blob: 4347039ecc183d538c23f32019e5213da2ebf2f4 [file] [log] [blame]
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001#include <linux/ceph/ceph_debug.h>
Greg Farnum40819f62010-08-02 15:34:23 -07002
3#include <linux/file.h>
4#include <linux/namei.h>
Yan, Zhengeb13e832014-03-09 23:16:40 +08005#include <linux/random.h>
Greg Farnum40819f62010-08-02 15:34:23 -07006
7#include "super.h"
8#include "mds_client.h"
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07009#include <linux/ceph/pagelist.h>
Greg Farnum40819f62010-08-02 15:34:23 -070010
Yan, Zhengeb13e832014-03-09 23:16:40 +080011static u64 lock_secret;
Yan, Zheng9280be22014-10-14 10:33:35 +080012static int ceph_lock_wait_for_completion(struct ceph_mds_client *mdsc,
13 struct ceph_mds_request *req);
Yan, Zhengeb13e832014-03-09 23:16:40 +080014
15static inline u64 secure_addr(void *addr)
16{
17 u64 v = lock_secret ^ (u64)(unsigned long)addr;
18 /*
19 * Set the most significant bit, so that MDS knows the 'owner'
20 * is sufficient to identify the owner of lock. (old code uses
21 * both 'owner' and 'pid')
22 */
23 v |= (1ULL << 63);
24 return v;
25}
26
27void __init ceph_flock_init(void)
28{
29 get_random_bytes(&lock_secret, sizeof(lock_secret));
30}
31
Greg Farnum40819f62010-08-02 15:34:23 -070032/**
33 * Implement fcntl and flock locking functions.
34 */
35static int ceph_lock_message(u8 lock_type, u16 operation, struct file *file,
Herb Shiu637ae8d2010-11-23 13:42:23 -080036 int cmd, u8 wait, struct file_lock *fl)
Greg Farnum40819f62010-08-02 15:34:23 -070037{
Al Viro496ad9a2013-01-23 17:07:38 -050038 struct inode *inode = file_inode(file);
Yan, Zhengeb13e832014-03-09 23:16:40 +080039 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Greg Farnum40819f62010-08-02 15:34:23 -070040 struct ceph_mds_request *req;
41 int err;
Herb Shiu637ae8d2010-11-23 13:42:23 -080042 u64 length = 0;
Yan, Zhengeb13e832014-03-09 23:16:40 +080043 u64 owner;
Greg Farnum40819f62010-08-02 15:34:23 -070044
Yan, Zheng9280be22014-10-14 10:33:35 +080045 if (operation != CEPH_MDS_OP_SETFILELOCK || cmd == CEPH_LOCK_UNLOCK)
46 wait = 0;
47
Greg Farnum40819f62010-08-02 15:34:23 -070048 req = ceph_mdsc_create_request(mdsc, operation, USE_AUTH_MDS);
49 if (IS_ERR(req))
50 return PTR_ERR(req);
Sage Weil70b666c2011-05-27 09:24:26 -070051 req->r_inode = inode;
52 ihold(inode);
Yan, Zheng3bd58142014-04-27 09:17:45 +080053 req->r_num_caps = 1;
Greg Farnum40819f62010-08-02 15:34:23 -070054
Herb Shiu637ae8d2010-11-23 13:42:23 -080055 /* mds requires start and length rather than start and end */
56 if (LLONG_MAX == fl->fl_end)
57 length = 0;
58 else
59 length = fl->fl_end - fl->fl_start + 1;
60
Jeff Layton130d1f92014-05-09 14:13:04 -040061 owner = secure_addr(fl->fl_owner);
Yan, Zhengeb13e832014-03-09 23:16:40 +080062
63 dout("ceph_lock_message: rule: %d, op: %d, owner: %llx, pid: %llu, "
64 "start: %llu, length: %llu, wait: %d, type: %d", (int)lock_type,
65 (int)operation, owner, (u64)fl->fl_pid, fl->fl_start, length,
66 wait, fl->fl_type);
Herb Shiu637ae8d2010-11-23 13:42:23 -080067
Greg Farnum40819f62010-08-02 15:34:23 -070068 req->r_args.filelock_change.rule = lock_type;
69 req->r_args.filelock_change.type = cmd;
Yan, Zhengeb13e832014-03-09 23:16:40 +080070 req->r_args.filelock_change.owner = cpu_to_le64(owner);
Herb Shiu637ae8d2010-11-23 13:42:23 -080071 req->r_args.filelock_change.pid = cpu_to_le64((u64)fl->fl_pid);
Herb Shiu637ae8d2010-11-23 13:42:23 -080072 req->r_args.filelock_change.start = cpu_to_le64(fl->fl_start);
Greg Farnum40819f62010-08-02 15:34:23 -070073 req->r_args.filelock_change.length = cpu_to_le64(length);
74 req->r_args.filelock_change.wait = wait;
75
Yan, Zheng9280be22014-10-14 10:33:35 +080076 if (wait)
77 req->r_wait_for_completion = ceph_lock_wait_for_completion;
78
Greg Farnum40819f62010-08-02 15:34:23 -070079 err = ceph_mdsc_do_request(mdsc, inode, req);
Herb Shiua5b10622010-11-23 13:58:29 -080080
Yan, Zhengeb13e832014-03-09 23:16:40 +080081 if (operation == CEPH_MDS_OP_GETFILELOCK) {
Herb Shiua5b10622010-11-23 13:58:29 -080082 fl->fl_pid = le64_to_cpu(req->r_reply_info.filelock_reply->pid);
83 if (CEPH_LOCK_SHARED == req->r_reply_info.filelock_reply->type)
84 fl->fl_type = F_RDLCK;
85 else if (CEPH_LOCK_EXCL == req->r_reply_info.filelock_reply->type)
86 fl->fl_type = F_WRLCK;
87 else
88 fl->fl_type = F_UNLCK;
89
90 fl->fl_start = le64_to_cpu(req->r_reply_info.filelock_reply->start);
91 length = le64_to_cpu(req->r_reply_info.filelock_reply->start) +
92 le64_to_cpu(req->r_reply_info.filelock_reply->length);
93 if (length >= 1)
94 fl->fl_end = length -1;
95 else
96 fl->fl_end = 0;
97
98 }
Greg Farnum40819f62010-08-02 15:34:23 -070099 ceph_mdsc_put_request(req);
100 dout("ceph_lock_message: rule: %d, op: %d, pid: %llu, start: %llu, "
Sage Weil0c1f91f2011-05-25 14:56:12 -0700101 "length: %llu, wait: %d, type: %d, err code %d", (int)lock_type,
Herb Shiu637ae8d2010-11-23 13:42:23 -0800102 (int)operation, (u64)fl->fl_pid, fl->fl_start,
103 length, wait, fl->fl_type, err);
Greg Farnum40819f62010-08-02 15:34:23 -0700104 return err;
105}
106
Yan, Zheng9280be22014-10-14 10:33:35 +0800107static int ceph_lock_wait_for_completion(struct ceph_mds_client *mdsc,
108 struct ceph_mds_request *req)
109{
110 struct ceph_mds_request *intr_req;
111 struct inode *inode = req->r_inode;
112 int err, lock_type;
113
114 BUG_ON(req->r_op != CEPH_MDS_OP_SETFILELOCK);
115 if (req->r_args.filelock_change.rule == CEPH_LOCK_FCNTL)
116 lock_type = CEPH_LOCK_FCNTL_INTR;
117 else if (req->r_args.filelock_change.rule == CEPH_LOCK_FLOCK)
118 lock_type = CEPH_LOCK_FLOCK_INTR;
119 else
120 BUG_ON(1);
121 BUG_ON(req->r_args.filelock_change.type == CEPH_LOCK_UNLOCK);
122
123 err = wait_for_completion_interruptible(&req->r_completion);
124 if (!err)
125 return 0;
126
127 dout("ceph_lock_wait_for_completion: request %llu was interrupted\n",
128 req->r_tid);
129
130 intr_req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETFILELOCK,
131 USE_AUTH_MDS);
132 if (IS_ERR(intr_req))
133 return PTR_ERR(intr_req);
134
135 intr_req->r_inode = inode;
136 ihold(inode);
137 intr_req->r_num_caps = 1;
138
139 intr_req->r_args.filelock_change = req->r_args.filelock_change;
140 intr_req->r_args.filelock_change.rule = lock_type;
141 intr_req->r_args.filelock_change.type = CEPH_LOCK_UNLOCK;
142
143 err = ceph_mdsc_do_request(mdsc, inode, intr_req);
144 ceph_mdsc_put_request(intr_req);
145
146 if (err && err != -ERESTARTSYS)
147 return err;
148
149 wait_for_completion(&req->r_completion);
150 return 0;
151}
152
Greg Farnum40819f62010-08-02 15:34:23 -0700153/**
154 * Attempt to set an fcntl lock.
155 * For now, this just goes away to the server. Later it may be more awesome.
156 */
157int ceph_lock(struct file *file, int cmd, struct file_lock *fl)
158{
Greg Farnum40819f62010-08-02 15:34:23 -0700159 u8 lock_cmd;
160 int err;
161 u8 wait = 0;
162 u16 op = CEPH_MDS_OP_SETFILELOCK;
163
Yan, Zhengeb70c0c2014-03-04 15:50:06 +0800164 if (!(fl->fl_flags & FL_POSIX))
165 return -ENOLCK;
166 /* No mandatory locks */
167 if (__mandatory_lock(file->f_mapping->host) && fl->fl_type != F_UNLCK)
168 return -ENOLCK;
169
Yan, Zhengeb13e832014-03-09 23:16:40 +0800170 dout("ceph_lock, fl_owner: %p", fl->fl_owner);
Greg Farnum40819f62010-08-02 15:34:23 -0700171
172 /* set wait bit as appropriate, then make command as Ceph expects it*/
Yan, Zheng0e8e95d2014-03-04 15:42:24 +0800173 if (IS_GETLK(cmd))
Greg Farnum40819f62010-08-02 15:34:23 -0700174 op = CEPH_MDS_OP_GETFILELOCK;
Yan, Zheng0e8e95d2014-03-04 15:42:24 +0800175 else if (IS_SETLKW(cmd))
176 wait = 1;
Greg Farnum40819f62010-08-02 15:34:23 -0700177
178 if (F_RDLCK == fl->fl_type)
179 lock_cmd = CEPH_LOCK_SHARED;
180 else if (F_WRLCK == fl->fl_type)
181 lock_cmd = CEPH_LOCK_EXCL;
182 else
183 lock_cmd = CEPH_LOCK_UNLOCK;
184
Herb Shiu637ae8d2010-11-23 13:42:23 -0800185 err = ceph_lock_message(CEPH_LOCK_FCNTL, op, file, lock_cmd, wait, fl);
Greg Farnum40819f62010-08-02 15:34:23 -0700186 if (!err) {
Yan, Zhengeb13e832014-03-09 23:16:40 +0800187 if (op != CEPH_MDS_OP_GETFILELOCK) {
Herb Shiua5b10622010-11-23 13:58:29 -0800188 dout("mds locked, locking locally");
189 err = posix_lock_file(file, fl, NULL);
190 if (err && (CEPH_MDS_OP_SETFILELOCK == op)) {
Sage Weil0c1f91f2011-05-25 14:56:12 -0700191 /* undo! This should only happen if
192 * the kernel detects local
193 * deadlock. */
Herb Shiua5b10622010-11-23 13:58:29 -0800194 ceph_lock_message(CEPH_LOCK_FCNTL, op, file,
195 CEPH_LOCK_UNLOCK, 0, fl);
Sage Weil0c1f91f2011-05-25 14:56:12 -0700196 dout("got %d on posix_lock_file, undid lock",
197 err);
Herb Shiua5b10622010-11-23 13:58:29 -0800198 }
Greg Farnum40819f62010-08-02 15:34:23 -0700199 }
Greg Farnum40819f62010-08-02 15:34:23 -0700200 }
201 return err;
202}
203
204int ceph_flock(struct file *file, int cmd, struct file_lock *fl)
205{
Greg Farnum40819f62010-08-02 15:34:23 -0700206 u8 lock_cmd;
207 int err;
Yan, Zheng0e8e95d2014-03-04 15:42:24 +0800208 u8 wait = 0;
Greg Farnum40819f62010-08-02 15:34:23 -0700209
Yan, Zhengeb70c0c2014-03-04 15:50:06 +0800210 if (!(fl->fl_flags & FL_FLOCK))
211 return -ENOLCK;
212 /* No mandatory locks */
213 if (__mandatory_lock(file->f_mapping->host) && fl->fl_type != F_UNLCK)
214 return -ENOLCK;
215
Yan, Zhengeb13e832014-03-09 23:16:40 +0800216 dout("ceph_flock, fl_file: %p", fl->fl_file);
Greg Farnum40819f62010-08-02 15:34:23 -0700217
Yan, Zheng0e8e95d2014-03-04 15:42:24 +0800218 if (IS_SETLKW(cmd))
219 wait = 1;
220
221 if (F_RDLCK == fl->fl_type)
Greg Farnum40819f62010-08-02 15:34:23 -0700222 lock_cmd = CEPH_LOCK_SHARED;
Yan, Zheng0e8e95d2014-03-04 15:42:24 +0800223 else if (F_WRLCK == fl->fl_type)
Greg Farnum40819f62010-08-02 15:34:23 -0700224 lock_cmd = CEPH_LOCK_EXCL;
225 else
226 lock_cmd = CEPH_LOCK_UNLOCK;
Greg Farnum40819f62010-08-02 15:34:23 -0700227
228 err = ceph_lock_message(CEPH_LOCK_FLOCK, CEPH_MDS_OP_SETFILELOCK,
Herb Shiu637ae8d2010-11-23 13:42:23 -0800229 file, lock_cmd, wait, fl);
Greg Farnum40819f62010-08-02 15:34:23 -0700230 if (!err) {
231 err = flock_lock_file_wait(file, fl);
232 if (err) {
233 ceph_lock_message(CEPH_LOCK_FLOCK,
234 CEPH_MDS_OP_SETFILELOCK,
Herb Shiu637ae8d2010-11-23 13:42:23 -0800235 file, CEPH_LOCK_UNLOCK, 0, fl);
Greg Farnum40819f62010-08-02 15:34:23 -0700236 dout("got %d on flock_lock_file_wait, undid lock", err);
237 }
Greg Farnum40819f62010-08-02 15:34:23 -0700238 }
239 return err;
240}
241
Jeff Layton5263e312015-01-16 15:05:55 -0500242/*
243 * Fills in the passed counter variables, so you can prepare pagelist metadata
244 * before calling ceph_encode_locks.
Greg Farnum40819f62010-08-02 15:34:23 -0700245 */
246void ceph_count_locks(struct inode *inode, int *fcntl_count, int *flock_count)
247{
Jeff Laytone084c1b2015-02-16 14:32:03 -0500248 struct file_lock *lock;
Jeff Layton5263e312015-01-16 15:05:55 -0500249 struct file_lock_context *ctx;
Greg Farnum40819f62010-08-02 15:34:23 -0700250
251 *fcntl_count = 0;
252 *flock_count = 0;
253
Jeff Layton5263e312015-01-16 15:05:55 -0500254 ctx = inode->i_flctx;
255 if (ctx) {
Jeff Laytone084c1b2015-02-16 14:32:03 -0500256 spin_lock(&ctx->flc_lock);
257 list_for_each_entry(lock, &ctx->flc_posix, fl_list)
258 ++(*fcntl_count);
259 list_for_each_entry(lock, &ctx->flc_flock, fl_list)
260 ++(*flock_count);
261 spin_unlock(&ctx->flc_lock);
Greg Farnum40819f62010-08-02 15:34:23 -0700262 }
263 dout("counted %d flock locks and %d fcntl locks",
264 *flock_count, *fcntl_count);
265}
266
267/**
Jim Schutt39be95e2013-05-15 13:03:35 -0500268 * Encode the flock and fcntl locks for the given inode into the ceph_filelock
Jeff Layton1c8c6012013-06-21 08:58:15 -0400269 * array. Must be called with inode->i_lock already held.
Jim Schutt39be95e2013-05-15 13:03:35 -0500270 * If we encounter more of a specific lock type than expected, return -ENOSPC.
Greg Farnum40819f62010-08-02 15:34:23 -0700271 */
Jim Schutt39be95e2013-05-15 13:03:35 -0500272int ceph_encode_locks_to_buffer(struct inode *inode,
273 struct ceph_filelock *flocks,
274 int num_fcntl_locks, int num_flock_locks)
Greg Farnum40819f62010-08-02 15:34:23 -0700275{
276 struct file_lock *lock;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500277 struct file_lock_context *ctx = inode->i_flctx;
Greg Farnum40819f62010-08-02 15:34:23 -0700278 int err = 0;
Greg Farnumfca44512010-09-17 10:24:02 -0700279 int seen_fcntl = 0;
280 int seen_flock = 0;
Jim Schutt39be95e2013-05-15 13:03:35 -0500281 int l = 0;
Greg Farnum40819f62010-08-02 15:34:23 -0700282
283 dout("encoding %d flock and %d fcntl locks", num_flock_locks,
284 num_fcntl_locks);
Jim Schutt39be95e2013-05-15 13:03:35 -0500285
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500286 if (!ctx)
287 return 0;
Jeff Layton5263e312015-01-16 15:05:55 -0500288
Jeff Layton6109c852015-01-16 15:05:57 -0500289 spin_lock(&ctx->flc_lock);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500290 list_for_each_entry(lock, &ctx->flc_flock, fl_list) {
291 ++seen_fcntl;
292 if (seen_fcntl > num_fcntl_locks) {
293 err = -ENOSPC;
294 goto fail;
Greg Farnum40819f62010-08-02 15:34:23 -0700295 }
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500296 err = lock_to_ceph_filelock(lock, &flocks[l]);
297 if (err)
298 goto fail;
299 ++l;
300 }
301 list_for_each_entry(lock, &ctx->flc_flock, fl_list) {
302 ++seen_flock;
303 if (seen_flock > num_flock_locks) {
304 err = -ENOSPC;
305 goto fail;
306 }
307 err = lock_to_ceph_filelock(lock, &flocks[l]);
308 if (err)
309 goto fail;
310 ++l;
Greg Farnum40819f62010-08-02 15:34:23 -0700311 }
312fail:
Jeff Layton6109c852015-01-16 15:05:57 -0500313 spin_unlock(&ctx->flc_lock);
Greg Farnum40819f62010-08-02 15:34:23 -0700314 return err;
315}
316
Jim Schutt39be95e2013-05-15 13:03:35 -0500317/**
318 * Copy the encoded flock and fcntl locks into the pagelist.
319 * Format is: #fcntl locks, sequential fcntl locks, #flock locks,
320 * sequential flock locks.
321 * Returns zero on success.
322 */
323int ceph_locks_to_pagelist(struct ceph_filelock *flocks,
324 struct ceph_pagelist *pagelist,
325 int num_fcntl_locks, int num_flock_locks)
326{
327 int err = 0;
328 __le32 nlocks;
329
330 nlocks = cpu_to_le32(num_fcntl_locks);
331 err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
332 if (err)
333 goto out_fail;
334
335 err = ceph_pagelist_append(pagelist, flocks,
336 num_fcntl_locks * sizeof(*flocks));
337 if (err)
338 goto out_fail;
339
340 nlocks = cpu_to_le32(num_flock_locks);
341 err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
342 if (err)
343 goto out_fail;
344
345 err = ceph_pagelist_append(pagelist,
346 &flocks[num_fcntl_locks],
347 num_flock_locks * sizeof(*flocks));
348out_fail:
349 return err;
350}
351
Greg Farnum40819f62010-08-02 15:34:23 -0700352/*
353 * Given a pointer to a lock, convert it to a ceph filelock
354 */
355int lock_to_ceph_filelock(struct file_lock *lock,
356 struct ceph_filelock *cephlock)
357{
358 int err = 0;
Greg Farnum40819f62010-08-02 15:34:23 -0700359 cephlock->start = cpu_to_le64(lock->fl_start);
360 cephlock->length = cpu_to_le64(lock->fl_end - lock->fl_start + 1);
361 cephlock->client = cpu_to_le64(0);
Yan, Zhengeb13e832014-03-09 23:16:40 +0800362 cephlock->pid = cpu_to_le64((u64)lock->fl_pid);
Jeff Layton130d1f92014-05-09 14:13:04 -0400363 cephlock->owner = cpu_to_le64(secure_addr(lock->fl_owner));
Greg Farnum40819f62010-08-02 15:34:23 -0700364
365 switch (lock->fl_type) {
366 case F_RDLCK:
367 cephlock->type = CEPH_LOCK_SHARED;
368 break;
369 case F_WRLCK:
370 cephlock->type = CEPH_LOCK_EXCL;
371 break;
372 case F_UNLCK:
373 cephlock->type = CEPH_LOCK_UNLOCK;
374 break;
375 default:
376 dout("Have unknown lock type %d", lock->fl_type);
377 err = -EINVAL;
378 }
379
380 return err;
381}