601	struct inode *dinode = d_inode(dentry);
 602	struct inode *tinode = req->r_target_inode;
 603	int result = req->r_err ? req->r_err :
 604			le32_to_cpu(req->r_reply_info.head->result);
 605
 606	WARN_ON_ONCE(dinode && tinode && dinode != tinode);
 607
 608	/* MDS changed -- caller must resubmit */
 609	if (result == -EJUKEBOX)
 610		goto out;
 611
 612	mapping_set_error(req->r_parent->i_mapping, result);
 613
 614	if (result) {
 615		struct ceph_path_info path_info = {0};
 616		char *path = ceph_mdsc_build_path(mdsc, req->r_dentry, &path_info, 0);
 617
 618		pr_warn_client(cl,
 619			"async create failure path=(%llx)%s result=%d!\n",
 620			path_info.vino.ino, IS_ERR(path) ? "<<bad>>" : path, result);
 621		ceph_mdsc_free_path_info(&path_info);
 622
 623		ceph_dir_clear_complete(req->r_parent);
 624		if (!d_unhashed(dentry))
 625			d_drop(dentry);
 626
 627		if (dinode) {
 628			mapping_set_error(dinode->i_mapping, result);
 629			ceph_inode_shutdown(dinode);
 630			wake_async_create_waiters(dinode, req->r_session);
 631		}
 632	}
 633
 634	if (tinode) {
 635		u64 ino = ceph_vino(tinode).ino;
 636
 637		if (req->r_deleg_ino != ino)
 638			pr_warn_client(cl,
 639				"inode number mismatch! err=%d deleg_ino=0x%llx target=0x%llx\n",
 640				req->r_err, req->r_deleg_ino, ino);
 641
 642		mapping_set_error(tinode->i_mapping, result);
 643		wake_async_create_waiters(tinode, req->r_session);
 644	} else if (!result) {
 645		pr_warn_client(cl, "no req->r_target_inode for 0x%llx\n",
 646			       req->r_deleg_ino);
 647	}
 648out:
 649	ceph_mdsc_release_dir_caps(req);
 650}
 651
 652static int ceph_finish_async_create(struct inode *dir, struct inode *inode,
 653				    struct dentry *dentry,
 654				    struct file *file, umode_t mode,
 655				    struct ceph_mds_request *req,
 656				    struct ceph_acl_sec_ctx *as_ctx,
 657				    struct ceph_file_layout *lo)
 658{
 659	int ret;
 660	char xattr_buf[4];
 661	struct ceph_mds_reply_inode in = { };
 662	struct ceph_mds_reply_info_in iinfo = { .in = &in };
 663	struct ceph_inode_info *ci = ceph_inode(dir);
 664	struct ceph_dentry_info *di = ceph_dentry(dentry);
 665	struct timespec64 now;
 666	struct ceph_string *pool_ns;
 667	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
 668	struct ceph_client *cl = mdsc->fsc->client;
 669	struct ceph_vino vino = { .ino = req->r_deleg_ino,
 670				  .snap = CEPH_NOSNAP };
 671
 672	ktime_get_real_ts64(&now);
 673
 674	iinfo.inline_version = CEPH_INLINE_NONE;
 675	iinfo.change_attr = 1;
 676	ceph_encode_timespec64(&iinfo.btime, &now);
 677
 678	if (req->r_pagelist) {
 679		iinfo.xattr_len = req->r_pagelist->length;
 680		iinfo.xattr_data = req->r_pagelist->mapped_tail;
 681	} else {
 682		/* fake it */
 683		iinfo.xattr_len = ARRAY_SIZE(xattr_buf);
 684		iinfo.xattr_data = xattr_buf;
 685		memset(iinfo.xattr_data, 0, iinfo.xattr_len);
 686	}
 687
 688	in.ino = cpu_to_le64(vino.ino);
 689	in.snapid = cpu_to_le64(CEPH_NOSNAP);
 690	in.version = cpu_to_le64(1);	// ???
 691	in.cap.caps = in.cap.wanted = cpu_to_le32(CEPH_CAP_ALL_FILE);
 692	in.cap.cap_id = cpu_to_le64(1);
 693	in.cap.realm = cpu_to_le64(ci->i_snap_realm->ino);
 694	in.cap.flags = CEPH_CAP_FLAG_AUTH;
 695	in.ctime = in.mtime = in.atime = iinfo.btime;
 696	in.truncate_seq = cpu_to_le32(1);
 697	in.truncate_size = cpu_to_le64(-1ULL);
 698	in.xattr_version = cpu_to_le64(1);
 699	in.uid = cpu_to_le32(from_kuid(&init_user_ns,
 700				       mapped_fsuid(req->r_mnt_idmap,
 701						    &init_user_ns)));
 702	if (dir->i_mode & S_ISGID) {
 703		in.gid = cpu_to_le32(from_kgid(&init_user_ns, dir->i_gid));
 704
 705		/* Directories always inherit the setgid bit. */
 706		if (S_ISDIR(mode))
 707			mode |= S_ISGID;
 708	} else {
 709		in.gid = cpu_to_le32(from_kgid(&init_user_ns,
 710				     mapped_fsgid(req->r_mnt_idmap,
 711						  &init_user_ns)));
 712	}
 713	in.mode = cpu_to_le32((u32)mode);
 714
 715	in.nlink = cpu_to_le32(1);
 716	in.max_size = cpu_to_le64(lo->stripe_unit);
 717
 718	ceph_file_layout_to_legacy(lo, &in.layout);
 719	/* lo is private, so pool_ns can't change */
 720	pool_ns = rcu_dereference_raw(lo->pool_ns);
 721	if (pool_ns) {
 722		iinfo.pool_ns_len = pool_ns->len;
 723		iinfo.pool_ns_data = pool_ns->str;
 724	}
 725
 726	down_read(&mdsc->snap_rwsem);
 727	ret = ceph_fill_inode(inode, NULL, &iinfo, NULL, req->r_session,
 728			      req->r_fmode, NULL);
 729	up_read(&mdsc->snap_rwsem);
 730	if (ret) {
 731		doutc(cl, "failed to fill inode: %d\n", ret);
 732		ceph_dir_clear_complete(dir);
 733		if (!d_unhashed(dentry))
 734			d_drop(dentry);
 735		discard_new_inode(inode);
 736	} else {
 737		struct dentry *dn;
 738
 739		doutc(cl, "d_adding new inode 0x%llx to 0x%llx/%s\n",
 740		      vino.ino, ceph_ino(dir), dentry->d_name.name);
 741		ceph_dir_clear_ordered(dir);
 742		ceph_init_inode_acls(inode, as_ctx);
 743		if (inode_state_read_once(inode) & I_NEW) {
 744			/*
 745			 * If it's not I_NEW, then someone created this before
 746			 * we got here. Assume the server is aware of it at
 747			 * that point and don't worry about setting
 748			 * CEPH_I_ASYNC_CREATE.
 749			 */
 750			ceph_inode(inode)->i_ceph_flags = CEPH_I_ASYNC_CREATE;
 751			unlock_new_inode(inode);
 752		}
 753		if (d_in_lookup(dentry) || d_really_is_negative(dentry)) {
 754			if (!d_unhashed(dentry))
 755				d_drop(dentry);
 756			dn = d_splice_alias(inode, dentry);
 757			WARN_ON_ONCE(dn && dn != dentry);
 758		}
 759		file->f_mode |= FMODE_CREATED;
 760		ret = finish_open(file, dentry, ceph_open);
 761	}
 762
 763	spin_lock(&dentry->d_lock);
 764	clear_and_wake_up_bit(CEPH_DENTRY_ASYNC_CREATE_BIT, &di->flags);
 765	spin_unlock(&dentry->d_lock);
 766
 767	return ret;
 768}
 769
 770/*
 771 * Do a lookup + open with a single request.  If we get a non-existent
 772 * file or symlink, return 1 so the VFS can retry.
 773 */
 774int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
 775		     struct file *file, unsigned flags, umode_t mode)
 776{
 777	struct mnt_idmap *idmap = file_mnt_idmap(file);
 778	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(dir->i_sb);
 779	struct ceph_client *cl = fsc->client;
 780	struct ceph_mds_client *mdsc = fsc->mdsc;
 781	struct ceph_mds_request *req;
 782	struct inode *new_inode = NULL;
 783	struct dentry *dn;
 784	struct ceph_acl_sec_ctx as_ctx = {};
 785	bool try_async = ceph_test_mount_opt(fsc, ASYNC_DIROPS);
 786	int mask;
 787	int err;
 788	char *path;
 789
 790	doutc(cl, "%p %llx.%llx dentry %p '%pd' %s flags %d mode 0%o\n",
 791	      dir, ceph_vinop(dir), dentry, dentry,
 792	      d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode);
 793
 794	if (dentry->d_name.len > NAME_MAX)
 795		return -ENAMETOOLONG;
 796
 797	err = ceph_wait_on_conflict_unlink(dentry);
 798	if (err)
 799		return err;
 800	/*
 801	 * Do not truncate the file, since atomic_open is called before the
 802	 * permission check. The caller will do the truncation afterward.
 803	 */
 804	flags &= ~O_TRUNC;
 805
 806	dn = d_find_alias(dir);
 807	if (!dn) {
 808		try_async = false;
 809	} else {
 810		struct ceph_path_info path_info = {0};
 811		path = ceph_mdsc_build_path(mdsc, dn, &path_info, 0);
 812		if (IS_ERR(path)) {
 813			try_async = false;
 814			err = 0;
 815		} else {
 816			int fmode = ceph_flags_to_mode(flags);
 817
 818			mask = MAY_READ;
 819			if (fmode & CEPH_FILE_MODE_WR)
 820				mask |= MAY_WRITE;
 821			err = ceph_mds_check_access(mdsc, path, mask);
 822		}
 823		ceph_mdsc_free_path_info(&path_info);
 824		dput(dn);
 825
 826		/* For none EACCES cases will let the MDS do the mds auth check */
 827		if (err == -EACCES) {
 828			return err;
 829		} else if (err < 0) {
 830			try_async = false;
 831			err = 0;
 832		}
 833	}
 834
 835retry:
 836	if (flags & O_CREAT) {
 837		if (ceph_quota_is_max_files_exceeded(dir))
 838			return -EDQUOT;
 839
 840		new_inode = ceph_new_inode(dir, dentry, &mode, &as_ctx);
 841		if (IS_ERR(new_inode)) {
 842			err = PTR_ERR(new_inode);
 843			goto out_ctx;
 844		}
 845		/* Async create can't handle more than a page of xattrs */
 846		if (as_ctx.pagelist &&
 847		    !list_is_singular(&as_ctx.pagelist->head))
 848			try_async = false;
 849	} else if (!d_in_lookup(dentry)) {
 850		/* If it's not being looked up, it's negative */
 851		return -ENOENT;
 852	}
 853
 854	/* do the open */
 855	req = prepare_open_request(dir->i_sb, flags, mode);
 856	if (IS_ERR(req)) {
 857		err = PTR_ERR(req);
 858		goto out_ctx;
 859	}
 860	req->r_dentry = dget(dentry);
 861	req->r_num_caps = 2;
 862	mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
 863	if (ceph_security_xattr_wanted(dir))
 864		mask |= CEPH_CAP_XATTR_SHARED;
 865	req->r_args.open.mask = cpu_to_le32(mask);
 866	req->r_parent = dir;
 867	if (req->r_op == CEPH_MDS_OP_CREATE)
 868		req->r_mnt_idmap = mnt_idmap_get(idmap);
 869	ihold(dir);
 870	if (IS_ENCRYPTED(dir)) {
 871		set_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags);
 872		err = fscrypt_prepare_lookup_partial(dir, dentry);
 873		if (err < 0)
 874			goto out_req;
 875	}
 876
 877	if (flags & O_CREAT) {
 878		struct ceph_file_layout lo;
 879
 880		req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL |
 881				     CEPH_CAP_XATTR_EXCL;
 882		req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 883
 884		ceph_as_ctx_to_req(req, &as_ctx);
 885
 886		if (try_async && (req->r_dir_caps =
 887				  try_prep_async_create(dir, dentry, &lo,
 888							&req->r_deleg_ino))) {
 889			struct ceph_vino vino = { .ino = req->r_deleg_ino,
 890						  .snap = CEPH_NOSNAP };
 891			struct ceph_dentry_info *di = ceph_dentry(dentry);
 892
 893			set_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags);
 894			req->r_args.open.flags |= cpu_to_le32(CEPH_O_EXCL);
 895			req->r_callback = ceph_async_create_cb;
 896
 897			/* Hash inode before RPC */
 898			new_inode = ceph_get_inode(dir->i_sb, vino, new_inode);
 899			if (IS_ERR(new_inode)) {
 900				err = PTR_ERR(new_inode);
 901				new_inode = NULL;
 902				goto out_req;
 903			}
 904			WARN_ON_ONCE(!(inode_state_read_once(new_inode) & I_NEW));
 905
 906			spin_lock(&dentry->d_lock);
 907			di->flags |= CEPH_DENTRY_ASYNC_CREATE;
 908			spin_unlock(&dentry->d_lock);
 909
 910			err = ceph_mdsc_submit_request(mdsc, dir, req);
 911			if (!err) {
 912				err = ceph_finish_async_create(dir, new_inode,
 913							       dentry, file,
 914							       mode, req,
 915							       &as_ctx, &lo);
 916				new_inode = NULL;
 917			} else if (err == -EJUKEBOX) {
 918				restore_deleg_ino(dir, req->r_deleg_ino);
 919				ceph_mdsc_put_request(req);
 920				discard_new_inode(new_inode);
 921				ceph_release_acl_sec_ctx(&as_ctx);
 922				memset(&as_ctx, 0, sizeof(as_ctx));
 923				new_inode = NULL;
 924				try_async = false;
 925				ceph_put_string(rcu_dereference_raw(lo.pool_ns));
 926				goto retry;
 927			}
 928			ceph_put_string(rcu_dereference_raw(lo.pool_ns));
 929			goto out_req;
 930		}
 931	}
 932
 933	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 934	req->r_new_inode = new_inode;
 935	new_inode = NULL;
 936	err = ceph_mdsc_do_request(mdsc, (flags & O_CREAT) ? dir : NULL, req);
 937	if (err == -ENOENT) {
 938		dentry = ceph_handle_snapdir(req, dentry);
 939		if (IS_ERR(dentry)) {
 940			err = PTR_ERR(dentry);
 941			goto out_req;
 942		}
 943		err = 0;
 944	}
 945
 946	if (!err && (flags & O_CREAT) && !req->r_reply_info.head->is_dentry)
 947		err = ceph_handle_notrace_create(dir, dentry);
 948
 949	if (d_in_lookup(dentry)) {
 950		dn = ceph_finish_lookup(req, dentry, err);
 951		if (IS_ERR(dn))
 952			err = PTR_ERR(dn);
 953	} else {
 954		/* we were given a hashed negative dentry */
 955		dn = NULL;
 956	}
 957	if (err)
 958		goto out_req;
 959	if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) {
 960		/* make vfs retry on splice, ENOENT, or symlink */
 961		doutc(cl, "finish_no_open on dn %p\n", dn);
 962		err = finish_no_open(file, dn);
 963	} else {
 964		if (IS_ENCRYPTED(dir) &&
 965		    !fscrypt_has_permitted_context(dir, d_inode(dentry))) {
 966			pr_warn_client(cl,
 967				"Inconsistent encryption context (parent %llx:%llx child %llx:%llx)\n",
 968				ceph_vinop(dir), ceph_vinop(d_inode(dentry)));
 969			goto out_req;
 970		}
 971
 972		doutc(cl, "finish_open on dn %p\n", dn);
 973		if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) {
 974			struct inode *newino = d_inode(dentry);
 975
 976			cache_file_layout(dir, newino);
 977			ceph_init_inode_acls(newino, &as_ctx);
 978			file->f_mode |= FMODE_CREATED;
 979		}
 980		err = finish_open(file, dentry, ceph_open);
 981	}
 982out_req:
 983	ceph_mdsc_put_request(req);
 984	iput(new_inode);
 985out_ctx:
 986	ceph_release_acl_sec_ctx(&as_ctx);
 987	doutc(cl, "result=%d\n", err);
 988	return err;
 989}
 990
 991int ceph_release(struct inode *inode, struct file *file)
 992{
 993	struct ceph_client *cl = ceph_inode_to_client(inode);
 994	struct ceph_inode_info *ci = ceph_inode(inode);
 995
 996	if (S_ISDIR(inode->i_mode)) {
 997		struct ceph_dir_file_info *dfi = file->private_data;
 998		doutc(cl, "%p %llx.%llx dir file %p\n", inode,
 999		      ceph_vinop(inode), file);
1000		WARN_ON(!list_empty(&dfi->file_info.rw_contexts));
1001
1002		ceph_put_fmode(ci, dfi->file_info.fmode, 1);
1003
1004		if (dfi->last_readdir)
1005			ceph_mdsc_put_request(dfi->last_readdir);
1006		kfree(dfi->last_name);
1007		kfree(dfi->dir_info);
1008		kmem_cache_free(ceph_dir_file_cachep, dfi);
1009	} else {
1010		struct ceph_file_info *fi = file->private_data;
1011		doutc(cl, "%p %llx.%llx regular file %p\n", inode,
1012		      ceph_vinop(inode), file);
1013		WARN_ON(!list_empty(&fi->rw_contexts));
1014
1015		ceph_fscache_unuse_cookie(inode, file->f_mode & FMODE_WRITE);
1016		ceph_put_fmode(ci, fi->fmode, 1);
1017
1018		kmem_cache_free(ceph_file_cachep, fi);
1019	}
1020
1021	/* wake up anyone waiting for caps on this inode */
1022	wake_up_all(&ci->i_cap_wq);
1023	return 0;
1024}
1025
1026enum {
1027	HAVE_RETRIED = 1,
1028	CHECK_EOF =    2,
1029	READ_INLINE =  3,
1030};
1031
1032/*
1033 * Completely synchronous read and write methods.  Direct from __user
1034 * buffer to osd, or directly to user pages (if O_DIRECT).
1035 *
1036 * If the read spans object boundary, just do multiple reads.  (That's not
1037 * atomic, but good enough for now.)
1038 *
1039 * If we get a short result from the OSD, check against i_size; we need to
1040 * only return a short read to the caller if we hit EOF.
1041 */
1042ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
1043			 struct iov_iter *to, int *retry_op,
1044			 u64 *last_objver)
1045{
1046	struct ceph_inode_info *ci = ceph_inode(inode);
1047	struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
1048	struct ceph_client *cl = fsc->client;
1049	struct ceph_osd_client *osdc = &fsc->client->osdc;
1050	ssize_t ret;
1051	u64 off = *ki_pos;
1052	u64 len = iov_iter_count(to);
1053	u64 i_size = i_size_read(inode);
1054	bool sparse = IS_ENCRYPTED(inode) || ceph_test_mount_opt(fsc, SPARSEREAD);
1055	u64 objver = 0;
1056
1057	doutc(cl, "on inode %p %llx.%llx %llx~%llx\n", inode,
1058	      ceph_vinop(inode), *ki_pos, len);
1059
1060	if (ceph_inode_is_shutdown(inode))
1061		return -EIO;
1062
1063	if (!len || !i_size)
1064		return 0;
1065	/*
1066	 * flush any page cache pages in this range.  this
1067	 * will make concurrent normal and sync io slow,
1068	 * but it will at least behave sensibly when they are
1069	 * in sequence.
1070	 */
1071	ret = filemap_write_and_wait_range(inode->i_mapping,
1072					   off, off + len - 1);
1073	if (ret < 0)
1074		return ret;
1075
1076	ret = 0;
1077	while ((len = iov_iter_count(to)) > 0) {
1078		struct ceph_osd_request *req;
1079		struct page **pages;
1080		int num_pages;
1081		size_t page_off;
1082		bool more;
1083		int idx = 0;
1084		size_t left;
1085		struct ceph_osd_req_op *op;
1086		u64 read_off = off;
1087		u64 read_len = len;
1088		int extent_cnt;
1089
1090		/* determine new offset/length if encrypted */
1091		ceph_fscrypt_adjust_off_and_len(inode, &read_off, &read_len);
1092
1093		doutc(cl, "orig %llu~%llu reading %llu~%llu", off, len,
1094		      read_off, read_len);
1095
1096		req = ceph_osdc_new_request(osdc, &ci->i_layout,
1097					ci->i_vino, read_off, &read_len, 0, 1,
1098					sparse ? CEPH_OSD_OP_SPARSE_READ :
1099						 CEPH_OSD_OP_READ,
1100					CEPH_OSD_FLAG_READ,
1101					NULL, ci->i_truncate_seq,
1102					ci->i_truncate_size, false);
1103		if (IS_ERR(req)) {
1104			ret = PTR_ERR(req);
1105			break;
1106		}
1107
1108		/* adjust len downward if the request truncated the len */
1109		if (off + len > read_off + read_len)
1110			len = read_off + read_len - off;
1111		more = len < iov_iter_count(to);
1112
1113		op = &req->r_ops[0];
1114		if (sparse) {
1115			extent_cnt = __ceph_sparse_read_ext_count(inode, read_len);
1116			ret = ceph_alloc_sparse_ext_map(op, extent_cnt);
1117			if (ret) {
1118				ceph_osdc_put_request(req);
1119				break;
1120			}
1121		}
1122
1123		num_pages = calc_pages_for(read_off, read_len);
1124		page_off = offset_in_page(off);
1125		pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
1126		if (IS_ERR(pages)) {
1127			ceph_osdc_put_request(req);
1128			ret = PTR_ERR(pages);
1129			break;
1130		}
1131
1132		osd_req_op_extent_osd_data_pages(req, 0, pages, read_len,
1133						 offset_in_page(read_off),
1134						 false, true);
1135
1136		ceph_osdc_start_request(osdc, req);
1137		ret = ceph_osdc_wait_request(osdc, req);
1138
1139		ceph_update_read_metrics(&fsc->mdsc->metric,
1140					 req->r_start_latency,
1141					 req->r_end_latency,
1142					 read_len, ret);
1143
1144		if (ret > 0)
1145			objver = req->r_version;
1146
1147		i_size = i_size_read(inode);
1148		doutc(cl, "%llu~%llu got %zd i_size %llu%s\n", off, len,
1149		      ret, i_size, (more ? " MORE" : ""));
1150
1151		/* Fix it to go to end of extent map */
1152		if (sparse && ret >= 0)
1153			ret = ceph_sparse_ext_map_end(op);
1154		else if (ret == -ENOENT)
1155			ret = 0;
1156
1157		if (ret < 0) {
1158			ceph_osdc_put_request(req);
1159			if (ret == -EBLOCKLISTED)
1160				fsc->blocklisted = true;
1161			break;
1162		}
1163
1164		if (IS_ENCRYPTED(inode)) {
1165			int fret;
1166
1167			fret = ceph_fscrypt_decrypt_extents(inode, pages,
1168					read_off, op->extent.sparse_ext,
1169					op->extent.sparse_ext_cnt);
1170			if (fret < 0) {
1171				ret = fret;
1172				ceph_osdc_put_request(req);
1173				break;
1174			}
1175
1176			/* account for any partial block at the beginning */
1177			fret -= (off - read_off);
1178
1179			/*
1180			 * Short read after big offset adjustment?
1181			 * Nothing is usable, just call it a zero
1182			 * len read.
1183			 */
1184			fret = max(fret, 0);
1185
1186			/* account for partial block at the end */
1187			ret = min_t(ssize_t, fret, len);
1188		}
1189
1190		/* Short read but not EOF? Zero out the remainder. */
1191		if (ret < len && (off + ret < i_size)) {
1192			int zlen = min(len - ret, i_size - off - ret);
1193			int zoff = page_off + ret;
1194
1195			doutc(cl, "zero gap %llu~%llu\n", off + ret,
1196			      off + ret + zlen);
1197			ceph_zero_page_vector_range(zoff, zlen, pages);
1198			ret += zlen;
1199		}
1200
1201		if (off + ret > i_size)
1202			left = (i_size > off) ? i_size - off : 0;
1203		else
1204			left = ret;
1205
1206		while (left > 0) {
1207			size_t plen, copied;
1208
1209			plen = min_t(size_t, left, PAGE_SIZE - page_off);
1210			SetPageUptodate(pages[idx]);
1211			copied = copy_page_to_iter(pages[idx++],
1212						   page_off, plen, to);
1213			off += copied;
1214			left -= copied;
1215			page_off = 0;
1216			if (copied < plen) {
1217				ret = -EFAULT;
1218				break;
1219			}
1220		}
1221
1222		ceph_osdc_put_request(req);
1223
1224		if (off >= i_size || !more)
1225			break;
1226	}
1227
1228	if (ret > 0) {
1229		if (off >= i_size) {
1230			*retry_op = CHECK_EOF;
1231			ret = i_size - *ki_pos;
1232			*ki_pos = i_size;
1233		} else {
1234			ret = off - *ki_pos;
1235			*ki_pos = off;
1236		}
1237
1238		if (last_objver)
1239			*last_objver = objver;
1240	}
1241	doutc(cl, "result %zd retry_op %d\n", ret, *retry_op);
1242	return ret;
1243}
1244
1245static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *to,
1246			      int *retry_op)
1247{
1248	struct file *file = iocb->ki_filp;
1249	struct inode *inode = file_inode(file);
1250	struct ceph_client *cl = ceph_inode_to_client(inode);
1251
1252	doutc(cl, "on file %p %llx~%zx %s\n", file, iocb->ki_pos,
1253	      iov_iter_count(to),
1254	      (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
1255
1256	return __ceph_sync_read(inode, &iocb->ki_pos, to, retry_op, NULL);
1257}
1258
1259struct ceph_aio_request {
1260	struct kiocb *iocb;
1261	size_t total_len;
1262	bool write;
1263	bool should_dirty;
1264	int error;
1265	struct list_head osd_reqs;
1266	unsigned num_reqs;
1267	atomic_t pending_reqs;
1268	struct timespec64 mtime;
1269	struct ceph_cap_flush *prealloc_cf;
1270};
1271
1272struct ceph_aio_work {
1273	struct work_struct work;
1274	struct ceph_osd_request *req;
1275};
1276
1277static void ceph_aio_retry_work(struct work_struct *work);
1278
1279static void ceph_aio_complete(struct inode *inode,
1280			      struct ceph_aio_request *aio_req)
1281{
1282	struct ceph_client *cl = ceph_inode_to_client(inode);
1283	struct ceph_inode_info *ci = ceph_inode(inode);
1284	int ret;
1285
1286	if (!atomic_dec_and_test(&aio_req->pending_reqs))
1287		return;
1288
1289	if (aio_req->iocb->ki_flags & IOCB_DIRECT)
1290		inode_dio_end(inode);
1291
1292	ret = aio_req->error;
1293	if (!ret)
1294		ret = aio_req->total_len;
1295
1296	doutc(cl, "%p %llx.%llx rc %d\n", inode, ceph_vinop(inode), ret);
1297
1298	if (ret >= 0 && aio_req->write) {
1299		int dirty;
1300
1301		loff_t endoff = aio_req->iocb->ki_pos + aio_req->total_len;
1302		if (endoff > i_size_read(inode)) {
1303			if (ceph_inode_set_size(inode, endoff))
1304				ceph_check_caps(ci, CHECK_CAPS_AUTHONLY);
1305		}
1306
1307		spin_lock(&ci->i_ceph_lock);
1308		dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1309					       &aio_req->prealloc_cf);
1310		spin_unlock(&ci->i_ceph_lock);
1311		if (dirty)
1312			__mark_inode_dirty(inode, dirty);
1313
1314	}
1315
1316	ceph_put_cap_refs(ci, (aio_req->write ? CEPH_CAP_FILE_WR :
1317						CEPH_CAP_FILE_RD));
1318
1319	aio_req->iocb->ki_complete(aio_req->iocb, ret);
1320
1321	ceph_free_cap_flush(aio_req->prealloc_cf);
1322	kfree(aio_req);
1323}
1324
1325static void ceph_aio_complete_req(struct ceph_osd_request *req)
1326{
1327	int rc = req->r_result;
1328	struct inode *inode = req->r_inode;
1329	struct ceph_aio_request *aio_req = req->r_priv;
1330	struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0);
1331	struct ceph_osd_req_op *op = &req->r_ops[0];
1332	struct ceph_client_metric *metric = &ceph_sb_to_mdsc(inode->i_sb)->metric;
1333	unsigned int len = osd_data->bvec_pos.iter.bi_size;
1334	bool sparse = (op->op == CEPH_OSD_OP_SPARSE_READ);
1335	struct ceph_client *cl = ceph_inode_to_client(inode);
1336
1337	BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_BVECS);
1338	BUG_ON(!osd_data->num_bvecs);
1339
1340	doutc(cl, "req %p inode %p %llx.%llx, rc %d bytes %u\n", req,
1341	      inode, ceph_vinop(inode), rc, len);
1342
1343	if (rc == -EOLDSNAPC) {
1344		struct ceph_aio_work *aio_work;
1345		BUG_ON(!aio_req->write);
1346
1347		aio_work = kmalloc(sizeof(*aio_work), GFP_NOFS);
1348		if (aio_work) {
1349			INIT_WORK(&aio_work->work, ceph_aio_retry_work);
1350			aio_work->req = req;
1351			queue_work(ceph_inode_to_fs_client(inode)->inode_wq,
1352				   &aio_work->work);
1353			return;
1354		}
1355		rc = -ENOMEM;
1356	} else if (!aio_req->write) {
1357		if (sparse && rc >= 0)
1358			rc = ceph_sparse_ext_map_end(op);
1359		if (rc == -ENOENT)
1360			rc = 0;
1361		if (rc >= 0 && len > rc) {
1362			struct iov_iter i;
1363			int zlen = len - rc;
1364
1365			/*
1366			 * If read is satisfied by single OSD request,
1367			 * it can pass EOF. Otherwise read is within
1368			 * i_size.
1369			 */
1370			if (aio_req->num_reqs == 1) {
1371				loff_t i_size = i_size_read(inode);
1372				loff_t endoff = aio_req->iocb->ki_pos + rc;
1373				if (endoff < i_size)
1374					zlen = min_t(size_t, zlen,
1375						     i_size - endoff);
1376				aio_req->total_len = rc + zlen;
1377			}
1378
1379			iov_iter_bvec(&i, ITER_DEST, osd_data->bvec_pos.bvecs,
1380				      osd_data->num_bvecs, len);
1381			iov_iter_advance(&i, rc);
1382			iov_iter_zero(zlen, &i);
1383		}
1384	}
1385
1386	/* r_start_latency == 0 means the request was not submitted */
1387	if (req->r_start_latency) {
1388		if (aio_req->write)
1389			ceph_update_write_metrics(metric, req->r_start_latency,
1390						  req->r_end_latency, len, rc);
1391		else
1392			ceph_update_read_metrics(metric, req->r_start_latency,
1393						 req->r_end_latency, len, rc);
1394	}
1395
1396	put_bvecs(osd_data->bvec_pos.bvecs, osd_data->num_bvecs,
1397		  aio_req->should_dirty);
1398	ceph_osdc_put_request(req);
1399
1400	if (rc < 0)
1401		cmpxchg(&aio_req->error, 0, rc);
1402
1403	ceph_aio_complete(inode, aio_req);
1404	return;
1405}
1406
1407static void ceph_aio_retry_work(struct work_struct *work)
1408{
1409	struct ceph_aio_work *aio_work =
1410		container_of(work, struct ceph_aio_work, work);
1411	struct ceph_osd_request *orig_req = aio_work->req;
1412	struct ceph_aio_request *aio_req = orig_req->r_priv;
1413	struct inode *inode = orig_req->r_inode;
1414	struct ceph_inode_info *ci = ceph_inode(inode);
1415	struct ceph_snap_context *snapc;
1416	struct ceph_osd_request *req;
1417	int ret;
1418
1419	spin_lock(&ci->i_ceph_lock);
1420	if (__ceph_have_pending_cap_snap(ci)) {
1421		struct ceph_cap_snap *capsnap =
1422			list_last_entry(&ci->i_cap_snaps,
1423					struct ceph_cap_snap,
1424					ci_item);
1425		snapc = ceph_get_snap_context(capsnap->context);
1426	} else {
1427		BUG_ON(!ci->i_head_snapc);
1428		snapc = ceph_get_snap_context(ci->i_head_snapc);
1429	}
1430	spin_unlock(&ci->i_ceph_lock);
1431
1432	req = ceph_osdc_alloc_request(orig_req->r_osdc, snapc, 1,
1433			false, GFP_NOFS);
1434	if (!req) {
1435		ret = -ENOMEM;
1436		req = orig_req;
1437		goto out;
1438	}
1439
1440	req->r_flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE;
1441	ceph_oloc_copy(&req->r_base_oloc, &orig_req->r_base_oloc);
1442	ceph_oid_copy(&req->r_base_oid, &orig_req->r_base_oid);
1443
1444	req->r_ops[0] = orig_req->r_ops[0];
1445
1446	req->r_mtime = aio_req->mtime;
1447	req->r_data_offset = req->r_ops[0].extent.offset;
1448
1449	ret = ceph_osdc_alloc_messages(req, GFP_NOFS);
1450	if (ret) {
1451		ceph_osdc_put_request(req);
1452		req = orig_req;
1453		goto out;
1454	}
1455
1456	ceph_osdc_put_request(orig_req);
1457
1458	req->r_callback = ceph_aio_complete_req;
1459	req->r_inode = inode;
1460	req->r_priv = aio_req;
1461
1462	ceph_osdc_start_request(req->r_osdc, req);
1463out:
1464	if (ret < 0) {
1465		req->r_result = ret;
1466		ceph_aio_complete_req(req);
1467	}
1468
1469	ceph_put_snap_context(snapc);
1470	kfree(aio_work);
1471}
1472
1473static ssize_t
1474ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter,
1475		       struct ceph_snap_context *snapc,
1476		       struct ceph_cap_flush **pcf)
1477{
1478	struct file *file = iocb->ki_filp;
1479	struct inode *inode = file_inode(file);
1480	struct ceph_inode_info *ci = ceph_inode(inode);
1481	struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
1482	struct ceph_client *cl = fsc->client;
1483	struct ceph_client_metric *metric = &fsc->mdsc->metric;
1484	struct ceph_vino vino;
1485	struct ceph_osd_request *req;
1486	struct bio_vec *bvecs;
1487	struct ceph_aio_request *aio_req = NULL;
1488	int num_pages = 0;
1489	int flags;
1490	int ret = 0;
1491	struct timespec64 mtime = current_time(inode);
1492	size_t count = iov_iter_count(iter);
1493	loff_t pos = iocb->ki_pos;
1494	bool write = iov_iter_rw(iter) == WRITE;
1495	bool should_dirty = !write && user_backed_iter(iter);
1496	bool sparse = ceph_test_mount_opt(fsc, SPARSEREAD);
1497
1498	if (write && ceph_snap(file_inode(file)) != CEPH_NOSNAP)
1499		return -EROFS;
1500
1501	doutc(cl, "sync_direct_%s on file %p %lld~%u snapc %p seq %lld\n",
1502	      (write ? "write" : "read"), file, pos, (unsigned)count,
1503	      snapc, snapc ? snapc->seq : 0);
1504
1505	if (write) {
1506		int ret2;
1507
1508		ceph_fscache_invalidate(inode, true);
1509
1510		ret2 = invalidate_inode_pages2_range(inode->i_mapping,
1511					pos >> PAGE_SHIFT,
1512					(pos + count - 1) >> PAGE_SHIFT);
1513		if (ret2 < 0)
1514			doutc(cl, "invalidate_inode_pages2_range returned %d\n",
1515			      ret2);
1516
1517		flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE;
1518	} else {
1519		flags = CEPH_OSD_FLAG_READ;
1520	}
1521
1522	while (iov_iter_count(iter) > 0) {
1523		u64 size = iov_iter_count(iter);
1524		ssize_t len;
1525		struct ceph_osd_req_op *op;
1526		int readop = sparse ? CEPH_OSD_OP_SPARSE_READ : CEPH_OSD_OP_READ;
1527		int extent_cnt;
1528
1529		if (write)
1530			size = min_t(u64, size, fsc->mount_options->wsize);
1531		else
1532			size = min_t(u64, size, fsc->mount_options->rsize);
1533
1534		vino = ceph_vino(inode);
1535		req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1536					    vino, pos, &size, 0,
1537					    1,
1538					    write ? CEPH_OSD_OP_WRITE : readop,
1539					    flags, snapc,
1540					    ci->i_truncate_seq,
1541					    ci->i_truncate_size,
1542					    false);
1543		if (IS_ERR(req)) {
1544			ret = PTR_ERR(req);
1545			break;
1546		}
1547
1548		op = &req->r_ops[0];
1549		if (!write && sparse) {
1550			extent_cnt = __ceph_sparse_read_ext_count(inode, size);
1551			ret = ceph_alloc_sparse_ext_map(op, extent_cnt);
1552			if (ret) {
1553				ceph_osdc_put_request(req);
1554				break;
1555			}
1556		}
1557
1558		len = iter_get_bvecs_alloc(iter, size, &bvecs, &num_pages);
1559		if (len < 0) {
1560			ceph_osdc_put_request(req);
1561			ret = len;
1562			break;
1563		}
1564		if (len != size)
1565			osd_req_op_extent_update(req, 0, len);
1566
1567		osd_req_op_extent_osd_data_bvecs(req, 0, bvecs, num_pages, len);
1568
1569		/*
1570		 * To simplify error handling, allow AIO when IO within i_size
1571		 * or IO can be satisfied by single OSD request.
1572		 */
1573		if (pos == iocb->ki_pos && !is_sync_kiocb(iocb) &&
1574		    (len == count || pos + count <= i_size_read(inode))) {
1575			aio_req = kzalloc(sizeof(*aio_req), GFP_KERNEL);
1576			if (aio_req) {
1577				aio_req->iocb = iocb;
1578				aio_req->write = write;
1579				aio_req->should_dirty = should_dirty;
1580				INIT_LIST_HEAD(&aio_req->osd_reqs);
1581				if (write) {
1582					aio_req->mtime = mtime;
1583					swap(aio_req->prealloc_cf, *pcf);
1584				}
1585			}
1586			/* ignore error */
1587		}
1588
1589		if (write) {
1590			/*
1591			 * throw out any page cache pages in this range. this
1592			 * may block.
1593			 */
1594			truncate_inode_pages_range(inode->i_mapping, pos,
1595						   PAGE_ALIGN(pos + len) - 1);
1596
1597			req->r_mtime = mtime;
1598		}
1599
1600		if (aio_req) {
1601			aio_req->total_len += len;
1602			aio_req->num_reqs++;
1603			atomic_inc(&aio_req->pending_reqs);
1604
1605			req->r_callback = ceph_aio_complete_req;
1606			req->r_inode = inode;
1607			req->r_priv = aio_req;
1608			list_add_tail(&req->r_private_item, &aio_req->osd_reqs);
1609
1610			pos += len;
1611			continue;
1612		}
1613
1614		ceph_osdc_start_request(req->r_osdc, req);
1615		ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1616
1617		if (write)
1618			ceph_update_write_metrics(metric, req->r_start_latency,
1619						  req->r_end_latency, len, ret);
1620		else
1621			ceph_update_read_metrics(metric, req->r_start_latency,
1622						 req->r_end_latency, len, ret);
1623
1624		size = i_size_read(inode);
1625		if (!write) {
1626			if (sparse && ret >= 0)
1627				ret = ceph_sparse_ext_map_end(op);
1628			else if (ret == -ENOENT)
1629				ret = 0;
1630
1631			if (ret >= 0 && ret < len && pos + ret < size) {
1632				struct iov_iter i;
1633				int zlen = min_t(size_t, len - ret,
1634						 size - pos - ret);
1635
1636				iov_iter_bvec(&i, ITER_DEST, bvecs, num_pages, len);
1637				iov_iter_advance(&i, ret);
1638				iov_iter_zero(zlen, &i);
1639				ret += zlen;
1640			}
1641			if (ret >= 0)
1642				len = ret;
1643		}
1644
1645		put_bvecs(bvecs, num_pages, should_dirty);
1646		ceph_osdc_put_request(req);
1647		if (ret < 0)
1648			break;
1649
1650		pos += len;
1651		if (!write && pos >= size)
1652			break;
1653
1654		if (write && pos > size) {
1655			if (ceph_inode_set_size(inode, pos))
1656				ceph_check_caps(ceph_inode(inode),
1657						CHECK_CAPS_AUTHONLY);
1658		}
1659	}
1660
1661	if (aio_req) {
1662		LIST_HEAD(osd_reqs);
1663
1664		if (aio_req->num_reqs == 0) {
1665			kfree(aio_req);
1666			return ret;
1667		}
1668
1669		ceph_get_cap_refs(ci, write ? CEPH_CAP_FILE_WR :
1670					      CEPH_CAP_FILE_RD);
1671
1672		list_splice(&aio_req->osd_reqs, &osd_reqs);
1673		inode_dio_begin(inode);
1674		while (!list_empty(&osd_reqs)) {
1675			req = list_first_entry(&osd_reqs,
1676					       struct ceph_osd_request,
1677					       r_private_item);
1678			list_del_init(&req->r_private_item);
1679			if (ret >= 0)
1680				ceph_osdc_start_request(req->r_osdc, req);
1681			if (ret < 0) {
1682				req->r_result = ret;
1683				ceph_aio_complete_req(req);
1684			}
1685		}
1686		return -EIOCBQUEUED;
1687	}
1688
1689	if (ret != -EOLDSNAPC && pos > iocb->ki_pos) {
1690		ret = pos - iocb->ki_pos;
1691		iocb->ki_pos = pos;
1692	}
1693	return ret;
1694}
1695
1696/*
1697 * Synchronous write, straight from __user pointer or user pages.
1698 *
1699 * If write spans object boundary, just do multiple writes.  (For a
1700 * correct atomic write, we should e.g. take write locks on all
1701 * objects, rollback on failure, etc.)
1702 */
1703static ssize_t
1704ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
1705		struct ceph_snap_context *snapc)
1706{
1707	struct file *file = iocb->ki_filp;
1708	struct inode *inode = file_inode(file);
1709	struct ceph_inode_info *ci = ceph_inode(inode);
1710	struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
1711	struct ceph_client *cl = fsc->client;
1712	struct ceph_osd_client *osdc = &fsc->client->osdc;
1713	struct ceph_osd_request *req;
1714	struct page **pages;
1715	u64 len;
1716	int num_pages;
1717	int written = 0;
1718	int ret;
1719	bool check_caps = false;
1720	struct timespec64 mtime = current_time(inode);
1721	size_t count = iov_iter_count(from);
1722
1723	if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
1724		return -EROFS;
1725
1726	doutc(cl, "on file %p %lld~%u snapc %p seq %lld\n", file, pos,
1727	      (unsigned)count, snapc, snapc->seq);
1728
1729	ret = filemap_write_and_wait_range(inode->i_mapping,
1730					   pos, pos + count - 1);
1731	if (ret < 0)
1732		return ret;
1733
1734	ceph_fscache_invalidate(inode, false);
1735
1736	while ((len = iov_iter_count(from)) > 0) {
1737		size_t left;
1738		int n;
1739		u64 write_pos = pos;
1740		u64 write_len = len;
1741		u64 objnum, objoff;
1742		u32 xlen;
1743		u64 assert_ver = 0;
1744		bool rmw;
1745		bool first, last;
1746		struct iov_iter saved_iter = *from;
1747		size_t off;
1748
1749		ceph_fscrypt_adjust_off_and_len(inode, &write_pos, &write_len);
1750
1751		/* clamp the length to the end of first object */
1752		ceph_calc_file_object_mapping(&ci->i_layout, write_pos,
1753					      write_len, &objnum, &objoff,
1754					      &xlen);
1755		write_len = xlen;
1756
1757		/* adjust len downward if it goes beyond current object */
1758		if (pos + len > write_pos + write_len)
1759			len = write_pos + write_len - pos;
1760
1761		/*
1762		 * If we had to adjust the length or position to align with a
1763		 * crypto block, then we must do a read/modify/write cycle. We
1764		 * use a version assertion to redrive the thing if something
1765		 * changes in between.
1766		 */
1767		first = pos != write_pos;
1768		last = (pos + len) != (write_pos + write_len);
1769		rmw = first || last;
1770
1771		doutc(cl, "ino %llx %lld~%llu adjusted %lld~%llu -- %srmw\n",
1772		      ci->i_vino.ino, pos, len, write_pos, write_len,
1773		      rmw ? "" : "no ");
1774
1775		/*
1776		 * The data is emplaced into the page as it would be if it were
1777		 * in an array of pagecache pages.
1778		 */
1779		num_pages = calc_pages_for(write_pos, write_len);
1780		pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
1781		if (IS_ERR(pages)) {
1782			ret = PTR_ERR(pages);
1783			break;
1784		}
1785
1786		/* Do we need to preload the pages? */
1787		if (rmw) {
1788			u64 first_pos = write_pos;
1789			u64 last_pos = (write_pos + write_len) - CEPH_FSCRYPT_BLOCK_SIZE;
1790			u64 read_len = CEPH_FSCRYPT_BLOCK_SIZE;
1791			struct ceph_osd_req_op *op;
1792
1793			/* We should only need to do this for encrypted inodes */
1794			WARN_ON_ONCE(!IS_ENCRYPTED(inode));
1795
1796			/* No need to do two reads if first and last blocks are same */
1797			if (first && last_pos == first_pos)
1798				last = false;
1799
1800			/*
1801			 * Allocate a read request for one or two extents,
1802			 * depending on how the request was aligned.
1803			 */
1804			req = ceph_osdc_new_request(osdc, &ci->i_layout,
1805					ci->i_vino, first ? first_pos : last_pos,
1806					&read_len, 0, (first && last) ? 2 : 1,
1807					CEPH_OSD_OP_SPARSE_READ, CEPH_OSD_FLAG_READ,
1808					NULL, ci->i_truncate_seq,
1809					ci->i_truncate_size, false);
1810			if (IS_ERR(req)) {
1811				ceph_release_page_vector(pages, num_pages);
1812				ret = PTR_ERR(req);
1813				break;
1814			}
1815
1816			/* Something is misaligned! */
1817			if (read_len != CEPH_FSCRYPT_BLOCK_SIZE) {
1818				ceph_osdc_put_request(req);
1819				ceph_release_page_vector(pages, num_pages);
1820				ret = -EIO;
1821				break;
1822			}
1823
1824			/* Add extent for first block? */
1825			op = &req->r_ops[0];
1826
1827			if (first) {
1828				osd_req_op_extent_osd_data_pages(req, 0, pages,
1829							 CEPH_FSCRYPT_BLOCK_SIZE,
1830							 offset_in_page(first_pos),
1831							 false, false);
1832				/* We only expect a single extent here */
1833				ret = __ceph_alloc_sparse_ext_map(op, 1);
1834				if (ret) {
1835					ceph_osdc_put_request(req);
1836					ceph_release_page_vector(pages, num_pages);
1837					break;
1838				}
1839			}
1840
1841			/* Add extent for last block */
1842			if (last) {
1843				/* Init the other extent if first extent has been used */
1844				if (first) {
1845					op = &req->r_ops[1];
1846					osd_req_op_extent_init(req, 1,
1847							CEPH_OSD_OP_SPARSE_READ,
1848							last_pos, CEPH_FSCRYPT_BLOCK_SIZE,
1849							ci->i_truncate_size,
1850							ci->i_truncate_seq);
1851				}
1852
1853				ret = __ceph_alloc_sparse_ext_map(op, 1);
1854				if (ret) {
1855					ceph_osdc_put_request(req);
1856					ceph_release_page_vector(pages, num_pages);
1857					break;
1858				}
1859
1860				osd_req_op_extent_osd_data_pages(req, first ? 1 : 0,
1861							&pages[num_pages - 1],
1862							CEPH_FSCRYPT_BLOCK_SIZE,
1863							offset_in_page(last_pos),
1864							false, false);
1865			}
1866
1867			ceph_osdc_start_request(osdc, req);
1868			ret = ceph_osdc_wait_request(osdc, req);
1869
1870			/* FIXME: length field is wrong if there are 2 extents */
1871			ceph_update_read_metrics(&fsc->mdsc->metric,
1872						 req->r_start_latency,
1873						 req->r_end_latency,
1874						 read_len, ret);
1875
1876			/* Ok if object is not already present */
1877			if (ret == -ENOENT) {
1878				/*
1879				 * If there is no object, then we can't assert
1880				 * on its version. Set it to 0, and we'll use an
1881				 * exclusive create instead.
1882				 */
1883				ceph_osdc_put_request(req);
1884				ret = 0;
1885
1886				/*
1887				 * zero out the soon-to-be uncopied parts of the
1888				 * first and last pages.
1889				 */
1890				if (first)
1891					zero_user_segment(pages[0], 0,
1892							  offset_in_page(first_pos));
1893				if (last)
1894					zero_user_segment(pages[num_pages - 1],
1895							  offset_in_page(last_pos),
1896							  PAGE_SIZE);
1897			} else {
1898				if (ret < 0) {
1899					ceph_osdc_put_request(req);
1900					ceph_release_page_vector(pages, num_pages);
1901					break;
1902				}
1903
1904				op = &req->r_ops[0];
1905				if (op->extent.sparse_ext_cnt == 0) {
1906					if (first)
1907						zero_user_segment(pages[0], 0,
1908								  offset_in_page(first_pos));
1909					else
1910						zero_user_segment(pages[num_pages - 1],
1911								  offset_in_page(last_pos),
1912								  PAGE_SIZE);
1913				} else if (op->extent.sparse_ext_cnt != 1 ||
1914					   ceph_sparse_ext_map_end(op) !=
1915						CEPH_FSCRYPT_BLOCK_SIZE) {
1916					ret = -EIO;
1917					ceph_osdc_put_request(req);
1918					ceph_release_page_vector(pages, num_pages);
1919					break;
1920				}
1921
1922				if (first && last) {
1923					op = &req->r_ops[1];
1924					if (op->extent.sparse_ext_cnt == 0) {
1925						zero_user_segment(pages[num_pages - 1],
1926								  offset_in_page(last_pos),
1927								  PAGE_SIZE);
1928					} else if (op->extent.sparse_ext_cnt != 1 ||
1929						   ceph_sparse_ext_map_end(op) !=
1930							CEPH_FSCRYPT_BLOCK_SIZE) {
1931						ret = -EIO;
1932						ceph_osdc_put_request(req);
1933						ceph_release_page_vector(pages, num_pages);
1934						break;
1935					}
1936				}
1937
1938				/* Grab assert version. It must be non-zero. */
1939				assert_ver = req->r_version;
1940				WARN_ON_ONCE(ret > 0 && assert_ver == 0);
1941
1942				ceph_osdc_put_request(req);
1943				if (first) {
1944					ret = ceph_fscrypt_decrypt_block_inplace(inode,
1945							pages[0], CEPH_FSCRYPT_BLOCK_SIZE,
1946							offset_in_page(first_pos),
1947							first_pos >> CEPH_FSCRYPT_BLOCK_SHIFT);
1948					if (ret < 0) {
1949						ceph_release_page_vector(pages, num_pages);
1950						break;
1951					}
1952				}
1953				if (last) {
1954					ret = ceph_fscrypt_decrypt_block_inplace(inode,
1955							pages[num_pages - 1],
1956							CEPH_FSCRYPT_BLOCK_SIZE,
1957							offset_in_page(last_pos),
1958							last_pos >> CEPH_FSCRYPT_BLOCK_SHIFT);
1959					if (ret < 0) {
1960						ceph_release_page_vector(pages, num_pages);
1961						break;
1962					}
1963				}
1964			}
1965		}
1966
1967		left = len;
1968		off = offset_in_page(pos);
1969		for (n = 0; n < num_pages; n++) {
1970			size_t plen = min_t(size_t, left, PAGE_SIZE - off);
1971
1972			/* copy the data */
1973			ret = copy_page_from_iter(pages[n], off, plen, from);
1974			if (ret != plen) {
1975				ret = -EFAULT;
1976				break;
1977			}
1978			off = 0;
1979			left -= ret;
1980		}
1981		if (ret < 0) {
1982			doutc(cl, "write failed with %d\n", ret);
1983			ceph_release_page_vector(pages, num_pages);
1984			break;
1985		}
1986
1987		if (IS_ENCRYPTED(inode)) {
1988			ret = ceph_fscrypt_encrypt_pages(inode, pages,
1989							 write_pos, write_len);
1990			if (ret < 0) {
1991				doutc(cl, "encryption failed with %d\n", ret);
1992				ceph_release_page_vector(pages, num_pages);
1993				break;
1994			}
1995		}
1996
1997		req = ceph_osdc_new_request(osdc, &ci->i_layout,
1998					    ci->i_vino, write_pos, &write_len,
1999					    rmw ? 1 : 0, rmw ? 2 : 1,
2000					    CEPH_OSD_OP_WRITE,
2001					    CEPH_OSD_FLAG_WRITE,
2002					    snapc, ci->i_truncate_seq,
2003					    ci->i_truncate_size, false);
2004		if (IS_ERR(req)) {
2005			ret = PTR_ERR(req);
2006			ceph_release_page_vector(pages, num_pages);
2007			break;
2008		}
2009
2010		doutc(cl, "write op %lld~%llu\n", write_pos, write_len);
2011		osd_req_op_extent_osd_data_pages(req, rmw ? 1 : 0, pages, write_len,
2012						 offset_in_page(write_pos), false,
2013						 true);
2014		req->r_inode = inode;
2015		req->r_mtime = mtime;
2016
2017		/* Set up the assertion */
2018		if (rmw) {
2019			/*
2020			 * Set up the assertion. If we don't have a version
2021			 * number, then the object doesn't exist yet. Use an
2022			 * exclusive create instead of a version assertion in
2023			 * that case.
2024			 */
2025			if (assert_ver) {
2026				osd_req_op_init(req, 0, CEPH_OSD_OP_ASSERT_VER, 0);
2027				req->r_ops[0].assert_ver.ver = assert_ver;
2028			} else {
2029				osd_req_op_init(req, 0, CEPH_OSD_OP_CREATE,
2030						CEPH_OSD_OP_FLAG_EXCL);
2031			}
2032		}
2033
2034		ceph_osdc_start_request(osdc, req);
2035		ret = ceph_osdc_wait_request(osdc, req);
2036
2037		ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
2038					  req->r_end_latency, len, ret);
2039		ceph_osdc_put_request(req);
2040		if (ret != 0) {
2041			doutc(cl, "osd write returned %d\n", ret);
2042			/* Version changed! Must re-do the rmw cycle */
2043			if ((assert_ver && (ret == -ERANGE || ret == -EOVERFLOW)) ||
2044			    (!assert_ver && ret == -EEXIST)) {
2045				/* We should only ever see this on a rmw */
2046				WARN_ON_ONCE(!rmw);
2047
2048				/* The version should never go backward */
2049				WARN_ON_ONCE(ret == -EOVERFLOW);
2050
2051				*from = saved_iter;
2052
2053				/* FIXME: limit number of times we loop? */
2054				continue;
2055			}
2056			ceph_set_error_write(ci);
2057			break;
2058		}
2059
2060		ceph_clear_error_write(ci);
2061
2062		/*
2063		 * We successfully wrote to a range of the file. Declare
2064		 * that region of the pagecache invalid.
2065		 */
2066		ret = invalidate_inode_pages2_range(
2067				inode->i_mapping,
2068				pos >> PAGE_SHIFT,
2069				(pos + len - 1) >> PAGE_SHIFT);
2070		if (ret < 0) {
2071			doutc(cl, "invalidate_inode_pages2_range returned %d\n",
2072			      ret);
2073			ret = 0;
2074		}
2075		pos += len;
2076		written += len;
2077		doutc(cl, "written %d\n", written);
2078		if (pos > i_size_read(inode)) {
2079			check_caps = ceph_inode_set_size(inode, pos);
2080			if (check_caps)
2081				ceph_check_caps(ceph_inode(inode),
2082						CHECK_CAPS_AUTHONLY);
2083		}
2084
2085	}
2086
2087	if (ret != -EOLDSNAPC && written > 0) {
2088		ret = written;
2089		iocb->ki_pos = pos;
2090	}
2091	doutc(cl, "returning %d\n", ret);
2092	return ret;
2093}
2094
2095/*
2096 * Wrap generic_file_aio_read with checks for cap bits on the inode.
2097 * Atomically grab references, so that those bits are not released
2098 * back to the MDS mid-read.
2099 *
2100 * Hmm, the sync read case isn't actually async... should it be?
2101 */
2102static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to)
2103{
2104	struct file *filp = iocb->ki_filp;
2105	struct ceph_file_info *fi = filp->private_data;
2106	size_t len = iov_iter_count(to);
2107	struct inode *inode = file_inode(filp);
2108	struct ceph_inode_info *ci = ceph_inode(inode);
2109	bool direct_lock = iocb->ki_flags & IOCB_DIRECT;
2110	struct ceph_client *cl = ceph_inode_to_client(inode);
2111	ssize_t ret;
2112	int want = 0, got = 0;
2113	int retry_op = 0, read = 0;
2114
2115again:
2116	doutc(cl, "%llu~%u trying to get caps on %p %llx.%llx\n",
2117	      iocb->ki_pos, (unsigned)len, inode, ceph_vinop(inode));
2118
2119	if (ceph_inode_is_shutdown(inode))
2120		return -ESTALE;
2121
2122	ret = direct_lock ? ceph_start_io_direct(inode) :
2123			    ceph_start_io_read(inode);
2124	if (ret)
2125		return ret;
2126
2127	if (!(fi->flags & CEPH_F_SYNC) && !direct_lock)
2128		want |= CEPH_CAP_FILE_CACHE;
2129	if (fi->fmode & CEPH_FILE_MODE_LAZY)
2130		want |= CEPH_CAP_FILE_LAZYIO;
2131
2132	ret = ceph_get_caps(filp, CEPH_CAP_FILE_RD, want, -1, &got);
2133	if (ret < 0) {
2134		if (direct_lock)
2135			ceph_end_io_direct(inode);
2136		else
2137			ceph_end_io_read(inode);
2138		return ret;
2139	}
2140
2141	if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 ||
2142	    (iocb->ki_flags & IOCB_DIRECT) ||
2143	    (fi->flags & CEPH_F_SYNC)) {
2144
2145		doutc(cl, "sync %p %llx.%llx %llu~%u got cap refs on %s\n",
2146		      inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
2147		      ceph_cap_string(got));
2148
2149		if (!ceph_has_inline_data(ci)) {
2150			if (!retry_op &&
2151			    (iocb->ki_flags & IOCB_DIRECT) &&
2152			    !IS_ENCRYPTED(inode)) {
2153				ret = ceph_direct_read_write(iocb, to,
2154							     NULL, NULL);
2155				if (ret >= 0 && ret < len)
2156					retry_op = CHECK_EOF;
2157			} else {
2158				ret = ceph_sync_read(iocb, to, &retry_op);
2159			}
2160		} else {
2161			retry_op = READ_INLINE;
2162		}
2163	} else {
2164		CEPH_DEFINE_RW_CONTEXT(rw_ctx, got);
2165		doutc(cl, "async %p %llx.%llx %llu~%u got cap refs on %s\n",
2166		      inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
2167		      ceph_cap_string(got));
2168		ceph_add_rw_context(fi, &rw_ctx);
2169		ret = generic_file_read_iter(iocb, to);
2170		ceph_del_rw_context(fi, &rw_ctx);
2171	}
2172
2173	doutc(cl, "%p %llx.%llx dropping cap refs on %s = %d\n",
2174	      inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
2175	ceph_put_cap_refs(ci, got);
2176
2177	if (direct_lock)
2178		ceph_end_io_direct(inode);
2179	else
2180		ceph_end_io_read(inode);
2181
2182	if (retry_op > HAVE_RETRIED && ret >= 0) {
2183		int statret;
2184		struct page *page = NULL;
2185		loff_t i_size;
2186		int mask = CEPH_STAT_CAP_SIZE;
2187		if (retry_op == READ_INLINE) {
2188			page = __page_cache_alloc(GFP_KERNEL);
2189			if (!page)
2190				return -ENOMEM;
2191
2192			mask = CEPH_STAT_CAP_INLINE_DATA;
2193		}
2194
2195		statret = __ceph_do_getattr(inode, page, mask, !!page);
2196		if (statret < 0) {
2197			if (page)
2198				__free_page(page);
2199			if (statret == -ENODATA) {
2200				BUG_ON(retry_op != READ_INLINE);
2201				goto again;
2202			}
2203			return statret;
2204		}
2205
2206		i_size = i_size_read(inode);
2207		if (retry_op == READ_INLINE) {
2208			BUG_ON(ret > 0 || read > 0);
2209			if (iocb->ki_pos < i_size &&
2210			    iocb->ki_pos < PAGE_SIZE) {
2211				loff_t end = min_t(loff_t, i_size,
2212						   iocb->ki_pos + len);
2213				end = min_t(loff_t, end, PAGE_SIZE);
2214				if (statret < end)
2215					zero_user_segment(page, statret, end);
2216				ret = copy_page_to_iter(page,
2217						iocb->ki_pos & ~PAGE_MASK,
2218						end - iocb->ki_pos, to);
2219				iocb->ki_pos += ret;
2220				read += ret;
2221			}
2222			if (iocb->ki_pos < i_size && read < len) {
2223				size_t zlen = min_t(size_t, len - read,
2224						    i_size - iocb->ki_pos);
2225				ret = iov_iter_zero(zlen, to);
2226				iocb->ki_pos += ret;
2227				read += ret;
2228			}
2229			__free_pages(page, 0);
2230			return read;
2231		}
2232
2233		/* hit EOF or hole? */
2234		if (retry_op == CHECK_EOF && iocb->ki_pos < i_size &&
2235		    ret < len) {
2236			doutc(cl, "may hit hole, ppos %lld < size %lld, reading more\n",
2237			      iocb->ki_pos, i_size);
2238
2239			read += ret;
2240			len -= ret;
2241			retry_op = HAVE_RETRIED;
2242			goto again;
2243		}
2244	}
2245
2246	if (ret >= 0)
2247		ret += read;
2248
2249	return ret;
2250}
2251
2252/*
2253 * Wrap filemap_splice_read with checks for cap bits on the inode.
2254 * Atomically grab references, so that those bits are not released
2255 * back to the MDS mid-read.
2256 */
2257static ssize_t ceph_splice_read(struct file *in, loff_t *ppos,
2258				struct pipe_inode_info *pipe,
2259				size_t len, unsigned int flags)
2260{
2261	struct ceph_file_info *fi = in->private_data;
2262	struct inode *inode = file_inode(in);
2263	struct ceph_inode_info *ci = ceph_inode(inode);
2264	ssize_t ret;
2265	int want = 0, got = 0;
2266	CEPH_DEFINE_RW_CONTEXT(rw_ctx, 0);
2267
2268	dout("splice_read %p %llx.%llx %llu~%zu trying to get caps on %p\n",
2269	     inode, ceph_vinop(inode), *ppos, len, inode);
2270
2271	if (ceph_inode_is_shutdown(inode))
2272		return -ESTALE;
2273
2274	if (ceph_has_inline_data(ci) ||
2275	    (fi->flags & CEPH_F_SYNC))
2276		return copy_splice_read(in, ppos, pipe, len, flags);
2277
2278	ret = ceph_start_io_read(inode);
2279	if (ret)
2280		return ret;
2281
2282	want = CEPH_CAP_FILE_CACHE;
2283	if (fi->fmode & CEPH_FILE_MODE_LAZY)
2284		want |= CEPH_CAP_FILE_LAZYIO;
2285
2286	ret = ceph_get_caps(in, CEPH_CAP_FILE_RD, want, -1, &got);
2287	if (ret < 0)
2288		goto out_end;
2289
2290	if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) == 0) {
2291		dout("splice_read/sync %p %llx.%llx %llu~%zu got cap refs on %s\n",
2292		     inode, ceph_vinop(inode), *ppos, len,
2293		     ceph_cap_string(got));
2294
2295		ceph_put_cap_refs(ci, got);
2296		ceph_end_io_read(inode);
2297		return copy_splice_read(in, ppos, pipe, len, flags);
2298	}
2299
2300	dout("splice_read %p %llx.%llx %llu~%zu got cap refs on %s\n",
2301	     inode, ceph_vinop(inode), *ppos, len, ceph_cap_string(got));
2302
2303	rw_ctx.caps = got;
2304	ceph_add_rw_context(fi, &rw_ctx);
2305	ret = filemap_splice_read(in, ppos, pipe, len, flags);
2306	ceph_del_rw_context(fi, &rw_ctx);
2307
2308	dout("splice_read %p %llx.%llx dropping cap refs on %s = %zd\n",
2309	     inode, ceph_vinop(inode), ceph_cap_string(got), ret);
2310
2311	ceph_put_cap_refs(ci, got);
2312out_end:
2313	ceph_end_io_read(inode);
2314	return ret;
2315}
2316
2317/*
2318 * Take cap references to avoid releasing caps to MDS mid-write.
2319 *
2320 * If we are synchronous, and write with an old snap context, the OSD
2321 * may return EOLDSNAPC.  In that case, retry the write.. _after_
2322 * dropping our cap refs and allowing the pending snap to logically
2323 * complete _before_ this write occurs.
2324 *
2325 * If we are near ENOSPC, write synchronously.
2326 */
2327static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
2328{
2329	struct file *file = iocb->ki_filp;
2330	struct ceph_file_info *fi = file->private_data;
2331	struct inode *inode = file_inode(file);
2332	struct ceph_inode_info *ci = ceph_inode(inode);
2333	struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
2334	struct ceph_client *cl = fsc->client;
2335	struct ceph_osd_client *osdc = &fsc->client->osdc;
2336	struct ceph_cap_flush *prealloc_cf;
2337	ssize_t count, written = 0;
2338	int err, want = 0, got;
2339	bool direct_lock = false;
2340	u32 map_flags;
2341	u64 pool_flags;
2342	loff_t pos;
2343	loff_t limit = max(i_size_read(inode), fsc->max_file_size);
2344
2345	if (ceph_inode_is_shutdown(inode))
2346		return -ESTALE;
2347
2348	if (ceph_snap(inode) != CEPH_NOSNAP)
2349		return -EROFS;
2350
2351	prealloc_cf = ceph_alloc_cap_flush();
2352	if (!prealloc_cf)
2353		return -ENOMEM;
2354
2355	if ((iocb->ki_flags & (IOCB_DIRECT | IOCB_APPEND)) == IOCB_DIRECT)
2356		direct_lock = true;
2357
2358retry_snap:
2359	err = direct_lock ? ceph_start_io_direct(inode) :
2360			    ceph_start_io_write(inode);
2361	if (err)
2362		goto out_unlocked;
2363
2364	if (iocb->ki_flags & IOCB_APPEND) {
2365		err = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
2366		if (err < 0)
2367			goto out;
2368	}
2369
2370	err = generic_write_checks(iocb, from);
2371	if (err <= 0)
2372		goto out;
2373
2374	pos = iocb->ki_pos;
2375	if (unlikely(pos >= limit)) {
2376		err = -EFBIG;
2377		goto out;
2378	} else {
2379		iov_iter_truncate(from, limit - pos);
2380	}
2381
2382	count = iov_iter_count(from);
2383	if (ceph_quota_is_max_bytes_exceeded(inode, pos + count)) {
2384		err = -EDQUOT;
2385		goto out;
2386	}
2387
2388	down_read(&osdc->lock);
2389	map_flags = osdc->osdmap->flags;
2390	pool_flags = ceph_pg_pool_flags(osdc->osdmap, ci->i_layout.pool_id);
2391	up_read(&osdc->lock);
2392	if ((map_flags & CEPH_OSDMAP_FULL) ||
2393	    (pool_flags & CEPH_POOL_FLAG_FULL)) {
2394		err = -ENOSPC;
2395		goto out;
2396	}
2397
2398	err = file_remove_privs(file);
2399	if (err)
2400		goto out;
2401
2402	doutc(cl, "%p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
2403	      inode, ceph_vinop(inode), pos, count,
2404	      i_size_read(inode));
2405	if (!(fi->flags & CEPH_F_SYNC) && !direct_lock)
2406		want |= CEPH_CAP_FILE_BUFFER;
2407	if (fi->fmode & CEPH_FILE_MODE_LAZY)
2408		want |= CEPH_CAP_FILE_LAZYIO;
2409	got = 0;
2410	err = ceph_get_caps(file, CEPH_CAP_FILE_WR, want, pos + count, &got);
2411	if (err < 0)
2412		goto out;
2413
2414	err = file_update_time(file);
2415	if (err)
2416		goto out_caps;
2417
2418	inode_inc_iversion_raw(inode);
2419
2420	doutc(cl, "%p %llx.%llx %llu~%zd got cap refs on %s\n",
2421	      inode, ceph_vinop(inode), pos, count, ceph_cap_string(got));
2422
2423	if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 ||
2424	    (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC) ||
2425	    (ci->i_ceph_flags & CEPH_I_ERROR_WRITE)) {
2426		struct ceph_snap_context *snapc;
2427		struct iov_iter data;
2428
2429		spin_lock(&ci->i_ceph_lock);
2430		if (__ceph_have_pending_cap_snap(ci)) {
2431			struct ceph_cap_snap *capsnap =
2432					list_last_entry(&ci->i_cap_snaps,
2433							struct ceph_cap_snap,
2434							ci_item);
2435			snapc = ceph_get_snap_context(capsnap->context);
2436		} else {
2437			BUG_ON(!ci->i_head_snapc);
2438			snapc = ceph_get_snap_context(ci->i_head_snapc);
2439		}
2440		spin_unlock(&ci->i_ceph_lock);
2441
2442		/* we might need to revert back to that point */
2443		data = *from;
2444		if ((iocb->ki_flags & IOCB_DIRECT) && !IS_ENCRYPTED(inode))
2445			written = ceph_direct_read_write(iocb, &data, snapc,
2446							 &prealloc_cf);
2447		else
2448			written = ceph_sync_write(iocb, &data, pos, snapc);
2449		if (direct_lock)
2450			ceph_end_io_direct(inode);
2451		else
2452			ceph_end_io_write(inode);
2453		if (written > 0)
2454			iov_iter_advance(from, written);
2455		ceph_put_snap_context(snapc);
2456	} else {
2457		/*
2458		 * No need to acquire the i_truncate_mutex. Because
2459		 * the MDS revokes Fwb caps before sending truncate
2460		 * message to us. We can't get Fwb cap while there
2461		 * are pending vmtruncate. So write and vmtruncate
2462		 * can not run at the same time
2463		 */
2464		written = generic_perform_write(iocb, from);
2465		ceph_end_io_write(inode);
2466	}
2467
2468	if (written >= 0) {
2469		int dirty;
2470
2471		spin_lock(&ci->i_ceph_lock);
2472		dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
2473					       &prealloc_cf);
2474		spin_unlock(&ci->i_ceph_lock);
2475		if (dirty)
2476			__mark_inode_dirty(inode, dirty);
2477		if (ceph_quota_is_max_bytes_approaching(inode, iocb->ki_pos))
2478			ceph_check_caps(ci, CHECK_CAPS_FLUSH);
2479	}
2480
2481	doutc(cl, "%p %llx.%llx %llu~%u  dropping cap refs on %s\n",
2482	      inode, ceph_vinop(inode), pos, (unsigned)count,
2483	      ceph_cap_string(got));
2484	ceph_put_cap_refs(ci, got);
2485
2486	if (written == -EOLDSNAPC) {
2487		doutc(cl, "%p %llx.%llx %llu~%u" "got EOLDSNAPC, retrying\n",
2488		      inode, ceph_vinop(inode), pos, (unsigned)count);
2489		goto retry_snap;
2490	}
2491
2492	if (written >= 0) {
2493		if ((map_flags & CEPH_OSDMAP_NEARFULL) ||
2494		    (pool_flags & CEPH_POOL_FLAG_NEARFULL))
2495			iocb->ki_flags |= IOCB_DSYNC;
2496		written = generic_write_sync(iocb, written);
2497	}
2498
2499	goto out_unlocked;
2500out_caps:
2501	ceph_put_cap_refs(ci, got);
2502out:
2503	if (direct_lock)
2504		ceph_end_io_direct(inode);
2505	else
2506		ceph_end_io_write(inode);
2507out_unlocked:
2508	ceph_free_cap_flush(prealloc_cf);
2509	return written ? written : err;
2510}
2511
2512/*
2513 * llseek.  be sure to verify file size on SEEK_END.
2514 */
2515static loff_t ceph_llseek(struct file *file, loff_t offset, int whence)
2516{
2517	if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) {
2518		struct inode *inode = file_inode(file);
2519		int ret;
2520
2521		ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
2522		if (ret < 0)
2523			return ret;
2524	}
2525	return generic_file_llseek(file, offset, whence);
2526}
2527
2528static inline void ceph_zero_partial_page(struct inode *inode,
2529		loff_t offset, size_t size)
2530{
2531	struct folio *folio;
2532
2533	folio = filemap_lock_folio(inode->i_mapping, offset >> PAGE_SHIFT);
2534	if (IS_ERR(folio))
2535		return;
2536
2537	folio_wait_writeback(folio);
2538	folio_zero_range(folio, offset_in_folio(folio, offset), size);
2539	folio_unlock(folio);
2540	folio_put(folio);
2541}
2542
2543static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset,
2544				      loff_t length)
2545{
2546	loff_t nearly = round_up(offset, PAGE_SIZE);
2547	if (offset < nearly) {
2548		loff_t size = nearly - offset;
2549		if (length < size)
2550			size = length;
2551		ceph_zero_partial_page(inode, offset, size);
2552		offset += size;
2553		length -= size;
2554	}
2555	if (length >= PAGE_SIZE) {
2556		loff_t size = round_down(length, PAGE_SIZE);
2557		truncate_pagecache_range(inode, offset, offset + size - 1);
2558		offset += size;
2559		length -= size;
2560	}
2561	if (length)
2562		ceph_zero_partial_page(inode, offset, length);
2563}
2564
2565static int ceph_zero_partial_object(struct inode *inode,
2566				    loff_t offset, loff_t *length)
2567{
2568	struct ceph_inode_info *ci = ceph_inode(inode);
2569	struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
2570	struct ceph_osd_request *req;
2571	struct ceph_snap_context *snapc;
2572	int ret = 0;
2573	loff_t zero = 0;
2574	int op;
2575
2576	if (ceph_inode_is_shutdown(inode))
2577		return -EIO;
2578
2579	if (!length) {
2580		op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE;
2581		length = &zero;
2582	} else {
2583		op = CEPH_OSD_OP_ZERO;
2584	}
2585
2586	spin_lock(&ci->i_ceph_lock);
2587	if (__ceph_have_pending_cap_snap(ci)) {
2588		struct ceph_cap_snap *capsnap =
2589				list_last_entry(&ci->i_cap_snaps,
2590						struct ceph_cap_snap,
2591						ci_item);
2592		snapc = ceph_get_snap_context(capsnap->context);
2593	} else {
2594		BUG_ON(!ci->i_head_snapc);
2595		snapc = ceph_get_snap_context(ci->i_head_snapc);
2596	}
2597	spin_unlock(&ci->i_ceph_lock);
2598
2599	req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
2600					ceph_vino(inode),
2601					offset, length,
2602					0, 1, op,
2603					CEPH_OSD_FLAG_WRITE,
2604					snapc, 0, 0, false);
2605	if (IS_ERR(req)) {
2606		ret = PTR_ERR(req);
2607		goto out;
2608	}
2609
2610	req->r_mtime = inode_get_mtime(inode);
2611	ceph_osdc_start_request(&fsc->client->osdc, req);
2612	ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
2613	if (ret == -ENOENT)
2614		ret = 0;
2615	ceph_osdc_put_request(req);
2616
2617out:
2618	ceph_put_snap_context(snapc);
2619	return ret;
2620}
2621
2622static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length)
2623{
2624	int ret = 0;
2625	struct ceph_inode_info *ci = ceph_inode(inode);
2626	s32 stripe_unit = ci->i_layout.stripe_unit;
2627	s32 stripe_count = ci->i_layout.stripe_count;
2628	s32 object_size = ci->i_layout.object_size;
2629	u64 object_set_size = (u64) object_size * stripe_count;
2630	u64 nearly, t;
2631
2632	/* round offset up to next period boundary */
2633	nearly = offset + object_set_size - 1;
2634	t = nearly;
2635	nearly -= do_div(t, object_set_size);
2636
2637	while (length && offset < nearly) {
2638		loff_t size = length;
2639		ret = ceph_zero_partial_object(inode, offset, &size);
2640		if (ret < 0)
2641			return ret;
2642		offset += size;
2643		length -= size;
2644	}
2645	while (length >= object_set_size) {
2646		int i;
2647		loff_t pos = offset;
2648		for (i = 0; i < stripe_count; ++i) {
2649			ret = ceph_zero_partial_object(inode, pos, NULL);
2650			if (ret < 0)
2651				return ret;
2652			pos += stripe_unit;
2653		}
2654		offset += object_set_size;
2655		length -= object_set_size;
2656	}
2657	while (length) {
2658		loff_t size = length;
2659		ret = ceph_zero_partial_object(inode, offset, &size);
2660		if (ret < 0)
2661			return ret;
2662		offset += size;
2663		length -= size;
2664	}
2665	return ret;
2666}
2667
2668static long ceph_fallocate(struct file *file, int mode,
2669				loff_t offset, loff_t length)
2670{
2671	struct ceph_file_info *fi = file->private_data;
2672	struct inode *inode = file_inode(file);
2673	struct ceph_inode_info *ci = ceph_inode(inode);
2674	struct ceph_cap_flush *prealloc_cf;
2675	struct ceph_client *cl = ceph_inode_to_client(inode);
2676	int want, got = 0;
2677	int dirty;
2678	int ret = 0;
2679	loff_t endoff = 0;
2680	loff_t size;
2681
2682	doutc(cl, "%p %llx.%llx mode %x, offset %llu length %llu\n",
2683	      inode, ceph_vinop(inode), mode, offset, length);
2684
2685	if (mode != (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2686		return -EOPNOTSUPP;
2687
2688	if (!S_ISREG(inode->i_mode))
2689		return -EOPNOTSUPP;
2690
2691	if (IS_ENCRYPTED(inode))
2692		return -EOPNOTSUPP;
2693
2694	prealloc_cf = ceph_alloc_cap_flush();
2695	if (!prealloc_cf)
2696		return -ENOMEM;
2697
2698	inode_lock(inode);
2699
2700	if (ceph_snap(inode) != CEPH_NOSNAP) {
2701		ret = -EROFS;
2702		goto unlock;
2703	}
2704
2705	size = i_size_read(inode);
2706
2707	/* Are we punching a hole beyond EOF? */
2708	if (offset >= size)
2709		goto unlock;
2710	if ((offset + length) > size)
2711		length = size - offset;
2712
2713	if (fi->fmode & CEPH_FILE_MODE_LAZY)
2714		want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
2715	else
2716		want = CEPH_CAP_FILE_BUFFER;
2717
2718	ret = ceph_get_caps(file, CEPH_CAP_FILE_WR, want, endoff, &got);
2719	if (ret < 0)
2720		goto unlock;
2721
2722	ret = file_modified(file);
2723	if (ret)
2724		goto put_caps;
2725
2726	filemap_invalidate_lock(inode->i_mapping);
2727	ceph_fscache_invalidate(inode, false);
2728	ceph_zero_pagecache_range(inode, offset, length);
2729	ret = ceph_zero_objects(inode, offset, length);
2730
2731	if (!ret) {
2732		spin_lock(&ci->i_ceph_lock);
2733		dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
2734					       &prealloc_cf);
2735		spin_unlock(&ci->i_ceph_lock);
2736		if (dirty)
2737			__mark_inode_dirty(inode, dirty);
2738	}
2739	filemap_invalidate_unlock(inode->i_mapping);
2740
2741put_caps:
2742	ceph_put_cap_refs(ci, got);
2743unlock:
2744	inode_unlock(inode);
2745	ceph_free_cap_flush(prealloc_cf);
2746	return ret;
2747}
2748
2749/*
2750 * This function tries to get FILE_WR capabilities for dst_ci and FILE_RD for
2751 * src_ci.  Two attempts are made to obtain both caps, and an error is return if
2752 * this fails; zero is returned on success.
2753 */
2754static int get_rd_wr_caps(struct file *src_filp, int *src_got,
2755			  struct file *dst_filp,
2756			  loff_t dst_endoff, int *dst_got)
2757{
2758	int ret = 0;
2759	bool retrying = false;
2760
2761retry_caps:
2762	ret = ceph_get_caps(dst_filp, CEPH_CAP_FILE_WR, CEPH_CAP_FILE_BUFFER,
2763			    dst_endoff, dst_got);
2764	if (ret < 0)
2765		return ret;
2766
2767	/*
2768	 * Since we're already holding the FILE_WR capability for the dst file,
2769	 * we would risk a deadlock by using ceph_get_caps.  Thus, we'll do some
2770	 * retry dance instead to try to get both capabilities.
2771	 */
2772	ret = ceph_try_get_caps(file_inode(src_filp),
2773				CEPH_CAP_FILE_RD, CEPH_CAP_FILE_SHARED,
2774				false, src_got);
2775	if (ret <= 0) {
2776		/* Start by dropping dst_ci caps and getting src_ci caps */
2777		ceph_put_cap_refs(ceph_inode(file_inode(dst_filp)), *dst_got);
2778		if (retrying) {
2779			if (!ret)
2780				/* ceph_try_get_caps masks EAGAIN */
2781				ret = -EAGAIN;
2782			return ret;
2783		}
2784		ret = ceph_get_caps(src_filp, CEPH_CAP_FILE_RD,
2785				    CEPH_CAP_FILE_SHARED, -1, src_got);
2786		if (ret < 0)
2787			return ret;
2788		/*... drop src_ci caps too, and retry */
2789		ceph_put_cap_refs(ceph_inode(file_inode(src_filp)), *src_got);
2790		retrying = true;
2791		goto retry_caps;
2792	}
2793	return ret;
2794}
2795
2796static void put_rd_wr_caps(struct ceph_inode_info *src_ci, int src_got,
2797			   struct ceph_inode_info *dst_ci, int dst_got)
2798{
2799	ceph_put_cap_refs(src_ci, src_got);
2800	ceph_put_cap_refs(dst_ci, dst_got);
2801}
2802
2803/*
2804 * This function does several size-related checks, returning an error if:
2805 *  - source file is smaller than off+len
2806 *  - destination file size is not OK (inode_newsize_ok())
2807 *  - max bytes quotas is exceeded
2808 */
2809static int is_file_size_ok(struct inode *src_inode, struct inode *dst_inode,
2810			   loff_t src_off, loff_t dst_off, size_t len)
2811{
2812	struct ceph_client *cl = ceph_inode_to_client(src_inode);
2813	loff_t size, endoff;
2814
2815	size = i_size_read(src_inode);
2816	/*
2817	 * Don't copy beyond source file EOF.  Instead of simply setting length
2818	 * to (size - src_off), just drop to VFS default implementation, as the
2819	 * local i_size may be stale due to other clients writing to the source
2820	 * inode.
2821	 */
2822	if (src_off + len > size) {
2823		doutc(cl, "Copy beyond EOF (%llu + %zu > %llu)\n", src_off,
2824		      len, size);
2825		return -EOPNOTSUPP;
2826	}
2827	size = i_size_read(dst_inode);
2828
2829	endoff = dst_off + len;
2830	if (inode_newsize_ok(dst_inode, endoff))
2831		return -EOPNOTSUPP;
2832
2833	if (ceph_quota_is_max_bytes_exceeded(dst_inode, endoff))
2834		return -EDQUOT;
2835
2836	return 0;
2837}
2838
2839static struct ceph_osd_request *
2840ceph_alloc_copyfrom_request(struct ceph_osd_client *osdc,
2841			    u64 src_snapid,
2842			    struct ceph_object_id *src_oid,
2843			    struct ceph_object_locator *src_oloc,
2844			    struct ceph_object_id *dst_oid,
2845			    struct ceph_object_locator *dst_oloc,
2846			    u32 truncate_seq, u64 truncate_size)
2847{
2848	struct ceph_osd_request *req;
2849	int ret;
2850	u32 src_fadvise_flags =
2851		CEPH_OSD_OP_FLAG_FADVISE_SEQUENTIAL |
2852		CEPH_OSD_OP_FLAG_FADVISE_NOCACHE;
2853	u32 dst_fadvise_flags =
2854		CEPH_OSD_OP_FLAG_FADVISE_SEQUENTIAL |
2855		CEPH_OSD_OP_FLAG_FADVISE_DONTNEED;
2856
2857	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_KERNEL);
2858	if (!req)
2859		return ERR_PTR(-ENOMEM);
2860
2861	req->r_flags = CEPH_OSD_FLAG_WRITE;
2862
2863	ceph_oloc_copy(&req->r_t.base_oloc, dst_oloc);
2864	ceph_oid_copy(&req->r_t.base_oid, dst_oid);
2865
2866	ret = osd_req_op_copy_from_init(req, src_snapid, 0,
2867					src_oid, src_oloc,
2868					src_fadvise_flags,
2869					dst_fadvise_flags,
2870					truncate_seq,
2871					truncate_size,
2872					CEPH_OSD_COPY_FROM_FLAG_TRUNCATE_SEQ);
2873	if (ret)
2874		goto out;
2875
2876	ret = ceph_osdc_alloc_messages(req, GFP_KERNEL);
2877	if (ret)
2878		goto out;
2879
2880	return req;
2881
2882out:
2883	ceph_osdc_put_request(req);
2884	return ERR_PTR(ret);
2885}
2886
2887static ssize_t ceph_do_objects_copy(struct ceph_inode_info *src_ci, u64 *src_off,
2888				    struct ceph_inode_info *dst_ci, u64 *dst_off,
2889				    struct ceph_fs_client *fsc,
2890				    size_t len, unsigned int flags)
2891{
2892	struct ceph_object_locator src_oloc, dst_oloc;
2893	struct ceph_object_id src_oid, dst_oid;
2894	struct ceph_osd_client *osdc;
2895	struct ceph_osd_request *req;
2896	ssize_t bytes = 0;
2897	u64 src_objnum, src_objoff, dst_objnum, dst_objoff;
2898	u32 src_objlen, dst_objlen;
2899	u32 object_size = src_ci->i_layout.object_size;
2900	struct ceph_client *cl = fsc->client;
2901	int ret;
2902
2903	src_oloc.pool = src_ci->i_layout.pool_id;
2904	src_oloc.pool_ns = ceph_try_get_string(src_ci->i_layout.pool_ns);
2905	dst_oloc.pool = dst_ci->i_layout.pool_id;
2906	dst_oloc.pool_ns = ceph_try_get_string(dst_ci->i_layout.pool_ns);
2907	osdc = &fsc->client->osdc;
2908
2909	while (len >= object_size) {
2910		ceph_calc_file_object_mapping(&src_ci->i_layout, *src_off,
2911					      object_size, &src_objnum,
2912					      &src_objoff, &src_objlen);
2913		ceph_calc_file_object_mapping(&dst_ci->i_layout, *dst_off,
2914					      object_size, &dst_objnum,
2915					      &dst_objoff, &dst_objlen);
2916		ceph_oid_init(&src_oid);
2917		ceph_oid_printf(&src_oid, "%llx.%08llx",
2918				src_ci->i_vino.ino, src_objnum);
2919		ceph_oid_init(&dst_oid);
2920		ceph_oid_printf(&dst_oid, "%llx.%08llx",
2921				dst_ci->i_vino.ino, dst_objnum);
2922		/* Do an object remote copy */
2923		req = ceph_alloc_copyfrom_request(osdc, src_ci->i_vino.snap,
2924						  &src_oid, &src_oloc,
2925						  &dst_oid, &dst_oloc,
2926						  dst_ci->i_truncate_seq,
2927						  dst_ci->i_truncate_size);
2928		if (IS_ERR(req))
2929			ret = PTR_ERR(req);
2930		else {
2931			ceph_osdc_start_request(osdc, req);
2932			ret = ceph_osdc_wait_request(osdc, req);
2933			ceph_update_copyfrom_metrics(&fsc->mdsc->metric,
2934						     req->r_start_latency,
2935						     req->r_end_latency,
2936						     object_size, ret);
2937			ceph_osdc_put_request(req);
2938		}
2939		if (ret) {
2940			if (ret == -EOPNOTSUPP) {
2941				fsc->have_copy_from2 = false;
2942				pr_notice_client(cl,
2943					"OSDs don't support copy-from2; disabling copy offload\n");
2944			}
2945			doutc(cl, "returned %d\n", ret);
2946			if (bytes <= 0)
2947				bytes = ret;
2948			goto out;
2949		}
2950		len -= object_size;
2951		bytes += object_size;
2952		*src_off += object_size;
2953		*dst_off += object_size;
2954	}
2955
2956out:
2957	ceph_oloc_destroy(&src_oloc);
2958	ceph_oloc_destroy(&dst_oloc);
2959	return bytes;
2960}
2961
2962static ssize_t __ceph_copy_file_range(struct file *src_file, loff_t src_off,
2963				      struct file *dst_file, loff_t dst_off,
2964				      size_t len, unsigned int flags)
2965{
2966	struct inode *src_inode = file_inode(src_file);
2967	struct inode *dst_inode = file_inode(dst_file);
2968	struct ceph_inode_info *src_ci = ceph_inode(src_inode);
2969	struct ceph_inode_info *dst_ci = ceph_inode(dst_inode);
2970	struct ceph_cap_flush *prealloc_cf;
2971	struct ceph_fs_client *src_fsc = ceph_inode_to_fs_client(src_inode);
2972	struct ceph_client *cl = src_fsc->client;
2973	loff_t size;
2974	ssize_t ret = -EIO, bytes;
2975	u64 src_objnum, dst_objnum, src_objoff, dst_objoff;
2976	u32 src_objlen, dst_objlen;
2977	int src_got = 0, dst_got = 0, err, dirty;
2978
2979	if (src_inode->i_sb != dst_inode->i_sb) {
2980		struct ceph_fs_client *dst_fsc = ceph_inode_to_fs_client(dst_inode);
2981
2982		if (ceph_fsid_compare(&src_fsc->client->fsid,
2983				      &dst_fsc->client->fsid)) {
2984			dout("Copying files across clusters: src: %pU dst: %pU\n",
2985			     &src_fsc->client->fsid, &dst_fsc->client->fsid);
2986			return -EXDEV;
2987		}
2988	}
2989	if (ceph_snap(dst_inode) != CEPH_NOSNAP)
2990		return -EROFS;
2991
2992	/*
2993	 * Some of the checks below will return -EOPNOTSUPP, which will force a
2994	 * fallback to the default VFS copy_file_range implementation.  This is
2995	 * desirable in several cases (for ex, the 'len' is smaller than the
2996	 * size of the objects, or in cases where that would be more
2997	 * efficient).
2998	 */
2999
3000	if (ceph_test_mount_opt(src_fsc, NOCOPYFROM))
3001		return -EOPNOTSUPP;
3002
3003	if (!src_fsc->have_copy_from2)
3004		return -EOPNOTSUPP;
3005
3006	/*
3007	 * Striped file layouts require that we copy partial objects, but the
3008	 * OSD copy-from operation only supports full-object copies.  Limit
3009	 * this to non-striped file layouts for now.
3010	 */
3011	if ((src_ci->i_layout.stripe_unit != dst_ci->i_layout.stripe_unit) ||
3012	    (src_ci->i_layout.stripe_count != 1) ||
3013	    (dst_ci->i_layout.stripe_count != 1) ||
3014	    (src_ci->i_layout.object_size != dst_ci->i_layout.object_size)) {
3015		doutc(cl, "Invalid src/dst files layout\n");
3016		return -EOPNOTSUPP;
3017	}
3018
3019	/* Every encrypted inode gets its own key, so we can't offload them */
3020	if (IS_ENCRYPTED(src_inode) || IS_ENCRYPTED(dst_inode))
3021		return -EOPNOTSUPP;
3022
3023	if (len < src_ci->i_layout.object_size)
3024		return -EOPNOTSUPP; /* no remote copy will be done */
3025
3026	prealloc_cf = ceph_alloc_cap_flush();
3027	if (!prealloc_cf)
3028		return -ENOMEM;
3029
3030	/* Start by sync'ing the source and destination files */
3031	ret = file_write_and_wait_range(src_file, src_off, (src_off + len));
3032	if (ret < 0) {
3033		doutc(cl, "failed to write src file (%zd)\n", ret);
3034		goto out;
3035	}
3036	ret = file_write_and_wait_range(dst_file, dst_off, (dst_off + len));
3037	if (ret < 0) {
3038		doutc(cl, "failed to write dst file (%zd)\n", ret);
3039		goto out;
3040	}
3041
3042	/*
3043	 * We need FILE_WR caps for dst_ci and FILE_RD for src_ci as other
3044	 * clients may have dirty data in their caches.  And OSDs know nothing
3045	 * about caps, so they can't safely do the remote object copies.
3046	 */
3047	err = get_rd_wr_caps(src_file, &src_got,
3048			     dst_file, (dst_off + len), &dst_got);
3049	if (err < 0) {
3050		doutc(cl, "get_rd_wr_caps returned %d\n", err);
3051		ret = -EOPNOTSUPP;
3052		goto out;
3053	}
3054
3055	ret = is_file_size_ok(src_inode, dst_inode, src_off, dst_off, len);
3056	if (ret < 0)
3057		goto out_caps;
3058
3059	/* Drop dst file cached pages */
3060	ceph_fscache_invalidate(dst_inode, false);
3061	ret = invalidate_inode_pages2_range(dst_inode->i_mapping,
3062					    dst_off >> PAGE_SHIFT,
3063					    (dst_off + len) >> PAGE_SHIFT);
3064	if (ret < 0) {
3065		doutc(cl, "Failed to invalidate inode pages (%zd)\n",
3066			    ret);
3067		ret = 0; /* XXX */
3068	}
3069	ceph_calc_file_object_mapping(&src_ci->i_layout, src_off,
3070				      src_ci->i_layout.object_size,
3071				      &src_objnum, &src_objoff, &src_objlen);
3072	ceph_calc_file_object_mapping(&dst_ci->i_layout, dst_off,
3073				      dst_ci->i_layout.object_size,
3074				      &dst_objnum, &dst_objoff, &dst_objlen);
3075	/* object-level offsets need to the same */
3076	if (src_objoff != dst_objoff) {
3077		ret = -EOPNOTSUPP;
3078		goto out_caps;
3079	}
3080
3081	/*
3082	 * Do a manual copy if the object offset isn't object aligned.
3083	 * 'src_objlen' contains the bytes left until the end of the object,
3084	 * starting at the src_off
3085	 */
3086	if (src_objoff) {
3087		doutc(cl, "Initial partial copy of %u bytes\n", src_objlen);
3088
3089		/*
3090		 * we need to temporarily drop all caps as we'll be calling
3091		 * {read,write}_iter, which will get caps again.
3092		 */
3093		put_rd_wr_caps(src_ci, src_got, dst_ci, dst_got);
3094		ret = splice_file_range(src_file, &src_off, dst_file, &dst_off,
3095					src_objlen);
3096		/* Abort on short copies or on error */
3097		if (ret < (long)src_objlen) {
3098			doutc(cl, "Failed partial copy (%zd)\n", ret);
3099			goto out;
3100		}
3101		len -= ret;
3102		err = get_rd_wr_caps(src_file, &src_got,
3103				     dst_file, (dst_off + len), &dst_got);
3104		if (err < 0)
3105			goto out;
3106		err = is_file_size_ok(src_inode, dst_inode,
3107				      src_off, dst_off, len);
3108		if (err < 0)
3109			goto out_caps;
3110	}
3111
3112	size = i_size_read(dst_inode);
3113	bytes = ceph_do_objects_copy(src_ci, &src_off, dst_ci, &dst_off,
3114				     src_fsc, len, flags);
3115	if (bytes <= 0) {
3116		if (!ret)
3117			ret = bytes;
3118		goto out_caps;
3119	}
3120	doutc(cl, "Copied %zu bytes out of %zu\n", bytes, len);
3121	len -= bytes;
3122	ret += bytes;
3123
3124	file_update_time(dst_file);
3125	inode_inc_iversion_raw(dst_inode);
3126
3127	if (dst_off > size) {
3128		/* Let the MDS know about dst file size change */
3129		if (ceph_inode_set_size(dst_inode, dst_off) ||
3130		    ceph_quota_is_max_bytes_approaching(dst_inode, dst_off))
3131			ceph_check_caps(dst_ci, CHECK_CAPS_AUTHONLY | CHECK_CAPS_FLUSH);
3132	}
3133	/* Mark Fw dirty */
3134	spin_lock(&dst_ci->i_ceph_lock);
3135	dirty = __ceph_mark_dirty_caps(dst_ci, CEPH_CAP_FILE_WR, &prealloc_cf);
3136	spin_unlock(&dst_ci->i_ceph_lock);
3137	if (dirty)
3138		__mark_inode_dirty(dst_inode, dirty);
3139
3140out_caps:
3141	put_rd_wr_caps(src_ci, src_got, dst_ci, dst_got);
3142
3143	/*
3144	 * Do the final manual copy if we still have some bytes left, unless
3145	 * there were errors in remote object copies (len >= object_size).
3146	 */
3147	if (len && (len < src_ci->i_layout.object_size)) {
3148		doutc(cl, "Final partial copy of %zu bytes\n", len);
3149		bytes = splice_file_range(src_file, &src_off, dst_file,
3150					  &dst_off, len);
3151		if (bytes > 0)
3152			ret += bytes;
3153		else
3154			doutc(cl, "Failed partial copy (%zd)\n", bytes);
3155	}
3156
3157out:
3158	ceph_free_cap_flush(prealloc_cf);
3159
3160	return ret;
3161}
3162
3163static ssize_t ceph_copy_file_range(struct file *src_file, loff_t src_off,
3164				    struct file *dst_file, loff_t dst_off,
3165				    size_t len, unsigned int flags)
3166{
3167	ssize_t ret;
3168
3169	ret = __ceph_copy_file_range(src_file, src_off, dst_file, dst_off,
3170				     len, flags);
3171
3172	if (ret == -EOPNOTSUPP || ret == -EXDEV)
3173		ret = splice_copy_file_range(src_file, src_off, dst_file,
3174					     dst_off, len);
3175	return ret;
3176}
3177
3178const struct file_operations ceph_file_fops = {
3179	.open = ceph_open,
3180	.release = ceph_release,
3181	.llseek = ceph_llseek,
3182	.read_iter = ceph_read_iter,
3183	.write_iter = ceph_write_iter,
3184	.mmap_prepare = ceph_mmap_prepare,
3185	.fsync = ceph_fsync,
3186	.lock = ceph_lock,
3187	.setlease = simple_nosetlease,
3188	.flock = ceph_flock,
3189	.splice_read = ceph_splice_read,
3190	.splice_write = iter_file_splice_write,
3191	.unlocked_ioctl = ceph_ioctl,
3192	.compat_ioctl = compat_ptr_ioctl,
3193	.fallocate	= ceph_fallocate,
3194	.copy_file_range = ceph_copy_file_range,
3195};