Back to home page

OSCL-LXR

 
 

    


0001 // SPDX-License-Identifier: GPL-2.0
0002 /*
0003  * linux/fs/lockd/svclock.c
0004  *
0005  * Handling of server-side locks, mostly of the blocked variety.
0006  * This is the ugliest part of lockd because we tread on very thin ice.
0007  * GRANT and CANCEL calls may get stuck, meet in mid-flight, etc.
0008  * IMNSHO introducing the grant callback into the NLM protocol was one
0009  * of the worst ideas Sun ever had. Except maybe for the idea of doing
0010  * NFS file locking at all.
0011  *
0012  * I'm trying hard to avoid race conditions by protecting most accesses
0013  * to a file's list of blocked locks through a semaphore. The global
0014  * list of blocked locks is not protected in this fashion however.
0015  * Therefore, some functions (such as the RPC callback for the async grant
0016  * call) move blocked locks towards the head of the list *while some other
0017  * process might be traversing it*. This should not be a problem in
0018  * practice, because this will only cause functions traversing the list
0019  * to visit some blocks twice.
0020  *
0021  * Copyright (C) 1996, Olaf Kirch <okir@monad.swb.de>
0022  */
0023 
0024 #include <linux/types.h>
0025 #include <linux/slab.h>
0026 #include <linux/errno.h>
0027 #include <linux/kernel.h>
0028 #include <linux/sched.h>
0029 #include <linux/sunrpc/clnt.h>
0030 #include <linux/sunrpc/svc_xprt.h>
0031 #include <linux/lockd/nlm.h>
0032 #include <linux/lockd/lockd.h>
0033 #include <linux/kthread.h>
0034 #include <linux/exportfs.h>
0035 
0036 #define NLMDBG_FACILITY     NLMDBG_SVCLOCK
0037 
0038 #ifdef CONFIG_LOCKD_V4
0039 #define nlm_deadlock    nlm4_deadlock
0040 #else
0041 #define nlm_deadlock    nlm_lck_denied
0042 #endif
0043 
0044 static void nlmsvc_release_block(struct nlm_block *block);
0045 static void nlmsvc_insert_block(struct nlm_block *block, unsigned long);
0046 static void nlmsvc_remove_block(struct nlm_block *block);
0047 
0048 static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock);
0049 static void nlmsvc_freegrantargs(struct nlm_rqst *call);
0050 static const struct rpc_call_ops nlmsvc_grant_ops;
0051 
0052 /*
0053  * The list of blocked locks to retry
0054  */
0055 static LIST_HEAD(nlm_blocked);
0056 static DEFINE_SPINLOCK(nlm_blocked_lock);
0057 
0058 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
0059 static const char *nlmdbg_cookie2a(const struct nlm_cookie *cookie)
0060 {
0061     /*
0062      * We can get away with a static buffer because this is only called
0063      * from lockd, which is single-threaded.
0064      */
0065     static char buf[2*NLM_MAXCOOKIELEN+1];
0066     unsigned int i, len = sizeof(buf);
0067     char *p = buf;
0068 
0069     len--;  /* allow for trailing \0 */
0070     if (len < 3)
0071         return "???";
0072     for (i = 0 ; i < cookie->len ; i++) {
0073         if (len < 2) {
0074             strcpy(p-3, "...");
0075             break;
0076         }
0077         sprintf(p, "%02x", cookie->data[i]);
0078         p += 2;
0079         len -= 2;
0080     }
0081     *p = '\0';
0082 
0083     return buf;
0084 }
0085 #endif
0086 
0087 /*
0088  * Insert a blocked lock into the global list
0089  */
0090 static void
0091 nlmsvc_insert_block_locked(struct nlm_block *block, unsigned long when)
0092 {
0093     struct nlm_block *b;
0094     struct list_head *pos;
0095 
0096     dprintk("lockd: nlmsvc_insert_block(%p, %ld)\n", block, when);
0097     if (list_empty(&block->b_list)) {
0098         kref_get(&block->b_count);
0099     } else {
0100         list_del_init(&block->b_list);
0101     }
0102 
0103     pos = &nlm_blocked;
0104     if (when != NLM_NEVER) {
0105         if ((when += jiffies) == NLM_NEVER)
0106             when ++;
0107         list_for_each(pos, &nlm_blocked) {
0108             b = list_entry(pos, struct nlm_block, b_list);
0109             if (time_after(b->b_when,when) || b->b_when == NLM_NEVER)
0110                 break;
0111         }
0112         /* On normal exit from the loop, pos == &nlm_blocked,
0113          * so we will be adding to the end of the list - good
0114          */
0115     }
0116 
0117     list_add_tail(&block->b_list, pos);
0118     block->b_when = when;
0119 }
0120 
0121 static void nlmsvc_insert_block(struct nlm_block *block, unsigned long when)
0122 {
0123     spin_lock(&nlm_blocked_lock);
0124     nlmsvc_insert_block_locked(block, when);
0125     spin_unlock(&nlm_blocked_lock);
0126 }
0127 
0128 /*
0129  * Remove a block from the global list
0130  */
0131 static inline void
0132 nlmsvc_remove_block(struct nlm_block *block)
0133 {
0134     if (!list_empty(&block->b_list)) {
0135         spin_lock(&nlm_blocked_lock);
0136         list_del_init(&block->b_list);
0137         spin_unlock(&nlm_blocked_lock);
0138         nlmsvc_release_block(block);
0139     }
0140 }
0141 
0142 /*
0143  * Find a block for a given lock
0144  */
0145 static struct nlm_block *
0146 nlmsvc_lookup_block(struct nlm_file *file, struct nlm_lock *lock)
0147 {
0148     struct nlm_block    *block;
0149     struct file_lock    *fl;
0150 
0151     dprintk("lockd: nlmsvc_lookup_block f=%p pd=%d %Ld-%Ld ty=%d\n",
0152                 file, lock->fl.fl_pid,
0153                 (long long)lock->fl.fl_start,
0154                 (long long)lock->fl.fl_end, lock->fl.fl_type);
0155     list_for_each_entry(block, &nlm_blocked, b_list) {
0156         fl = &block->b_call->a_args.lock.fl;
0157         dprintk("lockd: check f=%p pd=%d %Ld-%Ld ty=%d cookie=%s\n",
0158                 block->b_file, fl->fl_pid,
0159                 (long long)fl->fl_start,
0160                 (long long)fl->fl_end, fl->fl_type,
0161                 nlmdbg_cookie2a(&block->b_call->a_args.cookie));
0162         if (block->b_file == file && nlm_compare_locks(fl, &lock->fl)) {
0163             kref_get(&block->b_count);
0164             return block;
0165         }
0166     }
0167 
0168     return NULL;
0169 }
0170 
0171 static inline int nlm_cookie_match(struct nlm_cookie *a, struct nlm_cookie *b)
0172 {
0173     if (a->len != b->len)
0174         return 0;
0175     if (memcmp(a->data, b->data, a->len))
0176         return 0;
0177     return 1;
0178 }
0179 
0180 /*
0181  * Find a block with a given NLM cookie.
0182  */
0183 static inline struct nlm_block *
0184 nlmsvc_find_block(struct nlm_cookie *cookie)
0185 {
0186     struct nlm_block *block;
0187 
0188     list_for_each_entry(block, &nlm_blocked, b_list) {
0189         if (nlm_cookie_match(&block->b_call->a_args.cookie,cookie))
0190             goto found;
0191     }
0192 
0193     return NULL;
0194 
0195 found:
0196     dprintk("nlmsvc_find_block(%s): block=%p\n", nlmdbg_cookie2a(cookie), block);
0197     kref_get(&block->b_count);
0198     return block;
0199 }
0200 
0201 /*
0202  * Create a block and initialize it.
0203  *
0204  * Note: we explicitly set the cookie of the grant reply to that of
0205  * the blocked lock request. The spec explicitly mentions that the client
0206  * should _not_ rely on the callback containing the same cookie as the
0207  * request, but (as I found out later) that's because some implementations
0208  * do just this. Never mind the standards comittees, they support our
0209  * logging industries.
0210  *
0211  * 10 years later: I hope we can safely ignore these old and broken
0212  * clients by now. Let's fix this so we can uniquely identify an incoming
0213  * GRANTED_RES message by cookie, without having to rely on the client's IP
0214  * address. --okir
0215  */
0216 static struct nlm_block *
0217 nlmsvc_create_block(struct svc_rqst *rqstp, struct nlm_host *host,
0218             struct nlm_file *file, struct nlm_lock *lock,
0219             struct nlm_cookie *cookie)
0220 {
0221     struct nlm_block    *block;
0222     struct nlm_rqst     *call = NULL;
0223 
0224     call = nlm_alloc_call(host);
0225     if (call == NULL)
0226         return NULL;
0227 
0228     /* Allocate memory for block, and initialize arguments */
0229     block = kzalloc(sizeof(*block), GFP_KERNEL);
0230     if (block == NULL)
0231         goto failed;
0232     kref_init(&block->b_count);
0233     INIT_LIST_HEAD(&block->b_list);
0234     INIT_LIST_HEAD(&block->b_flist);
0235 
0236     if (!nlmsvc_setgrantargs(call, lock))
0237         goto failed_free;
0238 
0239     /* Set notifier function for VFS, and init args */
0240     call->a_args.lock.fl.fl_flags |= FL_SLEEP;
0241     call->a_args.lock.fl.fl_lmops = &nlmsvc_lock_operations;
0242     nlmclnt_next_cookie(&call->a_args.cookie);
0243 
0244     dprintk("lockd: created block %p...\n", block);
0245 
0246     /* Create and initialize the block */
0247     block->b_daemon = rqstp->rq_server;
0248     block->b_host   = host;
0249     block->b_file   = file;
0250     file->f_count++;
0251 
0252     /* Add to file's list of blocks */
0253     list_add(&block->b_flist, &file->f_blocks);
0254 
0255     /* Set up RPC arguments for callback */
0256     block->b_call = call;
0257     call->a_flags   = RPC_TASK_ASYNC;
0258     call->a_block = block;
0259 
0260     return block;
0261 
0262 failed_free:
0263     kfree(block);
0264 failed:
0265     nlmsvc_release_call(call);
0266     return NULL;
0267 }
0268 
0269 /*
0270  * Delete a block.
0271  * It is the caller's responsibility to check whether the file
0272  * can be closed hereafter.
0273  */
0274 static int nlmsvc_unlink_block(struct nlm_block *block)
0275 {
0276     int status;
0277     dprintk("lockd: unlinking block %p...\n", block);
0278 
0279     /* Remove block from list */
0280     status = locks_delete_block(&block->b_call->a_args.lock.fl);
0281     nlmsvc_remove_block(block);
0282     return status;
0283 }
0284 
0285 static void nlmsvc_free_block(struct kref *kref)
0286 {
0287     struct nlm_block *block = container_of(kref, struct nlm_block, b_count);
0288     struct nlm_file     *file = block->b_file;
0289 
0290     dprintk("lockd: freeing block %p...\n", block);
0291 
0292     /* Remove block from file's list of blocks */
0293     list_del_init(&block->b_flist);
0294     mutex_unlock(&file->f_mutex);
0295 
0296     nlmsvc_freegrantargs(block->b_call);
0297     nlmsvc_release_call(block->b_call);
0298     nlm_release_file(block->b_file);
0299     kfree(block);
0300 }
0301 
0302 static void nlmsvc_release_block(struct nlm_block *block)
0303 {
0304     if (block != NULL)
0305         kref_put_mutex(&block->b_count, nlmsvc_free_block, &block->b_file->f_mutex);
0306 }
0307 
0308 /*
0309  * Loop over all blocks and delete blocks held by
0310  * a matching host.
0311  */
0312 void nlmsvc_traverse_blocks(struct nlm_host *host,
0313             struct nlm_file *file,
0314             nlm_host_match_fn_t match)
0315 {
0316     struct nlm_block *block, *next;
0317 
0318 restart:
0319     mutex_lock(&file->f_mutex);
0320     list_for_each_entry_safe(block, next, &file->f_blocks, b_flist) {
0321         if (!match(block->b_host, host))
0322             continue;
0323         /* Do not destroy blocks that are not on
0324          * the global retry list - why? */
0325         if (list_empty(&block->b_list))
0326             continue;
0327         kref_get(&block->b_count);
0328         mutex_unlock(&file->f_mutex);
0329         nlmsvc_unlink_block(block);
0330         nlmsvc_release_block(block);
0331         goto restart;
0332     }
0333     mutex_unlock(&file->f_mutex);
0334 }
0335 
0336 static struct nlm_lockowner *
0337 nlmsvc_get_lockowner(struct nlm_lockowner *lockowner)
0338 {
0339     refcount_inc(&lockowner->count);
0340     return lockowner;
0341 }
0342 
0343 void nlmsvc_put_lockowner(struct nlm_lockowner *lockowner)
0344 {
0345     if (!refcount_dec_and_lock(&lockowner->count, &lockowner->host->h_lock))
0346         return;
0347     list_del(&lockowner->list);
0348     spin_unlock(&lockowner->host->h_lock);
0349     nlmsvc_release_host(lockowner->host);
0350     kfree(lockowner);
0351 }
0352 
0353 static struct nlm_lockowner *__nlmsvc_find_lockowner(struct nlm_host *host, pid_t pid)
0354 {
0355     struct nlm_lockowner *lockowner;
0356     list_for_each_entry(lockowner, &host->h_lockowners, list) {
0357         if (lockowner->pid != pid)
0358             continue;
0359         return nlmsvc_get_lockowner(lockowner);
0360     }
0361     return NULL;
0362 }
0363 
0364 static struct nlm_lockowner *nlmsvc_find_lockowner(struct nlm_host *host, pid_t pid)
0365 {
0366     struct nlm_lockowner *res, *new = NULL;
0367 
0368     spin_lock(&host->h_lock);
0369     res = __nlmsvc_find_lockowner(host, pid);
0370 
0371     if (res == NULL) {
0372         spin_unlock(&host->h_lock);
0373         new = kmalloc(sizeof(*res), GFP_KERNEL);
0374         spin_lock(&host->h_lock);
0375         res = __nlmsvc_find_lockowner(host, pid);
0376         if (res == NULL && new != NULL) {
0377             res = new;
0378             /* fs/locks.c will manage the refcount through lock_ops */
0379             refcount_set(&new->count, 1);
0380             new->pid = pid;
0381             new->host = nlm_get_host(host);
0382             list_add(&new->list, &host->h_lockowners);
0383             new = NULL;
0384         }
0385     }
0386 
0387     spin_unlock(&host->h_lock);
0388     kfree(new);
0389     return res;
0390 }
0391 
0392 void
0393 nlmsvc_release_lockowner(struct nlm_lock *lock)
0394 {
0395     if (lock->fl.fl_owner)
0396         nlmsvc_put_lockowner(lock->fl.fl_owner);
0397 }
0398 
0399 void nlmsvc_locks_init_private(struct file_lock *fl, struct nlm_host *host,
0400                         pid_t pid)
0401 {
0402     fl->fl_owner = nlmsvc_find_lockowner(host, pid);
0403 }
0404 
0405 /*
0406  * Initialize arguments for GRANTED call. The nlm_rqst structure
0407  * has been cleared already.
0408  */
0409 static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock)
0410 {
0411     locks_copy_lock(&call->a_args.lock.fl, &lock->fl);
0412     memcpy(&call->a_args.lock.fh, &lock->fh, sizeof(call->a_args.lock.fh));
0413     call->a_args.lock.caller = utsname()->nodename;
0414     call->a_args.lock.oh.len = lock->oh.len;
0415 
0416     /* set default data area */
0417     call->a_args.lock.oh.data = call->a_owner;
0418     call->a_args.lock.svid = ((struct nlm_lockowner *)lock->fl.fl_owner)->pid;
0419 
0420     if (lock->oh.len > NLMCLNT_OHSIZE) {
0421         void *data = kmalloc(lock->oh.len, GFP_KERNEL);
0422         if (!data)
0423             return 0;
0424         call->a_args.lock.oh.data = (u8 *) data;
0425     }
0426 
0427     memcpy(call->a_args.lock.oh.data, lock->oh.data, lock->oh.len);
0428     return 1;
0429 }
0430 
0431 static void nlmsvc_freegrantargs(struct nlm_rqst *call)
0432 {
0433     if (call->a_args.lock.oh.data != call->a_owner)
0434         kfree(call->a_args.lock.oh.data);
0435 
0436     locks_release_private(&call->a_args.lock.fl);
0437 }
0438 
0439 /*
0440  * Deferred lock request handling for non-blocking lock
0441  */
0442 static __be32
0443 nlmsvc_defer_lock_rqst(struct svc_rqst *rqstp, struct nlm_block *block)
0444 {
0445     __be32 status = nlm_lck_denied_nolocks;
0446 
0447     block->b_flags |= B_QUEUED;
0448 
0449     nlmsvc_insert_block(block, NLM_TIMEOUT);
0450 
0451     block->b_cache_req = &rqstp->rq_chandle;
0452     if (rqstp->rq_chandle.defer) {
0453         block->b_deferred_req =
0454             rqstp->rq_chandle.defer(block->b_cache_req);
0455         if (block->b_deferred_req != NULL)
0456             status = nlm_drop_reply;
0457     }
0458     dprintk("lockd: nlmsvc_defer_lock_rqst block %p flags %d status %d\n",
0459         block, block->b_flags, ntohl(status));
0460 
0461     return status;
0462 }
0463 
0464 /*
0465  * Attempt to establish a lock, and if it can't be granted, block it
0466  * if required.
0467  */
0468 __be32
0469 nlmsvc_lock(struct svc_rqst *rqstp, struct nlm_file *file,
0470         struct nlm_host *host, struct nlm_lock *lock, int wait,
0471         struct nlm_cookie *cookie, int reclaim)
0472 {
0473 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
0474     struct inode        *inode = nlmsvc_file_inode(file);
0475 #endif
0476     struct nlm_block    *block = NULL;
0477     int         error;
0478     int         mode;
0479     int         async_block = 0;
0480     __be32          ret;
0481 
0482     dprintk("lockd: nlmsvc_lock(%s/%ld, ty=%d, pi=%d, %Ld-%Ld, bl=%d)\n",
0483                 inode->i_sb->s_id, inode->i_ino,
0484                 lock->fl.fl_type, lock->fl.fl_pid,
0485                 (long long)lock->fl.fl_start,
0486                 (long long)lock->fl.fl_end,
0487                 wait);
0488 
0489     if (nlmsvc_file_file(file)->f_op->lock) {
0490         async_block = wait;
0491         wait = 0;
0492     }
0493 
0494     /* Lock file against concurrent access */
0495     mutex_lock(&file->f_mutex);
0496     /* Get existing block (in case client is busy-waiting)
0497      * or create new block
0498      */
0499     block = nlmsvc_lookup_block(file, lock);
0500     if (block == NULL) {
0501         block = nlmsvc_create_block(rqstp, host, file, lock, cookie);
0502         ret = nlm_lck_denied_nolocks;
0503         if (block == NULL)
0504             goto out;
0505         lock = &block->b_call->a_args.lock;
0506     } else
0507         lock->fl.fl_flags &= ~FL_SLEEP;
0508 
0509     if (block->b_flags & B_QUEUED) {
0510         dprintk("lockd: nlmsvc_lock deferred block %p flags %d\n",
0511                             block, block->b_flags);
0512         if (block->b_granted) {
0513             nlmsvc_unlink_block(block);
0514             ret = nlm_granted;
0515             goto out;
0516         }
0517         if (block->b_flags & B_TIMED_OUT) {
0518             nlmsvc_unlink_block(block);
0519             ret = nlm_lck_denied;
0520             goto out;
0521         }
0522         ret = nlm_drop_reply;
0523         goto out;
0524     }
0525 
0526     if (locks_in_grace(SVC_NET(rqstp)) && !reclaim) {
0527         ret = nlm_lck_denied_grace_period;
0528         goto out;
0529     }
0530     if (reclaim && !locks_in_grace(SVC_NET(rqstp))) {
0531         ret = nlm_lck_denied_grace_period;
0532         goto out;
0533     }
0534 
0535     if (!wait)
0536         lock->fl.fl_flags &= ~FL_SLEEP;
0537     mode = lock_to_openmode(&lock->fl);
0538     error = vfs_lock_file(file->f_file[mode], F_SETLK, &lock->fl, NULL);
0539     lock->fl.fl_flags &= ~FL_SLEEP;
0540 
0541     dprintk("lockd: vfs_lock_file returned %d\n", error);
0542     switch (error) {
0543         case 0:
0544             ret = nlm_granted;
0545             goto out;
0546         case -EAGAIN:
0547             /*
0548              * If this is a blocking request for an
0549              * already pending lock request then we need
0550              * to put it back on lockd's block list
0551              */
0552             if (wait)
0553                 break;
0554             ret = async_block ? nlm_lck_blocked : nlm_lck_denied;
0555             goto out;
0556         case FILE_LOCK_DEFERRED:
0557             if (wait)
0558                 break;
0559             /* Filesystem lock operation is in progress
0560                Add it to the queue waiting for callback */
0561             ret = nlmsvc_defer_lock_rqst(rqstp, block);
0562             goto out;
0563         case -EDEADLK:
0564             ret = nlm_deadlock;
0565             goto out;
0566         default:            /* includes ENOLCK */
0567             ret = nlm_lck_denied_nolocks;
0568             goto out;
0569     }
0570 
0571     ret = nlm_lck_blocked;
0572 
0573     /* Append to list of blocked */
0574     nlmsvc_insert_block(block, NLM_NEVER);
0575 out:
0576     mutex_unlock(&file->f_mutex);
0577     nlmsvc_release_block(block);
0578     dprintk("lockd: nlmsvc_lock returned %u\n", ret);
0579     return ret;
0580 }
0581 
0582 /*
0583  * Test for presence of a conflicting lock.
0584  */
0585 __be32
0586 nlmsvc_testlock(struct svc_rqst *rqstp, struct nlm_file *file,
0587         struct nlm_host *host, struct nlm_lock *lock,
0588         struct nlm_lock *conflock, struct nlm_cookie *cookie)
0589 {
0590     int         error;
0591     int         mode;
0592     __be32          ret;
0593 
0594     dprintk("lockd: nlmsvc_testlock(%s/%ld, ty=%d, %Ld-%Ld)\n",
0595                 nlmsvc_file_inode(file)->i_sb->s_id,
0596                 nlmsvc_file_inode(file)->i_ino,
0597                 lock->fl.fl_type,
0598                 (long long)lock->fl.fl_start,
0599                 (long long)lock->fl.fl_end);
0600 
0601     if (locks_in_grace(SVC_NET(rqstp))) {
0602         ret = nlm_lck_denied_grace_period;
0603         goto out;
0604     }
0605 
0606     mode = lock_to_openmode(&lock->fl);
0607     error = vfs_test_lock(file->f_file[mode], &lock->fl);
0608     if (error) {
0609         /* We can't currently deal with deferred test requests */
0610         if (error == FILE_LOCK_DEFERRED)
0611             WARN_ON_ONCE(1);
0612 
0613         ret = nlm_lck_denied_nolocks;
0614         goto out;
0615     }
0616 
0617     if (lock->fl.fl_type == F_UNLCK) {
0618         ret = nlm_granted;
0619         goto out;
0620     }
0621 
0622     dprintk("lockd: conflicting lock(ty=%d, %Ld-%Ld)\n",
0623         lock->fl.fl_type, (long long)lock->fl.fl_start,
0624         (long long)lock->fl.fl_end);
0625     conflock->caller = "somehost";  /* FIXME */
0626     conflock->len = strlen(conflock->caller);
0627     conflock->oh.len = 0;       /* don't return OH info */
0628     conflock->svid = lock->fl.fl_pid;
0629     conflock->fl.fl_type = lock->fl.fl_type;
0630     conflock->fl.fl_start = lock->fl.fl_start;
0631     conflock->fl.fl_end = lock->fl.fl_end;
0632     locks_release_private(&lock->fl);
0633 
0634     ret = nlm_lck_denied;
0635 out:
0636     return ret;
0637 }
0638 
0639 /*
0640  * Remove a lock.
0641  * This implies a CANCEL call: We send a GRANT_MSG, the client replies
0642  * with a GRANT_RES call which gets lost, and calls UNLOCK immediately
0643  * afterwards. In this case the block will still be there, and hence
0644  * must be removed.
0645  */
0646 __be32
0647 nlmsvc_unlock(struct net *net, struct nlm_file *file, struct nlm_lock *lock)
0648 {
0649     int error = 0;
0650 
0651     dprintk("lockd: nlmsvc_unlock(%s/%ld, pi=%d, %Ld-%Ld)\n",
0652                 nlmsvc_file_inode(file)->i_sb->s_id,
0653                 nlmsvc_file_inode(file)->i_ino,
0654                 lock->fl.fl_pid,
0655                 (long long)lock->fl.fl_start,
0656                 (long long)lock->fl.fl_end);
0657 
0658     /* First, cancel any lock that might be there */
0659     nlmsvc_cancel_blocked(net, file, lock);
0660 
0661     lock->fl.fl_type = F_UNLCK;
0662     if (file->f_file[O_RDONLY])
0663         error = vfs_lock_file(file->f_file[O_RDONLY], F_SETLK,
0664                     &lock->fl, NULL);
0665     if (file->f_file[O_WRONLY])
0666         error = vfs_lock_file(file->f_file[O_WRONLY], F_SETLK,
0667                     &lock->fl, NULL);
0668 
0669     return (error < 0)? nlm_lck_denied_nolocks : nlm_granted;
0670 }
0671 
0672 /*
0673  * Cancel a previously blocked request.
0674  *
0675  * A cancel request always overrides any grant that may currently
0676  * be in progress.
0677  * The calling procedure must check whether the file can be closed.
0678  */
0679 __be32
0680 nlmsvc_cancel_blocked(struct net *net, struct nlm_file *file, struct nlm_lock *lock)
0681 {
0682     struct nlm_block    *block;
0683     int status = 0;
0684     int mode;
0685 
0686     dprintk("lockd: nlmsvc_cancel(%s/%ld, pi=%d, %Ld-%Ld)\n",
0687                 nlmsvc_file_inode(file)->i_sb->s_id,
0688                 nlmsvc_file_inode(file)->i_ino,
0689                 lock->fl.fl_pid,
0690                 (long long)lock->fl.fl_start,
0691                 (long long)lock->fl.fl_end);
0692 
0693     if (locks_in_grace(net))
0694         return nlm_lck_denied_grace_period;
0695 
0696     mutex_lock(&file->f_mutex);
0697     block = nlmsvc_lookup_block(file, lock);
0698     mutex_unlock(&file->f_mutex);
0699     if (block != NULL) {
0700         mode = lock_to_openmode(&lock->fl);
0701         vfs_cancel_lock(block->b_file->f_file[mode],
0702                 &block->b_call->a_args.lock.fl);
0703         status = nlmsvc_unlink_block(block);
0704         nlmsvc_release_block(block);
0705     }
0706     return status ? nlm_lck_denied : nlm_granted;
0707 }
0708 
0709 /*
0710  * This is a callback from the filesystem for VFS file lock requests.
0711  * It will be used if lm_grant is defined and the filesystem can not
0712  * respond to the request immediately.
0713  * For SETLK or SETLKW request it will get the local posix lock.
0714  * In all cases it will move the block to the head of nlm_blocked q where
0715  * nlmsvc_retry_blocked() can send back a reply for SETLKW or revisit the
0716  * deferred rpc for GETLK and SETLK.
0717  */
0718 static void
0719 nlmsvc_update_deferred_block(struct nlm_block *block, int result)
0720 {
0721     block->b_flags |= B_GOT_CALLBACK;
0722     if (result == 0)
0723         block->b_granted = 1;
0724     else
0725         block->b_flags |= B_TIMED_OUT;
0726 }
0727 
0728 static int nlmsvc_grant_deferred(struct file_lock *fl, int result)
0729 {
0730     struct nlm_block *block;
0731     int rc = -ENOENT;
0732 
0733     spin_lock(&nlm_blocked_lock);
0734     list_for_each_entry(block, &nlm_blocked, b_list) {
0735         if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) {
0736             dprintk("lockd: nlmsvc_notify_blocked block %p flags %d\n",
0737                             block, block->b_flags);
0738             if (block->b_flags & B_QUEUED) {
0739                 if (block->b_flags & B_TIMED_OUT) {
0740                     rc = -ENOLCK;
0741                     break;
0742                 }
0743                 nlmsvc_update_deferred_block(block, result);
0744             } else if (result == 0)
0745                 block->b_granted = 1;
0746 
0747             nlmsvc_insert_block_locked(block, 0);
0748             svc_wake_up(block->b_daemon);
0749             rc = 0;
0750             break;
0751         }
0752     }
0753     spin_unlock(&nlm_blocked_lock);
0754     if (rc == -ENOENT)
0755         printk(KERN_WARNING "lockd: grant for unknown block\n");
0756     return rc;
0757 }
0758 
0759 /*
0760  * Unblock a blocked lock request. This is a callback invoked from the
0761  * VFS layer when a lock on which we blocked is removed.
0762  *
0763  * This function doesn't grant the blocked lock instantly, but rather moves
0764  * the block to the head of nlm_blocked where it can be picked up by lockd.
0765  */
0766 static void
0767 nlmsvc_notify_blocked(struct file_lock *fl)
0768 {
0769     struct nlm_block    *block;
0770 
0771     dprintk("lockd: VFS unblock notification for block %p\n", fl);
0772     spin_lock(&nlm_blocked_lock);
0773     list_for_each_entry(block, &nlm_blocked, b_list) {
0774         if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) {
0775             nlmsvc_insert_block_locked(block, 0);
0776             spin_unlock(&nlm_blocked_lock);
0777             svc_wake_up(block->b_daemon);
0778             return;
0779         }
0780     }
0781     spin_unlock(&nlm_blocked_lock);
0782     printk(KERN_WARNING "lockd: notification for unknown block!\n");
0783 }
0784 
0785 static fl_owner_t nlmsvc_get_owner(fl_owner_t owner)
0786 {
0787     return nlmsvc_get_lockowner(owner);
0788 }
0789 
0790 static void nlmsvc_put_owner(fl_owner_t owner)
0791 {
0792     nlmsvc_put_lockowner(owner);
0793 }
0794 
0795 const struct lock_manager_operations nlmsvc_lock_operations = {
0796     .lm_notify = nlmsvc_notify_blocked,
0797     .lm_grant = nlmsvc_grant_deferred,
0798     .lm_get_owner = nlmsvc_get_owner,
0799     .lm_put_owner = nlmsvc_put_owner,
0800 };
0801 
0802 /*
0803  * Try to claim a lock that was previously blocked.
0804  *
0805  * Note that we use both the RPC_GRANTED_MSG call _and_ an async
0806  * RPC thread when notifying the client. This seems like overkill...
0807  * Here's why:
0808  *  -   we don't want to use a synchronous RPC thread, otherwise
0809  *  we might find ourselves hanging on a dead portmapper.
0810  *  -   Some lockd implementations (e.g. HP) don't react to
0811  *  RPC_GRANTED calls; they seem to insist on RPC_GRANTED_MSG calls.
0812  */
0813 static void
0814 nlmsvc_grant_blocked(struct nlm_block *block)
0815 {
0816     struct nlm_file     *file = block->b_file;
0817     struct nlm_lock     *lock = &block->b_call->a_args.lock;
0818     int         mode;
0819     int         error;
0820     loff_t          fl_start, fl_end;
0821 
0822     dprintk("lockd: grant blocked lock %p\n", block);
0823 
0824     kref_get(&block->b_count);
0825 
0826     /* Unlink block request from list */
0827     nlmsvc_unlink_block(block);
0828 
0829     /* If b_granted is true this means we've been here before.
0830      * Just retry the grant callback, possibly refreshing the RPC
0831      * binding */
0832     if (block->b_granted) {
0833         nlm_rebind_host(block->b_host);
0834         goto callback;
0835     }
0836 
0837     /* Try the lock operation again */
0838     /* vfs_lock_file() can mangle fl_start and fl_end, but we need
0839      * them unchanged for the GRANT_MSG
0840      */
0841     lock->fl.fl_flags |= FL_SLEEP;
0842     fl_start = lock->fl.fl_start;
0843     fl_end = lock->fl.fl_end;
0844     mode = lock_to_openmode(&lock->fl);
0845     error = vfs_lock_file(file->f_file[mode], F_SETLK, &lock->fl, NULL);
0846     lock->fl.fl_flags &= ~FL_SLEEP;
0847     lock->fl.fl_start = fl_start;
0848     lock->fl.fl_end = fl_end;
0849 
0850     switch (error) {
0851     case 0:
0852         break;
0853     case FILE_LOCK_DEFERRED:
0854         dprintk("lockd: lock still blocked error %d\n", error);
0855         nlmsvc_insert_block(block, NLM_NEVER);
0856         nlmsvc_release_block(block);
0857         return;
0858     default:
0859         printk(KERN_WARNING "lockd: unexpected error %d in %s!\n",
0860                 -error, __func__);
0861         nlmsvc_insert_block(block, 10 * HZ);
0862         nlmsvc_release_block(block);
0863         return;
0864     }
0865 
0866 callback:
0867     /* Lock was granted by VFS. */
0868     dprintk("lockd: GRANTing blocked lock.\n");
0869     block->b_granted = 1;
0870 
0871     /* keep block on the list, but don't reattempt until the RPC
0872      * completes or the submission fails
0873      */
0874     nlmsvc_insert_block(block, NLM_NEVER);
0875 
0876     /* Call the client -- use a soft RPC task since nlmsvc_retry_blocked
0877      * will queue up a new one if this one times out
0878      */
0879     error = nlm_async_call(block->b_call, NLMPROC_GRANTED_MSG,
0880                 &nlmsvc_grant_ops);
0881 
0882     /* RPC submission failed, wait a bit and retry */
0883     if (error < 0)
0884         nlmsvc_insert_block(block, 10 * HZ);
0885 }
0886 
0887 /*
0888  * This is the callback from the RPC layer when the NLM_GRANTED_MSG
0889  * RPC call has succeeded or timed out.
0890  * Like all RPC callbacks, it is invoked by the rpciod process, so it
0891  * better not sleep. Therefore, we put the blocked lock on the nlm_blocked
0892  * chain once more in order to have it removed by lockd itself (which can
0893  * then sleep on the file semaphore without disrupting e.g. the nfs client).
0894  */
0895 static void nlmsvc_grant_callback(struct rpc_task *task, void *data)
0896 {
0897     struct nlm_rqst     *call = data;
0898     struct nlm_block    *block = call->a_block;
0899     unsigned long       timeout;
0900 
0901     dprintk("lockd: GRANT_MSG RPC callback\n");
0902 
0903     spin_lock(&nlm_blocked_lock);
0904     /* if the block is not on a list at this point then it has
0905      * been invalidated. Don't try to requeue it.
0906      *
0907      * FIXME: it's possible that the block is removed from the list
0908      * after this check but before the nlmsvc_insert_block. In that
0909      * case it will be added back. Perhaps we need better locking
0910      * for nlm_blocked?
0911      */
0912     if (list_empty(&block->b_list))
0913         goto out;
0914 
0915     /* Technically, we should down the file semaphore here. Since we
0916      * move the block towards the head of the queue only, no harm
0917      * can be done, though. */
0918     if (task->tk_status < 0) {
0919         /* RPC error: Re-insert for retransmission */
0920         timeout = 10 * HZ;
0921     } else {
0922         /* Call was successful, now wait for client callback */
0923         timeout = 60 * HZ;
0924     }
0925     nlmsvc_insert_block_locked(block, timeout);
0926     svc_wake_up(block->b_daemon);
0927 out:
0928     spin_unlock(&nlm_blocked_lock);
0929 }
0930 
0931 /*
0932  * FIXME: nlmsvc_release_block() grabs a mutex.  This is not allowed for an
0933  * .rpc_release rpc_call_op
0934  */
0935 static void nlmsvc_grant_release(void *data)
0936 {
0937     struct nlm_rqst     *call = data;
0938     nlmsvc_release_block(call->a_block);
0939 }
0940 
0941 static const struct rpc_call_ops nlmsvc_grant_ops = {
0942     .rpc_call_done = nlmsvc_grant_callback,
0943     .rpc_release = nlmsvc_grant_release,
0944 };
0945 
0946 /*
0947  * We received a GRANT_RES callback. Try to find the corresponding
0948  * block.
0949  */
0950 void
0951 nlmsvc_grant_reply(struct nlm_cookie *cookie, __be32 status)
0952 {
0953     struct nlm_block    *block;
0954 
0955     dprintk("grant_reply: looking for cookie %x, s=%d \n",
0956         *(unsigned int *)(cookie->data), status);
0957     if (!(block = nlmsvc_find_block(cookie)))
0958         return;
0959 
0960     if (status == nlm_lck_denied_grace_period) {
0961         /* Try again in a couple of seconds */
0962         nlmsvc_insert_block(block, 10 * HZ);
0963     } else {
0964         /*
0965          * Lock is now held by client, or has been rejected.
0966          * In both cases, the block should be removed.
0967          */
0968         nlmsvc_unlink_block(block);
0969     }
0970     nlmsvc_release_block(block);
0971 }
0972 
0973 /* Helper function to handle retry of a deferred block.
0974  * If it is a blocking lock, call grant_blocked.
0975  * For a non-blocking lock or test lock, revisit the request.
0976  */
0977 static void
0978 retry_deferred_block(struct nlm_block *block)
0979 {
0980     if (!(block->b_flags & B_GOT_CALLBACK))
0981         block->b_flags |= B_TIMED_OUT;
0982     nlmsvc_insert_block(block, NLM_TIMEOUT);
0983     dprintk("revisit block %p flags %d\n",  block, block->b_flags);
0984     if (block->b_deferred_req) {
0985         block->b_deferred_req->revisit(block->b_deferred_req, 0);
0986         block->b_deferred_req = NULL;
0987     }
0988 }
0989 
0990 /*
0991  * Retry all blocked locks that have been notified. This is where lockd
0992  * picks up locks that can be granted, or grant notifications that must
0993  * be retransmitted.
0994  */
0995 unsigned long
0996 nlmsvc_retry_blocked(void)
0997 {
0998     unsigned long   timeout = MAX_SCHEDULE_TIMEOUT;
0999     struct nlm_block *block;
1000 
1001     spin_lock(&nlm_blocked_lock);
1002     while (!list_empty(&nlm_blocked) && !kthread_should_stop()) {
1003         block = list_entry(nlm_blocked.next, struct nlm_block, b_list);
1004 
1005         if (block->b_when == NLM_NEVER)
1006             break;
1007         if (time_after(block->b_when, jiffies)) {
1008             timeout = block->b_when - jiffies;
1009             break;
1010         }
1011         spin_unlock(&nlm_blocked_lock);
1012 
1013         dprintk("nlmsvc_retry_blocked(%p, when=%ld)\n",
1014             block, block->b_when);
1015         if (block->b_flags & B_QUEUED) {
1016             dprintk("nlmsvc_retry_blocked delete block (%p, granted=%d, flags=%d)\n",
1017                 block, block->b_granted, block->b_flags);
1018             retry_deferred_block(block);
1019         } else
1020             nlmsvc_grant_blocked(block);
1021         spin_lock(&nlm_blocked_lock);
1022     }
1023     spin_unlock(&nlm_blocked_lock);
1024 
1025     return timeout;
1026 }