fs/nfs/pagelist.c

   1 /*
   2  * linux/fs/nfs/pagelist.c
   3  *
   4  * A set of helper functions for managing NFS read and write requests.
   5  * The main purpose of these routines is to provide support for the
   6  * coalescing of several requests into a single RPC call.
   7  *
   8  * Copyright 2000, 2001 (c) Trond Myklebust <trond.myklebust@fys.uio.no>
   9  *
  10  */
  11
  12 #include <linux/slab.h>
  13 #include <linux/file.h>
  14 #include <linux/sched.h>
  15 #include <linux/sunrpc/clnt.h>
  16 #include <linux/nfs.h>
  17 #include <linux/nfs3.h>
  18 #include <linux/nfs4.h>
  19 #include <linux/nfs_page.h>
  20 #include <linux/nfs_fs.h>
  21 #include <linux/nfs_mount.h>
  22 #include <linux/export.h>
  23
  24 #include "internal.h"
  25 #include "pnfs.h"
  26
  27 static struct kmem_cache *nfs_page_cachep;
  28
  29 bool nfs_pgarray_set(struct nfs_page_array *p, unsigned int pagecount)
  30 {
  31         p->npages = pagecount;
  32         if (pagecount <= ARRAY_SIZE(p->page_array))
  33                 p->pagevec = p->page_array;
  34         else {
  35                 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL);
  36                 if (!p->pagevec)
  37                         p->npages = 0;
  38         }
  39         return p->pagevec != NULL;
  40 }
  41
  42 void nfs_pgheader_init(struct nfs_pageio_descriptor *desc,
  43                        struct nfs_pgio_header *hdr,
  44                        void (*release)(struct nfs_pgio_header *hdr))
  45 {
  46         hdr->req = nfs_list_entry(desc->pg_list.next);
  47         hdr->inode = desc->pg_inode;
  48         hdr->cred = hdr->req->wb_context->cred;
  49         hdr->io_start = req_offset(hdr->req);
  50         hdr->good_bytes = desc->pg_count;
  51         hdr->dreq = desc->pg_dreq;
  52         hdr->release = release;
  53         hdr->completion_ops = desc->pg_completion_ops;
  54         if (hdr->completion_ops->init_hdr)
  55                 hdr->completion_ops->init_hdr(hdr);
  56 }
  57 EXPORT_SYMBOL_GPL(nfs_pgheader_init);
  58
  59 void nfs_set_pgio_error(struct nfs_pgio_header *hdr, int error, loff_t pos)
  60 {
  61         spin_lock(&hdr->lock);
  62         if (pos < hdr->io_start + hdr->good_bytes) {
  63                 set_bit(NFS_IOHDR_ERROR, &hdr->flags);
  64                 clear_bit(NFS_IOHDR_EOF, &hdr->flags);
  65                 hdr->good_bytes = pos - hdr->io_start;
  66                 hdr->error = error;
  67         }
  68         spin_unlock(&hdr->lock);
  69 }
  70
  71 static inline struct nfs_page *
  72 nfs_page_alloc(void)
  73 {
  74         struct nfs_page *p = kmem_cache_zalloc(nfs_page_cachep, GFP_NOIO);
  75         if (p)
  76                 INIT_LIST_HEAD(&p->wb_list);
  77         return p;
  78 }
  79
  80 static inline void
  81 nfs_page_free(struct nfs_page *p)
  82 {
  83         kmem_cache_free(nfs_page_cachep, p);
  84 }
  85
  86 /**
  87  * nfs_create_request - Create an NFS read/write request.
  88  * @ctx: open context to use
  89  * @inode: inode to which the request is attached
  90  * @page: page to write
  91  * @offset: starting offset within the page for the write
  92  * @count: number of bytes to read/write
  93  *
  94  * The page must be locked by the caller. This makes sure we never
  95  * create two different requests for the same page.
  96  * User should ensure it is safe to sleep in this function.
  97  */
  98 struct nfs_page *
  99 nfs_create_request(struct nfs_open_context *ctx, struct inode *inode,
 100                    struct page *page,
 101                    unsigned int offset, unsigned int count)
 102 {
 103         struct nfs_page         *req;
 104
 105         /* try to allocate the request struct */
 106         req = nfs_page_alloc();
 107         if (req == NULL)
 108                 return ERR_PTR(-ENOMEM);
 109
 110         /* get lock context early so we can deal with alloc failures */
 111         req->wb_lock_context = nfs_get_lock_context(ctx);
 112         if (req->wb_lock_context == NULL) {
 113                 nfs_page_free(req);
 114                 return ERR_PTR(-ENOMEM);
 115         }
 116
 117         /* Initialize the request struct. Initially, we assume a
 118          * long write-back delay. This will be adjusted in
 119          * update_nfs_request below if the region is not locked. */
 120         req->wb_page    = page;
 121         req->wb_index   = page_file_index(page);
 122         page_cache_get(page);
 123         req->wb_offset  = offset;
 124         req->wb_pgbase  = offset;
 125         req->wb_bytes   = count;
 126         req->wb_context = get_nfs_open_context(ctx);
 127         kref_init(&req->wb_kref);
 128         return req;
 129 }
 130
 131 /**
 132  * nfs_unlock_request - Unlock request and wake up sleepers.
 133  * @req:
 134  */
 135 void nfs_unlock_request(struct nfs_page *req)
 136 {
 137         if (!NFS_WBACK_BUSY(req)) {
 138                 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
 139                 BUG();
 140         }
 141         smp_mb__before_clear_bit();
 142         clear_bit(PG_BUSY, &req->wb_flags);
 143         smp_mb__after_clear_bit();
 144         wake_up_bit(&req->wb_flags, PG_BUSY);
 145 }
 146
 147 /**
 148  * nfs_unlock_and_release_request - Unlock request and release the nfs_page
 149  * @req:
 150  */
 151 void nfs_unlock_and_release_request(struct nfs_page *req)
 152 {
 153         nfs_unlock_request(req);
 154         nfs_release_request(req);
 155 }
 156
 157 /*
 158  * nfs_clear_request - Free up all resources allocated to the request
 159  * @req:
 160  *
 161  * Release page and open context resources associated with a read/write
 162  * request after it has completed.
 163  */
 164 static void nfs_clear_request(struct nfs_page *req)
 165 {
 166         struct page *page = req->wb_page;
 167         struct nfs_open_context *ctx = req->wb_context;
 168         struct nfs_lock_context *l_ctx = req->wb_lock_context;
 169
 170         if (page != NULL) {
 171                 page_cache_release(page);
 172                 req->wb_page = NULL;
 173         }
 174         if (l_ctx != NULL) {
 175                 nfs_put_lock_context(l_ctx);
 176                 req->wb_lock_context = NULL;
 177         }
 178         if (ctx != NULL) {
 179                 put_nfs_open_context(ctx);
 180                 req->wb_context = NULL;
 181         }
 182 }
 183
 184
 185 /**
 186  * nfs_release_request - Release the count on an NFS read/write request
 187  * @req: request to release
 188  *
 189  * Note: Should never be called with the spinlock held!
 190  */
 191 static void nfs_free_request(struct kref *kref)
 192 {
 193         struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
 194
 195         /* Release struct file and open context */
 196         nfs_clear_request(req);
 197         nfs_page_free(req);
 198 }
 199
 200 void nfs_release_request(struct nfs_page *req)
 201 {
 202         kref_put(&req->wb_kref, nfs_free_request);
 203 }
 204
 205 static int nfs_wait_bit_uninterruptible(void *word)
 206 {
 207         io_schedule();
 208         return 0;
 209 }
 210
 211 /**
 212  * nfs_wait_on_request - Wait for a request to complete.
 213  * @req: request to wait upon.
 214  *
 215  * Interruptible by fatal signals only.
 216  * The user is responsible for holding a count on the request.
 217  */
 218 int
 219 nfs_wait_on_request(struct nfs_page *req)
 220 {
 221         return wait_on_bit(&req->wb_flags, PG_BUSY,
 222                         nfs_wait_bit_uninterruptible,
 223                         TASK_UNINTERRUPTIBLE);
 224 }
 225
 226 bool nfs_generic_pg_test(struct nfs_pageio_descriptor *desc, struct nfs_page *prev, struct nfs_page *req)
 227 {
 228         /*
 229          * FIXME: ideally we should be able to coalesce all requests
 230          * that are not block boundary aligned, but currently this
 231          * is problematic for the case of bsize < PAGE_CACHE_SIZE,
 232          * since nfs_flush_multi and nfs_pagein_multi assume you
 233          * can have only one struct nfs_page.
 234          */
 235         if (desc->pg_bsize < PAGE_SIZE)
 236                 return 0;
 237
 238         return desc->pg_count + req->wb_bytes <= desc->pg_bsize;
 239 }
 240 EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
 241
 242 /**
 243  * nfs_pageio_init - initialise a page io descriptor
 244  * @desc: pointer to descriptor
 245  * @inode: pointer to inode
 246  * @doio: pointer to io function
 247  * @bsize: io block size
 248  * @io_flags: extra parameters for the io function
 249  */
 250 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
 251                      struct inode *inode,
 252                      const struct nfs_pageio_ops *pg_ops,
 253                      const struct nfs_pgio_completion_ops *compl_ops,
 254                      size_t bsize,
 255                      int io_flags)
 256 {
 257         INIT_LIST_HEAD(&desc->pg_list);
 258         desc->pg_bytes_written = 0;
 259         desc->pg_count = 0;
 260         desc->pg_bsize = bsize;
 261         desc->pg_base = 0;
 262         desc->pg_moreio = 0;
 263         desc->pg_recoalesce = 0;
 264         desc->pg_inode = inode;
 265         desc->pg_ops = pg_ops;
 266         desc->pg_completion_ops = compl_ops;
 267         desc->pg_ioflags = io_flags;
 268         desc->pg_error = 0;
 269         desc->pg_lseg = NULL;
 270         desc->pg_dreq = NULL;
 271 }
 272 EXPORT_SYMBOL_GPL(nfs_pageio_init);
 273
 274 /**
 275  * nfs_can_coalesce_requests - test two requests for compatibility
 276  * @prev: pointer to nfs_page
 277  * @req: pointer to nfs_page
 278  *
 279  * The nfs_page structures 'prev' and 'req' are compared to ensure that the
 280  * page data area they describe is contiguous, and that their RPC
 281  * credentials, NFSv4 open state, and lockowners are the same.
 282  *
 283  * Return 'true' if this is the case, else return 'false'.
 284  */
 285 static bool nfs_can_coalesce_requests(struct nfs_page *prev,
 286                                       struct nfs_page *req,
 287                                       struct nfs_pageio_descriptor *pgio)
 288 {
 289         if (req->wb_context->cred != prev->wb_context->cred)
 290                 return false;
 291         if (req->wb_lock_context->lockowner != prev->wb_lock_context->lockowner)
 292                 return false;
 293         if (req->wb_context->state != prev->wb_context->state)
 294                 return false;
 295         if (req->wb_pgbase != 0)
 296                 return false;
 297         if (prev->wb_pgbase + prev->wb_bytes != PAGE_CACHE_SIZE)
 298                 return false;
 299         if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
 300                 return false;
 301         return pgio->pg_ops->pg_test(pgio, prev, req);
 302 }
 303
 304 /**
 305  * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
 306  * @desc: destination io descriptor
 307  * @req: request
 308  *
 309  * Returns true if the request 'req' was successfully coalesced into the
 310  * existing list of pages 'desc'.
 311  */
 312 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
 313                                      struct nfs_page *req)
 314 {
 315         if (desc->pg_count != 0) {
 316                 struct nfs_page *prev;
 317
 318                 prev = nfs_list_entry(desc->pg_list.prev);
 319                 if (!nfs_can_coalesce_requests(prev, req, desc))
 320                         return 0;
 321         } else {
 322                 if (desc->pg_ops->pg_init)
 323                         desc->pg_ops->pg_init(desc, req);
 324                 desc->pg_base = req->wb_pgbase;
 325         }
 326         nfs_list_remove_request(req);
 327         nfs_list_add_request(req, &desc->pg_list);
 328         desc->pg_count += req->wb_bytes;
 329         return 1;
 330 }
 331
 332 /*
 333  * Helper for nfs_pageio_add_request and nfs_pageio_complete
 334  */
 335 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
 336 {
 337         if (!list_empty(&desc->pg_list)) {
 338                 int error = desc->pg_ops->pg_doio(desc);
 339                 if (error < 0)
 340                         desc->pg_error = error;
 341                 else
 342                         desc->pg_bytes_written += desc->pg_count;
 343         }
 344         if (list_empty(&desc->pg_list)) {
 345                 desc->pg_count = 0;
 346                 desc->pg_base = 0;
 347         }
 348 }
 349
 350 /**
 351  * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
 352  * @desc: destination io descriptor
 353  * @req: request
 354  *
 355  * Returns true if the request 'req' was successfully coalesced into the
 356  * existing list of pages 'desc'.
 357  */
 358 static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
 359                            struct nfs_page *req)
 360 {
 361         while (!nfs_pageio_do_add_request(desc, req)) {
 362                 desc->pg_moreio = 1;
 363                 nfs_pageio_doio(desc);
 364                 if (desc->pg_error < 0)
 365                         return 0;
 366                 desc->pg_moreio = 0;
 367                 if (desc->pg_recoalesce)
 368                         return 0;
 369         }
 370         return 1;
 371 }
 372
 373 static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
 374 {
 375         LIST_HEAD(head);
 376
 377         do {
 378                 list_splice_init(&desc->pg_list, &head);
 379                 desc->pg_bytes_written -= desc->pg_count;
 380                 desc->pg_count = 0;
 381                 desc->pg_base = 0;
 382                 desc->pg_recoalesce = 0;
 383
 384                 while (!list_empty(&head)) {
 385                         struct nfs_page *req;
 386
 387                         req = list_first_entry(&head, struct nfs_page, wb_list);
 388                         nfs_list_remove_request(req);
 389                         if (__nfs_pageio_add_request(desc, req))
 390                                 continue;
 391                         if (desc->pg_error < 0)
 392                                 return 0;
 393                         break;
 394                 }
 395         } while (desc->pg_recoalesce);
 396         return 1;
 397 }
 398
 399 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
 400                 struct nfs_page *req)
 401 {
 402         int ret;
 403
 404         do {
 405                 ret = __nfs_pageio_add_request(desc, req);
 406                 if (ret)
 407                         break;
 408                 if (desc->pg_error < 0)
 409                         break;
 410                 ret = nfs_do_recoalesce(desc);
 411         } while (ret);
 412         return ret;
 413 }
 414 EXPORT_SYMBOL_GPL(nfs_pageio_add_request);
 415
 416 /**
 417  * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
 418  * @desc: pointer to io descriptor
 419  */
 420 void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
 421 {
 422         for (;;) {
 423                 nfs_pageio_doio(desc);
 424                 if (!desc->pg_recoalesce)
 425                         break;
 426                 if (!nfs_do_recoalesce(desc))
 427                         break;
 428         }
 429 }
 430 EXPORT_SYMBOL_GPL(nfs_pageio_complete);
 431
 432 /**
 433  * nfs_pageio_cond_complete - Conditional I/O completion
 434  * @desc: pointer to io descriptor
 435  * @index: page index
 436  *
 437  * It is important to ensure that processes don't try to take locks
 438  * on non-contiguous ranges of pages as that might deadlock. This
 439  * function should be called before attempting to wait on a locked
 440  * nfs_page. It will complete the I/O if the page index 'index'
 441  * is not contiguous with the existing list of pages in 'desc'.
 442  */
 443 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
 444 {
 445         if (!list_empty(&desc->pg_list)) {
 446                 struct nfs_page *prev = nfs_list_entry(desc->pg_list.prev);
 447                 if (index != prev->wb_index + 1)
 448                         nfs_pageio_complete(desc);
 449         }
 450 }
 451
 452 int __init nfs_init_nfspagecache(void)
 453 {
 454         nfs_page_cachep = kmem_cache_create("nfs_page",
 455                                             sizeof(struct nfs_page),
 456                                             0, SLAB_HWCACHE_ALIGN,
 457                                             NULL);
 458         if (nfs_page_cachep == NULL)
 459                 return -ENOMEM;
 460
 461         return 0;
 462 }
 463
 464 void nfs_destroy_nfspagecache(void)
 465 {
 466         kmem_cache_destroy(nfs_page_cachep);
 467 }
 468