Back to home page

OSCL-LXR

 
 

    


0001 // SPDX-License-Identifier: GPL-2.0-only
0002 /*
0003  * net/sunrpc/cache.c
0004  *
0005  * Generic code for various authentication-related caches
0006  * used by sunrpc clients and servers.
0007  *
0008  * Copyright (C) 2002 Neil Brown <neilb@cse.unsw.edu.au>
0009  */
0010 
0011 #include <linux/types.h>
0012 #include <linux/fs.h>
0013 #include <linux/file.h>
0014 #include <linux/slab.h>
0015 #include <linux/signal.h>
0016 #include <linux/sched.h>
0017 #include <linux/kmod.h>
0018 #include <linux/list.h>
0019 #include <linux/module.h>
0020 #include <linux/ctype.h>
0021 #include <linux/string_helpers.h>
0022 #include <linux/uaccess.h>
0023 #include <linux/poll.h>
0024 #include <linux/seq_file.h>
0025 #include <linux/proc_fs.h>
0026 #include <linux/net.h>
0027 #include <linux/workqueue.h>
0028 #include <linux/mutex.h>
0029 #include <linux/pagemap.h>
0030 #include <asm/ioctls.h>
0031 #include <linux/sunrpc/types.h>
0032 #include <linux/sunrpc/cache.h>
0033 #include <linux/sunrpc/stats.h>
0034 #include <linux/sunrpc/rpc_pipe_fs.h>
0035 #include <trace/events/sunrpc.h>
0036 
0037 #include "netns.h"
0038 #include "fail.h"
0039 
0040 #define  RPCDBG_FACILITY RPCDBG_CACHE
0041 
0042 static bool cache_defer_req(struct cache_req *req, struct cache_head *item);
0043 static void cache_revisit_request(struct cache_head *item);
0044 
0045 static void cache_init(struct cache_head *h, struct cache_detail *detail)
0046 {
0047     time64_t now = seconds_since_boot();
0048     INIT_HLIST_NODE(&h->cache_list);
0049     h->flags = 0;
0050     kref_init(&h->ref);
0051     h->expiry_time = now + CACHE_NEW_EXPIRY;
0052     if (now <= detail->flush_time)
0053         /* ensure it isn't already expired */
0054         now = detail->flush_time + 1;
0055     h->last_refresh = now;
0056 }
0057 
0058 static void cache_fresh_unlocked(struct cache_head *head,
0059                 struct cache_detail *detail);
0060 
0061 static struct cache_head *sunrpc_cache_find_rcu(struct cache_detail *detail,
0062                         struct cache_head *key,
0063                         int hash)
0064 {
0065     struct hlist_head *head = &detail->hash_table[hash];
0066     struct cache_head *tmp;
0067 
0068     rcu_read_lock();
0069     hlist_for_each_entry_rcu(tmp, head, cache_list) {
0070         if (!detail->match(tmp, key))
0071             continue;
0072         if (test_bit(CACHE_VALID, &tmp->flags) &&
0073             cache_is_expired(detail, tmp))
0074             continue;
0075         tmp = cache_get_rcu(tmp);
0076         rcu_read_unlock();
0077         return tmp;
0078     }
0079     rcu_read_unlock();
0080     return NULL;
0081 }
0082 
0083 static void sunrpc_begin_cache_remove_entry(struct cache_head *ch,
0084                         struct cache_detail *cd)
0085 {
0086     /* Must be called under cd->hash_lock */
0087     hlist_del_init_rcu(&ch->cache_list);
0088     set_bit(CACHE_CLEANED, &ch->flags);
0089     cd->entries --;
0090 }
0091 
0092 static void sunrpc_end_cache_remove_entry(struct cache_head *ch,
0093                       struct cache_detail *cd)
0094 {
0095     cache_fresh_unlocked(ch, cd);
0096     cache_put(ch, cd);
0097 }
0098 
0099 static struct cache_head *sunrpc_cache_add_entry(struct cache_detail *detail,
0100                          struct cache_head *key,
0101                          int hash)
0102 {
0103     struct cache_head *new, *tmp, *freeme = NULL;
0104     struct hlist_head *head = &detail->hash_table[hash];
0105 
0106     new = detail->alloc();
0107     if (!new)
0108         return NULL;
0109     /* must fully initialise 'new', else
0110      * we might get lose if we need to
0111      * cache_put it soon.
0112      */
0113     cache_init(new, detail);
0114     detail->init(new, key);
0115 
0116     spin_lock(&detail->hash_lock);
0117 
0118     /* check if entry appeared while we slept */
0119     hlist_for_each_entry_rcu(tmp, head, cache_list,
0120                  lockdep_is_held(&detail->hash_lock)) {
0121         if (!detail->match(tmp, key))
0122             continue;
0123         if (test_bit(CACHE_VALID, &tmp->flags) &&
0124             cache_is_expired(detail, tmp)) {
0125             sunrpc_begin_cache_remove_entry(tmp, detail);
0126             trace_cache_entry_expired(detail, tmp);
0127             freeme = tmp;
0128             break;
0129         }
0130         cache_get(tmp);
0131         spin_unlock(&detail->hash_lock);
0132         cache_put(new, detail);
0133         return tmp;
0134     }
0135 
0136     hlist_add_head_rcu(&new->cache_list, head);
0137     detail->entries++;
0138     cache_get(new);
0139     spin_unlock(&detail->hash_lock);
0140 
0141     if (freeme)
0142         sunrpc_end_cache_remove_entry(freeme, detail);
0143     return new;
0144 }
0145 
0146 struct cache_head *sunrpc_cache_lookup_rcu(struct cache_detail *detail,
0147                        struct cache_head *key, int hash)
0148 {
0149     struct cache_head *ret;
0150 
0151     ret = sunrpc_cache_find_rcu(detail, key, hash);
0152     if (ret)
0153         return ret;
0154     /* Didn't find anything, insert an empty entry */
0155     return sunrpc_cache_add_entry(detail, key, hash);
0156 }
0157 EXPORT_SYMBOL_GPL(sunrpc_cache_lookup_rcu);
0158 
0159 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch);
0160 
0161 static void cache_fresh_locked(struct cache_head *head, time64_t expiry,
0162                    struct cache_detail *detail)
0163 {
0164     time64_t now = seconds_since_boot();
0165     if (now <= detail->flush_time)
0166         /* ensure it isn't immediately treated as expired */
0167         now = detail->flush_time + 1;
0168     head->expiry_time = expiry;
0169     head->last_refresh = now;
0170     smp_wmb(); /* paired with smp_rmb() in cache_is_valid() */
0171     set_bit(CACHE_VALID, &head->flags);
0172 }
0173 
0174 static void cache_fresh_unlocked(struct cache_head *head,
0175                  struct cache_detail *detail)
0176 {
0177     if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
0178         cache_revisit_request(head);
0179         cache_dequeue(detail, head);
0180     }
0181 }
0182 
0183 static void cache_make_negative(struct cache_detail *detail,
0184                 struct cache_head *h)
0185 {
0186     set_bit(CACHE_NEGATIVE, &h->flags);
0187     trace_cache_entry_make_negative(detail, h);
0188 }
0189 
0190 static void cache_entry_update(struct cache_detail *detail,
0191                    struct cache_head *h,
0192                    struct cache_head *new)
0193 {
0194     if (!test_bit(CACHE_NEGATIVE, &new->flags)) {
0195         detail->update(h, new);
0196         trace_cache_entry_update(detail, h);
0197     } else {
0198         cache_make_negative(detail, h);
0199     }
0200 }
0201 
0202 struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
0203                        struct cache_head *new, struct cache_head *old, int hash)
0204 {
0205     /* The 'old' entry is to be replaced by 'new'.
0206      * If 'old' is not VALID, we update it directly,
0207      * otherwise we need to replace it
0208      */
0209     struct cache_head *tmp;
0210 
0211     if (!test_bit(CACHE_VALID, &old->flags)) {
0212         spin_lock(&detail->hash_lock);
0213         if (!test_bit(CACHE_VALID, &old->flags)) {
0214             cache_entry_update(detail, old, new);
0215             cache_fresh_locked(old, new->expiry_time, detail);
0216             spin_unlock(&detail->hash_lock);
0217             cache_fresh_unlocked(old, detail);
0218             return old;
0219         }
0220         spin_unlock(&detail->hash_lock);
0221     }
0222     /* We need to insert a new entry */
0223     tmp = detail->alloc();
0224     if (!tmp) {
0225         cache_put(old, detail);
0226         return NULL;
0227     }
0228     cache_init(tmp, detail);
0229     detail->init(tmp, old);
0230 
0231     spin_lock(&detail->hash_lock);
0232     cache_entry_update(detail, tmp, new);
0233     hlist_add_head(&tmp->cache_list, &detail->hash_table[hash]);
0234     detail->entries++;
0235     cache_get(tmp);
0236     cache_fresh_locked(tmp, new->expiry_time, detail);
0237     cache_fresh_locked(old, 0, detail);
0238     spin_unlock(&detail->hash_lock);
0239     cache_fresh_unlocked(tmp, detail);
0240     cache_fresh_unlocked(old, detail);
0241     cache_put(old, detail);
0242     return tmp;
0243 }
0244 EXPORT_SYMBOL_GPL(sunrpc_cache_update);
0245 
0246 static inline int cache_is_valid(struct cache_head *h)
0247 {
0248     if (!test_bit(CACHE_VALID, &h->flags))
0249         return -EAGAIN;
0250     else {
0251         /* entry is valid */
0252         if (test_bit(CACHE_NEGATIVE, &h->flags))
0253             return -ENOENT;
0254         else {
0255             /*
0256              * In combination with write barrier in
0257              * sunrpc_cache_update, ensures that anyone
0258              * using the cache entry after this sees the
0259              * updated contents:
0260              */
0261             smp_rmb();
0262             return 0;
0263         }
0264     }
0265 }
0266 
0267 static int try_to_negate_entry(struct cache_detail *detail, struct cache_head *h)
0268 {
0269     int rv;
0270 
0271     spin_lock(&detail->hash_lock);
0272     rv = cache_is_valid(h);
0273     if (rv == -EAGAIN) {
0274         cache_make_negative(detail, h);
0275         cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY,
0276                    detail);
0277         rv = -ENOENT;
0278     }
0279     spin_unlock(&detail->hash_lock);
0280     cache_fresh_unlocked(h, detail);
0281     return rv;
0282 }
0283 
0284 /*
0285  * This is the generic cache management routine for all
0286  * the authentication caches.
0287  * It checks the currency of a cache item and will (later)
0288  * initiate an upcall to fill it if needed.
0289  *
0290  *
0291  * Returns 0 if the cache_head can be used, or cache_puts it and returns
0292  * -EAGAIN if upcall is pending and request has been queued
0293  * -ETIMEDOUT if upcall failed or request could not be queue or
0294  *           upcall completed but item is still invalid (implying that
0295  *           the cache item has been replaced with a newer one).
0296  * -ENOENT if cache entry was negative
0297  */
0298 int cache_check(struct cache_detail *detail,
0299             struct cache_head *h, struct cache_req *rqstp)
0300 {
0301     int rv;
0302     time64_t refresh_age, age;
0303 
0304     /* First decide return status as best we can */
0305     rv = cache_is_valid(h);
0306 
0307     /* now see if we want to start an upcall */
0308     refresh_age = (h->expiry_time - h->last_refresh);
0309     age = seconds_since_boot() - h->last_refresh;
0310 
0311     if (rqstp == NULL) {
0312         if (rv == -EAGAIN)
0313             rv = -ENOENT;
0314     } else if (rv == -EAGAIN ||
0315            (h->expiry_time != 0 && age > refresh_age/2)) {
0316         dprintk("RPC:       Want update, refage=%lld, age=%lld\n",
0317                 refresh_age, age);
0318         switch (detail->cache_upcall(detail, h)) {
0319         case -EINVAL:
0320             rv = try_to_negate_entry(detail, h);
0321             break;
0322         case -EAGAIN:
0323             cache_fresh_unlocked(h, detail);
0324             break;
0325         }
0326     }
0327 
0328     if (rv == -EAGAIN) {
0329         if (!cache_defer_req(rqstp, h)) {
0330             /*
0331              * Request was not deferred; handle it as best
0332              * we can ourselves:
0333              */
0334             rv = cache_is_valid(h);
0335             if (rv == -EAGAIN)
0336                 rv = -ETIMEDOUT;
0337         }
0338     }
0339     if (rv)
0340         cache_put(h, detail);
0341     return rv;
0342 }
0343 EXPORT_SYMBOL_GPL(cache_check);
0344 
0345 /*
0346  * caches need to be periodically cleaned.
0347  * For this we maintain a list of cache_detail and
0348  * a current pointer into that list and into the table
0349  * for that entry.
0350  *
0351  * Each time cache_clean is called it finds the next non-empty entry
0352  * in the current table and walks the list in that entry
0353  * looking for entries that can be removed.
0354  *
0355  * An entry gets removed if:
0356  * - The expiry is before current time
0357  * - The last_refresh time is before the flush_time for that cache
0358  *
0359  * later we might drop old entries with non-NEVER expiry if that table
0360  * is getting 'full' for some definition of 'full'
0361  *
0362  * The question of "how often to scan a table" is an interesting one
0363  * and is answered in part by the use of the "nextcheck" field in the
0364  * cache_detail.
0365  * When a scan of a table begins, the nextcheck field is set to a time
0366  * that is well into the future.
0367  * While scanning, if an expiry time is found that is earlier than the
0368  * current nextcheck time, nextcheck is set to that expiry time.
0369  * If the flush_time is ever set to a time earlier than the nextcheck
0370  * time, the nextcheck time is then set to that flush_time.
0371  *
0372  * A table is then only scanned if the current time is at least
0373  * the nextcheck time.
0374  *
0375  */
0376 
0377 static LIST_HEAD(cache_list);
0378 static DEFINE_SPINLOCK(cache_list_lock);
0379 static struct cache_detail *current_detail;
0380 static int current_index;
0381 
0382 static void do_cache_clean(struct work_struct *work);
0383 static struct delayed_work cache_cleaner;
0384 
0385 void sunrpc_init_cache_detail(struct cache_detail *cd)
0386 {
0387     spin_lock_init(&cd->hash_lock);
0388     INIT_LIST_HEAD(&cd->queue);
0389     spin_lock(&cache_list_lock);
0390     cd->nextcheck = 0;
0391     cd->entries = 0;
0392     atomic_set(&cd->writers, 0);
0393     cd->last_close = 0;
0394     cd->last_warn = -1;
0395     list_add(&cd->others, &cache_list);
0396     spin_unlock(&cache_list_lock);
0397 
0398     /* start the cleaning process */
0399     queue_delayed_work(system_power_efficient_wq, &cache_cleaner, 0);
0400 }
0401 EXPORT_SYMBOL_GPL(sunrpc_init_cache_detail);
0402 
0403 void sunrpc_destroy_cache_detail(struct cache_detail *cd)
0404 {
0405     cache_purge(cd);
0406     spin_lock(&cache_list_lock);
0407     spin_lock(&cd->hash_lock);
0408     if (current_detail == cd)
0409         current_detail = NULL;
0410     list_del_init(&cd->others);
0411     spin_unlock(&cd->hash_lock);
0412     spin_unlock(&cache_list_lock);
0413     if (list_empty(&cache_list)) {
0414         /* module must be being unloaded so its safe to kill the worker */
0415         cancel_delayed_work_sync(&cache_cleaner);
0416     }
0417 }
0418 EXPORT_SYMBOL_GPL(sunrpc_destroy_cache_detail);
0419 
0420 /* clean cache tries to find something to clean
0421  * and cleans it.
0422  * It returns 1 if it cleaned something,
0423  *            0 if it didn't find anything this time
0424  *           -1 if it fell off the end of the list.
0425  */
0426 static int cache_clean(void)
0427 {
0428     int rv = 0;
0429     struct list_head *next;
0430 
0431     spin_lock(&cache_list_lock);
0432 
0433     /* find a suitable table if we don't already have one */
0434     while (current_detail == NULL ||
0435         current_index >= current_detail->hash_size) {
0436         if (current_detail)
0437             next = current_detail->others.next;
0438         else
0439             next = cache_list.next;
0440         if (next == &cache_list) {
0441             current_detail = NULL;
0442             spin_unlock(&cache_list_lock);
0443             return -1;
0444         }
0445         current_detail = list_entry(next, struct cache_detail, others);
0446         if (current_detail->nextcheck > seconds_since_boot())
0447             current_index = current_detail->hash_size;
0448         else {
0449             current_index = 0;
0450             current_detail->nextcheck = seconds_since_boot()+30*60;
0451         }
0452     }
0453 
0454     /* find a non-empty bucket in the table */
0455     while (current_detail &&
0456            current_index < current_detail->hash_size &&
0457            hlist_empty(&current_detail->hash_table[current_index]))
0458         current_index++;
0459 
0460     /* find a cleanable entry in the bucket and clean it, or set to next bucket */
0461 
0462     if (current_detail && current_index < current_detail->hash_size) {
0463         struct cache_head *ch = NULL;
0464         struct cache_detail *d;
0465         struct hlist_head *head;
0466         struct hlist_node *tmp;
0467 
0468         spin_lock(&current_detail->hash_lock);
0469 
0470         /* Ok, now to clean this strand */
0471 
0472         head = &current_detail->hash_table[current_index];
0473         hlist_for_each_entry_safe(ch, tmp, head, cache_list) {
0474             if (current_detail->nextcheck > ch->expiry_time)
0475                 current_detail->nextcheck = ch->expiry_time+1;
0476             if (!cache_is_expired(current_detail, ch))
0477                 continue;
0478 
0479             sunrpc_begin_cache_remove_entry(ch, current_detail);
0480             trace_cache_entry_expired(current_detail, ch);
0481             rv = 1;
0482             break;
0483         }
0484 
0485         spin_unlock(&current_detail->hash_lock);
0486         d = current_detail;
0487         if (!ch)
0488             current_index ++;
0489         spin_unlock(&cache_list_lock);
0490         if (ch)
0491             sunrpc_end_cache_remove_entry(ch, d);
0492     } else
0493         spin_unlock(&cache_list_lock);
0494 
0495     return rv;
0496 }
0497 
0498 /*
0499  * We want to regularly clean the cache, so we need to schedule some work ...
0500  */
0501 static void do_cache_clean(struct work_struct *work)
0502 {
0503     int delay;
0504 
0505     if (list_empty(&cache_list))
0506         return;
0507 
0508     if (cache_clean() == -1)
0509         delay = round_jiffies_relative(30*HZ);
0510     else
0511         delay = 5;
0512 
0513     queue_delayed_work(system_power_efficient_wq, &cache_cleaner, delay);
0514 }
0515 
0516 
0517 /*
0518  * Clean all caches promptly.  This just calls cache_clean
0519  * repeatedly until we are sure that every cache has had a chance to
0520  * be fully cleaned
0521  */
0522 void cache_flush(void)
0523 {
0524     while (cache_clean() != -1)
0525         cond_resched();
0526     while (cache_clean() != -1)
0527         cond_resched();
0528 }
0529 EXPORT_SYMBOL_GPL(cache_flush);
0530 
0531 void cache_purge(struct cache_detail *detail)
0532 {
0533     struct cache_head *ch = NULL;
0534     struct hlist_head *head = NULL;
0535     int i = 0;
0536 
0537     spin_lock(&detail->hash_lock);
0538     if (!detail->entries) {
0539         spin_unlock(&detail->hash_lock);
0540         return;
0541     }
0542 
0543     dprintk("RPC: %d entries in %s cache\n", detail->entries, detail->name);
0544     for (i = 0; i < detail->hash_size; i++) {
0545         head = &detail->hash_table[i];
0546         while (!hlist_empty(head)) {
0547             ch = hlist_entry(head->first, struct cache_head,
0548                      cache_list);
0549             sunrpc_begin_cache_remove_entry(ch, detail);
0550             spin_unlock(&detail->hash_lock);
0551             sunrpc_end_cache_remove_entry(ch, detail);
0552             spin_lock(&detail->hash_lock);
0553         }
0554     }
0555     spin_unlock(&detail->hash_lock);
0556 }
0557 EXPORT_SYMBOL_GPL(cache_purge);
0558 
0559 
0560 /*
0561  * Deferral and Revisiting of Requests.
0562  *
0563  * If a cache lookup finds a pending entry, we
0564  * need to defer the request and revisit it later.
0565  * All deferred requests are stored in a hash table,
0566  * indexed by "struct cache_head *".
0567  * As it may be wasteful to store a whole request
0568  * structure, we allow the request to provide a
0569  * deferred form, which must contain a
0570  * 'struct cache_deferred_req'
0571  * This cache_deferred_req contains a method to allow
0572  * it to be revisited when cache info is available
0573  */
0574 
0575 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
0576 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
0577 
0578 #define DFR_MAX 300 /* ??? */
0579 
0580 static DEFINE_SPINLOCK(cache_defer_lock);
0581 static LIST_HEAD(cache_defer_list);
0582 static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
0583 static int cache_defer_cnt;
0584 
0585 static void __unhash_deferred_req(struct cache_deferred_req *dreq)
0586 {
0587     hlist_del_init(&dreq->hash);
0588     if (!list_empty(&dreq->recent)) {
0589         list_del_init(&dreq->recent);
0590         cache_defer_cnt--;
0591     }
0592 }
0593 
0594 static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
0595 {
0596     int hash = DFR_HASH(item);
0597 
0598     INIT_LIST_HEAD(&dreq->recent);
0599     hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
0600 }
0601 
0602 static void setup_deferral(struct cache_deferred_req *dreq,
0603                struct cache_head *item,
0604                int count_me)
0605 {
0606 
0607     dreq->item = item;
0608 
0609     spin_lock(&cache_defer_lock);
0610 
0611     __hash_deferred_req(dreq, item);
0612 
0613     if (count_me) {
0614         cache_defer_cnt++;
0615         list_add(&dreq->recent, &cache_defer_list);
0616     }
0617 
0618     spin_unlock(&cache_defer_lock);
0619 
0620 }
0621 
0622 struct thread_deferred_req {
0623     struct cache_deferred_req handle;
0624     struct completion completion;
0625 };
0626 
0627 static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
0628 {
0629     struct thread_deferred_req *dr =
0630         container_of(dreq, struct thread_deferred_req, handle);
0631     complete(&dr->completion);
0632 }
0633 
0634 static void cache_wait_req(struct cache_req *req, struct cache_head *item)
0635 {
0636     struct thread_deferred_req sleeper;
0637     struct cache_deferred_req *dreq = &sleeper.handle;
0638 
0639     sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
0640     dreq->revisit = cache_restart_thread;
0641 
0642     setup_deferral(dreq, item, 0);
0643 
0644     if (!test_bit(CACHE_PENDING, &item->flags) ||
0645         wait_for_completion_interruptible_timeout(
0646             &sleeper.completion, req->thread_wait) <= 0) {
0647         /* The completion wasn't completed, so we need
0648          * to clean up
0649          */
0650         spin_lock(&cache_defer_lock);
0651         if (!hlist_unhashed(&sleeper.handle.hash)) {
0652             __unhash_deferred_req(&sleeper.handle);
0653             spin_unlock(&cache_defer_lock);
0654         } else {
0655             /* cache_revisit_request already removed
0656              * this from the hash table, but hasn't
0657              * called ->revisit yet.  It will very soon
0658              * and we need to wait for it.
0659              */
0660             spin_unlock(&cache_defer_lock);
0661             wait_for_completion(&sleeper.completion);
0662         }
0663     }
0664 }
0665 
0666 static void cache_limit_defers(void)
0667 {
0668     /* Make sure we haven't exceed the limit of allowed deferred
0669      * requests.
0670      */
0671     struct cache_deferred_req *discard = NULL;
0672 
0673     if (cache_defer_cnt <= DFR_MAX)
0674         return;
0675 
0676     spin_lock(&cache_defer_lock);
0677 
0678     /* Consider removing either the first or the last */
0679     if (cache_defer_cnt > DFR_MAX) {
0680         if (prandom_u32() & 1)
0681             discard = list_entry(cache_defer_list.next,
0682                          struct cache_deferred_req, recent);
0683         else
0684             discard = list_entry(cache_defer_list.prev,
0685                          struct cache_deferred_req, recent);
0686         __unhash_deferred_req(discard);
0687     }
0688     spin_unlock(&cache_defer_lock);
0689     if (discard)
0690         discard->revisit(discard, 1);
0691 }
0692 
0693 #if IS_ENABLED(CONFIG_FAIL_SUNRPC)
0694 static inline bool cache_defer_immediately(void)
0695 {
0696     return !fail_sunrpc.ignore_cache_wait &&
0697         should_fail(&fail_sunrpc.attr, 1);
0698 }
0699 #else
0700 static inline bool cache_defer_immediately(void)
0701 {
0702     return false;
0703 }
0704 #endif
0705 
0706 /* Return true if and only if a deferred request is queued. */
0707 static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
0708 {
0709     struct cache_deferred_req *dreq;
0710 
0711     if (!cache_defer_immediately()) {
0712         cache_wait_req(req, item);
0713         if (!test_bit(CACHE_PENDING, &item->flags))
0714             return false;
0715     }
0716 
0717     dreq = req->defer(req);
0718     if (dreq == NULL)
0719         return false;
0720     setup_deferral(dreq, item, 1);
0721     if (!test_bit(CACHE_PENDING, &item->flags))
0722         /* Bit could have been cleared before we managed to
0723          * set up the deferral, so need to revisit just in case
0724          */
0725         cache_revisit_request(item);
0726 
0727     cache_limit_defers();
0728     return true;
0729 }
0730 
0731 static void cache_revisit_request(struct cache_head *item)
0732 {
0733     struct cache_deferred_req *dreq;
0734     struct list_head pending;
0735     struct hlist_node *tmp;
0736     int hash = DFR_HASH(item);
0737 
0738     INIT_LIST_HEAD(&pending);
0739     spin_lock(&cache_defer_lock);
0740 
0741     hlist_for_each_entry_safe(dreq, tmp, &cache_defer_hash[hash], hash)
0742         if (dreq->item == item) {
0743             __unhash_deferred_req(dreq);
0744             list_add(&dreq->recent, &pending);
0745         }
0746 
0747     spin_unlock(&cache_defer_lock);
0748 
0749     while (!list_empty(&pending)) {
0750         dreq = list_entry(pending.next, struct cache_deferred_req, recent);
0751         list_del_init(&dreq->recent);
0752         dreq->revisit(dreq, 0);
0753     }
0754 }
0755 
0756 void cache_clean_deferred(void *owner)
0757 {
0758     struct cache_deferred_req *dreq, *tmp;
0759     struct list_head pending;
0760 
0761 
0762     INIT_LIST_HEAD(&pending);
0763     spin_lock(&cache_defer_lock);
0764 
0765     list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
0766         if (dreq->owner == owner) {
0767             __unhash_deferred_req(dreq);
0768             list_add(&dreq->recent, &pending);
0769         }
0770     }
0771     spin_unlock(&cache_defer_lock);
0772 
0773     while (!list_empty(&pending)) {
0774         dreq = list_entry(pending.next, struct cache_deferred_req, recent);
0775         list_del_init(&dreq->recent);
0776         dreq->revisit(dreq, 1);
0777     }
0778 }
0779 
0780 /*
0781  * communicate with user-space
0782  *
0783  * We have a magic /proc file - /proc/net/rpc/<cachename>/channel.
0784  * On read, you get a full request, or block.
0785  * On write, an update request is processed.
0786  * Poll works if anything to read, and always allows write.
0787  *
0788  * Implemented by linked list of requests.  Each open file has
0789  * a ->private that also exists in this list.  New requests are added
0790  * to the end and may wakeup and preceding readers.
0791  * New readers are added to the head.  If, on read, an item is found with
0792  * CACHE_UPCALLING clear, we free it from the list.
0793  *
0794  */
0795 
0796 static DEFINE_SPINLOCK(queue_lock);
0797 
0798 struct cache_queue {
0799     struct list_head    list;
0800     int         reader; /* if 0, then request */
0801 };
0802 struct cache_request {
0803     struct cache_queue  q;
0804     struct cache_head   *item;
0805     char            * buf;
0806     int         len;
0807     int         readers;
0808 };
0809 struct cache_reader {
0810     struct cache_queue  q;
0811     int         offset; /* if non-0, we have a refcnt on next request */
0812 };
0813 
0814 static int cache_request(struct cache_detail *detail,
0815                    struct cache_request *crq)
0816 {
0817     char *bp = crq->buf;
0818     int len = PAGE_SIZE;
0819 
0820     detail->cache_request(detail, crq->item, &bp, &len);
0821     if (len < 0)
0822         return -E2BIG;
0823     return PAGE_SIZE - len;
0824 }
0825 
0826 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
0827               loff_t *ppos, struct cache_detail *cd)
0828 {
0829     struct cache_reader *rp = filp->private_data;
0830     struct cache_request *rq;
0831     struct inode *inode = file_inode(filp);
0832     int err;
0833 
0834     if (count == 0)
0835         return 0;
0836 
0837     inode_lock(inode); /* protect against multiple concurrent
0838                   * readers on this file */
0839  again:
0840     spin_lock(&queue_lock);
0841     /* need to find next request */
0842     while (rp->q.list.next != &cd->queue &&
0843            list_entry(rp->q.list.next, struct cache_queue, list)
0844            ->reader) {
0845         struct list_head *next = rp->q.list.next;
0846         list_move(&rp->q.list, next);
0847     }
0848     if (rp->q.list.next == &cd->queue) {
0849         spin_unlock(&queue_lock);
0850         inode_unlock(inode);
0851         WARN_ON_ONCE(rp->offset);
0852         return 0;
0853     }
0854     rq = container_of(rp->q.list.next, struct cache_request, q.list);
0855     WARN_ON_ONCE(rq->q.reader);
0856     if (rp->offset == 0)
0857         rq->readers++;
0858     spin_unlock(&queue_lock);
0859 
0860     if (rq->len == 0) {
0861         err = cache_request(cd, rq);
0862         if (err < 0)
0863             goto out;
0864         rq->len = err;
0865     }
0866 
0867     if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
0868         err = -EAGAIN;
0869         spin_lock(&queue_lock);
0870         list_move(&rp->q.list, &rq->q.list);
0871         spin_unlock(&queue_lock);
0872     } else {
0873         if (rp->offset + count > rq->len)
0874             count = rq->len - rp->offset;
0875         err = -EFAULT;
0876         if (copy_to_user(buf, rq->buf + rp->offset, count))
0877             goto out;
0878         rp->offset += count;
0879         if (rp->offset >= rq->len) {
0880             rp->offset = 0;
0881             spin_lock(&queue_lock);
0882             list_move(&rp->q.list, &rq->q.list);
0883             spin_unlock(&queue_lock);
0884         }
0885         err = 0;
0886     }
0887  out:
0888     if (rp->offset == 0) {
0889         /* need to release rq */
0890         spin_lock(&queue_lock);
0891         rq->readers--;
0892         if (rq->readers == 0 &&
0893             !test_bit(CACHE_PENDING, &rq->item->flags)) {
0894             list_del(&rq->q.list);
0895             spin_unlock(&queue_lock);
0896             cache_put(rq->item, cd);
0897             kfree(rq->buf);
0898             kfree(rq);
0899         } else
0900             spin_unlock(&queue_lock);
0901     }
0902     if (err == -EAGAIN)
0903         goto again;
0904     inode_unlock(inode);
0905     return err ? err :  count;
0906 }
0907 
0908 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
0909                  size_t count, struct cache_detail *cd)
0910 {
0911     ssize_t ret;
0912 
0913     if (count == 0)
0914         return -EINVAL;
0915     if (copy_from_user(kaddr, buf, count))
0916         return -EFAULT;
0917     kaddr[count] = '\0';
0918     ret = cd->cache_parse(cd, kaddr, count);
0919     if (!ret)
0920         ret = count;
0921     return ret;
0922 }
0923 
0924 static ssize_t cache_downcall(struct address_space *mapping,
0925                   const char __user *buf,
0926                   size_t count, struct cache_detail *cd)
0927 {
0928     char *write_buf;
0929     ssize_t ret = -ENOMEM;
0930 
0931     if (count >= 32768) { /* 32k is max userland buffer, lets check anyway */
0932         ret = -EINVAL;
0933         goto out;
0934     }
0935 
0936     write_buf = kvmalloc(count + 1, GFP_KERNEL);
0937     if (!write_buf)
0938         goto out;
0939 
0940     ret = cache_do_downcall(write_buf, buf, count, cd);
0941     kvfree(write_buf);
0942 out:
0943     return ret;
0944 }
0945 
0946 static ssize_t cache_write(struct file *filp, const char __user *buf,
0947                size_t count, loff_t *ppos,
0948                struct cache_detail *cd)
0949 {
0950     struct address_space *mapping = filp->f_mapping;
0951     struct inode *inode = file_inode(filp);
0952     ssize_t ret = -EINVAL;
0953 
0954     if (!cd->cache_parse)
0955         goto out;
0956 
0957     inode_lock(inode);
0958     ret = cache_downcall(mapping, buf, count, cd);
0959     inode_unlock(inode);
0960 out:
0961     return ret;
0962 }
0963 
0964 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
0965 
0966 static __poll_t cache_poll(struct file *filp, poll_table *wait,
0967                    struct cache_detail *cd)
0968 {
0969     __poll_t mask;
0970     struct cache_reader *rp = filp->private_data;
0971     struct cache_queue *cq;
0972 
0973     poll_wait(filp, &queue_wait, wait);
0974 
0975     /* alway allow write */
0976     mask = EPOLLOUT | EPOLLWRNORM;
0977 
0978     if (!rp)
0979         return mask;
0980 
0981     spin_lock(&queue_lock);
0982 
0983     for (cq= &rp->q; &cq->list != &cd->queue;
0984          cq = list_entry(cq->list.next, struct cache_queue, list))
0985         if (!cq->reader) {
0986             mask |= EPOLLIN | EPOLLRDNORM;
0987             break;
0988         }
0989     spin_unlock(&queue_lock);
0990     return mask;
0991 }
0992 
0993 static int cache_ioctl(struct inode *ino, struct file *filp,
0994                unsigned int cmd, unsigned long arg,
0995                struct cache_detail *cd)
0996 {
0997     int len = 0;
0998     struct cache_reader *rp = filp->private_data;
0999     struct cache_queue *cq;
1000 
1001     if (cmd != FIONREAD || !rp)
1002         return -EINVAL;
1003 
1004     spin_lock(&queue_lock);
1005 
1006     /* only find the length remaining in current request,
1007      * or the length of the next request
1008      */
1009     for (cq= &rp->q; &cq->list != &cd->queue;
1010          cq = list_entry(cq->list.next, struct cache_queue, list))
1011         if (!cq->reader) {
1012             struct cache_request *cr =
1013                 container_of(cq, struct cache_request, q);
1014             len = cr->len - rp->offset;
1015             break;
1016         }
1017     spin_unlock(&queue_lock);
1018 
1019     return put_user(len, (int __user *)arg);
1020 }
1021 
1022 static int cache_open(struct inode *inode, struct file *filp,
1023               struct cache_detail *cd)
1024 {
1025     struct cache_reader *rp = NULL;
1026 
1027     if (!cd || !try_module_get(cd->owner))
1028         return -EACCES;
1029     nonseekable_open(inode, filp);
1030     if (filp->f_mode & FMODE_READ) {
1031         rp = kmalloc(sizeof(*rp), GFP_KERNEL);
1032         if (!rp) {
1033             module_put(cd->owner);
1034             return -ENOMEM;
1035         }
1036         rp->offset = 0;
1037         rp->q.reader = 1;
1038 
1039         spin_lock(&queue_lock);
1040         list_add(&rp->q.list, &cd->queue);
1041         spin_unlock(&queue_lock);
1042     }
1043     if (filp->f_mode & FMODE_WRITE)
1044         atomic_inc(&cd->writers);
1045     filp->private_data = rp;
1046     return 0;
1047 }
1048 
1049 static int cache_release(struct inode *inode, struct file *filp,
1050              struct cache_detail *cd)
1051 {
1052     struct cache_reader *rp = filp->private_data;
1053 
1054     if (rp) {
1055         spin_lock(&queue_lock);
1056         if (rp->offset) {
1057             struct cache_queue *cq;
1058             for (cq= &rp->q; &cq->list != &cd->queue;
1059                  cq = list_entry(cq->list.next, struct cache_queue, list))
1060                 if (!cq->reader) {
1061                     container_of(cq, struct cache_request, q)
1062                         ->readers--;
1063                     break;
1064                 }
1065             rp->offset = 0;
1066         }
1067         list_del(&rp->q.list);
1068         spin_unlock(&queue_lock);
1069 
1070         filp->private_data = NULL;
1071         kfree(rp);
1072 
1073     }
1074     if (filp->f_mode & FMODE_WRITE) {
1075         atomic_dec(&cd->writers);
1076         cd->last_close = seconds_since_boot();
1077     }
1078     module_put(cd->owner);
1079     return 0;
1080 }
1081 
1082 
1083 
1084 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1085 {
1086     struct cache_queue *cq, *tmp;
1087     struct cache_request *cr;
1088     struct list_head dequeued;
1089 
1090     INIT_LIST_HEAD(&dequeued);
1091     spin_lock(&queue_lock);
1092     list_for_each_entry_safe(cq, tmp, &detail->queue, list)
1093         if (!cq->reader) {
1094             cr = container_of(cq, struct cache_request, q);
1095             if (cr->item != ch)
1096                 continue;
1097             if (test_bit(CACHE_PENDING, &ch->flags))
1098                 /* Lost a race and it is pending again */
1099                 break;
1100             if (cr->readers != 0)
1101                 continue;
1102             list_move(&cr->q.list, &dequeued);
1103         }
1104     spin_unlock(&queue_lock);
1105     while (!list_empty(&dequeued)) {
1106         cr = list_entry(dequeued.next, struct cache_request, q.list);
1107         list_del(&cr->q.list);
1108         cache_put(cr->item, detail);
1109         kfree(cr->buf);
1110         kfree(cr);
1111     }
1112 }
1113 
1114 /*
1115  * Support routines for text-based upcalls.
1116  * Fields are separated by spaces.
1117  * Fields are either mangled to quote space tab newline slosh with slosh
1118  * or a hexified with a leading \x
1119  * Record is terminated with newline.
1120  *
1121  */
1122 
1123 void qword_add(char **bpp, int *lp, char *str)
1124 {
1125     char *bp = *bpp;
1126     int len = *lp;
1127     int ret;
1128 
1129     if (len < 0) return;
1130 
1131     ret = string_escape_str(str, bp, len, ESCAPE_OCTAL, "\\ \n\t");
1132     if (ret >= len) {
1133         bp += len;
1134         len = -1;
1135     } else {
1136         bp += ret;
1137         len -= ret;
1138         *bp++ = ' ';
1139         len--;
1140     }
1141     *bpp = bp;
1142     *lp = len;
1143 }
1144 EXPORT_SYMBOL_GPL(qword_add);
1145 
1146 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1147 {
1148     char *bp = *bpp;
1149     int len = *lp;
1150 
1151     if (len < 0) return;
1152 
1153     if (len > 2) {
1154         *bp++ = '\\';
1155         *bp++ = 'x';
1156         len -= 2;
1157         while (blen && len >= 2) {
1158             bp = hex_byte_pack(bp, *buf++);
1159             len -= 2;
1160             blen--;
1161         }
1162     }
1163     if (blen || len<1) len = -1;
1164     else {
1165         *bp++ = ' ';
1166         len--;
1167     }
1168     *bpp = bp;
1169     *lp = len;
1170 }
1171 EXPORT_SYMBOL_GPL(qword_addhex);
1172 
1173 static void warn_no_listener(struct cache_detail *detail)
1174 {
1175     if (detail->last_warn != detail->last_close) {
1176         detail->last_warn = detail->last_close;
1177         if (detail->warn_no_listener)
1178             detail->warn_no_listener(detail, detail->last_close != 0);
1179     }
1180 }
1181 
1182 static bool cache_listeners_exist(struct cache_detail *detail)
1183 {
1184     if (atomic_read(&detail->writers))
1185         return true;
1186     if (detail->last_close == 0)
1187         /* This cache was never opened */
1188         return false;
1189     if (detail->last_close < seconds_since_boot() - 30)
1190         /*
1191          * We allow for the possibility that someone might
1192          * restart a userspace daemon without restarting the
1193          * server; but after 30 seconds, we give up.
1194          */
1195          return false;
1196     return true;
1197 }
1198 
1199 /*
1200  * register an upcall request to user-space and queue it up for read() by the
1201  * upcall daemon.
1202  *
1203  * Each request is at most one page long.
1204  */
1205 static int cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1206 {
1207     char *buf;
1208     struct cache_request *crq;
1209     int ret = 0;
1210 
1211     if (test_bit(CACHE_CLEANED, &h->flags))
1212         /* Too late to make an upcall */
1213         return -EAGAIN;
1214 
1215     buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1216     if (!buf)
1217         return -EAGAIN;
1218 
1219     crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1220     if (!crq) {
1221         kfree(buf);
1222         return -EAGAIN;
1223     }
1224 
1225     crq->q.reader = 0;
1226     crq->buf = buf;
1227     crq->len = 0;
1228     crq->readers = 0;
1229     spin_lock(&queue_lock);
1230     if (test_bit(CACHE_PENDING, &h->flags)) {
1231         crq->item = cache_get(h);
1232         list_add_tail(&crq->q.list, &detail->queue);
1233         trace_cache_entry_upcall(detail, h);
1234     } else
1235         /* Lost a race, no longer PENDING, so don't enqueue */
1236         ret = -EAGAIN;
1237     spin_unlock(&queue_lock);
1238     wake_up(&queue_wait);
1239     if (ret == -EAGAIN) {
1240         kfree(buf);
1241         kfree(crq);
1242     }
1243     return ret;
1244 }
1245 
1246 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1247 {
1248     if (test_and_set_bit(CACHE_PENDING, &h->flags))
1249         return 0;
1250     return cache_pipe_upcall(detail, h);
1251 }
1252 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1253 
1254 int sunrpc_cache_pipe_upcall_timeout(struct cache_detail *detail,
1255                      struct cache_head *h)
1256 {
1257     if (!cache_listeners_exist(detail)) {
1258         warn_no_listener(detail);
1259         trace_cache_entry_no_listener(detail, h);
1260         return -EINVAL;
1261     }
1262     return sunrpc_cache_pipe_upcall(detail, h);
1263 }
1264 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall_timeout);
1265 
1266 /*
1267  * parse a message from user-space and pass it
1268  * to an appropriate cache
1269  * Messages are, like requests, separated into fields by
1270  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1271  *
1272  * Message is
1273  *   reply cachename expiry key ... content....
1274  *
1275  * key and content are both parsed by cache
1276  */
1277 
1278 int qword_get(char **bpp, char *dest, int bufsize)
1279 {
1280     /* return bytes copied, or -1 on error */
1281     char *bp = *bpp;
1282     int len = 0;
1283 
1284     while (*bp == ' ') bp++;
1285 
1286     if (bp[0] == '\\' && bp[1] == 'x') {
1287         /* HEX STRING */
1288         bp += 2;
1289         while (len < bufsize - 1) {
1290             int h, l;
1291 
1292             h = hex_to_bin(bp[0]);
1293             if (h < 0)
1294                 break;
1295 
1296             l = hex_to_bin(bp[1]);
1297             if (l < 0)
1298                 break;
1299 
1300             *dest++ = (h << 4) | l;
1301             bp += 2;
1302             len++;
1303         }
1304     } else {
1305         /* text with \nnn octal quoting */
1306         while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1307             if (*bp == '\\' &&
1308                 isodigit(bp[1]) && (bp[1] <= '3') &&
1309                 isodigit(bp[2]) &&
1310                 isodigit(bp[3])) {
1311                 int byte = (*++bp -'0');
1312                 bp++;
1313                 byte = (byte << 3) | (*bp++ - '0');
1314                 byte = (byte << 3) | (*bp++ - '0');
1315                 *dest++ = byte;
1316                 len++;
1317             } else {
1318                 *dest++ = *bp++;
1319                 len++;
1320             }
1321         }
1322     }
1323 
1324     if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1325         return -1;
1326     while (*bp == ' ') bp++;
1327     *bpp = bp;
1328     *dest = '\0';
1329     return len;
1330 }
1331 EXPORT_SYMBOL_GPL(qword_get);
1332 
1333 
1334 /*
1335  * support /proc/net/rpc/$CACHENAME/content
1336  * as a seqfile.
1337  * We call ->cache_show passing NULL for the item to
1338  * get a header, then pass each real item in the cache
1339  */
1340 
1341 static void *__cache_seq_start(struct seq_file *m, loff_t *pos)
1342 {
1343     loff_t n = *pos;
1344     unsigned int hash, entry;
1345     struct cache_head *ch;
1346     struct cache_detail *cd = m->private;
1347 
1348     if (!n--)
1349         return SEQ_START_TOKEN;
1350     hash = n >> 32;
1351     entry = n & ((1LL<<32) - 1);
1352 
1353     hlist_for_each_entry_rcu(ch, &cd->hash_table[hash], cache_list)
1354         if (!entry--)
1355             return ch;
1356     n &= ~((1LL<<32) - 1);
1357     do {
1358         hash++;
1359         n += 1LL<<32;
1360     } while(hash < cd->hash_size &&
1361         hlist_empty(&cd->hash_table[hash]));
1362     if (hash >= cd->hash_size)
1363         return NULL;
1364     *pos = n+1;
1365     return hlist_entry_safe(rcu_dereference_raw(
1366                 hlist_first_rcu(&cd->hash_table[hash])),
1367                 struct cache_head, cache_list);
1368 }
1369 
1370 static void *cache_seq_next(struct seq_file *m, void *p, loff_t *pos)
1371 {
1372     struct cache_head *ch = p;
1373     int hash = (*pos >> 32);
1374     struct cache_detail *cd = m->private;
1375 
1376     if (p == SEQ_START_TOKEN)
1377         hash = 0;
1378     else if (ch->cache_list.next == NULL) {
1379         hash++;
1380         *pos += 1LL<<32;
1381     } else {
1382         ++*pos;
1383         return hlist_entry_safe(rcu_dereference_raw(
1384                     hlist_next_rcu(&ch->cache_list)),
1385                     struct cache_head, cache_list);
1386     }
1387     *pos &= ~((1LL<<32) - 1);
1388     while (hash < cd->hash_size &&
1389            hlist_empty(&cd->hash_table[hash])) {
1390         hash++;
1391         *pos += 1LL<<32;
1392     }
1393     if (hash >= cd->hash_size)
1394         return NULL;
1395     ++*pos;
1396     return hlist_entry_safe(rcu_dereference_raw(
1397                 hlist_first_rcu(&cd->hash_table[hash])),
1398                 struct cache_head, cache_list);
1399 }
1400 
1401 void *cache_seq_start_rcu(struct seq_file *m, loff_t *pos)
1402     __acquires(RCU)
1403 {
1404     rcu_read_lock();
1405     return __cache_seq_start(m, pos);
1406 }
1407 EXPORT_SYMBOL_GPL(cache_seq_start_rcu);
1408 
1409 void *cache_seq_next_rcu(struct seq_file *file, void *p, loff_t *pos)
1410 {
1411     return cache_seq_next(file, p, pos);
1412 }
1413 EXPORT_SYMBOL_GPL(cache_seq_next_rcu);
1414 
1415 void cache_seq_stop_rcu(struct seq_file *m, void *p)
1416     __releases(RCU)
1417 {
1418     rcu_read_unlock();
1419 }
1420 EXPORT_SYMBOL_GPL(cache_seq_stop_rcu);
1421 
1422 static int c_show(struct seq_file *m, void *p)
1423 {
1424     struct cache_head *cp = p;
1425     struct cache_detail *cd = m->private;
1426 
1427     if (p == SEQ_START_TOKEN)
1428         return cd->cache_show(m, cd, NULL);
1429 
1430     ifdebug(CACHE)
1431         seq_printf(m, "# expiry=%lld refcnt=%d flags=%lx\n",
1432                convert_to_wallclock(cp->expiry_time),
1433                kref_read(&cp->ref), cp->flags);
1434     cache_get(cp);
1435     if (cache_check(cd, cp, NULL))
1436         /* cache_check does a cache_put on failure */
1437         seq_puts(m, "# ");
1438     else {
1439         if (cache_is_expired(cd, cp))
1440             seq_puts(m, "# ");
1441         cache_put(cp, cd);
1442     }
1443 
1444     return cd->cache_show(m, cd, cp);
1445 }
1446 
1447 static const struct seq_operations cache_content_op = {
1448     .start  = cache_seq_start_rcu,
1449     .next   = cache_seq_next_rcu,
1450     .stop   = cache_seq_stop_rcu,
1451     .show   = c_show,
1452 };
1453 
1454 static int content_open(struct inode *inode, struct file *file,
1455             struct cache_detail *cd)
1456 {
1457     struct seq_file *seq;
1458     int err;
1459 
1460     if (!cd || !try_module_get(cd->owner))
1461         return -EACCES;
1462 
1463     err = seq_open(file, &cache_content_op);
1464     if (err) {
1465         module_put(cd->owner);
1466         return err;
1467     }
1468 
1469     seq = file->private_data;
1470     seq->private = cd;
1471     return 0;
1472 }
1473 
1474 static int content_release(struct inode *inode, struct file *file,
1475         struct cache_detail *cd)
1476 {
1477     int ret = seq_release(inode, file);
1478     module_put(cd->owner);
1479     return ret;
1480 }
1481 
1482 static int open_flush(struct inode *inode, struct file *file,
1483             struct cache_detail *cd)
1484 {
1485     if (!cd || !try_module_get(cd->owner))
1486         return -EACCES;
1487     return nonseekable_open(inode, file);
1488 }
1489 
1490 static int release_flush(struct inode *inode, struct file *file,
1491             struct cache_detail *cd)
1492 {
1493     module_put(cd->owner);
1494     return 0;
1495 }
1496 
1497 static ssize_t read_flush(struct file *file, char __user *buf,
1498               size_t count, loff_t *ppos,
1499               struct cache_detail *cd)
1500 {
1501     char tbuf[22];
1502     size_t len;
1503 
1504     len = snprintf(tbuf, sizeof(tbuf), "%llu\n",
1505             convert_to_wallclock(cd->flush_time));
1506     return simple_read_from_buffer(buf, count, ppos, tbuf, len);
1507 }
1508 
1509 static ssize_t write_flush(struct file *file, const char __user *buf,
1510                size_t count, loff_t *ppos,
1511                struct cache_detail *cd)
1512 {
1513     char tbuf[20];
1514     char *ep;
1515     time64_t now;
1516 
1517     if (*ppos || count > sizeof(tbuf)-1)
1518         return -EINVAL;
1519     if (copy_from_user(tbuf, buf, count))
1520         return -EFAULT;
1521     tbuf[count] = 0;
1522     simple_strtoul(tbuf, &ep, 0);
1523     if (*ep && *ep != '\n')
1524         return -EINVAL;
1525     /* Note that while we check that 'buf' holds a valid number,
1526      * we always ignore the value and just flush everything.
1527      * Making use of the number leads to races.
1528      */
1529 
1530     now = seconds_since_boot();
1531     /* Always flush everything, so behave like cache_purge()
1532      * Do this by advancing flush_time to the current time,
1533      * or by one second if it has already reached the current time.
1534      * Newly added cache entries will always have ->last_refresh greater
1535      * that ->flush_time, so they don't get flushed prematurely.
1536      */
1537 
1538     if (cd->flush_time >= now)
1539         now = cd->flush_time + 1;
1540 
1541     cd->flush_time = now;
1542     cd->nextcheck = now;
1543     cache_flush();
1544 
1545     if (cd->flush)
1546         cd->flush();
1547 
1548     *ppos += count;
1549     return count;
1550 }
1551 
1552 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1553                  size_t count, loff_t *ppos)
1554 {
1555     struct cache_detail *cd = pde_data(file_inode(filp));
1556 
1557     return cache_read(filp, buf, count, ppos, cd);
1558 }
1559 
1560 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1561                   size_t count, loff_t *ppos)
1562 {
1563     struct cache_detail *cd = pde_data(file_inode(filp));
1564 
1565     return cache_write(filp, buf, count, ppos, cd);
1566 }
1567 
1568 static __poll_t cache_poll_procfs(struct file *filp, poll_table *wait)
1569 {
1570     struct cache_detail *cd = pde_data(file_inode(filp));
1571 
1572     return cache_poll(filp, wait, cd);
1573 }
1574 
1575 static long cache_ioctl_procfs(struct file *filp,
1576                    unsigned int cmd, unsigned long arg)
1577 {
1578     struct inode *inode = file_inode(filp);
1579     struct cache_detail *cd = pde_data(inode);
1580 
1581     return cache_ioctl(inode, filp, cmd, arg, cd);
1582 }
1583 
1584 static int cache_open_procfs(struct inode *inode, struct file *filp)
1585 {
1586     struct cache_detail *cd = pde_data(inode);
1587 
1588     return cache_open(inode, filp, cd);
1589 }
1590 
1591 static int cache_release_procfs(struct inode *inode, struct file *filp)
1592 {
1593     struct cache_detail *cd = pde_data(inode);
1594 
1595     return cache_release(inode, filp, cd);
1596 }
1597 
1598 static const struct proc_ops cache_channel_proc_ops = {
1599     .proc_lseek = no_llseek,
1600     .proc_read  = cache_read_procfs,
1601     .proc_write = cache_write_procfs,
1602     .proc_poll  = cache_poll_procfs,
1603     .proc_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1604     .proc_open  = cache_open_procfs,
1605     .proc_release   = cache_release_procfs,
1606 };
1607 
1608 static int content_open_procfs(struct inode *inode, struct file *filp)
1609 {
1610     struct cache_detail *cd = pde_data(inode);
1611 
1612     return content_open(inode, filp, cd);
1613 }
1614 
1615 static int content_release_procfs(struct inode *inode, struct file *filp)
1616 {
1617     struct cache_detail *cd = pde_data(inode);
1618 
1619     return content_release(inode, filp, cd);
1620 }
1621 
1622 static const struct proc_ops content_proc_ops = {
1623     .proc_open  = content_open_procfs,
1624     .proc_read  = seq_read,
1625     .proc_lseek = seq_lseek,
1626     .proc_release   = content_release_procfs,
1627 };
1628 
1629 static int open_flush_procfs(struct inode *inode, struct file *filp)
1630 {
1631     struct cache_detail *cd = pde_data(inode);
1632 
1633     return open_flush(inode, filp, cd);
1634 }
1635 
1636 static int release_flush_procfs(struct inode *inode, struct file *filp)
1637 {
1638     struct cache_detail *cd = pde_data(inode);
1639 
1640     return release_flush(inode, filp, cd);
1641 }
1642 
1643 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1644                 size_t count, loff_t *ppos)
1645 {
1646     struct cache_detail *cd = pde_data(file_inode(filp));
1647 
1648     return read_flush(filp, buf, count, ppos, cd);
1649 }
1650 
1651 static ssize_t write_flush_procfs(struct file *filp,
1652                   const char __user *buf,
1653                   size_t count, loff_t *ppos)
1654 {
1655     struct cache_detail *cd = pde_data(file_inode(filp));
1656 
1657     return write_flush(filp, buf, count, ppos, cd);
1658 }
1659 
1660 static const struct proc_ops cache_flush_proc_ops = {
1661     .proc_open  = open_flush_procfs,
1662     .proc_read  = read_flush_procfs,
1663     .proc_write = write_flush_procfs,
1664     .proc_release   = release_flush_procfs,
1665     .proc_lseek = no_llseek,
1666 };
1667 
1668 static void remove_cache_proc_entries(struct cache_detail *cd)
1669 {
1670     if (cd->procfs) {
1671         proc_remove(cd->procfs);
1672         cd->procfs = NULL;
1673     }
1674 }
1675 
1676 #ifdef CONFIG_PROC_FS
1677 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1678 {
1679     struct proc_dir_entry *p;
1680     struct sunrpc_net *sn;
1681 
1682     sn = net_generic(net, sunrpc_net_id);
1683     cd->procfs = proc_mkdir(cd->name, sn->proc_net_rpc);
1684     if (cd->procfs == NULL)
1685         goto out_nomem;
1686 
1687     p = proc_create_data("flush", S_IFREG | 0600,
1688                  cd->procfs, &cache_flush_proc_ops, cd);
1689     if (p == NULL)
1690         goto out_nomem;
1691 
1692     if (cd->cache_request || cd->cache_parse) {
1693         p = proc_create_data("channel", S_IFREG | 0600, cd->procfs,
1694                      &cache_channel_proc_ops, cd);
1695         if (p == NULL)
1696             goto out_nomem;
1697     }
1698     if (cd->cache_show) {
1699         p = proc_create_data("content", S_IFREG | 0400, cd->procfs,
1700                      &content_proc_ops, cd);
1701         if (p == NULL)
1702             goto out_nomem;
1703     }
1704     return 0;
1705 out_nomem:
1706     remove_cache_proc_entries(cd);
1707     return -ENOMEM;
1708 }
1709 #else /* CONFIG_PROC_FS */
1710 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1711 {
1712     return 0;
1713 }
1714 #endif
1715 
1716 void __init cache_initialize(void)
1717 {
1718     INIT_DEFERRABLE_WORK(&cache_cleaner, do_cache_clean);
1719 }
1720 
1721 int cache_register_net(struct cache_detail *cd, struct net *net)
1722 {
1723     int ret;
1724 
1725     sunrpc_init_cache_detail(cd);
1726     ret = create_cache_proc_entries(cd, net);
1727     if (ret)
1728         sunrpc_destroy_cache_detail(cd);
1729     return ret;
1730 }
1731 EXPORT_SYMBOL_GPL(cache_register_net);
1732 
1733 void cache_unregister_net(struct cache_detail *cd, struct net *net)
1734 {
1735     remove_cache_proc_entries(cd);
1736     sunrpc_destroy_cache_detail(cd);
1737 }
1738 EXPORT_SYMBOL_GPL(cache_unregister_net);
1739 
1740 struct cache_detail *cache_create_net(const struct cache_detail *tmpl, struct net *net)
1741 {
1742     struct cache_detail *cd;
1743     int i;
1744 
1745     cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1746     if (cd == NULL)
1747         return ERR_PTR(-ENOMEM);
1748 
1749     cd->hash_table = kcalloc(cd->hash_size, sizeof(struct hlist_head),
1750                  GFP_KERNEL);
1751     if (cd->hash_table == NULL) {
1752         kfree(cd);
1753         return ERR_PTR(-ENOMEM);
1754     }
1755 
1756     for (i = 0; i < cd->hash_size; i++)
1757         INIT_HLIST_HEAD(&cd->hash_table[i]);
1758     cd->net = net;
1759     return cd;
1760 }
1761 EXPORT_SYMBOL_GPL(cache_create_net);
1762 
1763 void cache_destroy_net(struct cache_detail *cd, struct net *net)
1764 {
1765     kfree(cd->hash_table);
1766     kfree(cd);
1767 }
1768 EXPORT_SYMBOL_GPL(cache_destroy_net);
1769 
1770 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1771                  size_t count, loff_t *ppos)
1772 {
1773     struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1774 
1775     return cache_read(filp, buf, count, ppos, cd);
1776 }
1777 
1778 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1779                   size_t count, loff_t *ppos)
1780 {
1781     struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1782 
1783     return cache_write(filp, buf, count, ppos, cd);
1784 }
1785 
1786 static __poll_t cache_poll_pipefs(struct file *filp, poll_table *wait)
1787 {
1788     struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1789 
1790     return cache_poll(filp, wait, cd);
1791 }
1792 
1793 static long cache_ioctl_pipefs(struct file *filp,
1794                   unsigned int cmd, unsigned long arg)
1795 {
1796     struct inode *inode = file_inode(filp);
1797     struct cache_detail *cd = RPC_I(inode)->private;
1798 
1799     return cache_ioctl(inode, filp, cmd, arg, cd);
1800 }
1801 
1802 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1803 {
1804     struct cache_detail *cd = RPC_I(inode)->private;
1805 
1806     return cache_open(inode, filp, cd);
1807 }
1808 
1809 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1810 {
1811     struct cache_detail *cd = RPC_I(inode)->private;
1812 
1813     return cache_release(inode, filp, cd);
1814 }
1815 
1816 const struct file_operations cache_file_operations_pipefs = {
1817     .owner      = THIS_MODULE,
1818     .llseek     = no_llseek,
1819     .read       = cache_read_pipefs,
1820     .write      = cache_write_pipefs,
1821     .poll       = cache_poll_pipefs,
1822     .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1823     .open       = cache_open_pipefs,
1824     .release    = cache_release_pipefs,
1825 };
1826 
1827 static int content_open_pipefs(struct inode *inode, struct file *filp)
1828 {
1829     struct cache_detail *cd = RPC_I(inode)->private;
1830 
1831     return content_open(inode, filp, cd);
1832 }
1833 
1834 static int content_release_pipefs(struct inode *inode, struct file *filp)
1835 {
1836     struct cache_detail *cd = RPC_I(inode)->private;
1837 
1838     return content_release(inode, filp, cd);
1839 }
1840 
1841 const struct file_operations content_file_operations_pipefs = {
1842     .open       = content_open_pipefs,
1843     .read       = seq_read,
1844     .llseek     = seq_lseek,
1845     .release    = content_release_pipefs,
1846 };
1847 
1848 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1849 {
1850     struct cache_detail *cd = RPC_I(inode)->private;
1851 
1852     return open_flush(inode, filp, cd);
1853 }
1854 
1855 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1856 {
1857     struct cache_detail *cd = RPC_I(inode)->private;
1858 
1859     return release_flush(inode, filp, cd);
1860 }
1861 
1862 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1863                 size_t count, loff_t *ppos)
1864 {
1865     struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1866 
1867     return read_flush(filp, buf, count, ppos, cd);
1868 }
1869 
1870 static ssize_t write_flush_pipefs(struct file *filp,
1871                   const char __user *buf,
1872                   size_t count, loff_t *ppos)
1873 {
1874     struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1875 
1876     return write_flush(filp, buf, count, ppos, cd);
1877 }
1878 
1879 const struct file_operations cache_flush_operations_pipefs = {
1880     .open       = open_flush_pipefs,
1881     .read       = read_flush_pipefs,
1882     .write      = write_flush_pipefs,
1883     .release    = release_flush_pipefs,
1884     .llseek     = no_llseek,
1885 };
1886 
1887 int sunrpc_cache_register_pipefs(struct dentry *parent,
1888                  const char *name, umode_t umode,
1889                  struct cache_detail *cd)
1890 {
1891     struct dentry *dir = rpc_create_cache_dir(parent, name, umode, cd);
1892     if (IS_ERR(dir))
1893         return PTR_ERR(dir);
1894     cd->pipefs = dir;
1895     return 0;
1896 }
1897 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1898 
1899 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1900 {
1901     if (cd->pipefs) {
1902         rpc_remove_cache_dir(cd->pipefs);
1903         cd->pipefs = NULL;
1904     }
1905 }
1906 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1907 
1908 void sunrpc_cache_unhash(struct cache_detail *cd, struct cache_head *h)
1909 {
1910     spin_lock(&cd->hash_lock);
1911     if (!hlist_unhashed(&h->cache_list)){
1912         sunrpc_begin_cache_remove_entry(h, cd);
1913         spin_unlock(&cd->hash_lock);
1914         sunrpc_end_cache_remove_entry(h, cd);
1915     } else
1916         spin_unlock(&cd->hash_lock);
1917 }
1918 EXPORT_SYMBOL_GPL(sunrpc_cache_unhash);