xref: /openbmc/linux/fs/nfs/pnfs.c (revision 55e43d6abd078ed6d219902ce8cb4d68e3c993ba)
1  /*
2   *  pNFS functions to call and manage layout drivers.
3   *
4   *  Copyright (c) 2002 [year of first publication]
5   *  The Regents of the University of Michigan
6   *  All Rights Reserved
7   *
8   *  Dean Hildebrand <dhildebz@umich.edu>
9   *
10   *  Permission is granted to use, copy, create derivative works, and
11   *  redistribute this software and such derivative works for any purpose,
12   *  so long as the name of the University of Michigan is not used in
13   *  any advertising or publicity pertaining to the use or distribution
14   *  of this software without specific, written prior authorization. If
15   *  the above copyright notice or any other identification of the
16   *  University of Michigan is included in any copy of any portion of
17   *  this software, then the disclaimer below must also be included.
18   *
19   *  This software is provided as is, without representation or warranty
20   *  of any kind either express or implied, including without limitation
21   *  the implied warranties of merchantability, fitness for a particular
22   *  purpose, or noninfringement.  The Regents of the University of
23   *  Michigan shall not be liable for any damages, including special,
24   *  indirect, incidental, or consequential damages, with respect to any
25   *  claim arising out of or in connection with the use of the software,
26   *  even if it has been or is hereafter advised of the possibility of
27   *  such damages.
28   */
29  
30  #include <linux/nfs_fs.h>
31  #include <linux/nfs_page.h>
32  #include <linux/module.h>
33  #include <linux/sort.h>
34  #include "internal.h"
35  #include "pnfs.h"
36  #include "iostat.h"
37  #include "nfs4trace.h"
38  #include "delegation.h"
39  #include "nfs42.h"
40  #include "nfs4_fs.h"
41  
42  #define NFSDBG_FACILITY		NFSDBG_PNFS
43  #define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
44  
45  /* Locking:
46   *
47   * pnfs_spinlock:
48   *      protects pnfs_modules_tbl.
49   */
50  static DEFINE_SPINLOCK(pnfs_spinlock);
51  
52  /*
53   * pnfs_modules_tbl holds all pnfs modules
54   */
55  static LIST_HEAD(pnfs_modules_tbl);
56  
57  static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo);
58  static void pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
59  		struct list_head *free_me,
60  		const struct pnfs_layout_range *range,
61  		u32 seq);
62  static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
63  		                struct list_head *tmp_list);
64  
65  /* Return the registered pnfs layout driver module matching given id */
66  static struct pnfs_layoutdriver_type *
find_pnfs_driver_locked(u32 id)67  find_pnfs_driver_locked(u32 id)
68  {
69  	struct pnfs_layoutdriver_type *local;
70  
71  	list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
72  		if (local->id == id)
73  			goto out;
74  	local = NULL;
75  out:
76  	dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
77  	return local;
78  }
79  
80  static struct pnfs_layoutdriver_type *
find_pnfs_driver(u32 id)81  find_pnfs_driver(u32 id)
82  {
83  	struct pnfs_layoutdriver_type *local;
84  
85  	spin_lock(&pnfs_spinlock);
86  	local = find_pnfs_driver_locked(id);
87  	if (local != NULL && !try_module_get(local->owner)) {
88  		dprintk("%s: Could not grab reference on module\n", __func__);
89  		local = NULL;
90  	}
91  	spin_unlock(&pnfs_spinlock);
92  	return local;
93  }
94  
pnfs_find_layoutdriver(u32 id)95  const struct pnfs_layoutdriver_type *pnfs_find_layoutdriver(u32 id)
96  {
97  	return find_pnfs_driver(id);
98  }
99  
pnfs_put_layoutdriver(const struct pnfs_layoutdriver_type * ld)100  void pnfs_put_layoutdriver(const struct pnfs_layoutdriver_type *ld)
101  {
102  	if (ld)
103  		module_put(ld->owner);
104  }
105  
106  void
unset_pnfs_layoutdriver(struct nfs_server * nfss)107  unset_pnfs_layoutdriver(struct nfs_server *nfss)
108  {
109  	if (nfss->pnfs_curr_ld) {
110  		if (nfss->pnfs_curr_ld->clear_layoutdriver)
111  			nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
112  		/* Decrement the MDS count. Purge the deviceid cache if zero */
113  		if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count))
114  			nfs4_deviceid_purge_client(nfss->nfs_client);
115  		module_put(nfss->pnfs_curr_ld->owner);
116  	}
117  	nfss->pnfs_curr_ld = NULL;
118  }
119  
120  /*
121   * When the server sends a list of layout types, we choose one in the order
122   * given in the list below.
123   *
124   * FIXME: should this list be configurable in some fashion? module param?
125   * 	  mount option? something else?
126   */
127  static const u32 ld_prefs[] = {
128  	LAYOUT_SCSI,
129  	LAYOUT_BLOCK_VOLUME,
130  	LAYOUT_OSD2_OBJECTS,
131  	LAYOUT_FLEX_FILES,
132  	LAYOUT_NFSV4_1_FILES,
133  	0
134  };
135  
136  static int
ld_cmp(const void * e1,const void * e2)137  ld_cmp(const void *e1, const void *e2)
138  {
139  	u32 ld1 = *((u32 *)e1);
140  	u32 ld2 = *((u32 *)e2);
141  	int i;
142  
143  	for (i = 0; ld_prefs[i] != 0; i++) {
144  		if (ld1 == ld_prefs[i])
145  			return -1;
146  
147  		if (ld2 == ld_prefs[i])
148  			return 1;
149  	}
150  	return 0;
151  }
152  
153  /*
154   * Try to set the server's pnfs module to the pnfs layout type specified by id.
155   * Currently only one pNFS layout driver per filesystem is supported.
156   *
157   * @ids array of layout types supported by MDS.
158   */
159  void
set_pnfs_layoutdriver(struct nfs_server * server,const struct nfs_fh * mntfh,struct nfs_fsinfo * fsinfo)160  set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
161  		      struct nfs_fsinfo *fsinfo)
162  {
163  	struct pnfs_layoutdriver_type *ld_type = NULL;
164  	u32 id;
165  	int i;
166  
167  	if (fsinfo->nlayouttypes == 0)
168  		goto out_no_driver;
169  	if (!(server->nfs_client->cl_exchange_flags &
170  		 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
171  		printk(KERN_ERR "NFS: %s: cl_exchange_flags 0x%x\n",
172  			__func__, server->nfs_client->cl_exchange_flags);
173  		goto out_no_driver;
174  	}
175  
176  	sort(fsinfo->layouttype, fsinfo->nlayouttypes,
177  		sizeof(*fsinfo->layouttype), ld_cmp, NULL);
178  
179  	for (i = 0; i < fsinfo->nlayouttypes; i++) {
180  		id = fsinfo->layouttype[i];
181  		ld_type = find_pnfs_driver(id);
182  		if (!ld_type) {
183  			request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX,
184  					id);
185  			ld_type = find_pnfs_driver(id);
186  		}
187  		if (ld_type)
188  			break;
189  	}
190  
191  	if (!ld_type) {
192  		dprintk("%s: No pNFS module found!\n", __func__);
193  		goto out_no_driver;
194  	}
195  
196  	server->pnfs_curr_ld = ld_type;
197  	if (ld_type->set_layoutdriver
198  	    && ld_type->set_layoutdriver(server, mntfh)) {
199  		printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
200  			"driver %u.\n", __func__, id);
201  		module_put(ld_type->owner);
202  		goto out_no_driver;
203  	}
204  	/* Bump the MDS count */
205  	atomic_inc(&server->nfs_client->cl_mds_count);
206  
207  	dprintk("%s: pNFS module for %u set\n", __func__, id);
208  	return;
209  
210  out_no_driver:
211  	dprintk("%s: Using NFSv4 I/O\n", __func__);
212  	server->pnfs_curr_ld = NULL;
213  }
214  
215  int
pnfs_register_layoutdriver(struct pnfs_layoutdriver_type * ld_type)216  pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
217  {
218  	int status = -EINVAL;
219  	struct pnfs_layoutdriver_type *tmp;
220  
221  	if (ld_type->id == 0) {
222  		printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
223  		return status;
224  	}
225  	if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
226  		printk(KERN_ERR "NFS: %s Layout driver must provide "
227  		       "alloc_lseg and free_lseg.\n", __func__);
228  		return status;
229  	}
230  
231  	spin_lock(&pnfs_spinlock);
232  	tmp = find_pnfs_driver_locked(ld_type->id);
233  	if (!tmp) {
234  		list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
235  		status = 0;
236  		dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
237  			ld_type->name);
238  	} else {
239  		printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
240  			__func__, ld_type->id);
241  	}
242  	spin_unlock(&pnfs_spinlock);
243  
244  	return status;
245  }
246  EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
247  
248  void
pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type * ld_type)249  pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
250  {
251  	dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
252  	spin_lock(&pnfs_spinlock);
253  	list_del(&ld_type->pnfs_tblid);
254  	spin_unlock(&pnfs_spinlock);
255  }
256  EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
257  
258  /*
259   * pNFS client layout cache
260   */
261  
262  /* Need to hold i_lock if caller does not already hold reference */
263  void
pnfs_get_layout_hdr(struct pnfs_layout_hdr * lo)264  pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo)
265  {
266  	refcount_inc(&lo->plh_refcount);
267  }
268  
269  static struct pnfs_layout_hdr *
pnfs_alloc_layout_hdr(struct inode * ino,gfp_t gfp_flags)270  pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
271  {
272  	struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
273  	return ld->alloc_layout_hdr(ino, gfp_flags);
274  }
275  
276  static void
pnfs_free_layout_hdr(struct pnfs_layout_hdr * lo)277  pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
278  {
279  	struct nfs_server *server = NFS_SERVER(lo->plh_inode);
280  	struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
281  
282  	if (test_and_clear_bit(NFS_LAYOUT_HASHED, &lo->plh_flags)) {
283  		struct nfs_client *clp = server->nfs_client;
284  
285  		spin_lock(&clp->cl_lock);
286  		list_del_rcu(&lo->plh_layouts);
287  		spin_unlock(&clp->cl_lock);
288  	}
289  	put_cred(lo->plh_lc_cred);
290  	return ld->free_layout_hdr(lo);
291  }
292  
293  static void
pnfs_detach_layout_hdr(struct pnfs_layout_hdr * lo)294  pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo)
295  {
296  	struct nfs_inode *nfsi = NFS_I(lo->plh_inode);
297  	dprintk("%s: freeing layout cache %p\n", __func__, lo);
298  	nfsi->layout = NULL;
299  	/* Reset MDS Threshold I/O counters */
300  	nfsi->write_io = 0;
301  	nfsi->read_io = 0;
302  }
303  
304  void
pnfs_put_layout_hdr(struct pnfs_layout_hdr * lo)305  pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
306  {
307  	struct inode *inode;
308  	unsigned long i_state;
309  
310  	if (!lo)
311  		return;
312  	inode = lo->plh_inode;
313  	pnfs_layoutreturn_before_put_layout_hdr(lo);
314  
315  	if (refcount_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
316  		if (!list_empty(&lo->plh_segs))
317  			WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n");
318  		pnfs_detach_layout_hdr(lo);
319  		i_state = inode->i_state;
320  		spin_unlock(&inode->i_lock);
321  		pnfs_free_layout_hdr(lo);
322  		/* Notify pnfs_destroy_layout_final() that we're done */
323  		if (i_state & (I_FREEING | I_CLEAR))
324  			wake_up_var(lo);
325  	}
326  }
327  
328  static struct inode *
pnfs_grab_inode_layout_hdr(struct pnfs_layout_hdr * lo)329  pnfs_grab_inode_layout_hdr(struct pnfs_layout_hdr *lo)
330  {
331  	struct inode *inode = igrab(lo->plh_inode);
332  	if (inode)
333  		return inode;
334  	set_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags);
335  	return NULL;
336  }
337  
338  /*
339   * Compare 2 layout stateid sequence ids, to see which is newer,
340   * taking into account wraparound issues.
341   */
pnfs_seqid_is_newer(u32 s1,u32 s2)342  static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
343  {
344  	return (s32)(s1 - s2) > 0;
345  }
346  
pnfs_barrier_update(struct pnfs_layout_hdr * lo,u32 newseq)347  static void pnfs_barrier_update(struct pnfs_layout_hdr *lo, u32 newseq)
348  {
349  	if (pnfs_seqid_is_newer(newseq, lo->plh_barrier) || !lo->plh_barrier)
350  		lo->plh_barrier = newseq;
351  }
352  
353  static void
pnfs_set_plh_return_info(struct pnfs_layout_hdr * lo,enum pnfs_iomode iomode,u32 seq)354  pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode,
355  			 u32 seq)
356  {
357  	if (lo->plh_return_iomode != 0 && lo->plh_return_iomode != iomode)
358  		iomode = IOMODE_ANY;
359  	lo->plh_return_iomode = iomode;
360  	set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
361  	/*
362  	 * We must set lo->plh_return_seq to avoid livelocks with
363  	 * pnfs_layout_need_return()
364  	 */
365  	if (seq == 0)
366  		seq = be32_to_cpu(lo->plh_stateid.seqid);
367  	if (!lo->plh_return_seq || pnfs_seqid_is_newer(seq, lo->plh_return_seq))
368  		lo->plh_return_seq = seq;
369  	pnfs_barrier_update(lo, seq);
370  }
371  
372  static void
pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr * lo)373  pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr *lo)
374  {
375  	struct pnfs_layout_segment *lseg;
376  	lo->plh_return_iomode = 0;
377  	lo->plh_return_seq = 0;
378  	clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
379  	list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
380  		if (!test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
381  			continue;
382  		pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
383  	}
384  }
385  
pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr * lo)386  static void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo)
387  {
388  	clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
389  	clear_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags);
390  	smp_mb__after_atomic();
391  	wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
392  	rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
393  }
394  
395  static void
pnfs_clear_lseg_state(struct pnfs_layout_segment * lseg,struct list_head * free_me)396  pnfs_clear_lseg_state(struct pnfs_layout_segment *lseg,
397  		struct list_head *free_me)
398  {
399  	clear_bit(NFS_LSEG_ROC, &lseg->pls_flags);
400  	clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
401  	if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags))
402  		pnfs_lseg_dec_and_remove_zero(lseg, free_me);
403  	if (test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
404  		pnfs_lseg_dec_and_remove_zero(lseg, free_me);
405  }
406  
407  /*
408   * Update the seqid of a layout stateid after receiving
409   * NFS4ERR_OLD_STATEID
410   */
nfs4_layout_refresh_old_stateid(nfs4_stateid * dst,struct pnfs_layout_range * dst_range,struct inode * inode)411  bool nfs4_layout_refresh_old_stateid(nfs4_stateid *dst,
412  		struct pnfs_layout_range *dst_range,
413  		struct inode *inode)
414  {
415  	struct pnfs_layout_hdr *lo;
416  	struct pnfs_layout_range range = {
417  		.iomode = IOMODE_ANY,
418  		.offset = 0,
419  		.length = NFS4_MAX_UINT64,
420  	};
421  	bool ret = false;
422  	LIST_HEAD(head);
423  	int err;
424  
425  	spin_lock(&inode->i_lock);
426  	lo = NFS_I(inode)->layout;
427  	if (lo &&  pnfs_layout_is_valid(lo) &&
428  	    nfs4_stateid_match_other(dst, &lo->plh_stateid)) {
429  		/* Is our call using the most recent seqid? If so, bump it */
430  		if (!nfs4_stateid_is_newer(&lo->plh_stateid, dst)) {
431  			nfs4_stateid_seqid_inc(dst);
432  			ret = true;
433  			goto out;
434  		}
435  		/* Try to update the seqid to the most recent */
436  		err = pnfs_mark_matching_lsegs_return(lo, &head, &range, 0);
437  		if (err != -EBUSY) {
438  			dst->seqid = lo->plh_stateid.seqid;
439  			*dst_range = range;
440  			ret = true;
441  		}
442  	}
443  out:
444  	spin_unlock(&inode->i_lock);
445  	pnfs_free_lseg_list(&head);
446  	return ret;
447  }
448  
449  /*
450   * Mark a pnfs_layout_hdr and all associated layout segments as invalid
451   *
452   * In order to continue using the pnfs_layout_hdr, a full recovery
453   * is required.
454   * Note that caller must hold inode->i_lock.
455   */
456  int
pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr * lo,struct list_head * lseg_list)457  pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo,
458  		struct list_head *lseg_list)
459  {
460  	struct pnfs_layout_range range = {
461  		.iomode = IOMODE_ANY,
462  		.offset = 0,
463  		.length = NFS4_MAX_UINT64,
464  	};
465  	struct pnfs_layout_segment *lseg, *next;
466  
467  	set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
468  	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
469  		pnfs_clear_lseg_state(lseg, lseg_list);
470  	pnfs_clear_layoutreturn_info(lo);
471  	pnfs_free_returned_lsegs(lo, lseg_list, &range, 0);
472  	set_bit(NFS_LAYOUT_DRAIN, &lo->plh_flags);
473  	if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) &&
474  	    !test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
475  		pnfs_clear_layoutreturn_waitbit(lo);
476  	return !list_empty(&lo->plh_segs);
477  }
478  
479  static int
pnfs_iomode_to_fail_bit(u32 iomode)480  pnfs_iomode_to_fail_bit(u32 iomode)
481  {
482  	return iomode == IOMODE_RW ?
483  		NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
484  }
485  
486  static void
pnfs_layout_set_fail_bit(struct pnfs_layout_hdr * lo,int fail_bit)487  pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
488  {
489  	lo->plh_retry_timestamp = jiffies;
490  	if (!test_and_set_bit(fail_bit, &lo->plh_flags))
491  		refcount_inc(&lo->plh_refcount);
492  }
493  
494  static void
pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr * lo,int fail_bit)495  pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
496  {
497  	if (test_and_clear_bit(fail_bit, &lo->plh_flags))
498  		refcount_dec(&lo->plh_refcount);
499  }
500  
501  static void
pnfs_layout_io_set_failed(struct pnfs_layout_hdr * lo,u32 iomode)502  pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
503  {
504  	struct inode *inode = lo->plh_inode;
505  	struct pnfs_layout_range range = {
506  		.iomode = iomode,
507  		.offset = 0,
508  		.length = NFS4_MAX_UINT64,
509  	};
510  	LIST_HEAD(head);
511  
512  	spin_lock(&inode->i_lock);
513  	pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
514  	pnfs_mark_matching_lsegs_return(lo, &head, &range, 0);
515  	spin_unlock(&inode->i_lock);
516  	pnfs_free_lseg_list(&head);
517  	dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
518  			iomode == IOMODE_RW ?  "RW" : "READ");
519  }
520  
521  static bool
pnfs_layout_io_test_failed(struct pnfs_layout_hdr * lo,u32 iomode)522  pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
523  {
524  	unsigned long start, end;
525  	int fail_bit = pnfs_iomode_to_fail_bit(iomode);
526  
527  	if (test_bit(fail_bit, &lo->plh_flags) == 0)
528  		return false;
529  	end = jiffies;
530  	start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
531  	if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
532  		/* It is time to retry the failed layoutgets */
533  		pnfs_layout_clear_fail_bit(lo, fail_bit);
534  		return false;
535  	}
536  	return true;
537  }
538  
539  static void
pnfs_init_lseg(struct pnfs_layout_hdr * lo,struct pnfs_layout_segment * lseg,const struct pnfs_layout_range * range,const nfs4_stateid * stateid)540  pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg,
541  		const struct pnfs_layout_range *range,
542  		const nfs4_stateid *stateid)
543  {
544  	INIT_LIST_HEAD(&lseg->pls_list);
545  	INIT_LIST_HEAD(&lseg->pls_lc_list);
546  	INIT_LIST_HEAD(&lseg->pls_commits);
547  	refcount_set(&lseg->pls_refcount, 1);
548  	set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
549  	lseg->pls_layout = lo;
550  	lseg->pls_range = *range;
551  	lseg->pls_seq = be32_to_cpu(stateid->seqid);
552  }
553  
pnfs_free_lseg(struct pnfs_layout_segment * lseg)554  static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
555  {
556  	if (lseg != NULL) {
557  		struct inode *inode = lseg->pls_layout->plh_inode;
558  		NFS_SERVER(inode)->pnfs_curr_ld->free_lseg(lseg);
559  	}
560  }
561  
562  static void
pnfs_layout_remove_lseg(struct pnfs_layout_hdr * lo,struct pnfs_layout_segment * lseg)563  pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
564  		struct pnfs_layout_segment *lseg)
565  {
566  	WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
567  	list_del_init(&lseg->pls_list);
568  	/* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
569  	refcount_dec(&lo->plh_refcount);
570  	if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
571  		return;
572  	if (list_empty(&lo->plh_segs) &&
573  	    !test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
574  	    !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
575  		if (atomic_read(&lo->plh_outstanding) == 0)
576  			set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
577  		clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
578  	}
579  }
580  
581  static bool
pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr * lo,struct pnfs_layout_segment * lseg)582  pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr *lo,
583  		struct pnfs_layout_segment *lseg)
584  {
585  	if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
586  	    pnfs_layout_is_valid(lo)) {
587  		pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
588  		list_move_tail(&lseg->pls_list, &lo->plh_return_segs);
589  		return true;
590  	}
591  	return false;
592  }
593  
594  void
pnfs_put_lseg(struct pnfs_layout_segment * lseg)595  pnfs_put_lseg(struct pnfs_layout_segment *lseg)
596  {
597  	struct pnfs_layout_hdr *lo;
598  	struct inode *inode;
599  
600  	if (!lseg)
601  		return;
602  
603  	dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
604  		refcount_read(&lseg->pls_refcount),
605  		test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
606  
607  	lo = lseg->pls_layout;
608  	inode = lo->plh_inode;
609  
610  	if (refcount_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
611  		pnfs_get_layout_hdr(lo);
612  		pnfs_layout_remove_lseg(lo, lseg);
613  		if (pnfs_cache_lseg_for_layoutreturn(lo, lseg))
614  			lseg = NULL;
615  		spin_unlock(&inode->i_lock);
616  		pnfs_free_lseg(lseg);
617  		pnfs_put_layout_hdr(lo);
618  	}
619  }
620  EXPORT_SYMBOL_GPL(pnfs_put_lseg);
621  
622  /*
623   * is l2 fully contained in l1?
624   *   start1                             end1
625   *   [----------------------------------)
626   *           start2           end2
627   *           [----------------)
628   */
629  static bool
pnfs_lseg_range_contained(const struct pnfs_layout_range * l1,const struct pnfs_layout_range * l2)630  pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
631  		 const struct pnfs_layout_range *l2)
632  {
633  	u64 start1 = l1->offset;
634  	u64 end1 = pnfs_end_offset(start1, l1->length);
635  	u64 start2 = l2->offset;
636  	u64 end2 = pnfs_end_offset(start2, l2->length);
637  
638  	return (start1 <= start2) && (end1 >= end2);
639  }
640  
pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment * lseg,struct list_head * tmp_list)641  static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
642  		struct list_head *tmp_list)
643  {
644  	if (!refcount_dec_and_test(&lseg->pls_refcount))
645  		return false;
646  	pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
647  	list_add(&lseg->pls_list, tmp_list);
648  	return true;
649  }
650  
651  /* Returns 1 if lseg is removed from list, 0 otherwise */
mark_lseg_invalid(struct pnfs_layout_segment * lseg,struct list_head * tmp_list)652  static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
653  			     struct list_head *tmp_list)
654  {
655  	int rv = 0;
656  
657  	if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
658  		/* Remove the reference keeping the lseg in the
659  		 * list.  It will now be removed when all
660  		 * outstanding io is finished.
661  		 */
662  		dprintk("%s: lseg %p ref %d\n", __func__, lseg,
663  			refcount_read(&lseg->pls_refcount));
664  		if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
665  			rv = 1;
666  	}
667  	return rv;
668  }
669  
670  static bool
pnfs_should_free_range(const struct pnfs_layout_range * lseg_range,const struct pnfs_layout_range * recall_range)671  pnfs_should_free_range(const struct pnfs_layout_range *lseg_range,
672  		 const struct pnfs_layout_range *recall_range)
673  {
674  	return (recall_range->iomode == IOMODE_ANY ||
675  		lseg_range->iomode == recall_range->iomode) &&
676  	       pnfs_lseg_range_intersecting(lseg_range, recall_range);
677  }
678  
679  static bool
pnfs_match_lseg_recall(const struct pnfs_layout_segment * lseg,const struct pnfs_layout_range * recall_range,u32 seq)680  pnfs_match_lseg_recall(const struct pnfs_layout_segment *lseg,
681  		const struct pnfs_layout_range *recall_range,
682  		u32 seq)
683  {
684  	if (seq != 0 && pnfs_seqid_is_newer(lseg->pls_seq, seq))
685  		return false;
686  	if (recall_range == NULL)
687  		return true;
688  	return pnfs_should_free_range(&lseg->pls_range, recall_range);
689  }
690  
691  /**
692   * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
693   * @lo: layout header containing the lsegs
694   * @tmp_list: list head where doomed lsegs should go
695   * @recall_range: optional recall range argument to match (may be NULL)
696   * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
697   *
698   * Walk the list of lsegs in the layout header, and tear down any that should
699   * be destroyed. If "recall_range" is specified then the segment must match
700   * that range. If "seq" is non-zero, then only match segments that were handed
701   * out at or before that sequence.
702   *
703   * Returns number of matching invalid lsegs remaining in list after scanning
704   * it and purging them.
705   */
706  int
pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr * lo,struct list_head * tmp_list,const struct pnfs_layout_range * recall_range,u32 seq)707  pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
708  			    struct list_head *tmp_list,
709  			    const struct pnfs_layout_range *recall_range,
710  			    u32 seq)
711  {
712  	struct pnfs_layout_segment *lseg, *next;
713  	struct nfs_server *server = NFS_SERVER(lo->plh_inode);
714  	int remaining = 0;
715  
716  	dprintk("%s:Begin lo %p\n", __func__, lo);
717  
718  	if (list_empty(&lo->plh_segs))
719  		return 0;
720  	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
721  		if (pnfs_match_lseg_recall(lseg, recall_range, seq)) {
722  			dprintk("%s: freeing lseg %p iomode %d seq %u "
723  				"offset %llu length %llu\n", __func__,
724  				lseg, lseg->pls_range.iomode, lseg->pls_seq,
725  				lseg->pls_range.offset, lseg->pls_range.length);
726  			if (mark_lseg_invalid(lseg, tmp_list))
727  				continue;
728  			remaining++;
729  			pnfs_lseg_cancel_io(server, lseg);
730  		}
731  	dprintk("%s:Return %i\n", __func__, remaining);
732  	return remaining;
733  }
734  
735  static void
pnfs_free_returned_lsegs(struct pnfs_layout_hdr * lo,struct list_head * free_me,const struct pnfs_layout_range * range,u32 seq)736  pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
737  		struct list_head *free_me,
738  		const struct pnfs_layout_range *range,
739  		u32 seq)
740  {
741  	struct pnfs_layout_segment *lseg, *next;
742  
743  	list_for_each_entry_safe(lseg, next, &lo->plh_return_segs, pls_list) {
744  		if (pnfs_match_lseg_recall(lseg, range, seq))
745  			list_move_tail(&lseg->pls_list, free_me);
746  	}
747  }
748  
749  /* note free_me must contain lsegs from a single layout_hdr */
750  void
pnfs_free_lseg_list(struct list_head * free_me)751  pnfs_free_lseg_list(struct list_head *free_me)
752  {
753  	struct pnfs_layout_segment *lseg, *tmp;
754  
755  	if (list_empty(free_me))
756  		return;
757  
758  	list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
759  		list_del(&lseg->pls_list);
760  		pnfs_free_lseg(lseg);
761  	}
762  }
763  
__pnfs_destroy_layout(struct nfs_inode * nfsi)764  static struct pnfs_layout_hdr *__pnfs_destroy_layout(struct nfs_inode *nfsi)
765  {
766  	struct pnfs_layout_hdr *lo;
767  	LIST_HEAD(tmp_list);
768  
769  	spin_lock(&nfsi->vfs_inode.i_lock);
770  	lo = nfsi->layout;
771  	if (lo) {
772  		pnfs_get_layout_hdr(lo);
773  		pnfs_mark_layout_stateid_invalid(lo, &tmp_list);
774  		pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
775  		pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
776  		spin_unlock(&nfsi->vfs_inode.i_lock);
777  		pnfs_free_lseg_list(&tmp_list);
778  		nfs_commit_inode(&nfsi->vfs_inode, 0);
779  		pnfs_put_layout_hdr(lo);
780  	} else
781  		spin_unlock(&nfsi->vfs_inode.i_lock);
782  	return lo;
783  }
784  
pnfs_destroy_layout(struct nfs_inode * nfsi)785  void pnfs_destroy_layout(struct nfs_inode *nfsi)
786  {
787  	__pnfs_destroy_layout(nfsi);
788  }
789  EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
790  
pnfs_layout_removed(struct nfs_inode * nfsi,struct pnfs_layout_hdr * lo)791  static bool pnfs_layout_removed(struct nfs_inode *nfsi,
792  				struct pnfs_layout_hdr *lo)
793  {
794  	bool ret;
795  
796  	spin_lock(&nfsi->vfs_inode.i_lock);
797  	ret = nfsi->layout != lo;
798  	spin_unlock(&nfsi->vfs_inode.i_lock);
799  	return ret;
800  }
801  
pnfs_destroy_layout_final(struct nfs_inode * nfsi)802  void pnfs_destroy_layout_final(struct nfs_inode *nfsi)
803  {
804  	struct pnfs_layout_hdr *lo = __pnfs_destroy_layout(nfsi);
805  
806  	if (lo)
807  		wait_var_event(lo, pnfs_layout_removed(nfsi, lo));
808  }
809  
810  static bool
pnfs_layout_add_bulk_destroy_list(struct inode * inode,struct list_head * layout_list)811  pnfs_layout_add_bulk_destroy_list(struct inode *inode,
812  		struct list_head *layout_list)
813  {
814  	struct pnfs_layout_hdr *lo;
815  	bool ret = false;
816  
817  	spin_lock(&inode->i_lock);
818  	lo = NFS_I(inode)->layout;
819  	if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
820  		pnfs_get_layout_hdr(lo);
821  		list_add(&lo->plh_bulk_destroy, layout_list);
822  		ret = true;
823  	}
824  	spin_unlock(&inode->i_lock);
825  	return ret;
826  }
827  
828  /* Caller must hold rcu_read_lock and clp->cl_lock */
829  static int
pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client * clp,struct nfs_server * server,struct list_head * layout_list)830  pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
831  		struct nfs_server *server,
832  		struct list_head *layout_list)
833  	__must_hold(&clp->cl_lock)
834  	__must_hold(RCU)
835  {
836  	struct pnfs_layout_hdr *lo, *next;
837  	struct inode *inode;
838  
839  	list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
840  		if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags) ||
841  		    test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) ||
842  		    !list_empty(&lo->plh_bulk_destroy))
843  			continue;
844  		/* If the sb is being destroyed, just bail */
845  		if (!nfs_sb_active(server->super))
846  			break;
847  		inode = pnfs_grab_inode_layout_hdr(lo);
848  		if (inode != NULL) {
849  			if (test_and_clear_bit(NFS_LAYOUT_HASHED, &lo->plh_flags))
850  				list_del_rcu(&lo->plh_layouts);
851  			if (pnfs_layout_add_bulk_destroy_list(inode,
852  						layout_list))
853  				continue;
854  			rcu_read_unlock();
855  			spin_unlock(&clp->cl_lock);
856  			iput(inode);
857  		} else {
858  			rcu_read_unlock();
859  			spin_unlock(&clp->cl_lock);
860  		}
861  		nfs_sb_deactive(server->super);
862  		spin_lock(&clp->cl_lock);
863  		rcu_read_lock();
864  		return -EAGAIN;
865  	}
866  	return 0;
867  }
868  
869  static int
pnfs_layout_free_bulk_destroy_list(struct list_head * layout_list,bool is_bulk_recall)870  pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
871  		bool is_bulk_recall)
872  {
873  	struct pnfs_layout_hdr *lo;
874  	struct inode *inode;
875  	LIST_HEAD(lseg_list);
876  	int ret = 0;
877  
878  	while (!list_empty(layout_list)) {
879  		lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
880  				plh_bulk_destroy);
881  		dprintk("%s freeing layout for inode %lu\n", __func__,
882  			lo->plh_inode->i_ino);
883  		inode = lo->plh_inode;
884  
885  		pnfs_layoutcommit_inode(inode, false);
886  
887  		spin_lock(&inode->i_lock);
888  		list_del_init(&lo->plh_bulk_destroy);
889  		if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) {
890  			if (is_bulk_recall)
891  				set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
892  			ret = -EAGAIN;
893  		}
894  		spin_unlock(&inode->i_lock);
895  		pnfs_free_lseg_list(&lseg_list);
896  		/* Free all lsegs that are attached to commit buckets */
897  		nfs_commit_inode(inode, 0);
898  		pnfs_put_layout_hdr(lo);
899  		nfs_iput_and_deactive(inode);
900  	}
901  	return ret;
902  }
903  
904  int
pnfs_destroy_layouts_byfsid(struct nfs_client * clp,struct nfs_fsid * fsid,bool is_recall)905  pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
906  		struct nfs_fsid *fsid,
907  		bool is_recall)
908  {
909  	struct nfs_server *server;
910  	LIST_HEAD(layout_list);
911  
912  	spin_lock(&clp->cl_lock);
913  	rcu_read_lock();
914  restart:
915  	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
916  		if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
917  			continue;
918  		if (pnfs_layout_bulk_destroy_byserver_locked(clp,
919  				server,
920  				&layout_list) != 0)
921  			goto restart;
922  	}
923  	rcu_read_unlock();
924  	spin_unlock(&clp->cl_lock);
925  
926  	if (list_empty(&layout_list))
927  		return 0;
928  	return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
929  }
930  
931  int
pnfs_destroy_layouts_byclid(struct nfs_client * clp,bool is_recall)932  pnfs_destroy_layouts_byclid(struct nfs_client *clp,
933  		bool is_recall)
934  {
935  	struct nfs_server *server;
936  	LIST_HEAD(layout_list);
937  
938  	spin_lock(&clp->cl_lock);
939  	rcu_read_lock();
940  restart:
941  	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
942  		if (pnfs_layout_bulk_destroy_byserver_locked(clp,
943  					server,
944  					&layout_list) != 0)
945  			goto restart;
946  	}
947  	rcu_read_unlock();
948  	spin_unlock(&clp->cl_lock);
949  
950  	if (list_empty(&layout_list))
951  		return 0;
952  	return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
953  }
954  
955  /*
956   * Called by the state manager to remove all layouts established under an
957   * expired lease.
958   */
959  void
pnfs_destroy_all_layouts(struct nfs_client * clp)960  pnfs_destroy_all_layouts(struct nfs_client *clp)
961  {
962  	nfs4_deviceid_mark_client_invalid(clp);
963  	nfs4_deviceid_purge_client(clp);
964  
965  	pnfs_destroy_layouts_byclid(clp, false);
966  }
967  
968  static void
pnfs_set_layout_cred(struct pnfs_layout_hdr * lo,const struct cred * cred)969  pnfs_set_layout_cred(struct pnfs_layout_hdr *lo, const struct cred *cred)
970  {
971  	const struct cred *old;
972  
973  	if (cred && cred_fscmp(lo->plh_lc_cred, cred) != 0) {
974  		old = xchg(&lo->plh_lc_cred, get_cred(cred));
975  		put_cred(old);
976  	}
977  }
978  
979  /* update lo->plh_stateid with new if is more recent */
980  void
pnfs_set_layout_stateid(struct pnfs_layout_hdr * lo,const nfs4_stateid * new,const struct cred * cred,bool update_barrier)981  pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
982  			const struct cred *cred, bool update_barrier)
983  {
984  	u32 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
985  	u32 newseq = be32_to_cpu(new->seqid);
986  
987  	if (!pnfs_layout_is_valid(lo)) {
988  		pnfs_set_layout_cred(lo, cred);
989  		nfs4_stateid_copy(&lo->plh_stateid, new);
990  		lo->plh_barrier = newseq;
991  		pnfs_clear_layoutreturn_info(lo);
992  		clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
993  		return;
994  	}
995  
996  	if (pnfs_seqid_is_newer(newseq, oldseq))
997  		nfs4_stateid_copy(&lo->plh_stateid, new);
998  
999  	if (update_barrier) {
1000  		pnfs_barrier_update(lo, newseq);
1001  		return;
1002  	}
1003  	/*
1004  	 * Because of wraparound, we want to keep the barrier
1005  	 * "close" to the current seqids. We really only want to
1006  	 * get here from a layoutget call.
1007  	 */
1008  	if (atomic_read(&lo->plh_outstanding) == 1)
1009  		 pnfs_barrier_update(lo, be32_to_cpu(lo->plh_stateid.seqid));
1010  }
1011  
1012  static bool
pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr * lo,const nfs4_stateid * stateid)1013  pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
1014  		const nfs4_stateid *stateid)
1015  {
1016  	u32 seqid = be32_to_cpu(stateid->seqid);
1017  
1018  	return lo->plh_barrier && pnfs_seqid_is_newer(lo->plh_barrier, seqid);
1019  }
1020  
1021  /* lget is set to 1 if called from inside send_layoutget call chain */
1022  static bool
pnfs_layoutgets_blocked(const struct pnfs_layout_hdr * lo)1023  pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
1024  {
1025  	return lo->plh_block_lgets ||
1026  		test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
1027  }
1028  
1029  static struct nfs_server *
pnfs_find_server(struct inode * inode,struct nfs_open_context * ctx)1030  pnfs_find_server(struct inode *inode, struct nfs_open_context *ctx)
1031  {
1032  	struct nfs_server *server;
1033  
1034  	if (inode) {
1035  		server = NFS_SERVER(inode);
1036  	} else {
1037  		struct dentry *parent_dir = dget_parent(ctx->dentry);
1038  		server = NFS_SERVER(parent_dir->d_inode);
1039  		dput(parent_dir);
1040  	}
1041  	return server;
1042  }
1043  
nfs4_free_pages(struct page ** pages,size_t size)1044  static void nfs4_free_pages(struct page **pages, size_t size)
1045  {
1046  	int i;
1047  
1048  	if (!pages)
1049  		return;
1050  
1051  	for (i = 0; i < size; i++) {
1052  		if (!pages[i])
1053  			break;
1054  		__free_page(pages[i]);
1055  	}
1056  	kfree(pages);
1057  }
1058  
nfs4_alloc_pages(size_t size,gfp_t gfp_flags)1059  static struct page **nfs4_alloc_pages(size_t size, gfp_t gfp_flags)
1060  {
1061  	struct page **pages;
1062  	int i;
1063  
1064  	pages = kmalloc_array(size, sizeof(struct page *), gfp_flags);
1065  	if (!pages) {
1066  		dprintk("%s: can't alloc array of %zu pages\n", __func__, size);
1067  		return NULL;
1068  	}
1069  
1070  	for (i = 0; i < size; i++) {
1071  		pages[i] = alloc_page(gfp_flags);
1072  		if (!pages[i]) {
1073  			dprintk("%s: failed to allocate page\n", __func__);
1074  			nfs4_free_pages(pages, i);
1075  			return NULL;
1076  		}
1077  	}
1078  
1079  	return pages;
1080  }
1081  
1082  static struct nfs4_layoutget *
pnfs_alloc_init_layoutget_args(struct inode * ino,struct nfs_open_context * ctx,const nfs4_stateid * stateid,const struct pnfs_layout_range * range,gfp_t gfp_flags)1083  pnfs_alloc_init_layoutget_args(struct inode *ino,
1084  	   struct nfs_open_context *ctx,
1085  	   const nfs4_stateid *stateid,
1086  	   const struct pnfs_layout_range *range,
1087  	   gfp_t gfp_flags)
1088  {
1089  	struct nfs_server *server = pnfs_find_server(ino, ctx);
1090  	size_t max_reply_sz = server->pnfs_curr_ld->max_layoutget_response;
1091  	size_t max_pages = max_response_pages(server);
1092  	struct nfs4_layoutget *lgp;
1093  
1094  	dprintk("--> %s\n", __func__);
1095  
1096  	lgp = kzalloc(sizeof(*lgp), gfp_flags);
1097  	if (lgp == NULL)
1098  		return NULL;
1099  
1100  	if (max_reply_sz) {
1101  		size_t npages = (max_reply_sz + PAGE_SIZE - 1) >> PAGE_SHIFT;
1102  		if (npages < max_pages)
1103  			max_pages = npages;
1104  	}
1105  
1106  	lgp->args.layout.pages = nfs4_alloc_pages(max_pages, gfp_flags);
1107  	if (!lgp->args.layout.pages) {
1108  		kfree(lgp);
1109  		return NULL;
1110  	}
1111  	lgp->args.layout.pglen = max_pages * PAGE_SIZE;
1112  	lgp->res.layoutp = &lgp->args.layout;
1113  
1114  	/* Don't confuse uninitialised result and success */
1115  	lgp->res.status = -NFS4ERR_DELAY;
1116  
1117  	lgp->args.minlength = PAGE_SIZE;
1118  	if (lgp->args.minlength > range->length)
1119  		lgp->args.minlength = range->length;
1120  	if (ino) {
1121  		loff_t i_size = i_size_read(ino);
1122  
1123  		if (range->iomode == IOMODE_READ) {
1124  			if (range->offset >= i_size)
1125  				lgp->args.minlength = 0;
1126  			else if (i_size - range->offset < lgp->args.minlength)
1127  				lgp->args.minlength = i_size - range->offset;
1128  		}
1129  	}
1130  	lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
1131  	pnfs_copy_range(&lgp->args.range, range);
1132  	lgp->args.type = server->pnfs_curr_ld->id;
1133  	lgp->args.inode = ino;
1134  	lgp->args.ctx = get_nfs_open_context(ctx);
1135  	nfs4_stateid_copy(&lgp->args.stateid, stateid);
1136  	lgp->gfp_flags = gfp_flags;
1137  	lgp->cred = ctx->cred;
1138  	return lgp;
1139  }
1140  
pnfs_layoutget_free(struct nfs4_layoutget * lgp)1141  void pnfs_layoutget_free(struct nfs4_layoutget *lgp)
1142  {
1143  	size_t max_pages = lgp->args.layout.pglen / PAGE_SIZE;
1144  
1145  	nfs4_free_pages(lgp->args.layout.pages, max_pages);
1146  	pnfs_put_layout_hdr(lgp->lo);
1147  	put_nfs_open_context(lgp->args.ctx);
1148  	kfree(lgp);
1149  }
1150  
pnfs_clear_layoutcommit(struct inode * inode,struct list_head * head)1151  static void pnfs_clear_layoutcommit(struct inode *inode,
1152  		struct list_head *head)
1153  {
1154  	struct nfs_inode *nfsi = NFS_I(inode);
1155  	struct pnfs_layout_segment *lseg, *tmp;
1156  
1157  	if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1158  		return;
1159  	list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
1160  		if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1161  			continue;
1162  		pnfs_lseg_dec_and_remove_zero(lseg, head);
1163  	}
1164  }
1165  
pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr * lo,const nfs4_stateid * arg_stateid,const struct pnfs_layout_range * range,const nfs4_stateid * stateid)1166  void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo,
1167  		const nfs4_stateid *arg_stateid,
1168  		const struct pnfs_layout_range *range,
1169  		const nfs4_stateid *stateid)
1170  {
1171  	struct inode *inode = lo->plh_inode;
1172  	LIST_HEAD(freeme);
1173  
1174  	spin_lock(&inode->i_lock);
1175  	if (!nfs4_stateid_match_other(&lo->plh_stateid, arg_stateid))
1176  		goto out_unlock;
1177  	if (stateid && pnfs_layout_is_valid(lo)) {
1178  		u32 seq = be32_to_cpu(arg_stateid->seqid);
1179  
1180  		pnfs_mark_matching_lsegs_invalid(lo, &freeme, range, seq);
1181  		pnfs_free_returned_lsegs(lo, &freeme, range, seq);
1182  		pnfs_set_layout_stateid(lo, stateid, NULL, true);
1183  	} else
1184  		pnfs_mark_layout_stateid_invalid(lo, &freeme);
1185  out_unlock:
1186  	pnfs_clear_layoutreturn_waitbit(lo);
1187  	spin_unlock(&inode->i_lock);
1188  	pnfs_free_lseg_list(&freeme);
1189  
1190  }
1191  
1192  static bool
pnfs_prepare_layoutreturn(struct pnfs_layout_hdr * lo,nfs4_stateid * stateid,const struct cred ** cred,enum pnfs_iomode * iomode)1193  pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo,
1194  		nfs4_stateid *stateid,
1195  		const struct cred **cred,
1196  		enum pnfs_iomode *iomode)
1197  {
1198  	/* Serialise LAYOUTGET/LAYOUTRETURN */
1199  	if (atomic_read(&lo->plh_outstanding) != 0 && lo->plh_return_seq == 0)
1200  		return false;
1201  	if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
1202  		return false;
1203  	set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
1204  	pnfs_get_layout_hdr(lo);
1205  	nfs4_stateid_copy(stateid, &lo->plh_stateid);
1206  	*cred = get_cred(lo->plh_lc_cred);
1207  	if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
1208  		if (lo->plh_return_seq != 0)
1209  			stateid->seqid = cpu_to_be32(lo->plh_return_seq);
1210  		if (iomode != NULL)
1211  			*iomode = lo->plh_return_iomode;
1212  		pnfs_clear_layoutreturn_info(lo);
1213  	} else if (iomode != NULL)
1214  		*iomode = IOMODE_ANY;
1215  	pnfs_barrier_update(lo, be32_to_cpu(stateid->seqid));
1216  	return true;
1217  }
1218  
1219  static void
pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args * args,struct pnfs_layout_hdr * lo,const nfs4_stateid * stateid,enum pnfs_iomode iomode)1220  pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args *args,
1221  		struct pnfs_layout_hdr *lo,
1222  		const nfs4_stateid *stateid,
1223  		enum pnfs_iomode iomode)
1224  {
1225  	struct inode *inode = lo->plh_inode;
1226  
1227  	args->layout_type = NFS_SERVER(inode)->pnfs_curr_ld->id;
1228  	args->inode = inode;
1229  	args->range.iomode = iomode;
1230  	args->range.offset = 0;
1231  	args->range.length = NFS4_MAX_UINT64;
1232  	args->layout = lo;
1233  	nfs4_stateid_copy(&args->stateid, stateid);
1234  }
1235  
1236  static int
pnfs_send_layoutreturn(struct pnfs_layout_hdr * lo,const nfs4_stateid * stateid,const struct cred ** pcred,enum pnfs_iomode iomode,bool sync)1237  pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo,
1238  		       const nfs4_stateid *stateid,
1239  		       const struct cred **pcred,
1240  		       enum pnfs_iomode iomode,
1241  		       bool sync)
1242  {
1243  	struct inode *ino = lo->plh_inode;
1244  	struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1245  	struct nfs4_layoutreturn *lrp;
1246  	const struct cred *cred = *pcred;
1247  	int status = 0;
1248  
1249  	*pcred = NULL;
1250  	lrp = kzalloc(sizeof(*lrp), nfs_io_gfp_mask());
1251  	if (unlikely(lrp == NULL)) {
1252  		status = -ENOMEM;
1253  		spin_lock(&ino->i_lock);
1254  		pnfs_clear_layoutreturn_waitbit(lo);
1255  		spin_unlock(&ino->i_lock);
1256  		put_cred(cred);
1257  		pnfs_put_layout_hdr(lo);
1258  		goto out;
1259  	}
1260  
1261  	pnfs_init_layoutreturn_args(&lrp->args, lo, stateid, iomode);
1262  	lrp->args.ld_private = &lrp->ld_private;
1263  	lrp->clp = NFS_SERVER(ino)->nfs_client;
1264  	lrp->cred = cred;
1265  	if (ld->prepare_layoutreturn)
1266  		ld->prepare_layoutreturn(&lrp->args);
1267  
1268  	status = nfs4_proc_layoutreturn(lrp, sync);
1269  out:
1270  	dprintk("<-- %s status: %d\n", __func__, status);
1271  	return status;
1272  }
1273  
1274  static bool
pnfs_layout_segments_returnable(struct pnfs_layout_hdr * lo,enum pnfs_iomode iomode,u32 seq)1275  pnfs_layout_segments_returnable(struct pnfs_layout_hdr *lo,
1276  				enum pnfs_iomode iomode,
1277  				u32 seq)
1278  {
1279  	struct pnfs_layout_range recall_range = {
1280  		.length = NFS4_MAX_UINT64,
1281  		.iomode = iomode,
1282  	};
1283  	return pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs,
1284  					       &recall_range, seq) != -EBUSY;
1285  }
1286  
1287  /* Return true if layoutreturn is needed */
1288  static bool
pnfs_layout_need_return(struct pnfs_layout_hdr * lo)1289  pnfs_layout_need_return(struct pnfs_layout_hdr *lo)
1290  {
1291  	if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1292  		return false;
1293  	return pnfs_layout_segments_returnable(lo, lo->plh_return_iomode,
1294  					       lo->plh_return_seq);
1295  }
1296  
pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr * lo)1297  static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo)
1298  {
1299  	struct inode *inode= lo->plh_inode;
1300  
1301  	if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1302  		return;
1303  	spin_lock(&inode->i_lock);
1304  	if (pnfs_layout_need_return(lo)) {
1305  		const struct cred *cred;
1306  		nfs4_stateid stateid;
1307  		enum pnfs_iomode iomode;
1308  		bool send;
1309  
1310  		send = pnfs_prepare_layoutreturn(lo, &stateid, &cred, &iomode);
1311  		spin_unlock(&inode->i_lock);
1312  		if (send) {
1313  			/* Send an async layoutreturn so we dont deadlock */
1314  			pnfs_send_layoutreturn(lo, &stateid, &cred, iomode, false);
1315  		}
1316  	} else
1317  		spin_unlock(&inode->i_lock);
1318  }
1319  
1320  /*
1321   * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
1322   * when the layout segment list is empty.
1323   *
1324   * Note that a pnfs_layout_hdr can exist with an empty layout segment
1325   * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
1326   * deviceid is marked invalid.
1327   */
1328  int
_pnfs_return_layout(struct inode * ino)1329  _pnfs_return_layout(struct inode *ino)
1330  {
1331  	struct pnfs_layout_hdr *lo = NULL;
1332  	struct nfs_inode *nfsi = NFS_I(ino);
1333  	struct pnfs_layout_range range = {
1334  		.iomode		= IOMODE_ANY,
1335  		.offset		= 0,
1336  		.length		= NFS4_MAX_UINT64,
1337  	};
1338  	LIST_HEAD(tmp_list);
1339  	const struct cred *cred;
1340  	nfs4_stateid stateid;
1341  	int status = 0;
1342  	bool send, valid_layout;
1343  
1344  	dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
1345  
1346  	spin_lock(&ino->i_lock);
1347  	lo = nfsi->layout;
1348  	if (!lo) {
1349  		spin_unlock(&ino->i_lock);
1350  		dprintk("NFS: %s no layout to return\n", __func__);
1351  		goto out;
1352  	}
1353  	/* Reference matched in nfs4_layoutreturn_release */
1354  	pnfs_get_layout_hdr(lo);
1355  	/* Is there an outstanding layoutreturn ? */
1356  	if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1357  		spin_unlock(&ino->i_lock);
1358  		if (wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1359  					TASK_UNINTERRUPTIBLE))
1360  			goto out_put_layout_hdr;
1361  		spin_lock(&ino->i_lock);
1362  	}
1363  	valid_layout = pnfs_layout_is_valid(lo);
1364  	pnfs_clear_layoutcommit(ino, &tmp_list);
1365  	pnfs_mark_matching_lsegs_return(lo, &tmp_list, &range, 0);
1366  
1367  	if (NFS_SERVER(ino)->pnfs_curr_ld->return_range)
1368  		NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
1369  
1370  	/* Don't send a LAYOUTRETURN if list was initially empty */
1371  	if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) ||
1372  			!valid_layout) {
1373  		spin_unlock(&ino->i_lock);
1374  		dprintk("NFS: %s no layout segments to return\n", __func__);
1375  		goto out_wait_layoutreturn;
1376  	}
1377  
1378  	send = pnfs_prepare_layoutreturn(lo, &stateid, &cred, NULL);
1379  	spin_unlock(&ino->i_lock);
1380  	if (send)
1381  		status = pnfs_send_layoutreturn(lo, &stateid, &cred, IOMODE_ANY, true);
1382  out_wait_layoutreturn:
1383  	wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN, TASK_UNINTERRUPTIBLE);
1384  out_put_layout_hdr:
1385  	pnfs_free_lseg_list(&tmp_list);
1386  	pnfs_put_layout_hdr(lo);
1387  out:
1388  	dprintk("<-- %s status: %d\n", __func__, status);
1389  	return status;
1390  }
1391  
1392  int
pnfs_commit_and_return_layout(struct inode * inode)1393  pnfs_commit_and_return_layout(struct inode *inode)
1394  {
1395  	struct pnfs_layout_hdr *lo;
1396  	int ret;
1397  
1398  	spin_lock(&inode->i_lock);
1399  	lo = NFS_I(inode)->layout;
1400  	if (lo == NULL) {
1401  		spin_unlock(&inode->i_lock);
1402  		return 0;
1403  	}
1404  	pnfs_get_layout_hdr(lo);
1405  	/* Block new layoutgets and read/write to ds */
1406  	lo->plh_block_lgets++;
1407  	spin_unlock(&inode->i_lock);
1408  	filemap_fdatawait(inode->i_mapping);
1409  	ret = pnfs_layoutcommit_inode(inode, true);
1410  	if (ret == 0)
1411  		ret = _pnfs_return_layout(inode);
1412  	spin_lock(&inode->i_lock);
1413  	lo->plh_block_lgets--;
1414  	spin_unlock(&inode->i_lock);
1415  	pnfs_put_layout_hdr(lo);
1416  	return ret;
1417  }
1418  
pnfs_roc(struct inode * ino,struct nfs4_layoutreturn_args * args,struct nfs4_layoutreturn_res * res,const struct cred * cred)1419  bool pnfs_roc(struct inode *ino,
1420  		struct nfs4_layoutreturn_args *args,
1421  		struct nfs4_layoutreturn_res *res,
1422  		const struct cred *cred)
1423  {
1424  	struct nfs_inode *nfsi = NFS_I(ino);
1425  	struct nfs_open_context *ctx;
1426  	struct nfs4_state *state;
1427  	struct pnfs_layout_hdr *lo;
1428  	struct pnfs_layout_segment *lseg, *next;
1429  	const struct cred *lc_cred;
1430  	nfs4_stateid stateid;
1431  	enum pnfs_iomode iomode = 0;
1432  	bool layoutreturn = false, roc = false;
1433  	bool skip_read = false;
1434  
1435  	if (!nfs_have_layout(ino))
1436  		return false;
1437  retry:
1438  	rcu_read_lock();
1439  	spin_lock(&ino->i_lock);
1440  	lo = nfsi->layout;
1441  	if (!lo || !pnfs_layout_is_valid(lo) ||
1442  	    test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1443  		lo = NULL;
1444  		goto out_noroc;
1445  	}
1446  	pnfs_get_layout_hdr(lo);
1447  	if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1448  		spin_unlock(&ino->i_lock);
1449  		rcu_read_unlock();
1450  		wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1451  				TASK_UNINTERRUPTIBLE);
1452  		pnfs_put_layout_hdr(lo);
1453  		goto retry;
1454  	}
1455  
1456  	/* no roc if we hold a delegation */
1457  	if (nfs4_check_delegation(ino, FMODE_READ)) {
1458  		if (nfs4_check_delegation(ino, FMODE_WRITE))
1459  			goto out_noroc;
1460  		skip_read = true;
1461  	}
1462  
1463  	list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1464  		state = ctx->state;
1465  		if (state == NULL)
1466  			continue;
1467  		/* Don't return layout if there is open file state */
1468  		if (state->state & FMODE_WRITE)
1469  			goto out_noroc;
1470  		if (state->state & FMODE_READ)
1471  			skip_read = true;
1472  	}
1473  
1474  
1475  	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) {
1476  		if (skip_read && lseg->pls_range.iomode == IOMODE_READ)
1477  			continue;
1478  		/* If we are sending layoutreturn, invalidate all valid lsegs */
1479  		if (!test_and_clear_bit(NFS_LSEG_ROC, &lseg->pls_flags))
1480  			continue;
1481  		/*
1482  		 * Note: mark lseg for return so pnfs_layout_remove_lseg
1483  		 * doesn't invalidate the layout for us.
1484  		 */
1485  		set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1486  		if (!mark_lseg_invalid(lseg, &lo->plh_return_segs))
1487  			continue;
1488  		pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
1489  	}
1490  
1491  	if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1492  		goto out_noroc;
1493  
1494  	/* ROC in two conditions:
1495  	 * 1. there are ROC lsegs
1496  	 * 2. we don't send layoutreturn
1497  	 */
1498  	/* lo ref dropped in pnfs_roc_release() */
1499  	layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &lc_cred, &iomode);
1500  	/* If the creds don't match, we can't compound the layoutreturn */
1501  	if (!layoutreturn || cred_fscmp(cred, lc_cred) != 0)
1502  		goto out_noroc;
1503  
1504  	roc = layoutreturn;
1505  	pnfs_init_layoutreturn_args(args, lo, &stateid, iomode);
1506  	res->lrs_present = 0;
1507  	layoutreturn = false;
1508  	put_cred(lc_cred);
1509  
1510  out_noroc:
1511  	spin_unlock(&ino->i_lock);
1512  	rcu_read_unlock();
1513  	pnfs_layoutcommit_inode(ino, true);
1514  	if (roc) {
1515  		struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1516  		if (ld->prepare_layoutreturn)
1517  			ld->prepare_layoutreturn(args);
1518  		pnfs_put_layout_hdr(lo);
1519  		return true;
1520  	}
1521  	if (layoutreturn)
1522  		pnfs_send_layoutreturn(lo, &stateid, &lc_cred, iomode, true);
1523  	pnfs_put_layout_hdr(lo);
1524  	return false;
1525  }
1526  
pnfs_roc_done(struct rpc_task * task,struct nfs4_layoutreturn_args ** argpp,struct nfs4_layoutreturn_res ** respp,int * ret)1527  int pnfs_roc_done(struct rpc_task *task, struct nfs4_layoutreturn_args **argpp,
1528  		  struct nfs4_layoutreturn_res **respp, int *ret)
1529  {
1530  	struct nfs4_layoutreturn_args *arg = *argpp;
1531  	int retval = -EAGAIN;
1532  
1533  	if (!arg)
1534  		return 0;
1535  	/* Handle Layoutreturn errors */
1536  	switch (*ret) {
1537  	case 0:
1538  		retval = 0;
1539  		break;
1540  	case -NFS4ERR_NOMATCHING_LAYOUT:
1541  		/* Was there an RPC level error? If not, retry */
1542  		if (task->tk_rpc_status == 0)
1543  			break;
1544  		/* If the call was not sent, let caller handle it */
1545  		if (!RPC_WAS_SENT(task))
1546  			return 0;
1547  		/*
1548  		 * Otherwise, assume the call succeeded and
1549  		 * that we need to release the layout
1550  		 */
1551  		*ret = 0;
1552  		(*respp)->lrs_present = 0;
1553  		retval = 0;
1554  		break;
1555  	case -NFS4ERR_DELAY:
1556  		/* Let the caller handle the retry */
1557  		*ret = -NFS4ERR_NOMATCHING_LAYOUT;
1558  		return 0;
1559  	case -NFS4ERR_OLD_STATEID:
1560  		if (!nfs4_layout_refresh_old_stateid(&arg->stateid,
1561  						     &arg->range, arg->inode))
1562  			break;
1563  		*ret = -NFS4ERR_NOMATCHING_LAYOUT;
1564  		return -EAGAIN;
1565  	}
1566  	*argpp = NULL;
1567  	*respp = NULL;
1568  	return retval;
1569  }
1570  
pnfs_roc_release(struct nfs4_layoutreturn_args * args,struct nfs4_layoutreturn_res * res,int ret)1571  void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
1572  		struct nfs4_layoutreturn_res *res,
1573  		int ret)
1574  {
1575  	struct pnfs_layout_hdr *lo = args->layout;
1576  	struct inode *inode = args->inode;
1577  	const nfs4_stateid *res_stateid = NULL;
1578  	struct nfs4_xdr_opaque_data *ld_private = args->ld_private;
1579  
1580  	switch (ret) {
1581  	case -NFS4ERR_NOMATCHING_LAYOUT:
1582  		spin_lock(&inode->i_lock);
1583  		if (pnfs_layout_is_valid(lo) &&
1584  		    nfs4_stateid_match_other(&args->stateid, &lo->plh_stateid))
1585  			pnfs_set_plh_return_info(lo, args->range.iomode, 0);
1586  		pnfs_clear_layoutreturn_waitbit(lo);
1587  		spin_unlock(&inode->i_lock);
1588  		break;
1589  	case 0:
1590  		if (res->lrs_present)
1591  			res_stateid = &res->stateid;
1592  		fallthrough;
1593  	default:
1594  		pnfs_layoutreturn_free_lsegs(lo, &args->stateid, &args->range,
1595  					     res_stateid);
1596  	}
1597  	trace_nfs4_layoutreturn_on_close(args->inode, &args->stateid, ret);
1598  	if (ld_private && ld_private->ops && ld_private->ops->free)
1599  		ld_private->ops->free(ld_private);
1600  	pnfs_put_layout_hdr(lo);
1601  }
1602  
pnfs_wait_on_layoutreturn(struct inode * ino,struct rpc_task * task)1603  bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task)
1604  {
1605  	struct nfs_inode *nfsi = NFS_I(ino);
1606          struct pnfs_layout_hdr *lo;
1607          bool sleep = false;
1608  
1609  	/* we might not have grabbed lo reference. so need to check under
1610  	 * i_lock */
1611          spin_lock(&ino->i_lock);
1612          lo = nfsi->layout;
1613          if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1614                  rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
1615                  sleep = true;
1616  	}
1617          spin_unlock(&ino->i_lock);
1618          return sleep;
1619  }
1620  
1621  /*
1622   * Compare two layout segments for sorting into layout cache.
1623   * We want to preferentially return RW over RO layouts, so ensure those
1624   * are seen first.
1625   */
1626  static s64
pnfs_lseg_range_cmp(const struct pnfs_layout_range * l1,const struct pnfs_layout_range * l2)1627  pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
1628  	   const struct pnfs_layout_range *l2)
1629  {
1630  	s64 d;
1631  
1632  	/* high offset > low offset */
1633  	d = l1->offset - l2->offset;
1634  	if (d)
1635  		return d;
1636  
1637  	/* short length > long length */
1638  	d = l2->length - l1->length;
1639  	if (d)
1640  		return d;
1641  
1642  	/* read > read/write */
1643  	return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
1644  }
1645  
1646  static bool
pnfs_lseg_range_is_after(const struct pnfs_layout_range * l1,const struct pnfs_layout_range * l2)1647  pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
1648  		const struct pnfs_layout_range *l2)
1649  {
1650  	return pnfs_lseg_range_cmp(l1, l2) > 0;
1651  }
1652  
1653  static bool
pnfs_lseg_no_merge(struct pnfs_layout_segment * lseg,struct pnfs_layout_segment * old)1654  pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
1655  		struct pnfs_layout_segment *old)
1656  {
1657  	return false;
1658  }
1659  
1660  void
pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr * lo,struct pnfs_layout_segment * lseg,bool (* is_after)(const struct pnfs_layout_range *,const struct pnfs_layout_range *),bool (* do_merge)(struct pnfs_layout_segment *,struct pnfs_layout_segment *),struct list_head * free_me)1661  pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1662  		   struct pnfs_layout_segment *lseg,
1663  		   bool (*is_after)(const struct pnfs_layout_range *,
1664  			   const struct pnfs_layout_range *),
1665  		   bool (*do_merge)(struct pnfs_layout_segment *,
1666  			   struct pnfs_layout_segment *),
1667  		   struct list_head *free_me)
1668  {
1669  	struct pnfs_layout_segment *lp, *tmp;
1670  
1671  	dprintk("%s:Begin\n", __func__);
1672  
1673  	list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
1674  		if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
1675  			continue;
1676  		if (do_merge(lseg, lp)) {
1677  			mark_lseg_invalid(lp, free_me);
1678  			continue;
1679  		}
1680  		if (is_after(&lseg->pls_range, &lp->pls_range))
1681  			continue;
1682  		list_add_tail(&lseg->pls_list, &lp->pls_list);
1683  		dprintk("%s: inserted lseg %p "
1684  			"iomode %d offset %llu length %llu before "
1685  			"lp %p iomode %d offset %llu length %llu\n",
1686  			__func__, lseg, lseg->pls_range.iomode,
1687  			lseg->pls_range.offset, lseg->pls_range.length,
1688  			lp, lp->pls_range.iomode, lp->pls_range.offset,
1689  			lp->pls_range.length);
1690  		goto out;
1691  	}
1692  	list_add_tail(&lseg->pls_list, &lo->plh_segs);
1693  	dprintk("%s: inserted lseg %p "
1694  		"iomode %d offset %llu length %llu at tail\n",
1695  		__func__, lseg, lseg->pls_range.iomode,
1696  		lseg->pls_range.offset, lseg->pls_range.length);
1697  out:
1698  	pnfs_get_layout_hdr(lo);
1699  
1700  	dprintk("%s:Return\n", __func__);
1701  }
1702  EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
1703  
1704  static void
pnfs_layout_insert_lseg(struct pnfs_layout_hdr * lo,struct pnfs_layout_segment * lseg,struct list_head * free_me)1705  pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1706  		   struct pnfs_layout_segment *lseg,
1707  		   struct list_head *free_me)
1708  {
1709  	struct inode *inode = lo->plh_inode;
1710  	struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
1711  
1712  	if (ld->add_lseg != NULL)
1713  		ld->add_lseg(lo, lseg, free_me);
1714  	else
1715  		pnfs_generic_layout_insert_lseg(lo, lseg,
1716  				pnfs_lseg_range_is_after,
1717  				pnfs_lseg_no_merge,
1718  				free_me);
1719  }
1720  
1721  static struct pnfs_layout_hdr *
alloc_init_layout_hdr(struct inode * ino,struct nfs_open_context * ctx,gfp_t gfp_flags)1722  alloc_init_layout_hdr(struct inode *ino,
1723  		      struct nfs_open_context *ctx,
1724  		      gfp_t gfp_flags)
1725  {
1726  	struct pnfs_layout_hdr *lo;
1727  
1728  	lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
1729  	if (!lo)
1730  		return NULL;
1731  	refcount_set(&lo->plh_refcount, 1);
1732  	INIT_LIST_HEAD(&lo->plh_layouts);
1733  	INIT_LIST_HEAD(&lo->plh_segs);
1734  	INIT_LIST_HEAD(&lo->plh_return_segs);
1735  	INIT_LIST_HEAD(&lo->plh_bulk_destroy);
1736  	lo->plh_inode = ino;
1737  	lo->plh_lc_cred = get_cred(ctx->cred);
1738  	lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID;
1739  	return lo;
1740  }
1741  
1742  static struct pnfs_layout_hdr *
pnfs_find_alloc_layout(struct inode * ino,struct nfs_open_context * ctx,gfp_t gfp_flags)1743  pnfs_find_alloc_layout(struct inode *ino,
1744  		       struct nfs_open_context *ctx,
1745  		       gfp_t gfp_flags)
1746  	__releases(&ino->i_lock)
1747  	__acquires(&ino->i_lock)
1748  {
1749  	struct nfs_inode *nfsi = NFS_I(ino);
1750  	struct pnfs_layout_hdr *new = NULL;
1751  
1752  	dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1753  
1754  	if (nfsi->layout != NULL)
1755  		goto out_existing;
1756  	spin_unlock(&ino->i_lock);
1757  	new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
1758  	spin_lock(&ino->i_lock);
1759  
1760  	if (likely(nfsi->layout == NULL)) {	/* Won the race? */
1761  		nfsi->layout = new;
1762  		return new;
1763  	} else if (new != NULL)
1764  		pnfs_free_layout_hdr(new);
1765  out_existing:
1766  	pnfs_get_layout_hdr(nfsi->layout);
1767  	return nfsi->layout;
1768  }
1769  
1770  /*
1771   * iomode matching rules:
1772   * iomode	lseg	strict match
1773   *                      iomode
1774   * -----	-----	------ -----
1775   * ANY		READ	N/A    true
1776   * ANY		RW	N/A    true
1777   * RW		READ	N/A    false
1778   * RW		RW	N/A    true
1779   * READ		READ	N/A    true
1780   * READ		RW	true   false
1781   * READ		RW	false  true
1782   */
1783  static bool
pnfs_lseg_range_match(const struct pnfs_layout_range * ls_range,const struct pnfs_layout_range * range,bool strict_iomode)1784  pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
1785  		 const struct pnfs_layout_range *range,
1786  		 bool strict_iomode)
1787  {
1788  	struct pnfs_layout_range range1;
1789  
1790  	if ((range->iomode == IOMODE_RW &&
1791  	     ls_range->iomode != IOMODE_RW) ||
1792  	    (range->iomode != ls_range->iomode &&
1793  	     strict_iomode) ||
1794  	    !pnfs_lseg_range_intersecting(ls_range, range))
1795  		return false;
1796  
1797  	/* range1 covers only the first byte in the range */
1798  	range1 = *range;
1799  	range1.length = 1;
1800  	return pnfs_lseg_range_contained(ls_range, &range1);
1801  }
1802  
1803  /*
1804   * lookup range in layout
1805   */
1806  static struct pnfs_layout_segment *
pnfs_find_lseg(struct pnfs_layout_hdr * lo,struct pnfs_layout_range * range,bool strict_iomode)1807  pnfs_find_lseg(struct pnfs_layout_hdr *lo,
1808  		struct pnfs_layout_range *range,
1809  		bool strict_iomode)
1810  {
1811  	struct pnfs_layout_segment *lseg, *ret = NULL;
1812  
1813  	dprintk("%s:Begin\n", __func__);
1814  
1815  	list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
1816  		if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
1817  		    pnfs_lseg_range_match(&lseg->pls_range, range,
1818  					  strict_iomode)) {
1819  			ret = pnfs_get_lseg(lseg);
1820  			break;
1821  		}
1822  	}
1823  
1824  	dprintk("%s:Return lseg %p ref %d\n",
1825  		__func__, ret, ret ? refcount_read(&ret->pls_refcount) : 0);
1826  	return ret;
1827  }
1828  
1829  /*
1830   * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1831   * to the MDS or over pNFS
1832   *
1833   * The nfs_inode read_io and write_io fields are cumulative counters reset
1834   * when there are no layout segments. Note that in pnfs_update_layout iomode
1835   * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1836   * WRITE request.
1837   *
1838   * A return of true means use MDS I/O.
1839   *
1840   * From rfc 5661:
1841   * If a file's size is smaller than the file size threshold, data accesses
1842   * SHOULD be sent to the metadata server.  If an I/O request has a length that
1843   * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1844   * server.  If both file size and I/O size are provided, the client SHOULD
1845   * reach or exceed  both thresholds before sending its read or write
1846   * requests to the data server.
1847   */
pnfs_within_mdsthreshold(struct nfs_open_context * ctx,struct inode * ino,int iomode)1848  static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1849  				     struct inode *ino, int iomode)
1850  {
1851  	struct nfs4_threshold *t = ctx->mdsthreshold;
1852  	struct nfs_inode *nfsi = NFS_I(ino);
1853  	loff_t fsize = i_size_read(ino);
1854  	bool size = false, size_set = false, io = false, io_set = false, ret = false;
1855  
1856  	if (t == NULL)
1857  		return ret;
1858  
1859  	dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1860  		__func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1861  
1862  	switch (iomode) {
1863  	case IOMODE_READ:
1864  		if (t->bm & THRESHOLD_RD) {
1865  			dprintk("%s fsize %llu\n", __func__, fsize);
1866  			size_set = true;
1867  			if (fsize < t->rd_sz)
1868  				size = true;
1869  		}
1870  		if (t->bm & THRESHOLD_RD_IO) {
1871  			dprintk("%s nfsi->read_io %llu\n", __func__,
1872  				nfsi->read_io);
1873  			io_set = true;
1874  			if (nfsi->read_io < t->rd_io_sz)
1875  				io = true;
1876  		}
1877  		break;
1878  	case IOMODE_RW:
1879  		if (t->bm & THRESHOLD_WR) {
1880  			dprintk("%s fsize %llu\n", __func__, fsize);
1881  			size_set = true;
1882  			if (fsize < t->wr_sz)
1883  				size = true;
1884  		}
1885  		if (t->bm & THRESHOLD_WR_IO) {
1886  			dprintk("%s nfsi->write_io %llu\n", __func__,
1887  				nfsi->write_io);
1888  			io_set = true;
1889  			if (nfsi->write_io < t->wr_io_sz)
1890  				io = true;
1891  		}
1892  		break;
1893  	}
1894  	if (size_set && io_set) {
1895  		if (size && io)
1896  			ret = true;
1897  	} else if (size || io)
1898  		ret = true;
1899  
1900  	dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1901  	return ret;
1902  }
1903  
pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr * lo)1904  static int pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
1905  {
1906  	/*
1907  	 * send layoutcommit as it can hold up layoutreturn due to lseg
1908  	 * reference
1909  	 */
1910  	pnfs_layoutcommit_inode(lo->plh_inode, false);
1911  	return wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
1912  				   nfs_wait_bit_killable,
1913  				   TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
1914  }
1915  
nfs_layoutget_begin(struct pnfs_layout_hdr * lo)1916  static void nfs_layoutget_begin(struct pnfs_layout_hdr *lo)
1917  {
1918  	atomic_inc(&lo->plh_outstanding);
1919  }
1920  
nfs_layoutget_end(struct pnfs_layout_hdr * lo)1921  static void nfs_layoutget_end(struct pnfs_layout_hdr *lo)
1922  {
1923  	if (atomic_dec_and_test(&lo->plh_outstanding) &&
1924  	    test_and_clear_bit(NFS_LAYOUT_DRAIN, &lo->plh_flags))
1925  		wake_up_bit(&lo->plh_flags, NFS_LAYOUT_DRAIN);
1926  }
1927  
pnfs_is_first_layoutget(struct pnfs_layout_hdr * lo)1928  static bool pnfs_is_first_layoutget(struct pnfs_layout_hdr *lo)
1929  {
1930  	return test_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags);
1931  }
1932  
pnfs_clear_first_layoutget(struct pnfs_layout_hdr * lo)1933  static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1934  {
1935  	unsigned long *bitlock = &lo->plh_flags;
1936  
1937  	clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1938  	smp_mb__after_atomic();
1939  	wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1940  }
1941  
_add_to_server_list(struct pnfs_layout_hdr * lo,struct nfs_server * server)1942  static void _add_to_server_list(struct pnfs_layout_hdr *lo,
1943  				struct nfs_server *server)
1944  {
1945  	if (!test_and_set_bit(NFS_LAYOUT_HASHED, &lo->plh_flags)) {
1946  		struct nfs_client *clp = server->nfs_client;
1947  
1948  		/* The lo must be on the clp list if there is any
1949  		 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1950  		 */
1951  		spin_lock(&clp->cl_lock);
1952  		list_add_tail_rcu(&lo->plh_layouts, &server->layouts);
1953  		spin_unlock(&clp->cl_lock);
1954  	}
1955  }
1956  
1957  /*
1958   * Layout segment is retreived from the server if not cached.
1959   * The appropriate layout segment is referenced and returned to the caller.
1960   */
1961  struct pnfs_layout_segment *
pnfs_update_layout(struct inode * ino,struct nfs_open_context * ctx,loff_t pos,u64 count,enum pnfs_iomode iomode,bool strict_iomode,gfp_t gfp_flags)1962  pnfs_update_layout(struct inode *ino,
1963  		   struct nfs_open_context *ctx,
1964  		   loff_t pos,
1965  		   u64 count,
1966  		   enum pnfs_iomode iomode,
1967  		   bool strict_iomode,
1968  		   gfp_t gfp_flags)
1969  {
1970  	struct pnfs_layout_range arg = {
1971  		.iomode = iomode,
1972  		.offset = pos,
1973  		.length = count,
1974  	};
1975  	unsigned pg_offset;
1976  	struct nfs_server *server = NFS_SERVER(ino);
1977  	struct nfs_client *clp = server->nfs_client;
1978  	struct pnfs_layout_hdr *lo = NULL;
1979  	struct pnfs_layout_segment *lseg = NULL;
1980  	struct nfs4_layoutget *lgp;
1981  	nfs4_stateid stateid;
1982  	long timeout = 0;
1983  	unsigned long giveup = jiffies + (clp->cl_lease_time << 1);
1984  	bool first;
1985  
1986  	if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
1987  		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1988  				 PNFS_UPDATE_LAYOUT_NO_PNFS);
1989  		goto out;
1990  	}
1991  
1992  	if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
1993  		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1994  				 PNFS_UPDATE_LAYOUT_MDSTHRESH);
1995  		goto out;
1996  	}
1997  
1998  lookup_again:
1999  	if (!nfs4_valid_open_stateid(ctx->state)) {
2000  		trace_pnfs_update_layout(ino, pos, count,
2001  					 iomode, lo, lseg,
2002  					 PNFS_UPDATE_LAYOUT_INVALID_OPEN);
2003  		lseg = ERR_PTR(-EIO);
2004  		goto out;
2005  	}
2006  
2007  	lseg = ERR_PTR(nfs4_client_recover_expired_lease(clp));
2008  	if (IS_ERR(lseg))
2009  		goto out;
2010  	first = false;
2011  	spin_lock(&ino->i_lock);
2012  	lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
2013  	if (lo == NULL) {
2014  		spin_unlock(&ino->i_lock);
2015  		lseg = ERR_PTR(-ENOMEM);
2016  		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2017  				 PNFS_UPDATE_LAYOUT_NOMEM);
2018  		goto out;
2019  	}
2020  
2021  	/* Do we even need to bother with this? */
2022  	if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
2023  		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2024  				 PNFS_UPDATE_LAYOUT_BULK_RECALL);
2025  		dprintk("%s matches recall, use MDS\n", __func__);
2026  		goto out_unlock;
2027  	}
2028  
2029  	/* if LAYOUTGET already failed once we don't try again */
2030  	if (pnfs_layout_io_test_failed(lo, iomode)) {
2031  		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2032  				 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
2033  		goto out_unlock;
2034  	}
2035  
2036  	/*
2037  	 * If the layout segment list is empty, but there are outstanding
2038  	 * layoutget calls, then they might be subject to a layoutrecall.
2039  	 */
2040  	if (test_bit(NFS_LAYOUT_DRAIN, &lo->plh_flags) &&
2041  	    atomic_read(&lo->plh_outstanding) != 0) {
2042  		spin_unlock(&ino->i_lock);
2043  		lseg = ERR_PTR(wait_on_bit(&lo->plh_flags, NFS_LAYOUT_DRAIN,
2044  					   TASK_KILLABLE));
2045  		if (IS_ERR(lseg))
2046  			goto out_put_layout_hdr;
2047  		pnfs_put_layout_hdr(lo);
2048  		goto lookup_again;
2049  	}
2050  
2051  	/*
2052  	 * Because we free lsegs when sending LAYOUTRETURN, we need to wait
2053  	 * for LAYOUTRETURN.
2054  	 */
2055  	if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
2056  		spin_unlock(&ino->i_lock);
2057  		dprintk("%s wait for layoutreturn\n", __func__);
2058  		lseg = ERR_PTR(pnfs_prepare_to_retry_layoutget(lo));
2059  		if (!IS_ERR(lseg)) {
2060  			pnfs_put_layout_hdr(lo);
2061  			dprintk("%s retrying\n", __func__);
2062  			trace_pnfs_update_layout(ino, pos, count, iomode, lo,
2063  						 lseg,
2064  						 PNFS_UPDATE_LAYOUT_RETRY);
2065  			goto lookup_again;
2066  		}
2067  		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2068  					 PNFS_UPDATE_LAYOUT_RETURN);
2069  		goto out_put_layout_hdr;
2070  	}
2071  
2072  	lseg = pnfs_find_lseg(lo, &arg, strict_iomode);
2073  	if (lseg) {
2074  		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2075  				PNFS_UPDATE_LAYOUT_FOUND_CACHED);
2076  		goto out_unlock;
2077  	}
2078  
2079  	/*
2080  	 * Choose a stateid for the LAYOUTGET. If we don't have a layout
2081  	 * stateid, or it has been invalidated, then we must use the open
2082  	 * stateid.
2083  	 */
2084  	if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
2085  		int status;
2086  
2087  		/*
2088  		 * The first layoutget for the file. Need to serialize per
2089  		 * RFC 5661 Errata 3208.
2090  		 */
2091  		if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
2092  				     &lo->plh_flags)) {
2093  			spin_unlock(&ino->i_lock);
2094  			lseg = ERR_PTR(wait_on_bit(&lo->plh_flags,
2095  						NFS_LAYOUT_FIRST_LAYOUTGET,
2096  						TASK_KILLABLE));
2097  			if (IS_ERR(lseg))
2098  				goto out_put_layout_hdr;
2099  			pnfs_put_layout_hdr(lo);
2100  			dprintk("%s retrying\n", __func__);
2101  			goto lookup_again;
2102  		}
2103  
2104  		spin_unlock(&ino->i_lock);
2105  		first = true;
2106  		status = nfs4_select_rw_stateid(ctx->state,
2107  					iomode == IOMODE_RW ? FMODE_WRITE : FMODE_READ,
2108  					NULL, &stateid, NULL);
2109  		if (status != 0) {
2110  			lseg = ERR_PTR(status);
2111  			trace_pnfs_update_layout(ino, pos, count,
2112  					iomode, lo, lseg,
2113  					PNFS_UPDATE_LAYOUT_INVALID_OPEN);
2114  			nfs4_schedule_stateid_recovery(server, ctx->state);
2115  			pnfs_clear_first_layoutget(lo);
2116  			pnfs_put_layout_hdr(lo);
2117  			goto lookup_again;
2118  		}
2119  		spin_lock(&ino->i_lock);
2120  	} else {
2121  		nfs4_stateid_copy(&stateid, &lo->plh_stateid);
2122  	}
2123  
2124  	if (pnfs_layoutgets_blocked(lo)) {
2125  		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2126  				PNFS_UPDATE_LAYOUT_BLOCKED);
2127  		goto out_unlock;
2128  	}
2129  	nfs_layoutget_begin(lo);
2130  	spin_unlock(&ino->i_lock);
2131  
2132  	_add_to_server_list(lo, server);
2133  
2134  	pg_offset = arg.offset & ~PAGE_MASK;
2135  	if (pg_offset) {
2136  		arg.offset -= pg_offset;
2137  		arg.length += pg_offset;
2138  	}
2139  	if (arg.length != NFS4_MAX_UINT64)
2140  		arg.length = PAGE_ALIGN(arg.length);
2141  
2142  	lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &stateid, &arg, gfp_flags);
2143  	if (!lgp) {
2144  		lseg = ERR_PTR(-ENOMEM);
2145  		trace_pnfs_update_layout(ino, pos, count, iomode, lo, NULL,
2146  					 PNFS_UPDATE_LAYOUT_NOMEM);
2147  		nfs_layoutget_end(lo);
2148  		goto out_put_layout_hdr;
2149  	}
2150  
2151  	lgp->lo = lo;
2152  	pnfs_get_layout_hdr(lo);
2153  
2154  	lseg = nfs4_proc_layoutget(lgp, &timeout);
2155  	trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2156  				 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
2157  	nfs_layoutget_end(lo);
2158  	if (IS_ERR(lseg)) {
2159  		switch(PTR_ERR(lseg)) {
2160  		case -EBUSY:
2161  			if (time_after(jiffies, giveup))
2162  				lseg = NULL;
2163  			break;
2164  		case -ERECALLCONFLICT:
2165  		case -EAGAIN:
2166  			break;
2167  		case -ENODATA:
2168  			/* The server returned NFS4ERR_LAYOUTUNAVAILABLE */
2169  			pnfs_layout_set_fail_bit(
2170  				lo, pnfs_iomode_to_fail_bit(iomode));
2171  			lseg = NULL;
2172  			goto out_put_layout_hdr;
2173  		default:
2174  			if (!nfs_error_is_fatal(PTR_ERR(lseg))) {
2175  				pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2176  				lseg = NULL;
2177  			}
2178  			goto out_put_layout_hdr;
2179  		}
2180  		if (lseg) {
2181  			if (first)
2182  				pnfs_clear_first_layoutget(lo);
2183  			trace_pnfs_update_layout(ino, pos, count,
2184  				iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY);
2185  			pnfs_put_layout_hdr(lo);
2186  			goto lookup_again;
2187  		}
2188  	} else {
2189  		pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2190  	}
2191  
2192  out_put_layout_hdr:
2193  	if (first)
2194  		pnfs_clear_first_layoutget(lo);
2195  	trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2196  				 PNFS_UPDATE_LAYOUT_EXIT);
2197  	pnfs_put_layout_hdr(lo);
2198  out:
2199  	dprintk("%s: inode %s/%llu pNFS layout segment %s for "
2200  			"(%s, offset: %llu, length: %llu)\n",
2201  			__func__, ino->i_sb->s_id,
2202  			(unsigned long long)NFS_FILEID(ino),
2203  			IS_ERR_OR_NULL(lseg) ? "not found" : "found",
2204  			iomode==IOMODE_RW ?  "read/write" : "read-only",
2205  			(unsigned long long)pos,
2206  			(unsigned long long)count);
2207  	return lseg;
2208  out_unlock:
2209  	spin_unlock(&ino->i_lock);
2210  	goto out_put_layout_hdr;
2211  }
2212  EXPORT_SYMBOL_GPL(pnfs_update_layout);
2213  
2214  static bool
pnfs_sanity_check_layout_range(struct pnfs_layout_range * range)2215  pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
2216  {
2217  	switch (range->iomode) {
2218  	case IOMODE_READ:
2219  	case IOMODE_RW:
2220  		break;
2221  	default:
2222  		return false;
2223  	}
2224  	if (range->offset == NFS4_MAX_UINT64)
2225  		return false;
2226  	if (range->length == 0)
2227  		return false;
2228  	if (range->length != NFS4_MAX_UINT64 &&
2229  	    range->length > NFS4_MAX_UINT64 - range->offset)
2230  		return false;
2231  	return true;
2232  }
2233  
2234  static struct pnfs_layout_hdr *
_pnfs_grab_empty_layout(struct inode * ino,struct nfs_open_context * ctx)2235  _pnfs_grab_empty_layout(struct inode *ino, struct nfs_open_context *ctx)
2236  {
2237  	struct pnfs_layout_hdr *lo;
2238  
2239  	spin_lock(&ino->i_lock);
2240  	lo = pnfs_find_alloc_layout(ino, ctx, nfs_io_gfp_mask());
2241  	if (!lo)
2242  		goto out_unlock;
2243  	if (!test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags))
2244  		goto out_unlock;
2245  	if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
2246  		goto out_unlock;
2247  	if (pnfs_layoutgets_blocked(lo))
2248  		goto out_unlock;
2249  	if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags))
2250  		goto out_unlock;
2251  	nfs_layoutget_begin(lo);
2252  	spin_unlock(&ino->i_lock);
2253  	_add_to_server_list(lo, NFS_SERVER(ino));
2254  	return lo;
2255  
2256  out_unlock:
2257  	spin_unlock(&ino->i_lock);
2258  	pnfs_put_layout_hdr(lo);
2259  	return NULL;
2260  }
2261  
_lgopen_prepare_attached(struct nfs4_opendata * data,struct nfs_open_context * ctx)2262  static void _lgopen_prepare_attached(struct nfs4_opendata *data,
2263  				     struct nfs_open_context *ctx)
2264  {
2265  	struct inode *ino = data->dentry->d_inode;
2266  	struct pnfs_layout_range rng = {
2267  		.iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2268  			  IOMODE_RW: IOMODE_READ,
2269  		.offset = 0,
2270  		.length = NFS4_MAX_UINT64,
2271  	};
2272  	struct nfs4_layoutget *lgp;
2273  	struct pnfs_layout_hdr *lo;
2274  
2275  	/* Heuristic: don't send layoutget if we have cached data */
2276  	if (rng.iomode == IOMODE_READ &&
2277  	   (i_size_read(ino) == 0 || ino->i_mapping->nrpages != 0))
2278  		return;
2279  
2280  	lo = _pnfs_grab_empty_layout(ino, ctx);
2281  	if (!lo)
2282  		return;
2283  	lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &current_stateid, &rng,
2284  					     nfs_io_gfp_mask());
2285  	if (!lgp) {
2286  		pnfs_clear_first_layoutget(lo);
2287  		nfs_layoutget_end(lo);
2288  		pnfs_put_layout_hdr(lo);
2289  		return;
2290  	}
2291  	lgp->lo = lo;
2292  	data->lgp = lgp;
2293  	data->o_arg.lg_args = &lgp->args;
2294  	data->o_res.lg_res = &lgp->res;
2295  }
2296  
_lgopen_prepare_floating(struct nfs4_opendata * data,struct nfs_open_context * ctx)2297  static void _lgopen_prepare_floating(struct nfs4_opendata *data,
2298  				     struct nfs_open_context *ctx)
2299  {
2300  	struct inode *ino = data->dentry->d_inode;
2301  	struct pnfs_layout_range rng = {
2302  		.iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2303  			  IOMODE_RW: IOMODE_READ,
2304  		.offset = 0,
2305  		.length = NFS4_MAX_UINT64,
2306  	};
2307  	struct nfs4_layoutget *lgp;
2308  
2309  	lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &current_stateid, &rng,
2310  					     nfs_io_gfp_mask());
2311  	if (!lgp)
2312  		return;
2313  	data->lgp = lgp;
2314  	data->o_arg.lg_args = &lgp->args;
2315  	data->o_res.lg_res = &lgp->res;
2316  }
2317  
pnfs_lgopen_prepare(struct nfs4_opendata * data,struct nfs_open_context * ctx)2318  void pnfs_lgopen_prepare(struct nfs4_opendata *data,
2319  			 struct nfs_open_context *ctx)
2320  {
2321  	struct nfs_server *server = NFS_SERVER(data->dir->d_inode);
2322  
2323  	if (!(pnfs_enabled_sb(server) &&
2324  	      server->pnfs_curr_ld->flags & PNFS_LAYOUTGET_ON_OPEN))
2325  		return;
2326  	/* Could check on max_ops, but currently hardcoded high enough */
2327  	if (!nfs_server_capable(data->dir->d_inode, NFS_CAP_LGOPEN))
2328  		return;
2329  	if (data->lgp)
2330  		return;
2331  	if (data->state)
2332  		_lgopen_prepare_attached(data, ctx);
2333  	else
2334  		_lgopen_prepare_floating(data, ctx);
2335  }
2336  
pnfs_parse_lgopen(struct inode * ino,struct nfs4_layoutget * lgp,struct nfs_open_context * ctx)2337  void pnfs_parse_lgopen(struct inode *ino, struct nfs4_layoutget *lgp,
2338  		       struct nfs_open_context *ctx)
2339  {
2340  	struct pnfs_layout_hdr *lo;
2341  	struct pnfs_layout_segment *lseg;
2342  	struct nfs_server *srv = NFS_SERVER(ino);
2343  	u32 iomode;
2344  
2345  	if (!lgp)
2346  		return;
2347  	dprintk("%s: entered with status %i\n", __func__, lgp->res.status);
2348  	if (lgp->res.status) {
2349  		switch (lgp->res.status) {
2350  		default:
2351  			break;
2352  		/*
2353  		 * Halt lgopen attempts if the server doesn't recognise
2354  		 * the "current stateid" value, the layout type, or the
2355  		 * layoutget operation as being valid.
2356  		 * Also if it complains about too many ops in the compound
2357  		 * or of the request/reply being too big.
2358  		 */
2359  		case -NFS4ERR_BAD_STATEID:
2360  		case -NFS4ERR_NOTSUPP:
2361  		case -NFS4ERR_REP_TOO_BIG:
2362  		case -NFS4ERR_REP_TOO_BIG_TO_CACHE:
2363  		case -NFS4ERR_REQ_TOO_BIG:
2364  		case -NFS4ERR_TOO_MANY_OPS:
2365  		case -NFS4ERR_UNKNOWN_LAYOUTTYPE:
2366  			srv->caps &= ~NFS_CAP_LGOPEN;
2367  		}
2368  		return;
2369  	}
2370  	if (!lgp->lo) {
2371  		lo = _pnfs_grab_empty_layout(ino, ctx);
2372  		if (!lo)
2373  			return;
2374  		lgp->lo = lo;
2375  	} else
2376  		lo = lgp->lo;
2377  
2378  	lseg = pnfs_layout_process(lgp);
2379  	if (!IS_ERR(lseg)) {
2380  		iomode = lgp->args.range.iomode;
2381  		pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2382  		pnfs_put_lseg(lseg);
2383  	}
2384  }
2385  
nfs4_lgopen_release(struct nfs4_layoutget * lgp)2386  void nfs4_lgopen_release(struct nfs4_layoutget *lgp)
2387  {
2388  	if (lgp != NULL) {
2389  		if (lgp->lo) {
2390  			pnfs_clear_first_layoutget(lgp->lo);
2391  			nfs_layoutget_end(lgp->lo);
2392  		}
2393  		pnfs_layoutget_free(lgp);
2394  	}
2395  }
2396  
2397  struct pnfs_layout_segment *
pnfs_layout_process(struct nfs4_layoutget * lgp)2398  pnfs_layout_process(struct nfs4_layoutget *lgp)
2399  {
2400  	struct pnfs_layout_hdr *lo = lgp->lo;
2401  	struct nfs4_layoutget_res *res = &lgp->res;
2402  	struct pnfs_layout_segment *lseg;
2403  	struct inode *ino = lo->plh_inode;
2404  	LIST_HEAD(free_me);
2405  
2406  	if (!pnfs_sanity_check_layout_range(&res->range))
2407  		return ERR_PTR(-EINVAL);
2408  
2409  	/* Inject layout blob into I/O device driver */
2410  	lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
2411  	if (IS_ERR_OR_NULL(lseg)) {
2412  		if (!lseg)
2413  			lseg = ERR_PTR(-ENOMEM);
2414  
2415  		dprintk("%s: Could not allocate layout: error %ld\n",
2416  		       __func__, PTR_ERR(lseg));
2417  		return lseg;
2418  	}
2419  
2420  	pnfs_init_lseg(lo, lseg, &res->range, &res->stateid);
2421  
2422  	spin_lock(&ino->i_lock);
2423  	if (pnfs_layoutgets_blocked(lo)) {
2424  		dprintk("%s forget reply due to state\n", __func__);
2425  		goto out_forget;
2426  	}
2427  
2428  	if (test_bit(NFS_LAYOUT_DRAIN, &lo->plh_flags) &&
2429  	    !pnfs_is_first_layoutget(lo))
2430  		goto out_forget;
2431  
2432  	if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
2433  		/* existing state ID, make sure the sequence number matches. */
2434  		if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
2435  			if (!pnfs_layout_is_valid(lo))
2436  				lo->plh_barrier = 0;
2437  			dprintk("%s forget reply due to sequence\n", __func__);
2438  			goto out_forget;
2439  		}
2440  		pnfs_set_layout_stateid(lo, &res->stateid, lgp->cred, false);
2441  	} else if (pnfs_layout_is_valid(lo)) {
2442  		/*
2443  		 * We got an entirely new state ID.  Mark all segments for the
2444  		 * inode invalid, and retry the layoutget
2445  		 */
2446  		struct pnfs_layout_range range = {
2447  			.iomode = IOMODE_ANY,
2448  			.length = NFS4_MAX_UINT64,
2449  		};
2450  		pnfs_mark_matching_lsegs_return(lo, &free_me, &range, 0);
2451  		goto out_forget;
2452  	} else {
2453  		/* We have a completely new layout */
2454  		pnfs_set_layout_stateid(lo, &res->stateid, lgp->cred, true);
2455  	}
2456  
2457  	pnfs_get_lseg(lseg);
2458  	pnfs_layout_insert_lseg(lo, lseg, &free_me);
2459  
2460  
2461  	if (res->return_on_close)
2462  		set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
2463  
2464  	spin_unlock(&ino->i_lock);
2465  	pnfs_free_lseg_list(&free_me);
2466  	return lseg;
2467  
2468  out_forget:
2469  	spin_unlock(&ino->i_lock);
2470  	lseg->pls_layout = lo;
2471  	NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
2472  	return ERR_PTR(-EAGAIN);
2473  }
2474  
2475  /**
2476   * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
2477   * @lo: pointer to layout header
2478   * @tmp_list: list header to be used with pnfs_free_lseg_list()
2479   * @return_range: describe layout segment ranges to be returned
2480   * @seq: stateid seqid to match
2481   *
2482   * This function is mainly intended for use by layoutrecall. It attempts
2483   * to free the layout segment immediately, or else to mark it for return
2484   * as soon as its reference count drops to zero.
2485   *
2486   * Returns
2487   * - 0: a layoutreturn needs to be scheduled.
2488   * - EBUSY: there are layout segment that are still in use.
2489   * - ENOENT: there are no layout segments that need to be returned.
2490   */
2491  int
pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr * lo,struct list_head * tmp_list,const struct pnfs_layout_range * return_range,u32 seq)2492  pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
2493  				struct list_head *tmp_list,
2494  				const struct pnfs_layout_range *return_range,
2495  				u32 seq)
2496  {
2497  	struct pnfs_layout_segment *lseg, *next;
2498  	struct nfs_server *server = NFS_SERVER(lo->plh_inode);
2499  	int remaining = 0;
2500  
2501  	dprintk("%s:Begin lo %p\n", __func__, lo);
2502  
2503  	assert_spin_locked(&lo->plh_inode->i_lock);
2504  
2505  	if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
2506  		tmp_list = &lo->plh_return_segs;
2507  
2508  	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
2509  		if (pnfs_match_lseg_recall(lseg, return_range, seq)) {
2510  			dprintk("%s: marking lseg %p iomode %d "
2511  				"offset %llu length %llu\n", __func__,
2512  				lseg, lseg->pls_range.iomode,
2513  				lseg->pls_range.offset,
2514  				lseg->pls_range.length);
2515  			if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
2516  				tmp_list = &lo->plh_return_segs;
2517  			if (mark_lseg_invalid(lseg, tmp_list))
2518  				continue;
2519  			remaining++;
2520  			set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
2521  			pnfs_lseg_cancel_io(server, lseg);
2522  		}
2523  
2524  	if (remaining) {
2525  		pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2526  		return -EBUSY;
2527  	}
2528  
2529  	if (!list_empty(&lo->plh_return_segs)) {
2530  		pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2531  		return 0;
2532  	}
2533  
2534  	return -ENOENT;
2535  }
2536  
2537  static void
pnfs_mark_layout_for_return(struct inode * inode,const struct pnfs_layout_range * range)2538  pnfs_mark_layout_for_return(struct inode *inode,
2539  			    const struct pnfs_layout_range *range)
2540  {
2541  	struct pnfs_layout_hdr *lo;
2542  	bool return_now = false;
2543  
2544  	spin_lock(&inode->i_lock);
2545  	lo = NFS_I(inode)->layout;
2546  	if (!pnfs_layout_is_valid(lo)) {
2547  		spin_unlock(&inode->i_lock);
2548  		return;
2549  	}
2550  	pnfs_set_plh_return_info(lo, range->iomode, 0);
2551  	/*
2552  	 * mark all matching lsegs so that we are sure to have no live
2553  	 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
2554  	 * for how it works.
2555  	 */
2556  	if (pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, range, 0) != -EBUSY) {
2557  		const struct cred *cred;
2558  		nfs4_stateid stateid;
2559  		enum pnfs_iomode iomode;
2560  
2561  		return_now = pnfs_prepare_layoutreturn(lo, &stateid, &cred, &iomode);
2562  		spin_unlock(&inode->i_lock);
2563  		if (return_now)
2564  			pnfs_send_layoutreturn(lo, &stateid, &cred, iomode, false);
2565  	} else {
2566  		spin_unlock(&inode->i_lock);
2567  		nfs_commit_inode(inode, 0);
2568  	}
2569  }
2570  
pnfs_error_mark_layout_for_return(struct inode * inode,struct pnfs_layout_segment * lseg)2571  void pnfs_error_mark_layout_for_return(struct inode *inode,
2572  				       struct pnfs_layout_segment *lseg)
2573  {
2574  	struct pnfs_layout_range range = {
2575  		.iomode = lseg->pls_range.iomode,
2576  		.offset = 0,
2577  		.length = NFS4_MAX_UINT64,
2578  	};
2579  
2580  	pnfs_mark_layout_for_return(inode, &range);
2581  }
2582  EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
2583  
2584  static bool
pnfs_layout_can_be_returned(struct pnfs_layout_hdr * lo)2585  pnfs_layout_can_be_returned(struct pnfs_layout_hdr *lo)
2586  {
2587  	return pnfs_layout_is_valid(lo) &&
2588  		!test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) &&
2589  		!test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
2590  }
2591  
2592  static struct pnfs_layout_segment *
pnfs_find_first_lseg(struct pnfs_layout_hdr * lo,const struct pnfs_layout_range * range,enum pnfs_iomode iomode)2593  pnfs_find_first_lseg(struct pnfs_layout_hdr *lo,
2594  		     const struct pnfs_layout_range *range,
2595  		     enum pnfs_iomode iomode)
2596  {
2597  	struct pnfs_layout_segment *lseg;
2598  
2599  	list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
2600  		if (!test_bit(NFS_LSEG_VALID, &lseg->pls_flags))
2601  			continue;
2602  		if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
2603  			continue;
2604  		if (lseg->pls_range.iomode != iomode && iomode != IOMODE_ANY)
2605  			continue;
2606  		if (pnfs_lseg_range_intersecting(&lseg->pls_range, range))
2607  			return lseg;
2608  	}
2609  	return NULL;
2610  }
2611  
2612  /* Find open file states whose mode matches that of the range */
2613  static bool
pnfs_should_return_unused_layout(struct pnfs_layout_hdr * lo,const struct pnfs_layout_range * range)2614  pnfs_should_return_unused_layout(struct pnfs_layout_hdr *lo,
2615  				 const struct pnfs_layout_range *range)
2616  {
2617  	struct list_head *head;
2618  	struct nfs_open_context *ctx;
2619  	fmode_t mode = 0;
2620  
2621  	if (!pnfs_layout_can_be_returned(lo) ||
2622  	    !pnfs_find_first_lseg(lo, range, range->iomode))
2623  		return false;
2624  
2625  	head = &NFS_I(lo->plh_inode)->open_files;
2626  	list_for_each_entry_rcu(ctx, head, list) {
2627  		if (ctx->state)
2628  			mode |= ctx->state->state & (FMODE_READ|FMODE_WRITE);
2629  	}
2630  
2631  	switch (range->iomode) {
2632  	default:
2633  		break;
2634  	case IOMODE_READ:
2635  		mode &= ~FMODE_WRITE;
2636  		break;
2637  	case IOMODE_RW:
2638  		if (pnfs_find_first_lseg(lo, range, IOMODE_READ))
2639  			mode &= ~FMODE_READ;
2640  	}
2641  	return mode == 0;
2642  }
2643  
pnfs_layout_return_unused_byserver(struct nfs_server * server,void * data)2644  static int pnfs_layout_return_unused_byserver(struct nfs_server *server,
2645  					      void *data)
2646  {
2647  	const struct pnfs_layout_range *range = data;
2648  	const struct cred *cred;
2649  	struct pnfs_layout_hdr *lo;
2650  	struct inode *inode;
2651  	nfs4_stateid stateid;
2652  	enum pnfs_iomode iomode;
2653  
2654  restart:
2655  	rcu_read_lock();
2656  	list_for_each_entry_rcu(lo, &server->layouts, plh_layouts) {
2657  		inode = lo->plh_inode;
2658  		if (!inode || !pnfs_layout_can_be_returned(lo) ||
2659  		    test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
2660  			continue;
2661  		spin_lock(&inode->i_lock);
2662  		if (!lo->plh_inode ||
2663  		    !pnfs_should_return_unused_layout(lo, range)) {
2664  			spin_unlock(&inode->i_lock);
2665  			continue;
2666  		}
2667  		pnfs_get_layout_hdr(lo);
2668  		pnfs_set_plh_return_info(lo, range->iomode, 0);
2669  		if (pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs,
2670  						    range, 0) != 0 ||
2671  		    !pnfs_prepare_layoutreturn(lo, &stateid, &cred, &iomode)) {
2672  			spin_unlock(&inode->i_lock);
2673  			rcu_read_unlock();
2674  			pnfs_put_layout_hdr(lo);
2675  			cond_resched();
2676  			goto restart;
2677  		}
2678  		spin_unlock(&inode->i_lock);
2679  		rcu_read_unlock();
2680  		pnfs_send_layoutreturn(lo, &stateid, &cred, iomode, false);
2681  		pnfs_put_layout_hdr(lo);
2682  		cond_resched();
2683  		goto restart;
2684  	}
2685  	rcu_read_unlock();
2686  	return 0;
2687  }
2688  
2689  void
pnfs_layout_return_unused_byclid(struct nfs_client * clp,enum pnfs_iomode iomode)2690  pnfs_layout_return_unused_byclid(struct nfs_client *clp,
2691  				 enum pnfs_iomode iomode)
2692  {
2693  	struct pnfs_layout_range range = {
2694  		.iomode = iomode,
2695  		.offset = 0,
2696  		.length = NFS4_MAX_UINT64,
2697  	};
2698  
2699  	nfs_client_for_each_server(clp, pnfs_layout_return_unused_byserver,
2700  			&range);
2701  }
2702  
2703  void
pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor * pgio)2704  pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor *pgio)
2705  {
2706  	if (pgio->pg_lseg == NULL ||
2707  	    test_bit(NFS_LSEG_VALID, &pgio->pg_lseg->pls_flags))
2708  		return;
2709  	pnfs_put_lseg(pgio->pg_lseg);
2710  	pgio->pg_lseg = NULL;
2711  }
2712  EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout);
2713  
2714  /*
2715   * Check for any intersection between the request and the pgio->pg_lseg,
2716   * and if none, put this pgio->pg_lseg away.
2717   */
2718  void
pnfs_generic_pg_check_range(struct nfs_pageio_descriptor * pgio,struct nfs_page * req)2719  pnfs_generic_pg_check_range(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2720  {
2721  	if (pgio->pg_lseg && !pnfs_lseg_request_intersecting(pgio->pg_lseg, req)) {
2722  		pnfs_put_lseg(pgio->pg_lseg);
2723  		pgio->pg_lseg = NULL;
2724  	}
2725  }
2726  EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_range);
2727  
2728  void
pnfs_generic_pg_init_read(struct nfs_pageio_descriptor * pgio,struct nfs_page * req)2729  pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2730  {
2731  	u64 rd_size;
2732  
2733  	pnfs_generic_pg_check_layout(pgio);
2734  	pnfs_generic_pg_check_range(pgio, req);
2735  	if (pgio->pg_lseg == NULL) {
2736  		if (pgio->pg_dreq == NULL)
2737  			rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
2738  		else
2739  			rd_size = nfs_dreq_bytes_left(pgio->pg_dreq,
2740  						      req_offset(req));
2741  
2742  		pgio->pg_lseg =
2743  			pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req),
2744  					   req_offset(req), rd_size,
2745  					   IOMODE_READ, false,
2746  					   nfs_io_gfp_mask());
2747  		if (IS_ERR(pgio->pg_lseg)) {
2748  			pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2749  			pgio->pg_lseg = NULL;
2750  			return;
2751  		}
2752  	}
2753  	/* If no lseg, fall back to read through mds */
2754  	if (pgio->pg_lseg == NULL)
2755  		nfs_pageio_reset_read_mds(pgio);
2756  
2757  }
2758  EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
2759  
2760  void
pnfs_generic_pg_init_write(struct nfs_pageio_descriptor * pgio,struct nfs_page * req,u64 wb_size)2761  pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
2762  			   struct nfs_page *req, u64 wb_size)
2763  {
2764  	pnfs_generic_pg_check_layout(pgio);
2765  	pnfs_generic_pg_check_range(pgio, req);
2766  	if (pgio->pg_lseg == NULL) {
2767  		pgio->pg_lseg =
2768  			pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req),
2769  					   req_offset(req), wb_size, IOMODE_RW,
2770  					   false, nfs_io_gfp_mask());
2771  		if (IS_ERR(pgio->pg_lseg)) {
2772  			pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2773  			pgio->pg_lseg = NULL;
2774  			return;
2775  		}
2776  	}
2777  	/* If no lseg, fall back to write through mds */
2778  	if (pgio->pg_lseg == NULL)
2779  		nfs_pageio_reset_write_mds(pgio);
2780  }
2781  EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
2782  
2783  void
pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor * desc)2784  pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
2785  {
2786  	if (desc->pg_lseg) {
2787  		pnfs_put_lseg(desc->pg_lseg);
2788  		desc->pg_lseg = NULL;
2789  	}
2790  }
2791  EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
2792  
2793  /*
2794   * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
2795   * of bytes (maximum @req->wb_bytes) that can be coalesced.
2796   */
2797  size_t
pnfs_generic_pg_test(struct nfs_pageio_descriptor * pgio,struct nfs_page * prev,struct nfs_page * req)2798  pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
2799  		     struct nfs_page *prev, struct nfs_page *req)
2800  {
2801  	unsigned int size;
2802  	u64 seg_end, req_start, seg_left;
2803  
2804  	size = nfs_generic_pg_test(pgio, prev, req);
2805  	if (!size)
2806  		return 0;
2807  
2808  	/*
2809  	 * 'size' contains the number of bytes left in the current page (up
2810  	 * to the original size asked for in @req->wb_bytes).
2811  	 *
2812  	 * Calculate how many bytes are left in the layout segment
2813  	 * and if there are less bytes than 'size', return that instead.
2814  	 *
2815  	 * Please also note that 'end_offset' is actually the offset of the
2816  	 * first byte that lies outside the pnfs_layout_range. FIXME?
2817  	 *
2818  	 */
2819  	if (pgio->pg_lseg) {
2820  		seg_end = pnfs_end_offset(pgio->pg_lseg->pls_range.offset,
2821  				     pgio->pg_lseg->pls_range.length);
2822  		req_start = req_offset(req);
2823  
2824  		/* start of request is past the last byte of this segment */
2825  		if (req_start >= seg_end)
2826  			return 0;
2827  
2828  		/* adjust 'size' iff there are fewer bytes left in the
2829  		 * segment than what nfs_generic_pg_test returned */
2830  		seg_left = seg_end - req_start;
2831  		if (seg_left < size)
2832  			size = (unsigned int)seg_left;
2833  	}
2834  
2835  	return size;
2836  }
2837  EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
2838  
pnfs_write_done_resend_to_mds(struct nfs_pgio_header * hdr)2839  int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
2840  {
2841  	struct nfs_pageio_descriptor pgio;
2842  
2843  	/* Resend all requests through the MDS */
2844  	nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
2845  			      hdr->completion_ops);
2846  	return nfs_pageio_resend(&pgio, hdr);
2847  }
2848  EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
2849  
pnfs_ld_handle_write_error(struct nfs_pgio_header * hdr)2850  static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
2851  {
2852  
2853  	dprintk("pnfs write error = %d\n", hdr->pnfs_error);
2854  	if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
2855  	    PNFS_LAYOUTRET_ON_ERROR) {
2856  		pnfs_return_layout(hdr->inode);
2857  	}
2858  	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
2859  		hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
2860  }
2861  
2862  /*
2863   * Called by non rpc-based layout drivers
2864   */
pnfs_ld_write_done(struct nfs_pgio_header * hdr)2865  void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
2866  {
2867  	if (likely(!hdr->pnfs_error)) {
2868  		pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
2869  				hdr->mds_offset + hdr->res.count);
2870  		hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
2871  	}
2872  	trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
2873  	if (unlikely(hdr->pnfs_error))
2874  		pnfs_ld_handle_write_error(hdr);
2875  	hdr->mds_ops->rpc_release(hdr);
2876  }
2877  EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
2878  
2879  static void
pnfs_write_through_mds(struct nfs_pageio_descriptor * desc,struct nfs_pgio_header * hdr)2880  pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
2881  		struct nfs_pgio_header *hdr)
2882  {
2883  	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2884  
2885  	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2886  		list_splice_tail_init(&hdr->pages, &mirror->pg_list);
2887  		nfs_pageio_reset_write_mds(desc);
2888  		mirror->pg_recoalesce = 1;
2889  	}
2890  	hdr->completion_ops->completion(hdr);
2891  }
2892  
2893  static enum pnfs_try_status
pnfs_try_to_write_data(struct nfs_pgio_header * hdr,const struct rpc_call_ops * call_ops,struct pnfs_layout_segment * lseg,int how)2894  pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
2895  			const struct rpc_call_ops *call_ops,
2896  			struct pnfs_layout_segment *lseg,
2897  			int how)
2898  {
2899  	struct inode *inode = hdr->inode;
2900  	enum pnfs_try_status trypnfs;
2901  	struct nfs_server *nfss = NFS_SERVER(inode);
2902  
2903  	hdr->mds_ops = call_ops;
2904  
2905  	dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
2906  		inode->i_ino, hdr->args.count, hdr->args.offset, how);
2907  	trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
2908  	if (trypnfs != PNFS_NOT_ATTEMPTED)
2909  		nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
2910  	dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2911  	return trypnfs;
2912  }
2913  
2914  static void
pnfs_do_write(struct nfs_pageio_descriptor * desc,struct nfs_pgio_header * hdr,int how)2915  pnfs_do_write(struct nfs_pageio_descriptor *desc,
2916  	      struct nfs_pgio_header *hdr, int how)
2917  {
2918  	const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2919  	struct pnfs_layout_segment *lseg = desc->pg_lseg;
2920  	enum pnfs_try_status trypnfs;
2921  
2922  	trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
2923  	switch (trypnfs) {
2924  	case PNFS_NOT_ATTEMPTED:
2925  		pnfs_write_through_mds(desc, hdr);
2926  		break;
2927  	case PNFS_ATTEMPTED:
2928  		break;
2929  	case PNFS_TRY_AGAIN:
2930  		/* cleanup hdr and prepare to redo pnfs */
2931  		if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2932  			struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2933  			list_splice_init(&hdr->pages, &mirror->pg_list);
2934  			mirror->pg_recoalesce = 1;
2935  		}
2936  		hdr->mds_ops->rpc_release(hdr);
2937  	}
2938  }
2939  
pnfs_writehdr_free(struct nfs_pgio_header * hdr)2940  static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
2941  {
2942  	pnfs_put_lseg(hdr->lseg);
2943  	nfs_pgio_header_free(hdr);
2944  }
2945  
2946  int
pnfs_generic_pg_writepages(struct nfs_pageio_descriptor * desc)2947  pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
2948  {
2949  	struct nfs_pgio_header *hdr;
2950  	int ret;
2951  
2952  	hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2953  	if (!hdr) {
2954  		desc->pg_error = -ENOMEM;
2955  		return desc->pg_error;
2956  	}
2957  	nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
2958  
2959  	hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
2960  	ret = nfs_generic_pgio(desc, hdr);
2961  	if (!ret)
2962  		pnfs_do_write(desc, hdr, desc->pg_ioflags);
2963  
2964  	return ret;
2965  }
2966  EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
2967  
pnfs_read_done_resend_to_mds(struct nfs_pgio_header * hdr)2968  int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
2969  {
2970  	struct nfs_pageio_descriptor pgio;
2971  
2972  	/* Resend all requests through the MDS */
2973  	nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
2974  	return nfs_pageio_resend(&pgio, hdr);
2975  }
2976  EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
2977  
pnfs_ld_handle_read_error(struct nfs_pgio_header * hdr)2978  static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
2979  {
2980  	dprintk("pnfs read error = %d\n", hdr->pnfs_error);
2981  	if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
2982  	    PNFS_LAYOUTRET_ON_ERROR) {
2983  		pnfs_return_layout(hdr->inode);
2984  	}
2985  	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
2986  		hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
2987  }
2988  
2989  /*
2990   * Called by non rpc-based layout drivers
2991   */
pnfs_ld_read_done(struct nfs_pgio_header * hdr)2992  void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
2993  {
2994  	if (likely(!hdr->pnfs_error))
2995  		hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
2996  	trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
2997  	if (unlikely(hdr->pnfs_error))
2998  		pnfs_ld_handle_read_error(hdr);
2999  	hdr->mds_ops->rpc_release(hdr);
3000  }
3001  EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
3002  
3003  static void
pnfs_read_through_mds(struct nfs_pageio_descriptor * desc,struct nfs_pgio_header * hdr)3004  pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
3005  		struct nfs_pgio_header *hdr)
3006  {
3007  	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
3008  
3009  	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
3010  		list_splice_tail_init(&hdr->pages, &mirror->pg_list);
3011  		nfs_pageio_reset_read_mds(desc);
3012  		mirror->pg_recoalesce = 1;
3013  	}
3014  	hdr->completion_ops->completion(hdr);
3015  }
3016  
3017  /*
3018   * Call the appropriate parallel I/O subsystem read function.
3019   */
3020  static enum pnfs_try_status
pnfs_try_to_read_data(struct nfs_pgio_header * hdr,const struct rpc_call_ops * call_ops,struct pnfs_layout_segment * lseg)3021  pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
3022  		       const struct rpc_call_ops *call_ops,
3023  		       struct pnfs_layout_segment *lseg)
3024  {
3025  	struct inode *inode = hdr->inode;
3026  	struct nfs_server *nfss = NFS_SERVER(inode);
3027  	enum pnfs_try_status trypnfs;
3028  
3029  	hdr->mds_ops = call_ops;
3030  
3031  	dprintk("%s: Reading ino:%lu %u@%llu\n",
3032  		__func__, inode->i_ino, hdr->args.count, hdr->args.offset);
3033  
3034  	trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
3035  	if (trypnfs != PNFS_NOT_ATTEMPTED)
3036  		nfs_inc_stats(inode, NFSIOS_PNFS_READ);
3037  	dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
3038  	return trypnfs;
3039  }
3040  
3041  /* Resend all requests through pnfs. */
pnfs_read_resend_pnfs(struct nfs_pgio_header * hdr,unsigned int mirror_idx)3042  void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr,
3043  			   unsigned int mirror_idx)
3044  {
3045  	struct nfs_pageio_descriptor pgio;
3046  
3047  	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
3048  		/* Prevent deadlocks with layoutreturn! */
3049  		pnfs_put_lseg(hdr->lseg);
3050  		hdr->lseg = NULL;
3051  
3052  		nfs_pageio_init_read(&pgio, hdr->inode, false,
3053  					hdr->completion_ops);
3054  		pgio.pg_mirror_idx = mirror_idx;
3055  		hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr);
3056  	}
3057  }
3058  EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
3059  
3060  static void
pnfs_do_read(struct nfs_pageio_descriptor * desc,struct nfs_pgio_header * hdr)3061  pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
3062  {
3063  	const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
3064  	struct pnfs_layout_segment *lseg = desc->pg_lseg;
3065  	enum pnfs_try_status trypnfs;
3066  
3067  	trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
3068  	switch (trypnfs) {
3069  	case PNFS_NOT_ATTEMPTED:
3070  		pnfs_read_through_mds(desc, hdr);
3071  		break;
3072  	case PNFS_ATTEMPTED:
3073  		break;
3074  	case PNFS_TRY_AGAIN:
3075  		/* cleanup hdr and prepare to redo pnfs */
3076  		if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
3077  			struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
3078  			list_splice_init(&hdr->pages, &mirror->pg_list);
3079  			mirror->pg_recoalesce = 1;
3080  		}
3081  		hdr->mds_ops->rpc_release(hdr);
3082  	}
3083  }
3084  
pnfs_readhdr_free(struct nfs_pgio_header * hdr)3085  static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
3086  {
3087  	pnfs_put_lseg(hdr->lseg);
3088  	nfs_pgio_header_free(hdr);
3089  }
3090  
3091  int
pnfs_generic_pg_readpages(struct nfs_pageio_descriptor * desc)3092  pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
3093  {
3094  	struct nfs_pgio_header *hdr;
3095  	int ret;
3096  
3097  	hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
3098  	if (!hdr) {
3099  		desc->pg_error = -ENOMEM;
3100  		return desc->pg_error;
3101  	}
3102  	nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
3103  	hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
3104  	ret = nfs_generic_pgio(desc, hdr);
3105  	if (!ret)
3106  		pnfs_do_read(desc, hdr);
3107  	return ret;
3108  }
3109  EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
3110  
pnfs_clear_layoutcommitting(struct inode * inode)3111  static void pnfs_clear_layoutcommitting(struct inode *inode)
3112  {
3113  	unsigned long *bitlock = &NFS_I(inode)->flags;
3114  
3115  	clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
3116  	smp_mb__after_atomic();
3117  	wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
3118  }
3119  
3120  /*
3121   * There can be multiple RW segments.
3122   */
pnfs_list_write_lseg(struct inode * inode,struct list_head * listp)3123  static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
3124  {
3125  	struct pnfs_layout_segment *lseg;
3126  
3127  	list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
3128  		if (lseg->pls_range.iomode == IOMODE_RW &&
3129  		    test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
3130  			list_add(&lseg->pls_lc_list, listp);
3131  	}
3132  }
3133  
pnfs_list_write_lseg_done(struct inode * inode,struct list_head * listp)3134  static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
3135  {
3136  	struct pnfs_layout_segment *lseg, *tmp;
3137  
3138  	/* Matched by references in pnfs_set_layoutcommit */
3139  	list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
3140  		list_del_init(&lseg->pls_lc_list);
3141  		pnfs_put_lseg(lseg);
3142  	}
3143  
3144  	pnfs_clear_layoutcommitting(inode);
3145  }
3146  
pnfs_set_lo_fail(struct pnfs_layout_segment * lseg)3147  void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
3148  {
3149  	pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
3150  }
3151  EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
3152  
3153  void
pnfs_set_layoutcommit(struct inode * inode,struct pnfs_layout_segment * lseg,loff_t end_pos)3154  pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
3155  		loff_t end_pos)
3156  {
3157  	struct nfs_inode *nfsi = NFS_I(inode);
3158  	bool mark_as_dirty = false;
3159  
3160  	spin_lock(&inode->i_lock);
3161  	if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
3162  		nfsi->layout->plh_lwb = end_pos;
3163  		mark_as_dirty = true;
3164  		dprintk("%s: Set layoutcommit for inode %lu ",
3165  			__func__, inode->i_ino);
3166  	} else if (end_pos > nfsi->layout->plh_lwb)
3167  		nfsi->layout->plh_lwb = end_pos;
3168  	if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
3169  		/* references matched in nfs4_layoutcommit_release */
3170  		pnfs_get_lseg(lseg);
3171  	}
3172  	spin_unlock(&inode->i_lock);
3173  	dprintk("%s: lseg %p end_pos %llu\n",
3174  		__func__, lseg, nfsi->layout->plh_lwb);
3175  
3176  	/* if pnfs_layoutcommit_inode() runs between inode locks, the next one
3177  	 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
3178  	if (mark_as_dirty)
3179  		mark_inode_dirty_sync(inode);
3180  }
3181  EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
3182  
pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data * data)3183  void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
3184  {
3185  	struct nfs_server *nfss = NFS_SERVER(data->args.inode);
3186  
3187  	if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
3188  		nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
3189  	pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
3190  }
3191  
3192  /*
3193   * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
3194   * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
3195   * data to disk to allow the server to recover the data if it crashes.
3196   * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
3197   * is off, and a COMMIT is sent to a data server, or
3198   * if WRITEs to a data server return NFS_DATA_SYNC.
3199   */
3200  int
pnfs_layoutcommit_inode(struct inode * inode,bool sync)3201  pnfs_layoutcommit_inode(struct inode *inode, bool sync)
3202  {
3203  	struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
3204  	struct nfs4_layoutcommit_data *data;
3205  	struct nfs_inode *nfsi = NFS_I(inode);
3206  	loff_t end_pos;
3207  	int status;
3208  
3209  	if (!pnfs_layoutcommit_outstanding(inode))
3210  		return 0;
3211  
3212  	dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
3213  
3214  	status = -EAGAIN;
3215  	if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
3216  		if (!sync)
3217  			goto out;
3218  		status = wait_on_bit_lock_action(&nfsi->flags,
3219  				NFS_INO_LAYOUTCOMMITTING,
3220  				nfs_wait_bit_killable,
3221  				TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
3222  		if (status)
3223  			goto out;
3224  	}
3225  
3226  	status = -ENOMEM;
3227  	/* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
3228  	data = kzalloc(sizeof(*data), nfs_io_gfp_mask());
3229  	if (!data)
3230  		goto clear_layoutcommitting;
3231  
3232  	status = 0;
3233  	spin_lock(&inode->i_lock);
3234  	if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
3235  		goto out_unlock;
3236  
3237  	INIT_LIST_HEAD(&data->lseg_list);
3238  	pnfs_list_write_lseg(inode, &data->lseg_list);
3239  
3240  	end_pos = nfsi->layout->plh_lwb;
3241  
3242  	nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
3243  	data->cred = get_cred(nfsi->layout->plh_lc_cred);
3244  	spin_unlock(&inode->i_lock);
3245  
3246  	data->args.inode = inode;
3247  	nfs_fattr_init(&data->fattr);
3248  	data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
3249  	data->res.fattr = &data->fattr;
3250  	if (end_pos != 0)
3251  		data->args.lastbytewritten = end_pos - 1;
3252  	else
3253  		data->args.lastbytewritten = U64_MAX;
3254  	data->res.server = NFS_SERVER(inode);
3255  
3256  	if (ld->prepare_layoutcommit) {
3257  		status = ld->prepare_layoutcommit(&data->args);
3258  		if (status) {
3259  			put_cred(data->cred);
3260  			spin_lock(&inode->i_lock);
3261  			set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
3262  			if (end_pos > nfsi->layout->plh_lwb)
3263  				nfsi->layout->plh_lwb = end_pos;
3264  			goto out_unlock;
3265  		}
3266  	}
3267  
3268  
3269  	status = nfs4_proc_layoutcommit(data, sync);
3270  out:
3271  	if (status)
3272  		mark_inode_dirty_sync(inode);
3273  	dprintk("<-- %s status %d\n", __func__, status);
3274  	return status;
3275  out_unlock:
3276  	spin_unlock(&inode->i_lock);
3277  	kfree(data);
3278  clear_layoutcommitting:
3279  	pnfs_clear_layoutcommitting(inode);
3280  	goto out;
3281  }
3282  EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
3283  
3284  int
pnfs_generic_sync(struct inode * inode,bool datasync)3285  pnfs_generic_sync(struct inode *inode, bool datasync)
3286  {
3287  	return pnfs_layoutcommit_inode(inode, true);
3288  }
3289  EXPORT_SYMBOL_GPL(pnfs_generic_sync);
3290  
pnfs_mdsthreshold_alloc(void)3291  struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
3292  {
3293  	struct nfs4_threshold *thp;
3294  
3295  	thp = kzalloc(sizeof(*thp), nfs_io_gfp_mask());
3296  	if (!thp) {
3297  		dprintk("%s mdsthreshold allocation failed\n", __func__);
3298  		return NULL;
3299  	}
3300  	return thp;
3301  }
3302  
3303  #if IS_ENABLED(CONFIG_NFS_V4_2)
3304  int
pnfs_report_layoutstat(struct inode * inode,gfp_t gfp_flags)3305  pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
3306  {
3307  	struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
3308  	struct nfs_server *server = NFS_SERVER(inode);
3309  	struct nfs_inode *nfsi = NFS_I(inode);
3310  	struct nfs42_layoutstat_data *data;
3311  	struct pnfs_layout_hdr *hdr;
3312  	int status = 0;
3313  
3314  	if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
3315  		goto out;
3316  
3317  	if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
3318  		goto out;
3319  
3320  	if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
3321  		goto out;
3322  
3323  	spin_lock(&inode->i_lock);
3324  	if (!NFS_I(inode)->layout) {
3325  		spin_unlock(&inode->i_lock);
3326  		goto out_clear_layoutstats;
3327  	}
3328  	hdr = NFS_I(inode)->layout;
3329  	pnfs_get_layout_hdr(hdr);
3330  	spin_unlock(&inode->i_lock);
3331  
3332  	data = kzalloc(sizeof(*data), gfp_flags);
3333  	if (!data) {
3334  		status = -ENOMEM;
3335  		goto out_put;
3336  	}
3337  
3338  	data->args.fh = NFS_FH(inode);
3339  	data->args.inode = inode;
3340  	status = ld->prepare_layoutstats(&data->args);
3341  	if (status)
3342  		goto out_free;
3343  
3344  	status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);
3345  
3346  out:
3347  	dprintk("%s returns %d\n", __func__, status);
3348  	return status;
3349  
3350  out_free:
3351  	kfree(data);
3352  out_put:
3353  	pnfs_put_layout_hdr(hdr);
3354  out_clear_layoutstats:
3355  	smp_mb__before_atomic();
3356  	clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
3357  	smp_mb__after_atomic();
3358  	goto out;
3359  }
3360  EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
3361  #endif
3362  
3363  unsigned int layoutstats_timer;
3364  module_param(layoutstats_timer, uint, 0644);
3365  EXPORT_SYMBOL_GPL(layoutstats_timer);
3366